WEBVTT 00:00:00.000 --> 00:00:01.976 align:middle line:90% [SQUEAKING] 00:00:01.976 --> 00:00:03.452 align:middle line:90% [RUSTLING] 00:00:03.452 --> 00:00:05.420 align:middle line:90% [CLICKING] 00:00:05.420 --> 00:00:12.280 align:middle line:90% 00:00:12.280 --> 00:00:13.780 align:middle line:84% HAZHIR RAHMANDAD: How was last week? 00:00:13.780 --> 00:00:17.720 align:middle line:84% I unfortunately, somehow this got disconnected, 00:00:17.720 --> 00:00:22.940 align:middle line:84% the recording, so I couldn't watch the session. 00:00:22.940 --> 00:00:28.960 align:middle line:84% But what did you guys discuss and how did it go? 00:00:28.960 --> 00:00:32.140 align:middle line:90% AUDIENCE: We talked about-- oh. 00:00:32.140 --> 00:00:33.640 align:middle line:84% Oh, were you looking at [INAUDIBLE]? 00:00:33.640 --> 00:00:34.260 align:middle line:84% HAZHIR RAHMANDAD: It doesn't matter. 00:00:34.260 --> 00:00:35.440 align:middle line:90% Go for it. 00:00:35.440 --> 00:00:37.280 align:middle line:84% AUDIENCE: We started by talking about when 00:00:37.280 --> 00:00:41.097 align:middle line:84% are system dynamics useful, this general idea when 00:00:41.097 --> 00:00:42.680 align:middle line:84% working towards models, which she said 00:00:42.680 --> 00:00:44.880 align:middle line:84% might be different from yours and [INAUDIBLE] but you 00:00:44.880 --> 00:00:46.420 align:middle line:84% wanted to get [INAUDIBLE] perspective on that. 00:00:46.420 --> 00:00:46.860 align:middle line:90% HAZHIR RAHMANDAD: That's great. 00:00:46.860 --> 00:00:49.480 align:middle line:84% AUDIENCE: Then we started going model by model. 00:00:49.480 --> 00:00:51.960 align:middle line:90% [INTERPOSING VOICES] 00:00:51.960 --> 00:00:53.200 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 00:00:53.200 --> 00:00:55.235 align:middle line:90% Cool. 00:00:55.235 --> 00:00:55.860 align:middle line:90% AUDIENCE: Yeah. 00:00:55.860 --> 00:01:00.448 align:middle line:84% We were talking about the debate, 00:01:00.448 --> 00:01:02.740 align:middle line:84% I guess, between whether to start with stocks and flows 00:01:02.740 --> 00:01:05.860 align:middle line:90% versus the feedback loops. 00:01:05.860 --> 00:01:09.600 align:middle line:84% And [INAUDIBLE] was commenting that there's no one answer. 00:01:09.600 --> 00:01:12.820 align:middle line:84% And for mine he said that I would 00:01:12.820 --> 00:01:15.540 align:middle line:84% start with stocks and flows and he 00:01:15.540 --> 00:01:18.700 align:middle line:84% that I should start with the feedback 00:01:18.700 --> 00:01:22.320 align:middle line:84% story and [INAUDIBLE] feedback on that, 00:01:22.320 --> 00:01:25.240 align:middle line:90% and then go towards [INAUDIBLE]. 00:01:25.240 --> 00:01:26.300 align:middle line:90% 00:01:26.300 --> 00:01:31.500 align:middle line:84% So there was an interesting discussion about where 00:01:31.500 --> 00:01:34.140 align:middle line:90% to begin from the Viking Age. 00:01:34.140 --> 00:01:37.940 align:middle line:84% And he said, sometimes the stock and flow stories 00:01:37.940 --> 00:01:41.858 align:middle line:84% are clear and other times the feedback [INAUDIBLE]. 00:01:41.858 --> 00:01:43.900 align:middle line:84% HAZHIR RAHMANDAD: Yeah, that's actually generally 00:01:43.900 --> 00:01:47.300 align:middle line:84% an important architectural choice at the beginning 00:01:47.300 --> 00:01:48.760 align:middle line:90% about where to start. 00:01:48.760 --> 00:01:52.180 align:middle line:84% And it's absolutely the case that for some problems 00:01:52.180 --> 00:01:54.740 align:middle line:84% it is easier to go with the stocks and flows 00:01:54.740 --> 00:01:59.000 align:middle line:84% and sometimes with causal loops would add value too 00:01:59.000 --> 00:02:00.600 align:middle line:90% as a starting point. 00:02:00.600 --> 00:02:01.960 align:middle line:90% Cool. 00:02:01.960 --> 00:02:05.600 align:middle line:84% I didn't see your model in the folder from last week. 00:02:05.600 --> 00:02:06.600 align:middle line:90% AUDIENCE: Yeah. 00:02:06.600 --> 00:02:07.975 align:middle line:84% HAZHIR RAHMANDAD: It's all right, 00:02:07.975 --> 00:02:11.935 align:middle line:84% I just-- so you had something, put it for this week. 00:02:11.935 --> 00:02:12.560 align:middle line:90% AUDIENCE: Yeah. 00:02:12.560 --> 00:02:15.800 align:middle line:90% 00:02:15.800 --> 00:02:18.260 align:middle line:84% HAZHIR RAHMANDAD: Other observations? 00:02:18.260 --> 00:02:22.440 align:middle line:84% Anything else from last session to share? 00:02:22.440 --> 00:02:26.880 align:middle line:84% AUDIENCE: Nelson, I thought it was a useful session. 00:02:26.880 --> 00:02:32.000 align:middle line:84% We spent more time on thinking about problem formulation 00:02:32.000 --> 00:02:36.160 align:middle line:84% and when is modeling useful, what types of problems. 00:02:36.160 --> 00:02:38.840 align:middle line:84% We talked about a few formulations. 00:02:38.840 --> 00:02:52.600 align:middle line:84% But I think the majority of the time was why model at all. 00:02:52.600 --> 00:02:54.020 align:middle line:90% So that was helpful for me. 00:02:54.020 --> 00:02:55.900 align:middle line:84% HAZHIR RAHMANDAD: Yeah, absolutely. 00:02:55.900 --> 00:03:00.740 align:middle line:84% And Nelson has a very good taste in focusing on really 00:03:00.740 --> 00:03:02.640 align:middle line:84% what is the value added of the model. 00:03:02.640 --> 00:03:05.800 align:middle line:84% So that's a great opportunity to get that. 00:03:05.800 --> 00:03:06.300 align:middle line:90% Cool. 00:03:06.300 --> 00:03:09.660 align:middle line:90% 00:03:09.660 --> 00:03:12.245 align:middle line:90% Other thoughts, observations? 00:03:12.245 --> 00:03:14.620 align:middle line:84% AUDIENCE: I also thought that the part of that discussion 00:03:14.620 --> 00:03:18.880 align:middle line:84% on not only just what problem are you trying to solve for who, 00:03:18.880 --> 00:03:23.380 align:middle line:84% but then also what makes it dynamic specifically 00:03:23.380 --> 00:03:26.925 align:middle line:84% or what is your hypothesis that makes 00:03:26.925 --> 00:03:28.300 align:middle line:84% you think that system dynamics is 00:03:28.300 --> 00:03:33.620 align:middle line:84% going to be specifically useful for it versus any other talk. 00:03:33.620 --> 00:03:35.020 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 00:03:35.020 --> 00:03:37.380 align:middle line:84% And it's a conversation we haven't really 00:03:37.380 --> 00:03:40.580 align:middle line:84% focused on in this class because basically each week we 00:03:40.580 --> 00:03:43.740 align:middle line:84% gave a problem that you want to focus on. 00:03:43.740 --> 00:03:47.580 align:middle line:84% And then for your personal projects 00:03:47.580 --> 00:03:51.140 align:middle line:84% I gave the feedback not at the level 00:03:51.140 --> 00:03:54.380 align:middle line:84% of let's completely challenge the problem you're bringing in, 00:03:54.380 --> 00:03:56.120 align:middle line:84% but at the level of, OK, if that is 00:03:56.120 --> 00:03:59.000 align:middle line:84% the problem you want to tackle, let's 00:03:59.000 --> 00:04:03.920 align:middle line:84% try to mold it into something that you can build a model for. 00:04:03.920 --> 00:04:06.360 align:middle line:84% And in many cases, you could argue 00:04:06.360 --> 00:04:10.940 align:middle line:84% that those problems are not the optimal dynamic problem. 00:04:10.940 --> 00:04:14.740 align:middle line:84% So I think that's an important consideration. 00:04:14.740 --> 00:04:18.480 align:middle line:84% In many ways you can build models in Vensim 00:04:18.480 --> 00:04:24.040 align:middle line:84% that are not dynamically that interesting or that complex. 00:04:24.040 --> 00:04:26.100 align:middle line:90% They are accounting models. 00:04:26.100 --> 00:04:28.580 align:middle line:84% They work, they are useful for some purposes, 00:04:28.580 --> 00:04:30.480 align:middle line:84% but you could also build a similar model 00:04:30.480 --> 00:04:36.360 align:middle line:84% in an Excel spreadsheet and it does the same job, give or take. 00:04:36.360 --> 00:04:42.240 align:middle line:84% And that is not really where the heart of system dynamics is 00:04:42.240 --> 00:04:44.340 align:middle line:84% and where the real value of modeling is. 00:04:44.340 --> 00:04:46.660 align:middle line:84% But there is nothing wrong with building those models, 00:04:46.660 --> 00:04:49.140 align:middle line:84% and in some cases, they are actually very useful, 00:04:49.140 --> 00:04:53.330 align:middle line:84% just as Excel models are very useful for a lot of activities. 00:04:53.330 --> 00:04:55.830 align:middle line:84% Now, whether it is more useful to build them 00:04:55.830 --> 00:04:59.390 align:middle line:84% in Vensim versus in some spreadsheet, 00:04:59.390 --> 00:05:02.850 align:middle line:84% then it becomes a more important question. 00:05:02.850 --> 00:05:06.490 align:middle line:84% That is, for some of these accounting type of models, 00:05:06.490 --> 00:05:10.550 align:middle line:84% like models that have basically stocks and flows 00:05:10.550 --> 00:05:15.670 align:middle line:84% without much feedback, sometimes it 00:05:15.670 --> 00:05:20.230 align:middle line:84% works easier to just build them in a spreadsheet, 00:05:20.230 --> 00:05:22.350 align:middle line:84% and sometimes it is actually useful still 00:05:22.350 --> 00:05:28.370 align:middle line:84% to have the explicit stock and flow structure in front of you. 00:05:28.370 --> 00:05:31.810 align:middle line:84% It makes thinking about them a little bit more clear. 00:05:31.810 --> 00:05:32.310 align:middle line:90% Good. 00:05:32.310 --> 00:05:37.390 align:middle line:84% Other observations from last week? 00:05:37.390 --> 00:05:39.230 align:middle line:84% AUDIENCE: He also mentioned that he's 00:05:39.230 --> 00:05:43.250 align:middle line:84% in favor of very simple models, in contrast to you and Jonathan. 00:05:43.250 --> 00:05:43.990 align:middle line:90% Oh. 00:05:43.990 --> 00:05:44.510 align:middle line:90% Yeah. 00:05:44.510 --> 00:05:47.630 align:middle line:84% So that was an interesting observation to me. 00:05:47.630 --> 00:05:51.510 align:middle line:84% And I've seen his papers, but two [INAUDIBLE] models, 00:05:51.510 --> 00:05:54.650 align:middle line:84% I think you're getting saturated in terms 00:05:54.650 --> 00:05:57.870 align:middle line:84% of attacking theoretical environment, maybe, 00:05:57.870 --> 00:05:59.230 align:middle line:90% I'm not really sure. 00:05:59.230 --> 00:06:02.610 align:middle line:84% When you're thinking of even comparing 00:06:02.610 --> 00:06:06.250 align:middle line:84% two theories, the interaction of the different mechanisms in two 00:06:06.250 --> 00:06:11.110 align:middle line:84% theories, the model suddenly gets big and how to tackle that. 00:06:11.110 --> 00:06:15.370 align:middle line:84% HAZHIR RAHMANDAD: Yeah, I would say there are really 00:06:15.370 --> 00:06:17.530 align:middle line:84% two different consideration sets that 00:06:17.530 --> 00:06:21.890 align:middle line:84% come into this, what's the right level of detail and complexity 00:06:21.890 --> 00:06:23.130 align:middle line:90% for a model. 00:06:23.130 --> 00:06:25.870 align:middle line:84% The most important one is what's the purpose of the model. 00:06:25.870 --> 00:06:29.950 align:middle line:84% So if the purpose from a spectrum of theory-making, 00:06:29.950 --> 00:06:32.990 align:middle line:84% theory-building, to policy analysis, 00:06:32.990 --> 00:06:35.730 align:middle line:84% you think of that as spectrum, the more you 00:06:35.730 --> 00:06:39.890 align:middle line:84% are towards the theory-building, the simpler models 00:06:39.890 --> 00:06:42.990 align:middle line:84% add more value for that situation. 00:06:42.990 --> 00:06:47.010 align:middle line:84% And then the more you move towards the policy analysis, 00:06:47.010 --> 00:06:50.450 align:middle line:84% you would need more complex, more detailed models. 00:06:50.450 --> 00:06:54.630 align:middle line:84% So that's one spectrum, one dimension. 00:06:54.630 --> 00:06:59.710 align:middle line:84% Empirical grounding in terms of numerical data 00:06:59.710 --> 00:07:02.310 align:middle line:84% is another dimension that becomes important. 00:07:02.310 --> 00:07:06.270 align:middle line:84% Again, when you want to have more empirically 00:07:06.270 --> 00:07:09.190 align:middle line:84% well-grounded models that are estimated based 00:07:09.190 --> 00:07:13.710 align:middle line:84% on real-world data, usually, not always, but usually very simple 00:07:13.710 --> 00:07:15.370 align:middle line:90% models don't do a great job. 00:07:15.370 --> 00:07:20.490 align:middle line:84% But that's at least a starting very simple model. 00:07:20.490 --> 00:07:22.530 align:middle line:84% Sometimes after doing a lot of modeling, 00:07:22.530 --> 00:07:26.670 align:middle line:84% you can come up with a very simplified version that 00:07:26.670 --> 00:07:31.590 align:middle line:84% can still do a good job of capturing the detailed data 00:07:31.590 --> 00:07:33.150 align:middle line:90% that you may have. 00:07:33.150 --> 00:07:34.890 align:middle line:84% So that's kind of a second dimension. 00:07:34.890 --> 00:07:37.430 align:middle line:84% And then the third dimension is the audience. 00:07:37.430 --> 00:07:42.790 align:middle line:84% So if you are targeting organizational studies 00:07:42.790 --> 00:07:45.790 align:middle line:84% audiences, basically large models 00:07:45.790 --> 00:07:47.830 align:middle line:90% are not going to work for you. 00:07:47.830 --> 00:07:49.870 align:middle line:84% It's just they are not interested. 00:07:49.870 --> 00:07:52.690 align:middle line:84% They don't have the capacity to consume that. 00:07:52.690 --> 00:07:56.050 align:middle line:84% It's just not the type of thing that 00:07:56.050 --> 00:07:57.930 align:middle line:90% would work for that audience. 00:07:57.930 --> 00:08:01.310 align:middle line:84% You go into mathematical epidemiology, 00:08:01.310 --> 00:08:05.590 align:middle line:84% they are much more receptive to these more complex models. 00:08:05.590 --> 00:08:09.350 align:middle line:84% You go to sustainability, energy, economics, 00:08:09.350 --> 00:08:13.570 align:middle line:84% they are more receptive to complex models. 00:08:13.570 --> 00:08:17.310 align:middle line:84% So that's another dimension to think about. 00:08:17.310 --> 00:08:21.250 align:middle line:84% So for a typical business school PhD 00:08:21.250 --> 00:08:23.910 align:middle line:84% student who is targeting business school audiences, 00:08:23.910 --> 00:08:29.930 align:middle line:84% I think Nelson's point about smaller models 00:08:29.930 --> 00:08:32.890 align:middle line:84% is a very good advice for the typical audience they 00:08:32.890 --> 00:08:35.650 align:middle line:90% are dealing with. 00:08:35.650 --> 00:08:39.049 align:middle line:84% So I think my experience has been 00:08:39.049 --> 00:08:41.190 align:middle line:84% you have to know your audience and your purpose, 00:08:41.190 --> 00:08:46.890 align:middle line:84% and based on that, decide on the complexity of the model. 00:08:46.890 --> 00:08:51.550 align:middle line:84% There is a little bit of a process step 00:08:51.550 --> 00:08:56.830 align:middle line:84% as well that, when you start building a model, 00:08:56.830 --> 00:08:59.970 align:middle line:84% you usually don't know what are the best simplifications. 00:08:59.970 --> 00:09:05.590 align:middle line:84% So usually you end up building something more complex, 00:09:05.590 --> 00:09:08.750 align:middle line:84% and then you distill the main insights 00:09:08.750 --> 00:09:13.030 align:middle line:84% from that more complex model and try to get to something simpler. 00:09:13.030 --> 00:09:21.110 align:middle line:84% So some of my PhD thesis work I built this very detailed model 00:09:21.110 --> 00:09:24.330 align:middle line:84% of software development projects, 00:09:24.330 --> 00:09:31.790 align:middle line:84% multiple releases, and working on cases studies in an IT firm. 00:09:31.790 --> 00:09:34.130 align:middle line:84% It was a fairly detailed large model. 00:09:34.130 --> 00:09:37.170 align:middle line:84% I built a management flight simulator on top of it, 00:09:37.170 --> 00:09:39.730 align:middle line:84% but none of that could go into a management journal. 00:09:39.730 --> 00:09:44.170 align:middle line:84% And then I spent quite a bit of time reducing that model 00:09:44.170 --> 00:09:50.330 align:middle line:84% into a core theoretical insight type of a model that was two 00:09:50.330 --> 00:09:52.130 align:middle line:90% orders of magnitude simpler. 00:09:52.130 --> 00:09:57.770 align:middle line:84% We were going from really 2,000 equations to 20 equations. 00:09:57.770 --> 00:10:01.290 align:middle line:84% But I couldn't build that two-order-of-magnitude-simpler 00:10:01.290 --> 00:10:03.570 align:middle line:90% model as my first model. 00:10:03.570 --> 00:10:09.570 align:middle line:84% Now, maybe somebody who had done at the time a lot more modeling 00:10:09.570 --> 00:10:12.330 align:middle line:84% and had seen a lot more and knew the theory that they are trying 00:10:12.330 --> 00:10:16.130 align:middle line:84% to contribute to could build a version of that simpler 00:10:16.130 --> 00:10:18.150 align:middle line:90% model or model from scratch. 00:10:18.150 --> 00:10:21.490 align:middle line:84% And these days I do build very simple models from scratch, 00:10:21.490 --> 00:10:23.170 align:middle line:84% because I know the theory and I know 00:10:23.170 --> 00:10:25.170 align:middle line:84% what is missing in this space and what 00:10:25.170 --> 00:10:26.850 align:middle line:90% would be a contribution. 00:10:26.850 --> 00:10:32.030 align:middle line:84% As a starting junior researcher, I didn't have that big picture. 00:10:32.030 --> 00:10:34.770 align:middle line:84% So this journey through a complex model 00:10:34.770 --> 00:10:38.830 align:middle line:84% back into a simpler model was maybe the feasible path for me. 00:10:38.830 --> 00:10:41.550 align:middle line:84% So those are some of the considerations. 00:10:41.550 --> 00:10:43.150 align:middle line:90% [INTERPOSING VOICES] 00:10:43.150 --> 00:10:44.750 align:middle line:90% AUDIENCE: Very quickly. 00:10:44.750 --> 00:10:48.890 align:middle line:84% Nelson also mentioned that when you're adding another structure, 00:10:48.890 --> 00:10:51.950 align:middle line:84% your understanding of the model should go hand in hand 00:10:51.950 --> 00:10:54.550 align:middle line:90% with the extension of the model. 00:10:54.550 --> 00:10:56.470 align:middle line:84% And what you were saying is that since you 00:10:56.470 --> 00:10:57.990 align:middle line:84% have gone through that path and you 00:10:57.990 --> 00:11:02.065 align:middle line:84% have understood all the details, you were able to simplify. 00:11:02.065 --> 00:11:04.190 align:middle line:84% HAZHIR RAHMANDAD: Some of it is that but some of it 00:11:04.190 --> 00:11:10.310 align:middle line:84% is, as you build a model, because you're reflecting what 00:11:10.310 --> 00:11:15.350 align:middle line:84% you think is important in an empirical setting, that set 00:11:15.350 --> 00:11:17.910 align:middle line:84% of important mechanisms in an empirical setting 00:11:17.910 --> 00:11:20.830 align:middle line:84% do not necessarily overlap with the theoretically 00:11:20.830 --> 00:11:23.150 align:middle line:84% interesting mechanisms, because what 00:11:23.150 --> 00:11:28.470 align:middle line:84% is theoretically interesting in researcher space is what is new. 00:11:28.470 --> 00:11:32.190 align:middle line:84% So basically you need to find a way 00:11:32.190 --> 00:11:36.570 align:middle line:84% of marrying what you see as important in the real world, 00:11:36.570 --> 00:11:41.810 align:middle line:84% if you're doing a case-based detailed modeling with what you 00:11:41.810 --> 00:11:45.010 align:middle line:84% have learned from the theory as, OK, everybody has already 00:11:45.010 --> 00:11:48.570 align:middle line:84% talked about project dynamics, so there is nothing new. 00:11:48.570 --> 00:11:52.330 align:middle line:84% Putting down the project dynamics structure 00:11:52.330 --> 00:11:59.130 align:middle line:84% is not a contribution and even multiple release type 00:11:59.130 --> 00:12:01.970 align:middle line:84% a-- so you have a platform and you're 00:12:01.970 --> 00:12:05.850 align:middle line:84% building different projects on top of that platform per se 00:12:05.850 --> 00:12:07.350 align:middle line:90% is not that new. 00:12:07.350 --> 00:12:09.290 align:middle line:84% But then there are some dynamics that 00:12:09.290 --> 00:12:13.710 align:middle line:84% come from-- so I didn't have that big picture. 00:12:13.710 --> 00:12:18.130 align:middle line:84% I first needed to extract all of the mechanisms 00:12:18.130 --> 00:12:22.730 align:middle line:84% that empirically I was seeing as interesting, relevant, 00:12:22.730 --> 00:12:25.330 align:middle line:84% build them into my model and then 00:12:25.330 --> 00:12:30.210 align:middle line:84% decide that, OK, I throw out 90% of these because they are not, 00:12:30.210 --> 00:12:33.530 align:middle line:84% research-wise, giving me a novel angle 00:12:33.530 --> 00:12:36.250 align:middle line:84% to go and sell to the academic audiences 00:12:36.250 --> 00:12:40.270 align:middle line:84% and take that 10% that is new, that also 00:12:40.270 --> 00:12:44.190 align:middle line:84% is relatively important in my story, 00:12:44.190 --> 00:12:46.550 align:middle line:90% and it's not the whole story. 00:12:46.550 --> 00:12:48.630 align:middle line:84% Empirically, if I wanted to estimate 00:12:48.630 --> 00:12:51.910 align:middle line:84% what was happening in the case, maybe 00:12:51.910 --> 00:12:56.350 align:middle line:84% those theoretical mechanisms that I picked up 00:12:56.350 --> 00:13:01.690 align:middle line:84% for my theory paper would have explained 10% of the variance. 00:13:01.690 --> 00:13:05.810 align:middle line:84% They weren't the only thing or even the primary thing, 00:13:05.810 --> 00:13:07.470 align:middle line:84% but they were important enough that I 00:13:07.470 --> 00:13:10.890 align:middle line:84% could bring qualitative data to support them. 00:13:10.890 --> 00:13:14.550 align:middle line:84% I could tell a internally consistent story 00:13:14.550 --> 00:13:17.950 align:middle line:84% in a simple model that why they would potentially be relevant 00:13:17.950 --> 00:13:20.650 align:middle line:90% even beyond this specific case. 00:13:20.650 --> 00:13:24.830 align:middle line:84% And that made the paper ultimately. 00:13:24.830 --> 00:13:27.310 align:middle line:90% So I don't know. 00:13:27.310 --> 00:13:29.250 align:middle line:90% It's not an easy journey. 00:13:29.250 --> 00:13:34.470 align:middle line:84% It's the path that I took is not necessarily 00:13:34.470 --> 00:13:38.090 align:middle line:84% the right allocation of your limited time 00:13:38.090 --> 00:13:42.970 align:middle line:84% if you are trying to maximize your publications, going 00:13:42.970 --> 00:13:45.730 align:middle line:84% with a simple model from the start, 00:13:45.730 --> 00:13:49.590 align:middle line:84% if you're targeting management audiences, is better. 00:13:49.590 --> 00:13:51.790 align:middle line:84% But then how do you get to that simple model? 00:13:51.790 --> 00:13:54.550 align:middle line:84% You need to know the literature pretty well. 00:13:54.550 --> 00:13:57.190 align:middle line:84% You need to know what are the models that are around 00:13:57.190 --> 00:14:04.850 align:middle line:84% and then try to find an angle that has not been approached 00:14:04.850 --> 00:14:07.330 align:middle line:90% and used before. 00:14:07.330 --> 00:14:08.750 align:middle line:90% Good. 00:14:08.750 --> 00:14:12.350 align:middle line:84% So we have a few models here and maybe more are coming. 00:14:12.350 --> 00:14:15.550 align:middle line:84% And my Dropbox is a little bit weird today, 00:14:15.550 --> 00:14:19.570 align:middle line:84% but now it says up to date, so hopefully it's all working. 00:14:19.570 --> 00:14:27.110 align:middle line:84% So I think we have Ross and Cathy's first submissions. 00:14:27.110 --> 00:14:31.370 align:middle line:84% So I'll start with those and then-- and also Serchin's. 00:14:31.370 --> 00:14:35.520 align:middle line:84% And we will then go to the ones that we had seen 00:14:35.520 --> 00:14:37.800 align:middle line:90% in the previous week as well. 00:14:37.800 --> 00:14:41.880 align:middle line:90% This is Ross's model. 00:14:41.880 --> 00:14:50.960 align:middle line:84% So, Ross, would you like to give us 00:14:50.960 --> 00:14:55.300 align:middle line:84% a quick overview of what you're doing in this model? 00:14:55.300 --> 00:14:58.040 align:middle line:84% So the audience, all of us probably 00:14:58.040 --> 00:15:00.440 align:middle line:90% a little bit faster than-- 00:15:00.440 --> 00:15:02.080 align:middle line:84% AUDIENCE: So the one with inertia 00:15:02.080 --> 00:15:06.900 align:middle line:84% are the accessibility of the product as observed by customers 00:15:06.900 --> 00:15:12.880 align:middle line:84% and then the prestige associated with the brand. 00:15:12.880 --> 00:15:17.200 align:middle line:84% Accessibility is a function of how much of the demand 00:15:17.200 --> 00:15:20.760 align:middle line:84% was met with the sales and what portion of the sales 00:15:20.760 --> 00:15:24.920 align:middle line:84% was from-- for the price-sensitive customers. 00:15:24.920 --> 00:15:29.560 align:middle line:84% And then the prestige comes from the relative price 00:15:29.560 --> 00:15:34.540 align:middle line:84% of your brand to the competitor, and then quality 00:15:34.540 --> 00:15:37.140 align:middle line:90% and then the accessibility. 00:15:37.140 --> 00:15:42.140 align:middle line:90% The demand is on the lower side. 00:15:42.140 --> 00:15:44.220 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 00:15:44.220 --> 00:15:45.780 align:middle line:90% AUDIENCE: I wanted to make it-- 00:15:45.780 --> 00:15:48.180 align:middle line:84% when the prestige goes to zero, it 00:15:48.180 --> 00:15:51.060 align:middle line:90% acts like just a common goods. 00:15:51.060 --> 00:15:53.720 align:middle line:84% HAZHIR RAHMANDAD: So let me first, big picture, 00:15:53.720 --> 00:15:55.640 align:middle line:84% make sure I'm following what you're doing. 00:15:55.640 --> 00:15:59.980 align:middle line:84% So you're looking at a kind of demand and supply 00:15:59.980 --> 00:16:01.700 align:middle line:84% or maybe just the demand for a type 00:16:01.700 --> 00:16:06.220 align:middle line:84% of good that could be luxury and therefore prestige 00:16:06.220 --> 00:16:08.120 align:middle line:90% matters in terms of demand. 00:16:08.120 --> 00:16:11.900 align:middle line:84% You're dividing the population into two groups, those 00:16:11.900 --> 00:16:15.100 align:middle line:84% who are more sensitive about the price 00:16:15.100 --> 00:16:17.180 align:middle line:84% to quality ratio, and then the ones 00:16:17.180 --> 00:16:20.700 align:middle line:84% that for whom prestige also matters. 00:16:20.700 --> 00:16:23.660 align:middle line:84% Then you are modeling demand coming 00:16:23.660 --> 00:16:26.140 align:middle line:84% from both of these groups based on the characteristics 00:16:26.140 --> 00:16:28.060 align:middle line:84% of the product, those characteristics 00:16:28.060 --> 00:16:32.600 align:middle line:84% that matter here are the prestige, the price, 00:16:32.600 --> 00:16:34.280 align:middle line:90% and the accessibility. 00:16:34.280 --> 00:16:36.560 align:middle line:90% Is there also quality, you said? 00:16:36.560 --> 00:16:39.680 align:middle line:84% AUDIENCE: Yeah, quality is cost, but it's [INAUDIBLE] here. 00:16:39.680 --> 00:16:42.360 align:middle line:84% And then the-- I think quantity is also 00:16:42.360 --> 00:16:44.440 align:middle line:84% something-- quantity supplied is also 00:16:44.440 --> 00:16:46.203 align:middle line:84% something you can control at the company. 00:16:46.203 --> 00:16:47.120 align:middle line:90% HAZHIR RAHMANDAD: Yes. 00:16:47.120 --> 00:16:47.620 align:middle line:90% OK. 00:16:47.620 --> 00:16:52.605 align:middle line:84% And then the company has control over price and the quantity. 00:16:52.605 --> 00:16:54.980 align:middle line:84% AUDIENCE: Price, quantity, quality, the quality is fixed. 00:16:54.980 --> 00:16:57.140 align:middle line:84% HAZHIR RAHMANDAD: OK, for now we are fixing the quality. 00:16:57.140 --> 00:16:57.280 align:middle line:90% AUDIENCE: Yeah. 00:16:57.280 --> 00:16:58.738 align:middle line:84% HAZHIR RAHMANDAD: That makes sense. 00:16:58.738 --> 00:17:00.680 align:middle line:90% So let's not-- good. 00:17:00.680 --> 00:17:02.920 align:middle line:84% And basically then there is the question 00:17:02.920 --> 00:17:05.760 align:middle line:84% of the trade-offs the company faces 00:17:05.760 --> 00:17:10.079 align:middle line:84% in terms of the overall demand they would expect, 00:17:10.079 --> 00:17:13.960 align:middle line:84% given these two policy levers at their disposal. 00:17:13.960 --> 00:17:15.880 align:middle line:84% You are not looking at competition 00:17:15.880 --> 00:17:18.280 align:middle line:84% or are you looking at competition? 00:17:18.280 --> 00:17:22.760 align:middle line:84% AUDIENCE: I could, but it seems too complex. 00:17:22.760 --> 00:17:24.599 align:middle line:84% HAZHIR RAHMANDAD: Yeah, I think that's fine. 00:17:24.599 --> 00:17:27.500 align:middle line:84% I don't want to push you in that direction. 00:17:27.500 --> 00:17:28.900 align:middle line:90% Just asking. 00:17:28.900 --> 00:17:29.580 align:middle line:90% Great. 00:17:29.580 --> 00:17:32.780 align:middle line:84% So this is basically a single firm. 00:17:32.780 --> 00:17:41.700 align:middle line:84% And there is always a baseline external option, competition, 00:17:41.700 --> 00:17:46.180 align:middle line:84% in the sense that when you are calculating demand, people-- 00:17:46.180 --> 00:17:49.500 align:middle line:84% the reason they may go to not buy from you 00:17:49.500 --> 00:17:51.820 align:middle line:84% is that they have some alternative out there. 00:17:51.820 --> 00:17:55.320 align:middle line:84% In that sense, it's not completely competition-free, 00:17:55.320 --> 00:17:57.440 align:middle line:84% but the competition is not responding to you. 00:17:57.440 --> 00:17:58.320 align:middle line:90% OK, good. 00:17:58.320 --> 00:18:00.480 align:middle line:84% So now I think I understand the scope. 00:18:00.480 --> 00:18:05.040 align:middle line:84% Let's now look at the structure of the model. 00:18:05.040 --> 00:18:08.820 align:middle line:84% So you have the prestige coming from the effect 00:18:08.820 --> 00:18:16.680 align:middle line:84% of accessibility, effect of price and quality. 00:18:16.680 --> 00:18:18.860 align:middle line:84% And for now we are just ignoring quality 00:18:18.860 --> 00:18:21.940 align:middle line:90% because that's fixed throughout. 00:18:21.940 --> 00:18:26.080 align:middle line:84% And you're using a multiplicative function 00:18:26.080 --> 00:18:29.600 align:middle line:90% for the price. 00:18:29.600 --> 00:18:32.480 align:middle line:84% And then another multiplicative, I 00:18:32.480 --> 00:18:36.400 align:middle line:84% assume-- oh, this one is a power function. 00:18:36.400 --> 00:18:45.680 align:middle line:84% So you're saying prestige goes up based on-- 00:18:45.680 --> 00:18:46.480 align:middle line:90% oops. 00:18:46.480 --> 00:18:47.660 align:middle line:90% This is more complex. 00:18:47.660 --> 00:18:51.560 align:middle line:90% So effective quality. 00:18:51.560 --> 00:18:59.360 align:middle line:84% Let's put that out, effective RP. 00:18:59.360 --> 00:19:01.780 align:middle line:90% RP is the relative price. 00:19:01.780 --> 00:19:07.000 align:middle line:84% So it is some fixed number plus relative price effect divided 00:19:07.000 --> 00:19:14.240 align:middle line:84% by 1 plus that fixed number plus relative price effect multiplied 00:19:14.240 --> 00:19:18.900 align:middle line:90% by effective accessibility. 00:19:18.900 --> 00:19:23.120 align:middle line:84% So this overall indicated prestige scales 00:19:23.120 --> 00:19:27.580 align:middle line:84% with accessibility effect directly. 00:19:27.580 --> 00:19:35.540 align:middle line:84% And then effect of price divided by 1 plus effect of price 00:19:35.540 --> 00:19:39.100 align:middle line:90% with some additional component. 00:19:39.100 --> 00:19:40.560 align:middle line:90% This is a little bit complex. 00:19:40.560 --> 00:19:44.140 align:middle line:84% Can you tell us how you were thinking about it, 00:19:44.140 --> 00:19:46.940 align:middle line:84% why you ended up with this formulation? 00:19:46.940 --> 00:19:51.900 align:middle line:84% AUDIENCE: I guess I wanted to find a function that starts-- 00:19:51.900 --> 00:19:55.340 align:middle line:84% that is bounded by 0 to 1, and that has a domain of 0 00:19:55.340 --> 00:19:56.140 align:middle line:90% to infinity. 00:19:56.140 --> 00:19:58.820 align:middle line:84% HAZHIR RAHMANDAD: So first, bounded by 0 and 1 00:19:58.820 --> 00:20:00.840 align:middle line:84% because that's how you want to define prestige. 00:20:00.840 --> 00:20:01.340 align:middle line:90% That's good. 00:20:01.340 --> 00:20:03.600 align:middle line:84% I think I like that practice in general. 00:20:03.600 --> 00:20:04.220 align:middle line:90% OK. 00:20:04.220 --> 00:20:04.845 align:middle line:90% AUDIENCE: Yeah. 00:20:04.845 --> 00:20:07.500 align:middle line:84% And then the I think the relative price [INAUDIBLE] 00:20:07.500 --> 00:20:11.580 align:middle line:84% with quality is just some positive quantity that 00:20:11.580 --> 00:20:12.342 align:middle line:90% can be bounded. 00:20:12.342 --> 00:20:13.300 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 00:20:13.300 --> 00:20:14.300 align:middle line:90% OK. 00:20:14.300 --> 00:20:18.460 align:middle line:84% So I would say, perhaps a multiplicative effect 00:20:18.460 --> 00:20:20.780 align:middle line:84% that has the three effects multiplied 00:20:20.780 --> 00:20:22.800 align:middle line:90% by each other or two-- 00:20:22.800 --> 00:20:24.920 align:middle line:90% putting aside the quality-- 00:20:24.920 --> 00:20:29.600 align:middle line:84% would have worked just fine because if you 00:20:29.600 --> 00:20:32.100 align:middle line:84% keep each of them between 0 and 1, 00:20:32.100 --> 00:20:34.600 align:middle line:90% the total will remain 0 and 1. 00:20:34.600 --> 00:20:37.560 align:middle line:84% So this non-linear transformation 00:20:37.560 --> 00:20:41.680 align:middle line:84% that you have under this equation, because it then 00:20:41.680 --> 00:20:43.520 align:middle line:84% gets combined with another nonlinear 00:20:43.520 --> 00:20:45.800 align:middle line:84% transformation in the previous step, 00:20:45.800 --> 00:20:47.780 align:middle line:90% becomes very hard to imagine. 00:20:47.780 --> 00:20:50.880 align:middle line:84% So when I look at this, I don't know 00:20:50.880 --> 00:20:57.160 align:middle line:84% how to mentally calculate what will happen and think 00:20:57.160 --> 00:20:58.340 align:middle line:90% about what will happen. 00:20:58.340 --> 00:21:02.780 align:middle line:84% So that transparency is really helpful for yourself 00:21:02.780 --> 00:21:05.880 align:middle line:84% when you get into diagnosing why the model behaves 00:21:05.880 --> 00:21:06.980 align:middle line:90% the way it does. 00:21:06.980 --> 00:21:10.000 align:middle line:84% So I would probably go for each of the effects 00:21:10.000 --> 00:21:14.440 align:middle line:84% being between 0 and 1, and then multiplying the three of them 00:21:14.440 --> 00:21:17.680 align:middle line:84% to get your indicated prestige, or even doing 00:21:17.680 --> 00:21:20.140 align:middle line:84% a weighted average, if you think multiplication 00:21:20.140 --> 00:21:24.060 align:middle line:84% is too strict because, the way multiplication works, 00:21:24.060 --> 00:21:28.500 align:middle line:84% it says, basically it reduces the probability of getting 00:21:28.500 --> 00:21:29.720 align:middle line:90% something close to 1. 00:21:29.720 --> 00:21:34.300 align:middle line:84% So if you have two numbers that are 0.5, you get 0.25. 00:21:34.300 --> 00:21:36.960 align:middle line:84% And if you have three numbers that are 0.5, 00:21:36.960 --> 00:21:41.420 align:middle line:90% now you get less than 12%. 00:21:41.420 --> 00:21:44.880 align:middle line:84% So it just brings down the mass of the distribution, 00:21:44.880 --> 00:21:46.320 align:middle line:90% each factor that comes in. 00:21:46.320 --> 00:21:48.940 align:middle line:84% So that may indicate something that 00:21:48.940 --> 00:21:51.820 align:middle line:84% is more average of the ones or weighted average 00:21:51.820 --> 00:21:55.020 align:middle line:84% of the ones based on each of them having an effect, maybe 00:21:55.020 --> 00:21:57.420 align:middle line:90% a better choice. 00:21:57.420 --> 00:21:58.240 align:middle line:90% And that's OK. 00:21:58.240 --> 00:22:01.060 align:middle line:84% That still keeps it between 0 and 1, as long as each of them 00:22:01.060 --> 00:22:02.700 align:middle line:90% are between 0 and 1. 00:22:02.700 --> 00:22:05.840 align:middle line:90% But putting that aside. 00:22:05.840 --> 00:22:10.760 align:middle line:84% Now, accessibility, assuming that it is between 0 and 1, 00:22:10.760 --> 00:22:15.360 align:middle line:84% then the effect of accessibility will remain between 0 and 1 00:22:15.360 --> 00:22:16.400 align:middle line:90% in this equation. 00:22:16.400 --> 00:22:19.360 align:middle line:84% And you have 1 minus it because the more accessible something 00:22:19.360 --> 00:22:22.740 align:middle line:90% is, the less prestigious it is. 00:22:22.740 --> 00:22:25.900 align:middle line:84% So that's the idea that you are trying to highlight here. 00:22:25.900 --> 00:22:28.080 align:middle line:84% And you have this power function. 00:22:28.080 --> 00:22:30.440 align:middle line:90% That's all fine, I think. 00:22:30.440 --> 00:22:33.480 align:middle line:84% But because accessibility, assuming that accessibility 00:22:33.480 --> 00:22:37.140 align:middle line:84% is between 0 and 1, then raised to a power, 00:22:37.140 --> 00:22:38.740 align:middle line:90% it will remain between 0 and 1. 00:22:38.740 --> 00:22:40.200 align:middle line:90% It does what you want to do. 00:22:40.200 --> 00:22:42.940 align:middle line:84% The effect of price the way you currently have, 00:22:42.940 --> 00:22:45.060 align:middle line:84% it wouldn't remain between 0 and 1. 00:22:45.060 --> 00:22:49.680 align:middle line:84% So that's the place that an alternative formulation 00:22:49.680 --> 00:22:54.240 align:middle line:84% between 0 and 1 would do a little bit better if you went 00:22:54.240 --> 00:22:55.742 align:middle line:90% with that indicated prestige. 00:22:55.742 --> 00:22:58.200 align:middle line:84% I think the reason you went with the current formulation is 00:22:58.200 --> 00:23:01.240 align:middle line:84% that your current effect of relative price 00:23:01.240 --> 00:23:03.320 align:middle line:84% was not bounded between 0 and 1, so you 00:23:03.320 --> 00:23:05.900 align:middle line:84% needed to bring it into the 0 and 1 range. 00:23:05.900 --> 00:23:08.220 align:middle line:84% So what you have done all makes sense. 00:23:08.220 --> 00:23:12.560 align:middle line:84% It's just that it's not the easiest 00:23:12.560 --> 00:23:17.410 align:middle line:84% to consume for yourself and for anybody who looks at them all. 00:23:17.410 --> 00:23:23.110 align:middle line:84% And then you have an asymmetric adjustment time, I assume. 00:23:23.110 --> 00:23:26.230 align:middle line:84% But otherwise this is a first order smooth. 00:23:26.230 --> 00:23:27.590 align:middle line:90% Good. 00:23:27.590 --> 00:23:30.670 align:middle line:84% So prestige is determined this way. 00:23:30.670 --> 00:23:35.310 align:middle line:90% Let's now look at accessibility. 00:23:35.310 --> 00:23:39.230 align:middle line:84% So you're tracking demand to sales. 00:23:39.230 --> 00:23:48.150 align:middle line:84% And you're saying, if sales was close to demand then 00:23:48.150 --> 00:23:52.070 align:middle line:84% accessibility was close to 1 and therefore 00:23:52.070 --> 00:23:54.470 align:middle line:90% this is not very prestigious. 00:23:54.470 --> 00:24:00.190 align:middle line:84% If sales was much lower than demand, then 00:24:00.190 --> 00:24:01.830 align:middle line:90% that's creating prestige. 00:24:01.830 --> 00:24:03.230 align:middle line:90% I think that's good. 00:24:03.230 --> 00:24:05.630 align:middle line:84% And you don't need any non-linearity in this part 00:24:05.630 --> 00:24:08.310 align:middle line:84% because you will have a non-linearity later 00:24:08.310 --> 00:24:09.330 align:middle line:90% on the right-hand side. 00:24:09.330 --> 00:24:11.270 align:middle line:84% So it is good to keep it clean and well 00:24:11.270 --> 00:24:14.763 align:middle line:84% defined so that when you see a value of 0.5 00:24:14.763 --> 00:24:17.490 align:middle line:84% you know it means half of the demand 00:24:17.490 --> 00:24:20.810 align:middle line:84% was served on average in the recent times. 00:24:20.810 --> 00:24:27.130 align:middle line:84% Now, it is important to use the same time constant here 00:24:27.130 --> 00:24:28.150 align:middle line:90% for this averaging. 00:24:28.150 --> 00:24:30.450 align:middle line:84% I mean, it's not critical, but it is a good idea 00:24:30.450 --> 00:24:36.530 align:middle line:84% because you're basically saying, sales to demand 00:24:36.530 --> 00:24:38.210 align:middle line:90% is getting averaged. 00:24:38.210 --> 00:24:41.490 align:middle line:84% Now, you didn't need necessarily to create 00:24:41.490 --> 00:24:45.510 align:middle line:84% these two stocks to track the indicated accessibility. 00:24:45.510 --> 00:24:50.850 align:middle line:84% You could just calculate current sales divided by current demand 00:24:50.850 --> 00:24:54.570 align:middle line:84% and then take a smooth of that function. 00:24:54.570 --> 00:24:58.710 align:middle line:84% So it is the current accessibility, 00:24:58.710 --> 00:25:01.370 align:middle line:84% and then you're taking an average over the season 00:25:01.370 --> 00:25:02.430 align:middle line:90% duration. 00:25:02.430 --> 00:25:04.650 align:middle line:84% So that will have saved you a stock 00:25:04.650 --> 00:25:07.410 align:middle line:84% and a little bit of a structure that 00:25:07.410 --> 00:25:11.970 align:middle line:90% may look unfamiliar to people. 00:25:11.970 --> 00:25:14.190 align:middle line:84% I don't think you're using any of these two 00:25:14.190 --> 00:25:15.610 align:middle line:90% stocks in any other place. 00:25:15.610 --> 00:25:17.090 align:middle line:84% If you are, then you'd need them, 00:25:17.090 --> 00:25:20.030 align:middle line:84% but you're not using anywhere else. 00:25:20.030 --> 00:25:26.110 align:middle line:84% So that would have been an easier, cleaner formulation. 00:25:26.110 --> 00:25:33.770 align:middle line:84% So sales for action price sensitivity, OK. 00:25:33.770 --> 00:25:38.990 align:middle line:84% So now sales recorded has this thing that 00:25:38.990 --> 00:25:44.070 align:middle line:84% is sales multiplied by fraction price-sensitive demand. 00:25:44.070 --> 00:25:48.790 align:middle line:84% So this is-- what's the story here? 00:25:48.790 --> 00:25:53.690 align:middle line:84% AUDIENCE: I wanted to see, of the demand, 00:25:53.690 --> 00:25:58.075 align:middle line:84% how much went to the price-sensitive customers. 00:25:58.075 --> 00:25:58.950 align:middle line:90% HAZHIR RAHMANDAD: OK. 00:25:58.950 --> 00:26:02.350 align:middle line:84% So then the sales that you're tracking here 00:26:02.350 --> 00:26:05.150 align:middle line:84% is only the fraction of sales that is going 00:26:05.150 --> 00:26:07.710 align:middle line:90% to price-sensitive customers. 00:26:07.710 --> 00:26:09.967 align:middle line:90% AUDIENCE: Yeah. 00:26:09.967 --> 00:26:12.050 align:middle line:84% HAZHIR RAHMANDAD: Well, because you're dividing it 00:26:12.050 --> 00:26:16.230 align:middle line:84% by total demand, not by demand from price-sensitive customers. 00:26:16.230 --> 00:26:16.770 align:middle line:90% Correct? 00:26:16.770 --> 00:26:16.930 align:middle line:90% AUDIENCE: Yeah. 00:26:16.930 --> 00:26:17.210 align:middle line:90% What's 00:26:17.210 --> 00:26:17.750 align:middle line:84% HAZHIR RAHMANDAD: So what's the story? 00:26:17.750 --> 00:26:19.390 align:middle line:90% Why are you doing it that way? 00:26:19.390 --> 00:26:22.010 align:middle line:90% AUDIENCE: It's actually a model. 00:26:22.010 --> 00:26:26.210 align:middle line:84% I think it is a multiplication of two [INAUDIBLE] 00:26:26.210 --> 00:26:33.970 align:middle line:84% First was the quantity sold to the sensitive segment 00:26:33.970 --> 00:26:38.890 align:middle line:84% over the quantity sold times quantity sold over demand. 00:26:38.890 --> 00:26:44.370 align:middle line:84% So how much of excess demand there 00:26:44.370 --> 00:26:46.450 align:middle line:84% were and other demand, how much went 00:26:46.450 --> 00:26:48.313 align:middle line:84% to the price-sensitive customers. 00:26:48.313 --> 00:26:49.230 align:middle line:90% HAZHIR RAHMANDAD: Yes. 00:26:49.230 --> 00:26:51.630 align:middle line:84% But why do you want that concept? 00:26:51.630 --> 00:26:54.730 align:middle line:84% Why is that driving accessibility? 00:26:54.730 --> 00:26:58.210 align:middle line:84% AUDIENCE: I want it to be always in excess demand, 00:26:58.210 --> 00:27:01.650 align:middle line:84% so that you have more prestige than of the things that 00:27:01.650 --> 00:27:06.473 align:middle line:84% are sold, that they don't go to the price-sensitive customers. 00:27:06.473 --> 00:27:07.890 align:middle line:84% HAZHIR RAHMANDAD: I see, so you're 00:27:07.890 --> 00:27:17.310 align:middle line:84% saying a prestigious product could have full sales to-- 00:27:17.310 --> 00:27:20.470 align:middle line:84% I mean, supply and demand can match, that is, 00:27:20.470 --> 00:27:23.270 align:middle line:84% sales could be the same as demand 00:27:23.270 --> 00:27:32.590 align:middle line:84% if it is all of it going to folks who are kind of snobs. 00:27:32.590 --> 00:27:40.150 align:middle line:84% So that wouldn't reduce the prestige factor for the product. 00:27:40.150 --> 00:27:42.630 align:middle line:84% What does reduce the prestige factor 00:27:42.630 --> 00:27:47.030 align:middle line:84% is how much of the price-sensitive customers 00:27:47.030 --> 00:27:50.150 align:middle line:90% are going to get the product. 00:27:50.150 --> 00:27:55.830 align:middle line:90% OK, so if that's the argument. 00:27:55.830 --> 00:28:00.790 align:middle line:84% Then I'm following what you're doing here. 00:28:00.790 --> 00:28:04.390 align:middle line:84% So it's the fraction of total demand 00:28:04.390 --> 00:28:12.730 align:middle line:84% that is sold to price-sensitive customers. 00:28:12.730 --> 00:28:13.350 align:middle line:90% That's good. 00:28:13.350 --> 00:28:15.030 align:middle line:90% OK, that makes sense. 00:28:15.030 --> 00:28:17.930 align:middle line:90% 00:28:17.930 --> 00:28:21.250 align:middle line:84% Let's now come down to the demand side. 00:28:21.250 --> 00:28:26.190 align:middle line:84% So you have effect of these different factors on demand. 00:28:26.190 --> 00:28:30.370 align:middle line:90% So there is a baseline demand. 00:28:30.370 --> 00:28:34.930 align:middle line:90% There is effect of-- 00:28:34.930 --> 00:28:36.625 align:middle line:90% what is trend? 00:28:36.625 --> 00:28:39.250 align:middle line:84% AUDIENCE: I just wanted to make it some exogenous trend, like-- 00:28:39.250 --> 00:28:40.292 align:middle line:90% HAZHIR RAHMANDAD: Oh, OK. 00:28:40.292 --> 00:28:42.530 align:middle line:90% So this demand is exogenous? 00:28:42.530 --> 00:28:44.830 align:middle line:84% AUDIENCE: No, just the trend part is [INAUDIBLE], 00:28:44.830 --> 00:28:45.975 align:middle line:90% but it's 6 to 1 right now. 00:28:45.975 --> 00:28:46.850 align:middle line:90% HAZHIR RAHMANDAD: OK. 00:28:46.850 --> 00:28:47.350 align:middle line:90% Yeah. 00:28:47.350 --> 00:28:48.290 align:middle line:90% Got it. 00:28:48.290 --> 00:28:50.370 align:middle line:90% OK. 00:28:50.370 --> 00:28:56.490 align:middle line:84% So then effective prestige and relative price 00:28:56.490 --> 00:29:03.690 align:middle line:90% are the two primary drivers. 00:29:03.690 --> 00:29:10.350 align:middle line:90% And are you creating-- 00:29:10.350 --> 00:29:11.890 align:middle line:90% yeah, this is soft-scripted. 00:29:11.890 --> 00:29:14.470 align:middle line:90% 00:29:14.470 --> 00:29:16.390 align:middle line:90% Is it soft-scripted? 00:29:16.390 --> 00:29:19.190 align:middle line:84% Yeah, it is on different types of demands. 00:29:19.190 --> 00:29:23.838 align:middle line:84% And these people have different-- 00:29:23.838 --> 00:29:25.630 align:middle line:84% AUDIENCE: They have different elasticities. 00:29:25.630 --> 00:29:28.010 align:middle line:84% HAZHIR RAHMANDAD: --elasticities for the price. 00:29:28.010 --> 00:29:35.750 align:middle line:84% So the E sensitivity to price for first group 00:29:35.750 --> 00:29:40.950 align:middle line:84% is more sensitive to price than the second group. 00:29:40.950 --> 00:29:53.310 align:middle line:84% And then the prestige effect is first group-- 00:29:53.310 --> 00:29:57.190 align:middle line:84% sorry, first group is who, are the normal people or is it 00:29:57.190 --> 00:29:58.230 align:middle line:90% snobs? 00:29:58.230 --> 00:30:02.113 align:middle line:84% AUDIENCE: I think the snobs are more sensitive to the prestige. 00:30:02.113 --> 00:30:03.030 align:middle line:90% HAZHIR RAHMANDAD: Yes. 00:30:03.030 --> 00:30:04.610 align:middle line:90% AUDIENCE: Very sensitive parts. 00:30:04.610 --> 00:30:05.527 align:middle line:90% HAZHIR RAHMANDAD: Yes. 00:30:05.527 --> 00:30:06.610 align:middle line:90% So this 0 point-- 00:30:06.610 --> 00:30:09.410 align:middle line:84% AUDIENCE: There is sensitivity to prestige. 00:30:09.410 --> 00:30:11.970 align:middle line:84% HAZHIR RAHMANDAD: So this is sensitivity to prestige 00:30:11.970 --> 00:30:13.930 align:middle line:84% or no, this is price sensitivity. 00:30:13.930 --> 00:30:15.970 align:middle line:84% AUDIENCE: I made the elasticity to change 00:30:15.970 --> 00:30:17.890 align:middle line:90% with the prestige [INAUDIBLE]. 00:30:17.890 --> 00:30:21.290 align:middle line:90% 00:30:21.290 --> 00:30:22.790 align:middle line:84% HAZHIR RAHMANDAD: I'm not following. 00:30:22.790 --> 00:30:24.890 align:middle line:90% Can you explain? 00:30:24.890 --> 00:30:28.607 align:middle line:84% AUDIENCE: So prestige, I accounted for it twice. 00:30:28.607 --> 00:30:30.190 align:middle line:84% HAZHIR RAHMANDAD: I see, I see, I see. 00:30:30.190 --> 00:30:34.330 align:middle line:84% So here you have baseline price multiplied 00:30:34.330 --> 00:30:40.210 align:middle line:84% by 1 minus sensitivity to prestige 00:30:40.210 --> 00:30:44.450 align:middle line:90% multiplied by prestige itself. 00:30:44.450 --> 00:30:47.930 align:middle line:84% So tell us, what was your thinking 00:30:47.930 --> 00:30:49.170 align:middle line:90% to create this equation? 00:30:49.170 --> 00:30:52.170 align:middle line:84% AUDIENCE: On the demand side, some people 00:30:52.170 --> 00:30:55.170 align:middle line:84% have higher elasticity, lower elasticity. 00:30:55.170 --> 00:30:57.850 align:middle line:84% And given that demand we're going 00:30:57.850 --> 00:31:01.130 align:middle line:84% to have different elasticity depending on the characteristic 00:31:01.130 --> 00:31:01.850 align:middle line:90% of [INAUDIBLE]. 00:31:01.850 --> 00:31:03.392 align:middle line:84% If it is a luxury brand, you're going 00:31:03.392 --> 00:31:08.310 align:middle line:90% to have more elastic demand-- 00:31:08.310 --> 00:31:09.450 align:middle line:90% inelastic demand. 00:31:09.450 --> 00:31:13.290 align:middle line:84% And then, yeah, there's that two dimension. 00:31:13.290 --> 00:31:14.290 align:middle line:90% HAZHIR RAHMANDAD: I see. 00:31:14.290 --> 00:31:16.270 align:middle line:84% So basically, you're saying the snobs 00:31:16.270 --> 00:31:20.230 align:middle line:84% are only snobs for products that are prestigious. 00:31:20.230 --> 00:31:27.830 align:middle line:84% If this product loses its prestigeness then 00:31:27.830 --> 00:31:29.010 align:middle line:90% they wouldn't be snobs. 00:31:29.010 --> 00:31:30.890 align:middle line:84% They would be price-sensitive about it. 00:31:30.890 --> 00:31:31.750 align:middle line:90% Got it. 00:31:31.750 --> 00:31:35.090 align:middle line:84% That's pretty thoughtful, interesting idea. 00:31:35.090 --> 00:31:36.950 align:middle line:84% And yeah, I think the way you have 00:31:36.950 --> 00:31:39.510 align:middle line:90% captured it does make sense. 00:31:39.510 --> 00:31:46.470 align:middle line:84% Now, given that prestige changes between 0 and 1, a prestige of 1 00:31:46.470 --> 00:31:48.590 align:middle line:90% in this case-- 00:31:48.590 --> 00:31:51.610 align:middle line:90% now it is 1 minus this factor. 00:31:51.610 --> 00:31:55.790 align:middle line:84% So the 0.8 in the other one now turns into 0.2. 00:31:55.790 --> 00:32:00.010 align:middle line:84% So the parameter names could be better specified 00:32:00.010 --> 00:32:02.850 align:middle line:84% because this sensitivity is actually not the sensitivity, 00:32:02.850 --> 00:32:05.410 align:middle line:84% it is 1 minus maximum sensitivity or something 00:32:05.410 --> 00:32:07.250 align:middle line:90% like that. 00:32:07.250 --> 00:32:08.330 align:middle line:90% OK. 00:32:08.330 --> 00:32:09.870 align:middle line:90% All of that makes sense. 00:32:09.870 --> 00:32:14.570 align:middle line:84% And then you have the price effect and the prestige effect 00:32:14.570 --> 00:32:15.190 align:middle line:90% separately. 00:32:15.190 --> 00:32:20.170 align:middle line:84% And the prestige has lower impact for the normal people. 00:32:20.170 --> 00:32:22.850 align:middle line:90% And these effects are-- 00:32:22.850 --> 00:32:24.210 align:middle line:90% oh. 00:32:24.210 --> 00:32:27.290 align:middle line:84% So here you're using a linear effect 00:32:27.290 --> 00:32:34.770 align:middle line:84% for the effective prestige on demand. 00:32:34.770 --> 00:32:42.250 align:middle line:84% So I think in system dynamics, community people wouldn't really 00:32:42.250 --> 00:32:45.250 align:middle line:84% have a strong opinion on this formulation. 00:32:45.250 --> 00:32:50.690 align:middle line:84% But if you talk with more econ-oriented folks, 00:32:50.690 --> 00:32:55.490 align:middle line:84% this linear effect is perhaps not as attractive 00:32:55.490 --> 00:33:01.480 align:middle line:84% as a power function because power functions directly 00:33:01.480 --> 00:33:05.520 align:middle line:90% translate into elasticities. 00:33:05.520 --> 00:33:07.640 align:middle line:84% So people have an intuition for them. 00:33:07.640 --> 00:33:09.100 align:middle line:90% Now linear is also simple. 00:33:09.100 --> 00:33:12.600 align:middle line:84% So you can build an intuition for what it does. 00:33:12.600 --> 00:33:20.920 align:middle line:84% Just a little bit of a consideration for alternatives. 00:33:20.920 --> 00:33:27.960 align:middle line:84% For price you do have the elasticity as defined. 00:33:27.960 --> 00:33:28.480 align:middle line:90% Interesting. 00:33:28.480 --> 00:33:31.040 align:middle line:84% And then, because you're using effective PR 00:33:31.040 --> 00:33:35.520 align:middle line:84% on E but with the negative sign here, 00:33:35.520 --> 00:33:38.240 align:middle line:84% so the larger it is the more-- the faster 00:33:38.240 --> 00:33:40.540 align:middle line:90% they reduce their demand. 00:33:40.540 --> 00:33:43.780 align:middle line:90% That makes sense. 00:33:43.780 --> 00:33:47.440 align:middle line:84% All of that is sensible, I think. 00:33:47.440 --> 00:33:50.180 align:middle line:84% So then what's the analysis you're doing here? 00:33:50.180 --> 00:33:57.980 align:middle line:84% You're looking at different policy choices of the firm 00:33:57.980 --> 00:34:01.340 align:middle line:90% to see how they maximize-- 00:34:01.340 --> 00:34:02.142 align:middle line:90% AUDIENCE: Yeah. 00:34:02.142 --> 00:34:03.100 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 00:34:03.100 --> 00:34:10.820 align:middle line:84% So this is their price, how it changes the demand and-- 00:34:10.820 --> 00:34:13.940 align:middle line:84% AUDIENCE: I feel like for now you can't really-- 00:34:13.940 --> 00:34:17.679 align:middle line:84% there's no feedback to price and quantity. 00:34:17.679 --> 00:34:19.580 align:middle line:90% HAZHIR RAHMANDAD: Uh-huh. 00:34:19.580 --> 00:34:23.739 align:middle line:84% AUDIENCE: I was thinking, shouldn't strategy 00:34:23.739 --> 00:34:25.858 align:middle line:84% be a function of the state [INAUDIBLE]? 00:34:25.858 --> 00:34:27.400 align:middle line:84% HAZHIR RAHMANDAD: Very good question. 00:34:27.400 --> 00:34:28.380 align:middle line:90% So this is kind of-- 00:34:28.380 --> 00:34:28.880 align:middle line:90% OK. 00:34:28.880 --> 00:34:32.100 align:middle line:84% So for now, let's say this is your model that 00:34:32.100 --> 00:34:35.739 align:middle line:84% is for the core engine of generation of profits. 00:34:35.739 --> 00:34:37.239 align:middle line:90% We settle with this. 00:34:37.239 --> 00:34:41.273 align:middle line:84% We could talk about gaming and having a competition and so on. 00:34:41.273 --> 00:34:42.440 align:middle line:90% But I think that's too much. 00:34:42.440 --> 00:34:44.860 align:middle line:84% I think, given that you have one more week that you 00:34:44.860 --> 00:34:46.500 align:middle line:84% can work on this, if you want, you 00:34:46.500 --> 00:34:49.940 align:middle line:84% can go in a direction that is aligned 00:34:49.940 --> 00:34:54.300 align:middle line:84% with your interest in what is the optimal solution. 00:34:54.300 --> 00:34:58.800 align:middle line:84% So this is, in a dynamic system, you have, let's say, 00:34:58.800 --> 00:35:00.060 align:middle line:90% two policy levers. 00:35:00.060 --> 00:35:08.760 align:middle line:84% There is the price and there is the supply, how much you sell 00:35:08.760 --> 00:35:10.880 align:middle line:90% and what price you sell. 00:35:10.880 --> 00:35:14.280 align:middle line:84% So these two you could think of as fixed numbers 00:35:14.280 --> 00:35:18.680 align:middle line:84% for the whole simulation and you get a response 00:35:18.680 --> 00:35:20.600 align:middle line:90% in terms of profitability. 00:35:20.600 --> 00:35:22.920 align:middle line:84% That's the easiest way to start thinking 00:35:22.920 --> 00:35:26.800 align:middle line:90% about optimal solutions. 00:35:26.800 --> 00:35:28.340 align:middle line:90% And that is OK. 00:35:28.340 --> 00:35:31.160 align:middle line:90% That's a good starting point. 00:35:31.160 --> 00:35:32.900 align:middle line:90% You can do a quick analysis. 00:35:32.900 --> 00:35:35.240 align:middle line:84% You can create a 3D graph that has, 00:35:35.240 --> 00:35:40.420 align:middle line:84% for each value of price and each value of supply, 00:35:40.420 --> 00:35:43.220 align:middle line:84% what is your net profits in this market. 00:35:43.220 --> 00:35:45.680 align:middle line:84% And you can see a surface plot or a contour 00:35:45.680 --> 00:35:47.540 align:middle line:90% plot that summarizes that. 00:35:47.540 --> 00:35:50.300 align:middle line:84% And that becomes the output of the analysis. 00:35:50.300 --> 00:35:55.300 align:middle line:84% So that's a first order take on the optimal question. 00:35:55.300 --> 00:35:59.680 align:middle line:84% Now you can bring a rebuttal to that, 00:35:59.680 --> 00:36:02.040 align:middle line:84% or a kind of a complication to that. 00:36:02.040 --> 00:36:04.060 align:middle line:84% And that is what Ross was saying, that, 00:36:04.060 --> 00:36:07.120 align:middle line:90% oh, these things are dynamic. 00:36:07.120 --> 00:36:08.960 align:middle line:84% I could change the price over time. 00:36:08.960 --> 00:36:11.180 align:middle line:84% And I could change the supply over time 00:36:11.180 --> 00:36:17.940 align:middle line:84% to get myself into the right range for what 00:36:17.940 --> 00:36:20.260 align:middle line:90% is a better long term. 00:36:20.260 --> 00:36:24.440 align:middle line:84% So if I have 100 months total in this market, 00:36:24.440 --> 00:36:27.560 align:middle line:84% I can actually do much better than just having a fixed number. 00:36:27.560 --> 00:36:28.920 align:middle line:90% And that's absolutely correct. 00:36:28.920 --> 00:36:32.780 align:middle line:84% That's the more sophisticated version of the optimization. 00:36:32.780 --> 00:36:35.660 align:middle line:84% So that more sophisticated version, solving it 00:36:35.660 --> 00:36:37.280 align:middle line:90% becomes a harder problem. 00:36:37.280 --> 00:36:40.100 align:middle line:84% So it now becomes a dynamic programming problem, 00:36:40.100 --> 00:36:42.380 align:middle line:84% or a reinforcement learning problem, 00:36:42.380 --> 00:36:44.680 align:middle line:84% or some stochastic optimization problem, 00:36:44.680 --> 00:36:48.060 align:middle line:84% depending on what kind of a tool you want to solve that. 00:36:48.060 --> 00:36:50.400 align:middle line:84% Now, system dynamics doesn't have 00:36:50.400 --> 00:36:55.100 align:middle line:84% that toolbox as its center of gravity or attention. 00:36:55.100 --> 00:36:59.100 align:middle line:84% But basically any dynamic model can be optimally controlled. 00:36:59.100 --> 00:37:02.000 align:middle line:84% So you can think of it as an optimal control problem. 00:37:02.000 --> 00:37:05.560 align:middle line:84% So which of these tools you want to use 00:37:05.560 --> 00:37:07.960 align:middle line:84% to solve that problem partly depends 00:37:07.960 --> 00:37:09.840 align:middle line:84% on what you are comfortable with, what 00:37:09.840 --> 00:37:15.720 align:middle line:84% you're familiar with, partly on how complex the problem is. 00:37:15.720 --> 00:37:20.000 align:middle line:84% In the general space that we are talking about, 00:37:20.000 --> 00:37:25.920 align:middle line:84% these problems can be simplified into a mapping from the status 00:37:25.920 --> 00:37:27.160 align:middle line:90% space-- 00:37:27.160 --> 00:37:29.360 align:middle line:84% and we should define this status space-- 00:37:29.360 --> 00:37:30.900 align:middle line:90% into the action space. 00:37:30.900 --> 00:37:32.760 align:middle line:84% So the action space in Ross's problem 00:37:32.760 --> 00:37:36.160 align:middle line:84% is a two-dimensional set of, OK, what is the price 00:37:36.160 --> 00:37:42.560 align:middle line:84% and what is the supply production. 00:37:42.560 --> 00:37:46.400 align:middle line:84% Those two, at any point in time, the firm can change. 00:37:46.400 --> 00:37:49.900 align:middle line:84% And the status space is basically 00:37:49.900 --> 00:37:53.340 align:middle line:84% what are the stock values, the values of different stocks 00:37:53.340 --> 00:37:57.190 align:middle line:84% in the system that you can look at 00:37:57.190 --> 00:37:58.980 align:middle line:90% or the important stock values. 00:37:58.980 --> 00:38:01.420 align:middle line:90% Not everything is important. 00:38:01.420 --> 00:38:07.100 align:middle line:84% So creating that mapping optimally 00:38:07.100 --> 00:38:12.820 align:middle line:84% becomes an optimal control dynamic program problem 00:38:12.820 --> 00:38:15.500 align:middle line:90% that you can try to solve. 00:38:15.500 --> 00:38:18.580 align:middle line:84% Even for a simple model like this, 00:38:18.580 --> 00:38:22.340 align:middle line:84% it probably is not feasible to solve anything analytically. 00:38:22.340 --> 00:38:25.800 align:middle line:84% And then the numerical solutions are also non-trivial. 00:38:25.800 --> 00:38:29.060 align:middle line:84% That is, if you have a continuous state space 00:38:29.060 --> 00:38:32.100 align:middle line:84% and continuous action space, this 00:38:32.100 --> 00:38:36.420 align:middle line:84% is a computationally intractable problem, 00:38:36.420 --> 00:38:40.100 align:middle line:84% dynamic programming typically creates grids 00:38:40.100 --> 00:38:42.200 align:middle line:84% on this status space and says, for example, 00:38:42.200 --> 00:38:45.780 align:middle line:84% if you're at the low levels of current prestige 00:38:45.780 --> 00:38:49.200 align:middle line:90% and low levels of current-- 00:38:49.200 --> 00:38:52.680 align:middle line:84% and early time, because time is one of those state variables 00:38:52.680 --> 00:38:55.280 align:middle line:90% that is changing over time-- 00:38:55.280 --> 00:38:57.940 align:middle line:84% then maybe you should price it very high 00:38:57.940 --> 00:39:00.260 align:middle line:84% and then later bring it down or something like that. 00:39:00.260 --> 00:39:04.320 align:middle line:84% So that's one way of solving it, a dynamic program. 00:39:04.320 --> 00:39:08.760 align:middle line:84% Then there is this whole neural network-based methods 00:39:08.760 --> 00:39:11.860 align:middle line:84% for approximating that functional form. 00:39:11.860 --> 00:39:15.980 align:middle line:84% So basically we are looking for a function from a status space, 00:39:15.980 --> 00:39:17.880 align:middle line:84% let's say, a two-dimensional status space 00:39:17.880 --> 00:39:20.280 align:middle line:84% into a two-dimensional action space. 00:39:20.280 --> 00:39:24.000 align:middle line:84% So all the functions that can create 00:39:24.000 --> 00:39:26.840 align:middle line:84% from a two-dimensional space to a two-dimensional space 00:39:26.840 --> 00:39:28.200 align:middle line:90% are candidates. 00:39:28.200 --> 00:39:31.720 align:middle line:84% And now you're optimizing over that functional space. 00:39:31.720 --> 00:39:34.680 align:middle line:84% And doing that using neural networks 00:39:34.680 --> 00:39:37.780 align:middle line:90% is now the standard method. 00:39:37.780 --> 00:39:42.660 align:middle line:84% Basically, if you go, you want a scalable, high-quality methods, 00:39:42.660 --> 00:39:45.220 align:middle line:84% people will say, go reinforcement learning using 00:39:45.220 --> 00:39:52.020 align:middle line:84% neural nets as your approximators for the value 00:39:52.020 --> 00:39:54.780 align:middle line:84% function, in the lingo of dynamic programming 00:39:54.780 --> 00:39:59.580 align:middle line:84% or reinforcement learning, and that value function-- 00:39:59.580 --> 00:40:01.940 align:middle line:84% but all of it is pretty involved. 00:40:01.940 --> 00:40:04.220 align:middle line:84% That is, in the SD community, there 00:40:04.220 --> 00:40:06.620 align:middle line:84% are very few people who actually have done this 00:40:06.620 --> 00:40:09.300 align:middle line:84% and have built models to connect your Vensim 00:40:09.300 --> 00:40:11.680 align:middle line:84% model to the machinery for doing this. 00:40:11.680 --> 00:40:14.940 align:middle line:84% And even on the reinforcement learning side, 00:40:14.940 --> 00:40:16.800 align:middle line:84% this is not a fully solved problem. 00:40:16.800 --> 00:40:19.660 align:middle line:84% For every problem you need to actually create the machinery 00:40:19.660 --> 00:40:20.440 align:middle line:90% from scratch. 00:40:20.440 --> 00:40:24.780 align:middle line:84% It's not a black box, I throw it in and get the results. 00:40:24.780 --> 00:40:29.780 align:middle line:84% Then there are simpler hacks in between that would say, 00:40:29.780 --> 00:40:33.060 align:middle line:84% OK, I understand this is a complex problem, 00:40:33.060 --> 00:40:36.780 align:middle line:84% but based on my intuition for this system, maybe 00:40:36.780 --> 00:40:42.180 align:middle line:84% a linear function that says, the output, 00:40:42.180 --> 00:40:48.140 align:middle line:84% these actions is a linear function of my state variables, 00:40:48.140 --> 00:40:50.080 align:middle line:84% and I'll estimate those parameters 00:40:50.080 --> 00:40:53.240 align:middle line:84% of that linear function using an optimization. 00:40:53.240 --> 00:40:55.000 align:middle line:90% That can do well. 00:40:55.000 --> 00:40:56.880 align:middle line:90% And maybe I use that. 00:40:56.880 --> 00:40:58.600 align:middle line:84% So that you can do inside Vensim, 00:40:58.600 --> 00:41:02.200 align:middle line:84% basically come up with a flexible, functional form 00:41:02.200 --> 00:41:05.240 align:middle line:84% that says your price is a function 00:41:05.240 --> 00:41:11.660 align:middle line:84% of your couple of variables that are important here. 00:41:11.660 --> 00:41:15.000 align:middle line:84% In this case, I think the most important is the prestige 00:41:15.000 --> 00:41:18.020 align:middle line:90% and perhaps the-- and time. 00:41:18.020 --> 00:41:20.900 align:middle line:84% I think those two are the critical ones. 00:41:20.900 --> 00:41:24.120 align:middle line:84% So you can create a linear or a quadratic function 00:41:24.120 --> 00:41:27.260 align:middle line:84% with a few handful of free parameters, 00:41:27.260 --> 00:41:29.720 align:middle line:84% then use Vensim's optimization engine 00:41:29.720 --> 00:41:32.780 align:middle line:84% to maximize your profits by changing those parameters. 00:41:32.780 --> 00:41:35.380 align:middle line:84% It gives you a decent first-order approximation. 00:41:35.380 --> 00:41:38.700 align:middle line:90% So that's another way to do it. 00:41:38.700 --> 00:41:42.220 align:middle line:84% So that, I went through this because we have never 00:41:42.220 --> 00:41:46.380 align:middle line:84% had the opportunity to talk about optimal control 00:41:46.380 --> 00:41:47.400 align:middle line:90% type of things. 00:41:47.400 --> 00:41:49.720 align:middle line:84% But this is a simple enough problem. 00:41:49.720 --> 00:41:54.180 align:middle line:84% You can imagine doing it in this case. 00:41:54.180 --> 00:41:55.540 align:middle line:90% Now for you, Ross. 00:41:55.540 --> 00:41:57.820 align:middle line:84% It really depends on what you want to get out 00:41:57.820 --> 00:41:59.080 align:middle line:90% of it for the next week. 00:41:59.080 --> 00:42:02.540 align:middle line:84% I would say doing one of these spectrum of things 00:42:02.540 --> 00:42:06.220 align:middle line:84% from just a two-dimensional space where you're 00:42:06.220 --> 00:42:15.220 align:middle line:84% running different combinations of price and supply 00:42:15.220 --> 00:42:17.000 align:middle line:90% and you calculate the profit. 00:42:17.000 --> 00:42:19.980 align:middle line:84% So it is a fixed policy, you create that contour plot, 00:42:19.980 --> 00:42:24.780 align:middle line:84% that's a one level, all the way to more complex 00:42:24.780 --> 00:42:26.120 align:middle line:90% dynamic programming ones. 00:42:26.120 --> 00:42:27.740 align:middle line:84% But I think that's too much unless you 00:42:27.740 --> 00:42:31.000 align:middle line:84% are planning to do something more with that space in general. 00:42:31.000 --> 00:42:33.580 align:middle line:84% So this is an opportunity to learn that. 00:42:33.580 --> 00:42:37.320 align:middle line:84% The intermediate one would be running something inside Vensim 00:42:37.320 --> 00:42:40.530 align:middle line:84% that is a simple function connecting 00:42:40.530 --> 00:42:44.250 align:middle line:84% your inputs to your outputs, your status space to your action 00:42:44.250 --> 00:42:46.850 align:middle line:84% space, and optimizing those parameters using 00:42:46.850 --> 00:42:49.990 align:middle line:90% Vensim's optimization engine. 00:42:49.990 --> 00:42:54.570 align:middle line:84% I would say that is probably as far as you should push yourself 00:42:54.570 --> 00:42:56.650 align:middle line:84% if you're not building a capability 00:42:56.650 --> 00:42:59.170 align:middle line:90% longer-term from this exercise. 00:42:59.170 --> 00:43:00.110 align:middle line:90% OK, good. 00:43:00.110 --> 00:43:04.010 align:middle line:84% I think this is a nice model for what you're trying to do. 00:43:04.010 --> 00:43:05.990 align:middle line:84% We discussed a few things that you can improve, 00:43:05.990 --> 00:43:10.090 align:middle line:84% but I think it's in the right space. 00:43:10.090 --> 00:43:16.090 align:middle line:84% Let's next look at Cathy's model. 00:43:16.090 --> 00:43:20.150 align:middle line:84% AUDIENCE: And I'll say, it's not 100% formulated. 00:43:20.150 --> 00:43:21.012 align:middle line:90% This is structure. 00:43:21.012 --> 00:43:21.970 align:middle line:90% HAZHIR RAHMANDAD: Sure. 00:43:21.970 --> 00:43:28.850 align:middle line:84% So maybe tell us a little bit about what 00:43:28.850 --> 00:43:31.750 align:middle line:84% you're trying to capture, how you're trying to capture it. 00:43:31.750 --> 00:43:34.410 align:middle line:84% And what are the questions that we can focus on, 00:43:34.410 --> 00:43:37.130 align:middle line:84% the places where it adds more value for you. 00:43:37.130 --> 00:43:40.310 align:middle line:84% AUDIENCE: Yeah, so I am trying to model 00:43:40.310 --> 00:43:43.190 align:middle line:84% the trade-off between social and human capital 00:43:43.190 --> 00:43:47.590 align:middle line:84% and also the heterogeneity among readers, 00:43:47.590 --> 00:43:50.750 align:middle line:84% and to see if I can incorporate both of those 00:43:50.750 --> 00:43:57.030 align:middle line:84% and explain the divergence between buying and reading 00:43:57.030 --> 00:43:59.830 align:middle line:90% books rates. 00:43:59.830 --> 00:44:05.310 align:middle line:84% And so I started with this social influence feedback loop 00:44:05.310 --> 00:44:09.250 align:middle line:84% on the right here, where books become more attractive, 00:44:09.250 --> 00:44:13.290 align:middle line:84% people are buying more, and then because they're buying more, 00:44:13.290 --> 00:44:15.670 align:middle line:84% they're kind of showing them off, 00:44:15.670 --> 00:44:17.850 align:middle line:84% social pressure for book-buying should increase. 00:44:17.850 --> 00:44:20.510 align:middle line:84% And then the indicated social capital-- 00:44:20.510 --> 00:44:22.830 align:middle line:84% or the social capital that you get from buying a book 00:44:22.830 --> 00:44:25.510 align:middle line:90% should increase. 00:44:25.510 --> 00:44:30.910 align:middle line:84% And then that changes the calculus in terms of it's 00:44:30.910 --> 00:44:33.430 align:middle line:84% important for me to buy books versus it's important for me 00:44:33.430 --> 00:44:36.810 align:middle line:90% to buy and read them. 00:44:36.810 --> 00:44:40.850 align:middle line:84% And so it influences buying books and the fraction 00:44:40.850 --> 00:44:44.278 align:middle line:90% of books that are read. 00:44:44.278 --> 00:44:46.070 align:middle line:84% Then that goes to the attractiveness piece, 00:44:46.070 --> 00:44:49.842 align:middle line:84% but then it also influences the fraction of books that are read. 00:44:49.842 --> 00:44:52.050 align:middle line:84% So out of the books that are bought, how many of them 00:44:52.050 --> 00:44:52.550 align:middle line:90% are read? 00:44:52.550 --> 00:44:55.890 align:middle line:84% And that depends on both this relative importance, so 00:44:55.890 --> 00:44:58.330 align:middle line:84% how important is it for me to read the book that I bought? 00:44:58.330 --> 00:45:02.410 align:middle line:84% And then this two groups, low-frequency and high-frequency 00:45:02.410 --> 00:45:03.570 align:middle line:90% readers, I just divided-- 00:45:03.570 --> 00:45:07.750 align:middle line:84% I had three before, but it was suggested, two is best. 00:45:07.750 --> 00:45:10.370 align:middle line:84% So yeah, there are people who read-- 00:45:10.370 --> 00:45:12.270 align:middle line:84% I think over half the people read zero. 00:45:12.270 --> 00:45:14.370 align:middle line:90% And so I was switching it from-- 00:45:14.370 --> 00:45:18.190 align:middle line:84% I was dividing it into zero, low and high. 00:45:18.190 --> 00:45:21.590 align:middle line:84% But I think just low and high is [INAUDIBLE] around here. 00:45:21.590 --> 00:45:24.330 align:middle line:90% 00:45:24.330 --> 00:45:30.010 align:middle line:90% And then-- OK. 00:45:30.010 --> 00:45:32.170 align:middle line:90% So let me pause here. 00:45:32.170 --> 00:45:32.770 align:middle line:90% So yeah. 00:45:32.770 --> 00:45:33.050 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 00:45:33.050 --> 00:45:34.925 align:middle line:84% AUDIENCE: The low- and high-frequency readers 00:45:34.925 --> 00:45:38.470 align:middle line:90% [INAUDIBLE] just kind of this-- 00:45:38.470 --> 00:45:42.950 align:middle line:84% it's coming from this indicated fraction that is 00:45:42.950 --> 00:45:44.370 align:middle line:90% determined by the reading rate. 00:45:44.370 --> 00:45:46.850 align:middle line:84% And then that I actually want to formulate, 00:45:46.850 --> 00:45:47.830 align:middle line:90% because it's kind of-- 00:45:47.830 --> 00:45:51.750 align:middle line:84% I don't know how discreet to make it. 00:45:51.750 --> 00:45:54.270 align:middle line:84% There are definitions in the literature, 00:45:54.270 --> 00:45:57.070 align:middle line:84% but no high frequency readers are reading at least one 00:45:57.070 --> 00:45:58.443 align:middle line:90% or at least two books a month. 00:45:58.443 --> 00:46:00.610 align:middle line:84% HAZHIR RAHMANDAD: Yeah, but then you don't need to-- 00:46:00.610 --> 00:46:02.902 align:middle line:84% I mean the definition can be whatever you choose and it 00:46:02.902 --> 00:46:06.870 align:middle line:84% can be discreet, but the movement between these stocks 00:46:06.870 --> 00:46:08.350 align:middle line:90% wouldn't be discreet. 00:46:08.350 --> 00:46:10.390 align:middle line:84% So you don't need to worry too much 00:46:10.390 --> 00:46:12.765 align:middle line:90% about the discreet definitions. 00:46:12.765 --> 00:46:13.390 align:middle line:90% AUDIENCE: Yeah. 00:46:13.390 --> 00:46:15.935 align:middle line:90% 00:46:15.935 --> 00:46:16.810 align:middle line:90% HAZHIR RAHMANDAD: OK. 00:46:16.810 --> 00:46:17.310 align:middle line:90% Sounds good. 00:46:17.310 --> 00:46:20.170 align:middle line:84% So let's now go through some of these. 00:46:20.170 --> 00:46:23.670 align:middle line:84% So basically you have the high- and low-frequency readers. 00:46:23.670 --> 00:46:26.930 align:middle line:84% They each are reading a number per month. 00:46:26.930 --> 00:46:29.370 align:middle line:84% Is that how things are determined? 00:46:29.370 --> 00:46:33.310 align:middle line:84% So the reading rate, so right now-- 00:46:33.310 --> 00:46:37.970 align:middle line:90% 00:46:37.970 --> 00:46:42.090 align:middle line:84% AUDIENCE: So that big spot for Kristen should be subdivided 00:46:42.090 --> 00:46:43.570 align:middle line:90% or [INAUDIBLE]. 00:46:43.570 --> 00:46:48.290 align:middle line:84% But right now, it's in the-- it's not. 00:46:48.290 --> 00:46:51.010 align:middle line:84% HAZHIR RAHMANDAD: So books bought per person 00:46:51.010 --> 00:46:55.480 align:middle line:84% should be subscripted by high versus low frequency. 00:46:55.480 --> 00:46:57.730 align:middle line:84% AUDIENCE: Yeah, but then it's kind of double counting. 00:46:57.730 --> 00:47:02.950 align:middle line:84% HAZHIR RAHMANDAD: So yeah, I'm not quite sure if-- 00:47:02.950 --> 00:47:05.610 align:middle line:90% 00:47:05.610 --> 00:47:08.670 align:middle line:90% so you could have that-- 00:47:08.670 --> 00:47:13.930 align:middle line:90% 00:47:13.930 --> 00:47:19.090 align:middle line:84% so there's book buying rate and then you have book reading rate. 00:47:19.090 --> 00:47:22.950 align:middle line:84% So this comes from fraction of bought books that are read. 00:47:22.950 --> 00:47:37.590 align:middle line:90% 00:47:37.590 --> 00:47:41.470 align:middle line:84% So the worry I have is that you're 00:47:41.470 --> 00:47:47.590 align:middle line:84% calculating this fraction of bought books read somewhere, 00:47:47.590 --> 00:47:51.350 align:middle line:84% and then you're capturing book-buying rates somewhere 00:47:51.350 --> 00:47:52.150 align:middle line:90% else. 00:47:52.150 --> 00:47:59.110 align:middle line:84% And the two need to be fairly closely integrated. 00:47:59.110 --> 00:48:02.350 align:middle line:84% So if I increase the number of books 00:48:02.350 --> 00:48:06.070 align:middle line:84% that are bought to a very large number, 00:48:06.070 --> 00:48:09.230 align:middle line:84% the fraction of books that are read 00:48:09.230 --> 00:48:13.590 align:middle line:84% should go down in a very consistent manner. 00:48:13.590 --> 00:48:20.590 align:middle line:84% So if today I double the number of books that are bought by me, 00:48:20.590 --> 00:48:25.070 align:middle line:84% my reading rate should go close to half as a fraction. 00:48:25.070 --> 00:48:29.110 align:middle line:84% AUDIENCE: So there should be some kind of reading capacity? 00:48:29.110 --> 00:48:30.530 align:middle line:90% HAZHIR RAHMANDAD: Something. 00:48:30.530 --> 00:48:34.070 align:middle line:84% That's-- Right now it is not tightly connected. 00:48:34.070 --> 00:48:36.570 align:middle line:84% So you may get very weird behaviors that just 00:48:36.570 --> 00:48:39.770 align:middle line:90% don't make sense physically. 00:48:39.770 --> 00:48:42.630 align:middle line:84% So you want that tighter connection. 00:48:42.630 --> 00:48:46.130 align:middle line:84% Now how do you impose that tighter connection 00:48:46.130 --> 00:48:47.750 align:middle line:90% is non-trivial. 00:48:47.750 --> 00:48:50.910 align:middle line:84% So maybe capacity is a useful thing. 00:48:50.910 --> 00:48:53.550 align:middle line:84% So that is each of these two groups, 00:48:53.550 --> 00:48:58.210 align:middle line:84% you think of them as people who have this much time 00:48:58.210 --> 00:49:01.050 align:middle line:90% at a maximum for reading. 00:49:01.050 --> 00:49:06.210 align:middle line:84% And as you buy more books, you get 00:49:06.210 --> 00:49:08.970 align:middle line:84% closer to saturating that amount of time 00:49:08.970 --> 00:49:11.830 align:middle line:84% that you may allocate to reading. 00:49:11.830 --> 00:49:14.490 align:middle line:84% You can never exceed that amount. 00:49:14.490 --> 00:49:17.330 align:middle line:84% And that amount of time that you spend reading 00:49:17.330 --> 00:49:20.170 align:middle line:84% determines how many books you are going to read. 00:49:20.170 --> 00:49:22.530 align:middle line:84% Now, if you're not buying any books, 00:49:22.530 --> 00:49:24.650 align:middle line:90% you're not reading any books. 00:49:24.650 --> 00:49:28.170 align:middle line:84% So the lower bound is also should be there, 00:49:28.170 --> 00:49:30.330 align:middle line:90% but the upper bound saturates. 00:49:30.330 --> 00:49:33.150 align:middle line:84% And now if you're buying 100 books, 00:49:33.150 --> 00:49:38.750 align:middle line:84% you're not reading most of them per month, and that's fine. 00:49:38.750 --> 00:49:44.310 align:middle line:84% So making that connection tight and clean is really important. 00:49:44.310 --> 00:49:49.110 align:middle line:84% That might be the most important part of getting this model 00:49:49.110 --> 00:49:50.870 align:middle line:90% to behave meaningfully. 00:49:50.870 --> 00:49:52.370 align:middle line:84% AUDIENCE: That makes a lot of sense. 00:49:52.370 --> 00:49:55.350 align:middle line:84% I think I was so focused on those things being decoupled 00:49:55.350 --> 00:49:57.710 align:middle line:90% that I didn't think about that. 00:49:57.710 --> 00:49:59.830 align:middle line:84% HAZHIR RAHMANDAD: Yes, but decoupling 00:49:59.830 --> 00:50:01.650 align:middle line:84% comes from really physics of reading. 00:50:01.650 --> 00:50:04.550 align:middle line:84% So if the physics are not captured well, 00:50:04.550 --> 00:50:10.510 align:middle line:84% everything else will break down and do weird things. 00:50:10.510 --> 00:50:12.910 align:middle line:84% Then you have this attractiveness factor 00:50:12.910 --> 00:50:15.910 align:middle line:84% that you're saying, OK, buying books 00:50:15.910 --> 00:50:18.790 align:middle line:84% because of the social media creates 00:50:18.790 --> 00:50:23.010 align:middle line:84% a signal of attractiveness, prestige, 00:50:23.010 --> 00:50:26.750 align:middle line:84% whatever, that adds to the reasons for buying books. 00:50:26.750 --> 00:50:30.450 align:middle line:84% So basically, people have some desired level of reading 00:50:30.450 --> 00:50:34.170 align:middle line:84% that might be driving the buying of books 00:50:34.170 --> 00:50:37.530 align:middle line:84% before this attractiveness came in through the social media 00:50:37.530 --> 00:50:38.490 align:middle line:90% channel. 00:50:38.490 --> 00:50:40.250 align:middle line:84% And now you're saying, OK, now they 00:50:40.250 --> 00:50:44.730 align:middle line:84% are buying more books beyond what they normally 00:50:44.730 --> 00:50:47.850 align:middle line:90% read to show off to say that-- 00:50:47.850 --> 00:50:53.130 align:middle line:84% so to capture that you would anchor the demand, probably, 00:50:53.130 --> 00:50:56.910 align:middle line:84% to what they are reading right now or what they want to read. 00:50:56.910 --> 00:50:59.270 align:middle line:84% They have a desired level of actual reading. 00:50:59.270 --> 00:51:00.590 align:middle line:90% That's your initial anchor. 00:51:00.590 --> 00:51:05.250 align:middle line:84% And then you can adjust from that anchor towards a larger 00:51:05.250 --> 00:51:10.610 align:middle line:84% or smaller value of purchasing based on their social media 00:51:10.610 --> 00:51:14.330 align:middle line:84% prestige factor you're highlighting. 00:51:14.330 --> 00:51:18.650 align:middle line:84% AUDIENCE: Is that captured by the reference books bought 00:51:18.650 --> 00:51:20.390 align:middle line:84% per person and I can flesh it out, 00:51:20.390 --> 00:51:23.760 align:middle line:84% or is that-- are you referring to something different? 00:51:23.760 --> 00:51:26.240 align:middle line:84% HAZHIR RAHMANDAD: So these reference, 00:51:26.240 --> 00:51:29.120 align:middle line:84% SC reference, HC, these are what? 00:51:29.120 --> 00:51:31.080 align:middle line:84% AUDIENCE: Well, sorry, below those 00:51:31.080 --> 00:51:33.180 align:middle line:84% there's reference books about the person. 00:51:33.180 --> 00:51:33.320 align:middle line:90% HAZHIR RAHMANDAD: Oh. 00:51:33.320 --> 00:51:34.080 align:middle line:90% [INTERPOSING VOICES] 00:51:34.080 --> 00:51:36.220 align:middle line:84% AUDIENCE: Reference SC is reference Social Capital, 00:51:36.220 --> 00:51:36.840 align:middle line:90% HC is Human Capital. 00:51:36.840 --> 00:51:37.840 align:middle line:90% HAZHIR RAHMANDAD: I see. 00:51:37.840 --> 00:51:42.980 align:middle line:84% So the equation here is one of these complex ones. 00:51:42.980 --> 00:51:46.240 align:middle line:90% I feel intimidated by this. 00:51:46.240 --> 00:51:50.080 align:middle line:84% AUDIENCE: This is the attractiveness section. 00:51:50.080 --> 00:51:53.080 align:middle line:84% HAZHIR RAHMANDAD: So each of them, you have a sensitivity. 00:51:53.080 --> 00:51:58.120 align:middle line:84% You have a human capital gain via buying and reading divided 00:51:58.120 --> 00:52:00.170 align:middle line:90% by that reference. 00:52:00.170 --> 00:52:02.500 align:middle line:84% AUDIENCE: And then it's just a multiplicative. 00:52:02.500 --> 00:52:04.320 align:middle line:90% HAZHIR RAHMANDAD: OK, got it. 00:52:04.320 --> 00:52:08.380 align:middle line:84% So HC and SC are the high and low frequency readers? 00:52:08.380 --> 00:52:08.880 align:middle line:90% No. 00:52:08.880 --> 00:52:10.520 align:middle line:84% AUDIENCE: No, human capital and social capital. 00:52:10.520 --> 00:52:11.395 align:middle line:90% HAZHIR RAHMANDAD: OK. 00:52:11.395 --> 00:52:14.500 align:middle line:84% So you have the human capital and social capital factors. 00:52:14.500 --> 00:52:22.140 align:middle line:90% 00:52:22.140 --> 00:52:24.440 align:middle line:84% I don't know, this attractiveness of books. 00:52:24.440 --> 00:52:26.900 align:middle line:84% First of all, you need to separate it 00:52:26.900 --> 00:52:28.360 align:middle line:90% for these two groups. 00:52:28.360 --> 00:52:31.700 align:middle line:90% Right now it is a single value. 00:52:31.700 --> 00:52:35.380 align:middle line:84% Now, you may say the effect of prestige 00:52:35.380 --> 00:52:39.580 align:middle line:84% or social capital on that attractiveness could 00:52:39.580 --> 00:52:41.160 align:middle line:90% be the same for both groups. 00:52:41.160 --> 00:52:44.220 align:middle line:90% That's OK assumption. 00:52:44.220 --> 00:52:47.980 align:middle line:84% So you would calculate that effect of social capital 00:52:47.980 --> 00:52:52.100 align:middle line:84% on attractiveness and multiply it by the baseline book 00:52:52.100 --> 00:52:55.460 align:middle line:90% demand from each group. 00:52:55.460 --> 00:52:57.020 align:middle line:84% But the baseline book demand should 00:52:57.020 --> 00:52:58.440 align:middle line:90% be distinct for the two groups. 00:52:58.440 --> 00:53:03.820 align:middle line:84% Otherwise, it just doesn't make sense internally. 00:53:03.820 --> 00:53:07.740 align:middle line:84% So let's say you have these separate demands or baseline 00:53:07.740 --> 00:53:12.700 align:middle line:84% demands that is coming from, you're calling it human capital. 00:53:12.700 --> 00:53:17.140 align:middle line:84% But I wouldn't-- basically it is, yes, 00:53:17.140 --> 00:53:25.860 align:middle line:84% how many books they want to read left on their own devices. 00:53:25.860 --> 00:53:28.560 align:middle line:90% AUDIENCE: Yeah. 00:53:28.560 --> 00:53:31.460 align:middle line:84% HAZHIR RAHMANDAD: So you can do that. 00:53:31.460 --> 00:53:36.360 align:middle line:90% 00:53:36.360 --> 00:53:42.360 align:middle line:84% And then you get this social pressure effect 00:53:42.360 --> 00:53:46.600 align:middle line:84% that expands that number that are purchased. 00:53:46.600 --> 00:53:49.720 align:middle line:84% And then you can decide whether reading of books 00:53:49.720 --> 00:53:51.980 align:middle line:84% go up because I have bought more books. 00:53:51.980 --> 00:53:55.580 align:middle line:84% So maybe I read a little bit more or maybe I don't. 00:53:55.580 --> 00:53:57.140 align:middle line:84% You can make an assumption on that. 00:53:57.140 --> 00:53:58.900 align:middle line:84% You should have a parameter that says, 00:53:58.900 --> 00:54:01.600 align:middle line:84% what is the effect of buying more books 00:54:01.600 --> 00:54:04.560 align:middle line:84% or having access to more books on your reading 00:54:04.560 --> 00:54:08.040 align:middle line:84% rate for these two groups that would saturate 00:54:08.040 --> 00:54:11.420 align:middle line:84% to how much maximum time you are willing to allocate to it. 00:54:11.420 --> 00:54:13.960 align:middle line:84% So maybe if your baseline demand is 00:54:13.960 --> 00:54:17.240 align:middle line:84% for two books per month for the high readers 00:54:17.240 --> 00:54:21.460 align:middle line:84% and 0.2 books per month for the low readers, 00:54:21.460 --> 00:54:23.980 align:middle line:84% you're not going to more than double 00:54:23.980 --> 00:54:26.380 align:middle line:84% that amount of reading just because you have bought more 00:54:26.380 --> 00:54:27.720 align:middle line:90% books or something like that. 00:54:27.720 --> 00:54:30.980 align:middle line:84% And so you create that saturation point. 00:54:30.980 --> 00:54:34.760 align:middle line:84% So that's, I think, the key pieces you need to incorporate. 00:54:34.760 --> 00:54:36.820 align:middle line:84% Then there is the question of movement of people 00:54:36.820 --> 00:54:39.700 align:middle line:84% between low-frequency and high-frequency readers, 00:54:39.700 --> 00:54:42.700 align:middle line:84% and whether that is endogenous or exogenous. 00:54:42.700 --> 00:54:51.140 align:middle line:84% So if I'm seeing people in the low-frequency reader group who 00:54:51.140 --> 00:54:54.340 align:middle line:84% are reading more books than the baseline, 00:54:54.340 --> 00:54:56.460 align:middle line:84% my expectation is that a few of them 00:54:56.460 --> 00:55:02.580 align:middle line:84% will move to the high-frequency reader category. 00:55:02.580 --> 00:55:07.640 align:middle line:84% Alternatively, if I'm seeing them read less, 00:55:07.640 --> 00:55:10.420 align:middle line:84% I would expect some high-frequency ones 00:55:10.420 --> 00:55:12.200 align:middle line:90% to come to the other one. 00:55:12.200 --> 00:55:14.960 align:middle line:84% There is a source of endogeneity here, 00:55:14.960 --> 00:55:17.740 align:middle line:84% but your story is a little bit more exogenous, 00:55:17.740 --> 00:55:20.140 align:middle line:84% that there is a trend in general, 00:55:20.140 --> 00:55:21.760 align:middle line:84% that people are reading less books 00:55:21.760 --> 00:55:23.320 align:middle line:84% and moving from the high-frequency 00:55:23.320 --> 00:55:26.400 align:middle line:84% to low-frequency category just because they are busy doing 00:55:26.400 --> 00:55:30.000 align:middle line:84% other things, and maybe they're too much on the social media 00:55:30.000 --> 00:55:31.720 align:middle line:90% looking at other people's books. 00:55:31.720 --> 00:55:38.080 align:middle line:84% But that story then has an exogenous flow. 00:55:38.080 --> 00:55:41.920 align:middle line:84% And basically then your solution to the puzzle 00:55:41.920 --> 00:55:45.240 align:middle line:84% is this exogenous flow of people from high frequency 00:55:45.240 --> 00:55:49.840 align:middle line:84% to low frequency, combined with this social pressure 00:55:49.840 --> 00:55:58.720 align:middle line:84% dynamic that might be increasing the attractiveness of buying 00:55:58.720 --> 00:56:00.280 align:middle line:90% books. 00:56:00.280 --> 00:56:04.385 align:middle line:84% AUDIENCE: So try it with an exogenous transition? 00:56:04.385 --> 00:56:05.760 align:middle line:84% HAZHIR RAHMANDAD: Yeah, for now I 00:56:05.760 --> 00:56:07.660 align:middle line:84% think that's a good starting point. 00:56:07.660 --> 00:56:09.680 align:middle line:90% AUDIENCE: OK. 00:56:09.680 --> 00:56:12.900 align:middle line:84% HAZHIR RAHMANDAD: Again, there is-- 00:56:12.900 --> 00:56:19.420 align:middle line:84% so that is the effect coming from all the other factors. 00:56:19.420 --> 00:56:21.797 align:middle line:84% And then there is the effect of, oh, I have more books, 00:56:21.797 --> 00:56:22.880 align:middle line:90% so I'm reading more books. 00:56:22.880 --> 00:56:25.740 align:middle line:84% So maybe a few of these people become interested in books 00:56:25.740 --> 00:56:28.180 align:middle line:84% and actually go to the right-hand side. 00:56:28.180 --> 00:56:32.820 align:middle line:84% That effect is worse incorporating in your model 00:56:32.820 --> 00:56:36.300 align:middle line:84% as a second step once you have the first exogenous 00:56:36.300 --> 00:56:39.100 align:middle line:90% part working. 00:56:39.100 --> 00:56:39.720 align:middle line:90% Jason. 00:56:39.720 --> 00:56:41.700 align:middle line:90% AUDIENCE: You could make it-- 00:56:41.700 --> 00:56:45.620 align:middle line:84% you could tell another story where older generations 00:56:45.620 --> 00:56:47.160 align:middle line:90% are the high-frequency readers. 00:56:47.160 --> 00:56:50.220 align:middle line:90% They die and go away. 00:56:50.220 --> 00:56:54.420 align:middle line:84% AUDIENCE: Actually, older people, for the first time, 00:56:54.420 --> 00:56:59.103 align:middle line:84% are reading less books than the middle-aged groups. 00:56:59.103 --> 00:57:01.020 align:middle line:84% AUDIENCE: I was just trying to get more people 00:57:01.020 --> 00:57:02.897 align:middle line:90% in your low-frequency stuff. 00:57:02.897 --> 00:57:05.480 align:middle line:84% HAZHIR RAHMANDAD: Well, so you could tell those other stories. 00:57:05.480 --> 00:57:06.980 align:middle line:84% But I think at this stage you don't 00:57:06.980 --> 00:57:10.400 align:middle line:84% want to go into the demographic part. 00:57:10.400 --> 00:57:14.760 align:middle line:84% That's a whole different story structure. 00:57:14.760 --> 00:57:17.295 align:middle line:84% First get the very simple without demographic one. 00:57:17.295 --> 00:57:17.920 align:middle line:90% AUDIENCE: Yeah. 00:57:17.920 --> 00:57:22.560 align:middle line:84% AUDIENCE: I think the exogenous explanation for why these are 00:57:22.560 --> 00:57:28.560 align:middle line:84% diverging but we still have all the high-frequency people who 00:57:28.560 --> 00:57:34.080 align:middle line:84% are reading 150 books a year, is-- 00:57:34.080 --> 00:57:36.880 align:middle line:84% there's some reinforcing dynamic just 00:57:36.880 --> 00:57:39.480 align:middle line:84% among high-frequency readers where they're going 00:57:39.480 --> 00:57:43.520 align:middle line:90% from 20 or 40 to 100 or 150. 00:57:43.520 --> 00:57:46.080 align:middle line:84% And I think that's, in article that I've 00:57:46.080 --> 00:57:47.800 align:middle line:90% read, that's driven by-- 00:57:47.800 --> 00:57:51.920 align:middle line:84% or that's emphasized to be driven by e-books and audiobooks 00:57:51.920 --> 00:57:56.800 align:middle line:84% being so much more available and popular. 00:57:56.800 --> 00:58:00.760 align:middle line:84% So I think I could just have an exogenous model that 00:58:00.760 --> 00:58:06.960 align:middle line:84% would do this with that dynamic and the social capital then 00:58:06.960 --> 00:58:09.863 align:middle line:84% is driven by this small segment of the market. 00:58:09.863 --> 00:58:11.780 align:middle line:84% HAZHIR RAHMANDAD: So the story you are telling 00:58:11.780 --> 00:58:16.620 align:middle line:84% might suggest that people, how sensitive they 00:58:16.620 --> 00:58:19.060 align:middle line:84% are to this social capital mechanism 00:58:19.060 --> 00:58:20.740 align:middle line:84% may actually be differentiable between 00:58:20.740 --> 00:58:23.780 align:middle line:84% the low-frequency readers and high-frequency readers. 00:58:23.780 --> 00:58:25.260 align:middle line:84% That is, the high-frequency readers 00:58:25.260 --> 00:58:29.260 align:middle line:84% are more sensitive to the social capital story, 00:58:29.260 --> 00:58:32.460 align:middle line:84% and the low-frequency readers are not sensitive to it. 00:58:32.460 --> 00:58:35.660 align:middle line:84% And that could create a situation 00:58:35.660 --> 00:58:38.860 align:middle line:84% where the high-frequency readers buy more books 00:58:38.860 --> 00:58:40.780 align:middle line:90% and read more books. 00:58:40.780 --> 00:58:44.140 align:middle line:84% So they don't saturate as fast, even 00:58:44.140 --> 00:58:48.040 align:middle line:84% though they are spending a lot more time on reading books. 00:58:48.040 --> 00:58:50.700 align:middle line:90% 00:58:50.700 --> 00:58:52.820 align:middle line:84% So your model, I think, should allow 00:58:52.820 --> 00:58:56.780 align:middle line:84% you to generate that hypothesis as well. 00:58:56.780 --> 00:59:00.340 align:middle line:84% That is, it should have enough degrees of freedom in parameters 00:59:00.340 --> 00:59:03.180 align:middle line:90% to do that as well. 00:59:03.180 --> 00:59:04.540 align:middle line:90% AUDIENCE: OK, that's helpful. 00:59:04.540 --> 00:59:05.498 align:middle line:90% HAZHIR RAHMANDAD: Good. 00:59:05.498 --> 00:59:11.120 align:middle line:84% Other questions you may have on this 00:59:11.120 --> 00:59:13.160 align:middle line:84% AUDIENCE: I think it's OK that I was just 00:59:13.160 --> 00:59:15.040 align:middle line:90% finding the [INAUDIBLE]? 00:59:15.040 --> 00:59:17.160 align:middle line:84% HAZHIR RAHMANDAD: So I think the part that we 00:59:17.160 --> 00:59:22.508 align:middle line:84% haven't talked about is how this social capital gets constructed. 00:59:22.508 --> 00:59:24.300 align:middle line:84% AUDIENCE: Yeah, I have not formulated that. 00:59:24.300 --> 00:59:28.040 align:middle line:84% HAZHIR RAHMANDAD: Yeah, and that is not necessarily very easy 00:59:28.040 --> 00:59:34.080 align:middle line:84% as well because that's the more softer side of the model. 00:59:34.080 --> 00:59:38.240 align:middle line:90% So what are the thoughts? 00:59:38.240 --> 00:59:42.200 align:middle line:84% Anybody has suggestions for how Cathy 00:59:42.200 --> 00:59:46.760 align:middle line:84% can formulate the social capital in a simple way 00:59:46.760 --> 00:59:48.460 align:middle line:90% and well-grounded way? 00:59:48.460 --> 00:59:58.800 align:middle line:90% 00:59:58.800 --> 01:00:01.600 align:middle line:84% AUDIENCE: Is it going to be separate for high-frequency 01:00:01.600 --> 01:00:04.840 align:middle line:90% versus low-frequency? 01:00:04.840 --> 01:00:06.420 align:middle line:90% HAZHIR RAHMANDAD: Could be. 01:00:06.420 --> 01:00:09.860 align:middle line:84% I mean, that's a decision on the modeling side. 01:00:09.860 --> 01:00:25.860 align:middle line:90% 01:00:25.860 --> 01:00:28.380 align:middle line:84% AUDIENCE: Are you aware of any papers 01:00:28.380 --> 01:00:30.560 align:middle line:84% that have done on this [INAUDIBLE] model? 01:00:30.560 --> 01:00:32.900 align:middle line:90% I was wanting to-- 01:00:32.900 --> 01:00:34.380 align:middle line:84% HAZHIR RAHMANDAD: So it's actually 01:00:34.380 --> 01:00:38.700 align:middle line:84% a good question that is, what is a standard formulation that 01:00:38.700 --> 01:00:43.860 align:middle line:84% captures this idea of social pressure, 01:00:43.860 --> 01:00:46.360 align:middle line:90% role-setting, norm-setting. 01:00:46.360 --> 01:00:49.260 align:middle line:90% 01:00:49.260 --> 01:00:52.840 align:middle line:84% AUDIENCE: You could use social learning models. 01:00:52.840 --> 01:00:56.495 align:middle line:90% 01:00:56.495 --> 01:00:57.120 align:middle line:90% AUDIENCE: Yeah. 01:00:57.120 --> 01:01:00.342 align:middle line:90% 01:01:00.342 --> 01:01:01.300 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 01:01:01.300 --> 01:01:02.000 align:middle line:84% AUDIENCE: It's a little different. 01:01:02.000 --> 01:01:03.917 align:middle line:84% HAZHIR RAHMANDAD: It's a little bit different. 01:01:03.917 --> 01:01:05.630 align:middle line:90% But you have to adopt something. 01:01:05.630 --> 01:01:07.642 align:middle line:84% So, Oriana, did you have some ideas? 01:01:07.642 --> 01:01:09.850 align:middle line:84% AUDIENCE: I was thinking maybe you could do something 01:01:09.850 --> 01:01:17.770 align:middle line:84% like does social capital gain because the fraction 01:01:17.770 --> 01:01:22.810 align:middle line:84% of the books that you have over the average number of books 01:01:22.810 --> 01:01:28.650 align:middle line:84% per person if you have-- well, it wouldn't be linear, 01:01:28.650 --> 01:01:32.430 align:middle line:84% but if you have 10% more books than the average person, 01:01:32.430 --> 01:01:35.970 align:middle line:90% then you're 5% [INAUDIBLE]. 01:01:35.970 --> 01:01:39.210 align:middle line:84% But I actually don't know how that would go. 01:01:39.210 --> 01:01:41.650 align:middle line:84% You need to do a singular agent thinking 01:01:41.650 --> 01:01:44.090 align:middle line:84% about their own social capital and however 01:01:44.090 --> 01:01:46.355 align:middle line:90% [INAUDIBLE] a bigger model. 01:01:46.355 --> 01:01:48.230 align:middle line:84% HAZHIR RAHMANDAD: That's an interesting idea. 01:01:48.230 --> 01:01:51.610 align:middle line:84% So basically, you're saying, look at what's 01:01:51.610 --> 01:01:54.490 align:middle line:84% the norm in terms of the number of books per person 01:01:54.490 --> 01:02:02.230 align:middle line:84% that is being bought up, and then you try to anchor to that 01:02:02.230 --> 01:02:05.870 align:middle line:84% and go a little bit higher or potentially more 01:02:05.870 --> 01:02:08.530 align:middle line:84% than a little bit, depending on how aggressive you are. 01:02:08.530 --> 01:02:11.670 align:middle line:84% But average of those folks, they are trying to anchor to that 01:02:11.670 --> 01:02:14.070 align:middle line:90% and go a little bit above that. 01:02:14.070 --> 01:02:18.590 align:middle line:84% That's how they adjust their buying rate 01:02:18.590 --> 01:02:21.390 align:middle line:90% to gain social capital. 01:02:21.390 --> 01:02:25.190 align:middle line:84% So that's a decent way of tackling it. 01:02:25.190 --> 01:02:30.770 align:middle line:84% Now that creates a very simple, quick reinforcing loop. 01:02:30.770 --> 01:02:35.370 align:middle line:84% So basically people buy more books per person. 01:02:35.370 --> 01:02:37.370 align:middle line:84% That increases the goal for everybody. 01:02:37.370 --> 01:02:39.350 align:middle line:84% And everybody is going to do more. 01:02:39.350 --> 01:02:41.510 align:middle line:84% So you leave that on its own, people 01:02:41.510 --> 01:02:45.190 align:middle line:84% will buy infinite number of books very quickly. 01:02:45.190 --> 01:02:48.970 align:middle line:84% So that creates a weird dynamic that is not realistic. 01:02:48.970 --> 01:02:52.742 align:middle line:84% So in practice, what stops you from that? 01:02:52.742 --> 01:02:53.450 align:middle line:90% AUDIENCE: Income. 01:02:53.450 --> 01:02:54.510 align:middle line:90% HAZHIR RAHMANDAD: Exam? 01:02:54.510 --> 01:02:55.190 align:middle line:90% 01:02:55.190 --> 01:02:56.110 align:middle line:90% AUDIENCE: Income. 01:02:56.110 --> 01:03:01.010 align:middle line:84% HAZHIR RAHMANDAD: So you need to think about the bounding factors 01:03:01.010 --> 01:03:03.750 align:middle line:84% that slows down that reinforcing loop 01:03:03.750 --> 01:03:06.450 align:middle line:84% if you are going to use that reinforcing loop formulation 01:03:06.450 --> 01:03:07.870 align:middle line:90% that Oriana suggested. 01:03:07.870 --> 01:03:11.430 align:middle line:84% I think the reinforcing loop mechanism is OK. 01:03:11.430 --> 01:03:16.450 align:middle line:84% I think that's a decent perspective because people 01:03:16.450 --> 01:03:21.150 align:middle line:84% are doing peer comparisons and try to outperform their peers. 01:03:21.150 --> 01:03:22.910 align:middle line:84% You need the balancing mechanisms, 01:03:22.910 --> 01:03:26.050 align:middle line:84% either coming from your income or from your ability 01:03:26.050 --> 01:03:29.250 align:middle line:90% to read these books. 01:03:29.250 --> 01:03:31.010 align:middle line:90% So yeah. 01:03:31.010 --> 01:03:35.290 align:middle line:84% AUDIENCE: I think maybe another way you can stop it 01:03:35.290 --> 01:03:40.530 align:middle line:84% from getting too crazy is how much social capital you 01:03:40.530 --> 01:03:43.210 align:middle line:84% gain is dependent on how many people care 01:03:43.210 --> 01:03:45.190 align:middle line:90% or can compete on that scale. 01:03:45.190 --> 01:03:48.770 align:middle line:84% So at five books, you have the entire population 01:03:48.770 --> 01:03:50.070 align:middle line:90% of readers is competing. 01:03:50.070 --> 01:03:52.230 align:middle line:84% And so you'll see it go up pretty quick. 01:03:52.230 --> 01:03:55.010 align:middle line:84% But look, when you're buying 100 books, you're saying, 01:03:55.010 --> 01:03:57.390 align:middle line:84% the income, there's a very small number of people. 01:03:57.390 --> 01:03:58.910 align:middle line:84% And so then the number of people who 01:03:58.910 --> 01:04:01.630 align:middle line:84% really care about that book or that competition 01:04:01.630 --> 01:04:02.790 align:middle line:90% gets really small. 01:04:02.790 --> 01:04:07.748 align:middle line:84% So that should make it plateau and then eventually fall off. 01:04:07.748 --> 01:04:09.790 align:middle line:84% HAZHIR RAHMANDAD: Yeah, that's an interesting way 01:04:09.790 --> 01:04:11.210 align:middle line:90% of thinking about it as well. 01:04:11.210 --> 01:04:13.750 align:middle line:84% So but to do that, basically you now 01:04:13.750 --> 01:04:17.590 align:middle line:84% have to break down the population of readers 01:04:17.590 --> 01:04:19.590 align:middle line:90% into different buckets. 01:04:19.590 --> 01:04:21.990 align:middle line:84% So this is now you would have movement 01:04:21.990 --> 01:04:26.490 align:middle line:84% of people across these buckets of 10 books a month, 01:04:26.490 --> 01:04:29.360 align:middle line:90% 20 books a month, or whatever. 01:04:29.360 --> 01:04:32.070 align:middle line:84% AUDIENCE: [INAUDIBLE] at some point, 01:04:32.070 --> 01:04:34.570 align:middle line:90% you look at somebody who has-- 01:04:34.570 --> 01:04:38.190 align:middle line:84% they're buying just a number of books that's unreasonable 01:04:38.190 --> 01:04:41.430 align:middle line:84% and you think, oh, they're not actually reading those. 01:04:41.430 --> 01:04:44.030 align:middle line:84% It becomes kind of a cloud story where they actually 01:04:44.030 --> 01:04:45.910 align:middle line:90% get discounted. 01:04:45.910 --> 01:04:48.750 align:middle line:84% So maybe there's a way to, if you're buying too many 01:04:48.750 --> 01:04:50.090 align:middle line:90% that it's ridiculous. 01:04:50.090 --> 01:04:51.290 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 01:04:51.290 --> 01:04:53.050 align:middle line:84% So basically you need a feedback there. 01:04:53.050 --> 01:04:55.810 align:middle line:84% It could be because of the money. 01:04:55.810 --> 01:04:57.870 align:middle line:84% It could be because you think it is ridiculous, 01:04:57.870 --> 01:05:00.338 align:middle line:84% because you think for yourself, I'm not 01:05:00.338 --> 01:05:01.630 align:middle line:90% going to read all of this book. 01:05:01.630 --> 01:05:03.290 align:middle line:90% Why am I going to buy them? 01:05:03.290 --> 01:05:04.910 align:middle line:90% Any of those are fine. 01:05:04.910 --> 01:05:08.370 align:middle line:84% So basically you would have a decreasing return 01:05:08.370 --> 01:05:11.010 align:middle line:84% and, at some point, negative return 01:05:11.010 --> 01:05:14.410 align:middle line:84% on your utility from buying more books because you're losing 01:05:14.410 --> 01:05:17.210 align:middle line:90% money and not reading anything. 01:05:17.210 --> 01:05:20.090 align:middle line:84% AUDIENCE: There is a 2014 [INAUDIBLE] paper 01:05:20.090 --> 01:05:22.230 align:middle line:90% on the curse of publicity. 01:05:22.230 --> 01:05:27.090 align:middle line:84% It kind of related to this looking at prize-winning books. 01:05:27.090 --> 01:05:30.850 align:middle line:84% And they're saying that these prize-winners, because they're 01:05:30.850 --> 01:05:33.210 align:middle line:84% getting a lot of attraction, they actually 01:05:33.210 --> 01:05:35.770 align:middle line:84% reach to certain markets, people are not 01:05:35.770 --> 01:05:38.530 align:middle line:84% very interested in those books, but they've heard that quote, 01:05:38.530 --> 01:05:39.590 align:middle line:90% this is a famous book. 01:05:39.590 --> 01:05:40.970 align:middle line:90% So that's why. 01:05:40.970 --> 01:05:43.970 align:middle line:84% And they use Goodreads data showing 01:05:43.970 --> 01:05:46.170 align:middle line:84% that actually those types of book 01:05:46.170 --> 01:05:50.610 align:middle line:84% tends to get lower ratings compared to finally [INAUDIBLE] 01:05:50.610 --> 01:05:51.430 align:middle line:90% the price. 01:05:51.430 --> 01:05:51.810 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 01:05:51.810 --> 01:05:53.730 align:middle line:84% AUDIENCE: So if you want, I think that's social-- 01:05:53.730 --> 01:05:55.480 align:middle line:84% HAZHIR RAHMANDAD: That's very interesting. 01:05:55.480 --> 01:06:00.390 align:middle line:84% AUDIENCE: [INAUDIBLE] because I feel like, part of it 01:06:00.390 --> 01:06:04.250 align:middle line:84% isn't just necessarily the book-buying by itself. 01:06:04.250 --> 01:06:05.970 align:middle line:84% I mean, that's complicated too much, 01:06:05.970 --> 01:06:14.310 align:middle line:84% but I feel like the reason that now book-buying-- or now there's 01:06:14.310 --> 01:06:17.350 align:middle line:84% so much more social capital from buying books, 01:06:17.350 --> 01:06:19.630 align:middle line:84% does directly have to do with social media 01:06:19.630 --> 01:06:22.710 align:middle line:84% and does directly have to do with, oh, 01:06:22.710 --> 01:06:27.590 align:middle line:84% do you have 100% of the books that people are talking about. 01:06:27.590 --> 01:06:32.390 align:middle line:84% Whereas if you were to model this pre-2014, 01:06:32.390 --> 01:06:36.350 align:middle line:84% I don't know that the social pressure for book-buying 01:06:36.350 --> 01:06:39.957 align:middle line:84% would be directly determined by the book-buying rate. 01:06:39.957 --> 01:06:40.790 align:middle line:90% [INTERPOSING VOICES] 01:06:40.790 --> 01:06:41.770 align:middle line:90% AUDIENCE: Same. 01:06:41.770 --> 01:06:43.212 align:middle line:90% 01:06:43.212 --> 01:06:44.170 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 01:06:44.170 --> 01:06:47.390 align:middle line:84% And I think Cathy's hypothesis is also very much aligned 01:06:47.390 --> 01:06:51.790 align:middle line:84% that with social media and creating a different set 01:06:51.790 --> 01:06:54.650 align:middle line:84% of pressures that didn't exist, and now it 01:06:54.650 --> 01:07:01.730 align:middle line:84% creates this kind of separation between reading and buying. 01:07:01.730 --> 01:07:04.010 align:middle line:84% AUDIENCE: I just saw this student thesis 01:07:04.010 --> 01:07:08.075 align:middle line:90% that says that Goodreads is-- 01:07:08.075 --> 01:07:09.450 align:middle line:84% AUDIENCE: Performing [INAUDIBLE]. 01:07:09.450 --> 01:07:12.810 align:middle line:84% AUDIENCE: --some tracking from Lipton 01:07:12.810 --> 01:07:14.890 align:middle line:84% and Foucault's governmentality, Goodreads 01:07:14.890 --> 01:07:16.790 align:middle line:90% is a disciplinary technology. 01:07:16.790 --> 01:07:18.350 align:middle line:90% [LAUGHTER] 01:07:18.350 --> 01:07:18.850 align:middle line:90% [INAUDIBLE] 01:07:18.850 --> 01:07:20.870 align:middle line:90% HAZHIR RAHMANDAD: Cool. 01:07:20.870 --> 01:07:23.910 align:middle line:84% So let's now switch to Serchin's model. 01:07:23.910 --> 01:07:32.730 align:middle line:90% 01:07:32.730 --> 01:07:35.510 align:middle line:90% So Serchin. 01:07:35.510 --> 01:07:38.310 align:middle line:84% Please tell us what you're trying to do. 01:07:38.310 --> 01:07:41.090 align:middle line:84% This is-- let me first get a feel for-- 01:07:41.090 --> 01:07:41.590 align:middle line:90% OK. 01:07:41.590 --> 01:07:43.710 align:middle line:84% This is a little bit of a big model. 01:07:43.710 --> 01:07:49.720 align:middle line:84% So hopefully there are not too many views. 01:07:49.720 --> 01:07:51.600 align:middle line:90% [LAUGHTER] 01:07:51.600 --> 01:07:53.950 align:middle line:90% 01:07:53.950 --> 01:07:58.670 align:middle line:84% OK, so this might be well above what we can cover in the class 01:07:58.670 --> 01:08:01.830 align:middle line:84% because I think this is a model of the whole economy 01:08:01.830 --> 01:08:02.630 align:middle line:90% or something. 01:08:02.630 --> 01:08:06.990 align:middle line:84% But tell us where we should focus on 01:08:06.990 --> 01:08:13.710 align:middle line:84% so that it is useful for you and we can also 01:08:13.710 --> 01:08:17.350 align:middle line:90% have a good chat in the class. 01:08:17.350 --> 01:08:19.490 align:middle line:84% AUDIENCE: So first of all, good morning, Professor. 01:08:19.490 --> 01:08:22.430 align:middle line:90% Good morning, everyone. 01:08:22.430 --> 01:08:25.870 align:middle line:84% Maybe I should start with the problem definition. 01:08:25.870 --> 01:08:27.472 align:middle line:84% Maybe I can explain a little bit. 01:08:27.472 --> 01:08:28.430 align:middle line:90% HAZHIR RAHMANDAD: Sure. 01:08:28.430 --> 01:08:32.270 align:middle line:84% AUDIENCE: Because I couldn't attend the session 01:08:32.270 --> 01:08:35.189 align:middle line:84% that my problem definition is discussed. 01:08:35.189 --> 01:08:36.990 align:middle line:84% And I would like to thank everyone 01:08:36.990 --> 01:08:40.370 align:middle line:84% who took the time to comment on my problem definition. 01:08:40.370 --> 01:08:44.990 align:middle line:84% It was really useful while writing my final problem 01:08:44.990 --> 01:08:46.550 align:middle line:90% definition. 01:08:46.550 --> 01:08:51.130 align:middle line:84% So my main central question is that can 01:08:51.130 --> 01:08:53.930 align:middle line:84% the behavior of the firms, in order 01:08:53.930 --> 01:08:56.710 align:middle line:84% that are meant to tackle with uncertainty, 01:08:56.710 --> 01:08:59.890 align:middle line:84% can be a source of instability and create 01:08:59.890 --> 01:09:03.370 align:middle line:84% endogenous oscillations in the economy. 01:09:03.370 --> 01:09:08.370 align:middle line:84% Back in the day, Frank Knight and John Maynard Keynes 01:09:08.370 --> 01:09:12.290 align:middle line:84% made a distinction between risk and uncertainty. 01:09:12.290 --> 01:09:17.729 align:middle line:84% And in risk situations, we can list all possible outcomes 01:09:17.729 --> 01:09:20.450 align:middle line:84% and assign probabilities to them. 01:09:20.450 --> 01:09:23.189 align:middle line:84% On the other hand, in the case of uncertainty, 01:09:23.189 --> 01:09:27.330 align:middle line:84% we cannot even define the exhaustive list of outcomes. 01:09:27.330 --> 01:09:30.490 align:middle line:84% That can always be some black swan events. 01:09:30.490 --> 01:09:34.770 align:middle line:84% And we cannot assign a probability distribution. 01:09:34.770 --> 01:09:38.130 align:middle line:84% And in order to tackle with these situations, 01:09:38.130 --> 01:09:42.270 align:middle line:84% Keynes argues that economic agents act on animal spirits. 01:09:42.270 --> 01:09:45.250 align:middle line:84% And these are akin to what we might call 01:09:45.250 --> 01:09:50.240 align:middle line:90% as gut feelings or sentiments. 01:09:50.240 --> 01:09:53.240 align:middle line:84% In fact, sentiments are very important data 01:09:53.240 --> 01:09:57.960 align:middle line:84% that are closely followed by financial markets being real. 01:09:57.960 --> 01:10:02.000 align:middle line:84% For example, University of Michigan Sentiment Index 01:10:02.000 --> 01:10:08.200 align:middle line:84% or/and Consumer Confidence Index are two important sentiment 01:10:08.200 --> 01:10:09.740 align:middle line:90% indicators in the USA. 01:10:09.740 --> 01:10:14.160 align:middle line:84% And these intersect-- capture how optimistic or pessimistic 01:10:14.160 --> 01:10:19.000 align:middle line:84% economic agents feel about the present and future 01:10:19.000 --> 01:10:24.640 align:middle line:84% of the economy, which influence their spending and investment 01:10:24.640 --> 01:10:26.960 align:middle line:90% decisions. 01:10:26.960 --> 01:10:30.380 align:middle line:84% Apart from that, in mainstream economics, 01:10:30.380 --> 01:10:32.320 align:middle line:90% there are no buffer stocks. 01:10:32.320 --> 01:10:36.160 align:middle line:84% It is all about flows, as you probably know, 01:10:36.160 --> 01:10:40.020 align:middle line:84% which are assumed to be equal to each other all the time. 01:10:40.020 --> 01:10:44.440 align:middle line:84% So the demand is equal to supply and the economy 01:10:44.440 --> 01:10:48.700 align:middle line:90% is in equilibrium all the time. 01:10:48.700 --> 01:10:52.020 align:middle line:84% But in order to tackle with uncertainty, 01:10:52.020 --> 01:10:55.020 align:middle line:84% we know that economic agents use buffer stocks. 01:10:55.020 --> 01:10:57.540 align:middle line:84% For example, in order to avoid stock-outs 01:10:57.540 --> 01:11:00.060 align:middle line:84% in the case of unexpected demand, 01:11:00.060 --> 01:11:03.900 align:middle line:84% firms hold buffer inventories, or in order 01:11:03.900 --> 01:11:11.020 align:middle line:84% to be able to make payments in case of shortages in cash flow, 01:11:11.020 --> 01:11:14.980 align:middle line:84% they hold liquid assets, and so on and so forth. 01:11:14.980 --> 01:11:19.220 align:middle line:84% My hypothesis is that there are important feedback loops running 01:11:19.220 --> 01:11:22.700 align:middle line:84% through sentiments and buffer stocks. 01:11:22.700 --> 01:11:26.060 align:middle line:84% And these will generate macroeconomic fluctuations 01:11:26.060 --> 01:11:28.580 align:middle line:90% at a business cycle frequency. 01:11:28.580 --> 01:11:33.900 align:middle line:84% And one of them runs through these sentiments and decisions 01:11:33.900 --> 01:11:37.620 align:middle line:90% of the real sector. 01:11:37.620 --> 01:11:42.620 align:middle line:84% My goal is to show that stability is destabilizing, 01:11:42.620 --> 01:11:47.600 align:middle line:84% as argued by Minsky, and namely, even if the economy starts 01:11:47.600 --> 01:11:51.440 align:middle line:84% from its equilibrium, psychological factors 01:11:51.440 --> 01:11:55.080 align:middle line:84% would give rise to macroeconomic fluctuations in the presence 01:11:55.080 --> 01:11:59.720 align:middle line:84% of uncertainty, and the cycles would be amplified 01:11:59.720 --> 01:12:04.000 align:middle line:84% or dampened by the psychological parameters. 01:12:04.000 --> 01:12:07.760 align:middle line:84% So I tried to build this model in order 01:12:07.760 --> 01:12:12.680 align:middle line:84% to depict how the behavior of the real sector 01:12:12.680 --> 01:12:16.440 align:middle line:84% affects these cycles, or create these cycles. 01:12:16.440 --> 01:12:21.400 align:middle line:84% So the focus on the real sector, and therefore I 01:12:21.400 --> 01:12:25.240 align:middle line:84% simplified the behavior of the household and the [? banks ?] 01:12:25.240 --> 01:12:26.320 align:middle line:90% in the sector. 01:12:26.320 --> 01:12:29.900 align:middle line:84% And I assume the way the price mechanism. 01:12:29.900 --> 01:12:36.120 align:middle line:84% So all the adjustments is through the quantity adjustment. 01:12:36.120 --> 01:12:37.260 align:middle line:90% I don't know if-- 01:12:37.260 --> 01:12:38.093 align:middle line:90% [INTERPOSING VOICES] 01:12:38.093 --> 01:12:40.040 align:middle line:84% HAZHIR RAHMANDAD: That's very helpful. 01:12:40.040 --> 01:12:43.260 align:middle line:84% So if I understand correctly, basically you're saying, 01:12:43.260 --> 01:12:45.460 align:middle line:84% depending on the sentiment, you are 01:12:45.460 --> 01:12:48.040 align:middle line:84% changing your buffer stocks, your inventories, 01:12:48.040 --> 01:12:50.900 align:middle line:84% think about it, but also your cash in your bank 01:12:50.900 --> 01:12:52.593 align:middle line:90% account that you desire. 01:12:52.593 --> 01:12:53.260 align:middle line:90% AUDIENCE: Right. 01:12:53.260 --> 01:12:55.180 align:middle line:84% HAZHIR RAHMANDAD: And those changes 01:12:55.180 --> 01:12:59.820 align:middle line:84% have real impacts in the supply, and in the demand side 01:12:59.820 --> 01:13:05.860 align:middle line:84% primarily, which then creates more instability, volatility 01:13:05.860 --> 01:13:11.540 align:middle line:84% downstream, which leads to potential need for even 01:13:11.540 --> 01:13:14.020 align:middle line:90% more buffer stock. 01:13:14.020 --> 01:13:19.660 align:middle line:84% But at some point that becomes too much buffer stock 01:13:19.660 --> 01:13:24.060 align:middle line:84% or it becomes too much demand, because now you 01:13:24.060 --> 01:13:29.100 align:middle line:84% have built your buffer stock and then the demand goes down. 01:13:29.100 --> 01:13:34.420 align:middle line:84% So that becomes the trigger for the endogenous oscillations 01:13:34.420 --> 01:13:36.710 align:middle line:90% in the real economy. 01:13:36.710 --> 01:13:37.460 align:middle line:90% AUDIENCE: Exactly. 01:13:37.460 --> 01:13:38.335 align:middle line:90% HAZHIR RAHMANDAD: OK. 01:13:38.335 --> 01:13:39.680 align:middle line:90% AUDIENCE: So yes. 01:13:39.680 --> 01:13:42.800 align:middle line:84% And the finished good inventories 01:13:42.800 --> 01:13:44.220 align:middle line:90% is just one side of it. 01:13:44.220 --> 01:13:47.680 align:middle line:84% And apart from that, in the stabilized balance 01:13:47.680 --> 01:13:51.640 align:middle line:84% sheet of the firm, we had capital stock, inventories, 01:13:51.640 --> 01:13:54.440 align:middle line:84% liquidity assets on the asset side 01:13:54.440 --> 01:13:58.243 align:middle line:84% and loans and equity at the liability side. 01:13:58.243 --> 01:13:59.160 align:middle line:90% HAZHIR RAHMANDAD: Yes. 01:13:59.160 --> 01:14:03.720 align:middle line:84% AUDIENCE: I assume that if there were no uncertainty, 01:14:03.720 --> 01:14:07.320 align:middle line:84% the levels of these stocks would be equal to their inflows 01:14:07.320 --> 01:14:11.480 align:middle line:84% and outflows, and the firms would not 01:14:11.480 --> 01:14:14.240 align:middle line:84% hold any excess inventory as it is assumed 01:14:14.240 --> 01:14:15.580 align:middle line:90% in mainstream economics. 01:14:15.580 --> 01:14:18.960 align:middle line:84% But in order to tackle with uncertainty, 01:14:18.960 --> 01:14:21.320 align:middle line:90% firms do hold excess inventory. 01:14:21.320 --> 01:14:22.560 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 01:14:22.560 --> 01:14:26.320 align:middle line:84% AUDIENCE: And the excess, the normal level 01:14:26.320 --> 01:14:30.740 align:middle line:84% of these excess inventories, changes with the sentiment. 01:14:30.740 --> 01:14:34.320 align:middle line:84% So I also endogenize the sentiments 01:14:34.320 --> 01:14:41.420 align:middle line:84% which has a range between 0 and 2, with a normal level of 1. 01:14:41.420 --> 01:14:45.380 align:middle line:84% The normal level of excess inventories 01:14:45.380 --> 01:14:50.500 align:middle line:84% is equal to their level when sentiment is equal to 1. 01:14:50.500 --> 01:14:52.940 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 01:14:52.940 --> 01:14:53.960 align:middle line:90% Makes sense. 01:14:53.960 --> 01:14:58.660 align:middle line:84% So big picture, I think that's an interesting problem. 01:14:58.660 --> 01:15:02.280 align:middle line:84% Now, outside of the relevant literature, 01:15:02.280 --> 01:15:07.300 align:middle line:84% a lot of people have built models of instability 01:15:07.300 --> 01:15:13.220 align:middle line:84% in an economy, I think kind of old work in system dynamics 01:15:13.220 --> 01:15:18.400 align:middle line:84% in the '70s and early '80s has been focused on some of this. 01:15:18.400 --> 01:15:25.580 align:middle line:84% I don't remember having heard about the kind of sentiment 01:15:25.580 --> 01:15:33.340 align:middle line:84% based on the instability, based on the deviances of the demand 01:15:33.340 --> 01:15:35.540 align:middle line:84% growth rates, so I think that this 01:15:35.540 --> 01:15:38.680 align:middle line:84% is a new channel for that literature, 01:15:38.680 --> 01:15:41.400 align:middle line:84% but I don't know enough in the econ literature 01:15:41.400 --> 01:15:44.160 align:middle line:84% and how these are modeled and treated 01:15:44.160 --> 01:15:49.080 align:middle line:84% to have a meaningful advice on what 01:15:49.080 --> 01:15:53.920 align:middle line:84% is the right level of aggregation or what to include 01:15:53.920 --> 01:15:56.840 align:middle line:84% and what to exclude, so that your contribution is 01:15:56.840 --> 01:15:59.360 align:middle line:90% clear and novel. 01:15:59.360 --> 01:16:04.260 align:middle line:84% For an uneducated person in this space, 01:16:04.260 --> 01:16:08.760 align:middle line:84% like myself, I would say, focusing on the simplest 01:16:08.760 --> 01:16:13.520 align:middle line:84% structures that gets you the core idea-- allows 01:16:13.520 --> 01:16:17.960 align:middle line:84% you to show the core idea would be the right choice. 01:16:17.960 --> 01:16:23.040 align:middle line:84% So I'm not sure if you need to include all of those buffer 01:16:23.040 --> 01:16:28.920 align:middle line:84% stocks going over the capital, stock of the companies, 01:16:28.920 --> 01:16:33.420 align:middle line:84% and their balance sheets and earnings and so on. 01:16:33.420 --> 01:16:36.780 align:middle line:84% I would say, just pick, for example, the inventory 01:16:36.780 --> 01:16:42.500 align:middle line:84% and see how you can just capture the core effect 01:16:42.500 --> 01:16:45.800 align:middle line:90% within that inventory sector. 01:16:45.800 --> 01:16:49.620 align:middle line:84% But that might not be good enough 01:16:49.620 --> 01:16:53.820 align:middle line:84% or the state of the art for an econ audience 01:16:53.820 --> 01:17:01.020 align:middle line:84% who may say, oh, the inventory kind of adjustment pathway 01:17:01.020 --> 01:17:03.840 align:middle line:84% should be complemented with these other pathways. 01:17:03.840 --> 01:17:06.460 align:middle line:84% Otherwise, what you see for the inventory does not 01:17:06.460 --> 01:17:07.700 align:middle line:90% make much sense. 01:17:07.700 --> 01:17:12.300 align:middle line:84% So I don't know that, what is the right choice there. 01:17:12.300 --> 01:17:15.880 align:middle line:84% So with that caveat in mind I can see two different ways. 01:17:15.880 --> 01:17:19.905 align:middle line:84% One is let's stick to what you have here, 01:17:19.905 --> 01:17:21.780 align:middle line:84% because you have been looking at this problem 01:17:21.780 --> 01:17:24.420 align:middle line:84% and thinking about it, and you tell us 01:17:24.420 --> 01:17:27.560 align:middle line:84% where we can provide more meaningful feedback, 01:17:27.560 --> 01:17:30.660 align:middle line:84% because I don't think we have the time and also 01:17:30.660 --> 01:17:32.800 align:middle line:84% the expertise really to provide feedback 01:17:32.800 --> 01:17:35.520 align:middle line:90% on everything you have here. 01:17:35.520 --> 01:17:38.720 align:middle line:84% So that's one way we can move ahead. 01:17:38.720 --> 01:17:41.440 align:middle line:84% The alternative is to think together 01:17:41.440 --> 01:17:46.120 align:middle line:84% to more what would be a minimalistic model that 01:17:46.120 --> 01:17:50.200 align:middle line:84% can capture this instability without a lot 01:17:50.200 --> 01:17:52.362 align:middle line:84% of the additional structures you have. 01:17:52.362 --> 01:17:54.820 align:middle line:84% So that would be a little bit more conceptual conversation. 01:17:54.820 --> 01:17:56.680 align:middle line:90% So what is your choice? 01:17:56.680 --> 01:17:58.498 align:middle line:90% We can go with either of those. 01:17:58.498 --> 01:18:00.040 align:middle line:84% AUDIENCE: I would appreciate feedback 01:18:00.040 --> 01:18:06.220 align:middle line:84% about how I can formulate the endogenous sentiment. 01:18:06.220 --> 01:18:13.080 align:middle line:84% So here in this model I thought that it 01:18:13.080 --> 01:18:17.760 align:middle line:84% is a function of the growth rate, the average percent growth 01:18:17.760 --> 01:18:19.963 align:middle line:90% rate of the aggregate demand. 01:18:19.963 --> 01:18:20.880 align:middle line:90% HAZHIR RAHMANDAD: Yes. 01:18:20.880 --> 01:18:23.005 align:middle line:84% AUDIENCE: And it's supposed to be function of that. 01:18:23.005 --> 01:18:29.000 align:middle line:84% And it is a negative function of the volatility of the growth 01:18:29.000 --> 01:18:30.220 align:middle line:90% rate of the economy. 01:18:30.220 --> 01:18:33.020 align:middle line:90% HAZHIR RAHMANDAD: Yes. 01:18:33.020 --> 01:18:34.300 align:middle line:90% Got it. 01:18:34.300 --> 01:18:36.380 align:middle line:90% So one question. 01:18:36.380 --> 01:18:40.080 align:middle line:84% The sentiment index, do you have a more clear definition for it? 01:18:40.080 --> 01:18:42.400 align:middle line:84% This is kind of the Michigan Survey of Households 01:18:42.400 --> 01:18:44.440 align:middle line:84% and what's their sentiment about the economy? 01:18:44.440 --> 01:18:47.140 align:middle line:84% Is that what is kind of the idea? 01:18:47.140 --> 01:18:48.340 align:middle line:90% AUDIENCE: Yes, exactly. 01:18:48.340 --> 01:18:52.005 align:middle line:84% Or the Confidence index sector confidence. 01:18:52.005 --> 01:18:52.880 align:middle line:90% HAZHIR RAHMANDAD: OK. 01:18:52.880 --> 01:18:54.600 align:middle line:84% AUDIENCE: How optimistic or pessimistic-- 01:18:54.600 --> 01:18:55.060 align:middle line:90% [INTERPOSING VOICES] 01:18:55.060 --> 01:18:55.980 align:middle line:84% HAZHIR RAHMANDAD: Pessimistic that, yeah, 01:18:55.980 --> 01:18:56.860 align:middle line:90% the business climate. 01:18:56.860 --> 01:18:58.380 align:middle line:84% AUDIENCE: --picture of the economy. 01:18:58.380 --> 01:18:59.255 align:middle line:90% HAZHIR RAHMANDAD: OK. 01:18:59.255 --> 01:19:03.380 align:middle line:84% So I assume empirically these two factors that you have, 01:19:03.380 --> 01:19:06.780 align:middle line:84% that is the recent growth rate and the volatility 01:19:06.780 --> 01:19:12.860 align:middle line:84% in the recent months are the primary candidates 01:19:12.860 --> 01:19:14.160 align:middle line:90% for driving this effect. 01:19:14.160 --> 01:19:16.020 align:middle line:90% Is that a correct understanding? 01:19:16.020 --> 01:19:16.785 align:middle line:90% AUDIENCE: Yes. 01:19:16.785 --> 01:19:17.660 align:middle line:90% HAZHIR RAHMANDAD: OK. 01:19:17.660 --> 01:19:21.860 align:middle line:84% So then, big picture, you're doing the right thing. 01:19:21.860 --> 01:19:25.860 align:middle line:84% Now what would be the right formulations here, 01:19:25.860 --> 01:19:30.810 align:middle line:84% I would certainly go into the econ literature that 01:19:30.810 --> 01:19:34.570 align:middle line:84% looks at these and see what functional forms they 01:19:34.570 --> 01:19:40.810 align:middle line:84% use for estimating sentiment index based 01:19:40.810 --> 01:19:46.010 align:middle line:90% on growth rates and volatility. 01:19:46.010 --> 01:19:50.530 align:middle line:84% So volatility, typically, they use something 01:19:50.530 --> 01:19:56.250 align:middle line:84% like a standard deviation variance in whatever 01:19:56.250 --> 01:19:57.687 align:middle line:90% the underlying factor is. 01:19:57.687 --> 01:19:59.770 align:middle line:84% So if this is the growth rate they are looking at, 01:19:59.770 --> 01:20:03.450 align:middle line:84% they might be looking at the variance in the growth rate. 01:20:03.450 --> 01:20:05.470 align:middle line:90% So that is pretty common. 01:20:05.470 --> 01:20:13.570 align:middle line:84% And then I suspect they would be doing regressions 01:20:13.570 --> 01:20:16.610 align:middle line:84% on the empirical side for estimating sentiment 01:20:16.610 --> 01:20:19.490 align:middle line:90% as a function of these factors. 01:20:19.490 --> 01:20:22.150 align:middle line:84% So regressions are typically a linear function. 01:20:22.150 --> 01:20:25.370 align:middle line:84% So whether linear function is the best thing you can do here 01:20:25.370 --> 01:20:30.550 align:middle line:84% or something more bounded might be useful 01:20:30.550 --> 01:20:32.990 align:middle line:84% because you are trying to keep the sentiment 01:20:32.990 --> 01:20:35.150 align:middle line:90% index between 0 and 2. 01:20:35.150 --> 01:20:38.590 align:middle line:90% That's a little bit open to me. 01:20:38.590 --> 01:20:41.950 align:middle line:90% You could go either way. 01:20:41.950 --> 01:20:45.390 align:middle line:84% And then some weighted average of the effect 01:20:45.390 --> 01:20:49.830 align:middle line:84% of volatility and effect of growth 01:20:49.830 --> 01:20:53.530 align:middle line:84% is perhaps a good mix, a good way of doing it. 01:20:53.530 --> 01:20:56.750 align:middle line:84% So right now you have a lot of things 01:20:56.750 --> 01:20:59.890 align:middle line:84% going into this equation for the flow of sentiment index. 01:20:59.890 --> 01:21:05.690 align:middle line:84% So that's makes me a little bit worried in understanding it. 01:21:05.690 --> 01:21:11.410 align:middle line:84% I'm not saying it is wrong, I just, yeah. 01:21:11.410 --> 01:21:12.630 align:middle line:90% You see? 01:21:12.630 --> 01:21:14.670 align:middle line:84% AUDIENCE: So basically I'm formulating it 01:21:14.670 --> 01:21:20.648 align:middle line:84% as a logistic function with ranges between 0 and 2. 01:21:20.648 --> 01:21:21.690 align:middle line:90% HAZHIR RAHMANDAD: Got it. 01:21:21.690 --> 01:21:29.410 align:middle line:84% AUDIENCE: And average growth rate affects positively, 01:21:29.410 --> 01:21:36.730 align:middle line:84% and the volatility or the variance of the aggregate demand 01:21:36.730 --> 01:21:39.030 align:middle line:90% has a negative effect on that. 01:21:39.030 --> 01:21:40.170 align:middle line:90% HAZHIR RAHMANDAD: Got it. 01:21:40.170 --> 01:21:43.670 align:middle line:84% AUDIENCE: And in order to give that negative effect, 01:21:43.670 --> 01:21:46.930 align:middle line:84% I assume that there is a minimum variance, 01:21:46.930 --> 01:21:51.493 align:middle line:84% normal variance, that people or [INAUDIBLE] as normal. 01:21:51.493 --> 01:21:52.410 align:middle line:90% HAZHIR RAHMANDAD: Yes. 01:21:52.410 --> 01:21:56.890 align:middle line:84% AUDIENCE: So if it is below that normal level, 01:21:56.890 --> 01:21:59.110 align:middle line:90% sentiment index increases. 01:21:59.110 --> 01:22:04.010 align:middle line:84% And if it is above that level the sentiment index decreases. 01:22:04.010 --> 01:22:07.290 align:middle line:84% But I don't know if this formulation is too ad hoc. 01:22:07.290 --> 01:22:08.210 align:middle line:90% It works, but-- 01:22:08.210 --> 01:22:09.043 align:middle line:90% [INTERPOSING VOICES] 01:22:09.043 --> 01:22:11.250 align:middle line:84% HAZHIR RAHMANDAD: So I think, yeah. 01:22:11.250 --> 01:22:15.770 align:middle line:84% So very simple feedback here is that you definitely 01:22:15.770 --> 01:22:19.650 align:middle line:84% just need to create a indicated sentiment index 01:22:19.650 --> 01:22:21.930 align:middle line:84% and then adjust towards that with the time delay 01:22:21.930 --> 01:22:25.070 align:middle line:84% rather than this convoluted formulation that 01:22:25.070 --> 01:22:27.990 align:middle line:84% throws everything into the formulation for the stock. 01:22:27.990 --> 01:22:33.790 align:middle line:84% So that's just housekeeping norms that you should follow. 01:22:33.790 --> 01:22:36.710 align:middle line:84% Beyond that, the logistic function 01:22:36.710 --> 01:22:41.730 align:middle line:84% does make sense-- at its surface is sensible to me. 01:22:41.730 --> 01:22:45.410 align:middle line:84% It's not something I have looked into the literature for. 01:22:45.410 --> 01:22:47.910 align:middle line:84% So I don't know what are the alternatives people would 01:22:47.910 --> 01:22:50.690 align:middle line:90% have potentially recommended. 01:22:50.690 --> 01:22:52.170 align:middle line:90% But it is a robust function. 01:22:52.170 --> 01:22:55.590 align:middle line:84% So it works OK, keeps you in the bounds. 01:22:55.590 --> 01:22:57.190 align:middle line:90% So that's fine. 01:22:57.190 --> 01:22:59.950 align:middle line:84% Where I'm a little bit less clear 01:22:59.950 --> 01:23:06.830 align:middle line:84% is this average square deviation of demand growth rate. 01:23:06.830 --> 01:23:12.830 align:middle line:84% So where does variance come from in this model? 01:23:12.830 --> 01:23:15.930 align:middle line:90% So does it simulate right now? 01:23:15.930 --> 01:23:16.430 align:middle line:90% Yeah. 01:23:16.430 --> 01:23:19.630 align:middle line:84% So right now, if you simulate it, 01:23:19.630 --> 01:23:22.110 align:middle line:84% you're starting in some kind of equilibrium, 01:23:22.110 --> 01:23:26.170 align:middle line:90% which is generally a good idea. 01:23:26.170 --> 01:23:29.490 align:middle line:90% But then that means-- 01:23:29.490 --> 01:23:33.170 align:middle line:84% so but how do you trigger any behavior 01:23:33.170 --> 01:23:35.750 align:middle line:90% that is non-equilibrium? 01:23:35.750 --> 01:23:39.930 align:middle line:84% AUDIENCE: So there is the uncertainty switch 01:23:39.930 --> 01:23:43.550 align:middle line:84% to the bottom right here on the screen. 01:23:43.550 --> 01:23:45.730 align:middle line:90% No, sorry. 01:23:45.730 --> 01:23:48.730 align:middle line:90% It is a [INAUDIBLE] now. 01:23:48.730 --> 01:23:50.810 align:middle line:90% HAZHIR RAHMANDAD: Yep. 01:23:50.810 --> 01:23:51.310 align:middle line:90% I see. 01:23:51.310 --> 01:23:53.290 align:middle line:90% So if we turn it on-- 01:23:53.290 --> 01:23:55.210 align:middle line:90% AUDIENCE: Yes. 01:23:55.210 --> 01:23:56.393 align:middle line:90% And-- 01:23:56.393 --> 01:23:57.310 align:middle line:90% HAZHIR RAHMANDAD: Yes. 01:23:57.310 --> 01:24:01.450 align:middle line:84% AUDIENCE: --value of 1 creates answer. 01:24:01.450 --> 01:24:06.910 align:middle line:84% HAZHIR RAHMANDAD: Yeah, you can see oscillations coming from-- 01:24:06.910 --> 01:24:11.450 align:middle line:84% so what does the uncertainty switch do? 01:24:11.450 --> 01:24:13.570 align:middle line:90% What is it that it triggers? 01:24:13.570 --> 01:24:17.210 align:middle line:84% It triggers some excess capacity. 01:24:17.210 --> 01:24:18.090 align:middle line:90% AUDIENCE: Yes. 01:24:18.090 --> 01:24:19.150 align:middle line:90% HAZHIR RAHMANDAD: OK. 01:24:19.150 --> 01:24:23.550 align:middle line:84% And then that changes the capital stock. 01:24:23.550 --> 01:24:28.750 align:middle line:84% And then through that, everything starts to oscillate. 01:24:28.750 --> 01:24:36.610 align:middle line:84% And the oscillations create this standard deviation part. 01:24:36.610 --> 01:24:38.590 align:middle line:84% Now the calculation of standard deviation, 01:24:38.590 --> 01:24:42.190 align:middle line:84% I don't know how you have done it. 01:24:42.190 --> 01:24:44.070 align:middle line:84% Are you comfortable-- confident with how 01:24:44.070 --> 01:24:48.550 align:middle line:84% it is calculated in your model because that's not trivial? 01:24:48.550 --> 01:24:49.530 align:middle line:90% AUDIENCE: I'm not sure. 01:24:49.530 --> 01:24:49.990 align:middle line:90% [LAUGHS] 01:24:49.990 --> 01:24:50.948 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 01:24:50.948 --> 01:24:54.030 align:middle line:90% 01:24:54.030 --> 01:24:56.650 align:middle line:90% So when you have the mean-- 01:24:56.650 --> 01:24:59.830 align:middle line:84% so to calculate standard deviation 01:24:59.830 --> 01:25:11.710 align:middle line:84% you want to accumulate the second moment that is whatever 01:25:11.710 --> 01:25:17.650 align:middle line:84% you're tracking to the power of 2 01:25:17.650 --> 01:25:20.770 align:middle line:84% and then look at the difference between-- 01:25:20.770 --> 01:25:24.370 align:middle line:84% so the expected value versus expected value 01:25:24.370 --> 01:25:29.790 align:middle line:84% of x to the power of 2, the difference between those two. 01:25:29.790 --> 01:25:32.890 align:middle line:84% So it's expected value of x to the power of 2 01:25:32.890 --> 01:25:37.050 align:middle line:84% minus expected value to the power of 2. 01:25:37.050 --> 01:25:39.910 align:middle line:90% That is driving the variance. 01:25:39.910 --> 01:25:42.530 align:middle line:90% 01:25:42.530 --> 01:25:45.630 align:middle line:84% So yeah, we can take a look at that. 01:25:45.630 --> 01:25:48.910 align:middle line:84% That's nothing complex conceptually. 01:25:48.910 --> 01:25:52.470 align:middle line:84% It's just making sure that the mathematically-- 01:25:52.470 --> 01:25:56.890 align:middle line:90% 01:25:56.890 --> 01:25:59.890 align:middle line:84% so you have a perception adjustment 01:25:59.890 --> 01:26:02.890 align:middle line:84% times square deviance of demand growth 01:26:02.890 --> 01:26:06.490 align:middle line:84% rate minus average squared deviance of demand growth rate. 01:26:06.490 --> 01:26:08.030 align:middle line:90% Yeah, that looks right. 01:26:08.030 --> 01:26:17.750 align:middle line:84% If this average squared deviance is 01:26:17.750 --> 01:26:22.790 align:middle line:84% the difference between-- yeah, I think this is correct. 01:26:22.790 --> 01:26:28.430 align:middle line:84% So this gives you basically variance. 01:26:28.430 --> 01:26:30.630 align:middle line:84% The important thing is that the time constant should 01:26:30.630 --> 01:26:32.170 align:middle line:90% be the same for both of these. 01:26:32.170 --> 01:26:34.270 align:middle line:84% So this perceived average aggregate, yes, 01:26:34.270 --> 01:26:36.470 align:middle line:90% the time constant is the same. 01:26:36.470 --> 01:26:37.790 align:middle line:90% Good. 01:26:37.790 --> 01:26:38.330 align:middle line:90% OK. 01:26:38.330 --> 01:26:43.710 align:middle line:84% So the model seems to be doing something oscillatory. 01:26:43.710 --> 01:26:46.890 align:middle line:84% Now do you know where the oscillations are coming from? 01:26:46.890 --> 01:26:49.610 align:middle line:84% Have you tried to trace the oscillations? 01:26:49.610 --> 01:26:52.698 align:middle line:90% 01:26:52.698 --> 01:26:54.990 align:middle line:84% AUDIENCE: I think there are some feedback loops running 01:26:54.990 --> 01:26:57.450 align:middle line:90% through these sentiments. 01:26:57.450 --> 01:27:05.150 align:middle line:84% But because as the buffer stocks increase, 01:27:05.150 --> 01:27:07.170 align:middle line:90% the aggregate demand increases. 01:27:07.170 --> 01:27:11.350 align:middle line:84% So there are definitely positive feedback loops 01:27:11.350 --> 01:27:13.770 align:middle line:90% running through sentiments. 01:27:13.770 --> 01:27:15.693 align:middle line:84% But it's another thing I would like to ask. 01:27:15.693 --> 01:27:16.610 align:middle line:90% HAZHIR RAHMANDAD: Yes. 01:27:16.610 --> 01:27:16.850 align:middle line:90% [INTERPOSING VOICES] 01:27:16.850 --> 01:27:18.710 align:middle line:84% AUDIENCE: So there must be something, 01:27:18.710 --> 01:27:22.370 align:middle line:84% some negative feedback loops running also. 01:27:22.370 --> 01:27:26.930 align:middle line:84% But I cannot define any at the moment because I don't know how 01:27:26.930 --> 01:27:34.090 align:middle line:84% to give the direction to the causal relations between 01:27:34.090 --> 01:27:34.937 align:middle line:90% volatility and-- 01:27:34.937 --> 01:27:35.770 align:middle line:90% [INTERPOSING VOICES] 01:27:35.770 --> 01:27:38.590 align:middle line:84% HAZHIR RAHMANDAD: Yes, so that's exactly what I was asking. 01:27:38.590 --> 01:27:42.090 align:middle line:84% That is, you really need to understand what is 01:27:42.090 --> 01:27:46.090 align:middle line:90% driving your oscillatory mode. 01:27:46.090 --> 01:27:49.630 align:middle line:84% So right now I select sentiment index, for example, 01:27:49.630 --> 01:27:52.130 align:middle line:90% you can do the causal strips. 01:27:52.130 --> 01:27:55.090 align:middle line:84% So which of the drivers of sentiment index 01:27:55.090 --> 01:27:58.370 align:middle line:84% are likely to be driving its behavior? 01:27:58.370 --> 01:28:02.610 align:middle line:84% It's pretty clearly effective growth. 01:28:02.610 --> 01:28:04.630 align:middle line:84% And then you can look at effective growth. 01:28:04.630 --> 01:28:07.090 align:middle line:84% What is driving oscillations in effective growth, 01:28:07.090 --> 01:28:11.950 align:middle line:84% it's the expected perceived growth. 01:28:11.950 --> 01:28:18.390 align:middle line:84% And then that one is perceived average aggregate demand growth. 01:28:18.390 --> 01:28:21.370 align:middle line:84% That is a little bit more tricky. 01:28:21.370 --> 01:28:24.310 align:middle line:90% 01:28:24.310 --> 01:28:26.910 align:middle line:90% Let's see. 01:28:26.910 --> 01:28:33.190 align:middle line:84% Real aggregate demand, average aggregate demand growth rate, 01:28:33.190 --> 01:28:37.870 align:middle line:84% and it is coming from aggregate demand. 01:28:37.870 --> 01:28:40.870 align:middle line:84% Because this perceived average real aggregate demand 01:28:40.870 --> 01:28:44.690 align:middle line:84% is just a delay of this guy if I understand it correctly. 01:28:44.690 --> 01:28:48.167 align:middle line:84% So that would also come from the real aggregate demand. 01:28:48.167 --> 01:28:48.750 align:middle line:90% AUDIENCE: Yes. 01:28:48.750 --> 01:28:50.708 align:middle line:84% HAZHIR RAHMANDAD: So then real aggregate demand 01:28:50.708 --> 01:28:52.690 align:middle line:84% is coming from nominal aggregate demand. 01:28:52.690 --> 01:28:56.210 align:middle line:84% Nominal aggregate demand is coming from-- 01:28:56.210 --> 01:28:59.270 align:middle line:90% 01:28:59.270 --> 01:29:02.010 align:middle line:84% these two different things could be driving it. 01:29:02.010 --> 01:29:04.390 align:middle line:84% But you look at the scale of these, 01:29:04.390 --> 01:29:07.950 align:middle line:84% by far the household desired consumption is the biggest 01:29:07.950 --> 01:29:08.650 align:middle line:90% driver. 01:29:08.650 --> 01:29:11.180 align:middle line:84% So this planned investment expenditure 01:29:11.180 --> 01:29:13.300 align:middle line:90% is tiny compared to that. 01:29:13.300 --> 01:29:20.940 align:middle line:84% So the household is coming from nominal budget 01:29:20.940 --> 01:29:26.220 align:middle line:84% for living, that's coming from a standard of living, that's 01:29:26.220 --> 01:29:30.780 align:middle line:84% coming from household disposable income, that's 01:29:30.780 --> 01:29:34.180 align:middle line:90% coming from household income. 01:29:34.180 --> 01:29:39.600 align:middle line:84% Now household income is largely wages. 01:29:39.600 --> 01:29:45.086 align:middle line:84% Dividends matter, but not as big of a deal. 01:29:45.086 --> 01:29:47.440 align:middle line:84% Interest income is not a big deal. 01:29:47.440 --> 01:29:52.220 align:middle line:84% So wage is really what is driving the income. 01:29:52.220 --> 01:29:55.580 align:middle line:90% Wages coming from labor demand. 01:29:55.580 --> 01:29:58.320 align:middle line:84% Labor demand is coming from production. 01:29:58.320 --> 01:30:01.500 align:middle line:90% 01:30:01.500 --> 01:30:05.200 align:middle line:84% Production is coming from which of these two? 01:30:05.200 --> 01:30:09.520 align:middle line:84% So this is a little bit less clear. 01:30:09.520 --> 01:30:19.120 align:middle line:84% It's desired production versus production capacity. 01:30:19.120 --> 01:30:24.520 align:middle line:84% I think production capacity is also tied to desired production? 01:30:24.520 --> 01:30:29.467 align:middle line:84% AUDIENCE: So it defines the ceiling for product. 01:30:29.467 --> 01:30:30.300 align:middle line:90% [INTERPOSING VOICES] 01:30:30.300 --> 01:30:31.592 align:middle line:90% HAZHIR RAHMANDAD: Yes, exactly. 01:30:31.592 --> 01:30:34.980 align:middle line:84% AUDIENCE: At the short term, capital stock is fixed. 01:30:34.980 --> 01:30:39.682 align:middle line:84% So at most, that must-- or might be produced. 01:30:39.682 --> 01:30:40.640 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 01:30:40.640 --> 01:30:43.980 align:middle line:84% So OK, I'm going with desired production. 01:30:43.980 --> 01:30:46.000 align:middle line:84% This is one of the places that I'm not sure 01:30:46.000 --> 01:30:51.680 align:middle line:84% whether I'm tracing the right driver. 01:30:51.680 --> 01:30:58.480 align:middle line:84% Then desired production based on the average real aggregate 01:30:58.480 --> 01:30:59.460 align:middle line:90% demand. 01:30:59.460 --> 01:31:03.400 align:middle line:90% 01:31:03.400 --> 01:31:08.420 align:middle line:84% And desired inventory-- so I'm going 01:31:08.420 --> 01:31:10.280 align:middle line:84% with the average aggregate demand, 01:31:10.280 --> 01:31:13.220 align:middle line:84% even though the other one might be relevant as well. 01:31:13.220 --> 01:31:16.840 align:middle line:84% And this goes to real aggregate demand, 01:31:16.840 --> 01:31:23.467 align:middle line:84% which goes to nominal aggregate demand, which goes to-- 01:31:23.467 --> 01:31:24.300 align:middle line:90% [INTERPOSING VOICES] 01:31:24.300 --> 01:31:26.425 align:middle line:84% HAZHIR RAHMANDAD: We are now back to the household. 01:31:26.425 --> 01:31:32.260 align:middle line:84% So basically that feedback loop that we identified 01:31:32.260 --> 01:31:37.860 align:middle line:84% as the candidate goes through the household demand, which 01:31:37.860 --> 01:31:39.640 align:middle line:90% comes from household income. 01:31:39.640 --> 01:31:42.280 align:middle line:84% Household income comes from household wages. 01:31:42.280 --> 01:31:44.900 align:middle line:84% Household wages comes from the labor. 01:31:44.900 --> 01:31:47.180 align:middle line:90% Labor comes from the production. 01:31:47.180 --> 01:31:50.820 align:middle line:84% Desired production comes from household demand ultimately. 01:31:50.820 --> 01:31:53.020 align:middle line:90% So that's the feedback loop. 01:31:53.020 --> 01:31:59.780 align:middle line:84% That is the hypothesis for what is driving it. 01:31:59.780 --> 01:32:02.860 align:middle line:84% Now we need to cut that feedback loop 01:32:02.860 --> 01:32:06.160 align:middle line:84% somewhere to see whether that changes 01:32:06.160 --> 01:32:08.360 align:middle line:90% the behavior significantly. 01:32:08.360 --> 01:32:12.800 align:middle line:90% So I could look at the-- 01:32:12.800 --> 01:32:14.880 align:middle line:90% somewhere about households. 01:32:14.880 --> 01:32:19.586 align:middle line:84% AUDIENCE: And households is in the second view. 01:32:19.586 --> 01:32:24.920 align:middle line:84% HAZHIR RAHMANDAD: So we have consumption demand, for example, 01:32:24.920 --> 01:32:25.700 align:middle line:90% here. 01:32:25.700 --> 01:32:29.760 align:middle line:90% 01:32:29.760 --> 01:32:34.160 align:middle line:84% And household desired consumption. 01:32:34.160 --> 01:32:37.410 align:middle line:84% And then consumption demand is feeding back somewhere? 01:32:37.410 --> 01:32:39.160 align:middle line:84% AUDIENCE: No, no, I think it is redundant. 01:32:39.160 --> 01:32:40.020 align:middle line:90% I forgot that. 01:32:40.020 --> 01:32:46.360 align:middle line:84% HAZHIR RAHMANDAD: So let's say household desired consumption, 01:32:46.360 --> 01:32:49.600 align:middle line:84% we just override the default behavior. 01:32:49.600 --> 01:32:57.240 align:middle line:84% And we just say, let's make it constant. 01:32:57.240 --> 01:33:03.580 align:middle line:84% So now I just made it constant at its base value. 01:33:03.580 --> 01:33:07.560 align:middle line:84% And now we can see whether anything else oscillates. 01:33:07.560 --> 01:33:09.000 align:middle line:90% Nothing else oscillates. 01:33:09.000 --> 01:33:12.100 align:middle line:84% So that hypothesis gets an additional boost 01:33:12.100 --> 01:33:13.740 align:middle line:90% from this experiment. 01:33:13.740 --> 01:33:15.700 align:middle line:84% I just wanted to go through this because that's 01:33:15.700 --> 01:33:18.940 align:middle line:84% a process that you need to all follow 01:33:18.940 --> 01:33:24.500 align:middle line:84% and do to be able to identify what is driving your model's 01:33:24.500 --> 01:33:25.620 align:middle line:90% behavior. 01:33:25.620 --> 01:33:28.740 align:middle line:84% And in this case, it doesn't tell me 01:33:28.740 --> 01:33:31.580 align:middle line:84% that the hypothesis is absolutely correct because I 01:33:31.580 --> 01:33:35.940 align:middle line:84% didn't pick a link that is unique to that loop and nothing 01:33:35.940 --> 01:33:36.440 align:middle line:90% else. 01:33:36.440 --> 01:33:40.080 align:middle line:84% I just went into the loop and picked something, 01:33:40.080 --> 01:33:42.140 align:middle line:84% and that something might be overlapping 01:33:42.140 --> 01:33:46.320 align:middle line:84% with another loop that is important for this dynamic. 01:33:46.320 --> 01:33:49.100 align:middle line:84% But it gives me confidence that the household sector 01:33:49.100 --> 01:33:50.500 align:middle line:90% is really important. 01:33:50.500 --> 01:33:53.640 align:middle line:84% And now if I don't want to use this-- 01:33:53.640 --> 01:33:58.020 align:middle line:90% 01:33:58.020 --> 01:33:58.860 align:middle line:90% let me see. 01:33:58.860 --> 01:34:01.520 align:middle line:90% Reset all overrides. 01:34:01.520 --> 01:34:07.220 align:middle line:84% Yeah, if I don't want to focus only on this one. 01:34:07.220 --> 01:34:10.640 align:middle line:84% I could go somewhere else, like where the households get 01:34:10.640 --> 01:34:13.220 align:middle line:90% their wages, and test that. 01:34:13.220 --> 01:34:15.520 align:middle line:84% So if you test different parts of the loop 01:34:15.520 --> 01:34:18.520 align:middle line:84% and you get the same result, that gives you more confidence 01:34:18.520 --> 01:34:21.920 align:middle line:84% that this is the loop that is generating the behavior. 01:34:21.920 --> 01:34:25.840 align:middle line:84% OK, I think you all get the idea. 01:34:25.840 --> 01:34:28.800 align:middle line:84% This is really, really important. 01:34:28.800 --> 01:34:31.720 align:middle line:84% If you have a model, it is generating a behavior 01:34:31.720 --> 01:34:36.000 align:middle line:84% that you think is relevant, potentially interesting. 01:34:36.000 --> 01:34:39.000 align:middle line:84% You need to understand where that behavior comes from. 01:34:39.000 --> 01:34:42.960 align:middle line:84% Now the reason in this specific case it is super important, 01:34:42.960 --> 01:34:50.760 align:middle line:84% is that your whole hypothesis was about basically volatility 01:34:50.760 --> 01:34:54.560 align:middle line:90% driving these dynamics. 01:34:54.560 --> 01:34:57.800 align:middle line:84% And right now volatility doesn't play 01:34:57.800 --> 01:35:00.300 align:middle line:90% a role in your primary loop. 01:35:00.300 --> 01:35:07.420 align:middle line:84% In fact, my hypothesis is that if we disable the volatility 01:35:07.420 --> 01:35:11.380 align:middle line:90% side of the dynamics-- 01:35:11.380 --> 01:35:15.020 align:middle line:84% so if we turn on this switch back here-- 01:35:15.020 --> 01:35:17.220 align:middle line:90% where was the switch? 01:35:17.220 --> 01:35:19.020 align:middle line:84% AUDIENCE: And also [INAUDIBLE] switch. 01:35:19.020 --> 01:35:19.540 align:middle line:90% HAZHIR RAHMANDAD: Yes. 01:35:19.540 --> 01:35:20.248 align:middle line:90% AUDIENCE: Purple. 01:35:20.248 --> 01:35:21.600 align:middle line:90% HAZHIR RAHMANDAD: Here, I think. 01:35:21.600 --> 01:35:24.500 align:middle line:90% 01:35:24.500 --> 01:35:27.700 align:middle line:84% So now we have this pathway going. 01:35:27.700 --> 01:35:32.020 align:middle line:84% If I just take this demand growth 01:35:32.020 --> 01:35:40.900 align:middle line:84% rate standard deviation and just say, ah, make it a constant, 01:35:40.900 --> 01:35:41.540 align:middle line:90% you see? 01:35:41.540 --> 01:35:43.700 align:middle line:90% I cut this pathway. 01:35:43.700 --> 01:35:47.220 align:middle line:90% The dynamics are there. 01:35:47.220 --> 01:35:53.580 align:middle line:84% It is not stopping your structure. 01:35:53.580 --> 01:36:00.240 align:middle line:84% In fact, it's not even obvious is if the effect of growth 01:36:00.240 --> 01:36:04.800 align:middle line:84% will be also critical, although this 01:36:04.800 --> 01:36:08.080 align:middle line:90% is a little bit harder to say. 01:36:08.080 --> 01:36:10.640 align:middle line:84% Yeah, so this one does matter because it also 01:36:10.640 --> 01:36:14.000 align:middle line:84% cuts the shock to the system that 01:36:14.000 --> 01:36:15.820 align:middle line:90% came from the sentiment index. 01:36:15.820 --> 01:36:18.800 align:middle line:90% 01:36:18.800 --> 01:36:26.280 align:middle line:84% So anyway, the point is that, right now, 01:36:26.280 --> 01:36:29.480 align:middle line:84% the mechanism that is generating oscillations 01:36:29.480 --> 01:36:32.920 align:middle line:84% is more about the balancing loop of households changing 01:36:32.920 --> 01:36:36.360 align:middle line:84% their consumption, which changes the demand for the goods, which 01:36:36.360 --> 01:36:41.380 align:middle line:84% changes the labor needs and the wages that the households get, 01:36:41.380 --> 01:36:43.680 align:middle line:84% which changes the households needing goods. 01:36:43.680 --> 01:36:46.400 align:middle line:84% That's the dominant feedback loop here. 01:36:46.400 --> 01:36:48.160 align:middle line:84% That dominant feedback loop has nothing 01:36:48.160 --> 01:36:55.420 align:middle line:84% to do with your story that was related to dynamics 01:36:55.420 --> 01:37:00.980 align:middle line:90% of sentiments and volatility. 01:37:00.980 --> 01:37:04.020 align:middle line:84% So I'm not saying those don't create oscillations 01:37:04.020 --> 01:37:06.540 align:middle line:84% or can't create oscillations, but I'm 01:37:06.540 --> 01:37:11.260 align:middle line:84% saying your model has so many other moving parts that it 01:37:11.260 --> 01:37:13.540 align:middle line:84% is generating oscillation for reasons 01:37:13.540 --> 01:37:17.780 align:middle line:84% that you didn't care about as much, 01:37:17.780 --> 01:37:23.700 align:middle line:84% that are orthogonal to what you were defining as your problem. 01:37:23.700 --> 01:37:25.837 align:middle line:90% OK, I hope this was helpful. 01:37:25.837 --> 01:37:26.920 align:middle line:90% AUDIENCE: Yes, definitely. 01:37:26.920 --> 01:37:27.940 align:middle line:90% Thank you very much. 01:37:27.940 --> 01:37:30.965 align:middle line:84% HAZHIR RAHMANDAD: Let's take a break. 01:37:30.965 --> 01:37:33.340 align:middle line:84% We went a little bit over time, but let's take 10 minutes 01:37:33.340 --> 01:37:36.780 align:middle line:90% and come back. 01:37:36.780 --> 01:37:38.800 align:middle line:84% OK, I think we can get back together. 01:37:38.800 --> 01:37:40.840 align:middle line:84% I think it's kind of-- we can get started. 01:37:40.840 --> 01:37:46.980 align:middle line:84% I just opened Jake's model, which is on production of movies 01:37:46.980 --> 01:37:49.660 align:middle line:84% and series and competition between different locations 01:37:49.660 --> 01:37:55.270 align:middle line:90% for production. 01:37:55.270 --> 01:37:58.770 align:middle line:84% And the model basically has the story 01:37:58.770 --> 01:38:06.050 align:middle line:84% of two primary drivers of attractiveness of locations 01:38:06.050 --> 01:38:12.870 align:middle line:84% being talent, human capital, actors, and set designers, 01:38:12.870 --> 01:38:16.290 align:middle line:84% and all the kind of crew that make a movie, 01:38:16.290 --> 01:38:19.630 align:middle line:84% and local production infrastructure, 01:38:19.630 --> 01:38:24.310 align:middle line:84% which could be physical infrastructure and technology. 01:38:24.310 --> 01:38:25.330 align:middle line:90% Yeah. 01:38:25.330 --> 01:38:30.210 align:middle line:84% So those two, whether you have better options 01:38:30.210 --> 01:38:33.330 align:middle line:84% than alternative places makes you more attractive 01:38:33.330 --> 01:38:34.390 align:middle line:90% or less attractive. 01:38:34.390 --> 01:38:39.650 align:middle line:84% And then that determines what fraction of movies 01:38:39.650 --> 01:38:43.070 align:middle line:84% will decide to go with your location versus other locations. 01:38:43.070 --> 01:38:47.730 align:middle line:84% But then that creates attractiveness drive for talent 01:38:47.730 --> 01:38:54.230 align:middle line:84% to go to places that are getting a lot of production happening 01:38:54.230 --> 01:38:57.870 align:middle line:84% and for investment into the infrastructure for places 01:38:57.870 --> 01:38:59.070 align:middle line:90% that are getting. 01:38:59.070 --> 01:39:01.790 align:middle line:84% So that creates reinforcing loops that 01:39:01.790 --> 01:39:04.390 align:middle line:90% can generate bifurcation into-- 01:39:04.390 --> 01:39:07.910 align:middle line:90% that's the story. 01:39:07.910 --> 01:39:12.253 align:middle line:84% AUDIENCE: Does talent move away in this version of the model? 01:39:12.253 --> 01:39:13.170 align:middle line:90% HAZHIR RAHMANDAD: So-- 01:39:13.170 --> 01:39:15.890 align:middle line:84% AUDIENCE: The talent doesn't move here. 01:39:15.890 --> 01:39:16.390 align:middle line:90% No. 01:39:16.390 --> 01:39:20.270 align:middle line:84% It's the idea is that you'll have a local talent 01:39:20.270 --> 01:39:23.670 align:middle line:90% experienced workforce. 01:39:23.670 --> 01:39:26.090 align:middle line:84% It's not even necessarily number of people. 01:39:26.090 --> 01:39:30.710 align:middle line:84% It's person hours who have experience. 01:39:30.710 --> 01:39:32.390 align:middle line:84% HAZHIR RAHMANDAD: So this is actually 01:39:32.390 --> 01:39:37.790 align:middle line:84% a little bit of a how we define these resources is important. 01:39:37.790 --> 01:39:40.590 align:middle line:84% So one definition is number of people. 01:39:40.590 --> 01:39:45.070 align:middle line:84% The other is number of experience hours 01:39:45.070 --> 01:39:47.550 align:middle line:90% in the geography. 01:39:47.550 --> 01:39:52.390 align:middle line:84% And both of them are kind of a little bit leave me a squeamish, 01:39:52.390 --> 01:39:56.350 align:middle line:84% because people without experience is not perfect, 01:39:56.350 --> 01:40:00.450 align:middle line:84% and experienced others is too much 01:40:00.450 --> 01:40:04.970 align:middle line:84% giving credit to more experience because the return on experience 01:40:04.970 --> 01:40:07.430 align:middle line:90% is diminishing pretty rapidly. 01:40:07.430 --> 01:40:13.690 align:middle line:84% So if you want to capture both of them, 01:40:13.690 --> 01:40:18.090 align:middle line:84% one way, is to have a co-flow of people and the experience 01:40:18.090 --> 01:40:19.050 align:middle line:90% that they have. 01:40:19.050 --> 01:40:23.230 align:middle line:84% So that when people join, they join with limited experience. 01:40:23.230 --> 01:40:25.970 align:middle line:84% But then when they are in, they are gaining experience 01:40:25.970 --> 01:40:31.090 align:middle line:84% based on how many other productions they are engaged in. 01:40:31.090 --> 01:40:32.133 align:middle line:90% You could do that. 01:40:32.133 --> 01:40:33.550 align:middle line:84% I'm not saying you should do that. 01:40:33.550 --> 01:40:35.830 align:middle line:84% I'm just saying kind of conceptually, 01:40:35.830 --> 01:40:38.130 align:middle line:90% that gives you the flexibility. 01:40:38.130 --> 01:40:42.170 align:middle line:84% If you are not doing that, then maybe I 01:40:42.170 --> 01:40:45.810 align:middle line:84% would go with people, experienced people or something 01:40:45.810 --> 01:40:50.510 align:middle line:84% like that, so that I focus on what is the bottleneck 01:40:50.510 --> 01:40:55.510 align:middle line:84% is the number of experienced people or capable people. 01:40:55.510 --> 01:41:00.010 align:middle line:90% 01:41:00.010 --> 01:41:02.050 align:middle line:90% And then we were talking about-- 01:41:02.050 --> 01:41:04.470 align:middle line:84% OK, so the way Jake has formulated 01:41:04.470 --> 01:41:07.250 align:middle line:84% these is that for each component-- 01:41:07.250 --> 01:41:10.550 align:middle line:84% so you have a talent component, the way it is formulated 01:41:10.550 --> 01:41:13.370 align:middle line:84% is your local talent for that location. 01:41:13.370 --> 01:41:15.390 align:middle line:84% So this is subscripted by locations 01:41:15.390 --> 01:41:18.910 align:middle line:90% divided by total talent. 01:41:18.910 --> 01:41:23.390 align:middle line:84% And one feedback that I gave him on the previous week's version 01:41:23.390 --> 01:41:27.070 align:middle line:84% of the model was that maybe you can consider 01:41:27.070 --> 01:41:31.430 align:middle line:84% a denominator, a normalizing factor that is not 01:41:31.430 --> 01:41:34.350 align:middle line:90% tied to the total talent. 01:41:34.350 --> 01:41:40.510 align:middle line:84% And the argument there is the real attractiveness 01:41:40.510 --> 01:41:46.250 align:middle line:84% of Los Angeles versus New York, the attractiveness of Los 01:41:46.250 --> 01:41:49.790 align:middle line:84% Angeles per se is based on whether you have talent or not. 01:41:49.790 --> 01:41:53.050 align:middle line:84% It doesn't depend on whether New York has talent or not. 01:41:53.050 --> 01:41:56.110 align:middle line:84% So if talent in New York doubles, 01:41:56.110 --> 01:42:00.570 align:middle line:84% the attractiveness of Los Angeles alone should not change. 01:42:00.570 --> 01:42:03.030 align:middle line:84% The relative attractiveness certainly changes. 01:42:03.030 --> 01:42:05.490 align:middle line:84% But that relative attractiveness is already captures 01:42:05.490 --> 01:42:09.770 align:middle line:84% because now New York has twice as many talents, 01:42:09.770 --> 01:42:14.410 align:middle line:84% and therefore, it gets a bigger talent component coming 01:42:14.410 --> 01:42:17.910 align:middle line:90% from that numerator. 01:42:17.910 --> 01:42:20.570 align:middle line:84% You don't need to adjust the denominator. 01:42:20.570 --> 01:42:23.050 align:middle line:84% That was the argument I was making. 01:42:23.050 --> 01:42:25.170 align:middle line:84% And the reason it matters is that, 01:42:25.170 --> 01:42:29.970 align:middle line:84% in the current formulation, the talent component remains a 0 sum 01:42:29.970 --> 01:42:30.470 align:middle line:90% game. 01:42:30.470 --> 01:42:32.970 align:middle line:84% That is essentially the sum of the talent component 01:42:32.970 --> 01:42:35.870 align:middle line:90% across all of them, gets to 1. 01:42:35.870 --> 01:42:38.450 align:middle line:90% 01:42:38.450 --> 01:42:42.510 align:middle line:84% And for allocation problems, that's OK. 01:42:42.510 --> 01:42:45.150 align:middle line:84% You all, ultimately, you are allocating 01:42:45.150 --> 01:42:47.590 align:middle line:84% across a fixed amount of production 01:42:47.590 --> 01:42:51.030 align:middle line:84% among different places, so it will work fine. 01:42:51.030 --> 01:42:56.510 align:middle line:84% But if you had a situation where the production start was going 01:42:56.510 --> 01:43:03.830 align:middle line:84% to depend on availability of talent, and location, and so on, 01:43:03.830 --> 01:43:06.550 align:middle line:84% and you could increase or decrease the production start 01:43:06.550 --> 01:43:08.870 align:middle line:84% rate based on the availabilities, 01:43:08.870 --> 01:43:11.110 align:middle line:90% then things start to change. 01:43:11.110 --> 01:43:15.390 align:middle line:84% Then the absolute values matter more. 01:43:15.390 --> 01:43:17.830 align:middle line:84% And tracking those absolute values 01:43:17.830 --> 01:43:20.350 align:middle line:84% or normalized absolute values would 01:43:20.350 --> 01:43:22.470 align:middle line:90% be more useful in that sense. 01:43:22.470 --> 01:43:25.630 align:middle line:90% So again, not a big deal. 01:43:25.630 --> 01:43:28.890 align:middle line:84% And for your allocation problem it just works fine. 01:43:28.890 --> 01:43:32.810 align:middle line:84% But for thinking about the future of this model, 01:43:32.810 --> 01:43:34.990 align:middle line:90% I would have thought about. 01:43:34.990 --> 01:43:37.410 align:middle line:84% And very similar feedback on the other part, 01:43:37.410 --> 01:43:40.230 align:middle line:84% which is the local production infrastructure. 01:43:40.230 --> 01:43:43.630 align:middle line:84% And right now, you have decays for them, which are fine. 01:43:43.630 --> 01:43:48.250 align:middle line:84% Each of them can have a decay time. 01:43:48.250 --> 01:43:50.552 align:middle line:84% AUDIENCE: Another question [INAUDIBLE]. 01:43:50.552 --> 01:43:51.510 align:middle line:90% HAZHIR RAHMANDAD: Sure. 01:43:51.510 --> 01:43:53.177 align:middle line:84% AUDIENCE: Because I do feel like the tax 01:43:53.177 --> 01:43:55.770 align:middle line:90% credit is such an important-- 01:43:55.770 --> 01:43:59.206 align:middle line:90% [INTERPOSING VOICES] 01:43:59.206 --> 01:44:07.810 align:middle line:84% AUDIENCE: --would be, do I need to have an overall idea of costs 01:44:07.810 --> 01:44:09.297 align:middle line:90% associated with using local-- 01:44:09.297 --> 01:44:10.130 align:middle line:90% [INTERPOSING VOICES] 01:44:10.130 --> 01:44:12.213 align:middle line:84% HAZHIR RAHMANDAD: So this is a very good question. 01:44:12.213 --> 01:44:14.650 align:middle line:84% And I think this is really at the heart of this allocation 01:44:14.650 --> 01:44:20.450 align:middle line:84% problem is how productions decide where to locate. 01:44:20.450 --> 01:44:23.330 align:middle line:84% And that production deciding where to locate 01:44:23.330 --> 01:44:26.730 align:middle line:84% is a cost-benefit analysis, some kind of an optimization problem 01:44:26.730 --> 01:44:30.170 align:middle line:84% that I have two portfolios, one with this talent 01:44:30.170 --> 01:44:33.550 align:middle line:84% pool, this infrastructure and this tax credit, 01:44:33.550 --> 01:44:35.910 align:middle line:90% and maybe five other factors. 01:44:35.910 --> 01:44:37.430 align:middle line:90% And this is the other one. 01:44:37.430 --> 01:44:39.910 align:middle line:84% So how do I decide to allocate between them? 01:44:39.910 --> 01:44:43.310 align:middle line:84% Right now you have an ad hoc attractiveness that 01:44:43.310 --> 01:44:46.870 align:middle line:90% just divides and allocates. 01:44:46.870 --> 01:44:47.650 align:middle line:90% That's OK. 01:44:47.650 --> 01:44:49.750 align:middle line:90% It's a good first cut. 01:44:49.750 --> 01:44:53.270 align:middle line:84% I'm not going to say you shouldn't use that. 01:44:53.270 --> 01:44:55.990 align:middle line:84% But if you wanted to go the route of a little bit 01:44:55.990 --> 01:44:59.510 align:middle line:84% more rational decision-making process, 01:44:59.510 --> 01:45:02.710 align:middle line:84% you need to define a production function of sorts 01:45:02.710 --> 01:45:07.670 align:middle line:84% that says, OK, what would be the net quality, 01:45:07.670 --> 01:45:11.550 align:middle line:84% net profitability of my production, 01:45:11.550 --> 01:45:17.510 align:middle line:84% given this level of talent, this level of production 01:45:17.510 --> 01:45:21.390 align:middle line:84% infrastructure, and this level of tax credit. 01:45:21.390 --> 01:45:24.430 align:middle line:84% So if you can write a profit function 01:45:24.430 --> 01:45:27.590 align:middle line:84% and that profit function has costs and has benefits, 01:45:27.590 --> 01:45:30.150 align:middle line:90% then you can solve for-- 01:45:30.150 --> 01:45:34.790 align:middle line:84% between those different alternatives which one gives you 01:45:34.790 --> 01:45:37.110 align:middle line:90% the highest profit. 01:45:37.110 --> 01:45:40.890 align:middle line:84% So you can have expected profits from the five 01:45:40.890 --> 01:45:44.770 align:middle line:84% different locations based on calculating those profits. 01:45:44.770 --> 01:45:49.730 align:middle line:84% And you say, OK, and then you have a sensitivity parameter 01:45:49.730 --> 01:45:56.450 align:middle line:84% that says, how sensitive I'm to that expected profit? 01:45:56.450 --> 01:45:58.930 align:middle line:84% If I allocate 100% to the location 01:45:58.930 --> 01:46:03.090 align:middle line:84% that gives me the most profit, or I allocate 80% to that 01:46:03.090 --> 01:46:07.470 align:middle line:84% and 20% to the second one, or some kind of a other mix, 01:46:07.470 --> 01:46:10.250 align:middle line:84% and that could be a simple logistic function 01:46:10.250 --> 01:46:12.470 align:middle line:84% for distributing between these players. 01:46:12.470 --> 01:46:14.970 align:middle line:84% So that would be a little bit more rational. 01:46:14.970 --> 01:46:19.210 align:middle line:84% AUDIENCE: How would I do that sort of idea without-- it 01:46:19.210 --> 01:46:22.330 align:middle line:84% feels to me like then I would need a dozen 01:46:22.330 --> 01:46:29.790 align:middle line:84% additional parameters that tied specifically to benefit and cost 01:46:29.790 --> 01:46:32.033 align:middle line:90% [INAUDIBLE] 01:46:32.033 --> 01:46:33.450 align:middle line:84% HAZHIR RAHMANDAD: So right now you 01:46:33.450 --> 01:46:39.550 align:middle line:84% have those parameters embedded in how these sensitivities are 01:46:39.550 --> 01:46:40.250 align:middle line:90% happening. 01:46:40.250 --> 01:46:43.270 align:middle line:84% So there's sensitivity of the talent and sensitivity 01:46:43.270 --> 01:46:46.010 align:middle line:90% of infrastructure. 01:46:46.010 --> 01:46:49.290 align:middle line:84% And then you have this tax credit sensitivity. 01:46:49.290 --> 01:46:52.270 align:middle line:84% So these three are the drivers right now. 01:46:52.270 --> 01:46:55.630 align:middle line:84% In that alternative world you can basically 01:46:55.630 --> 01:46:58.190 align:middle line:84% have these three parameters determine 01:46:58.190 --> 01:47:02.430 align:middle line:84% the contribution of these guys to a profit function. 01:47:02.430 --> 01:47:08.310 align:middle line:84% And then you get the same result out of it. 01:47:08.310 --> 01:47:12.270 align:middle line:84% So the number of free parameters shouldn't necessarily 01:47:12.270 --> 01:47:14.190 align:middle line:90% change much. 01:47:14.190 --> 01:47:16.290 align:middle line:84% It is just where do they show up. 01:47:16.290 --> 01:47:19.030 align:middle line:84% Do they show up in a final decision function the way 01:47:19.030 --> 01:47:20.990 align:middle line:84% you currently have, or do they show up 01:47:20.990 --> 01:47:23.110 align:middle line:84% inside your profit function and then 01:47:23.110 --> 01:47:25.470 align:middle line:84% the decision comes from the profit function. 01:47:25.470 --> 01:47:27.870 align:middle line:84% AUDIENCE: So you're saying that there would just 01:47:27.870 --> 01:47:32.287 align:middle line:84% be one contribution to profitability? 01:47:32.287 --> 01:47:33.870 align:middle line:84% HAZHIR RAHMANDAD: Something like that. 01:47:33.870 --> 01:47:35.940 align:middle line:84% AUDIENCE: Like a cost and the [INAUDIBLE]. 01:47:35.940 --> 01:47:39.220 align:middle line:84% HAZHIR RAHMANDAD: So you could shift it into a cost 01:47:39.220 --> 01:47:39.840 align:middle line:90% and benefit. 01:47:39.840 --> 01:47:42.680 align:middle line:84% So the way I would potentially shift it into cost and benefit, 01:47:42.680 --> 01:47:50.460 align:middle line:84% I would say, OK, I have some kind of a production revenue 01:47:50.460 --> 01:47:53.580 align:middle line:84% side, which might be, let's say, a Cobb-Douglas 01:47:53.580 --> 01:47:59.380 align:middle line:84% function that says it is your talent to the power of alpha 01:47:59.380 --> 01:48:03.100 align:middle line:84% multiplied by your infrastructure 01:48:03.100 --> 01:48:04.700 align:middle line:90% to the power of beta. 01:48:04.700 --> 01:48:09.480 align:middle line:84% And these both can be normalized using the normalization process. 01:48:09.480 --> 01:48:15.260 align:middle line:84% The component that you have already can do that for you. 01:48:15.260 --> 01:48:19.400 align:middle line:84% And then so that's your revenue side. 01:48:19.400 --> 01:48:24.660 align:middle line:84% This can be multiplied by some kind of a additional parameter 01:48:24.660 --> 01:48:28.220 align:middle line:84% that says how much each unit of production 01:48:28.220 --> 01:48:33.440 align:middle line:84% turns into unit of dollar minus your costs. 01:48:33.440 --> 01:48:38.760 align:middle line:84% The costs in this case could be your tax. 01:48:38.760 --> 01:48:42.020 align:middle line:84% So you can just add the tax part to your cost function. 01:48:42.020 --> 01:48:46.020 align:middle line:84% That would be a very simple OK starting point. 01:48:46.020 --> 01:48:48.560 align:middle line:84% Then you will have this alpha and beta, 01:48:48.560 --> 01:48:52.440 align:middle line:84% that multiplier for what the transformation of production 01:48:52.440 --> 01:48:54.460 align:middle line:90% units into dollar units. 01:48:54.460 --> 01:48:56.660 align:middle line:84% And the tax is already in the dollar units. 01:48:56.660 --> 01:48:58.320 align:middle line:84% So you have three free parameters 01:48:58.320 --> 01:49:00.980 align:middle line:90% for that profit function. 01:49:00.980 --> 01:49:03.840 align:middle line:90% It would work fine. 01:49:03.840 --> 01:49:07.040 align:middle line:84% Then out of that you calculate expected profitability 01:49:07.040 --> 01:49:10.840 align:middle line:90% from each of the locations. 01:49:10.840 --> 01:49:15.960 align:middle line:84% Then there is a distribution of the demand to locations based 01:49:15.960 --> 01:49:17.960 align:middle line:90% on that expected profitability. 01:49:17.960 --> 01:49:21.200 align:middle line:84% That distribution has another free parameter 01:49:21.200 --> 01:49:24.400 align:middle line:84% that says how sensitive they are to profitability. 01:49:24.400 --> 01:49:28.800 align:middle line:84% And basically it becomes profitability-- 01:49:28.800 --> 01:49:34.180 align:middle line:84% it could be even, yeah, E to the power of profitability 01:49:34.180 --> 01:49:37.820 align:middle line:84% multiplied by that sensitivity parameter divided by sum of E 01:49:37.820 --> 01:49:41.700 align:middle line:84% to the power of profitability multiplied that by sensitivity 01:49:41.700 --> 01:49:43.520 align:middle line:90% for all of the locations. 01:49:43.520 --> 01:49:46.740 align:middle line:90% 01:49:46.740 --> 01:49:47.845 align:middle line:90% Does that make sense? 01:49:47.845 --> 01:49:48.720 align:middle line:90% AUDIENCE: I think so. 01:49:48.720 --> 01:49:49.420 align:middle line:90% Yeah. 01:49:49.420 --> 01:49:52.140 align:middle line:84% HAZHIR RAHMANDAD: So what you're doing is OK. 01:49:52.140 --> 01:49:53.520 align:middle line:90% It's nothing wrong. 01:49:53.520 --> 01:49:57.480 align:middle line:84% I just was, to try to make it a little bit more structured, 01:49:57.480 --> 01:50:00.120 align:middle line:84% it helps to think in terms of profits 01:50:00.120 --> 01:50:03.200 align:middle line:84% and making a decision based on different profitable things. 01:50:03.200 --> 01:50:03.860 align:middle line:90% AUDIENCE: Yes. 01:50:03.860 --> 01:50:11.740 align:middle line:84% HAZHIR RAHMANDAD: And then you said you added the adjustment 01:50:11.740 --> 01:50:13.745 align:middle line:90% based on crowding-- 01:50:13.745 --> 01:50:14.620 align:middle line:90% AUDIENCE: Congestion. 01:50:14.620 --> 01:50:15.912 align:middle line:90% HAZHIR RAHMANDAD: --congestion. 01:50:15.912 --> 01:50:17.800 align:middle line:90% So let's see. 01:50:17.800 --> 01:50:20.580 align:middle line:90% 01:50:20.580 --> 01:50:24.220 align:middle line:90% So basically now you have-- 01:50:24.220 --> 01:50:26.420 align:middle line:84% OK, this is congestion modifier that 01:50:26.420 --> 01:50:31.040 align:middle line:84% changes the attractiveness for each location. 01:50:31.040 --> 01:50:34.760 align:middle line:84% So in the formulation we've talked about, 01:50:34.760 --> 01:50:38.420 align:middle line:84% so profitability would go down if you have more congestion. 01:50:38.420 --> 01:50:41.840 align:middle line:84% So basically you can think of a congestion 01:50:41.840 --> 01:50:44.100 align:middle line:90% effect on your revenue side. 01:50:44.100 --> 01:50:45.620 align:middle line:84% Or you can think of it as a cost. 01:50:45.620 --> 01:50:48.260 align:middle line:84% So that would have another parameter. 01:50:48.260 --> 01:50:51.060 align:middle line:84% Or it could be a multiplier of your revenue. 01:50:51.060 --> 01:50:53.160 align:middle line:84% That is, if you have to wait longer 01:50:53.160 --> 01:50:56.500 align:middle line:84% you lose the optimal window of opportunity, 01:50:56.500 --> 01:50:57.580 align:middle line:90% something like that. 01:50:57.580 --> 01:50:59.180 align:middle line:84% Or it's just your cost of capital. 01:50:59.180 --> 01:51:02.080 align:middle line:84% So that congestion multiplier can be multiplied 01:51:02.080 --> 01:51:03.720 align:middle line:90% by your revenue side. 01:51:03.720 --> 01:51:07.120 align:middle line:84% That Cobb-Douglas function, that's one way to do it. 01:51:07.120 --> 01:51:09.460 align:middle line:84% A bit ad hoc, but not too bad, I think. 01:51:09.460 --> 01:51:11.840 align:middle line:84% AUDIENCE: Yeah, or just raises the cost 01:51:11.840 --> 01:51:14.563 align:middle line:84% of using the infrastructure [INAUDIBLE]. 01:51:14.563 --> 01:51:16.980 align:middle line:84% HAZHIR RAHMANDAD: Yeah, so right now, in that formulation, 01:51:16.980 --> 01:51:20.480 align:middle line:84% we didn't talk about cost of using the infrastructure. 01:51:20.480 --> 01:51:22.420 align:middle line:84% That would be another way to do it. 01:51:22.420 --> 01:51:26.600 align:middle line:84% That is to have explicit prices for both of these talent 01:51:26.600 --> 01:51:30.720 align:middle line:84% and infrastructure, and when the demand exceeds supply, 01:51:30.720 --> 01:51:32.080 align:middle line:90% the prices are changing. 01:51:32.080 --> 01:51:34.420 align:middle line:84% Then you don't need the explicit-- there's 01:51:34.420 --> 01:51:36.620 align:middle line:90% a separate congestion modifier. 01:51:36.620 --> 01:51:39.400 align:middle line:84% You just have higher prices in this location. 01:51:39.400 --> 01:51:43.420 align:middle line:84% So that would require you to model the prices endogenously 01:51:43.420 --> 01:51:47.140 align:middle line:84% and explicitly but then removes the need for capturing 01:51:47.140 --> 01:51:50.900 align:middle line:90% congestion as explicitly. 01:51:50.900 --> 01:51:56.220 align:middle line:84% Congestion effect will show up in the price change. 01:51:56.220 --> 01:52:01.940 align:middle line:84% That is probably closer to what an economist likes to see. 01:52:01.940 --> 01:52:03.940 align:middle line:84% It's a little bit more work on the modeling 01:52:03.940 --> 01:52:06.820 align:middle line:84% side and a little bit more complexity. 01:52:06.820 --> 01:52:08.060 align:middle line:90% So I don't know. 01:52:08.060 --> 01:52:12.800 align:middle line:84% AUDIENCE: I mean, can you do that without the data? 01:52:12.800 --> 01:52:14.300 align:middle line:84% HAZHIR RAHMANDAD: So all of these we 01:52:14.300 --> 01:52:17.820 align:middle line:84% are talking about doing things without data right now. 01:52:17.820 --> 01:52:20.220 align:middle line:84% Basically you can do all of that. 01:52:20.220 --> 01:52:23.880 align:middle line:84% Then there is the question of, OK, I've built a model, 01:52:23.880 --> 01:52:28.000 align:middle line:84% I have a price in it, but my price behaves very weirdly, 01:52:28.000 --> 01:52:31.260 align:middle line:84% or my wages for talent is behaving very weirdly. 01:52:31.260 --> 01:52:33.360 align:middle line:90% What should I do? 01:52:33.360 --> 01:52:38.660 align:middle line:84% Then perhaps it is just parameters. 01:52:38.660 --> 01:52:43.760 align:middle line:84% That is, your price elasticity might be too sharp. 01:52:43.760 --> 01:52:45.100 align:middle line:90% You should reduce that. 01:52:45.100 --> 01:52:48.200 align:middle line:84% So maybe by changing parameters you get it into a mode 01:52:48.200 --> 01:52:49.960 align:middle line:84% that you say, OK, now these ranges 01:52:49.960 --> 01:52:55.720 align:middle line:84% of 20%, 30% difference in wages between LA versus Toronto 01:52:55.720 --> 01:52:57.140 align:middle line:90% is reasonable. 01:52:57.140 --> 01:52:58.220 align:middle line:90% So let me go with that. 01:52:58.220 --> 01:53:00.280 align:middle line:90% That's a good range. 01:53:00.280 --> 01:53:01.620 align:middle line:90% And I think that's doable. 01:53:01.620 --> 01:53:03.680 align:middle line:84% Probably if you play with it a little bit 01:53:03.680 --> 01:53:07.520 align:middle line:84% you will get it to that range as well. 01:53:07.520 --> 01:53:14.280 align:middle line:84% I'm not necessarily pushing you towards doing the full price 01:53:14.280 --> 01:53:15.240 align:middle line:90% part. 01:53:15.240 --> 01:53:20.080 align:middle line:84% If you go that route, then your Cobb-Douglas production function 01:53:20.080 --> 01:53:24.100 align:middle line:84% or any other production function would have a cost side to it 01:53:24.100 --> 01:53:29.460 align:middle line:84% and a revenue side to it, and the net profit that comes out 01:53:29.460 --> 01:53:34.540 align:middle line:84% of that whole piece will inform which of the locations people 01:53:34.540 --> 01:53:36.280 align:middle line:90% will be picking. 01:53:36.280 --> 01:53:39.860 align:middle line:90% 01:53:39.860 --> 01:53:44.340 align:middle line:84% Also then it may change the whole production decision 01:53:44.340 --> 01:53:47.340 align:middle line:84% of whether you do production or not. 01:53:47.340 --> 01:53:51.900 align:middle line:84% And that becomes even a more involved thing 01:53:51.900 --> 01:53:54.540 align:middle line:84% if you want to make the production decision also 01:53:54.540 --> 01:53:55.320 align:middle line:90% endogenous. 01:53:55.320 --> 01:53:56.940 align:middle line:84% But you could make it endogenous. 01:53:56.940 --> 01:53:59.840 align:middle line:84% If every location I'm going to lose money, 01:53:59.840 --> 01:54:04.260 align:middle line:84% I'm not as stupid to do this movie at all. 01:54:04.260 --> 01:54:09.360 align:middle line:84% So you could capture that as well. 01:54:09.360 --> 01:54:11.060 align:middle line:84% It becomes a little bit more complex 01:54:11.060 --> 01:54:15.800 align:middle line:84% because now it is not just an allocation problem, 01:54:15.800 --> 01:54:18.660 align:middle line:84% it is the total demand problem as well. 01:54:18.660 --> 01:54:23.080 align:middle line:84% But that total demand depends on what are the options available. 01:54:23.080 --> 01:54:29.480 align:middle line:84% So I think the basic core components there are there. 01:54:29.480 --> 01:54:33.160 align:middle line:84% You can push it in these directions if you want. 01:54:33.160 --> 01:54:35.680 align:middle line:84% Let's just take a look at the congestion part 01:54:35.680 --> 01:54:37.660 align:middle line:90% because we didn't look at it. 01:54:37.660 --> 01:54:39.210 align:middle line:90% Is it a different view or here? 01:54:39.210 --> 01:54:40.460 align:middle line:90% AUDIENCE: No, it's just above. 01:54:40.460 --> 01:54:41.142 align:middle line:90% It's at the top. 01:54:41.142 --> 01:54:42.100 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 01:54:42.100 --> 01:54:43.200 align:middle line:90% Got it. 01:54:43.200 --> 01:54:46.800 align:middle line:84% So congestion modifier right now is 01:54:46.800 --> 01:54:50.340 align:middle line:84% looking at the infrastructure, capacity, 01:54:50.340 --> 01:54:53.220 align:middle line:90% and infrastructure utilization. 01:54:53.220 --> 01:54:56.700 align:middle line:84% So for utilization to make sense, 01:54:56.700 --> 01:55:03.240 align:middle line:84% you need some unit of production per capacity per infrastructure 01:55:03.240 --> 01:55:06.200 align:middle line:90% unit that is normal. 01:55:06.200 --> 01:55:08.560 align:middle line:84% And then if you have more than that you 01:55:08.560 --> 01:55:11.520 align:middle line:90% get into the congestion space. 01:55:11.520 --> 01:55:17.580 align:middle line:84% So that, yes, you have that needed per infrastructure. 01:55:17.580 --> 01:55:19.160 align:middle line:90% So that works. 01:55:19.160 --> 01:55:22.020 align:middle line:90% And then you have a sensitivity. 01:55:22.020 --> 01:55:25.700 align:middle line:84% And what is the minimum congestion modifier? 01:55:25.700 --> 01:55:28.420 align:middle line:84% So you're saying there is a maximum of this minimum 01:55:28.420 --> 01:55:30.920 align:middle line:90% and E to the power of this guy. 01:55:30.920 --> 01:55:33.980 align:middle line:84% So you don't want this congestion modifier 01:55:33.980 --> 01:55:37.500 align:middle line:90% to go below a minimum. 01:55:37.500 --> 01:55:41.960 align:middle line:84% And then congestion modifier, where did it operate? 01:55:41.960 --> 01:55:44.840 align:middle line:84% So it operated on the attractiveness. 01:55:44.840 --> 01:55:48.300 align:middle line:90% So it was multiplied by that. 01:55:48.300 --> 01:55:51.540 align:middle line:84% So at low level of congestion modifier means 01:55:51.540 --> 01:55:55.900 align:middle line:84% the attractiveness of this location has gone down. 01:55:55.900 --> 01:56:01.060 align:middle line:84% So putting a lower bound on it is basically saying, 01:56:01.060 --> 01:56:04.780 align:middle line:84% it can't go below that level because of congestion. 01:56:04.780 --> 01:56:05.900 align:middle line:90% That's OK. 01:56:05.900 --> 01:56:09.320 align:middle line:84% But then you want to also make sure you have an upper bound. 01:56:09.320 --> 01:56:10.900 align:middle line:84% That is, the congestion should not 01:56:10.900 --> 01:56:15.900 align:middle line:84% go above-- this modifier should not go above 1, perhaps, 01:56:15.900 --> 01:56:18.810 align:middle line:84% because you can't make it more attractive than just 01:56:18.810 --> 01:56:23.450 align:middle line:84% some maximum by just not having congestion. 01:56:23.450 --> 01:56:27.510 align:middle line:84% So whether that comes from this-- 01:56:27.510 --> 01:56:31.930 align:middle line:84% so you have max of 0 and infrastructure utilization. 01:56:31.930 --> 01:56:35.170 align:middle line:84% So if infrastructure utilization is 0, 01:56:35.170 --> 01:56:39.130 align:middle line:84% this becomes 0 congestion sensitivity, multiplied 01:56:39.130 --> 01:56:44.490 align:middle line:90% by that 0, and then minus 1. 01:56:44.490 --> 01:56:46.050 align:middle line:90% AUDIENCE: It's the utilization. 01:56:46.050 --> 01:56:48.250 align:middle line:90% Here you can go above. 01:56:48.250 --> 01:56:50.250 align:middle line:90% That's the only thing here. 01:56:50.250 --> 01:56:51.410 align:middle line:90% HAZHIR RAHMANDAD: Yes. 01:56:51.410 --> 01:56:56.742 align:middle line:84% So if utilization is 0, this whole thing gets minus 1. 01:56:56.742 --> 01:56:57.950 align:middle line:90% AUDIENCE: Then it'll go to 0. 01:56:57.950 --> 01:57:00.790 align:middle line:90% 01:57:00.790 --> 01:57:01.290 align:middle line:90% Max. 01:57:01.290 --> 01:57:04.130 align:middle line:90% 01:57:04.130 --> 01:57:11.150 align:middle line:84% HAZHIR RAHMANDAD: So but max 0 of this guy-- 01:57:11.150 --> 01:57:11.650 align:middle line:90% OK. 01:57:11.650 --> 01:57:13.550 align:middle line:90% This is the parentheses for max. 01:57:13.550 --> 01:57:14.490 align:middle line:90% OK. 01:57:14.490 --> 01:57:18.350 align:middle line:84% This is the parentheses for E to the power. 01:57:18.350 --> 01:57:21.270 align:middle line:84% So inside the E to the power parentheses 01:57:21.270 --> 01:57:24.070 align:middle line:84% we have congestion sensitivity, some number multiplied 01:57:24.070 --> 01:57:29.230 align:middle line:90% by a number that is 0. 01:57:29.230 --> 01:57:31.350 align:middle line:90% So then it is 0 and it is 1. 01:57:31.350 --> 01:57:34.630 align:middle line:84% So that gives you the modifier of 1. 01:57:34.630 --> 01:57:36.270 align:middle line:90% Got it. 01:57:36.270 --> 01:57:40.710 align:middle line:90% Yeah, it's perhaps working fine. 01:57:40.710 --> 01:57:45.890 align:middle line:90% I'm not seeing any problem here. 01:57:45.890 --> 01:57:46.730 align:middle line:90% OK, good. 01:57:46.730 --> 01:57:51.110 align:middle line:84% I think that is, for what you are doing right now, 01:57:51.110 --> 01:57:52.010 align:middle line:90% this is working OK. 01:57:52.010 --> 01:57:57.830 align:middle line:84% Now you had imported data for LA production days. 01:57:57.830 --> 01:58:02.830 align:middle line:84% Pretty interesting to think about how this data might 01:58:02.830 --> 01:58:07.310 align:middle line:90% go with what you are building. 01:58:07.310 --> 01:58:10.450 align:middle line:84% Part of the challenge is you're showing the LA part, 01:58:10.450 --> 01:58:13.150 align:middle line:84% but you're not showing anybody else. 01:58:13.150 --> 01:58:14.170 align:middle line:90% So-- 01:58:14.170 --> 01:58:19.450 align:middle line:84% AUDIENCE: I do have data on total productions that I have, 01:58:19.450 --> 01:58:21.610 align:middle line:84% and I also don't know how to use that 01:58:21.610 --> 01:58:25.290 align:middle line:84% as the input flow of production, but I do have that. 01:58:25.290 --> 01:58:28.370 align:middle line:84% HAZHIR RAHMANDAD: OK, so that's a little bit 01:58:28.370 --> 01:58:30.210 align:middle line:90% of a technical Vensim thing. 01:58:30.210 --> 01:58:32.170 align:middle line:84% So the technical Vensim thing, you 01:58:32.170 --> 01:58:37.850 align:middle line:84% need to make that as so you can import it into the DSS Vensim, 01:58:37.850 --> 01:58:44.097 align:middle line:84% you can import that data just by going to-- 01:58:44.097 --> 01:58:44.930 align:middle line:90% [INTERPOSING VOICES] 01:58:44.930 --> 01:58:46.430 align:middle line:84% HAZHIR RAHMANDAD: --Tools somewhere. 01:58:46.430 --> 01:58:47.570 align:middle line:90% Oh, you already have it. 01:58:47.570 --> 01:58:48.258 align:middle line:90% AUDIENCE: Yeah. 01:58:48.258 --> 01:58:50.050 align:middle line:84% HAZHIR RAHMANDAD: What's the variable name? 01:58:50.050 --> 01:58:51.770 align:middle line:84% AUDIENCE: It should be both of these. 01:58:51.770 --> 01:58:53.730 align:middle line:90% But [INAUDIBLE]. 01:58:53.730 --> 01:58:54.690 align:middle line:90% I don't know. 01:58:54.690 --> 01:58:55.010 align:middle line:90% HAZHIR RAHMANDAD: Oh. 01:58:55.010 --> 01:58:55.570 align:middle line:90% [INTERPOSING VOICES] 01:58:55.570 --> 01:58:56.445 align:middle line:90% HAZHIR RAHMANDAD: OK. 01:58:56.445 --> 01:58:59.490 align:middle line:84% What is the exact name you have in the-- 01:58:59.490 --> 01:59:00.650 align:middle line:90% movies? 01:59:00.650 --> 01:59:02.910 align:middle line:84% AUDIENCE: Yeah, I think that's the name of the column. 01:59:02.910 --> 01:59:03.490 align:middle line:90% And then-- 01:59:03.490 --> 01:59:05.090 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 01:59:05.090 --> 01:59:11.010 align:middle line:84% So if you have it here and you look at it. 01:59:11.010 --> 01:59:14.390 align:middle line:84% So if the spelling is correct, it will show up. 01:59:14.390 --> 01:59:17.040 align:middle line:90% It's just as simple as this. 01:59:17.040 --> 01:59:18.510 align:middle line:90% AUDIENCE: Can we run it again? 01:59:18.510 --> 01:59:20.990 align:middle line:84% HAZHIR RAHMANDAD: Oh, maybe we should run it again. 01:59:20.990 --> 01:59:25.670 align:middle line:84% Well, I should make it a data variable. 01:59:25.670 --> 01:59:28.990 align:middle line:90% So it should be data. 01:59:28.990 --> 01:59:31.990 align:middle line:90% 01:59:31.990 --> 01:59:33.990 align:middle line:90% AUDIENCE: And then too much. 01:59:33.990 --> 01:59:35.675 align:middle line:90% HAZHIR RAHMANDAD: Hmm? 01:59:35.675 --> 01:59:37.162 align:middle line:90% AUDIENCE: Might be capitalized? 01:59:37.162 --> 01:59:38.370 align:middle line:90% I don't know if that matters. 01:59:38.370 --> 01:59:41.430 align:middle line:90% 01:59:41.430 --> 01:59:42.930 align:middle line:84% HAZHIR RAHMANDAD: It didn't show up, 01:59:42.930 --> 01:59:45.783 align:middle line:84% but interestingly, it didn't also give us an error. 01:59:45.783 --> 01:59:46.950 align:middle line:90% AUDIENCE: It actually gave-- 01:59:46.950 --> 01:59:48.450 align:middle line:90% HAZHIR RAHMANDAD: Ah, I just-- 01:59:48.450 --> 01:59:50.430 align:middle line:84% AUDIENCE: Maybe put both of them [INAUDIBLE]. 01:59:50.430 --> 01:59:52.347 align:middle line:84% HAZHIR RAHMANDAD: Yeah, so now it is aborting. 01:59:52.347 --> 01:59:55.750 align:middle line:84% So it didn't do the simulation because it didn't find movies 01:59:55.750 --> 01:59:59.270 align:middle line:84% in the VDF file in-- well, we should say 01:59:59.270 --> 02:00:01.010 align:middle line:90% where to read it from as well. 02:00:01.010 --> 02:00:03.930 align:middle line:90% So data is coming from where? 02:00:03.930 --> 02:00:04.430 align:middle line:90% So-- 02:00:04.430 --> 02:00:05.565 align:middle line:90% AUDIENCE: The two. 02:00:05.565 --> 02:00:06.690 align:middle line:90% HAZHIR RAHMANDAD: This one? 02:00:06.690 --> 02:00:07.630 align:middle line:90% AUDIENCE: Yeah. 02:00:07.630 --> 02:00:08.505 align:middle line:90% HAZHIR RAHMANDAD: OK. 02:00:08.505 --> 02:00:12.770 align:middle line:90% 02:00:12.770 --> 02:00:14.890 align:middle line:90% Oh, it did work. 02:00:14.890 --> 02:00:18.140 align:middle line:84% So now you have the movies built into it. 02:00:18.140 --> 02:00:19.390 align:middle line:90% AUDIENCE: How did you do that? 02:00:19.390 --> 02:00:22.090 align:middle line:90% [LAUGHTER] 02:00:22.090 --> 02:00:24.270 align:middle line:84% HAZHIR RAHMANDAD: So there are two steps to it. 02:00:24.270 --> 02:00:26.090 align:middle line:84% So one was just defining the variable 02:00:26.090 --> 02:00:28.570 align:middle line:90% with the same exact name. 02:00:28.570 --> 02:00:32.970 align:middle line:84% The second was I went into the simulation control, 02:00:32.970 --> 02:00:37.250 align:middle line:84% inside simulation control to data. 02:00:37.250 --> 02:00:43.950 align:middle line:84% So I loaded the data set that you have as input into it. 02:00:43.950 --> 02:00:44.850 align:middle line:90% AUDIENCE: Got it. 02:00:44.850 --> 02:00:49.210 align:middle line:84% HAZHIR RAHMANDAD: And then once you have that you also make sure 02:00:49.210 --> 02:00:53.710 align:middle line:90% this variable is defined-- 02:00:53.710 --> 02:00:56.450 align:middle line:84% its equation type as a data equation. 02:00:56.450 --> 02:00:59.090 align:middle line:84% So that way Vensim knows it has to read 02:00:59.090 --> 02:01:03.010 align:middle line:84% some data for this variable from one of those data 02:01:03.010 --> 02:01:05.910 align:middle line:84% files that you have introduced as the source of the data. 02:01:05.910 --> 02:01:08.417 align:middle line:90% That's how it does it. 02:01:08.417 --> 02:01:09.250 align:middle line:90% AUDIENCE: Thank you. 02:01:09.250 --> 02:01:10.208 align:middle line:90% HAZHIR RAHMANDAD: Sure. 02:01:10.208 --> 02:01:13.090 align:middle line:84% And now you can just use it as an input into your capture. 02:01:13.090 --> 02:01:17.710 align:middle line:90% It's very straightforward. 02:01:17.710 --> 02:01:20.390 align:middle line:90% OK, I'll save this. 02:01:20.390 --> 02:01:22.237 align:middle line:90% You have it there. 02:01:22.237 --> 02:01:23.070 align:middle line:90% AUDIENCE: Thank you. 02:01:23.070 --> 02:01:24.028 align:middle line:90% HAZHIR RAHMANDAD: Sure. 02:01:24.028 --> 02:01:26.030 align:middle line:90% OK, so let's see. 02:01:26.030 --> 02:01:31.730 align:middle line:84% I think we have Ali, Jason, and Hassan. 02:01:31.730 --> 02:01:35.630 align:middle line:84% So let's see if we can go through all of them 02:01:35.630 --> 02:01:37.670 align:middle line:90% in the time we have. 02:01:37.670 --> 02:01:39.830 align:middle line:90% Try that. 02:01:39.830 --> 02:01:40.767 align:middle line:90% OK, Jason. 02:01:40.767 --> 02:01:41.350 align:middle line:90% AUDIENCE: Yes. 02:01:41.350 --> 02:01:44.190 align:middle line:84% HAZHIR RAHMANDAD: Tell us what you have done since last week 02:01:44.190 --> 02:01:44.790 align:middle line:90% and how-- 02:01:44.790 --> 02:01:47.030 align:middle line:90% AUDIENCE: Ah, totally changed. 02:01:47.030 --> 02:01:50.710 align:middle line:84% Well, this week I'm looking at supply side. 02:01:50.710 --> 02:01:56.550 align:middle line:84% Last week, I made a first pass at consumption. 02:01:56.550 --> 02:02:01.990 align:middle line:84% But I think I have a clearer story for this supply generation 02:02:01.990 --> 02:02:02.790 align:middle line:90% side. 02:02:02.790 --> 02:02:08.090 align:middle line:84% And then next time I'm going to try again on the demand side. 02:02:08.090 --> 02:02:10.370 align:middle line:84% HAZHIR RAHMANDAD: So you basically 02:02:10.370 --> 02:02:12.650 align:middle line:84% are not covering demand right now separate 02:02:12.650 --> 02:02:14.120 align:middle line:90% as an endogenous thing. 02:02:14.120 --> 02:02:14.870 align:middle line:90% AUDIENCE: Not yet. 02:02:14.870 --> 02:02:16.072 align:middle line:90% HAZHIR RAHMANDAD: OK. 02:02:16.072 --> 02:02:17.530 align:middle line:84% AUDIENCE: But I think I feel pretty 02:02:17.530 --> 02:02:20.970 align:middle line:84% good about the way forecast feeds into generating 02:02:20.970 --> 02:02:22.490 align:middle line:90% the protection level. 02:02:22.490 --> 02:02:23.050 align:middle line:90% So-- 02:02:23.050 --> 02:02:23.310 align:middle line:90% HAZHIR RAHMANDAD: OK. 02:02:23.310 --> 02:02:24.429 align:middle line:90% What's protection level? 02:02:24.429 --> 02:02:25.096 align:middle line:90% AUDIENCE: Right. 02:02:25.096 --> 02:02:26.750 align:middle line:90% So the idea is the following. 02:02:26.750 --> 02:02:31.050 align:middle line:90% 02:02:31.050 --> 02:02:33.690 align:middle line:84% The demand that's willing to pay more for a flight 02:02:33.690 --> 02:02:38.530 align:middle line:84% comes closer to the time of flight departs. 02:02:38.530 --> 02:02:42.250 align:middle line:84% And therefore the revenue manager, 02:02:42.250 --> 02:02:48.730 align:middle line:84% the person who is setting fares and protecting seats 02:02:48.730 --> 02:02:57.170 align:middle line:84% for bookings that come later has to protect some seats 02:02:57.170 --> 02:03:00.430 align:middle line:90% for those higher fares. 02:03:00.430 --> 02:03:02.810 align:middle line:84% So the protection level is, how many seats should I 02:03:02.810 --> 02:03:10.030 align:middle line:84% allocate when a flight is born to these higher fare markets 02:03:10.030 --> 02:03:13.930 align:middle line:84% before the flight even is published. 02:03:13.930 --> 02:03:20.190 align:middle line:84% And now if you have more demand in the early days, 02:03:20.190 --> 02:03:21.690 align:middle line:84% you might sell out of the low fares, 02:03:21.690 --> 02:03:27.310 align:middle line:84% but you have 20 or x whatever seats-- 02:03:27.310 --> 02:03:31.962 align:middle line:84% n number of seats is set apart for those later flights. 02:03:31.962 --> 02:03:33.670 align:middle line:84% HAZHIR RAHMANDAD: So the protection level 02:03:33.670 --> 02:03:39.310 align:middle line:90% is in seats or seats per flight. 02:03:39.310 --> 02:03:41.430 align:middle line:84% I assume they are kind of the same in this model 02:03:41.430 --> 02:03:43.690 align:middle line:84% because you're not looking at multiple flights. 02:03:43.690 --> 02:03:45.190 align:middle line:90% AUDIENCE: Yes. 02:03:45.190 --> 02:03:51.350 align:middle line:84% HAZHIR RAHMANDAD: So you start with an initial protection 02:03:51.350 --> 02:03:55.430 align:middle line:90% level, which is-- 02:03:55.430 --> 02:03:57.950 align:middle line:84% AUDIENCE: Whatever is indicated by the forecast. 02:03:57.950 --> 02:04:03.690 align:middle line:84% So this is based on-- so what is this mapping onto? 02:04:03.690 --> 02:04:05.570 align:middle line:90% What am I trying to get to here? 02:04:05.570 --> 02:04:09.330 align:middle line:84% There's a forecast which is generated by ticket sales. 02:04:09.330 --> 02:04:12.330 align:middle line:84% And then based on that forecast, you 02:04:12.330 --> 02:04:14.810 align:middle line:84% make a decision about how to set your protections 02:04:14.810 --> 02:04:17.570 align:middle line:90% for subsequent flights. 02:04:17.570 --> 02:04:19.810 align:middle line:84% Those protections are going to influence-- 02:04:19.810 --> 02:04:22.010 align:middle line:84% HAZHIR RAHMANDAD: Subsequent flights or this flight? 02:04:22.010 --> 02:04:23.218 align:middle line:90% AUDIENCE: Subsequent flights. 02:04:23.218 --> 02:04:25.840 align:middle line:90% 02:04:25.840 --> 02:04:29.890 align:middle line:84% And then those protection levels will 02:04:29.890 --> 02:04:35.310 align:middle line:84% determine future sales, which determine future forecasts. 02:04:35.310 --> 02:04:38.010 align:middle line:84% And you can get this spiral bound behavior 02:04:38.010 --> 02:04:40.970 align:middle line:84% where if you have a wrong initial belief 02:04:40.970 --> 02:04:46.030 align:middle line:84% or you misinterpret the booking screen, 02:04:46.030 --> 02:04:52.988 align:middle line:84% you can have very poorly aligned protection levels to your sales. 02:04:52.988 --> 02:04:54.530 align:middle line:84% HAZHIR RAHMANDAD: So let me just make 02:04:54.530 --> 02:04:56.950 align:middle line:84% sure I'm understanding protection level correctly. 02:04:56.950 --> 02:05:04.390 align:middle line:84% So you have a 300-seat flight a year before the start. 02:05:04.390 --> 02:05:08.510 align:middle line:84% A protection level of 250 means I'm opening 02:05:08.510 --> 02:05:11.418 align:middle line:90% 50 of those seats for sales. 02:05:11.418 --> 02:05:12.710 align:middle line:90% AUDIENCE: For low-ticket sales. 02:05:12.710 --> 02:05:16.727 align:middle line:84% HAZHIR RAHMANDAD: For whatever my current price is, correct? 02:05:16.727 --> 02:05:17.310 align:middle line:90% AUDIENCE: Yes. 02:05:17.310 --> 02:05:20.630 align:middle line:84% HAZHIR RAHMANDAD: And then where does your-- so basically you 02:05:20.630 --> 02:05:22.010 align:middle line:90% have control over two things. 02:05:22.010 --> 02:05:24.110 align:middle line:84% How many seats are open for sales 02:05:24.110 --> 02:05:27.090 align:middle line:84% and what is your price at that point in time. 02:05:27.090 --> 02:05:30.310 align:middle line:84% Are you capturing both of those or you're focusing only 02:05:30.310 --> 02:05:31.370 align:middle line:90% on the protection level? 02:05:31.370 --> 02:05:33.537 align:middle line:84% AUDIENCE: Right now I'm focusing only on protection, 02:05:33.537 --> 02:05:36.408 align:middle line:84% so I'm assuming there's only two price points on the plane, 02:05:36.408 --> 02:05:37.950 align:middle line:84% high pair and a low pair, and they're 02:05:37.950 --> 02:05:39.370 align:middle line:84% fixed for the life of the flight. 02:05:39.370 --> 02:05:41.390 align:middle line:84% We know that's not true, but let's assume 02:05:41.390 --> 02:05:44.410 align:middle line:84% those are the two kind of published levels. 02:05:44.410 --> 02:05:50.610 align:middle line:84% And I can take a booking of $100 today with certainty. 02:05:50.610 --> 02:05:56.230 align:middle line:84% Or I can wait for the possibility of $1,000 booking 02:05:56.230 --> 02:05:59.900 align:middle line:90% in 300 days from now. 02:05:59.900 --> 02:06:01.925 align:middle line:90% So those are the fare levels. 02:06:01.925 --> 02:06:03.800 align:middle line:84% HAZHIR RAHMANDAD: Yes, but these fare levels, 02:06:03.800 --> 02:06:06.800 align:middle line:84% they are not available to customers at all the time. 02:06:06.800 --> 02:06:08.660 align:middle line:90% Or are they? 02:06:08.660 --> 02:06:11.240 align:middle line:84% AUDIENCE: They are until the low fare sells out. 02:06:11.240 --> 02:06:13.160 align:middle line:84% That's what the protection level does. 02:06:13.160 --> 02:06:14.160 align:middle line:90% HAZHIR RAHMANDAD: I see. 02:06:14.160 --> 02:06:17.920 align:middle line:84% So I, as a customer, I can always buy at high fare. 02:06:17.920 --> 02:06:18.580 align:middle line:90% AUDIENCE: Yeah. 02:06:18.580 --> 02:06:21.620 align:middle line:84% HAZHIR RAHMANDAD: But I can only buy low fare 02:06:21.620 --> 02:06:22.933 align:middle line:90% within the protection level-- 02:06:22.933 --> 02:06:24.100 align:middle line:90% AUDIENCE: If it's available. 02:06:24.100 --> 02:06:25.517 align:middle line:84% HAZHIR RAHMANDAD: That is allowed. 02:06:25.517 --> 02:06:26.520 align:middle line:90% That is available to me. 02:06:26.520 --> 02:06:28.740 align:middle line:90% OK, got it, got it. 02:06:28.740 --> 02:06:29.500 align:middle line:90% So 02:06:29.500 --> 02:06:30.875 align:middle line:84% AUDIENCE: This is how they do it. 02:06:30.875 --> 02:06:32.060 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 02:06:32.060 --> 02:06:34.060 align:middle line:84% AUDIENCE: I mean, it gets really complicated-- 02:06:34.060 --> 02:06:35.640 align:middle line:84% HAZHIR RAHMANDAD: Because you have 20 levels. 02:06:35.640 --> 02:06:36.500 align:middle line:90% So it's not just-- 02:06:36.500 --> 02:06:37.960 align:middle line:84% AUDIENCE: Yeah, you have 20 levels. 02:06:37.960 --> 02:06:39.820 align:middle line:84% The price points are actually dynamic too 02:06:39.820 --> 02:06:42.580 align:middle line:84% because there's a separate pricing process based 02:06:42.580 --> 02:06:44.400 align:middle line:90% on competitive dynamics. 02:06:44.400 --> 02:06:47.080 align:middle line:84% And then those price points come in. 02:06:47.080 --> 02:06:50.580 align:middle line:84% And then there's an availability and optimization process that 02:06:50.580 --> 02:06:53.740 align:middle line:90% sets those protection levels. 02:06:53.740 --> 02:07:01.400 align:middle line:84% The whole point, though, is that the estimate of high-fare demand 02:07:01.400 --> 02:07:06.900 align:middle line:84% is through the filter of previous availability decisions. 02:07:06.900 --> 02:07:09.792 align:middle line:90% 02:07:09.792 --> 02:07:12.140 align:middle line:84% And it just continues feeding back forever. 02:07:12.140 --> 02:07:14.660 align:middle line:84% There's no check on forecast accuracy. 02:07:14.660 --> 02:07:17.920 align:middle line:84% There's no counterfactual setting. 02:07:17.920 --> 02:07:20.160 align:middle line:84% You make these availability decisions 02:07:20.160 --> 02:07:25.380 align:middle line:84% and then it just, the errors accumulate. 02:07:25.380 --> 02:07:28.140 align:middle line:84% HAZHIR RAHMANDAD: So let's see how that is implemented. 02:07:28.140 --> 02:07:30.460 align:middle line:84% That sounds good in theory, but let's see. 02:07:30.460 --> 02:07:32.560 align:middle line:84% AUDIENCE: So let's really focus just 02:07:32.560 --> 02:07:35.440 align:middle line:84% on the supply side, the right-hand side going 02:07:35.440 --> 02:07:38.060 align:middle line:84% from forecast into protection level. 02:07:38.060 --> 02:07:41.400 align:middle line:84% HAZHIR RAHMANDAD: So this forecast of high-fare demand 02:07:41.400 --> 02:07:43.360 align:middle line:90% is saying what exactly? 02:07:43.360 --> 02:07:50.280 align:middle line:84% It is saying how many total will be available will be-- 02:07:50.280 --> 02:07:53.960 align:middle line:84% how many total high-fare demand will come for this flight or? 02:07:53.960 --> 02:07:55.120 align:middle line:90% AUDIENCE: Yes. 02:07:55.120 --> 02:07:57.560 align:middle line:84% HAZHIR RAHMANDAD: So it is not from this point on. 02:07:57.560 --> 02:08:02.220 align:middle line:84% But this is for the whole history of that flight. 02:08:02.220 --> 02:08:03.640 align:middle line:90% AUDIENCE: Right. 02:08:03.640 --> 02:08:08.820 align:middle line:84% So how many units of high-fare demand 02:08:08.820 --> 02:08:13.380 align:middle line:84% can I expect for a particular flight? 02:08:13.380 --> 02:08:15.965 align:middle line:90% Yeah. 02:08:15.965 --> 02:08:16.840 align:middle line:90% HAZHIR RAHMANDAD: OK. 02:08:16.840 --> 02:08:19.900 align:middle line:90% 02:08:19.900 --> 02:08:20.760 align:middle line:90% Then what happens. 02:08:20.760 --> 02:08:21.802 align:middle line:90% AUDIENCE: It's a smooth-- 02:08:21.802 --> 02:08:24.040 align:middle line:84% So then that gives you an average. 02:08:24.040 --> 02:08:26.820 align:middle line:84% HAZHIR RAHMANDAD: But that is coming 02:08:26.820 --> 02:08:31.020 align:middle line:90% from this high-fare sales. 02:08:31.020 --> 02:08:31.780 align:middle line:90% AUDIENCE: Yeah. 02:08:31.780 --> 02:08:38.895 align:middle line:84% HAZHIR RAHMANDAD: And high-fare sales is in seats per month. 02:08:38.895 --> 02:08:40.020 align:middle line:90% AUDIENCE: Seats per flight. 02:08:40.020 --> 02:08:44.420 align:middle line:84% So I haven't formulated sales yet. 02:08:44.420 --> 02:08:46.140 align:middle line:90% I don't have sales yet. 02:08:46.140 --> 02:08:48.055 align:middle line:90% This is just-- 02:08:48.055 --> 02:08:49.680 align:middle line:84% HAZHIR RAHMANDAD: So what is this then? 02:08:49.680 --> 02:08:50.440 align:middle line:90% I'm just not-- 02:08:50.440 --> 02:08:54.160 align:middle line:90% 02:08:54.160 --> 02:08:57.440 align:middle line:84% I feel that there is a little bit of a unit inconsistency 02:08:57.440 --> 02:09:01.160 align:middle line:84% or unit confusion for me to reading through this 02:09:01.160 --> 02:09:07.040 align:middle line:84% because sales for me is in unit-- it's seats per month 02:09:07.040 --> 02:09:08.260 align:middle line:90% or something like that. 02:09:08.260 --> 02:09:12.560 align:middle line:84% It is a flow of things that are being sold, 02:09:12.560 --> 02:09:17.420 align:middle line:84% and then you are getting that-- you're using sales. 02:09:17.420 --> 02:09:19.180 align:middle line:90% I mean, this is called sales. 02:09:19.180 --> 02:09:21.080 align:middle line:84% Maybe it is not what you-- maybe it 02:09:21.080 --> 02:09:23.100 align:middle line:84% is cumulative sales or something else. 02:09:23.100 --> 02:09:24.140 align:middle line:90% I don't know what it-- 02:09:24.140 --> 02:09:25.640 align:middle line:90% AUDIENCE: It should be-- 02:09:25.640 --> 02:09:27.182 align:middle line:90% so I have the-- 02:09:27.182 --> 02:09:28.640 align:middle line:84% HAZHIR RAHMANDAD: And demand for me 02:09:28.640 --> 02:09:34.780 align:middle line:84% is also in the units of seats per month or something. 02:09:34.780 --> 02:09:38.640 align:middle line:90% 02:09:38.640 --> 02:09:41.140 align:middle line:84% Unless you have a different conception of demand. 02:09:41.140 --> 02:09:45.520 align:middle line:84% Because otherwise I get very confused what period of time 02:09:45.520 --> 02:09:46.420 align:middle line:90% we are talking about. 02:09:46.420 --> 02:09:49.960 align:middle line:84% Because you are running through the life of a single flight 02:09:49.960 --> 02:09:54.155 align:middle line:84% from 300 days before, or 360 days before until the date. 02:09:54.155 --> 02:09:54.780 align:middle line:90% AUDIENCE: Yeah. 02:09:54.780 --> 02:09:57.500 align:middle line:84% HAZHIR RAHMANDAD: And during this time, 02:09:57.500 --> 02:09:59.740 align:middle line:84% when you use the units of seats, I'm 02:09:59.740 --> 02:10:03.020 align:middle line:84% not sure what you're talking about for a flow 02:10:03.020 --> 02:10:04.633 align:middle line:90% variable like demand or sales. 02:10:04.633 --> 02:10:05.300 align:middle line:90% AUDIENCE: Right. 02:10:05.300 --> 02:10:06.758 align:middle line:84% HAZHIR RAHMANDAD: If you're talking 02:10:06.758 --> 02:10:08.540 align:middle line:84% about the number of seats remaining, which 02:10:08.540 --> 02:10:11.420 align:middle line:84% is a stock variable, I fully follow that, 02:10:11.420 --> 02:10:13.240 align:middle line:84% or the number of seats that have been sold, 02:10:13.240 --> 02:10:14.900 align:middle line:90% I can fully follow that. 02:10:14.900 --> 02:10:18.020 align:middle line:84% But as soon as you talk about sales or demand 02:10:18.020 --> 02:10:21.180 align:middle line:84% with a unit of seats, I get confused. 02:10:21.180 --> 02:10:22.700 align:middle line:90% AUDIENCE: Yeah. 02:10:22.700 --> 02:10:25.820 align:middle line:84% HAZHIR RAHMANDAD: Do you know what you're doing here 02:10:25.820 --> 02:10:27.720 align:middle line:84% or you're also confused about it? 02:10:27.720 --> 02:10:30.340 align:middle line:90% That's the question. 02:10:30.340 --> 02:10:32.700 align:middle line:90% AUDIENCE: So I might be-- 02:10:32.700 --> 02:10:35.420 align:middle line:84% I think I do have some confusion because I haven't formulated 02:10:35.420 --> 02:10:37.220 align:middle line:90% the demand side. 02:10:37.220 --> 02:10:39.840 align:middle line:84% HAZHIR RAHMANDAD: No, but you can have an exogenous demand. 02:10:39.840 --> 02:10:42.340 align:middle line:90% Just call it exogenous demand. 02:10:42.340 --> 02:10:44.560 align:middle line:84% What is the unit of that variable? 02:10:44.560 --> 02:10:47.820 align:middle line:90% 02:10:47.820 --> 02:10:50.306 align:middle line:90% AUDIENCE: Right. 02:10:50.306 --> 02:10:57.760 align:middle line:90% It is-- OK. 02:10:57.760 --> 02:11:10.260 align:middle line:84% It's units of demand, per this is the main confusion. 02:11:10.260 --> 02:11:15.260 align:middle line:84% It's people don't demand a particular flight. 02:11:15.260 --> 02:11:20.160 align:middle line:84% They demand a fare service x days from now. 02:11:20.160 --> 02:11:34.800 align:middle line:84% So it's units per kind of booking interval. 02:11:34.800 --> 02:11:36.880 align:middle line:84% HAZHIR RAHMANDAD: So I go online. 02:11:36.880 --> 02:11:41.760 align:middle line:84% I'm going to click to try to buy this flight. 02:11:41.760 --> 02:11:44.960 align:middle line:84% How many people like that happen every day is 02:11:44.960 --> 02:11:47.360 align:middle line:90% what I'm calling demand. 02:11:47.360 --> 02:11:52.340 align:middle line:84% Now you can have it demand for that whole competition 02:11:52.340 --> 02:11:54.500 align:middle line:84% between five different flights on the same route 02:11:54.500 --> 02:11:59.800 align:middle line:84% for the same day, then you can, yes, have a separate allocation. 02:11:59.800 --> 02:12:04.500 align:middle line:84% But right now we are focusing on a single flight. 02:12:04.500 --> 02:12:05.300 align:middle line:90% AUDIENCE: I'm not. 02:12:05.300 --> 02:12:06.840 align:middle line:90% So it's kind of evolved. 02:12:06.840 --> 02:12:14.640 align:middle line:84% I'm trying to forecast how many units of high-fare demand-- 02:12:14.640 --> 02:12:19.460 align:middle line:90% 02:12:19.460 --> 02:12:20.816 align:middle line:90% I see what you're saying. 02:12:20.816 --> 02:12:26.540 align:middle line:90% 02:12:26.540 --> 02:12:29.620 align:middle line:84% HAZHIR RAHMANDAD: So I think this confusion 02:12:29.620 --> 02:12:31.200 align:middle line:90% is pretty fundamental to this. 02:12:31.200 --> 02:12:33.820 align:middle line:84% That is, without solving that, you can't really 02:12:33.820 --> 02:12:37.520 align:middle line:84% build a model of supply and demand here 02:12:37.520 --> 02:12:42.860 align:middle line:84% if the units of supply and demand are not per unit of time. 02:12:42.860 --> 02:12:45.580 align:middle line:84% The version of the model you had last week 02:12:45.580 --> 02:12:48.960 align:middle line:84% seemed to be more consistent on the demand side, how 02:12:48.960 --> 02:12:52.544 align:middle line:90% it was thinking about it. 02:12:52.544 --> 02:13:00.680 align:middle line:84% Right now, the protection level is problematic for me 02:13:00.680 --> 02:13:03.440 align:middle line:84% because the protection, what you can protect, 02:13:03.440 --> 02:13:06.960 align:middle line:84% depends on how many you have already sold, 02:13:06.960 --> 02:13:09.645 align:middle line:90% if I understand it correctly. 02:13:09.645 --> 02:13:11.520 align:middle line:84% AUDIENCE: If you have a [INAUDIBLE] So right. 02:13:11.520 --> 02:13:14.440 align:middle line:90% 02:13:14.440 --> 02:13:16.800 align:middle line:84% HAZHIR RAHMANDAD: So you have seats that are unsold. 02:13:16.800 --> 02:13:17.080 align:middle line:90% AUDIENCE: Yes. 02:13:17.080 --> 02:13:18.747 align:middle line:84% HAZHIR RAHMANDAD: They can be protected. 02:13:18.747 --> 02:13:25.640 align:middle line:84% But once you have sold a seat it is out of the protection 02:13:25.640 --> 02:13:26.940 align:middle line:90% consideration. 02:13:26.940 --> 02:13:28.480 align:middle line:90% It is already sold. 02:13:28.480 --> 02:13:31.880 align:middle line:84% So the protection level, if by that you 02:13:31.880 --> 02:13:35.800 align:middle line:84% mean how many seats are sellable or not sellable 02:13:35.800 --> 02:13:40.600 align:middle line:84% from those that are generally still not sold, 02:13:40.600 --> 02:13:44.060 align:middle line:84% that should depend on how many you have sold. 02:13:44.060 --> 02:13:45.560 align:middle line:84% And right now, you don't have that. 02:13:45.560 --> 02:13:47.140 align:middle line:84% So you just are changing this thing 02:13:47.140 --> 02:13:49.420 align:middle line:90% without having any constraint. 02:13:49.420 --> 02:13:51.660 align:middle line:84% AUDIENCE: Because these are for different flights. 02:13:51.660 --> 02:13:54.800 align:middle line:84% The flight is born with a protection level. 02:13:54.800 --> 02:13:57.460 align:middle line:84% And in this current iteration, that protection level 02:13:57.460 --> 02:13:59.460 align:middle line:84% is fixed for the life of the flight. 02:13:59.460 --> 02:14:01.280 align:middle line:84% In reality it's a dynamic process. 02:14:01.280 --> 02:14:03.060 align:middle line:84% HAZHIR RAHMANDAD: So I'm now confused. 02:14:03.060 --> 02:14:05.040 align:middle line:84% You said these are for different flights. 02:14:05.040 --> 02:14:06.140 align:middle line:90% What do you mean? 02:14:06.140 --> 02:14:08.060 align:middle line:84% AUDIENCE: Because, OK, I think, yeah, 02:14:08.060 --> 02:14:18.940 align:middle line:84% this is the [INAUDIBLE] I think I need to get my units of demand 02:14:18.940 --> 02:14:26.220 align:middle line:84% correct, and then I can more clearly lay out how this-- 02:14:26.220 --> 02:14:29.660 align:middle line:84% HAZHIR RAHMANDAD: But are you modeling a single flight 02:14:29.660 --> 02:14:34.120 align:middle line:84% over the year before it, or now you're not doing that? 02:14:34.120 --> 02:14:36.940 align:middle line:90% AUDIENCE: I'm modeling multiple. 02:14:36.940 --> 02:14:39.180 align:middle line:90% Every day a new flight is born. 02:14:39.180 --> 02:14:45.280 align:middle line:84% And so I'm modeling multiple flights 02:14:45.280 --> 02:14:47.358 align:middle line:90% over their entire lifetime. 02:14:47.358 --> 02:14:49.400 align:middle line:84% HAZHIR RAHMANDAD: So each flight has [INAUDIBLE]? 02:14:49.400 --> 02:14:51.500 align:middle line:84% Each flight is one protection level? 02:14:51.500 --> 02:14:52.533 align:middle line:90% AUDIENCE: Yes. 02:14:52.533 --> 02:14:54.200 align:middle line:84% HAZHIR RAHMANDAD: But why did you switch 02:14:54.200 --> 02:14:55.660 align:middle line:90% to this level of analysis? 02:14:55.660 --> 02:14:59.040 align:middle line:84% Because last week you had a single flight 02:14:59.040 --> 02:15:01.003 align:middle line:90% over its one-year period. 02:15:01.003 --> 02:15:01.920 align:middle line:90% AUDIENCE: I switched-- 02:15:01.920 --> 02:15:03.100 align:middle line:84% HAZHIR RAHMANDAD: That I was following. 02:15:03.100 --> 02:15:04.900 align:middle line:84% Now I'm not following what you are doing. 02:15:04.900 --> 02:15:07.240 align:middle line:84% AUDIENCE: So the idea is, I feel like I 02:15:07.240 --> 02:15:12.560 align:middle line:84% need to have both horizons accounted 02:15:12.560 --> 02:15:15.960 align:middle line:84% for if I want to capture this feedback effect. 02:15:15.960 --> 02:15:30.320 align:middle line:84% Because the goal is to show how these incorrect assumptions 02:15:30.320 --> 02:15:31.720 align:middle line:90% can-- 02:15:31.720 --> 02:15:35.440 align:middle line:84% HAZHIR RAHMANDAD: So I'm OK with switching the level of analysis 02:15:35.440 --> 02:15:37.740 align:middle line:90% if that's justified. 02:15:37.740 --> 02:15:44.030 align:middle line:84% But I need to then understand what exactly 02:15:44.030 --> 02:15:46.410 align:middle line:90% this level of analysis is. 02:15:46.410 --> 02:15:50.110 align:middle line:90% So you have multiple flights. 02:15:50.110 --> 02:15:53.310 align:middle line:84% And based on the previous flights and their protection 02:15:53.310 --> 02:16:01.430 align:middle line:84% levels, you were getting some total fares, high-fare 02:16:01.430 --> 02:16:06.070 align:middle line:90% versus low-fare of the demand. 02:16:06.070 --> 02:16:09.590 align:middle line:84% And then based on that you're changing your protection 02:16:09.590 --> 02:16:11.127 align:middle line:90% level for the next flights. 02:16:11.127 --> 02:16:11.710 align:middle line:90% AUDIENCE: Yes. 02:16:11.710 --> 02:16:14.168 align:middle line:84% HAZHIR RAHMANDAD: So that's the story you are telling here. 02:16:14.168 --> 02:16:15.890 align:middle line:84% OK, so I was confused at the beginning. 02:16:15.890 --> 02:16:17.390 align:middle line:84% I was starting with the assumption 02:16:17.390 --> 02:16:19.290 align:middle line:84% that you are looking at a single flight. 02:16:19.290 --> 02:16:21.550 align:middle line:84% So a lot of my comments and questions 02:16:21.550 --> 02:16:24.903 align:middle line:84% might be misplaced now given what you were trying to do. 02:16:24.903 --> 02:16:25.570 align:middle line:90% AUDIENCE: Sorry. 02:16:25.570 --> 02:16:26.795 align:middle line:90% It's hard for me-- yeah. 02:16:26.795 --> 02:16:27.670 align:middle line:90% HAZHIR RAHMANDAD: OK. 02:16:27.670 --> 02:16:33.549 align:middle line:84% So now with that definition of protection level, 02:16:33.549 --> 02:16:37.330 align:middle line:84% the total sales you haven't yet formulated, you said. 02:16:37.330 --> 02:16:39.610 align:middle line:84% AUDIENCE: Right, because then I need some kind of rule 02:16:39.610 --> 02:16:42.549 align:middle line:84% for what I call diversion itself. 02:16:42.549 --> 02:16:45.129 align:middle line:84% So a diversion is when someone who's willing 02:16:45.129 --> 02:16:46.528 align:middle line:90% to pay a high fare-- 02:16:46.528 --> 02:16:47.570 align:middle line:90% HAZHIR RAHMANDAD: Got it. 02:16:47.570 --> 02:16:50.530 align:middle line:84% AUDIENCE: --ends up buying down because it's available. 02:16:50.530 --> 02:16:52.450 align:middle line:84% So you could be a flexible customer, but you-- 02:16:52.450 --> 02:16:52.809 align:middle line:90% [INTERPOSING VOICES] 02:16:52.809 --> 02:16:53.490 align:middle line:90% AUDIENCE: --at the end. 02:16:53.490 --> 02:16:53.910 align:middle line:90% HAZHIR RAHMANDAD: Got it. 02:16:53.910 --> 02:16:54.650 align:middle line:90% [INTERPOSING VOICES] 02:16:54.650 --> 02:16:56.230 align:middle line:84% AUDIENCE: So I haven't done that yet. 02:16:56.230 --> 02:16:57.272 align:middle line:90% HAZHIR RAHMANDAD: Got it. 02:16:57.272 --> 02:17:00.453 align:middle line:84% AUDIENCE: But the point is that we know that behavior happens. 02:17:00.453 --> 02:17:01.370 align:middle line:90% HAZHIR RAHMANDAD: Yes. 02:17:01.370 --> 02:17:05.730 align:middle line:84% So basically you need now some input 02:17:05.730 --> 02:17:08.809 align:middle line:84% in terms of customers who are willing to pay 02:17:08.809 --> 02:17:11.889 align:middle line:84% the high fare versus customers who are only 02:17:11.889 --> 02:17:13.510 align:middle line:90% willing to pay the low fare. 02:17:13.510 --> 02:17:14.290 align:middle line:90% AUDIENCE: Yes. 02:17:14.290 --> 02:17:16.670 align:middle line:84% HAZHIR RAHMANDAD: You need these two types of customers. 02:17:16.670 --> 02:17:17.545 align:middle line:90% AUDIENCE: [INAUDIBLE] 02:17:17.545 --> 02:17:19.889 align:middle line:84% HAZHIR RAHMANDAD: They, for each flight, 02:17:19.889 --> 02:17:23.290 align:middle line:84% they would have a realization, a random realization 02:17:23.290 --> 02:17:28.170 align:middle line:84% that you don't know, but you try to make a forecast based 02:17:28.170 --> 02:17:31.090 align:middle line:84% on what you could observe in the previous round. 02:17:31.090 --> 02:17:32.490 align:middle line:90% AUDIENCE: Yes. 02:17:32.490 --> 02:17:35.830 align:middle line:84% HAZHIR RAHMANDAD: And then I'm talking about rounds, 02:17:35.830 --> 02:17:37.330 align:middle line:84% even though this is continuous time, 02:17:37.330 --> 02:17:39.590 align:middle line:90% but it's OK to think about it. 02:17:39.590 --> 02:17:42.270 align:middle line:90% You can make that translation. 02:17:42.270 --> 02:17:48.469 align:middle line:84% And you then adjust your high-fare versus lo-fare 02:17:48.469 --> 02:17:52.910 align:middle line:84% for demand projection based on the observed 02:17:52.910 --> 02:17:56.209 align:middle line:90% values of realized sales. 02:17:56.209 --> 02:17:57.110 align:middle line:90% AUDIENCE: Yes. 02:17:57.110 --> 02:17:58.790 align:middle line:84% HAZHIR RAHMANDAD: And then you change the protection level 02:17:58.790 --> 02:17:59.530 align:middle line:90% accordingly. 02:17:59.530 --> 02:18:01.709 align:middle line:84% AUDIENCE: Yeah, and the rule for changing-- 02:18:01.709 --> 02:18:05.770 align:middle line:84% setting the protection level and the delays, 02:18:05.770 --> 02:18:08.910 align:middle line:84% because the thing is, you take in bookings during the daytime 02:18:08.910 --> 02:18:12.230 align:middle line:84% and then overnight, you run this optimization process where 02:18:12.230 --> 02:18:17.190 align:middle line:84% you reset the protection levels for tomorrow. 02:18:17.190 --> 02:18:19.230 align:middle line:84% I realize now, because I haven't right now, 02:18:19.230 --> 02:18:24.030 align:middle line:84% that the protections aren't dynamic yet 02:18:24.030 --> 02:18:27.030 align:middle line:90% over the life of the flight. 02:18:27.030 --> 02:18:29.930 align:middle line:90% That's not fully captured. 02:18:29.930 --> 02:18:34.030 align:middle line:90% But yeah, I like the-- 02:18:34.030 --> 02:18:37.179 align:middle line:84% so I feel good going from demand and protection. 02:18:37.179 --> 02:18:38.096 align:middle line:90% HAZHIR RAHMANDAD: Yes. 02:18:38.096 --> 02:18:41.850 align:middle line:84% Now there is this kind of a source of confusion for you, 02:18:41.850 --> 02:18:46.570 align:middle line:84% that will persist, that comes from two different levels 02:18:46.570 --> 02:18:49.850 align:middle line:84% of analysis, one as the single flight 02:18:49.850 --> 02:18:52.730 align:middle line:90% versus a portfolio of flights. 02:18:52.730 --> 02:18:57.010 align:middle line:84% So single flight is easier, more operationally specific 02:18:57.010 --> 02:18:58.850 align:middle line:84% to go to the details of it because you 02:18:58.850 --> 02:19:02.150 align:middle line:84% can say what is happening for every part of it. 02:19:02.150 --> 02:19:07.090 align:middle line:84% But then it doesn't give you the multi-flight feedback 02:19:07.090 --> 02:19:09.570 align:middle line:90% that you are looking for. 02:19:09.570 --> 02:19:12.230 align:middle line:84% AUDIENCE: So demand isn't really for a single flight. 02:19:12.230 --> 02:19:16.187 align:middle line:90% 02:19:16.187 --> 02:19:17.770 align:middle line:84% HAZHIR RAHMANDAD: So the approximation 02:19:17.770 --> 02:19:21.250 align:middle line:84% that the demand is for a single flight is not 02:19:21.250 --> 02:19:24.570 align:middle line:84% a bad approximation, because somebody wants 02:19:24.570 --> 02:19:26.830 align:middle line:90% to go somewhere on some date. 02:19:26.830 --> 02:19:28.730 align:middle line:84% Now, yes, people have some flexibility 02:19:28.730 --> 02:19:31.629 align:middle line:84% around those dates and some flexibility 02:19:31.629 --> 02:19:34.549 align:middle line:84% whether they take the trip or not, but all of those 02:19:34.549 --> 02:19:41.230 align:middle line:84% can be collapsed into a realized demand, which 02:19:41.230 --> 02:19:44.889 align:middle line:84% you were trying to model in the previous week version. 02:19:44.889 --> 02:19:47.549 align:middle line:90% AUDIENCE: So how do you think? 02:19:47.549 --> 02:19:53.510 align:middle line:84% So because of this problem of single-flight booking over time 02:19:53.510 --> 02:19:58.390 align:middle line:84% versus having multiple flights booking over time, 02:19:58.390 --> 02:20:02.110 align:middle line:84% I think I need to have both kind of-- 02:20:02.110 --> 02:20:04.790 align:middle line:84% HAZHIR RAHMANDAD: So for your longer-term question 02:20:04.790 --> 02:20:07.870 align:middle line:90% of connecting a single-- 02:20:07.870 --> 02:20:09.870 align:middle line:84% this decision of protection level 02:20:09.870 --> 02:20:12.730 align:middle line:84% setting to the experience across multiple flights, 02:20:12.730 --> 02:20:14.210 align:middle line:90% you need multiple flights. 02:20:14.210 --> 02:20:16.950 align:middle line:84% Otherwise you can't close that loop. 02:20:16.950 --> 02:20:20.070 align:middle line:84% That goes through learning from your experience. 02:20:20.070 --> 02:20:23.170 align:middle line:84% So for that reason this version is fine. 02:20:23.170 --> 02:20:24.170 align:middle line:90% It is helpful. 02:20:24.170 --> 02:20:24.910 align:middle line:90% AUDIENCE: Yeah. 02:20:24.910 --> 02:20:26.350 align:middle line:84% HAZHIR RAHMANDAD: But I'm worried 02:20:26.350 --> 02:20:33.730 align:middle line:84% whether you can manage this kind of a more abstract level 02:20:33.730 --> 02:20:37.610 align:middle line:84% of formulating the protection level 02:20:37.610 --> 02:20:41.170 align:middle line:84% as it continuously changes from feedback 02:20:41.170 --> 02:20:44.130 align:middle line:84% across multiple flights, whether that 02:20:44.130 --> 02:20:48.270 align:middle line:84% is going to be easier to manage or, I mean, 02:20:48.270 --> 02:20:52.010 align:middle line:84% it wouldn't be easier whether you would come up with something 02:20:52.010 --> 02:20:54.810 align:middle line:90% that works fine for you. 02:20:54.810 --> 02:20:58.930 align:middle line:84% Or you may take the longer route of building 02:20:58.930 --> 02:21:05.090 align:middle line:84% a single-flight model, fully getting that to work, 02:21:05.090 --> 02:21:07.410 align:middle line:84% then moving from that into something 02:21:07.410 --> 02:21:10.630 align:middle line:84% more aggregate, or perhaps even not more aggregate, 02:21:10.630 --> 02:21:14.490 align:middle line:84% but kind of a sequential that, OK, I did this. 02:21:14.490 --> 02:21:16.890 align:middle line:84% Now, from this I learned for the next flight, 02:21:16.890 --> 02:21:19.530 align:middle line:84% and then from the next flight I learned from the next flight 02:21:19.530 --> 02:21:21.350 align:middle line:84% after that or something like that. 02:21:21.350 --> 02:21:26.570 align:middle line:84% So that way you would capture the feedback 02:21:26.570 --> 02:21:28.730 align:middle line:90% across multiple flights. 02:21:28.730 --> 02:21:31.370 align:middle line:84% So that's another way of doing it. 02:21:31.370 --> 02:21:37.350 align:middle line:90% 02:21:37.350 --> 02:21:41.110 align:middle line:84% I think given how much confused we 02:21:41.110 --> 02:21:44.070 align:middle line:84% have been so far, go with the one 02:21:44.070 --> 02:21:47.930 align:middle line:84% that you can conceptualize very clearly and cleanly, 02:21:47.930 --> 02:21:50.530 align:middle line:84% even if that is only for a single flight. 02:21:50.530 --> 02:21:53.990 align:middle line:84% I don't think-- until you can do the single flight well, 02:21:53.990 --> 02:21:58.452 align:middle line:84% it would be very hard to go into the higher level of aggregation. 02:21:58.452 --> 02:22:00.910 align:middle line:84% AUDIENCE: But I don't think the single-flight case is going 02:22:00.910 --> 02:22:02.410 align:middle line:90% to have the feedback I want. 02:22:02.410 --> 02:22:03.830 align:middle line:90% HAZHIR RAHMANDAD: It doesn't. 02:22:03.830 --> 02:22:07.110 align:middle line:84% Don't capture that feedback in your first round. 02:22:07.110 --> 02:22:08.632 align:middle line:90% It's not working. 02:22:08.632 --> 02:22:10.090 align:middle line:84% Right now you don't have the sales. 02:22:10.090 --> 02:22:12.150 align:middle line:90% You don't have the demand. 02:22:12.150 --> 02:22:14.030 align:middle line:84% You don't have even the protection level 02:22:14.030 --> 02:22:17.830 align:middle line:90% right for the single flight. 02:22:17.830 --> 02:22:22.390 align:middle line:84% So I would get the single-flight mechanics 02:22:22.390 --> 02:22:32.810 align:middle line:84% right with the demand changing dynamically, what 02:22:32.810 --> 02:22:36.090 align:middle line:84% you were doing last week, and with the protection level 02:22:36.090 --> 02:22:40.050 align:middle line:84% changing dynamically, as what you're doing this week. 02:22:40.050 --> 02:22:42.750 align:middle line:84% Now, if you want to keep the price only at two tiers, 02:22:42.750 --> 02:22:43.430 align:middle line:90% that's fine. 02:22:43.430 --> 02:22:45.988 align:middle line:90% 02:22:45.988 --> 02:22:46.530 align:middle line:90% AUDIENCE: OK. 02:22:46.530 --> 02:22:49.355 align:middle line:90% 02:22:49.355 --> 02:22:50.230 align:middle line:90% HAZHIR RAHMANDAD: OK. 02:22:50.230 --> 02:22:50.830 align:middle line:90% We can-- 02:22:50.830 --> 02:22:51.690 align:middle line:90% AUDIENCE: No, no. 02:22:51.690 --> 02:22:52.430 align:middle line:90% It makes sense. 02:22:52.430 --> 02:22:54.847 align:middle line:84% HAZHIR RAHMANDAD: Yeah, we can have a little bit more chat 02:22:54.847 --> 02:22:55.950 align:middle line:90% later, the two of us. 02:22:55.950 --> 02:22:57.990 align:middle line:84% I want to get to at least a couple more models. 02:22:57.990 --> 02:22:58.950 align:middle line:90% So let's see. 02:22:58.950 --> 02:23:03.670 align:middle line:84% We have Hassan's-- so eligibility 1, 2, 3. 02:23:03.670 --> 02:23:05.110 align:middle line:90% Which one should I open up? 02:23:05.110 --> 02:23:07.443 align:middle line:84% AUDIENCE: Can we look at the first one and the third one 02:23:07.443 --> 02:23:08.070 align:middle line:90% after that? 02:23:08.070 --> 02:23:08.530 align:middle line:90% AUDIENCE: [INAUDIBLE] 02:23:08.530 --> 02:23:09.863 align:middle line:90% HAZHIR RAHMANDAD: Just pick one. 02:23:09.863 --> 02:23:10.890 align:middle line:90% We have 10 minutes. 02:23:10.890 --> 02:23:12.530 align:middle line:90% AUDIENCE: OK. 02:23:12.530 --> 02:23:13.330 align:middle line:90% Sure. 02:23:13.330 --> 02:23:15.530 align:middle line:84% Let's take a look at one and then I'll-- 02:23:15.530 --> 02:23:19.030 align:middle line:84% so I've been quite lost with this model. 02:23:19.030 --> 02:23:21.670 align:middle line:84% I think you've seen the previous iteration. 02:23:21.670 --> 02:23:23.770 align:middle line:84% I tried to do some compartmental thing. 02:23:23.770 --> 02:23:27.150 align:middle line:84% Then I started imagining as simple as a co-flow 02:23:27.150 --> 02:23:32.550 align:middle line:84% as I can, which is patients dependent on their smoking 02:23:32.550 --> 02:23:34.910 align:middle line:84% behavior will become eligible for screening, 02:23:34.910 --> 02:23:42.210 align:middle line:84% and then dependent on their actual health, 02:23:42.210 --> 02:23:45.550 align:middle line:84% which is the nodules, they may show symptoms 02:23:45.550 --> 02:23:49.610 align:middle line:84% or show positive in whatever they show up for screening. 02:23:49.610 --> 02:23:53.550 align:middle line:84% But then this triggered the idea that this co-flow, what 02:23:53.550 --> 02:23:56.270 align:middle line:84% it has with it is that it's not average attributes 02:23:56.270 --> 02:23:57.190 align:middle line:90% for [INAUDIBLE]. 02:23:57.190 --> 02:23:58.148 align:middle line:90% HAZHIR RAHMANDAD: Wait. 02:23:58.148 --> 02:24:01.270 align:middle line:84% So you're talking about population 02:24:01.270 --> 02:24:04.390 align:middle line:84% of people who maybe smoking or not smoking 02:24:04.390 --> 02:24:08.790 align:middle line:84% and then be eligible for this screening test. 02:24:08.790 --> 02:24:10.630 align:middle line:84% And then when they are screened they 02:24:10.630 --> 02:24:14.150 align:middle line:84% could be put into the category of diagnosed 02:24:14.150 --> 02:24:17.665 align:middle line:90% but incorrectly diagnosed or-- 02:24:17.665 --> 02:24:19.290 align:middle line:84% AUDIENCE: I'm not even going there yet. 02:24:19.290 --> 02:24:22.510 align:middle line:84% HAZHIR RAHMANDAD: So they can be screened. 02:24:22.510 --> 02:24:25.720 align:middle line:84% AUDIENCE: If they are screened they either get a negative, 02:24:25.720 --> 02:24:27.940 align:middle line:84% you could go home, or they are sent to treatment. 02:24:27.940 --> 02:24:28.800 align:middle line:84% HAZHIR RAHMANDAD: Sent to treatment. 02:24:28.800 --> 02:24:30.280 align:middle line:84% AUDIENCE: Then there's either something in between the two, 02:24:30.280 --> 02:24:30.960 align:middle line:90% but [INAUDIBLE]. 02:24:30.960 --> 02:24:31.877 align:middle line:90% HAZHIR RAHMANDAD: Yes. 02:24:31.877 --> 02:24:34.640 align:middle line:84% OK, so for that, why are you using co-flows? 02:24:34.640 --> 02:24:36.820 align:middle line:84% I'm not following what the co-flows do for you. 02:24:36.820 --> 02:24:39.940 align:middle line:84% AUDIENCE: Because the chance of developing lung cancer 02:24:39.940 --> 02:24:42.420 align:middle line:90% depends on smoking history. 02:24:42.420 --> 02:24:43.540 align:middle line:90% HAZHIR RAHMANDAD: OK. 02:24:43.540 --> 02:24:46.300 align:middle line:84% AUDIENCE: Which means if people have smoked for more 02:24:46.300 --> 02:24:50.120 align:middle line:84% and for longer there's a higher chance of developing 02:24:50.120 --> 02:24:50.620 align:middle line:90% [INAUDIBLE]. 02:24:50.620 --> 02:24:53.540 align:middle line:84% HAZHIR RAHMANDAD: So this SP PY, what is it? 02:24:53.540 --> 02:24:56.227 align:middle line:84% AUDIENCE: So smoking population, PY is pack years. 02:24:56.227 --> 02:24:57.060 align:middle line:90% [INTERPOSING VOICES] 02:24:57.060 --> 02:24:58.880 align:middle line:84% HAZHIR RAHMANDAD: So you are capturing-- 02:24:58.880 --> 02:25:01.742 align:middle line:90% you're trying to track how-- 02:25:01.742 --> 02:25:02.700 align:middle line:90% AUDIENCE: Much smoking. 02:25:02.700 --> 02:25:04.520 align:middle line:84% HAZHIR RAHMANDAD: --much smoking these people are doing. 02:25:04.520 --> 02:25:05.420 align:middle line:90% AUDIENCE: Yeah. 02:25:05.420 --> 02:25:07.587 align:middle line:84% HAZHIR RAHMANDAD: This is actually an important kind 02:25:07.587 --> 02:25:09.760 align:middle line:90% of a general lesson here. 02:25:09.760 --> 02:25:13.520 align:middle line:84% So here Hassan is trying to do two things. 02:25:13.520 --> 02:25:18.140 align:middle line:84% One is to have a flow of people between different states, 02:25:18.140 --> 02:25:21.540 align:middle line:84% but also tracking their frequency 02:25:21.540 --> 02:25:25.440 align:middle line:84% of doing some activity that impacts their ability 02:25:25.440 --> 02:25:26.900 align:middle line:90% to move up and down. 02:25:26.900 --> 02:25:29.480 align:middle line:90% 02:25:29.480 --> 02:25:32.300 align:middle line:84% There are two ways to manage this type of a problem. 02:25:32.300 --> 02:25:35.080 align:middle line:84% One is to break the population from the beginning 02:25:35.080 --> 02:25:40.020 align:middle line:84% into high smokers, low smokers type of a situation 02:25:40.020 --> 02:25:43.320 align:middle line:84% so that you always have these high smokers 02:25:43.320 --> 02:25:49.040 align:middle line:84% being their own flows, moving across different states. 02:25:49.040 --> 02:25:50.725 align:middle line:90% AUDIENCE: Could we have flows-- 02:25:50.725 --> 02:25:52.100 align:middle line:84% HAZHIR RAHMANDAD: You could have. 02:25:52.100 --> 02:25:54.820 align:middle line:84% So high smokers can become lower smokers, 02:25:54.820 --> 02:25:58.447 align:middle line:84% not really if it is-- but low smokers can become high smokers. 02:25:58.447 --> 02:25:59.280 align:middle line:90% [INTERPOSING VOICES] 02:25:59.280 --> 02:26:01.260 align:middle line:84% HAZHIR RAHMANDAD: Yes, so that can always happen. 02:26:01.260 --> 02:26:02.593 align:middle line:90% AUDIENCE: Because it starts to-- 02:26:02.593 --> 02:26:04.640 align:middle line:84% HAZHIR RAHMANDAD: Yes, so basically start-- 02:26:04.640 --> 02:26:06.980 align:middle line:84% The alternative that you're trying to do, 02:26:06.980 --> 02:26:10.920 align:middle line:84% which is using co-flows becomes tricky. 02:26:10.920 --> 02:26:13.500 align:middle line:84% I don't recommend it in this case. 02:26:13.500 --> 02:26:15.540 align:middle line:84% In simpler cases, it is possible. 02:26:15.540 --> 02:26:21.860 align:middle line:84% But when you have these many different groups that people 02:26:21.860 --> 02:26:28.660 align:middle line:84% be non-screened, screened, under treatment, 02:26:28.660 --> 02:26:32.780 align:middle line:84% the density of people in terms of the smoking levels 02:26:32.780 --> 02:26:33.635 align:middle line:90% will change. 02:26:33.635 --> 02:26:34.260 align:middle line:90% AUDIENCE: Yeah. 02:26:34.260 --> 02:26:36.640 align:middle line:84% HAZHIR RAHMANDAD: That is, those who have smoked more 02:26:36.640 --> 02:26:42.580 align:middle line:84% are more likely to get a screen, more likely to get treatment. 02:26:42.580 --> 02:26:47.400 align:middle line:84% And getting the co-flows right, that is, 02:26:47.400 --> 02:26:50.180 align:middle line:84% distributing the protection in the correct way, 02:26:50.180 --> 02:26:52.100 align:middle line:90% becomes extremely hard. 02:26:52.100 --> 02:26:54.640 align:middle line:84% There are ways for simpler cases to do it, 02:26:54.640 --> 02:26:58.440 align:middle line:84% and I have a few papers that actually introduce those, 02:26:58.440 --> 02:27:01.920 align:middle line:84% but I don't recommend it for this case. 02:27:01.920 --> 02:27:03.423 align:middle line:84% It is too complex to do it that way. 02:27:03.423 --> 02:27:05.340 align:middle line:84% AUDIENCE: I've actually practiced a little bit 02:27:05.340 --> 02:27:08.007 align:middle line:84% with an exponential distribution assumption and a truncated-- 02:27:08.007 --> 02:27:08.840 align:middle line:90% [INTERPOSING VOICES] 02:27:08.840 --> 02:27:10.798 align:middle line:84% HAZHIR RAHMANDAD: Yes, you can try to do those, 02:27:10.798 --> 02:27:13.580 align:middle line:90% but don't do it. 02:27:13.580 --> 02:27:16.660 align:middle line:84% I've done enough of this mess to know that it is 02:27:16.660 --> 02:27:20.840 align:middle line:90% too complex to try this pass. 02:27:20.840 --> 02:27:23.120 align:middle line:84% So it will take you in the wrong direction. 02:27:23.120 --> 02:27:26.680 align:middle line:84% Go with the simple alternative that 02:27:26.680 --> 02:27:30.420 align:middle line:84% says people have different levels of smoking 02:27:30.420 --> 02:27:31.980 align:middle line:90% history, whatever is important. 02:27:31.980 --> 02:27:36.360 align:middle line:84% If it is the cumulative history, you can have people who have had 02:27:36.360 --> 02:27:42.580 align:middle line:84% under 1,000 packs in their life, 1,000 to 5,000, 5,000 to 10,000, 02:27:42.580 --> 02:27:45.400 align:middle line:84% whatever is the categories that make sense. 02:27:45.400 --> 02:27:47.300 align:middle line:84% AUDIENCE: But my question still persist, 02:27:47.300 --> 02:27:50.680 align:middle line:84% which is if I have this low-smoking population, some 02:27:50.680 --> 02:27:53.240 align:middle line:84% of them through time, become high smoking. 02:27:53.240 --> 02:27:54.320 align:middle line:90% HAZHIR RAHMANDAD: Yes. 02:27:54.320 --> 02:27:59.440 align:middle line:84% AUDIENCE: How do I pick how many of them are moving on to that? 02:27:59.440 --> 02:28:02.080 align:middle line:84% HAZHIR RAHMANDAD: That's based on how many packs per day 02:28:02.080 --> 02:28:04.160 align:middle line:90% they are smoking because. 02:28:04.160 --> 02:28:05.180 align:middle line:90% It's not a co-flow. 02:28:05.180 --> 02:28:07.520 align:middle line:84% Now you have a number of these people 02:28:07.520 --> 02:28:10.660 align:middle line:84% that every year go to the next category. 02:28:10.660 --> 02:28:14.360 align:middle line:84% You need to make some assumption about the distribution of people 02:28:14.360 --> 02:28:16.390 align:middle line:90% in these categories. 02:28:16.390 --> 02:28:17.140 align:middle line:90% But that's doable. 02:28:17.140 --> 02:28:18.807 align:middle line:84% AUDIENCE: Just to ask one more question. 02:28:18.807 --> 02:28:21.780 align:middle line:84% So the second line of co-flows I have down there, which is we 02:28:21.780 --> 02:28:25.260 align:middle line:84% know that the rate of developing nodules 02:28:25.260 --> 02:28:27.783 align:middle line:84% depends on that accumulated smoking history. 02:28:27.783 --> 02:28:28.700 align:middle line:90% HAZHIR RAHMANDAD: Yes. 02:28:28.700 --> 02:28:30.617 align:middle line:84% AUDIENCE: So if I have two stocks, one of them 02:28:30.617 --> 02:28:33.900 align:middle line:84% are low-smoking intensity, the other one is high smoking. 02:28:33.900 --> 02:28:38.540 align:middle line:84% So each one of them will have different rates of nodules 02:28:38.540 --> 02:28:40.360 align:middle line:90% being built in there. 02:28:40.360 --> 02:28:42.220 align:middle line:84% HAZHIR RAHMANDAD: Why do you need to track 02:28:42.220 --> 02:28:43.780 align:middle line:90% the nodules separately? 02:28:43.780 --> 02:28:47.940 align:middle line:84% If it is tied to your cumulative smoking 02:28:47.940 --> 02:28:50.282 align:middle line:84% and you are capturing your cumulative smoking-- 02:28:50.282 --> 02:28:51.740 align:middle line:84% AUDIENCE: So the nodule size is not 02:28:51.740 --> 02:28:55.540 align:middle line:84% dependent on the smoking, the nodule growth rate is depending 02:28:55.540 --> 02:29:00.980 align:middle line:84% on smoking, which then the chance of someone showing 02:29:00.980 --> 02:29:02.860 align:middle line:84% symptoms depends on the nodule size, 02:29:02.860 --> 02:29:06.540 align:middle line:84% and the chance of them being tested positive 02:29:06.540 --> 02:29:09.120 align:middle line:84% when they go through screening, it depends upon the size. 02:29:09.120 --> 02:29:10.700 align:middle line:90% That's what I was thinking. 02:29:10.700 --> 02:29:11.617 align:middle line:90% HAZHIR RAHMANDAD: Yes. 02:29:11.617 --> 02:29:14.340 align:middle line:84% AUDIENCE: Cancer doesn't grow on itself like [INAUDIBLE]? 02:29:14.340 --> 02:29:17.460 align:middle line:84% AUDIENCE: Lung cancer is, the biggest driver is smoking. 02:29:17.460 --> 02:29:19.762 align:middle line:84% So there's nothing else but smoking, essentially. 02:29:19.762 --> 02:29:20.720 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 02:29:20.720 --> 02:29:28.600 align:middle line:84% So there might be an argument for doing the co-flow for that. 02:29:28.600 --> 02:29:30.220 align:middle line:84% AUDIENCE: [INAUDIBLE] lung health? 02:29:30.220 --> 02:29:37.120 align:middle line:84% HAZHIR RAHMANDAD: Yeah, basically tracking these nodules 02:29:37.120 --> 02:29:39.400 align:middle line:90% that are evolving. 02:29:39.400 --> 02:29:42.200 align:middle line:84% And it should be completely parallel to the population 02:29:42.200 --> 02:29:43.000 align:middle line:90% groups. 02:29:43.000 --> 02:29:48.520 align:middle line:84% For each population group you're tracking their nodules. 02:29:48.520 --> 02:29:52.120 align:middle line:84% But you need to be careful because when 02:29:52.120 --> 02:29:58.280 align:middle line:84% you get to route people into different states, 02:29:58.280 --> 02:30:03.200 align:middle line:84% you can't easily make it dependent on the nodules. 02:30:03.200 --> 02:30:09.680 align:middle line:84% You need to make it dependent on the stocks that they have. 02:30:09.680 --> 02:30:13.100 align:middle line:84% That is, the nodules should flow with the average. 02:30:13.100 --> 02:30:15.620 align:middle line:84% Now, if you don't like that, that is, you say, 02:30:15.620 --> 02:30:18.980 align:middle line:84% that's not how things happen in the practice, 02:30:18.980 --> 02:30:23.420 align:middle line:84% then you may need to reformulate the population groups based 02:30:23.420 --> 02:30:26.260 align:middle line:84% on nodules, not based on the history of smoking 02:30:26.260 --> 02:30:29.700 align:middle line:84% or how many packs they have had or something like that. 02:30:29.700 --> 02:30:33.460 align:middle line:84% OK, so that would say, I would have populations 02:30:33.460 --> 02:30:35.620 align:middle line:84% with this number of nodules and populations 02:30:35.620 --> 02:30:39.620 align:middle line:84% with this number of nodules and then-- 02:30:39.620 --> 02:30:41.140 align:middle line:84% AUDIENCE: So stratifying them first. 02:30:41.140 --> 02:30:42.557 align:middle line:84% HAZHIR RAHMANDAD: Stratifying them 02:30:42.557 --> 02:30:44.720 align:middle line:84% based on nodules in the first place. 02:30:44.720 --> 02:30:48.400 align:middle line:84% Or if you need both history and nodules, 02:30:48.400 --> 02:30:52.100 align:middle line:84% a two-dimensional stratification, a matrix, 02:30:52.100 --> 02:30:56.460 align:middle line:84% that so that the movement on the nodule space 02:30:56.460 --> 02:31:01.140 align:middle line:84% is tracked based on the physics of your stocks and flows. 02:31:01.140 --> 02:31:04.340 align:middle line:90% That's my recommendation. 02:31:04.340 --> 02:31:10.340 align:middle line:84% Don't go with the co-flow if the rates of outflows 02:31:10.340 --> 02:31:11.760 align:middle line:90% will depend on it. 02:31:11.760 --> 02:31:12.400 align:middle line:90% Yeah. 02:31:12.400 --> 02:31:15.360 align:middle line:84% AUDIENCE: So now that I think about it, 02:31:15.360 --> 02:31:18.720 align:middle line:84% why can't we go with the individual level 02:31:18.720 --> 02:31:19.920 align:middle line:90% testing [INAUDIBLE]? 02:31:19.920 --> 02:31:22.600 align:middle line:84% Because we can keep track of the stock 02:31:22.600 --> 02:31:24.558 align:middle line:84% of the nodule, the stock of the cancer history. 02:31:24.558 --> 02:31:26.017 align:middle line:84% HAZHIR RAHMANDAD: Yeah, absolutely. 02:31:26.017 --> 02:31:26.980 align:middle line:90% You can do that. 02:31:26.980 --> 02:31:29.880 align:middle line:84% So that would be an agent-based alternative. 02:31:29.880 --> 02:31:33.320 align:middle line:90% You can turn that agent base-- 02:31:33.320 --> 02:31:34.740 align:middle line:90% you can do it in Vensim. 02:31:34.740 --> 02:31:38.220 align:middle line:84% But it's not the optimal place to do agent-based models. 02:31:38.220 --> 02:31:38.960 align:middle line:90% But yeah-- 02:31:38.960 --> 02:31:39.400 align:middle line:90% [INTERPOSING VOICES] 02:31:39.400 --> 02:31:41.733 align:middle line:84% HAZHIR RAHMANDAD: --you would have a subscript of people 02:31:41.733 --> 02:31:43.160 align:middle line:90% basically. 02:31:43.160 --> 02:31:46.640 align:middle line:84% Now you could have representative agent type 02:31:46.640 --> 02:31:50.460 align:middle line:90% of a model that you would say-- 02:31:50.460 --> 02:31:55.760 align:middle line:90% 02:31:55.760 --> 02:31:58.660 align:middle line:84% it's actually a little bit tricky what you want to do it, 02:31:58.660 --> 02:32:02.480 align:middle line:84% because ultimately you care about the distribution of people 02:32:02.480 --> 02:32:04.560 align:middle line:84% and you can simulate as many people 02:32:04.560 --> 02:32:08.680 align:middle line:84% as you want that go through this and probably stochastically 02:32:08.680 --> 02:32:10.980 align:middle line:90% they move into different-- 02:32:10.980 --> 02:32:14.160 align:middle line:84% so that's a full-blown agent-based model. 02:32:14.160 --> 02:32:19.740 align:middle line:90% 02:32:19.740 --> 02:32:26.820 align:middle line:84% Some hybrid that tries to do that, but with less complexity, 02:32:26.820 --> 02:32:27.660 align:middle line:90% I don't know. 02:32:27.660 --> 02:32:29.980 align:middle line:84% That would be beyond what we can-- 02:32:29.980 --> 02:32:31.560 align:middle line:84% AUDIENCE: [INAUDIBLE] do right now. 02:32:31.560 --> 02:32:32.960 align:middle line:90% Then try to build upon. 02:32:32.960 --> 02:32:38.900 align:middle line:84% HAZHIR RAHMANDAD: Yeah, I would say really it is either stratify 02:32:38.900 --> 02:32:42.540 align:middle line:84% the population based on the dimensions that you need, 02:32:42.540 --> 02:32:46.700 align:middle line:84% which is in case maybe nodules and history of smoking. 02:32:46.700 --> 02:32:49.660 align:middle line:84% Then you would have a dimension for nodules, a dimension 02:32:49.660 --> 02:32:51.320 align:middle line:90% for history of smoking. 02:32:51.320 --> 02:32:54.100 align:middle line:84% And then within each of those two-- that is, 02:32:54.100 --> 02:32:57.120 align:middle line:84% once those two are set, what are the dimensions, 02:32:57.120 --> 02:33:00.560 align:middle line:84% the things that are once you set those dimensions, 02:33:00.560 --> 02:33:04.820 align:middle line:84% then the probability of movement is not going to-- 02:33:04.820 --> 02:33:06.840 align:middle line:84% I mean, only depends on those two things. 02:33:06.840 --> 02:33:08.367 align:middle line:90% That's the key. 02:33:08.367 --> 02:33:09.200 align:middle line:90% AUDIENCE: Thank you. 02:33:09.200 --> 02:33:10.620 align:middle line:90% HAZHIR RAHMANDAD: Sure. 02:33:10.620 --> 02:33:13.500 align:middle line:84% So I think we're almost out of time, 02:33:13.500 --> 02:33:18.160 align:middle line:84% but let me just do a quick look at Ali's model as well. 02:33:18.160 --> 02:33:20.700 align:middle line:84% So your model started more advanced, Ali. 02:33:20.700 --> 02:33:22.900 align:middle line:84% So what were you working on this week? 02:33:22.900 --> 02:33:25.640 align:middle line:84% AUDIENCE: I got a start with the punishment tier, punishment 02:33:25.640 --> 02:33:26.220 align:middle line:90% mechanism. 02:33:26.220 --> 02:33:28.083 align:middle line:84% So I duplicated the stock management. 02:33:28.083 --> 02:33:29.000 align:middle line:90% HAZHIR RAHMANDAD: Yes. 02:33:29.000 --> 02:33:30.542 align:middle line:84% AUDIENCE: And it's not sophisticated. 02:33:30.542 --> 02:33:33.500 align:middle line:84% So it's just, I get two digits at the top. 02:33:33.500 --> 02:33:36.180 align:middle line:84% This is the first agent and this, at the bottom, 02:33:36.180 --> 02:33:37.360 align:middle line:90% it's the second. 02:33:37.360 --> 02:33:39.040 align:middle line:84% But they're using the same resource. 02:33:39.040 --> 02:33:42.162 align:middle line:84% If you zoom out, at the very top I have the resources. 02:33:42.162 --> 02:33:43.120 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 02:33:43.120 --> 02:33:44.720 align:middle line:90% AUDIENCE: And they are-- 02:33:44.720 --> 02:33:48.020 align:middle line:84% HAZHIR RAHMANDAD: So all of that I saw yesterday. 02:33:48.020 --> 02:33:49.120 align:middle line:90% AUDIENCE: [INAUDIBLE] 02:33:49.120 --> 02:33:50.078 align:middle line:90% HAZHIR RAHMANDAD: Sure. 02:33:50.078 --> 02:33:52.907 align:middle line:84% So the question was, this week you were focusing on the-- 02:33:52.907 --> 02:33:55.240 align:middle line:84% AUDIENCE: Yeah, I wanted to capture the peer punishment. 02:33:55.240 --> 02:33:55.620 align:middle line:90% HAZHIR RAHMANDAD: Yes. 02:33:55.620 --> 02:33:57.245 align:middle line:84% AUDIENCE: And it's a little bit tricky. 02:33:57.245 --> 02:33:59.920 align:middle line:84% So in the literature it's like people saying that, 02:33:59.920 --> 02:34:05.510 align:middle line:84% OK, let's punish the deviators from the agreed-upon levels 02:34:05.510 --> 02:34:08.790 align:middle line:90% of extraction of that resource. 02:34:08.790 --> 02:34:10.570 align:middle line:90% So I have done that so far. 02:34:10.570 --> 02:34:12.130 align:middle line:90% And it's continuous. 02:34:12.130 --> 02:34:14.130 align:middle line:90% There's no lag in between. 02:34:14.130 --> 02:34:16.530 align:middle line:90% And it seems to work. 02:34:16.530 --> 02:34:19.110 align:middle line:90% But not [INAUDIBLE]. 02:34:19.110 --> 02:34:21.630 align:middle line:84% HAZHIR RAHMANDAD: So I think just big picture, 02:34:21.630 --> 02:34:25.830 align:middle line:84% it is a tough problem because the punishment question 02:34:25.830 --> 02:34:29.630 align:middle line:84% is very much a game theory type of a situation 02:34:29.630 --> 02:34:31.190 align:middle line:84% that you need to account for what 02:34:31.190 --> 02:34:33.590 align:middle line:84% the response of the other side might be. 02:34:33.590 --> 02:34:37.430 align:middle line:84% And in the typical SD models, we don't have the language 02:34:37.430 --> 02:34:39.470 align:middle line:90% for capturing those very well. 02:34:39.470 --> 02:34:43.250 align:middle line:84% Now you can think of these as strategies. 02:34:43.250 --> 02:34:46.390 align:middle line:84% That is, I have a couple of parameters 02:34:46.390 --> 02:34:51.510 align:middle line:84% that allow me to connect my current state of the world, 02:34:51.510 --> 02:34:56.750 align:middle line:84% how I see the other side and how I see myself, to whether I 02:34:56.750 --> 02:34:58.190 align:middle line:90% punish them or not. 02:34:58.190 --> 02:35:01.570 align:middle line:84% So if you define those two parameters, 02:35:01.570 --> 02:35:04.090 align:middle line:84% let's say, each actor can be defined 02:35:04.090 --> 02:35:09.530 align:middle line:84% with a strategy that consists of those two parameters. 02:35:09.530 --> 02:35:12.750 align:middle line:90% Now I have now two actors. 02:35:12.750 --> 02:35:16.250 align:middle line:84% Each of them, they have these two-dimensional strategy spaces 02:35:16.250 --> 02:35:18.150 align:middle line:90% that are those two parameters. 02:35:18.150 --> 02:35:19.890 align:middle line:84% They are playing with each other or they 02:35:19.890 --> 02:35:21.410 align:middle line:90% are going against each other. 02:35:21.410 --> 02:35:24.870 align:middle line:84% Now, given the other side's strategy, 02:35:24.870 --> 02:35:27.270 align:middle line:84% I can optimize my two parameters. 02:35:27.270 --> 02:35:32.210 align:middle line:84% So I can run an optimization that changes my strategy 02:35:32.210 --> 02:35:35.510 align:middle line:84% to be the best response given the strategy space 02:35:35.510 --> 02:35:36.330 align:middle line:90% I have defined. 02:35:36.330 --> 02:35:39.490 align:middle line:84% It's a linear functional form or something simple-- 02:35:39.490 --> 02:35:41.370 align:middle line:90% to what the other side does. 02:35:41.370 --> 02:35:45.650 align:middle line:84% Then the other side can optimize their response to me 02:35:45.650 --> 02:35:48.250 align:middle line:90% given what I'm doing. 02:35:48.250 --> 02:35:50.690 align:middle line:84% And if you do this a couple of times, 02:35:50.690 --> 02:35:53.050 align:middle line:84% you can converge-- maybe a few times-- 02:35:53.050 --> 02:35:55.850 align:middle line:84% you can converge to a Nash equilibrium 02:35:55.850 --> 02:36:00.090 align:middle line:84% type of a strategy that says, OK, this is stable. 02:36:00.090 --> 02:36:04.590 align:middle line:84% That is, both sides will converge to this strategy 02:36:04.590 --> 02:36:07.530 align:middle line:84% if their structure of decision-making 02:36:07.530 --> 02:36:11.430 align:middle line:84% followed that functional form you have put into your model. 02:36:11.430 --> 02:36:16.630 align:middle line:84% So that becomes one solution concept that you can use here. 02:36:16.630 --> 02:36:19.450 align:middle line:84% Otherwise, just ad hoc formulation, 02:36:19.450 --> 02:36:21.890 align:middle line:84% what you have already is doing something like that. 02:36:21.890 --> 02:36:23.630 align:middle line:84% That is, you have some parameters 02:36:23.630 --> 02:36:28.830 align:middle line:84% that generate a response in terms of punishment 02:36:28.830 --> 02:36:30.610 align:middle line:84% based on the state of the system. 02:36:30.610 --> 02:36:32.870 align:middle line:84% So you already have the first component. 02:36:32.870 --> 02:36:35.850 align:middle line:84% Now you could optimize the way I told you, 02:36:35.850 --> 02:36:37.390 align:middle line:90% but the optimization is tricky. 02:36:37.390 --> 02:36:39.770 align:middle line:84% You have to go through it iteratively. 02:36:39.770 --> 02:36:44.390 align:middle line:84% The first one does optimize on their two, three parameters. 02:36:44.390 --> 02:36:46.003 align:middle line:90% The second one is fixed. 02:36:46.003 --> 02:36:46.670 align:middle line:90% AUDIENCE: Right. 02:36:46.670 --> 02:36:49.710 align:middle line:84% HAZHIR RAHMANDAD: Then the second one looks at the-- 02:36:49.710 --> 02:36:54.150 align:middle line:84% so I have a 2011 paper in SDR that 02:36:54.150 --> 02:36:56.770 align:middle line:84% has the machinery for doing that inside Vensim. 02:36:56.770 --> 02:36:57.970 align:middle line:90% You can check it out. 02:36:57.970 --> 02:37:02.410 align:middle line:84% You can write some scripts that do that in an automated way. 02:37:02.410 --> 02:37:05.697 align:middle line:84% That's the solution type that you can use. 02:37:05.697 --> 02:37:06.530 align:middle line:90% AUDIENCE: Thank you. 02:37:06.530 --> 02:37:07.290 align:middle line:90% Thanks a lot. 02:37:07.290 --> 02:37:09.850 align:middle line:84% Yeah, this is all happening simultaneously. 02:37:09.850 --> 02:37:11.517 align:middle line:84% HAZHIR RAHMANDAD: Yes, but right now you 02:37:11.517 --> 02:37:14.667 align:middle line:84% are making the assumption that both use the same strategy. 02:37:14.667 --> 02:37:15.250 align:middle line:90% AUDIENCE: Yes. 02:37:15.250 --> 02:37:16.708 align:middle line:84% HAZHIR RAHMANDAD: And then you just 02:37:16.708 --> 02:37:18.470 align:middle line:84% look at the solution that comes out of it, 02:37:18.470 --> 02:37:21.670 align:middle line:84% and you're not trying to be internally consistent. 02:37:21.670 --> 02:37:25.450 align:middle line:84% An internally consistent strategy is one that if I use, 02:37:25.450 --> 02:37:29.110 align:middle line:84% I can't improve upon it by deviating from that strategy. 02:37:29.110 --> 02:37:31.170 align:middle line:90% I can't improve upon my-- 02:37:31.170 --> 02:37:35.210 align:middle line:84% and as such, what you do here might be completely stupid 02:37:35.210 --> 02:37:37.170 align:middle line:90% in terms of, yeah, there are-- 02:37:37.170 --> 02:37:41.370 align:middle line:84% I can easily improve upon this by less or more punishment, 02:37:41.370 --> 02:37:43.110 align:middle line:90% and you're not capturing that. 02:37:43.110 --> 02:37:47.170 align:middle line:84% So this is to push it into the next level. 02:37:47.170 --> 02:37:48.890 align:middle line:90% OK, great. 02:37:48.890 --> 02:37:50.270 align:middle line:90% I think we are out of time. 02:37:50.270 --> 02:37:53.230 align:middle line:84% And there will be a seminar happening here in five minutes, 02:37:53.230 --> 02:37:53.870 align:middle line:90% so thank you. 02:37:53.870 --> 02:37:56.320 align:middle line:90% [LAUGHS] 02:37:56.320 --> 02:37:59.000 align:middle line:90%