WEBVTT 00:00:00.000 --> 00:00:02.432 align:middle line:90% [SQUEAKING] 00:00:02.432 --> 00:00:04.364 align:middle line:90% [RUSTLING] 00:00:04.364 --> 00:00:07.262 align:middle line:90% [CLICKING] 00:00:07.262 --> 00:00:13.080 align:middle line:90% 00:00:13.080 --> 00:00:14.380 align:middle line:90% HAZHIR RAHMANDAD: Welcome. 00:00:14.380 --> 00:00:19.480 align:middle line:84% This is 15.879, System Dynamics Research Seminar. 00:00:19.480 --> 00:00:22.240 align:middle line:84% And this is a seminar series that each semester we 00:00:22.240 --> 00:00:23.820 align:middle line:90% have a variant of it. 00:00:23.820 --> 00:00:26.800 align:middle line:84% So we have about four different versions of this course 00:00:26.800 --> 00:00:31.000 align:middle line:84% that we iterate through over a two-year period. 00:00:31.000 --> 00:00:38.080 align:middle line:84% So this version is one focused on building classical modeling 00:00:38.080 --> 00:00:44.800 align:middle line:84% skills, covering the gap between what you get in 15.873 00:00:44.800 --> 00:00:48.040 align:middle line:84% or any variant of a full semester course on system 00:00:48.040 --> 00:00:51.640 align:middle line:84% dynamics and what you need to be able to do 00:00:51.640 --> 00:00:55.020 align:middle line:84% serious modeling in an actual consulting project, 00:00:55.020 --> 00:00:56.340 align:middle line:90% in research projects. 00:00:56.340 --> 00:00:59.440 align:middle line:84% This is a gap that typically students 00:00:59.440 --> 00:01:04.760 align:middle line:84% don't have much opportunities, in terms of resources, 00:01:04.760 --> 00:01:07.900 align:middle line:90% to do self-study on. 00:01:07.900 --> 00:01:10.440 align:middle line:84% We also don't have that many courses around-- 00:01:10.440 --> 00:01:14.260 align:middle line:84% across the world to cover the skill sets here. 00:01:14.260 --> 00:01:21.000 align:middle line:84% So I hope we have a fun seminar series throughout the semester 00:01:21.000 --> 00:01:21.500 align:middle line:90% together. 00:01:21.500 --> 00:01:24.040 align:middle line:84% But also, I hope that we can create some material that 00:01:24.040 --> 00:01:27.120 align:middle line:84% is useful, more broadly, for anybody 00:01:27.120 --> 00:01:31.640 align:middle line:84% who wants to learn system dynamics beyond the basics. 00:01:31.640 --> 00:01:34.380 align:middle line:84% So that's why we are actually recording these sessions. 00:01:34.380 --> 00:01:37.520 align:middle line:84% And we will share that through the OpenCourseWare Project 00:01:37.520 --> 00:01:38.220 align:middle line:90% of MIT. 00:01:38.220 --> 00:01:40.240 align:middle line:90% So that's the plan. 00:01:40.240 --> 00:01:42.800 align:middle line:84% But I guess, to get us started, it's a good idea 00:01:42.800 --> 00:01:46.520 align:middle line:84% to go around the room, get ourselves introduced, 00:01:46.520 --> 00:01:49.340 align:middle line:84% so that everybody knows everybody else. 00:01:49.340 --> 00:01:53.200 align:middle line:84% This is, for a research seminar, not very small, but still 00:01:53.200 --> 00:01:56.200 align:middle line:90% a fairly small class. 00:01:56.200 --> 00:01:58.050 align:middle line:90% Welcome, join us. 00:01:58.050 --> 00:02:00.510 align:middle line:90% So yeah, we can get us started. 00:02:00.510 --> 00:02:02.690 align:middle line:90% So I'm Hazhir Rahmandad. 00:02:02.690 --> 00:02:05.270 align:middle line:84% I'm on the faculty in the system dynamics group here. 00:02:05.270 --> 00:02:09.283 align:middle line:84% I've been around at MIT as a faculty member since 2015. 00:02:09.283 --> 00:02:11.450 align:middle line:84% I was at Virginia Tech in the industrial engineering 00:02:11.450 --> 00:02:13.490 align:middle line:84% department for a few years before that. 00:02:13.490 --> 00:02:16.650 align:middle line:90% And I did my PhD here at MIT. 00:02:16.650 --> 00:02:21.570 align:middle line:84% I do different types of research, some of it 00:02:21.570 --> 00:02:27.330 align:middle line:84% more on organizational dynamics and strategy, some of it 00:02:27.330 --> 00:02:31.450 align:middle line:84% on public policy, public health, and some also related 00:02:31.450 --> 00:02:32.970 align:middle line:90% to methods. 00:02:32.970 --> 00:02:34.590 align:middle line:90% So that's me. 00:02:34.590 --> 00:02:38.990 align:middle line:84% Maybe go around the room and also catch up with folks online. 00:02:38.990 --> 00:02:39.490 align:middle line:90% OK. 00:02:39.490 --> 00:02:42.250 align:middle line:84% As you can see, we have a very diverse group 00:02:42.250 --> 00:02:43.550 align:middle line:90% from everywhere in the world. 00:02:43.550 --> 00:02:45.690 align:middle line:90% So this is going to be fun. 00:02:45.690 --> 00:02:47.630 align:middle line:90% So today is our first session. 00:02:47.630 --> 00:02:50.190 align:middle line:84% I think, most of you, we have been in touch. 00:02:50.190 --> 00:02:55.350 align:middle line:84% I tried to share material on the class and both the syllabus, 00:02:55.350 --> 00:02:58.170 align:middle line:84% but also a little bit of a reading and some topics 00:02:58.170 --> 00:02:59.470 align:middle line:90% that we will discuss today. 00:02:59.470 --> 00:03:01.250 align:middle line:84% So today, I think maybe half of the class 00:03:01.250 --> 00:03:05.210 align:middle line:84% is more on the theory administrative stuff and half 00:03:05.210 --> 00:03:10.310 align:middle line:84% of it on discussing the problems that we have for this week. 00:03:10.310 --> 00:03:12.910 align:middle line:84% And then the coming weeks, it will be mostly hands-on. 00:03:12.910 --> 00:03:16.930 align:middle line:84% So basically, this is a very hands on class. 00:03:16.930 --> 00:03:23.313 align:middle line:84% So OK, on the administrative side, the class is three hours. 00:03:23.313 --> 00:03:24.730 align:middle line:84% We will take a break in the middle 00:03:24.730 --> 00:03:28.370 align:middle line:84% of the session in this room and also on the Zoom. 00:03:28.370 --> 00:03:31.530 align:middle line:84% You have the Zoom links on the syllabus. 00:03:31.530 --> 00:03:33.270 align:middle line:90% And this is a small class. 00:03:33.270 --> 00:03:35.590 align:middle line:90% So we will largely use emails. 00:03:35.590 --> 00:03:37.810 align:middle line:84% And so I've created an email list. 00:03:37.810 --> 00:03:40.050 align:middle line:84% And anybody who is not on the email list, 00:03:40.050 --> 00:03:41.310 align:middle line:90% please just send me an email. 00:03:41.310 --> 00:03:43.650 align:middle line:90% I'll add you to that. 00:03:43.650 --> 00:03:46.630 align:middle line:84% We also have a shared Dropbox for sharing the material. 00:03:46.630 --> 00:03:50.370 align:middle line:84% Again, this simplifies not worrying about Canvas access 00:03:50.370 --> 00:03:53.230 align:middle line:90% for people who are not at MIT. 00:03:53.230 --> 00:03:57.830 align:middle line:84% So if you don't have access to Canvas-- to Dropbox, again, 00:03:57.830 --> 00:03:58.670 align:middle line:90% just drop me a note. 00:03:58.670 --> 00:04:00.650 align:middle line:90% I'll add you. 00:04:00.650 --> 00:04:03.130 align:middle line:84% Grades, I don't think anybody here 00:04:03.130 --> 00:04:05.890 align:middle line:84% cares about grades for this specific course. 00:04:05.890 --> 00:04:09.230 align:middle line:84% In general, this is a learning opportunity. 00:04:09.230 --> 00:04:10.870 align:middle line:84% If you care about grades, let me know. 00:04:10.870 --> 00:04:14.270 align:middle line:90% So I can tell you more. 00:04:14.270 --> 00:04:17.370 align:middle line:84% But to the extent that there are grades for some of you 00:04:17.370 --> 00:04:19.230 align:middle line:84% who get grades out of this course, 00:04:19.230 --> 00:04:22.290 align:middle line:84% it's based on your weekly modeling work, assignments, 00:04:22.290 --> 00:04:25.810 align:middle line:84% and also active participation in the class. 00:04:25.810 --> 00:04:29.490 align:middle line:84% So as I mentioned, this class is being recorded. 00:04:29.490 --> 00:04:34.090 align:middle line:84% We actually have a full camera set here. 00:04:34.090 --> 00:04:36.430 align:middle line:90% And Tom is taking care of that. 00:04:36.430 --> 00:04:38.250 align:middle line:90% Thank you, Tom. 00:04:38.250 --> 00:04:42.770 align:middle line:84% So that's going to be later edited, 00:04:42.770 --> 00:04:46.090 align:middle line:84% put into a package that goes into MIT OpenCourseWare, 00:04:46.090 --> 00:04:49.210 align:middle line:84% along with the rest of the material from this class. 00:04:49.210 --> 00:04:52.230 align:middle line:84% So actually, we do have a media release form. 00:04:52.230 --> 00:04:54.890 align:middle line:84% I think many of you have already signed that. 00:04:54.890 --> 00:04:57.990 align:middle line:84% If you haven't, I have hard copies here. 00:04:57.990 --> 00:05:01.450 align:middle line:90% So feel free to pass it along. 00:05:01.450 --> 00:05:05.170 align:middle line:84% If anybody hasn't signed, you can share that 00:05:05.170 --> 00:05:09.518 align:middle line:90% with me, the signed version. 00:05:09.518 --> 00:05:11.310 align:middle line:84% And if you have any questions, let me know. 00:05:11.310 --> 00:05:14.650 align:middle line:90% We can talk about it. 00:05:14.650 --> 00:05:19.910 align:middle line:90% Then we have-- let's see. 00:05:19.910 --> 00:05:25.350 align:middle line:84% So the structure of the class is going to be pretty standard. 00:05:25.350 --> 00:05:31.050 align:middle line:84% So basically, before the class, you 00:05:31.050 --> 00:05:33.090 align:middle line:84% will have a modeling task each week 00:05:33.090 --> 00:05:38.050 align:middle line:84% to work on that I would provide it to you a couple of weeks 00:05:38.050 --> 00:05:39.890 align:middle line:90% ahead of time. 00:05:39.890 --> 00:05:42.890 align:middle line:84% For the next week modeling task, I 00:05:42.890 --> 00:05:46.550 align:middle line:84% think I provided it a little bit in the middle of this week. 00:05:46.550 --> 00:05:52.700 align:middle line:84% So the idea is that you work on modeling the thing that 00:05:52.700 --> 00:05:55.260 align:middle line:84% is upcoming in the coming week, but also 00:05:55.260 --> 00:05:58.440 align:middle line:84% think about the week after problem 00:05:58.440 --> 00:06:01.140 align:middle line:84% so that we have a conversation about that problem 00:06:01.140 --> 00:06:03.560 align:middle line:84% in the class in preparation for the next week. 00:06:03.560 --> 00:06:05.780 align:middle line:84% So today, we will talk about the problem 00:06:05.780 --> 00:06:07.740 align:middle line:84% that you will be working on in the coming week, 00:06:07.740 --> 00:06:10.820 align:middle line:84% in terms of the detailed modeling. 00:06:10.820 --> 00:06:14.740 align:middle line:84% And we will be working on some actual-- 00:06:14.740 --> 00:06:16.860 align:middle line:84% not modeling tasks today, today is mostly 00:06:16.860 --> 00:06:20.860 align:middle line:84% on architectural decisions in modeling projects. 00:06:20.860 --> 00:06:24.600 align:middle line:84% So the class will be basically broken down into two parts. 00:06:24.600 --> 00:06:26.720 align:middle line:84% The first part, the larger chunk, 00:06:26.720 --> 00:06:28.440 align:middle line:90% is focused on the modeling. 00:06:28.440 --> 00:06:30.460 align:middle line:84% And the way we do it is you folks 00:06:30.460 --> 00:06:32.380 align:middle line:84% will be submitting whatever you have 00:06:32.380 --> 00:06:34.720 align:middle line:90% created, in terms of models. 00:06:34.720 --> 00:06:38.560 align:middle line:84% And there is details of how to submit that in the syllabus 00:06:38.560 --> 00:06:41.340 align:middle line:84% and also inside each of those assignments. 00:06:41.340 --> 00:06:42.640 align:middle line:90% You put it on the Dropbox. 00:06:42.640 --> 00:06:44.840 align:middle line:90% We randomly pick a few of them. 00:06:44.840 --> 00:06:47.980 align:middle line:84% We go through the models together, we critique them, 00:06:47.980 --> 00:06:53.200 align:middle line:84% we think about places where things could be improved. 00:06:53.200 --> 00:06:56.660 align:middle line:84% And that process will hopefully help everybody 00:06:56.660 --> 00:07:00.740 align:middle line:90% learn some of those skills. 00:07:00.740 --> 00:07:03.900 align:middle line:84% So we will do that with as many of those models 00:07:03.900 --> 00:07:07.380 align:middle line:90% that we can in our limited time. 00:07:07.380 --> 00:07:12.820 align:middle line:84% And then we spend this smaller part of the end of the class 00:07:12.820 --> 00:07:14.940 align:middle line:84% discussing the next week's problem 00:07:14.940 --> 00:07:16.980 align:middle line:84% and thinking through the high level 00:07:16.980 --> 00:07:20.520 align:middle line:84% architectural choices, the definition of the problem, 00:07:20.520 --> 00:07:24.640 align:middle line:84% and so on together so that we are on the same page 00:07:24.640 --> 00:07:28.980 align:middle line:84% and we have converged on some of the harder architectural choices 00:07:28.980 --> 00:07:33.940 align:middle line:84% before you are left alone to try to tackle these modeling 00:07:33.940 --> 00:07:34.900 align:middle line:90% challenges. 00:07:34.900 --> 00:07:39.020 align:middle line:84% Now, this structure, I hope, is very helpful for you 00:07:39.020 --> 00:07:42.420 align:middle line:84% to learn from each other's challenges 00:07:42.420 --> 00:07:48.460 align:middle line:84% and also gain the confidence to tackle a full modeling 00:07:48.460 --> 00:07:50.780 align:middle line:84% project on your own because this is what 00:07:50.780 --> 00:07:52.300 align:middle line:90% you're going to do every week. 00:07:52.300 --> 00:07:54.060 align:middle line:84% Now, it will be challenging, depending 00:07:54.060 --> 00:07:57.180 align:middle line:84% on-- there is some variance in your background in modeling 00:07:57.180 --> 00:07:58.100 align:middle line:90% and prior experience. 00:07:58.100 --> 00:07:59.560 align:middle line:84% So some of you may find it easier. 00:07:59.560 --> 00:08:02.180 align:middle line:84% Some of you may find it really challenging. 00:08:02.180 --> 00:08:03.160 align:middle line:90% Stay with it. 00:08:03.160 --> 00:08:04.800 align:middle line:84% Hopefully, by the end of the semester, 00:08:04.800 --> 00:08:07.060 align:middle line:84% everybody here will be in good shape, 00:08:07.060 --> 00:08:11.280 align:middle line:84% in terms of building solid, relatively simple, 00:08:11.280 --> 00:08:15.300 align:middle line:84% but models that can be used in practice for research 00:08:15.300 --> 00:08:19.660 align:middle line:90% or consulting work. 00:08:19.660 --> 00:08:22.140 align:middle line:84% I'm also designing this partly with an eye 00:08:22.140 --> 00:08:25.700 align:middle line:84% towards future learners who want to do self-study 00:08:25.700 --> 00:08:26.960 align:middle line:90% in system dynamics. 00:08:26.960 --> 00:08:29.860 align:middle line:84% And the way I'm envisioning the material 00:08:29.860 --> 00:08:32.179 align:middle line:84% to be useful for them is that they 00:08:32.179 --> 00:08:35.380 align:middle line:84% go through these weekly material, 00:08:35.380 --> 00:08:40.260 align:middle line:84% they start by doing the assignment for each week, 00:08:40.260 --> 00:08:42.140 align:middle line:84% perhaps after watching the discussion 00:08:42.140 --> 00:08:47.580 align:middle line:84% on the architectural choices in the week before, 00:08:47.580 --> 00:08:52.400 align:middle line:84% then they review the models that you have submitted, 00:08:52.400 --> 00:08:56.240 align:middle line:84% especially focusing on the ones that we discuss in the class 00:08:56.240 --> 00:09:00.000 align:middle line:84% so that they have their own critique of those models. 00:09:00.000 --> 00:09:02.440 align:middle line:84% Then they watch how we critiqued it in the class. 00:09:02.440 --> 00:09:04.940 align:middle line:84% So that way, you calibrate your own way 00:09:04.940 --> 00:09:09.680 align:middle line:84% of critiquing a model with how the class does that. 00:09:09.680 --> 00:09:12.960 align:middle line:84% So that would be one kind of a feedback opportunity, 00:09:12.960 --> 00:09:15.860 align:middle line:84% even though we don't have live feedback for hopefully 00:09:15.860 --> 00:09:19.460 align:middle line:84% many, many learners who may use this material later. 00:09:19.460 --> 00:09:23.780 align:middle line:84% Then they will come back to their own models, 00:09:23.780 --> 00:09:25.960 align:middle line:84% based on what they have learned from that critique, 00:09:25.960 --> 00:09:30.620 align:middle line:84% and try to improve and get their models to as good a place 00:09:30.620 --> 00:09:31.640 align:middle line:90% as they can get. 00:09:31.640 --> 00:09:35.880 align:middle line:84% So this is a protocol for people who will use this in future, 00:09:35.880 --> 00:09:37.200 align:middle line:90% not really relevant to you. 00:09:37.200 --> 00:09:41.340 align:middle line:84% But I'm highlighting it because, as we go through the class, 00:09:41.340 --> 00:09:46.910 align:middle line:84% you can see how it may benefit others in the future as well. 00:09:46.910 --> 00:09:50.190 align:middle line:84% So that's a little bit of administrative background. 00:09:50.190 --> 00:09:53.210 align:middle line:84% Let's see if there are any questions on that. 00:09:53.210 --> 00:10:00.855 align:middle line:84% And then we can move into more content for our work. 00:10:00.855 --> 00:10:02.730 align:middle line:84% AUDIENCE: I'm taking the class as a listener. 00:10:02.730 --> 00:10:05.670 align:middle line:84% Is there any differentiation between those taking it 00:10:05.670 --> 00:10:06.210 align:middle line:90% for credit? 00:10:06.210 --> 00:10:10.470 align:middle line:84% HAZHIR RAHMANDAD: Yeah, so basically, no, in the sense 00:10:10.470 --> 00:10:13.010 align:middle line:84% that I'm not going to tell you, oh, 00:10:13.010 --> 00:10:16.210 align:middle line:84% why didn't you submit something or please do this or that? 00:10:16.210 --> 00:10:17.470 align:middle line:90% You should decide. 00:10:17.470 --> 00:10:20.490 align:middle line:84% You guys are here with different goals. 00:10:20.490 --> 00:10:22.590 align:middle line:84% Some of you are trying to do everything that we 00:10:22.590 --> 00:10:24.330 align:middle line:90% are going through the class. 00:10:24.330 --> 00:10:26.290 align:middle line:84% Some of you are more in the listener phase. 00:10:26.290 --> 00:10:28.150 align:middle line:90% I'm OK with that. 00:10:28.150 --> 00:10:30.130 align:middle line:84% In terms of participation in the class, 00:10:30.130 --> 00:10:33.350 align:middle line:84% be actively participating to add value. 00:10:33.350 --> 00:10:37.030 align:middle line:84% If you're here, you want to contribute. 00:10:37.030 --> 00:10:41.070 align:middle line:84% But for those who are taking this course for credit, 00:10:41.070 --> 00:10:44.110 align:middle line:84% then try to do as many of the assignments as you can. 00:10:44.110 --> 00:10:47.050 align:middle line:84% I think, if you can't do some week, 00:10:47.050 --> 00:10:49.030 align:middle line:84% just send me a note beforehand so 00:10:49.030 --> 00:10:51.910 align:middle line:84% that I don't call on you that week, 00:10:51.910 --> 00:10:53.410 align:middle line:84% you don't feel embarrassed about it. 00:10:53.410 --> 00:10:56.850 align:middle line:84% But other than that, that's all good. 00:10:56.850 --> 00:10:58.710 align:middle line:84% And if you're not taking it for credit, 00:10:58.710 --> 00:11:01.370 align:middle line:84% but you are here to learn, then do the assignments. 00:11:01.370 --> 00:11:02.970 align:middle line:84% It's going to be helpful for you. 00:11:02.970 --> 00:11:05.053 align:middle line:84% And we have more material to discuss in the class. 00:11:05.053 --> 00:11:06.471 align:middle line:84% AUDIENCE: So there's no difference 00:11:06.471 --> 00:11:08.870 align:middle line:84% when you're selecting which models to critique? 00:11:08.870 --> 00:11:12.630 align:middle line:84% HAZHIR RAHMANDAD: I am thinking that I would give a little bit 00:11:12.630 --> 00:11:14.230 align:middle line:84% more priority-- so in the probability 00:11:14.230 --> 00:11:16.730 align:middle line:84% distribution of being randomly selected, 00:11:16.730 --> 00:11:19.030 align:middle line:84% I'll give a little bit higher priority 00:11:19.030 --> 00:11:20.750 align:middle line:84% to those who are taking this for credit 00:11:20.750 --> 00:11:23.090 align:middle line:84% so that they have a higher chance of being picked 00:11:23.090 --> 00:11:26.250 align:middle line:84% and, therefore, getting a little bit more feedback on their work. 00:11:26.250 --> 00:11:28.110 align:middle line:84% But everybody who submits something 00:11:28.110 --> 00:11:31.010 align:middle line:84% will have a non-zero probability. 00:11:31.010 --> 00:11:33.690 align:middle line:84% And I haven't yet set up that spreadsheet. 00:11:33.690 --> 00:11:36.470 align:middle line:90% But I'll do that for next week. 00:11:36.470 --> 00:11:38.230 align:middle line:90% So that's the plan. 00:11:38.230 --> 00:11:39.330 align:middle line:90% Other questions? 00:11:39.330 --> 00:11:43.590 align:middle line:90% 00:11:43.590 --> 00:11:44.510 align:middle line:90% OK. 00:11:44.510 --> 00:11:46.870 align:middle line:84% Anybody online, anything that you want 00:11:46.870 --> 00:11:48.890 align:middle line:90% to share or ask any questions? 00:11:48.890 --> 00:11:52.150 align:middle line:90% 00:11:52.150 --> 00:11:55.430 align:middle line:90% OK. 00:11:55.430 --> 00:11:59.030 align:middle line:84% So when we talk about system dynamics modeling, 00:11:59.030 --> 00:12:03.150 align:middle line:84% we're actually talking about a big category of possible model 00:12:03.150 --> 00:12:04.510 align:middle line:90% types. 00:12:04.510 --> 00:12:06.350 align:middle line:90% And that is very important. 00:12:06.350 --> 00:12:09.990 align:middle line:84% What is the purpose of the model we are building 00:12:09.990 --> 00:12:15.810 align:middle line:84% would have a big impact on how detailed your model needs to be, 00:12:15.810 --> 00:12:17.570 align:middle line:84% how empirically grounded it needs to be, 00:12:17.570 --> 00:12:20.370 align:middle line:84% what kind of a process for model building you may use. 00:12:20.370 --> 00:12:23.383 align:middle line:84% So it is important to be mindful of those use cases. 00:12:23.383 --> 00:12:25.550 align:middle line:84% And we are not going to cover all of those use cases 00:12:25.550 --> 00:12:28.670 align:middle line:84% in this class, partly because some of those use cases 00:12:28.670 --> 00:12:31.230 align:middle line:84% require a lot of detailed work that is 00:12:31.230 --> 00:12:33.010 align:middle line:90% beyond the scope of this class. 00:12:33.010 --> 00:12:35.970 align:middle line:84% Some of them require a lot of working with stakeholders, 00:12:35.970 --> 00:12:38.310 align:middle line:84% group model building type of activities that is not well 00:12:38.310 --> 00:12:40.890 align:middle line:84% designed for what we can do in this class. 00:12:40.890 --> 00:12:44.370 align:middle line:84% But I want you to be aware of these different use cases 00:12:44.370 --> 00:12:47.590 align:middle line:84% because the process you learn here 00:12:47.590 --> 00:12:53.950 align:middle line:84% is optimized for a subset of these modeling practices. 00:12:53.950 --> 00:12:56.290 align:middle line:84% It is useful for all the other ones as well. 00:12:56.290 --> 00:13:00.470 align:middle line:84% But you need to know that there are additional components 00:13:00.470 --> 00:13:03.910 align:middle line:84% in your skill set that would be beneficial if, for example, you 00:13:03.910 --> 00:13:06.950 align:middle line:84% are doing modeling to bring a group together 00:13:06.950 --> 00:13:11.070 align:middle line:84% or you are doing modeling to do predictions in a very 00:13:11.070 --> 00:13:14.090 align:middle line:90% detailed, future-oriented task. 00:13:14.090 --> 00:13:20.150 align:middle line:84% So here are a few examples or key categories of uses 00:13:20.150 --> 00:13:21.950 align:middle line:90% of system dynamics models. 00:13:21.950 --> 00:13:24.330 align:middle line:84% So what I call process consultation, 00:13:24.330 --> 00:13:27.710 align:middle line:84% I'm thinking of it as a broad category, that models become 00:13:27.710 --> 00:13:33.390 align:middle line:84% useful artifacts for helping a group, a set of stakeholders, 00:13:33.390 --> 00:13:36.150 align:middle line:84% to come together, have a more productive conversation 00:13:36.150 --> 00:13:37.880 align:middle line:90% about a complex problem. 00:13:37.880 --> 00:13:40.560 align:middle line:84% So in this case, group model building 00:13:40.560 --> 00:13:46.040 align:middle line:84% would be one of those approaches to bring a group together 00:13:46.040 --> 00:13:48.700 align:middle line:84% to build a qualitative model, usually at the beginning, 00:13:48.700 --> 00:13:51.580 align:middle line:84% and then typically adding more quantitative models, 00:13:51.580 --> 00:13:53.880 align:middle line:84% though there are different schools of doing group model 00:13:53.880 --> 00:13:55.680 align:middle line:84% building, where you may start with actually 00:13:55.680 --> 00:13:58.000 align:middle line:84% a simple quantitative model from scratch. 00:13:58.000 --> 00:14:02.360 align:middle line:84% So that's a whole set of different goals from a modeling 00:14:02.360 --> 00:14:03.040 align:middle line:90% project. 00:14:03.040 --> 00:14:08.640 align:middle line:84% Building common mental models, surfacing different perspectives 00:14:08.640 --> 00:14:11.580 align:middle line:84% that are in a group of stakeholders, 00:14:11.580 --> 00:14:14.280 align:middle line:84% those are very important goals of a process consultation 00:14:14.280 --> 00:14:16.280 align:middle line:84% type of modeling project that might 00:14:16.280 --> 00:14:19.240 align:middle line:84% be distinct from a prediction-oriented modeling 00:14:19.240 --> 00:14:20.160 align:middle line:90% project. 00:14:20.160 --> 00:14:22.960 align:middle line:84% Then we have models that are trying 00:14:22.960 --> 00:14:25.240 align:middle line:84% to build generic high level insights 00:14:25.240 --> 00:14:27.240 align:middle line:90% about interesting dynamics. 00:14:27.240 --> 00:14:29.140 align:middle line:84% These could be very much theoretical. 00:14:29.140 --> 00:14:32.940 align:middle line:84% So much of what goes into academic journals, 00:14:32.940 --> 00:14:36.040 align:middle line:84% especially in the management space that use models, 00:14:36.040 --> 00:14:37.260 align:middle line:90% fall into this category. 00:14:37.260 --> 00:14:41.920 align:middle line:84% These are usually fairly simple models, but with a clear, 00:14:41.920 --> 00:14:45.200 align:middle line:84% neat insight that you need a model to appreciate that 00:14:45.200 --> 00:14:50.600 align:middle line:84% or mental models alone cannot handle that level of complexity. 00:14:50.600 --> 00:14:52.880 align:middle line:84% Usually, the idea is to get insights 00:14:52.880 --> 00:14:57.980 align:middle line:84% that are applicable across a variety of application areas. 00:14:57.980 --> 00:15:00.400 align:middle line:84% But you are not going to make any strong claims 00:15:00.400 --> 00:15:05.800 align:middle line:84% about that insight being the key to explaining 00:15:05.800 --> 00:15:09.220 align:middle line:90% a specific empirical case. 00:15:09.220 --> 00:15:14.360 align:middle line:84% You are not doing a close match with any empirical data 00:15:14.360 --> 00:15:16.320 align:middle line:90% in the quantitative sense. 00:15:16.320 --> 00:15:20.640 align:middle line:84% The theory testing variety of these kind of insight models 00:15:20.640 --> 00:15:22.600 align:middle line:84% tries to push it one level further-- that is, 00:15:22.600 --> 00:15:27.720 align:middle line:84% not only tell a story, but also test the strength of that story 00:15:27.720 --> 00:15:30.020 align:middle line:90% for explaining specific cases. 00:15:30.020 --> 00:15:35.400 align:middle line:84% So you may have data from a single case 00:15:35.400 --> 00:15:38.260 align:middle line:84% or multiple cases in a panel data set of sorts. 00:15:38.260 --> 00:15:40.380 align:middle line:84% And you will build a simple model, 00:15:40.380 --> 00:15:42.140 align:middle line:84% estimate that model with that data. 00:15:42.140 --> 00:15:44.040 align:middle line:84% And then the strengths of those parameters 00:15:44.040 --> 00:15:46.740 align:middle line:84% will inform how strong are the mechanisms. 00:15:46.740 --> 00:15:52.120 align:middle line:84% And you can go into a more quantitative space 00:15:52.120 --> 00:15:53.920 align:middle line:90% with models in this space. 00:15:53.920 --> 00:15:56.440 align:middle line:84% I would say that is in the minority, 00:15:56.440 --> 00:15:59.500 align:middle line:84% in terms of work that is already out there in the literature. 00:15:59.500 --> 00:16:02.660 align:middle line:84% There are domains, in pharmacokinetics for example, 00:16:02.660 --> 00:16:04.280 align:middle line:84% that you see these types of models 00:16:04.280 --> 00:16:08.860 align:middle line:84% and they are bread and butter of mathematical modeling. 00:16:08.860 --> 00:16:10.080 align:middle line:90% You estimate those models. 00:16:10.080 --> 00:16:12.480 align:middle line:84% And they are been informative for a bunch 00:16:12.480 --> 00:16:14.700 align:middle line:90% of actual practical questions. 00:16:14.700 --> 00:16:18.120 align:middle line:84% In management, there are fewer of these theory testing 00:16:18.120 --> 00:16:21.600 align:middle line:84% using dynamic models work, even though theory testing is 00:16:21.600 --> 00:16:24.340 align:middle line:84% the vast majority of literature in this space. 00:16:24.340 --> 00:16:25.880 align:middle line:84% So there is actually opportunities 00:16:25.880 --> 00:16:29.320 align:middle line:84% to build that, I think, in management research as well. 00:16:29.320 --> 00:16:33.440 align:middle line:84% And then you have use cases that are more applied, 00:16:33.440 --> 00:16:36.000 align:middle line:84% whether it is for predicting things, 00:16:36.000 --> 00:16:38.480 align:middle line:84% and you build a dynamic model to try to project 00:16:38.480 --> 00:16:39.900 align:middle line:90% what will happen in the future. 00:16:39.900 --> 00:16:42.720 align:middle line:84% We did examples of those around COVID 00:16:42.720 --> 00:16:45.800 align:middle line:90% with a bunch of collaborators. 00:16:45.800 --> 00:16:49.100 align:middle line:90% That's a pretty viable space. 00:16:49.100 --> 00:16:54.520 align:middle line:84% You need models that have a decent level of detail 00:16:54.520 --> 00:16:58.160 align:middle line:84% complexity to be able to basically capture 00:16:58.160 --> 00:17:01.440 align:middle line:84% the variables that are observable in the real world, 00:17:01.440 --> 00:17:04.400 align:middle line:84% but also to be feedback rich, very well 00:17:04.400 --> 00:17:06.140 align:middle line:90% grounded in the empirical data. 00:17:06.140 --> 00:17:10.040 align:middle line:84% Usually, you have to go through estimation of these models 00:17:10.040 --> 00:17:14.839 align:middle line:84% to be able to do some meaningful predictive work. 00:17:14.839 --> 00:17:17.960 align:middle line:84% The types of models we build for prediction purposes 00:17:17.960 --> 00:17:20.460 align:middle line:84% are most useful for longer term predictions. 00:17:20.460 --> 00:17:22.500 align:middle line:84% If you are doing short-term predictions, 00:17:22.500 --> 00:17:25.040 align:middle line:84% there are lots of other tools from machine learning 00:17:25.040 --> 00:17:29.080 align:middle line:84% and statistics that are better candidates 00:17:29.080 --> 00:17:31.650 align:middle line:84% because, in the short run, you don't 00:17:31.650 --> 00:17:35.410 align:middle line:84% need the long-term feedback mechanisms that we 00:17:35.410 --> 00:17:39.250 align:middle line:84% capture in our models to think about problems. 00:17:39.250 --> 00:17:40.890 align:middle line:84% Then you have a variety of models 00:17:40.890 --> 00:17:45.190 align:middle line:84% that are more focused on policy design, decision support tools. 00:17:45.190 --> 00:17:47.970 align:middle line:84% And there could be variants that are focused 00:17:47.970 --> 00:17:51.650 align:middle line:84% on a single decision, whether we go ahead with this investment 00:17:51.650 --> 00:17:56.250 align:middle line:84% or with this merger or with this project or not. 00:17:56.250 --> 00:17:57.810 align:middle line:84% For those types of decisions, you 00:17:57.810 --> 00:18:00.750 align:middle line:84% want a well calibrated model, fairly detailed. 00:18:00.750 --> 00:18:02.410 align:middle line:84% These are usually very big investments 00:18:02.410 --> 00:18:07.610 align:middle line:84% in terms of time and effort that goes into these models. 00:18:07.610 --> 00:18:10.890 align:middle line:84% So the decision needs to be actually pretty substantive. 00:18:10.890 --> 00:18:14.250 align:middle line:84% That is, if you're spending a couple hundred thousand dollars 00:18:14.250 --> 00:18:18.030 align:middle line:84% in building a model, the impact of that decision, 00:18:18.030 --> 00:18:20.490 align:middle line:84% or a delta in the quality of that decision, 00:18:20.490 --> 00:18:23.910 align:middle line:84% should be in millions of dollars to justify that. 00:18:23.910 --> 00:18:29.310 align:middle line:84% So for simple decisions, it may not be worth the effort. 00:18:29.310 --> 00:18:32.310 align:middle line:84% And that creates a whole area in consulting. 00:18:32.310 --> 00:18:36.110 align:middle line:84% For example, in dispute resolution, in legal cases, 00:18:36.110 --> 00:18:39.530 align:middle line:84% for big projects, that is where these detailed system dynamics 00:18:39.530 --> 00:18:43.210 align:middle line:84% models have historically been very successful. 00:18:43.210 --> 00:18:46.890 align:middle line:84% Then you have decision support, which is a little bit different, 00:18:46.890 --> 00:18:50.390 align:middle line:84% in the sense that it is more focused on a regular decision, 00:18:50.390 --> 00:18:53.370 align:middle line:84% that each decision, per se, may not 00:18:53.370 --> 00:18:56.510 align:middle line:84% be valued in millions of dollars, 00:18:56.510 --> 00:18:58.590 align:middle line:90% but it is taken repeatedly. 00:18:58.590 --> 00:19:00.830 align:middle line:84% And having an underlying dynamic model, 00:19:00.830 --> 00:19:02.770 align:middle line:84% whether it is for supply chain management 00:19:02.770 --> 00:19:06.590 align:middle line:84% or for production management and so on, can be beneficial. 00:19:06.590 --> 00:19:09.770 align:middle line:84% And a lot of operations management work 00:19:09.770 --> 00:19:11.630 align:middle line:90% uses these types of models. 00:19:11.630 --> 00:19:14.130 align:middle line:84% Many of them have components of optimization 00:19:14.130 --> 00:19:16.790 align:middle line:84% combined with components of dynamic modeling. 00:19:16.790 --> 00:19:18.830 align:middle line:90% So that's another related area. 00:19:18.830 --> 00:19:21.690 align:middle line:84% Again, well grounded in empirical data, 00:19:21.690 --> 00:19:25.770 align:middle line:84% usually has more detail than the generic insight type 00:19:25.770 --> 00:19:30.210 align:middle line:84% of models, more engagement and time 00:19:30.210 --> 00:19:32.502 align:middle line:84% spent in building and maintaining those models. 00:19:32.502 --> 00:19:34.210 align:middle line:84% Especially, the maintenance is important. 00:19:34.210 --> 00:19:37.490 align:middle line:84% And they need to be updated as things go. 00:19:37.490 --> 00:19:42.730 align:middle line:84% You can think about models that go into training modules, 00:19:42.730 --> 00:19:48.050 align:middle line:84% as management flight simulators or for working 00:19:48.050 --> 00:19:51.090 align:middle line:84% with the stakeholders to help them appreciate some 00:19:51.090 --> 00:19:53.090 align:middle line:90% of the interesting dynamics. 00:19:53.090 --> 00:19:55.450 align:middle line:84% Those models need to be very robust. 00:19:55.450 --> 00:19:57.610 align:middle line:84% They are usually a little bit more detailed 00:19:57.610 --> 00:19:59.650 align:middle line:84% than the generic insight models because you 00:19:59.650 --> 00:20:05.410 align:middle line:84% need the policy levers that the stakeholders care about 00:20:05.410 --> 00:20:09.090 align:middle line:84% to be well integrated into those models. 00:20:09.090 --> 00:20:13.130 align:middle line:84% But they are not usually calibrated to any single case, 00:20:13.130 --> 00:20:16.930 align:middle line:84% even though they need to be in the right empirical space for it 00:20:16.930 --> 00:20:17.710 align:middle line:90% to be useful. 00:20:17.710 --> 00:20:22.350 align:middle line:84% So as you can see-- and you can come up with other subsets. 00:20:22.350 --> 00:20:26.290 align:middle line:84% But I think these are covering some of the more common ones. 00:20:26.290 --> 00:20:28.970 align:middle line:90% They have different flavors. 00:20:28.970 --> 00:20:33.430 align:middle line:84% And in terms of the size and complexity of the model, 00:20:33.430 --> 00:20:36.730 align:middle line:84% you can think about small exploratory models 00:20:36.730 --> 00:20:39.630 align:middle line:84% that are more focused on understanding dynamics. 00:20:39.630 --> 00:20:43.610 align:middle line:84% And you create, explore some simple policies 00:20:43.610 --> 00:20:45.550 align:middle line:90% or you go one level further. 00:20:45.550 --> 00:20:49.090 align:middle line:84% You try to not only build theory, but also 00:20:49.090 --> 00:20:51.810 align:middle line:84% test theory when you calibrate some of these models. 00:20:51.810 --> 00:20:56.370 align:middle line:84% And then you go down into more calibrated, detailed models that 00:20:56.370 --> 00:21:00.250 align:middle line:84% are for the strategic decision support for single decisions 00:21:00.250 --> 00:21:04.690 align:middle line:84% or continuous types of decisions. 00:21:04.690 --> 00:21:09.530 align:middle line:84% So our focus, by and large, is in the smaller sized models 00:21:09.530 --> 00:21:14.850 align:middle line:84% in this class that could be used for helping explore 00:21:14.850 --> 00:21:17.810 align:middle line:84% different policies, building good intuitions 00:21:17.810 --> 00:21:23.010 align:middle line:84% about serious organizational challenge. 00:21:23.010 --> 00:21:26.380 align:middle line:84% They are actually very useful for even high level 00:21:26.380 --> 00:21:30.460 align:middle line:84% strategic consultation and thinking about problems. 00:21:30.460 --> 00:21:34.523 align:middle line:84% You get a lot of benefits of the more detailed, calibrated models 00:21:34.523 --> 00:21:35.940 align:middle line:84% in the earliest stages when you're 00:21:35.940 --> 00:21:40.440 align:middle line:84% building these higher level, but feedback rich models. 00:21:40.440 --> 00:21:43.460 align:middle line:84% They are also tools for process consultation, 00:21:43.460 --> 00:21:47.380 align:middle line:84% for engaging stakeholders, and so on. 00:21:47.380 --> 00:21:51.420 align:middle line:84% We have a version of this seminar that is focused only 00:21:51.420 --> 00:21:52.960 align:middle line:90% on calibration of models. 00:21:52.960 --> 00:21:58.180 align:middle line:84% So when you're interested in more detailed estimation work, 00:21:58.180 --> 00:21:59.380 align:middle line:90% there is a whole machinery. 00:21:59.380 --> 00:22:03.020 align:middle line:84% And it takes quite a bit of time to set that up and learn 00:22:03.020 --> 00:22:04.660 align:middle line:90% all the basics there. 00:22:04.660 --> 00:22:10.060 align:middle line:84% But this is where we are focused on for this class. 00:22:10.060 --> 00:22:14.500 align:middle line:84% Also, to give you a little bit more feel for time distribution 00:22:14.500 --> 00:22:19.540 align:middle line:84% and how much effort goes into different types of modeling 00:22:19.540 --> 00:22:22.060 align:middle line:84% efforts, you can think about, if you 00:22:22.060 --> 00:22:25.700 align:middle line:84% have a big enchilada of a model that 00:22:25.700 --> 00:22:28.260 align:middle line:84% has all the bells and whistles and is well calibrated 00:22:28.260 --> 00:22:33.300 align:middle line:84% and is detailed and so on, the biggest chunk of that work 00:22:33.300 --> 00:22:37.500 align:middle line:84% is actually going into the calibration work estimation, 00:22:37.500 --> 00:22:39.820 align:middle line:84% connecting it with the data, which 00:22:39.820 --> 00:22:44.440 align:middle line:84% requires you to go collect the data, estimate the model, 00:22:44.440 --> 00:22:45.620 align:middle line:90% do analysis. 00:22:45.620 --> 00:22:48.340 align:middle line:84% So that's actually the biggest chunk 00:22:48.340 --> 00:22:52.720 align:middle line:84% of the time that goes into those big strategic models, 00:22:52.720 --> 00:22:54.820 align:middle line:90% big consulting models. 00:22:54.820 --> 00:22:59.880 align:middle line:84% And also, these are things that you can break down into chunks. 00:22:59.880 --> 00:23:02.020 align:middle line:84% And some parts of the team could be 00:23:02.020 --> 00:23:04.248 align:middle line:84% focused on collecting the data and cleaning it up. 00:23:04.248 --> 00:23:06.540 align:middle line:84% And some part could be on the technical estimation side 00:23:06.540 --> 00:23:09.120 align:middle line:90% and some on the analysis side. 00:23:09.120 --> 00:23:11.980 align:middle line:84% So you usually tackle these problems in larger teams 00:23:11.980 --> 00:23:18.880 align:middle line:84% rather than a one man or woman show modeler situation. 00:23:18.880 --> 00:23:21.440 align:middle line:84% And then on the insight building, 00:23:21.440 --> 00:23:27.680 align:middle line:84% which is the smaller part of the modeling tasks, 00:23:27.680 --> 00:23:31.360 align:middle line:84% there is both the modeling and the analysis. 00:23:31.360 --> 00:23:35.580 align:middle line:84% And for this class, we think of modeling and analysis 00:23:35.580 --> 00:23:37.120 align:middle line:90% as very tightly connected. 00:23:37.120 --> 00:23:42.560 align:middle line:84% And if you went through the memo that I shared with you, 00:23:42.560 --> 00:23:46.700 align:middle line:84% you probably noticed that iterative modeling process, 00:23:46.700 --> 00:23:49.480 align:middle line:84% where you add a chunk, you understand that chunk, 00:23:49.480 --> 00:23:53.720 align:middle line:84% and then you go about adding the next chunk of the model 00:23:53.720 --> 00:23:55.660 align:middle line:90% and so on. 00:23:55.660 --> 00:23:58.060 align:middle line:84% Now, for this class, we are going 00:23:58.060 --> 00:24:02.020 align:middle line:84% to go through a modeling process that 00:24:02.020 --> 00:24:07.260 align:middle line:84% has some of the major components of any typical modeling 00:24:07.260 --> 00:24:07.960 align:middle line:90% initiative. 00:24:07.960 --> 00:24:11.660 align:middle line:84% But we focus on a subset of those activities. 00:24:11.660 --> 00:24:14.780 align:middle line:84% So basically, if you break down the modeling process 00:24:14.780 --> 00:24:18.040 align:middle line:84% into problem definition, building a qualitative model, 00:24:18.040 --> 00:24:21.220 align:middle line:84% building a simulation model, doing the calibration 00:24:21.220 --> 00:24:25.740 align:middle line:84% and estimation, and doing the analysis and insights, 00:24:25.740 --> 00:24:30.400 align:middle line:84% we are not spending too much time on problem definition 00:24:30.400 --> 00:24:34.020 align:middle line:84% because I usually try to come up with a decently well-defined 00:24:34.020 --> 00:24:35.600 align:middle line:90% problem definition for you. 00:24:35.600 --> 00:24:38.220 align:middle line:84% And then we will have a conversation 00:24:38.220 --> 00:24:45.020 align:middle line:84% in the pre-modeling session, where we discuss the problem 00:24:45.020 --> 00:24:46.580 align:middle line:90% definition. 00:24:46.580 --> 00:24:48.860 align:middle line:84% Sometimes, we may have reference modes, 00:24:48.860 --> 00:24:52.640 align:middle line:84% sometimes even not have very specific reference modes. 00:24:52.640 --> 00:24:54.660 align:middle line:84% Sometimes, we may discuss momentum policies 00:24:54.660 --> 00:24:56.500 align:middle line:90% when they make sense. 00:24:56.500 --> 00:24:58.220 align:middle line:90% Sometimes, we wouldn't. 00:24:58.220 --> 00:25:01.000 align:middle line:84% We usually wouldn't even go through a list of variables. 00:25:01.000 --> 00:25:04.380 align:middle line:84% So those are steps that are useful in defining the problem. 00:25:04.380 --> 00:25:09.140 align:middle line:84% We don't necessarily go through all of those in this class. 00:25:09.140 --> 00:25:11.920 align:middle line:84% By the way, anybody knows what are momentum policies? 00:25:11.920 --> 00:25:15.940 align:middle line:84% This is probably a new term for many of you. 00:25:15.940 --> 00:25:19.750 align:middle line:84% So momentum policies are, if you're 00:25:19.750 --> 00:25:21.750 align:middle line:84% working with a stakeholder, they usually 00:25:21.750 --> 00:25:25.770 align:middle line:84% come with a problem that is more pragmatic. 00:25:25.770 --> 00:25:28.510 align:middle line:90% They need to solve something. 00:25:28.510 --> 00:25:32.430 align:middle line:84% And usually, they have their existing priors 00:25:32.430 --> 00:25:34.710 align:middle line:84% on what would be a decent solution 00:25:34.710 --> 00:25:38.350 align:middle line:84% if they didn't do any modeling work with you. 00:25:38.350 --> 00:25:40.910 align:middle line:84% You want to elicit that so that you 00:25:40.910 --> 00:25:42.610 align:middle line:84% know what is their starting position, 00:25:42.610 --> 00:25:45.470 align:middle line:84% they know what is their starting position. 00:25:45.470 --> 00:25:47.970 align:middle line:84% And what are the benefits of doing that? 00:25:47.970 --> 00:25:50.750 align:middle line:84% First of all, it informs you what kind of policies 00:25:50.750 --> 00:25:54.670 align:middle line:84% you actually need to be able to reflect in your modeling work 00:25:54.670 --> 00:25:57.250 align:middle line:84% and analyze in your modeling work. 00:25:57.250 --> 00:26:00.590 align:middle line:84% So it tells you something about the structures that are needed 00:26:00.590 --> 00:26:03.330 align:middle line:90% to capture those policies. 00:26:03.330 --> 00:26:04.630 align:middle line:90% So that's number 1. 00:26:04.630 --> 00:26:08.890 align:middle line:84% Number 2 is that, in a few sessions or a few months, 00:26:08.890 --> 00:26:11.910 align:middle line:84% depending on how long an engagement you have, 00:26:11.910 --> 00:26:17.910 align:middle line:84% you probably have changed, in the process of modeling 00:26:17.910 --> 00:26:21.150 align:middle line:84% and working with your client, how they think about the problem 00:26:21.150 --> 00:26:25.470 align:middle line:84% and how they prioritize different policies. 00:26:25.470 --> 00:26:28.750 align:middle line:84% So you want to have put a stick in the ground that 00:26:28.750 --> 00:26:32.930 align:middle line:84% allows them to see how far they have come along, 00:26:32.930 --> 00:26:35.530 align:middle line:84% how much their thinking about the problem has changed, 00:26:35.530 --> 00:26:41.310 align:middle line:84% how they would now differently prioritize different policy 00:26:41.310 --> 00:26:44.550 align:middle line:84% levers so that they can appreciate what the modeling 00:26:44.550 --> 00:26:47.670 align:middle line:84% process has added to their thinking. 00:26:47.670 --> 00:26:50.870 align:middle line:84% This is important because the way we usually 00:26:50.870 --> 00:26:53.310 align:middle line:84% engage with stakeholders in a modeling project 00:26:53.310 --> 00:26:54.650 align:middle line:90% is very continuous. 00:26:54.650 --> 00:26:57.090 align:middle line:84% At every stage, you want them to be involved. 00:26:57.090 --> 00:26:58.830 align:middle line:84% Now in this class, you are not going 00:26:58.830 --> 00:27:01.810 align:middle line:84% through the process consultation type of modeling work. 00:27:01.810 --> 00:27:05.310 align:middle line:84% But you want engagement in any actual project 00:27:05.310 --> 00:27:07.930 align:middle line:84% because that's how people actually change their minds. 00:27:07.930 --> 00:27:09.790 align:middle line:84% They are not going to trust some black box 00:27:09.790 --> 00:27:11.430 align:middle line:90% model that you throw at them. 00:27:11.430 --> 00:27:15.990 align:middle line:84% But that means they gradually change their thinking. 00:27:15.990 --> 00:27:19.870 align:middle line:84% And three months later, everything seems obvious. 00:27:19.870 --> 00:27:21.910 align:middle line:84% So they may not appreciate how much 00:27:21.910 --> 00:27:26.270 align:middle line:84% the modeling process has helped them change their perspective. 00:27:26.270 --> 00:27:28.430 align:middle line:84% So having these momentum policies at the beginning 00:27:28.430 --> 00:27:31.310 align:middle line:84% helps them see the impact of the modeling project 00:27:31.310 --> 00:27:33.170 align:middle line:84% appreciate that much more clearly. 00:27:33.170 --> 00:27:35.950 align:middle line:84% Anyway, we are not doing momentum policies 00:27:35.950 --> 00:27:38.130 align:middle line:84% in any serious way in this class. 00:27:38.130 --> 00:27:39.190 align:middle line:90% We don't have clients. 00:27:39.190 --> 00:27:41.370 align:middle line:84% And this is not as much focused on that. 00:27:41.370 --> 00:27:43.750 align:middle line:84% But it is an important component in thinking 00:27:43.750 --> 00:27:47.310 align:middle line:84% about problem definition when you're working with clients. 00:27:47.310 --> 00:27:52.610 align:middle line:84% So typically, I give you some version of a qualitative model, 00:27:52.610 --> 00:27:54.890 align:middle line:84% sometimes with explicit causal loop diagrams, 00:27:54.890 --> 00:27:57.550 align:middle line:84% as we had for this week, even though we are not 00:27:57.550 --> 00:28:01.310 align:middle line:84% building models along those causal loop diagrams 00:28:01.310 --> 00:28:05.750 align:middle line:84% we had for this week, sometimes with just a verbal description 00:28:05.750 --> 00:28:08.430 align:middle line:84% of the mechanisms that are relevant. 00:28:08.430 --> 00:28:11.670 align:middle line:84% Either way, you have the outlines 00:28:11.670 --> 00:28:13.260 align:middle line:90% of the qualitative model. 00:28:13.260 --> 00:28:16.320 align:middle line:84% Your focus will be on building the simulation model that 00:28:16.320 --> 00:28:19.700 align:middle line:84% captures those dynamics and doing the analysis, 00:28:19.700 --> 00:28:20.920 align:middle line:90% building insights. 00:28:20.920 --> 00:28:26.040 align:middle line:84% And that analysis process is a very iterative process, 00:28:26.040 --> 00:28:30.320 align:middle line:84% in the sense that you yourself, in your week long journey, 00:28:30.320 --> 00:28:33.800 align:middle line:84% you may start with the simplest chunk 00:28:33.800 --> 00:28:36.560 align:middle line:90% that you think you can model. 00:28:36.560 --> 00:28:38.920 align:middle line:84% You model that, you understand that, 00:28:38.920 --> 00:28:43.860 align:middle line:84% and you need to be very specific about, OK, 00:28:43.860 --> 00:28:46.320 align:middle line:84% I'm leaving out a lot of things that are in the problem 00:28:46.320 --> 00:28:49.160 align:middle line:84% definition because I don't feel comfortable I can incorporate 00:28:49.160 --> 00:28:51.820 align:middle line:84% all of those in my first iteration of the model. 00:28:51.820 --> 00:28:54.040 align:middle line:90% That's the right way to do it. 00:28:54.040 --> 00:28:56.340 align:middle line:84% Don't try to incorporate everything, 00:28:56.340 --> 00:28:58.220 align:middle line:84% if you don't feel comfortable with that. 00:28:58.220 --> 00:29:00.760 align:middle line:84% And we'll talk about how you would feel comfortable 00:29:00.760 --> 00:29:03.560 align:middle line:90% or not as we go through this. 00:29:03.560 --> 00:29:06.640 align:middle line:84% It's completely legitimate, fine to come to the class 00:29:06.640 --> 00:29:10.680 align:middle line:84% and say, half of the things that you asked for in the problem 00:29:10.680 --> 00:29:12.580 align:middle line:84% definition are not part of my model 00:29:12.580 --> 00:29:15.400 align:middle line:84% because I didn't get a chance to get to that iteration 00:29:15.400 --> 00:29:16.340 align:middle line:90% to incorporate those. 00:29:16.340 --> 00:29:18.360 align:middle line:90% That's fine. 00:29:18.360 --> 00:29:20.800 align:middle line:84% Take this iterative approach seriously. 00:29:20.800 --> 00:29:23.100 align:middle line:84% And the iteration means there is another step. 00:29:23.100 --> 00:29:24.800 align:middle line:90% And that is the analysis step. 00:29:24.800 --> 00:29:26.820 align:middle line:84% And the analysis step is critical. 00:29:26.820 --> 00:29:27.820 align:middle line:90% We'll talk about it. 00:29:27.820 --> 00:29:31.760 align:middle line:84% It is by far the most underappreciated and 00:29:31.760 --> 00:29:34.500 align:middle line:84% underutilized part of the modeling process. 00:29:34.500 --> 00:29:38.800 align:middle line:84% I see it in lots of senior people submitting papers 00:29:38.800 --> 00:29:40.713 align:middle line:84% with models they have not understood. 00:29:40.713 --> 00:29:42.880 align:middle line:84% And if they have not fully understood your own model 00:29:42.880 --> 00:29:44.680 align:middle line:84% and why it is behaving the way it is, 00:29:44.680 --> 00:29:47.740 align:middle line:84% you can't really generate insights for others. 00:29:47.740 --> 00:29:52.680 align:middle line:84% This is so common that, if you just fix this issue 00:29:52.680 --> 00:29:58.440 align:middle line:84% in your modeling work, you will be ahead of 70%-80% of the pack 00:29:58.440 --> 00:30:02.720 align:middle line:84% in terms of what you can actually generate insight. 00:30:02.720 --> 00:30:06.000 align:middle line:84% OK, so that's a little bit of a big picture 00:30:06.000 --> 00:30:11.080 align:middle line:84% overview on model types, modeling process overview. 00:30:11.080 --> 00:30:15.280 align:middle line:84% Now, for the remainder of the theoretical part of the class 00:30:15.280 --> 00:30:18.060 align:middle line:84% today-- and as I said today, we have half of the class 00:30:18.060 --> 00:30:18.880 align:middle line:90% more on theory. 00:30:18.880 --> 00:30:22.140 align:middle line:84% In the future, we wouldn't have this much theory. 00:30:22.140 --> 00:30:25.040 align:middle line:84% So I would not be talking this much. 00:30:25.040 --> 00:30:25.820 align:middle line:90% For the remainder. 00:30:25.820 --> 00:30:29.480 align:middle line:84% I want to go through basically what you had in the memo, 00:30:29.480 --> 00:30:32.760 align:middle line:84% understanding that some people didn't have a chance 00:30:32.760 --> 00:30:34.700 align:middle line:90% to go through it in depth. 00:30:34.700 --> 00:30:36.720 align:middle line:84% But also, you may have questions. 00:30:36.720 --> 00:30:40.160 align:middle line:84% And we want to have a conversation around those. 00:30:40.160 --> 00:30:44.760 align:middle line:84% So I created this memo with the realization 00:30:44.760 --> 00:30:49.760 align:middle line:84% that there is a lot of art to modeling 00:30:49.760 --> 00:30:53.480 align:middle line:90% that is just hard to pin down. 00:30:53.480 --> 00:30:58.560 align:middle line:84% And it creates a very steep learning curve and also a very 00:30:58.560 --> 00:31:00.360 align:middle line:84% variable learning curve, depending 00:31:00.360 --> 00:31:02.240 align:middle line:84% on the individual and the opportunities 00:31:02.240 --> 00:31:05.180 align:middle line:84% they get to engage with modeling tasks. 00:31:05.180 --> 00:31:09.720 align:middle line:84% So you would see folks who may spent a couple of years 00:31:09.720 --> 00:31:12.640 align:middle line:84% in their learning journey and still not feel comfortable 00:31:12.640 --> 00:31:14.280 align:middle line:84% tackling the types of problems we are 00:31:14.280 --> 00:31:16.660 align:middle line:90% trying to tackle in this class. 00:31:16.660 --> 00:31:18.600 align:middle line:84% And part of that steep learning curve 00:31:18.600 --> 00:31:24.280 align:middle line:84% is that not every idea for how to do modeling well 00:31:24.280 --> 00:31:30.000 align:middle line:84% is well documented and even well specified. 00:31:30.000 --> 00:31:32.160 align:middle line:84% Experienced modelers don't necessarily 00:31:32.160 --> 00:31:34.700 align:middle line:84% have a very clear theory of how they do it. 00:31:34.700 --> 00:31:37.140 align:middle line:90% So some of it is like biking. 00:31:37.140 --> 00:31:40.060 align:middle line:84% You can't explain exactly how it happens. 00:31:40.060 --> 00:31:44.800 align:middle line:84% So I tried my best to verbalize and theorize around 00:31:44.800 --> 00:31:51.900 align:middle line:84% what some of those skills and artsy parts of the skills are 00:31:51.900 --> 00:31:54.120 align:middle line:84% and how you can actually implement those. 00:31:54.120 --> 00:31:55.960 align:middle line:84% And the way I found it most useful 00:31:55.960 --> 00:32:00.040 align:middle line:84% was to draw on the common mistakes and common challenges 00:32:00.040 --> 00:32:04.120 align:middle line:84% and pitfalls that I see early modelers deal with and try 00:32:04.120 --> 00:32:07.290 align:middle line:84% to say, OK, in response to this common challenge, what 00:32:07.290 --> 00:32:10.910 align:middle line:84% would be some of the solutions that may be helpful? 00:32:10.910 --> 00:32:16.510 align:middle line:84% So as such, it is actually a little bit subjective. 00:32:16.510 --> 00:32:19.530 align:middle line:84% It is coming from my experience of working 00:32:19.530 --> 00:32:24.010 align:middle line:84% with lots of students and folks who are doing modeling work 00:32:24.010 --> 00:32:27.930 align:middle line:84% and my own challenges learning these. 00:32:27.930 --> 00:32:31.610 align:middle line:84% But for whatever it is worth, I hope it is useful 00:32:31.610 --> 00:32:33.910 align:middle line:90% as you go through the process. 00:32:33.910 --> 00:32:36.850 align:middle line:84% Now, I'm hoping that I will be editing this memo 00:32:36.850 --> 00:32:38.970 align:middle line:84% over the semester based on our experiences 00:32:38.970 --> 00:32:42.287 align:middle line:84% as well, so that it becomes a more useful guide for others 00:32:42.287 --> 00:32:42.870 align:middle line:90% in the future. 00:32:42.870 --> 00:32:46.070 align:middle line:84% So if something is resonating, highlight that. 00:32:46.070 --> 00:32:48.030 align:middle line:84% But also, if something is not working for you-- 00:32:48.030 --> 00:32:50.610 align:middle line:84% you tried something and you didn't 00:32:50.610 --> 00:32:52.930 align:middle line:84% know what even it means the way I have 00:32:52.930 --> 00:32:54.850 align:middle line:90% expressed it-- bring it up. 00:32:54.850 --> 00:32:59.250 align:middle line:84% This is going to be helpful to make 00:32:59.250 --> 00:33:01.810 align:middle line:84% something more useful for the broader community. 00:33:01.810 --> 00:33:06.150 align:middle line:84% So big picture-- I, so far in the memo, 00:33:06.150 --> 00:33:12.050 align:middle line:84% have highlighted seven different failure modes of sorts, 00:33:12.050 --> 00:33:13.730 align:middle line:84% though you could aggregate some of them 00:33:13.730 --> 00:33:17.390 align:middle line:84% together and break down some of them separately. 00:33:17.390 --> 00:33:21.370 align:middle line:84% And for each, I'll go through some of the key ideas 00:33:21.370 --> 00:33:24.350 align:middle line:90% for solving those. 00:33:24.350 --> 00:33:26.710 align:middle line:84% So starting with unclear problem definition, 00:33:26.710 --> 00:33:31.850 align:middle line:84% perhaps one of the most common issues with modeling 00:33:31.850 --> 00:33:37.290 align:middle line:84% projects, and a couple of ideas for tackling it-- that 00:33:37.290 --> 00:33:40.510 align:middle line:84% is, starting with a clear problem definition. 00:33:40.510 --> 00:33:41.388 align:middle line:90% That's obvious. 00:33:41.388 --> 00:33:42.930 align:middle line:84% But how do you get to a clear problem 00:33:42.930 --> 00:33:46.250 align:middle line:90% definition is the question. 00:33:46.250 --> 00:33:48.570 align:middle line:84% Partly, you could do that by specifying 00:33:48.570 --> 00:33:50.650 align:middle line:90% important architectural choices. 00:33:50.650 --> 00:33:52.570 align:middle line:84% And we'll talk about some of those. 00:33:52.570 --> 00:33:58.210 align:middle line:84% So having a problem focus is key because it determines everything 00:33:58.210 --> 00:34:00.070 align:middle line:84% that you do in the modeling project, 00:34:00.070 --> 00:34:02.450 align:middle line:84% every choice you make-- what variable goes in 00:34:02.450 --> 00:34:04.370 align:middle line:84% and what variable doesn't go in, what 00:34:04.370 --> 00:34:08.350 align:middle line:84% is the boundary of the model, how detailed you need to get, 00:34:08.350 --> 00:34:10.610 align:middle line:84% whether you need to bring in quantitative data 00:34:10.610 --> 00:34:12.969 align:middle line:90% or qualitative data is enough. 00:34:12.969 --> 00:34:17.770 align:middle line:84% All these things are only answerable if you have 00:34:17.770 --> 00:34:19.889 align:middle line:90% a clear problem definition. 00:34:19.889 --> 00:34:22.610 align:middle line:84% When you don't and your problem definition 00:34:22.610 --> 00:34:25.530 align:middle line:84% is kind of wishy-washy, the risk is 00:34:25.530 --> 00:34:31.710 align:middle line:84% that you're trying to model a system, you think of the system, 00:34:31.710 --> 00:34:33.370 align:middle line:84% you have a view of the system, and then 00:34:33.370 --> 00:34:36.290 align:middle line:84% everything could be argued that needs 00:34:36.290 --> 00:34:38.210 align:middle line:90% to be included in the model. 00:34:38.210 --> 00:34:40.510 align:middle line:84% And then it creates-- usually, that approach, 00:34:40.510 --> 00:34:45.010 align:middle line:84% even if you are a very talented, capable modeler in terms 00:34:45.010 --> 00:34:47.650 align:middle line:84% of technicality of equations, you 00:34:47.650 --> 00:34:50.130 align:middle line:84% will end up adding so much detail complexity 00:34:50.130 --> 00:34:54.010 align:middle line:84% into your model that the model crumbles under its own weight. 00:34:54.010 --> 00:34:58.090 align:middle line:84% And also, it cannot communicate with anybody because it becomes 00:34:58.090 --> 00:35:02.090 align:middle line:84% such a big, unwieldy beast that nobody can understand what is 00:35:02.090 --> 00:35:03.250 align:middle line:90% happening under it. 00:35:03.250 --> 00:35:07.930 align:middle line:84% So the problem definition is your measure 00:35:07.930 --> 00:35:09.650 align:middle line:84% for deciding what is in and what is 00:35:09.650 --> 00:35:13.890 align:middle line:84% out and all of the architectural choices that go with it. 00:35:13.890 --> 00:35:16.430 align:middle line:84% Now, the way we define a problem, 00:35:16.430 --> 00:35:19.030 align:middle line:84% there are different approaches for defining a problem. 00:35:19.030 --> 00:35:23.730 align:middle line:84% And three of them, I think, are more common. 00:35:23.730 --> 00:35:26.310 align:middle line:84% Perhaps the first one, which is reference mode driven, 00:35:26.310 --> 00:35:28.830 align:middle line:84% is the most common in the system dynamics community. 00:35:28.830 --> 00:35:32.570 align:middle line:84% But it's by no means the only way to define a problem. 00:35:32.570 --> 00:35:34.970 align:middle line:84% But the basic idea there is that you 00:35:34.970 --> 00:35:39.210 align:middle line:84% have a historical behavioral mode, a reference mode. 00:35:39.210 --> 00:35:41.230 align:middle line:84% So things have been changing over time. 00:35:41.230 --> 00:35:43.310 align:middle line:90% Our profitability was growing. 00:35:43.310 --> 00:35:46.870 align:middle line:84% But now, it has stagnated over the last couple of years. 00:35:46.870 --> 00:35:51.410 align:middle line:84% You're trying to understand what drove those dynamics over time. 00:35:51.410 --> 00:35:54.750 align:middle line:84% So having a clear reference mode is a very good starting point 00:35:54.750 --> 00:35:58.010 align:middle line:84% because then it grounds you to whatever 00:35:58.010 --> 00:36:00.400 align:middle line:84% is needed to explain that mode of behavior 00:36:00.400 --> 00:36:03.820 align:middle line:84% or at least create plausible hypotheses for explaining 00:36:03.820 --> 00:36:05.080 align:middle line:90% that mode of behavior. 00:36:05.080 --> 00:36:08.940 align:middle line:84% And anything that is not relevant for that explanation 00:36:08.940 --> 00:36:12.940 align:middle line:84% would, for now, be out of the problem definition and model 00:36:12.940 --> 00:36:13.440 align:middle line:90% boundaries. 00:36:13.440 --> 00:36:18.140 align:middle line:84% So that's one way to tackle a problem definition-- pretty 00:36:18.140 --> 00:36:21.060 align:middle line:84% common, relatively easy in the sense 00:36:21.060 --> 00:36:23.480 align:middle line:84% that, once you have a reference mode, 00:36:23.480 --> 00:36:26.300 align:middle line:90% the next steps fall in place. 00:36:26.300 --> 00:36:30.040 align:middle line:84% Then you come up with qualitative stories of, OK, 00:36:30.040 --> 00:36:31.980 align:middle line:90% what might be the mechanisms. 00:36:31.980 --> 00:36:33.720 align:middle line:84% In collaboration with the stakeholders, 00:36:33.720 --> 00:36:37.940 align:middle line:84% you may generate a set of causal loop diagrams 00:36:37.940 --> 00:36:41.580 align:middle line:84% that help you come up with mechanisms 00:36:41.580 --> 00:36:45.180 align:middle line:84% you think may explain, for example, the exponential growth 00:36:45.180 --> 00:36:48.540 align:middle line:84% mode at the earliest phase and the slowdown 00:36:48.540 --> 00:36:52.380 align:middle line:84% separately in the later stages and so on. 00:36:52.380 --> 00:36:55.580 align:middle line:90% So that's method 1. 00:36:55.580 --> 00:36:59.540 align:middle line:84% The second method starts with mechanisms. 00:36:59.540 --> 00:37:03.020 align:middle line:84% That is, it says, oh, there are some interesting structural 00:37:03.020 --> 00:37:05.320 align:middle line:90% features in this problem space. 00:37:05.320 --> 00:37:07.620 align:middle line:84% So maybe there are two actors who 00:37:07.620 --> 00:37:09.780 align:middle line:90% are competing with each other. 00:37:09.780 --> 00:37:12.900 align:middle line:84% And that competition, I want to understand 00:37:12.900 --> 00:37:15.060 align:middle line:90% what are the implications. 00:37:15.060 --> 00:37:18.540 align:middle line:84% So it already puts you in a space 00:37:18.540 --> 00:37:22.800 align:middle line:84% you need to think about interaction between two actors. 00:37:22.800 --> 00:37:25.980 align:middle line:84% Now, whether you formalize it as a game theoretic case 00:37:25.980 --> 00:37:30.400 align:middle line:84% or something more heuristic, that's an architectural choice. 00:37:30.400 --> 00:37:37.700 align:middle line:84% But that limits the space of things you want to explore. 00:37:37.700 --> 00:37:41.960 align:middle line:84% You may have, for example, dynamics of word of mouth. 00:37:41.960 --> 00:37:45.320 align:middle line:84% And you're interested in its implications. 00:37:45.320 --> 00:37:49.020 align:middle line:84% So this gets you something similar to a Bass diffusion 00:37:49.020 --> 00:37:52.120 align:middle line:90% model in the simplest case. 00:37:52.120 --> 00:37:54.100 align:middle line:84% And then you may say, OK, what if we 00:37:54.100 --> 00:37:57.460 align:middle line:84% have another other components added to this-- for example, 00:37:57.460 --> 00:38:01.500 align:middle line:84% the expansion of potential adopters in a Bass diffusion 00:38:01.500 --> 00:38:06.460 align:middle line:84% model as the product becomes more viable, better 00:38:06.460 --> 00:38:09.040 align:middle line:84% in terms of its quality, its prices, and so on. 00:38:09.040 --> 00:38:11.180 align:middle line:84% So there is endogenous dynamics of improvement 00:38:11.180 --> 00:38:14.180 align:middle line:84% in the product, which then opens the door for a larger 00:38:14.180 --> 00:38:16.940 align:middle line:84% group of population to be interested in it. 00:38:16.940 --> 00:38:21.520 align:middle line:84% So once you have specified these mechanisms, 00:38:21.520 --> 00:38:23.000 align:middle line:84% now you have your marching orders. 00:38:23.000 --> 00:38:26.620 align:middle line:84% I want to build a generic model that captures those mechanisms 00:38:26.620 --> 00:38:30.500 align:middle line:84% and explore the implications in terms of some policy question 00:38:30.500 --> 00:38:33.420 align:middle line:90% or some theoretical question. 00:38:33.420 --> 00:38:35.180 align:middle line:84% This approach is actually very common. 00:38:35.180 --> 00:38:37.860 align:middle line:84% In econ modeling, this is typically 00:38:37.860 --> 00:38:39.700 align:middle line:90% how people build models. 00:38:39.700 --> 00:38:43.400 align:middle line:84% In much of the theoretical organization theory, 00:38:43.400 --> 00:38:45.980 align:middle line:84% this is how models are typically built. 00:38:45.980 --> 00:38:50.820 align:middle line:84% And in fact, you have canonical core models in this space, 00:38:50.820 --> 00:38:55.300 align:middle line:84% whether it might be the NK model or the multi-armed bandit model 00:38:55.300 --> 00:39:01.440 align:middle line:84% or the prisoner's dilemma model, that are basic building blocks. 00:39:01.440 --> 00:39:04.060 align:middle line:84% And then people add components to them. 00:39:04.060 --> 00:39:06.280 align:middle line:84% And that is how you make new contributions. 00:39:06.280 --> 00:39:09.400 align:middle line:84% That is, OK, that canonical model, take it, 00:39:09.400 --> 00:39:12.980 align:middle line:90% add this additional piece. 00:39:12.980 --> 00:39:16.220 align:middle line:84% Let's explore what will happen, how it changes our policy 00:39:16.220 --> 00:39:20.560 align:middle line:84% conclusions, how it changes our theoretical intuitions. 00:39:20.560 --> 00:39:22.515 align:middle line:90% So that's the second one. 00:39:22.515 --> 00:39:24.140 align:middle line:84% Sometimes, you can get from a reference 00:39:24.140 --> 00:39:29.500 align:middle line:84% mode to a core canonical model by capturing 00:39:29.500 --> 00:39:30.500 align:middle line:90% some of the mechanisms. 00:39:30.500 --> 00:39:31.720 align:middle line:90% That is common. 00:39:31.720 --> 00:39:34.740 align:middle line:84% So the two are not excluding each other, but different 00:39:34.740 --> 00:39:38.580 align:middle line:84% approaches to get to similar things, ultimately. 00:39:38.580 --> 00:39:40.340 align:middle line:84% Then you have a different approach 00:39:40.340 --> 00:39:43.020 align:middle line:84% that is more focused on the policy questions 00:39:43.020 --> 00:39:44.600 align:middle line:90% that your client is asking. 00:39:44.600 --> 00:39:49.380 align:middle line:84% So you have a stakeholder, a client, yourself. 00:39:49.380 --> 00:39:51.700 align:middle line:84% You have a few actual policy questions. 00:39:51.700 --> 00:39:52.510 align:middle line:90% Should I do x? 00:39:52.510 --> 00:39:54.230 align:middle line:90% Or should I do y? 00:39:54.230 --> 00:39:57.430 align:middle line:84% And then the question is, what model 00:39:57.430 --> 00:40:00.070 align:middle line:84% can help you address that question thoroughly 00:40:00.070 --> 00:40:02.970 align:middle line:90% and with decent confidence? 00:40:02.970 --> 00:40:06.830 align:middle line:84% And the ability to answer those questions 00:40:06.830 --> 00:40:09.750 align:middle line:84% becomes the driving force, the light 00:40:09.750 --> 00:40:13.870 align:middle line:84% that shows the way in terms of problem definition. 00:40:13.870 --> 00:40:15.990 align:middle line:84% It's attractive because the clients 00:40:15.990 --> 00:40:19.270 align:middle line:84% connect with this type of problem definition very well. 00:40:19.270 --> 00:40:22.710 align:middle line:84% It is challenging because it is hard to know 00:40:22.710 --> 00:40:28.070 align:middle line:84% what needs to be captured and at what level of detail 00:40:28.070 --> 00:40:30.790 align:middle line:84% to address those policy questions. 00:40:30.790 --> 00:40:33.670 align:middle line:84% So there will be lots of uncertainty 00:40:33.670 --> 00:40:35.150 align:middle line:84% about those architectural choices 00:40:35.150 --> 00:40:37.090 align:middle line:84% when you take the policy driven approach. 00:40:37.090 --> 00:40:39.190 align:middle line:84% And that is, in some way, the reason 00:40:39.190 --> 00:40:43.350 align:middle line:84% I don't necessarily recommend that for early modelers. 00:40:43.350 --> 00:40:45.870 align:middle line:84% But as you build your capabilities, 00:40:45.870 --> 00:40:49.390 align:middle line:84% you become more comfortable taking this approach 00:40:49.390 --> 00:40:51.670 align:middle line:90% to problem definition. 00:40:51.670 --> 00:40:53.070 align:middle line:90% So this is one way. 00:40:53.070 --> 00:40:55.830 align:middle line:84% Basically, try to define the problem early on, 00:40:55.830 --> 00:40:58.070 align:middle line:90% using any of these approaches. 00:40:58.070 --> 00:41:00.670 align:middle line:84% That way, you have a clear problem definition. 00:41:00.670 --> 00:41:04.150 align:middle line:84% Usually, you want to write it down in verbally. 00:41:04.150 --> 00:41:07.330 align:middle line:84% If you have reference modes, highlight that as well. 00:41:07.330 --> 00:41:10.030 align:middle line:84% And come back to it, potentially edit it, but also 00:41:10.030 --> 00:41:13.210 align:middle line:84% come back to it to decide, when you get into a decision, 00:41:13.210 --> 00:41:15.950 align:middle line:84% do I include this feature or not. 00:41:15.950 --> 00:41:17.770 align:middle line:84% Go back to your problem definition. 00:41:17.770 --> 00:41:20.730 align:middle line:84% Does it require you to include this feature? 00:41:20.730 --> 00:41:23.310 align:middle line:84% Only add things to your model when 00:41:23.310 --> 00:41:27.250 align:middle line:84% they're really necessary with your problem definition. 00:41:27.250 --> 00:41:29.590 align:middle line:84% And most often, you don't know whether something 00:41:29.590 --> 00:41:31.110 align:middle line:90% is really necessary. 00:41:31.110 --> 00:41:33.670 align:middle line:84% So that is why we have this iterative approach. 00:41:33.670 --> 00:41:35.990 align:middle line:84% You add components that, at the beginning, 00:41:35.990 --> 00:41:38.030 align:middle line:90% sound the most important. 00:41:38.030 --> 00:41:40.923 align:middle line:84% And you go to the next levels of importance. 00:41:40.923 --> 00:41:42.590 align:middle line:84% And at some point, you may add something 00:41:42.590 --> 00:41:44.310 align:middle line:90% and you see it makes no impact. 00:41:44.310 --> 00:41:48.030 align:middle line:84% In those cases, it is safe to take it out and keep your model 00:41:48.030 --> 00:41:49.510 align:middle line:90% simple. 00:41:49.510 --> 00:41:51.910 align:middle line:84% Sometimes, you may leave it in because your clients think 00:41:51.910 --> 00:41:52.690 align:middle line:90% it is important. 00:41:52.690 --> 00:41:55.148 align:middle line:84% So you need to show them that it doesn't make a difference. 00:41:55.148 --> 00:41:57.190 align:middle line:90% That's also legitimate. 00:41:57.190 --> 00:42:00.030 align:middle line:84% OK, so that's one approach to addressing this challenge. 00:42:00.030 --> 00:42:03.870 align:middle line:84% The second approach is to tackle your key architectural choices 00:42:03.870 --> 00:42:06.070 align:middle line:84% more explicitly, to think about them. 00:42:06.070 --> 00:42:09.630 align:middle line:84% And that helps you bring out the problem definition 00:42:09.630 --> 00:42:11.470 align:middle line:90% more explicitly. 00:42:11.470 --> 00:42:14.510 align:middle line:84% So thinking about what is the boundary of the mode, 00:42:14.510 --> 00:42:18.910 align:middle line:84% so what is in and what is out, in and out in terms 00:42:18.910 --> 00:42:22.550 align:middle line:84% of mechanisms, in terms of components. 00:42:22.550 --> 00:42:23.830 align:middle line:90% What works your model? 00:42:23.830 --> 00:42:27.750 align:middle line:84% As you think about this very important architectural choice, 00:42:27.750 --> 00:42:30.430 align:middle line:84% you will see that you have to be more clear what 00:42:30.430 --> 00:42:32.150 align:middle line:90% is your problem definition. 00:42:32.150 --> 00:42:35.390 align:middle line:84% That is how you can make this choice. 00:42:35.390 --> 00:42:41.350 align:middle line:84% So as you argue for yourself, do I include demographic change 00:42:41.350 --> 00:42:44.870 align:middle line:84% or not, do I include competition or not, 00:42:44.870 --> 00:42:51.470 align:middle line:84% do I include endogenous growth in my investment pool 00:42:51.470 --> 00:42:54.030 align:middle line:84% or not, or whatever is relevant to your problem, 00:42:54.030 --> 00:42:57.070 align:middle line:84% those types of questions will force 00:42:57.070 --> 00:43:00.470 align:middle line:84% you to think more clearly what is my problem that I'm 00:43:00.470 --> 00:43:04.670 align:middle line:84% trying to solve based on which I am 00:43:04.670 --> 00:43:07.790 align:middle line:84% inclined to go in this direction or that direction. 00:43:07.790 --> 00:43:11.190 align:middle line:84% And if you track that internal thinking 00:43:11.190 --> 00:43:16.310 align:middle line:84% and you put it down on the paper, 00:43:16.310 --> 00:43:18.470 align:middle line:84% you'll see whether you are converging 00:43:18.470 --> 00:43:20.630 align:middle line:84% around a single problem definition 00:43:20.630 --> 00:43:23.150 align:middle line:84% or for each decision you are drawing 00:43:23.150 --> 00:43:26.990 align:middle line:84% on different ideas and different problem definitions. 00:43:26.990 --> 00:43:30.230 align:middle line:84% If you see the latter-- that is, your drawing for each 00:43:30.230 --> 00:43:34.850 align:middle line:84% question on different potential problems-- that is a problem. 00:43:34.850 --> 00:43:37.950 align:middle line:84% That is, you need to actually leave out 00:43:37.950 --> 00:43:39.710 align:middle line:84% a bunch of those problem definitions 00:43:39.710 --> 00:43:43.670 align:middle line:84% and focus on a core one, a unifying one, or something 00:43:43.670 --> 00:43:46.640 align:middle line:90% that is a subset of those. 00:43:46.640 --> 00:43:49.120 align:middle line:84% So boundary of the model is very helpful, very important 00:43:49.120 --> 00:43:50.480 align:middle line:90% architectural choice. 00:43:50.480 --> 00:43:52.900 align:middle line:84% In our course, actually, the boundary of the model, 00:43:52.900 --> 00:43:57.080 align:middle line:84% we don't play with it too much because, again, we 00:43:57.080 --> 00:44:00.380 align:middle line:84% are going to define problems a little bit more clearly. 00:44:00.380 --> 00:44:03.380 align:middle line:84% So as such, this is an architectural choice 00:44:03.380 --> 00:44:07.220 align:middle line:84% we will discuss in the pre-sessions a little bit. 00:44:07.220 --> 00:44:12.080 align:middle line:84% But you wouldn't be too much worried about it in this class. 00:44:12.080 --> 00:44:15.420 align:middle line:84% Time horizon-- very simple, but also still very important. 00:44:15.420 --> 00:44:17.920 align:middle line:84% That is, over what time horizon we are we 00:44:17.920 --> 00:44:19.940 align:middle line:84% care about the dynamics and the outcomes. 00:44:19.940 --> 00:44:23.440 align:middle line:84% And that determines a lot of questions 00:44:23.440 --> 00:44:27.320 align:middle line:84% about what is a legitimate stock variable 00:44:27.320 --> 00:44:31.200 align:middle line:84% or source of inertia, what is the right level of aggregation. 00:44:31.200 --> 00:44:34.360 align:middle line:84% The time horizon typically sharpens 00:44:34.360 --> 00:44:36.860 align:middle line:84% what are the policy questions you care about, 00:44:36.860 --> 00:44:40.440 align:middle line:84% how long is the horizon over which 00:44:40.440 --> 00:44:41.940 align:middle line:90% the outcomes matter for you. 00:44:41.940 --> 00:44:47.720 align:middle line:84% And so you want to think about it clearly early on. 00:44:47.720 --> 00:44:52.040 align:middle line:84% Then the next very important architectural choice 00:44:52.040 --> 00:44:53.300 align:middle line:90% is the aggregation level. 00:44:53.300 --> 00:44:56.360 align:middle line:84% This is how detailed you want to make your model 00:44:56.360 --> 00:45:01.760 align:middle line:90% or how high level it would be. 00:45:01.760 --> 00:45:05.600 align:middle line:84% The heuristic that system dynamicists typically use 00:45:05.600 --> 00:45:11.360 align:middle line:84% is the more aggregate you can stay, the better for you. 00:45:11.360 --> 00:45:14.600 align:middle line:84% Now, that heuristic has good logics behind it. 00:45:14.600 --> 00:45:17.840 align:middle line:84% It helps you with building simple models quickly, 00:45:17.840 --> 00:45:20.920 align:middle line:84% get feedback, work with the stakeholders 00:45:20.920 --> 00:45:26.400 align:middle line:84% without overwhelming them with detailed complexity. 00:45:26.400 --> 00:45:31.920 align:middle line:84% Usually, it gives much more clear structural insights. 00:45:31.920 --> 00:45:36.080 align:middle line:84% The insights are not lost in detailed complexity. 00:45:36.080 --> 00:45:38.320 align:middle line:90% Analysis becomes much simpler. 00:45:38.320 --> 00:45:40.340 align:middle line:84% When you have higher level of aggregation, 00:45:40.340 --> 00:45:43.000 align:middle line:84% you don't need to figure out feedback loops that 00:45:43.000 --> 00:45:45.240 align:middle line:84% go through many thousands of variables. 00:45:45.240 --> 00:45:50.520 align:middle line:84% You can capture the same core ideas with dozens of variables. 00:45:50.520 --> 00:45:53.260 align:middle line:84% It makes analysis in other ways also simpler. 00:45:53.260 --> 00:45:55.440 align:middle line:84% If you're running a calibration and you 00:45:55.440 --> 00:45:58.300 align:middle line:84% need hundreds of millions of simulations, 00:45:58.300 --> 00:46:01.560 align:middle line:84% you can't do it in a model that has millions of variables 00:46:01.560 --> 00:46:02.960 align:middle line:90% in the first place. 00:46:02.960 --> 00:46:05.000 align:middle line:84% But you can easily do it in a model 00:46:05.000 --> 00:46:09.320 align:middle line:84% that has hundreds of variables or dozens of variables. 00:46:09.320 --> 00:46:12.640 align:middle line:84% So aggregation level, there are good arguments 00:46:12.640 --> 00:46:15.880 align:middle line:84% to go in the more aggregate level. 00:46:15.880 --> 00:46:17.600 align:middle line:84% But there are also good arguments 00:46:17.600 --> 00:46:19.300 align:middle line:84% to go into the more detailed level. 00:46:19.300 --> 00:46:23.880 align:middle line:84% And that is not what typical system dynamics practitioners 00:46:23.880 --> 00:46:25.200 align:middle line:90% would recommend. 00:46:25.200 --> 00:46:28.220 align:middle line:84% But in many other fields, people are focused on, 00:46:28.220 --> 00:46:31.640 align:middle line:84% actually, let's go one level deeper in understanding 00:46:31.640 --> 00:46:35.840 align:middle line:84% the microstructures that generate 00:46:35.840 --> 00:46:38.880 align:middle line:90% the dynamics we care about. 00:46:38.880 --> 00:46:42.080 align:middle line:84% Usually, the theoretical knowledge you have 00:46:42.080 --> 00:46:43.400 align:middle line:90% is more detailed. 00:46:43.400 --> 00:46:45.840 align:middle line:84% If you look at the physics of the problem, 00:46:45.840 --> 00:46:49.160 align:middle line:84% you know more about the next level of detail. 00:46:49.160 --> 00:46:52.760 align:middle line:84% And that sometimes means you have good data 00:46:52.760 --> 00:46:54.720 align:middle line:90% at lower level of aggregation. 00:46:54.720 --> 00:46:57.400 align:middle line:84% To utilize that data effectively, 00:46:57.400 --> 00:46:59.140 align:middle line:90% you benefit from disaggregation. 00:46:59.140 --> 00:47:01.080 align:middle line:84% If you aggregate, you are losing a lot 00:47:01.080 --> 00:47:03.680 align:middle line:90% of granularity in your data-- 00:47:03.680 --> 00:47:06.360 align:middle line:84% both quantitative data, but also qualitative data 00:47:06.360 --> 00:47:08.560 align:middle line:90% about how things actually work. 00:47:08.560 --> 00:47:11.820 align:middle line:84% And sometimes, to enforce internal consistency, 00:47:11.820 --> 00:47:14.020 align:middle line:84% you actually need to go one level deeper. 00:47:14.020 --> 00:47:18.880 align:middle line:84% For example, if you skip pricing mechanisms in models of markets 00:47:18.880 --> 00:47:21.880 align:middle line:84% and try to just use supply and demand 00:47:21.880 --> 00:47:26.440 align:middle line:84% to balance each other without an explicit price, things get hard. 00:47:26.440 --> 00:47:27.740 align:middle line:90% It's not impossible. 00:47:27.740 --> 00:47:34.920 align:middle line:84% But internal consistency of that model is harder to maintain. 00:47:34.920 --> 00:47:39.490 align:middle line:84% Sometimes, you benefit from it for having policy levers 00:47:39.490 --> 00:47:42.970 align:middle line:84% at the level of detail that actual decision makers care 00:47:42.970 --> 00:47:43.590 align:middle line:90% about. 00:47:43.590 --> 00:47:45.590 align:middle line:84% So if you aggregate a lot of things, 00:47:45.590 --> 00:47:48.210 align:middle line:84% the policy levers you have at that very aggregated level 00:47:48.210 --> 00:47:51.490 align:middle line:84% are different from what actually people on a day to day basis 00:47:51.490 --> 00:47:52.610 align:middle line:90% deal with. 00:47:52.610 --> 00:47:55.730 align:middle line:84% So I'm telling all of those to highlight the fact 00:47:55.730 --> 00:47:58.270 align:middle line:84% that aggregation level is not an obvious choice. 00:47:58.270 --> 00:48:01.930 align:middle line:84% It actually needs to be a very conscious choice. 00:48:01.930 --> 00:48:04.490 align:middle line:84% I generally-- yeah, go ahead Sepideh. 00:48:04.490 --> 00:48:07.490 align:middle line:84% AUDIENCE: I was wondering if the availability of data 00:48:07.490 --> 00:48:09.442 align:middle line:90% also affected aggregation level. 00:48:09.442 --> 00:48:10.650 align:middle line:90% HAZHIR RAHMANDAD: Absolutely. 00:48:10.650 --> 00:48:13.250 align:middle line:84% So that's one of the reasons-- typically, 00:48:13.250 --> 00:48:17.430 align:middle line:84% if you have very detailed data, especially if it is consistent-- 00:48:17.430 --> 00:48:19.830 align:middle line:84% that is, across different sectors of the model, 00:48:19.830 --> 00:48:23.290 align:middle line:84% you can use the same kind of-- if you have individual level 00:48:23.290 --> 00:48:26.130 align:middle line:84% data panel, for example, if you build 00:48:26.130 --> 00:48:28.030 align:middle line:84% the model at the population level, 00:48:28.030 --> 00:48:29.810 align:middle line:84% you lose a lot of statistical power 00:48:29.810 --> 00:48:33.970 align:middle line:84% because you don't leverage the heterogeneity across agents 00:48:33.970 --> 00:48:36.050 align:middle line:90% in estimating your mechanisms. 00:48:36.050 --> 00:48:38.770 align:middle line:84% It also means you are estimating mechanisms 00:48:38.770 --> 00:48:41.530 align:middle line:84% at a higher level of aggregation which 00:48:41.530 --> 00:48:45.030 align:middle line:84% are not the same as the lower individual level of aggregation, 00:48:45.030 --> 00:48:46.370 align:middle line:90% for example. 00:48:46.370 --> 00:48:48.410 align:middle line:84% So the definition of those effects 00:48:48.410 --> 00:48:51.430 align:middle line:84% are also qualitatively potentially different. 00:48:51.430 --> 00:48:53.170 align:middle line:84% Now, sometimes, they have a close 00:48:53.170 --> 00:48:57.230 align:middle line:84% matching with the lower level of aggregation coefficients. 00:48:57.230 --> 00:48:58.350 align:middle line:90% Sometimes, they don't. 00:48:58.350 --> 00:49:02.250 align:middle line:84% So absolutely, data is one of the reasons. 00:49:02.250 --> 00:49:04.630 align:middle line:90% But there is always a trade-off. 00:49:04.630 --> 00:49:07.650 align:middle line:84% You go into more detail, it becomes harder 00:49:07.650 --> 00:49:10.970 align:middle line:84% to have the broad model boundary that we talked about 00:49:10.970 --> 00:49:13.970 align:middle line:84% as another important architectural choice. 00:49:13.970 --> 00:49:17.510 align:middle line:90% So there is no right answer. 00:49:17.510 --> 00:49:19.010 align:middle line:84% It is really your problem definition 00:49:19.010 --> 00:49:21.510 align:middle line:84% that determines the level of aggregation. 00:49:21.510 --> 00:49:23.690 align:middle line:84% And as I mentioned, the big difference 00:49:23.690 --> 00:49:26.370 align:middle line:84% between very detailed operational models 00:49:26.370 --> 00:49:32.130 align:middle line:84% for operational decision making or even 00:49:32.130 --> 00:49:36.170 align:middle line:84% big one-time decisions is partly this issue, that you do 00:49:36.170 --> 00:49:39.730 align:middle line:84% need the additional detail, both to build confidence 00:49:39.730 --> 00:49:41.450 align:middle line:84% with your clients, but also to know 00:49:41.450 --> 00:49:46.730 align:middle line:84% that you have captured the key mechanisms at the right level. 00:49:46.730 --> 00:49:51.530 align:middle line:84% Once you have a feel for your boundary of the model, 00:49:51.530 --> 00:49:53.450 align:middle line:84% another important architectural choice 00:49:53.450 --> 00:49:54.990 align:middle line:90% is the sources of inertia. 00:49:54.990 --> 00:49:57.370 align:middle line:84% That is, what are the factors that 00:49:57.370 --> 00:50:00.250 align:middle line:84% create inertia in your model, given the time horizon 00:50:00.250 --> 00:50:01.710 align:middle line:90% and the boundary of the model? 00:50:01.710 --> 00:50:04.030 align:middle line:84% So if I have a population sector, 00:50:04.030 --> 00:50:08.290 align:middle line:84% there are obvious choices for thinking about candidate stock 00:50:08.290 --> 00:50:11.230 align:middle line:84% variables that create the sources of inertia and so on. 00:50:11.230 --> 00:50:12.970 align:middle line:84% But thinking about that, again, helps you 00:50:12.970 --> 00:50:15.330 align:middle line:84% with clarifying your problem definition 00:50:15.330 --> 00:50:18.670 align:middle line:84% and why you want to include some stock or not. 00:50:18.670 --> 00:50:19.170 align:middle line:90% OK. 00:50:19.170 --> 00:50:23.010 align:middle line:84% So that's a little bit of two approaches 00:50:23.010 --> 00:50:27.130 align:middle line:84% to tackling the big challenge of problem definitions 00:50:27.130 --> 00:50:28.530 align:middle line:90% that are unclear-- 00:50:28.530 --> 00:50:32.070 align:middle line:84% one starting with clarifying your problem definition. 00:50:32.070 --> 00:50:36.820 align:middle line:84% The other is thinking through your architectural choices. 00:50:36.820 --> 00:50:37.540 align:middle line:90% Alexander? 00:50:37.540 --> 00:50:39.082 align:middle line:84% AUDIENCE: Just kind of stepping back, 00:50:39.082 --> 00:50:41.500 align:middle line:84% could you discuss when it's most appropriate to do 00:50:41.500 --> 00:50:46.980 align:middle line:84% dynamic modeling as opposed to econometrics or optimization? 00:50:46.980 --> 00:50:50.100 align:middle line:84% HAZHIR RAHMANDAD: So typically, when 00:50:50.100 --> 00:50:55.980 align:middle line:84% you're trying to understand how things changed over time 00:50:55.980 --> 00:50:59.100 align:middle line:84% or would change over time, that's 00:50:59.100 --> 00:51:03.340 align:middle line:84% where the feedback mechanisms become critical. 00:51:03.340 --> 00:51:06.180 align:middle line:84% And therefore, you can benefit from the tools 00:51:06.180 --> 00:51:07.740 align:middle line:90% that we deal with. 00:51:07.740 --> 00:51:12.260 align:middle line:84% Now, the examples you gave, like econometrics modeling, 00:51:12.260 --> 00:51:15.540 align:middle line:84% is more the question of empirical grounding of the model 00:51:15.540 --> 00:51:16.600 align:middle line:90% that you're building. 00:51:16.600 --> 00:51:18.580 align:middle line:84% So there are questions about the assumptions 00:51:18.580 --> 00:51:19.800 align:middle line:90% that go into your model. 00:51:19.800 --> 00:51:23.660 align:middle line:84% So you could have models that, for example, are making 00:51:23.660 --> 00:51:25.380 align:middle line:90% the equilibrium assumption-- 00:51:25.380 --> 00:51:27.040 align:middle line:84% a stochastic equilibrium assumption, 00:51:27.040 --> 00:51:30.300 align:middle line:84% but they are still based on the equilibrium assumption. 00:51:30.300 --> 00:51:34.800 align:middle line:84% So they are interested in the long-term regularities 00:51:34.800 --> 00:51:37.980 align:middle line:84% or the forces that create those long-term regularities 00:51:37.980 --> 00:51:40.700 align:middle line:90% inside the system. 00:51:40.700 --> 00:51:44.860 align:middle line:84% And typical dynamic models care about capturing 00:51:44.860 --> 00:51:49.180 align:middle line:84% the transitions between those long-term equilibrium levels 00:51:49.180 --> 00:51:50.020 align:middle line:90% as well. 00:51:50.020 --> 00:51:53.700 align:middle line:84% So they would have the equilibriums built into them. 00:51:53.700 --> 00:51:55.480 align:middle line:84% Otherwise, the model wouldn't work. 00:51:55.480 --> 00:51:58.340 align:middle line:84% But also, you care about the dynamics, the transients, 00:51:58.340 --> 00:52:01.140 align:middle line:84% the challenges of moving from the current position 00:52:01.140 --> 00:52:02.540 align:middle line:90% to another position. 00:52:02.540 --> 00:52:06.700 align:middle line:90% So that's the key difference. 00:52:06.700 --> 00:52:08.180 align:middle line:84% Now, there is another difference. 00:52:08.180 --> 00:52:11.700 align:middle line:84% That is, your assumption about actors in the system 00:52:11.700 --> 00:52:16.240 align:middle line:84% also makes some variants of modeling approaches. 00:52:16.240 --> 00:52:18.820 align:middle line:84% So in system dynamics, we typically 00:52:18.820 --> 00:52:23.100 align:middle line:84% think about actors that are behaviorally realistic-- now, 00:52:23.100 --> 00:52:24.940 align:middle line:90% that is a very vague term. 00:52:24.940 --> 00:52:26.440 align:middle line:90% What exactly does that mean? 00:52:26.440 --> 00:52:29.900 align:middle line:84% You need to define it-- versus actors 00:52:29.900 --> 00:52:33.140 align:middle line:84% that are always rational in optimizing something 00:52:33.140 --> 00:52:34.620 align:middle line:90% very clearly. 00:52:34.620 --> 00:52:38.860 align:middle line:84% Now, it doesn't mean we should exclude rational actors 00:52:38.860 --> 00:52:39.880 align:middle line:90% from our modeling. 00:52:39.880 --> 00:52:42.040 align:middle line:84% In some cases, it is a good assumption. 00:52:42.040 --> 00:52:43.760 align:middle line:84% I'm right now building a model of cities. 00:52:43.760 --> 00:52:47.220 align:middle line:84% And I'm assuming firms are rational in how they decide 00:52:47.220 --> 00:52:51.980 align:middle line:84% to hire people and get capital because that's 00:52:51.980 --> 00:52:57.340 align:middle line:84% a much better approximation for my use case than an ad 00:52:57.340 --> 00:53:00.420 align:middle line:90% hoc behavioral heuristic. 00:53:00.420 --> 00:53:04.460 align:middle line:84% But I'm comfortable with letting some other parts of the model 00:53:04.460 --> 00:53:07.320 align:middle line:84% not be following that rationality assumption. 00:53:07.320 --> 00:53:11.300 align:middle line:84% So that's how you think of actors in your model also 00:53:11.300 --> 00:53:16.460 align:middle line:84% impacts these choices and whether this is the right tool 00:53:16.460 --> 00:53:19.180 align:middle line:90% and right fit for you. 00:53:19.180 --> 00:53:20.360 align:middle line:90% We could talk more about it. 00:53:20.360 --> 00:53:22.580 align:middle line:84% But I think, in practice, the theory 00:53:22.580 --> 00:53:25.280 align:middle line:84% is harder to say without having concrete examples. 00:53:25.280 --> 00:53:27.520 align:middle line:84% So I think, when we go into examples, 00:53:27.520 --> 00:53:28.760 align:middle line:90% things become more clear. 00:53:28.760 --> 00:53:31.500 align:middle line:90% 00:53:31.500 --> 00:53:32.380 align:middle line:90% OK. 00:53:32.380 --> 00:53:36.343 align:middle line:84% So the second big failure mode or challenge-- this one 00:53:36.343 --> 00:53:37.260 align:middle line:90% is not a failure mode. 00:53:37.260 --> 00:53:40.440 align:middle line:84% It's just you start and you don't know where to start. 00:53:40.440 --> 00:53:42.700 align:middle line:90% You just feel overwhelmed. 00:53:42.700 --> 00:53:45.140 align:middle line:84% You're staring at a blank screen and don't 00:53:45.140 --> 00:53:50.220 align:middle line:84% know how to add, what to add, how to go forward. 00:53:50.220 --> 00:53:53.820 align:middle line:84% And the major solution to it is this idea 00:53:53.820 --> 00:53:58.420 align:middle line:84% of iterating and adding structures in small chunks. 00:53:58.420 --> 00:54:00.760 align:middle line:84% So typically, for a novice modeler, 00:54:00.760 --> 00:54:03.680 align:middle line:84% I would say, add a single stock variable. 00:54:03.680 --> 00:54:07.060 align:middle line:84% That is, after considering the sources of inertia 00:54:07.060 --> 00:54:09.580 align:middle line:84% and your architectural choices and problem definition, 00:54:09.580 --> 00:54:10.900 align:middle line:90% pretty important. 00:54:10.900 --> 00:54:12.620 align:middle line:84% Add that stock variable and think 00:54:12.620 --> 00:54:16.140 align:middle line:84% about its inflows and outflows or even if it needs both inflows 00:54:16.140 --> 00:54:16.800 align:middle line:90% and outflows. 00:54:16.800 --> 00:54:18.540 align:middle line:84% Maybe it just needs some adjustment 00:54:18.540 --> 00:54:25.460 align:middle line:84% towards some underlying level, underlying goal. 00:54:25.460 --> 00:54:29.370 align:middle line:84% Put equations for that, analyze that component of the model, 00:54:29.370 --> 00:54:31.570 align:middle line:84% make sure you're understanding what is happening, 00:54:31.570 --> 00:54:33.390 align:middle line:84% whether it is a single stock or maybe just 00:54:33.390 --> 00:54:37.190 align:middle line:84% a couple of stocks that are coupled together. 00:54:37.190 --> 00:54:41.050 align:middle line:84% To analyze, you benefit from a few steps. 00:54:41.050 --> 00:54:44.950 align:middle line:84% For most model chunks, putting it in equilibrium 00:54:44.950 --> 00:54:46.230 align:middle line:90% is a good idea. 00:54:46.230 --> 00:54:49.970 align:middle line:84% Now, there are some exceptions to this general comment. 00:54:49.970 --> 00:54:54.750 align:middle line:84% But putting that piece of model in equilibrium 00:54:54.750 --> 00:55:00.790 align:middle line:84% allows you to understand its behavior much faster, much more 00:55:00.790 --> 00:55:04.470 align:middle line:84% cleanly because you can give it a shock, 00:55:04.470 --> 00:55:09.070 align:middle line:84% step function, ramp function, pulse function and see 00:55:09.070 --> 00:55:11.930 align:middle line:84% how it behaves, whereas, if it is not in equilibrium, 00:55:11.930 --> 00:55:13.970 align:middle line:84% it is already having its own dynamics. 00:55:13.970 --> 00:55:16.390 align:middle line:84% And now, whatever you add to it is confounded 00:55:16.390 --> 00:55:18.090 align:middle line:90% with those transient dynamics. 00:55:18.090 --> 00:55:22.750 align:middle line:84% It is much harder mentally to separate those from each other. 00:55:22.750 --> 00:55:25.310 align:middle line:84% And also, putting chunks of models in equilibrium 00:55:25.310 --> 00:55:28.790 align:middle line:84% will help you put your full model in equilibrium later on 00:55:28.790 --> 00:55:30.250 align:middle line:84% and do the same type of analysis, 00:55:30.250 --> 00:55:33.790 align:middle line:84% so you always have a model that can start from equilibrium. 00:55:33.790 --> 00:55:37.487 align:middle line:84% Not all models can be put in equilibrium analytically. 00:55:37.487 --> 00:55:39.070 align:middle line:84% When you have nonlinear relationships, 00:55:39.070 --> 00:55:40.850 align:middle line:84% sometimes you can solve that analytically. 00:55:40.850 --> 00:55:42.650 align:middle line:84% There are numerical methods for that. 00:55:42.650 --> 00:55:45.030 align:middle line:84% We can talk about it when we get into more 00:55:45.030 --> 00:55:50.030 align:middle line:84% challenging equilibrium questions, how to do that. 00:55:50.030 --> 00:55:53.550 align:middle line:84% Then you want to analyze what is happening in your model. 00:55:53.550 --> 00:55:58.150 align:middle line:84% And analysis here has really three important steps. 00:55:58.150 --> 00:56:00.330 align:middle line:84% So you pick the behavior you're trying to explain. 00:56:00.330 --> 00:56:02.230 align:middle line:90% So I added a chunk. 00:56:02.230 --> 00:56:05.390 align:middle line:84% Maybe it's a stock management chunk. 00:56:05.390 --> 00:56:07.630 align:middle line:90% I put it in equilibrium. 00:56:07.630 --> 00:56:08.870 align:middle line:90% Now, I give it a shock. 00:56:08.870 --> 00:56:11.210 align:middle line:90% I add a step function into it. 00:56:11.210 --> 00:56:13.430 align:middle line:84% Or I add a ramp function into it, 00:56:13.430 --> 00:56:17.870 align:middle line:84% into the demand that is coming into the system, for example. 00:56:17.870 --> 00:56:21.210 align:middle line:84% And I see that the model behaves in some way. 00:56:21.210 --> 00:56:25.050 align:middle line:84% Maybe it oscillates, maybe it grows, whatever is happening. 00:56:25.050 --> 00:56:27.890 align:middle line:84% You pick that behavior and come up with a hypothesis. 00:56:27.890 --> 00:56:29.830 align:middle line:84% Why do you think it is happening? 00:56:29.830 --> 00:56:33.170 align:middle line:84% Now, coming up with a hypothesis is not trivial. 00:56:33.170 --> 00:56:35.770 align:middle line:84% If you just look at your model and try to say, 00:56:35.770 --> 00:56:40.430 align:middle line:84% oh, I think it is because x, y, z, 99% you're wrong. 00:56:40.430 --> 00:56:43.590 align:middle line:90% So don't even go down that road. 00:56:43.590 --> 00:56:45.130 align:middle line:90% Start with causal tracing. 00:56:45.130 --> 00:56:47.450 align:middle line:84% Now, there are other ways to come up with hypotheses. 00:56:47.450 --> 00:56:53.450 align:middle line:84% But I think the good old causal tracing is pretty good. 00:56:53.450 --> 00:56:54.210 align:middle line:90% It works well. 00:56:54.210 --> 00:56:55.670 align:middle line:90% So what is causal tracing? 00:56:55.670 --> 00:56:57.210 align:middle line:90% You're looking at this variable. 00:56:57.210 --> 00:56:58.870 align:middle line:90% You double click on it. 00:56:58.870 --> 00:57:01.270 align:middle line:84% You look at the things that are driving that variable. 00:57:01.270 --> 00:57:02.750 align:middle line:84% In Vensim, in other software, there 00:57:02.750 --> 00:57:07.150 align:middle line:84% is a graph option that shows all of the drivers of that variable. 00:57:07.150 --> 00:57:10.670 align:middle line:84% So you go backwards and say, OK, among these three variables that 00:57:10.670 --> 00:57:13.350 align:middle line:84% are driving my core variable, which one 00:57:13.350 --> 00:57:17.030 align:middle line:84% is more clearly showing that same mode of behavior 00:57:17.030 --> 00:57:19.430 align:middle line:84% or driving that mode of behavior? 00:57:19.430 --> 00:57:21.410 align:middle line:84% Usually, one of them is a good candidate. 00:57:21.410 --> 00:57:24.350 align:middle line:84% And you can make that judgment call very easily 00:57:24.350 --> 00:57:26.390 align:middle line:90% mentally by just looking at it. 00:57:26.390 --> 00:57:29.110 align:middle line:84% Then pick that variable and go backwards and pick that variable 00:57:29.110 --> 00:57:30.430 align:middle line:90% and go backwards. 00:57:30.430 --> 00:57:31.750 align:middle line:90% You do this. 00:57:31.750 --> 00:57:36.270 align:middle line:84% Usually, there are two outcomes out of this tracing process. 00:57:36.270 --> 00:57:39.930 align:middle line:84% Sometimes, you land on a feedback loop. 00:57:39.930 --> 00:57:42.470 align:middle line:84% That is, at some point, you go back to one of the variables 00:57:42.470 --> 00:57:44.670 align:middle line:90% that you already have traced. 00:57:44.670 --> 00:57:47.750 align:middle line:84% So then that creates a feedback loop 00:57:47.750 --> 00:57:49.990 align:middle line:84% that is a good candidate for explaining 00:57:49.990 --> 00:57:52.150 align:middle line:90% that mode of behavior. 00:57:52.150 --> 00:57:55.050 align:middle line:84% Sometimes, it doesn't lead to a feedback loop. 00:57:55.050 --> 00:57:58.830 align:middle line:84% It just leads to an exogenous input coming into your model. 00:57:58.830 --> 00:58:01.550 align:middle line:84% It basically is telling you that the main driver 00:58:01.550 --> 00:58:04.250 align:middle line:84% of that mode of behavior is some exogenous thing. 00:58:04.250 --> 00:58:07.510 align:middle line:84% It might be the shock that you gave 00:58:07.510 --> 00:58:11.670 align:middle line:84% into exogenous demand that is primarily driving the behavior. 00:58:11.670 --> 00:58:17.750 align:middle line:84% So it is not boring, but not that groundbreaking as well. 00:58:17.750 --> 00:58:19.450 align:middle line:90% So that becomes a hypothesis. 00:58:19.450 --> 00:58:21.180 align:middle line:90% So this is the critical step. 00:58:21.180 --> 00:58:23.040 align:middle line:90% People go up to this step. 00:58:23.040 --> 00:58:25.100 align:middle line:84% And then they say, oh OK, I figured it out. 00:58:25.100 --> 00:58:27.000 align:middle line:84% And then they go to the next step. 00:58:27.000 --> 00:58:29.560 align:middle line:84% And they already have lost the game. 00:58:29.560 --> 00:58:33.240 align:middle line:84% So at this step, you need to test that hypothesis. 00:58:33.240 --> 00:58:35.280 align:middle line:84% What is testing of the hypothesis? 00:58:35.280 --> 00:58:37.440 align:middle line:84% So if I have come up with a feedback loop 00:58:37.440 --> 00:58:39.520 align:middle line:84% that I think explains that behavior, 00:58:39.520 --> 00:58:43.960 align:middle line:84% it means I need to go and disrupt that feedback 00:58:43.960 --> 00:58:47.780 align:middle line:84% loop without disrupting anything else in the system, 00:58:47.780 --> 00:58:50.720 align:middle line:84% without disrupting other important feedback loops, 00:58:50.720 --> 00:58:55.360 align:middle line:84% and see whether the behavior that I started with changes. 00:58:55.360 --> 00:58:58.080 align:middle line:84% That is at the heart of testing any hypothesis 00:58:58.080 --> 00:58:59.480 align:middle line:90% in the scientific method. 00:58:59.480 --> 00:59:02.240 align:middle line:84% You remove the causal factor and see 00:59:02.240 --> 00:59:04.920 align:middle line:90% whether the behavior changes. 00:59:04.920 --> 00:59:09.660 align:middle line:84% If it does, it is good evidence in support of your hypothesis. 00:59:09.660 --> 00:59:13.540 align:middle line:90% Again, you can't be 100% sure. 00:59:13.540 --> 00:59:16.140 align:middle line:84% But it is good enough evidence that, at this point, 00:59:16.140 --> 00:59:18.800 align:middle line:84% for a pragmatic modeler, you can say, OK, 00:59:18.800 --> 00:59:21.520 align:middle line:84% let's assume this is a good explanation. 00:59:21.520 --> 00:59:23.940 align:middle line:84% If not, then you need another hypothesis. 00:59:23.940 --> 00:59:25.780 align:middle line:84% You need to go back to your causal tracing, 00:59:25.780 --> 00:59:29.040 align:middle line:84% see where the logic didn't work, come up-- 00:59:29.040 --> 00:59:31.360 align:middle line:84% there is perhaps another pathway that you missed 00:59:31.360 --> 00:59:34.040 align:middle line:90% and go through that. 00:59:34.040 --> 00:59:36.840 align:middle line:84% So once you have done this for a couple of reference modes 00:59:36.840 --> 00:59:39.020 align:middle line:84% in the model, in the dynamics of the model, 00:59:39.020 --> 00:59:42.040 align:middle line:84% you should be able to feel comfortable that, OK, now, I 00:59:42.040 --> 00:59:44.480 align:middle line:90% understand my model. 00:59:44.480 --> 00:59:45.220 align:middle line:90% That's great. 00:59:45.220 --> 00:59:47.160 align:middle line:84% That's what you want to be, somebody 00:59:47.160 --> 00:59:48.760 align:middle line:90% who understands their model. 00:59:48.760 --> 00:59:50.960 align:middle line:84% But you, again, need to test that. 00:59:50.960 --> 00:59:56.160 align:middle line:84% So testing that step is just change the model 00:59:56.160 --> 00:59:58.860 align:middle line:84% parameters, all model parameters. 00:59:58.860 --> 01:00:00.360 align:middle line:84% In Vensim, you have a single button. 01:00:00.360 --> 01:00:01.100 align:middle line:90% You click. 01:00:01.100 --> 01:00:03.680 align:middle line:84% It changes all of the model parameters by 10%. 01:00:03.680 --> 01:00:09.707 align:middle line:84% And it reports to you the behavior of various outcomes 01:00:09.707 --> 01:00:10.540 align:middle line:90% that you care about. 01:00:10.540 --> 01:00:12.920 align:middle line:84% You can define what is the outcome you care about. 01:00:12.920 --> 01:00:16.400 align:middle line:84% Make predictions what will happen to those outcomes 01:00:16.400 --> 01:00:19.840 align:middle line:84% before doing this test, and then see whether your predictions are 01:00:19.840 --> 01:00:20.620 align:middle line:90% correct. 01:00:20.620 --> 01:00:23.040 align:middle line:84% That is, if I increase parameter x, 01:00:23.040 --> 01:00:27.200 align:middle line:84% whether my profits go up or come down. 01:00:27.200 --> 01:00:29.560 align:middle line:84% If your predictions are all correct, 01:00:29.560 --> 01:00:32.520 align:middle line:84% that means you have already built the intuition you want. 01:00:32.520 --> 01:00:34.920 align:middle line:84% If they are not largely correct, then it 01:00:34.920 --> 01:00:37.800 align:middle line:84% means you haven't done your homework completely yet. 01:00:37.800 --> 01:00:40.400 align:middle line:84% So that's the process of analysis. 01:00:40.400 --> 01:00:44.440 align:middle line:90% It may sound a lot. 01:00:44.440 --> 01:00:46.440 align:middle line:90% It is not trivial. 01:00:46.440 --> 01:00:49.920 align:middle line:84% That's why, in the diagram I was showing you, about half 01:00:49.920 --> 01:00:53.640 align:middle line:84% of the time should go into analysis. 01:00:53.640 --> 01:00:55.980 align:middle line:84% The modeling part of putting equations is important. 01:00:55.980 --> 01:00:58.720 align:middle line:84% But if you don't do the analysis, 01:00:58.720 --> 01:01:01.720 align:middle line:84% your modeling will become slower and slower in every iteration 01:01:01.720 --> 01:01:04.640 align:middle line:84% because now you are seeing all kinds of weird behaviors 01:01:04.640 --> 01:01:06.040 align:middle line:90% you don't understand. 01:01:06.040 --> 01:01:10.320 align:middle line:84% And then you end up with other failure modes we'll talk about. 01:01:10.320 --> 01:01:13.340 align:middle line:84% Any questions on this analysis process? 01:01:13.340 --> 01:01:14.680 align:middle line:90% This is very important. 01:01:14.680 --> 01:01:15.820 align:middle line:90% We will practice it. 01:01:15.820 --> 01:01:17.420 align:middle line:90% You will practice it every week. 01:01:17.420 --> 01:01:21.100 align:middle line:84% But this is a little bit of a high level description. 01:01:21.100 --> 01:01:24.520 align:middle line:84% We could go into details of it for specific applications. 01:01:24.520 --> 01:01:26.537 align:middle line:84% And folks online, please unmute yourself 01:01:26.537 --> 01:01:27.620 align:middle line:90% if you have any questions. 01:01:27.620 --> 01:01:32.780 align:middle line:84% If you want to join in and say anything, don't hesitate. 01:01:32.780 --> 01:01:37.120 align:middle line:90% 01:01:37.120 --> 01:01:43.240 align:middle line:84% OK, so let me now bring the next three pieces together and just 01:01:43.240 --> 01:01:46.520 align:middle line:84% talk about those challenges and some 01:01:46.520 --> 01:01:48.840 align:middle line:84% of the ideas for tackling them all together 01:01:48.840 --> 01:01:50.620 align:middle line:84% because they are closely related. 01:01:50.620 --> 01:01:56.400 align:middle line:84% So aggregation choices, we talked about them. 01:01:56.400 --> 01:02:00.800 align:middle line:84% Bad aggregation choices mess up your modeling project 01:02:00.800 --> 01:02:02.260 align:middle line:90% quite significantly. 01:02:02.260 --> 01:02:03.780 align:middle line:90% And they are hard to reverse. 01:02:03.780 --> 01:02:06.440 align:middle line:84% Once you have decided that I'm building a very aggregated 01:02:06.440 --> 01:02:10.980 align:middle line:84% model, going to a detailed individual level model, 01:02:10.980 --> 01:02:13.990 align:middle line:84% for example, will be basically building a new model. 01:02:13.990 --> 01:02:14.910 align:middle line:90% It's a new project. 01:02:14.910 --> 01:02:18.910 align:middle line:84% And so people have sunk cost fallacies. 01:02:18.910 --> 01:02:23.170 align:middle line:84% They wouldn't make that choice of changing architecture easily. 01:02:23.170 --> 01:02:26.090 align:middle line:84% So you may spend months on bad architecture. 01:02:26.090 --> 01:02:28.710 align:middle line:84% And then you just don't get what you want out of it. 01:02:28.710 --> 01:02:31.570 align:middle line:90% 01:02:31.570 --> 01:02:35.250 align:middle line:84% A very common problem I see, at the level of formulation 01:02:35.250 --> 01:02:38.930 align:middle line:84% and the level of putting variables on the diagram, 01:02:38.930 --> 01:02:43.970 align:middle line:84% is unclear constructs, folks coming up with variables 01:02:43.970 --> 01:02:47.530 align:middle line:84% that they are not quite clear what exactly the definition is 01:02:47.530 --> 01:02:48.690 align:middle line:90% for themselves. 01:02:48.690 --> 01:02:50.890 align:middle line:84% So that's another common challenge. 01:02:50.890 --> 01:02:52.930 align:middle line:84% And then bad stock selection is also 01:02:52.930 --> 01:02:54.710 align:middle line:90% very common, very challenging. 01:02:54.710 --> 01:02:57.870 align:middle line:84% If you don't have the best stock variables, 01:02:57.870 --> 01:03:01.390 align:middle line:84% everything else falls apart in your modeling work. 01:03:01.390 --> 01:03:05.250 align:middle line:84% So how can you tackle these challenges? 01:03:05.250 --> 01:03:09.810 align:middle line:84% So on the aggregation one, unfortunately, there 01:03:09.810 --> 01:03:11.950 align:middle line:90% are not many general rules. 01:03:11.950 --> 01:03:14.130 align:middle line:90% It is really problem specific. 01:03:14.130 --> 01:03:16.930 align:middle line:84% As I said, because of all the pros and cons of both 01:03:16.930 --> 01:03:19.210 align:middle line:84% aggregating and disaggregating, I'm 01:03:19.210 --> 01:03:23.930 align:middle line:84% not able to give you any general rules other than really 01:03:23.930 --> 01:03:25.630 align:middle line:90% think about it carefully. 01:03:25.630 --> 01:03:29.610 align:middle line:84% So one of the very common failure modes that I have seen 01:03:29.610 --> 01:03:33.290 align:middle line:84% is system dynamicists, classical ones, 01:03:33.290 --> 01:03:37.250 align:middle line:84% would just pick a very aggregate level of analysis 01:03:37.250 --> 01:03:42.010 align:middle line:84% and don't consider alternatives and close the doors 01:03:42.010 --> 01:03:44.690 align:middle line:84% to a lot of benefits they may get from disaggregation 01:03:44.690 --> 01:03:46.810 align:middle line:90% for specific applications. 01:03:46.810 --> 01:03:50.530 align:middle line:84% And then that forms their whole modeling journeys. 01:03:50.530 --> 01:03:54.570 align:middle line:84% And then you hear folks that are, for example, used to agent 01:03:54.570 --> 01:03:55.650 align:middle line:90% based modeling. 01:03:55.650 --> 01:03:58.170 align:middle line:84% And they see everything as, OK, I 01:03:58.170 --> 01:04:02.930 align:middle line:84% need to go model at the level of single individuals. 01:04:02.930 --> 01:04:06.370 align:middle line:84% So they miss opportunities to simplify their models 01:04:06.370 --> 01:04:09.070 align:middle line:84% significantly and benefit from aggregation. 01:04:09.070 --> 01:04:11.570 align:middle line:84% So the general failure mode you can avoid 01:04:11.570 --> 01:04:15.450 align:middle line:84% is to just think about this choice very explicitly. 01:04:15.450 --> 01:04:16.510 align:middle line:90% It's OK to be wrong. 01:04:16.510 --> 01:04:18.910 align:middle line:84% It's OK to be challenged with how to do it. 01:04:18.910 --> 01:04:20.690 align:middle line:84% But think about it at the very beginning 01:04:20.690 --> 01:04:23.530 align:middle line:90% of your modeling project. 01:04:23.530 --> 01:04:26.850 align:middle line:84% Then stay consistent with your level of aggregation 01:04:26.850 --> 01:04:28.933 align:middle line:84% across different components of the models. 01:04:28.933 --> 01:04:30.350 align:middle line:84% As we go through these iterations, 01:04:30.350 --> 01:04:32.130 align:middle line:90% you add chunks to your models. 01:04:32.130 --> 01:04:34.990 align:middle line:84% If some chunks are at a very detailed level of aggregation 01:04:34.990 --> 01:04:37.250 align:middle line:84% and some chunks are very aggregate, 01:04:37.250 --> 01:04:41.470 align:middle line:84% things are probably going to be problematic for you. 01:04:41.470 --> 01:04:44.490 align:middle line:84% It's just coupling between these different levels of aggregation 01:04:44.490 --> 01:04:46.170 align:middle line:90% becomes hard. 01:04:46.170 --> 01:04:48.310 align:middle line:84% You also don't benefit from the details. 01:04:48.310 --> 01:04:50.750 align:middle line:84% If you have part of a model that is very aggregate 01:04:50.750 --> 01:04:54.210 align:middle line:84% and, therefore, you think your level of confidence 01:04:54.210 --> 01:04:58.570 align:middle line:84% is whatever simulated plus and minus 50%, 01:04:58.570 --> 01:05:02.490 align:middle line:84% what's the point of optimizing another part of your model 01:05:02.490 --> 01:05:07.250 align:middle line:84% at the level of 1% variance if it interacts with a part that 01:05:07.250 --> 01:05:08.430 align:middle line:90% has so much uncertainty? 01:05:08.430 --> 01:05:11.430 align:middle line:84% It's just wasted modeling effort. 01:05:11.430 --> 01:05:13.770 align:middle line:84% So keeping things at the same level of aggregation 01:05:13.770 --> 01:05:17.550 align:middle line:84% avoids those types of wasted extra details. 01:05:17.550 --> 01:05:19.390 align:middle line:84% And this is also a very common problem. 01:05:19.390 --> 01:05:26.530 align:middle line:84% So I do a lot of epidemiological modeling and read through these. 01:05:26.530 --> 01:05:28.390 align:middle line:84% So you see models that are so detailed 01:05:28.390 --> 01:05:30.570 align:middle line:84% they model every individual in the country 01:05:30.570 --> 01:05:33.410 align:middle line:84% and how they move around and so on and so forth. 01:05:33.410 --> 01:05:36.650 align:middle line:84% So that's very detailed level of aggregation. 01:05:36.650 --> 01:05:40.170 align:middle line:84% And then they are missing huge feedback mechanisms 01:05:40.170 --> 01:05:42.770 align:middle line:84% about, for example, risk perceptions' impact 01:05:42.770 --> 01:05:44.070 align:middle line:90% on people's behavior. 01:05:44.070 --> 01:05:45.690 align:middle line:84% These are things that change behavior 01:05:45.690 --> 01:05:48.970 align:middle line:84% at the level of an order of magnitude. 01:05:48.970 --> 01:05:51.690 align:middle line:84% So if you're modeling the walking of people 01:05:51.690 --> 01:05:53.770 align:middle line:84% in the city, but missing a factor that 01:05:53.770 --> 01:05:56.610 align:middle line:84% is order of magnitude changing the dynamics, 01:05:56.610 --> 01:05:58.270 align:middle line:90% this is an unbalanced model. 01:05:58.270 --> 01:06:00.710 align:middle line:84% You're going to waste a lot of energy. 01:06:00.710 --> 01:06:03.170 align:middle line:84% You may even get it published in Science and Nature 01:06:03.170 --> 01:06:05.910 align:middle line:84% because people get impressed by a detailed model-- 01:06:05.910 --> 01:06:07.400 align:middle line:84% many, many people, not everybody. 01:06:07.400 --> 01:06:13.620 align:middle line:84% But the actual insights that come out of that model 01:06:13.620 --> 01:06:16.460 align:middle line:84% are probably useless because you have 01:06:16.460 --> 01:06:19.540 align:middle line:84% spent so much time on things that ultimately are, 01:06:19.540 --> 01:06:21.480 align:middle line:84% on the margin, making no difference 01:06:21.480 --> 01:06:25.080 align:middle line:84% and have lost track of things that make a huge difference. 01:06:25.080 --> 01:06:29.460 align:middle line:84% So this is a very common type of a problem, largely 01:06:29.460 --> 01:06:34.580 align:middle line:84% because people haven't done that careful trade-off 01:06:34.580 --> 01:06:37.220 align:middle line:84% analysis-- what is in, what is out, what level of detail 01:06:37.220 --> 01:06:40.300 align:middle line:90% I need to have. 01:06:40.300 --> 01:06:42.380 align:middle line:84% Usually, another thing that helps 01:06:42.380 --> 01:06:46.500 align:middle line:84% is to stay close to an actual case. 01:06:46.500 --> 01:06:48.960 align:middle line:84% That will help you with selecting the right variables. 01:06:48.960 --> 01:06:50.540 align:middle line:84% So this is more about the challenge 01:06:50.540 --> 01:06:54.060 align:middle line:84% of unclear and overlapping constructs. 01:06:54.060 --> 01:06:58.760 align:middle line:84% So if you have a case in mind, it will sharpen your thinking. 01:06:58.760 --> 01:07:00.500 align:middle line:84% It will allow you to say, OK, what 01:07:00.500 --> 01:07:06.860 align:middle line:84% I mean is, for example, how many other people you go and interact 01:07:06.860 --> 01:07:11.180 align:middle line:84% with in the community in my model of COVID-19 01:07:11.180 --> 01:07:12.480 align:middle line:90% on a daily basis. 01:07:12.480 --> 01:07:16.900 align:middle line:84% So by thinking about COVID-19, you can define your contact rate 01:07:16.900 --> 01:07:17.560 align:middle line:90% more clearly. 01:07:17.560 --> 01:07:19.840 align:middle line:84% You can say, OK, what do I mean by interact? 01:07:19.840 --> 01:07:21.080 align:middle line:90% I don't mean sending email. 01:07:21.080 --> 01:07:25.840 align:middle line:84% I mean physical contacts that are a few feet apart or less 01:07:25.840 --> 01:07:26.360 align:middle line:90% and so on. 01:07:26.360 --> 01:07:29.180 align:middle line:84% So you make the constructs much more clear 01:07:29.180 --> 01:07:31.140 align:middle line:90% when you think about a generic-- 01:07:31.140 --> 01:07:32.340 align:middle line:90% a specific case. 01:07:32.340 --> 01:07:34.200 align:middle line:84% Even if your model is going to be generic, 01:07:34.200 --> 01:07:38.700 align:middle line:84% having a specific case in mind really helps. 01:07:38.700 --> 01:07:42.480 align:middle line:84% Another thing is not only define your variables clearly-- 01:07:42.480 --> 01:07:44.160 align:middle line:84% that is, have clear definitions-- 01:07:44.160 --> 01:07:46.740 align:middle line:90% but avoid conceptual overlaps. 01:07:46.740 --> 01:07:52.180 align:middle line:84% So if you have something about morale and something 01:07:52.180 --> 01:07:55.700 align:middle line:84% about job satisfaction and another variable related 01:07:55.700 --> 01:07:59.140 align:middle line:84% to motivation, you start to see these constructs that 01:07:59.140 --> 01:08:01.620 align:middle line:90% are overlapping conceptually. 01:08:01.620 --> 01:08:04.460 align:middle line:84% And they will mess up your model because you 01:08:04.460 --> 01:08:07.700 align:middle line:84% will start tracking the same underlying idea 01:08:07.700 --> 01:08:09.820 align:middle line:90% through different variables. 01:08:09.820 --> 01:08:13.040 align:middle line:84% You create these loops that are not exactly the same, 01:08:13.040 --> 01:08:14.900 align:middle line:90% but largely the same. 01:08:14.900 --> 01:08:17.819 align:middle line:84% And nobody can really figure out what 01:08:17.819 --> 01:08:21.720 align:middle line:84% is the clear way of modeling and bringing them together. 01:08:21.720 --> 01:08:26.140 align:middle line:84% So avoiding overlapping concepts is very important. 01:08:26.140 --> 01:08:29.859 align:middle line:84% Then on the issues related to sources 01:08:29.859 --> 01:08:34.600 align:middle line:84% of inertia, stock selection, start with sources of inertia. 01:08:34.600 --> 01:08:36.270 align:middle line:84% And there are different ways to do that. 01:08:36.270 --> 01:08:38.020 align:middle line:84% We have talked about, in the basic course, 01:08:38.020 --> 01:08:40.040 align:middle line:84% about the snapshot test or freeze test. 01:08:40.040 --> 01:08:42.040 align:middle line:84% You just freeze the system or take a snapshot. 01:08:42.040 --> 01:08:43.120 align:middle line:90% What is observable? 01:08:43.120 --> 01:08:44.439 align:middle line:90% What is visible? 01:08:44.439 --> 01:08:45.920 align:middle line:90% Those are candidate stocks. 01:08:45.920 --> 01:08:48.020 align:middle line:84% But they are not always the right stock variables 01:08:48.020 --> 01:08:49.140 align:middle line:90% for your problem. 01:08:49.140 --> 01:08:53.020 align:middle line:84% Another source of insight into sources of inertia 01:08:53.020 --> 01:08:56.859 align:middle line:84% is perception processes, where people are learning 01:08:56.859 --> 01:09:00.120 align:middle line:90% about some information. 01:09:00.120 --> 01:09:02.399 align:middle line:84% It takes time to adjust their perceptions, 01:09:02.399 --> 01:09:05.840 align:middle line:84% adjust their assessments, adjust their measurements. 01:09:05.840 --> 01:09:08.420 align:middle line:90% That's another source of stocks. 01:09:08.420 --> 01:09:11.660 align:middle line:84% Any physical accumulation is typically 01:09:11.660 --> 01:09:15.279 align:middle line:84% a candidate, not necessarily the best one, but a candidate. 01:09:15.279 --> 01:09:17.960 align:middle line:84% Other settings give you common options. 01:09:17.960 --> 01:09:20.899 align:middle line:84% If you're modeling firms, you typically have people 01:09:20.899 --> 01:09:22.720 align:middle line:90% as employees or as customers. 01:09:22.720 --> 01:09:24.380 align:middle line:84% You have resources, whether it is 01:09:24.380 --> 01:09:27.340 align:middle line:90% in terms of machinery or money. 01:09:27.340 --> 01:09:29.660 align:middle line:84% And they give you stock and flow chains 01:09:29.660 --> 01:09:32.649 align:middle line:84% for how those resources are going through 01:09:32.649 --> 01:09:34.899 align:middle line:84% or products, how they are flowing through their system 01:09:34.899 --> 01:09:36.720 align:middle line:90% or tasks and projects. 01:09:36.720 --> 01:09:38.700 align:middle line:84% So you have good candidates to start 01:09:38.700 --> 01:09:42.620 align:middle line:84% with from your general knowledge of the space. 01:09:42.620 --> 01:09:45.580 align:middle line:84% So those are, again, good candidates. 01:09:45.580 --> 01:09:48.180 align:middle line:84% You generate a bunch of candidate stock variables 01:09:48.180 --> 01:09:51.580 align:middle line:84% or sources of inertia, and then prioritize them. 01:09:51.580 --> 01:09:53.760 align:middle line:84% And then once you have a source of inertia, 01:09:53.760 --> 01:09:55.880 align:middle line:84% you prioritize it to be high level. 01:09:55.880 --> 01:10:00.150 align:middle line:84% For example, I have people in my model of the world. 01:10:00.150 --> 01:10:03.570 align:middle line:84% Then you need to define what are the good stock variables. 01:10:03.570 --> 01:10:06.970 align:middle line:84% And at that level, some of it is about the aggregation level. 01:10:06.970 --> 01:10:09.330 align:middle line:84% So you may disaggregate or not disaggregate. 01:10:09.330 --> 01:10:12.030 align:middle line:84% You need to make some conscious choices there. 01:10:12.030 --> 01:10:15.110 align:middle line:84% But basically, pick the stock variables 01:10:15.110 --> 01:10:18.830 align:middle line:84% that are at the heart of those sources of inertia to the extent 01:10:18.830 --> 01:10:21.150 align:middle line:90% that you can. 01:10:21.150 --> 01:10:26.030 align:middle line:84% And then there is the MECE principle, mutually exclusive, 01:10:26.030 --> 01:10:29.270 align:middle line:84% collectively exhaustive principle, 01:10:29.270 --> 01:10:32.810 align:middle line:84% which can inform stock selection as well. 01:10:32.810 --> 01:10:36.630 align:middle line:84% So you want to stock variables that are mutually exclusive. 01:10:36.630 --> 01:10:39.710 align:middle line:84% Two stocks should not be able to hold 01:10:39.710 --> 01:10:42.050 align:middle line:84% the same person or the same widget 01:10:42.050 --> 01:10:44.350 align:middle line:84% or the same dollar at the same time. 01:10:44.350 --> 01:10:46.430 align:middle line:84% They should be mutually exclusive. 01:10:46.430 --> 01:10:48.290 align:middle line:84% But also, collectively exhaustive-- 01:10:48.290 --> 01:10:50.470 align:middle line:84% that is, they should cover the space that you're 01:10:50.470 --> 01:10:51.410 align:middle line:90% trying to capture. 01:10:51.410 --> 01:10:54.550 align:middle line:84% If it is people in your company, whether they 01:10:54.550 --> 01:10:58.110 align:middle line:84% are rookies versus experienced, that's one way to cut them. 01:10:58.110 --> 01:11:01.790 align:middle line:84% Whether they are based on their tenure, that's another way. 01:11:01.790 --> 01:11:05.410 align:middle line:84% Or based on their gender is another way. 01:11:05.410 --> 01:11:08.450 align:middle line:84% So what is relevant to your problem? 01:11:08.450 --> 01:11:10.530 align:middle line:90% But have that MECE principal. 01:11:10.530 --> 01:11:13.910 align:middle line:84% And in fact, the MECE principal, not as strictly, 01:11:13.910 --> 01:11:17.590 align:middle line:84% but qualitatively, is useful for thinking about other concepts 01:11:17.590 --> 01:11:18.090 align:middle line:90% as well. 01:11:18.090 --> 01:11:20.190 align:middle line:90% That is, when you have-- 01:11:20.190 --> 01:11:21.490 align:middle line:90% it avoids the overlap. 01:11:21.490 --> 01:11:27.270 align:middle line:84% So the MECE part-- the ME part is important for that. 01:11:27.270 --> 01:11:30.630 align:middle line:84% AUDIENCE: I have a question on the level of aggregation. 01:11:30.630 --> 01:11:33.830 align:middle line:84% I think this came up almost a year ago in the same room in one 01:11:33.830 --> 01:11:35.250 align:middle line:90% of the lunch seminars. 01:11:35.250 --> 01:11:38.552 align:middle line:84% And the idea was, if we have a model, trying 01:11:38.552 --> 01:11:41.010 align:middle line:84% to build a model in one level of aggregation-- for example, 01:11:41.010 --> 01:11:45.510 align:middle line:84% in population-- and we're having issues with the formulation 01:11:45.510 --> 01:11:47.710 align:middle line:84% or with the decision of going with the flow 01:11:47.710 --> 01:11:51.270 align:middle line:84% or stratification between two different levels, for example, 01:11:51.270 --> 01:11:56.790 align:middle line:84% is it ever reasonable to build a toy model on an individual level 01:11:56.790 --> 01:11:59.590 align:middle line:84% aggregation, for example, just to resolve 01:11:59.590 --> 01:12:02.130 align:middle line:84% that issue without formulation, and then come back to that? 01:12:02.130 --> 01:12:03.588 align:middle line:84% HAZHIR RAHMANDAD: Yeah, absolutely. 01:12:03.588 --> 01:12:06.710 align:middle line:84% So that's quite a good idea, in some cases, when 01:12:06.710 --> 01:12:10.710 align:middle line:84% you don't know whether your aggregate approximation is 01:12:10.710 --> 01:12:13.190 align:middle line:84% a good approximation for the more 01:12:13.190 --> 01:12:14.570 align:middle line:90% detailed level of aggregation. 01:12:14.570 --> 01:12:18.510 align:middle line:84% So building a chunk toy model, making that comparison, 01:12:18.510 --> 01:12:21.590 align:middle line:84% and making sure you have a good approximation 01:12:21.590 --> 01:12:24.690 align:middle line:84% will allow you to stay at your higher level of aggregation, 01:12:24.690 --> 01:12:30.150 align:middle line:84% but being also cognizant of what the lower level dynamics entail. 01:12:30.150 --> 01:12:32.270 align:middle line:84% There are also related ways that you 01:12:32.270 --> 01:12:35.150 align:middle line:84% can build more aggregate models by thinking 01:12:35.150 --> 01:12:38.410 align:middle line:84% about the individual level dynamics very explicitly. 01:12:38.410 --> 01:12:42.270 align:middle line:84% So we have a paper with Saeideh Fallah-Fini and colleagues, 01:12:42.270 --> 01:12:45.450 align:middle line:84% 2013, that formalizes that process. 01:12:45.450 --> 01:12:49.990 align:middle line:84% For example, if you want to look at weight dynamics 01:12:49.990 --> 01:12:54.710 align:middle line:84% in a population, you can use a model of individual weight 01:12:54.710 --> 01:12:58.270 align:middle line:84% dynamics, but then use that to derive 01:12:58.270 --> 01:13:01.750 align:middle line:84% the equations for the aggregate population level model 01:13:01.750 --> 01:13:02.650 align:middle line:90% explicitly. 01:13:02.650 --> 01:13:06.170 align:middle line:84% And so there are ways to make these even more formal, 01:13:06.170 --> 01:13:08.290 align:middle line:84% the connections between the level of aggregations, 01:13:08.290 --> 01:13:12.070 align:middle line:90% but not for every problem. 01:13:12.070 --> 01:13:15.190 align:middle line:90% Any other questions? 01:13:15.190 --> 01:13:19.190 align:middle line:84% OK, last-- not last, but one before last 01:13:19.190 --> 01:13:23.590 align:middle line:84% is basically the problem with bad formulations. 01:13:23.590 --> 01:13:25.470 align:middle line:90% It's a very generic term. 01:13:25.470 --> 01:13:28.570 align:middle line:84% And therefore, there is no single answer. 01:13:28.570 --> 01:13:30.510 align:middle line:84% But there are some ideas that help you 01:13:30.510 --> 01:13:32.910 align:middle line:90% with building good formulations. 01:13:32.910 --> 01:13:36.790 align:middle line:84% Probably the most important one is do not innovate. 01:13:36.790 --> 01:13:39.210 align:middle line:84% There is a whole host of formulations out there. 01:13:39.210 --> 01:13:40.990 align:middle line:90% Use them. 01:13:40.990 --> 01:13:45.550 align:middle line:84% Now, you may say, I don't know those good formulations. 01:13:45.550 --> 01:13:47.910 align:middle line:84% There are resources to actually pick them up, 01:13:47.910 --> 01:13:48.930 align:middle line:90% if you haven't already. 01:13:48.930 --> 01:13:51.150 align:middle line:84% I would say those folks in this class 01:13:51.150 --> 01:13:53.900 align:middle line:84% that you have already taken a course in system dynamics, 01:13:53.900 --> 01:13:56.100 align:middle line:84% you are familiar probably with 50% 01:13:56.100 --> 01:13:58.720 align:middle line:84% maybe more of standard formulations 01:13:58.720 --> 01:14:02.120 align:middle line:84% you could use in typical modeling challenges 01:14:02.120 --> 01:14:04.320 align:middle line:90% in this course and beyond. 01:14:04.320 --> 01:14:09.600 align:middle line:84% For going to 70%-80%, you can use the molecules that have been 01:14:09.600 --> 01:14:11.620 align:middle line:90% built by the community. 01:14:11.620 --> 01:14:18.880 align:middle line:84% Mostly, Jim Hines has invested in creating a coherent taxonomy 01:14:18.880 --> 01:14:21.460 align:middle line:90% of these formulations. 01:14:21.460 --> 01:14:23.820 align:middle line:84% You can see a screenshot of it here. 01:14:23.820 --> 01:14:25.700 align:middle line:84% But it's available through Vensim. 01:14:25.700 --> 01:14:29.320 align:middle line:84% It's available through the sdmolecules.org website. 01:14:29.320 --> 01:14:31.480 align:middle line:84% So basically, there are simple formulations. 01:14:31.480 --> 01:14:34.580 align:middle line:84% And it also has a hierarchy, logical hierarchy. 01:14:34.580 --> 01:14:36.540 align:middle line:84% It starts with a simple stock variable. 01:14:36.540 --> 01:14:42.080 align:middle line:84% And then a stock is the ancestor of a lot 01:14:42.080 --> 01:14:43.900 align:middle line:90% of different formulations. 01:14:43.900 --> 01:14:46.180 align:middle line:90% And you can see that ancestry. 01:14:46.180 --> 01:14:48.897 align:middle line:84% So proportional split, closing the gap, 01:14:48.897 --> 01:14:49.980 align:middle line:90% [? Bastof ?] is the stock. 01:14:49.980 --> 01:14:52.660 align:middle line:90% Those are the parents. 01:14:52.660 --> 01:14:56.120 align:middle line:84% And then you can see more and more complex 01:14:56.120 --> 01:15:00.416 align:middle line:90% items as you go down the-- 01:15:00.416 --> 01:15:03.080 align:middle line:90% let's look, for example. 01:15:03.080 --> 01:15:09.572 align:middle line:84% Here, we have the Hines experience coflow. 01:15:09.572 --> 01:15:10.780 align:middle line:90% So you would go through that. 01:15:10.780 --> 01:15:12.780 align:middle line:84% That might be new for most of the folks here. 01:15:12.780 --> 01:15:16.180 align:middle line:84% If you look at that formulation, you wouldn't recognize it. 01:15:16.180 --> 01:15:18.640 align:middle line:84% But it comes from Hines coflow, which may still 01:15:18.640 --> 01:15:21.040 align:middle line:84% be new for you, which comes from first-order smooth 01:15:21.040 --> 01:15:23.380 align:middle line:84% that you most likely already know about. 01:15:23.380 --> 01:15:25.560 align:middle line:84% So you may see that half of these things or more 01:15:25.560 --> 01:15:26.420 align:middle line:90% you already know. 01:15:26.420 --> 01:15:29.480 align:middle line:84% And then the new ones you can go through learning. 01:15:29.480 --> 01:15:32.640 align:middle line:84% And the way you would learn these molecules-- 01:15:32.640 --> 01:15:38.400 align:middle line:84% one way at least-- is, first, look at the causal diagram 01:15:38.400 --> 01:15:39.740 align:middle line:90% and write down the equations. 01:15:39.740 --> 01:15:41.760 align:middle line:84% Make sure you can write down the equations. 01:15:41.760 --> 01:15:45.520 align:middle line:84% Try to guess-- if there are ones that you already know, 01:15:45.520 --> 01:15:46.820 align:middle line:90% you should be able to. 01:15:46.820 --> 01:15:48.800 align:middle line:84% That's the definition of knowing the molecule. 01:15:48.800 --> 01:15:51.500 align:middle line:84% If you don't know, if it's a new one, 01:15:51.500 --> 01:15:54.080 align:middle line:84% try to guess what would be the formulations. 01:15:54.080 --> 01:15:56.740 align:middle line:84% And then identify what are the key inputs and outputs. 01:15:56.740 --> 01:16:00.200 align:middle line:84% So think of each as a chunk of a model that can be plugged 01:16:00.200 --> 01:16:03.020 align:middle line:90% into any new modeling task. 01:16:03.020 --> 01:16:04.600 align:middle line:84% So think, what are the pieces that 01:16:04.600 --> 01:16:06.840 align:middle line:84% are inputs-- that is, it should come 01:16:06.840 --> 01:16:09.620 align:middle line:84% from other chunks of your model-- and outputs-- 01:16:09.620 --> 01:16:12.233 align:middle line:84% that is, useful for other chunks of the models? 01:16:12.233 --> 01:16:14.400 align:middle line:84% And then there is a lot of things that don't usually 01:16:14.400 --> 01:16:18.080 align:middle line:84% interact with the rest of the model that's 01:16:18.080 --> 01:16:21.620 align:middle line:84% inherent to that piece, including a bunch of parameters. 01:16:21.620 --> 01:16:23.480 align:middle line:90% So know about those. 01:16:23.480 --> 01:16:26.343 align:middle line:84% Identify what's the meaning of those parameters. 01:16:26.343 --> 01:16:27.760 align:middle line:84% Then think about where you can use 01:16:27.760 --> 01:16:33.280 align:middle line:84% this chunk, in what kind of applications you can use those. 01:16:33.280 --> 01:16:35.560 align:middle line:84% And finally, predict the behavior 01:16:35.560 --> 01:16:41.660 align:middle line:84% of this chunk for a few standard inputs-- pulse, step, ramp, 01:16:41.660 --> 01:16:44.020 align:middle line:90% maybe oscillatory inputs. 01:16:44.020 --> 01:16:46.500 align:middle line:84% So if you change the inputs into one of these, 01:16:46.500 --> 01:16:49.610 align:middle line:84% what would be the expected-- what is your expected outputs? 01:16:49.610 --> 01:16:51.590 align:middle line:84% See whether you get that expected output. 01:16:51.590 --> 01:16:55.490 align:middle line:84% You can easily just copy and paste any of these 01:16:55.490 --> 01:17:00.290 align:middle line:84% and build a simple model of that chunk, 01:17:00.290 --> 01:17:04.890 align:middle line:84% test it, make sure you internalize an intuition 01:17:04.890 --> 01:17:06.150 align:middle line:90% for those behaviors. 01:17:06.150 --> 01:17:08.750 align:middle line:84% Because when you go into applying these molecules, 01:17:08.750 --> 01:17:10.610 align:middle line:84% you need to know what kind of behavior 01:17:10.610 --> 01:17:14.910 align:middle line:84% you expect to come out of each chunk of a molecule. 01:17:14.910 --> 01:17:16.390 align:middle line:84% If you don't have those intuitions, 01:17:16.390 --> 01:17:21.030 align:middle line:84% you don't really know what can be plugged into your model. 01:17:21.030 --> 01:17:24.210 align:middle line:84% And then you may end up thinking that, oh, maybe I 01:17:24.210 --> 01:17:26.230 align:middle line:84% should innovate and build a new structure. 01:17:26.230 --> 01:17:29.290 align:middle line:84% And that's a recipe for bad formulations. 01:17:29.290 --> 01:17:31.410 align:middle line:84% I think, for this class, by the end 01:17:31.410 --> 01:17:35.310 align:middle line:84% of the semester, maybe one formulation 01:17:35.310 --> 01:17:36.890 align:middle line:90% you are allowed to innovate. 01:17:36.890 --> 01:17:41.130 align:middle line:84% The rest you should be able to get from existing structures. 01:17:41.130 --> 01:17:43.970 align:middle line:84% I mean, that's about right, I think. 01:17:43.970 --> 01:17:47.430 align:middle line:84% Honestly, really new formulations, 01:17:47.430 --> 01:17:48.870 align:middle line:90% they are things to celebrate. 01:17:48.870 --> 01:17:51.290 align:middle line:90% They are not easy to come by. 01:17:51.290 --> 01:17:54.330 align:middle line:84% If they are really solving a meaningful problem, 01:17:54.330 --> 01:17:56.490 align:middle line:84% they should be added to the molecules. 01:17:56.490 --> 01:17:58.710 align:middle line:84% They should have their own publication. 01:17:58.710 --> 01:18:00.550 align:middle line:84% And this is how things also happen. 01:18:00.550 --> 01:18:03.830 align:middle line:84% That is, good, new formulations are rare. 01:18:03.830 --> 01:18:04.330 align:middle line:90% OK. 01:18:04.330 --> 01:18:09.790 align:middle line:84% So that's number 1 on the formulations, use existing ones. 01:18:09.790 --> 01:18:11.650 align:middle line:84% Replicating other classical models 01:18:11.650 --> 01:18:14.710 align:middle line:84% is another way to learn these formulations. 01:18:14.710 --> 01:18:18.410 align:middle line:84% And that's another good way of doing it. 01:18:18.410 --> 01:18:22.250 align:middle line:84% A few other tips is, when you want 01:18:22.250 --> 01:18:24.270 align:middle line:84% to build any chunk of a formulation, 01:18:24.270 --> 01:18:26.070 align:middle line:84% we are going through these iterations. 01:18:26.070 --> 01:18:27.750 align:middle line:90% So you have iteration 1. 01:18:27.750 --> 01:18:28.630 align:middle line:90% I have this stock. 01:18:28.630 --> 01:18:30.510 align:middle line:84% I'm trying to formulate its flow. 01:18:30.510 --> 01:18:36.450 align:middle line:84% I have identified some possible molecules 01:18:36.450 --> 01:18:39.758 align:middle line:84% that can be helpful for formulating it. 01:18:39.758 --> 01:18:42.050 align:middle line:84% Think through what is the objective of that chunk, what 01:18:42.050 --> 01:18:44.170 align:middle line:84% are its inputs and outputs and how 01:18:44.170 --> 01:18:47.270 align:middle line:84% those inputs and outputs should be connected with each other. 01:18:47.270 --> 01:18:49.090 align:middle line:84% That's the objective of that chunk, 01:18:49.090 --> 01:18:54.750 align:middle line:84% is creating some outputs with respect to some typical inputs. 01:18:54.750 --> 01:18:57.830 align:middle line:84% What are the acceptable ranges for those inputs and outputs? 01:18:57.830 --> 01:19:00.250 align:middle line:84% Make sure you impose those acceptable ranges 01:19:00.250 --> 01:19:02.170 align:middle line:84% within your equations for the output. 01:19:02.170 --> 01:19:03.770 align:middle line:84% For the input, it should be imposed 01:19:03.770 --> 01:19:06.410 align:middle line:84% in whatever chunk is providing the input 01:19:06.410 --> 01:19:08.230 align:middle line:90% to this part of your model. 01:19:08.230 --> 01:19:11.610 align:middle line:90% 01:19:11.610 --> 01:19:15.370 align:middle line:84% Then think about the regularities 01:19:15.370 --> 01:19:17.170 align:middle line:84% between inputs and outputs, whether they 01:19:17.170 --> 01:19:21.170 align:middle line:84% are the ones you want, whether it may do something weird, 01:19:21.170 --> 01:19:26.330 align:middle line:84% whether it is robust under extreme conditions. 01:19:26.330 --> 01:19:28.490 align:middle line:84% And then think about key mechanisms 01:19:28.490 --> 01:19:30.910 align:middle line:84% that this chunk is trying to capture. 01:19:30.910 --> 01:19:33.050 align:middle line:84% And do you have parameters that allow 01:19:33.050 --> 01:19:36.010 align:middle line:84% you to change the strength of those key mechanisms? 01:19:36.010 --> 01:19:39.610 align:middle line:84% So this is one of the places where I usually 01:19:39.610 --> 01:19:43.010 align:middle line:84% try not to use table functions because they 01:19:43.010 --> 01:19:45.810 align:middle line:84% limit my flexibility for changing 01:19:45.810 --> 01:19:47.110 align:middle line:90% the strength of mechanisms. 01:19:47.110 --> 01:19:49.670 align:middle line:84% So if there are simple analytical functions, 01:19:49.670 --> 01:19:53.210 align:middle line:84% I usually go with those rather than table functions that 01:19:53.210 --> 01:19:56.170 align:middle line:84% are very flexible, but are not easy to change 01:19:56.170 --> 01:19:58.870 align:middle line:84% because, in your analysis, in your calibration, 01:19:58.870 --> 01:20:00.610 align:middle line:84% you will need to be able to change 01:20:00.610 --> 01:20:02.810 align:middle line:84% the strength of different mechanisms. 01:20:02.810 --> 01:20:04.970 align:middle line:84% So those are some ideas for, when 01:20:04.970 --> 01:20:09.410 align:middle line:84% you get into building a chunk, think through these pieces 01:20:09.410 --> 01:20:10.810 align:middle line:90% explicitly. 01:20:10.810 --> 01:20:13.650 align:middle line:84% Then the other intuition is limit inputs 01:20:13.650 --> 01:20:15.510 align:middle line:90% to your formulations. 01:20:15.510 --> 01:20:18.850 align:middle line:84% Any single formulation equation, try 01:20:18.850 --> 01:20:22.210 align:middle line:84% to limit the number of inputs to three or fewer. 01:20:22.210 --> 01:20:24.370 align:middle line:90% What's the benefit of that? 01:20:24.370 --> 01:20:27.490 align:middle line:84% It helps you break down the logic in your model 01:20:27.490 --> 01:20:31.810 align:middle line:84% into smaller chunks that you can intuit and understand. 01:20:31.810 --> 01:20:34.250 align:middle line:84% It doesn't create surprises for you. 01:20:34.250 --> 01:20:36.730 align:middle line:84% It is easier to explain to others. 01:20:36.730 --> 01:20:39.170 align:middle line:84% You can look at those intermediate outcomes 01:20:39.170 --> 01:20:41.450 align:middle line:84% and diagnose when there is a problem. 01:20:41.450 --> 01:20:45.500 align:middle line:84% So they're all good reasons to limit the formulations 01:20:45.500 --> 01:20:47.620 align:middle line:90% to three or fewer. 01:20:47.620 --> 01:20:51.460 align:middle line:84% Discrete formulations-- for our typical continuous time 01:20:51.460 --> 01:20:54.660 align:middle line:84% continuous state space, system dynamics models 01:20:54.660 --> 01:20:57.500 align:middle line:84% are usually not very helpful because they 01:20:57.500 --> 01:21:00.940 align:middle line:84% create discontinuities that are hard to trace 01:21:00.940 --> 01:21:03.860 align:middle line:84% in terms of the impact on the behavior of the system. 01:21:03.860 --> 01:21:06.560 align:middle line:84% They usually reflect discrete thinking as well. 01:21:06.560 --> 01:21:10.740 align:middle line:84% People use if/then/elses very freely 01:21:10.740 --> 01:21:14.920 align:middle line:84% to capture their first order thinking about the problem 01:21:14.920 --> 01:21:17.900 align:middle line:84% without really going through those items 01:21:17.900 --> 01:21:19.520 align:middle line:90% on the top of the list. 01:21:19.520 --> 01:21:22.660 align:middle line:84% So to the extent possible-- now, sometimes, we 01:21:22.660 --> 01:21:24.340 align:middle line:90% do need min, max, and if. 01:21:24.340 --> 01:21:29.060 align:middle line:84% But generally, don't go with them as your first choice. 01:21:29.060 --> 01:21:31.220 align:middle line:84% And then ensuring unit consistency 01:21:31.220 --> 01:21:35.980 align:middle line:84% is probably the biggest, easiest thing 01:21:35.980 --> 01:21:38.940 align:middle line:84% you can do when you're building formulations. 01:21:38.940 --> 01:21:44.340 align:middle line:84% That's very simple, but very powerful. 01:21:44.340 --> 01:21:47.740 align:middle line:84% And one last item that was on our list 01:21:47.740 --> 01:22:01.880 align:middle line:84% is the big challenge of chasing a desired behavior. 01:22:01.880 --> 01:22:07.580 align:middle line:84% So this is how we get our PhD students in their general exams 01:22:07.580 --> 01:22:09.060 align:middle line:90% into trouble. 01:22:09.060 --> 01:22:13.380 align:middle line:84% Basically, you want the model to behave in some way 01:22:13.380 --> 01:22:15.600 align:middle line:84% because that's the reference mode, typically. 01:22:15.600 --> 01:22:17.940 align:middle line:90% That's the motivating thing. 01:22:17.940 --> 01:22:23.347 align:middle line:84% And then you are trying to get that behavior out of the model. 01:22:23.347 --> 01:22:25.180 align:middle line:84% But you don't know why the model is behaving 01:22:25.180 --> 01:22:26.700 align:middle line:90% the way it is behaving. 01:22:26.700 --> 01:22:29.580 align:middle line:84% So you end up changing the model parameters 01:22:29.580 --> 01:22:33.460 align:middle line:84% with the hope that it will work, it will give you 01:22:33.460 --> 01:22:37.780 align:middle line:84% the desired result. You do that for a while, 01:22:37.780 --> 01:22:40.380 align:middle line:90% usually to no avail. 01:22:40.380 --> 01:22:41.580 align:middle line:90% Then you get frustrated. 01:22:41.580 --> 01:22:43.680 align:middle line:84% And you say, OK, it is missing some structure. 01:22:43.680 --> 01:22:47.140 align:middle line:84% So you add some additional structure to fix that problem. 01:22:47.140 --> 01:22:49.940 align:middle line:84% And now, you have even burdened yourself 01:22:49.940 --> 01:22:51.620 align:middle line:84% with an even more complex problem 01:22:51.620 --> 01:22:53.100 align:middle line:90% that you don't understand. 01:22:53.100 --> 01:22:55.820 align:middle line:84% And this is the most frustrating and the biggest 01:22:55.820 --> 01:22:58.100 align:middle line:90% hog of your time. 01:22:58.100 --> 01:23:00.500 align:middle line:90% Never go down this road. 01:23:00.500 --> 01:23:03.540 align:middle line:84% If you want your model to behave in some way, 01:23:03.540 --> 01:23:06.940 align:middle line:84% you need to be able to basically set the parameters 01:23:06.940 --> 01:23:09.340 align:middle line:84% and it should behave the way you want it 01:23:09.340 --> 01:23:12.540 align:middle line:84% to because you have understood how the model behaves. 01:23:12.540 --> 01:23:14.880 align:middle line:84% You have gone through your analysis pieces. 01:23:14.880 --> 01:23:17.960 align:middle line:84% You know why it is behaving currently the way it does. 01:23:17.960 --> 01:23:22.080 align:middle line:84% And then you can say, OK, this behavior is what I want 01:23:22.080 --> 01:23:24.520 align:middle line:84% or, oh, I want it to behave this way. 01:23:24.520 --> 01:23:26.308 align:middle line:84% It just means I go to parameter x 01:23:26.308 --> 01:23:27.600 align:middle line:90% and increase it or decrease it. 01:23:27.600 --> 01:23:29.860 align:middle line:84% And it will behave the way it should 01:23:29.860 --> 01:23:31.480 align:middle line:90% because I understand my model. 01:23:31.480 --> 01:23:34.500 align:middle line:84% So always go through understanding rather than 01:23:34.500 --> 01:23:38.220 align:middle line:84% any kind of a blind search in the parameter space. 01:23:38.220 --> 01:23:39.580 align:middle line:90% Never do that. 01:23:39.580 --> 01:23:40.780 align:middle line:90% It's very common. 01:23:40.780 --> 01:23:42.580 align:middle line:90% It never works. 01:23:42.580 --> 01:23:47.240 align:middle line:84% OK, so that's that, I think, for our theoretical space. 01:23:47.240 --> 01:23:48.790 align:middle line:84% So let's see if there are questions. 01:23:48.790 --> 01:23:50.540 align:middle line:84% And then we can take a break and come back 01:23:50.540 --> 01:23:55.860 align:middle line:84% with the more practical looking at problems today together. 01:23:55.860 --> 01:23:57.320 align:middle line:90% Questions, thoughts? 01:23:57.320 --> 01:24:01.580 align:middle line:90% 01:24:01.580 --> 01:24:05.060 align:middle line:90% Let's come back together. 01:24:05.060 --> 01:24:08.380 align:middle line:84% And folks online, please join us as well. 01:24:08.380 --> 01:24:09.020 align:middle line:90% OK. 01:24:09.020 --> 01:24:10.700 align:middle line:84% So for the remainder of the class-- 01:24:10.700 --> 01:24:13.260 align:middle line:84% I think we have an hour and 10 minutes or so-- 01:24:13.260 --> 01:24:17.080 align:middle line:84% we will be basically talking about architectural choices, 01:24:17.080 --> 01:24:19.260 align:middle line:84% but in the context of a couple of examples 01:24:19.260 --> 01:24:21.840 align:middle line:84% that hopefully you had a chance to look at those. 01:24:21.840 --> 01:24:24.520 align:middle line:84% We will bring them up on the screen as well. 01:24:24.520 --> 01:24:28.740 align:middle line:84% So when I talk about architectural choices, 01:24:28.740 --> 01:24:34.700 align:middle line:84% basically, there is a set of interconnected questions. 01:24:34.700 --> 01:24:38.470 align:middle line:84% What's the problem definition and the audience 01:24:38.470 --> 01:24:40.090 align:middle line:90% for that problem definition? 01:24:40.090 --> 01:24:41.230 align:middle line:90% Who cares? 01:24:41.230 --> 01:24:42.930 align:middle line:84% Who are you trying to communicate with? 01:24:42.930 --> 01:24:47.670 align:middle line:84% Who are you trying to help or help them see the world 01:24:47.670 --> 01:24:49.710 align:middle line:90% a little bit differently? 01:24:49.710 --> 01:24:54.350 align:middle line:84% Then what's a good time horizon for the problem? 01:24:54.350 --> 01:24:57.935 align:middle line:84% What are the mechanisms that should be included in the model? 01:24:57.935 --> 01:24:59.310 align:middle line:84% What's the boundary of the model? 01:24:59.310 --> 01:25:01.030 align:middle line:90% And what should be left out? 01:25:01.030 --> 01:25:04.630 align:middle line:84% Leaving things out is always important. 01:25:04.630 --> 01:25:07.930 align:middle line:84% Then the level of aggregation question. 01:25:07.930 --> 01:25:11.290 align:middle line:84% And finally, key sources of inertia and the stock variables. 01:25:11.290 --> 01:25:13.370 align:middle line:84% So we have already gone through these. 01:25:13.370 --> 01:25:14.890 align:middle line:90% Now, let's go and practice them. 01:25:14.890 --> 01:25:17.110 align:middle line:84% You want to actually ask all of these questions 01:25:17.110 --> 01:25:19.790 align:middle line:84% for every modeling problem before you get started 01:25:19.790 --> 01:25:25.030 align:middle line:84% with putting structures on the Vensim or Stella 01:25:25.030 --> 01:25:28.950 align:middle line:84% or whatever your software package is. 01:25:28.950 --> 01:25:31.710 align:middle line:84% So we are doing this for two problems today. 01:25:31.710 --> 01:25:33.710 align:middle line:84% And then hopefully, we'll have also a little bit 01:25:33.710 --> 01:25:38.063 align:middle line:84% of time talking about the next week's problem and maybe 01:25:38.063 --> 01:25:39.730 align:middle line:84% some of the architectural choices there. 01:25:39.730 --> 01:25:41.590 align:middle line:84% Although, most of them are already 01:25:41.590 --> 01:25:43.630 align:middle line:84% baked into the problem definition. 01:25:43.630 --> 01:25:46.890 align:middle line:84% So the first one is the thoroughbred horse market. 01:25:46.890 --> 01:25:50.390 align:middle line:84% So most of these are coming from John Sterman's Business Dynamics 01:25:50.390 --> 01:25:51.050 align:middle line:90% textbook. 01:25:51.050 --> 01:25:56.510 align:middle line:84% I provided what you needed to learn about it in the problem 01:25:56.510 --> 01:26:00.310 align:middle line:90% definition PDF. 01:26:00.310 --> 01:26:05.930 align:middle line:84% But this is the story of thoroughbred horse market, 01:26:05.930 --> 01:26:09.390 align:middle line:84% the number of thoroughbreds born in the US. 01:26:09.390 --> 01:26:12.150 align:middle line:84% There are other reference modes in the problem definition 01:26:12.150 --> 01:26:13.710 align:middle line:90% that you have. 01:26:13.710 --> 01:26:17.750 align:middle line:84% Basically, it went through a boom and bust of a sort. 01:26:17.750 --> 01:26:22.910 align:middle line:84% This market became very common for people, 01:26:22.910 --> 01:26:27.350 align:middle line:84% for very wealthy people, to try to have these racehorses. 01:26:27.350 --> 01:26:31.990 align:middle line:84% And then at some point, it crashed. 01:26:31.990 --> 01:26:39.230 align:middle line:84% And basically, John uses this as an example to put flesh 01:26:39.230 --> 01:26:41.670 align:middle line:90% on the idea of bubbles. 01:26:41.670 --> 01:26:44.510 align:middle line:84% So there's a generic structure for bubbles, 01:26:44.510 --> 01:26:49.710 align:middle line:84% where people's demand for whatever they are looking at 01:26:49.710 --> 01:26:54.970 align:middle line:84% goes up because it is becoming more expensive 01:26:54.970 --> 01:26:58.070 align:middle line:84% and because they think there is room for growth. 01:26:58.070 --> 01:27:03.230 align:middle line:84% So you can think about Bitcoin's bubbles and crash, 01:27:03.230 --> 01:27:04.290 align:middle line:90% goes up and crashes. 01:27:04.290 --> 01:27:07.690 align:middle line:84% And every five years, we have had one of these big. 01:27:07.690 --> 01:27:10.130 align:middle line:84% Right now, we are at the top of the current one. 01:27:10.130 --> 01:27:12.070 align:middle line:84% But at some point, probably this will also 01:27:12.070 --> 01:27:15.310 align:middle line:84% go through the same mode of behavior. 01:27:15.310 --> 01:27:20.070 align:middle line:84% Either way, the thoroughbred horse market, then John 01:27:20.070 --> 01:27:22.850 align:middle line:84% goes through giving an answer to that challenge. 01:27:22.850 --> 01:27:25.890 align:middle line:84% So the challenge asks you to build a causal loop diagram. 01:27:25.890 --> 01:27:29.510 align:middle line:84% He builds a causal loop diagram in the reference manual 01:27:29.510 --> 01:27:32.270 align:middle line:90% for the class-- for the book. 01:27:32.270 --> 01:27:35.730 align:middle line:84% So this is the big picture of that. 01:27:35.730 --> 01:27:36.950 align:middle line:90% I think you have the PDFs. 01:27:36.950 --> 01:27:38.367 align:middle line:84% If this is not big enough, I think 01:27:38.367 --> 01:27:42.050 align:middle line:84% it's a little bit harder to read from the end of the room. 01:27:42.050 --> 01:27:47.750 align:middle line:84% But the basic story here is you have-- 01:27:47.750 --> 01:27:49.950 align:middle line:90% let's see, yeah. 01:27:49.950 --> 01:27:52.290 align:middle line:84% So you can think of the thoroughbred prices, 01:27:52.290 --> 01:27:59.390 align:middle line:84% thoroughbred racehorses, prices as a central variable here. 01:27:59.390 --> 01:28:01.490 align:middle line:84% And then you can think about what is driving that. 01:28:01.490 --> 01:28:04.050 align:middle line:84% Well, it is largely the demand versus supply. 01:28:04.050 --> 01:28:08.950 align:middle line:84% So you need to then figure out where the demand and supply are 01:28:08.950 --> 01:28:10.630 align:middle line:90% coming from. 01:28:10.630 --> 01:28:13.270 align:middle line:84% Supply is coming from breeders deciding 01:28:13.270 --> 01:28:20.310 align:middle line:84% whether they should invest into breeding and bringing 01:28:20.310 --> 01:28:21.830 align:middle line:90% in more horses. 01:28:21.830 --> 01:28:30.920 align:middle line:84% And there are delays, biological delays in horses coming to-- 01:28:30.920 --> 01:28:33.800 align:middle line:84% basically being born and being raised to the age 01:28:33.800 --> 01:28:38.280 align:middle line:90% that they can be in the market. 01:28:38.280 --> 01:28:40.720 align:middle line:84% There is also the expectation formation 01:28:40.720 --> 01:28:43.200 align:middle line:84% for what the price in future would be. 01:28:43.200 --> 01:28:46.620 align:middle line:84% So that expectation formation depends on the price itself. 01:28:46.620 --> 01:28:51.080 align:middle line:84% So there is a process for modeling or thinking 01:28:51.080 --> 01:28:54.940 align:middle line:84% about how the expectations are formed about future prices. 01:28:54.940 --> 01:28:58.180 align:middle line:84% There is an extrapolation process. 01:28:58.180 --> 01:29:00.920 align:middle line:84% And that future price expectations 01:29:00.920 --> 01:29:05.000 align:middle line:84% also impacts what people expect in terms of capital gains 01:29:05.000 --> 01:29:08.120 align:middle line:84% from having these horses that are appreciating 01:29:08.120 --> 01:29:10.720 align:middle line:84% in price, so you can actually sell them later and make money, 01:29:10.720 --> 01:29:15.200 align:middle line:84% not only from what you earn when they win races-- and some 01:29:15.200 --> 01:29:16.260 align:middle line:90% of them win races. 01:29:16.260 --> 01:29:23.620 align:middle line:84% And there is an active winning market for it. 01:29:23.620 --> 01:29:25.780 align:middle line:90% And that increases earnings. 01:29:25.780 --> 01:29:28.300 align:middle line:84% But also, there is the expected gains. 01:29:28.300 --> 01:29:30.400 align:middle line:84% And then as the prices go up, there 01:29:30.400 --> 01:29:34.920 align:middle line:84% is a status value associated with owning a racehorse. 01:29:34.920 --> 01:29:37.140 align:middle line:84% And that is another source of the demand. 01:29:37.140 --> 01:29:40.080 align:middle line:90% 01:29:40.080 --> 01:29:44.320 align:middle line:84% And in fact, some of the people, because they 01:29:44.320 --> 01:29:48.240 align:middle line:84% have multiple racehorses that are now very expensive, 01:29:48.240 --> 01:29:52.280 align:middle line:84% they may actually feel they are in a whole different category. 01:29:52.280 --> 01:29:58.740 align:middle line:84% And so that creates another source of pressure on demand. 01:29:58.740 --> 01:30:01.560 align:middle line:84% So those are the core mechanisms here. 01:30:01.560 --> 01:30:06.058 align:middle line:84% And so let's say we have the mechanism. 01:30:06.058 --> 01:30:07.600 align:middle line:84% So this is the boundary of the model. 01:30:07.600 --> 01:30:11.620 align:middle line:84% In a sense, we don't want to go beyond this boundary, 01:30:11.620 --> 01:30:14.560 align:middle line:84% add other structures that are relevant. 01:30:14.560 --> 01:30:16.640 align:middle line:84% For any real problem definition, we 01:30:16.640 --> 01:30:19.520 align:middle line:84% would actually need to consider what else is important 01:30:19.520 --> 01:30:21.620 align:middle line:84% and whether all of these things are important. 01:30:21.620 --> 01:30:24.000 align:middle line:84% But for now, let's just stick with that 01:30:24.000 --> 01:30:25.780 align:middle line:90% and think about audiences. 01:30:25.780 --> 01:30:30.600 align:middle line:84% So who would be an audience for this problem? 01:30:30.600 --> 01:30:33.060 align:middle line:84% And there is no single correct answer. 01:30:33.060 --> 01:30:34.700 align:middle line:90% But I'm curious. 01:30:34.700 --> 01:30:38.880 align:middle line:84% So share your thoughts on who would be a good audience. 01:30:38.880 --> 01:30:41.320 align:middle line:84% And then based on that, what would be the right problem 01:30:41.320 --> 01:30:42.700 align:middle line:90% definition for them? 01:30:42.700 --> 01:30:45.960 align:middle line:90% 01:30:45.960 --> 01:30:47.020 align:middle line:90% It's open. 01:30:47.020 --> 01:30:49.720 align:middle line:90% I'm not going to talk more. 01:30:49.720 --> 01:30:52.360 align:middle line:90% 01:30:52.360 --> 01:30:54.280 align:middle line:90% Oriana? 01:30:54.280 --> 01:30:57.440 align:middle line:84% AUDIENCE: Maybe one group could actually be high net worth 01:30:57.440 --> 01:31:01.120 align:middle line:84% individuals who would be interested in having horses 01:31:01.120 --> 01:31:06.480 align:middle line:84% or a variety of other types of investments 01:31:06.480 --> 01:31:09.100 align:middle line:84% like this, whether it's fine art or wines, like you mentioned. 01:31:09.100 --> 01:31:10.940 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 01:31:10.940 --> 01:31:15.200 align:middle line:84% STUDENT: In which case, sorry, you should probably be tracking 01:31:15.200 --> 01:31:17.660 align:middle line:90% prices and time horizon. 01:31:17.660 --> 01:31:19.600 align:middle line:84% Well, actually, I guess my concern, 01:31:19.600 --> 01:31:22.960 align:middle line:84% my question with time horizon is, what set off the bubble? 01:31:22.960 --> 01:31:26.840 align:middle line:84% Because it seems like it was in some level of equilibrium 01:31:26.840 --> 01:31:30.000 align:middle line:84% and closed off within very high net worth individuals 01:31:30.000 --> 01:31:34.800 align:middle line:84% until the Inflation problem made people who probably shouldn't 01:31:34.800 --> 01:31:37.980 align:middle line:84% have been in there to begin with think it was an inflation hedge. 01:31:37.980 --> 01:31:43.080 align:middle line:84% So you got an excess of people into the thoroughbred industry. 01:31:43.080 --> 01:31:45.540 align:middle line:84% So I guess, I'm a little confused about time horizon 01:31:45.540 --> 01:31:49.400 align:middle line:84% because it seems more of a response to some external shock 01:31:49.400 --> 01:31:52.110 align:middle line:84% rather than a completely inside model behavior. 01:31:52.110 --> 01:31:53.360 align:middle line:90% HAZHIR RAHMANDAD: Interesting. 01:31:53.360 --> 01:31:56.720 align:middle line:84% Yeah, so for bubble models, when they pop 01:31:56.720 --> 01:31:59.600 align:middle line:84% is usually a mix of endogenous dynamics 01:31:59.600 --> 01:32:03.380 align:middle line:84% that has made things too inflated. 01:32:03.380 --> 01:32:06.320 align:middle line:84% And people who are looking at the fundamentals cannot justify 01:32:06.320 --> 01:32:09.040 align:middle line:84% it, and then an external trigger, 01:32:09.040 --> 01:32:13.280 align:middle line:84% that kind of changes the minds of enough people on the margin 01:32:13.280 --> 01:32:15.540 align:middle line:84% that they decide to get out of the market. 01:32:15.540 --> 01:32:17.460 align:middle line:84% That puts a downward pressure on the price. 01:32:17.460 --> 01:32:21.040 align:middle line:84% And as soon as the price starts to slow down or come down, 01:32:21.040 --> 01:32:25.310 align:middle line:84% things will go awry in the opposite direction, 01:32:25.310 --> 01:32:26.310 align:middle line:90% bring down the market. 01:32:26.310 --> 01:32:29.030 align:middle line:84% So that, as you say, is not predictable 01:32:29.030 --> 01:32:30.850 align:middle line:84% because those external shocks may 01:32:30.850 --> 01:32:34.390 align:middle line:84% come at times that are not in your model boundary. 01:32:34.390 --> 01:32:37.170 align:middle line:84% But for the time horizon, you don't need 01:32:37.170 --> 01:32:38.590 align:middle line:90% to worry too much about those. 01:32:38.590 --> 01:32:41.210 align:middle line:84% You want to have long enough time horizon that 01:32:41.210 --> 01:32:45.290 align:middle line:84% captures the underlying pressures inside the system 01:32:45.290 --> 01:32:49.110 align:middle line:84% that could ultimately lead to a collapse, 01:32:49.110 --> 01:32:53.670 align:middle line:84% even without this external popping of the market. 01:32:53.670 --> 01:32:58.090 align:middle line:84% So external shocks can bring it forward or sometimes delay it 01:32:58.090 --> 01:33:03.010 align:middle line:84% because they make the space very attractive, 01:33:03.010 --> 01:33:06.730 align:middle line:84% even though the underlying fundamentals don't justify it. 01:33:06.730 --> 01:33:10.530 align:middle line:84% So with that, net worth individuals as you 01:33:10.530 --> 01:33:13.330 align:middle line:84% client, high net worth individuals, good. 01:33:13.330 --> 01:33:15.010 align:middle line:90% What's the problem for them? 01:33:15.010 --> 01:33:16.770 align:middle line:90% Should they invest or not? 01:33:16.770 --> 01:33:18.110 align:middle line:90% Something along those lines. 01:33:18.110 --> 01:33:21.670 align:middle line:84% So you are helping somebody make decisions on investment. 01:33:21.670 --> 01:33:24.610 align:middle line:84% That's one audience that is different, for example, 01:33:24.610 --> 01:33:28.290 align:middle line:84% from I want to publish a paper in a sociology journal 01:33:28.290 --> 01:33:33.970 align:middle line:84% about dynamics of bubbles in luxury goods. 01:33:33.970 --> 01:33:37.310 align:middle line:84% The two have overlapping-- in fact, 01:33:37.310 --> 01:33:39.770 align:middle line:84% they may both start with this diagram. 01:33:39.770 --> 01:33:41.270 align:middle line:84% But they are two different audiences 01:33:41.270 --> 01:33:44.430 align:middle line:84% who require different levels of rigor. 01:33:44.430 --> 01:33:48.530 align:middle line:84% So if we had these two different audiences, a sociology journal 01:33:48.530 --> 01:33:51.090 align:middle line:84% versus a high net worth individual who 01:33:51.090 --> 01:33:56.030 align:middle line:84% is interested in figuring out what to do in this market, 01:33:56.030 --> 01:34:00.890 align:middle line:84% whether to get in or get out, how 01:34:00.890 --> 01:34:04.250 align:middle line:84% would that impact your architectural choices? 01:34:04.250 --> 01:34:05.030 align:middle line:90% Any thoughts? 01:34:05.030 --> 01:34:08.010 align:middle line:84% What are the requirements that come with these two 01:34:08.010 --> 01:34:09.550 align:middle line:90% different audiences? 01:34:09.550 --> 01:34:16.330 align:middle line:90% 01:34:16.330 --> 01:34:20.250 align:middle line:84% Alexander has his hand is kind of-- 01:34:20.250 --> 01:34:23.330 align:middle line:84% AUDIENCE: I think, for the sociology journal, that's more 01:34:23.330 --> 01:34:25.790 align:middle line:90% of a theory building paper. 01:34:25.790 --> 01:34:27.570 align:middle line:84% And so I think it's better for the model 01:34:27.570 --> 01:34:31.130 align:middle line:84% to be simpler and not necessarily exactly calibrated 01:34:31.130 --> 01:34:33.710 align:middle line:84% to data, just as long as it corresponds to it. 01:34:33.710 --> 01:34:36.610 align:middle line:84% So it's more easily understandable by an audience 01:34:36.610 --> 01:34:38.990 align:middle line:84% that doesn't have a modeling background, typically. 01:34:38.990 --> 01:34:41.190 align:middle line:84% And then I think, on the other hand, 01:34:41.190 --> 01:34:43.730 align:middle line:84% if you're trying to suggest prices, 01:34:43.730 --> 01:34:47.370 align:middle line:84% you probably want to have much better calibration 01:34:47.370 --> 01:34:48.130 align:middle line:90% with the data. 01:34:48.130 --> 01:34:51.430 align:middle line:84% You probably want to have a deeper level of aggregation. 01:34:51.430 --> 01:34:53.350 align:middle line:84% Maybe there's different types of investors. 01:34:53.350 --> 01:34:56.250 align:middle line:84% Maybe there's different types of readers and so on. 01:34:56.250 --> 01:34:58.310 align:middle line:84% And so that way, you can maybe not-- 01:34:58.310 --> 01:35:00.030 align:middle line:84% maybe it would be difficult to predict. 01:35:00.030 --> 01:35:02.770 align:middle line:84% But you could have a much better idea 01:35:02.770 --> 01:35:05.613 align:middle line:84% of what the trajectory of these prices and such will be. 01:35:05.613 --> 01:35:06.530 align:middle line:90% HAZHIR RAHMANDAD: Yep. 01:35:06.530 --> 01:35:08.830 align:middle line:84% So those are very important distinctions. 01:35:08.830 --> 01:35:11.970 align:middle line:84% So definitely, if you go with a sociology journal, 01:35:11.970 --> 01:35:13.730 align:middle line:84% the purpose of the model is actually 01:35:13.730 --> 01:35:16.770 align:middle line:84% building intuition and communicating it 01:35:16.770 --> 01:35:19.110 align:middle line:90% with high bandwidth clarity. 01:35:19.110 --> 01:35:24.650 align:middle line:84% That means very simple model, just qualitative mechanisms 01:35:24.650 --> 01:35:30.010 align:middle line:84% at a high level, nothing beyond this level in terms 01:35:30.010 --> 01:35:32.350 align:middle line:90% of the qualitative model. 01:35:32.350 --> 01:35:34.650 align:middle line:84% And then the quantitative model should not 01:35:34.650 --> 01:35:38.230 align:middle line:84% have more complexity conceptually than this. 01:35:38.230 --> 01:35:42.370 align:middle line:84% And the level of aggregation should be high and so on. 01:35:42.370 --> 01:35:46.930 align:middle line:84% Whereas, if you are helping a high net worth individual 01:35:46.930 --> 01:35:50.850 align:middle line:84% make actual decisions with million dollar investments 01:35:50.850 --> 01:35:55.370 align:middle line:84% in this space, then you need good calibration, 01:35:55.370 --> 01:35:57.490 align:middle line:90% some additional details. 01:35:57.490 --> 01:36:00.690 align:middle line:84% You need a lot of actually external inputs 01:36:00.690 --> 01:36:02.570 align:middle line:84% into your model-- that is, things 01:36:02.570 --> 01:36:06.170 align:middle line:84% that are not fully endogenous-- because real world doesn't 01:36:06.170 --> 01:36:09.330 align:middle line:84% work only with endogenous dynamics we know about 01:36:09.330 --> 01:36:12.350 align:middle line:84% and care about in theoretical modeling, 01:36:12.350 --> 01:36:16.220 align:middle line:84% but also is driven by a lot of shocks from the general economy. 01:36:16.220 --> 01:36:19.233 align:middle line:84% In this case, it makes a big difference. 01:36:19.233 --> 01:36:20.900 align:middle line:84% The stock market makes a big difference. 01:36:20.900 --> 01:36:23.420 align:middle line:84% Interest rates make a difference. 01:36:23.420 --> 01:36:26.900 align:middle line:84% And you may need to trace down and see 01:36:26.900 --> 01:36:31.700 align:middle line:84% what is the stock of breeding horses that 01:36:31.700 --> 01:36:35.780 align:middle line:84% are in different parts of the country, 01:36:35.780 --> 01:36:37.040 align:middle line:90% how many breeders are active. 01:36:37.040 --> 01:36:39.860 align:middle line:84% So you may need to collect additional data. 01:36:39.860 --> 01:36:41.380 align:middle line:84% Your level of aggregation probably 01:36:41.380 --> 01:36:43.480 align:middle line:90% goes at least one level deeper. 01:36:43.480 --> 01:36:46.500 align:middle line:84% So the audience matters, even if you are just 01:36:46.500 --> 01:36:51.260 align:middle line:84% sticking with the same qualitative diagram. 01:36:51.260 --> 01:36:55.160 align:middle line:84% The time horizon actually could also vary. 01:36:55.160 --> 01:37:00.260 align:middle line:84% That is, for your generic model, would it be longer or shorter? 01:37:00.260 --> 01:37:01.380 align:middle line:90% AUDIENCE: [INAUDIBLE] 01:37:01.380 --> 01:37:03.500 align:middle line:90% HAZHIR RAHMANDAD: Why? 01:37:03.500 --> 01:37:06.260 align:middle line:84% AUDIENCE: [INAUDIBLE] to witness different types 01:37:06.260 --> 01:37:11.122 align:middle line:84% of dynamics that can arise from the structure [INAUDIBLE]. 01:37:11.122 --> 01:37:12.580 align:middle line:84% HAZHIR RAHMANDAD: Yeah, absolutely. 01:37:12.580 --> 01:37:14.900 align:middle line:84% So the theoretical model actually 01:37:14.900 --> 01:37:17.080 align:middle line:84% typically benefits from a longer time horizon. 01:37:17.080 --> 01:37:20.620 align:middle line:84% You want to see all of the implications play out. 01:37:20.620 --> 01:37:23.800 align:middle line:84% If you're capturing a feedback mechanism, 01:37:23.800 --> 01:37:27.780 align:middle line:84% you need to show them that it has an impact. 01:37:27.780 --> 01:37:29.240 align:middle line:90% So that's the reason. 01:37:29.240 --> 01:37:32.220 align:middle line:84% On the other hand, for the investment decision, 01:37:32.220 --> 01:37:38.580 align:middle line:84% ultimately, your horizon for decision is the next year or so. 01:37:38.580 --> 01:37:43.400 align:middle line:84% Now, you are also thinking about longer term risks. 01:37:43.400 --> 01:37:46.180 align:middle line:84% So do I stay or do I leave the market? 01:37:46.180 --> 01:37:49.280 align:middle line:84% And so maybe a few more years in the future. 01:37:49.280 --> 01:37:51.140 align:middle line:84% But projections beyond that are not 01:37:51.140 --> 01:37:53.600 align:middle line:84% relevant for your current investment decision. 01:37:53.600 --> 01:37:55.620 align:middle line:84% But you actually need to continuously 01:37:55.620 --> 01:37:57.685 align:middle line:84% feed data to calibrate the model, 01:37:57.685 --> 01:37:59.060 align:middle line:84% as long as you are in this market 01:37:59.060 --> 01:38:01.380 align:middle line:84% or considering this market, to always 01:38:01.380 --> 01:38:04.980 align:middle line:84% have a good one year ahead projection of what are 01:38:04.980 --> 01:38:06.760 align:middle line:90% the opportunities and risks. 01:38:06.760 --> 01:38:11.580 align:middle line:84% And that's a different time horizon, both backward 01:38:11.580 --> 01:38:12.880 align:middle line:90% looking and forward looking. 01:38:12.880 --> 01:38:17.140 align:middle line:90% 01:38:17.140 --> 01:38:22.980 align:middle line:84% So level of aggregation we talked about a little bit. 01:38:22.980 --> 01:38:26.540 align:middle line:84% Now, if we want to go into the question of stocks and sources 01:38:26.540 --> 01:38:29.140 align:middle line:84% of inertia, sources of inertia, we 01:38:29.140 --> 01:38:34.420 align:middle line:84% could talk about, for both of these potential clients, 01:38:34.420 --> 01:38:37.760 align:middle line:90% with similar answers. 01:38:37.760 --> 01:38:40.420 align:middle line:84% But when we get into stocks, it really 01:38:40.420 --> 01:38:43.380 align:middle line:84% makes a difference which one we want to focus on. 01:38:43.380 --> 01:38:50.420 align:middle line:84% So let's talk about sources of inertia. 01:38:50.420 --> 01:38:53.180 align:middle line:84% Actually, for this, maybe it's a good idea 01:38:53.180 --> 01:38:58.220 align:middle line:84% that you do it independently for maybe five minutes. 01:38:58.220 --> 01:39:01.020 align:middle line:84% What you want to do is to identify 01:39:01.020 --> 01:39:04.520 align:middle line:84% sources of inertia that are important, 01:39:04.520 --> 01:39:05.760 align:middle line:90% and then prioritize them. 01:39:05.760 --> 01:39:08.560 align:middle line:84% That is, if you were going through the iterations of model, 01:39:08.560 --> 01:39:13.470 align:middle line:84% analyze, model, analyze that we talked about, which ones would 01:39:13.470 --> 01:39:16.830 align:middle line:84% you think are the most important and which ones would 01:39:16.830 --> 01:39:18.590 align:middle line:90% be secondary and tertiary? 01:39:18.590 --> 01:39:21.430 align:middle line:84% It doesn't necessarily mean you will always tackle the most 01:39:21.430 --> 01:39:23.390 align:middle line:84% important first because it also partly 01:39:23.390 --> 01:39:27.233 align:middle line:84% is determined by how comfortable you are modeling those chunks. 01:39:27.233 --> 01:39:29.150 align:middle line:84% You usually want to start with the things that 01:39:29.150 --> 01:39:30.850 align:middle line:90% are easier for you to model. 01:39:30.850 --> 01:39:32.390 align:middle line:90% You know how to model those. 01:39:32.390 --> 01:39:34.170 align:middle line:90% So put that aside right now. 01:39:34.170 --> 01:39:36.910 align:middle line:84% Just focus on the importance in terms of the underlying 01:39:36.910 --> 01:39:38.470 align:middle line:90% dynamics. 01:39:38.470 --> 01:39:42.190 align:middle line:84% If you care about the audience, let's be more inclined 01:39:42.190 --> 01:39:45.710 align:middle line:84% towards the theoretical sociological journal audience 01:39:45.710 --> 01:39:50.870 align:middle line:90% than the net high worths-- 01:39:50.870 --> 01:39:56.750 align:middle line:84% high net worth individual clients that we hypothesized. 01:39:56.750 --> 01:39:58.690 align:middle line:90% OK, so five minutes on that. 01:39:58.690 --> 01:40:04.510 align:middle line:84% Then we'll get your thoughts and input on that. 01:40:04.510 --> 01:40:09.910 align:middle line:84% Let's start thinking about these together now. 01:40:09.910 --> 01:40:14.270 align:middle line:84% So maybe we can reach out to folks online 01:40:14.270 --> 01:40:19.310 align:middle line:84% and see if anybody online wants to share 01:40:19.310 --> 01:40:22.970 align:middle line:84% their thoughts on important sources of inertia here. 01:40:22.970 --> 01:40:34.510 align:middle line:90% 01:40:34.510 --> 01:40:39.912 align:middle line:84% You can volunteer yourself, just unmute and say, whoever. 01:40:39.912 --> 01:40:41.870 align:middle line:84% AUDIENCE: I think one of the sources of inertia 01:40:41.870 --> 01:40:45.710 align:middle line:84% is between when they decide to breed the horse 01:40:45.710 --> 01:40:48.190 align:middle line:84% and when the horse can actually go to the market. 01:40:48.190 --> 01:40:50.420 align:middle line:90% HAZHIR RAHMANDAD: OK. 01:40:50.420 --> 01:41:01.470 align:middle line:84% So basically, horse maturity delay, 01:41:01.470 --> 01:41:03.590 align:middle line:90% something along those lines. 01:41:03.590 --> 01:41:06.630 align:middle line:90% OK, good. 01:41:06.630 --> 01:41:09.870 align:middle line:90% What else? 01:41:09.870 --> 01:41:15.010 align:middle line:84% AUDIENCE: I was thinking about prices, demand and supply. 01:41:15.010 --> 01:41:17.710 align:middle line:84% But now, I'm thinking maybe I don't 01:41:17.710 --> 01:41:21.030 align:middle line:84% understand the clear difference between sources 01:41:21.030 --> 01:41:22.910 align:middle line:90% of inertia and stocks. 01:41:22.910 --> 01:41:24.170 align:middle line:90% HAZHIR RAHMANDAD: OK, good. 01:41:24.170 --> 01:41:27.450 align:middle line:84% So sources of inertia are a little bit higher level. 01:41:27.450 --> 01:41:34.190 align:middle line:84% That is, you may say, within the space of supply, 01:41:34.190 --> 01:41:39.150 align:middle line:84% I see the biological processes related to horses 01:41:39.150 --> 01:41:42.670 align:middle line:84% getting pregnant and having children. 01:41:42.670 --> 01:41:44.430 align:middle line:90% I don't know. 01:41:44.430 --> 01:41:46.682 align:middle line:84% I'm not using the right terminology. 01:41:46.682 --> 01:41:47.390 align:middle line:90% AUDIENCE: Calves. 01:41:47.390 --> 01:41:49.870 align:middle line:84% HAZHIR RAHMANDAD: Calves, yes, is the right-- 01:41:49.870 --> 01:41:51.130 align:middle line:90% is important. 01:41:51.130 --> 01:41:52.470 align:middle line:90% So that's kind of-- 01:41:52.470 --> 01:41:55.230 align:middle line:84% and then you can go into the specific stocks, 01:41:55.230 --> 01:41:57.910 align:middle line:84% if you wanted to capture that source of inertia. 01:41:57.910 --> 01:42:01.110 align:middle line:84% So the source of inertia is a little bit higher level 01:42:01.110 --> 01:42:02.790 align:middle line:90% and a little bit more vague. 01:42:02.790 --> 01:42:05.150 align:middle line:84% So you may say, in the demand, I think 01:42:05.150 --> 01:42:07.950 align:middle line:84% there is some perception processes that 01:42:07.950 --> 01:42:12.630 align:middle line:84% are important to be captured or something like that. 01:42:12.630 --> 01:42:16.150 align:middle line:84% I'm just breaking the process into two steps 01:42:16.150 --> 01:42:18.890 align:middle line:84% with sources of inertia and the stocks 01:42:18.890 --> 01:42:20.890 align:middle line:84% because, depending on the audience, 01:42:20.890 --> 01:42:22.630 align:middle line:84% the level of aggregation at which we 01:42:22.630 --> 01:42:27.670 align:middle line:84% model any of these sources of inertia could be different. 01:42:27.670 --> 01:42:29.390 align:middle line:90% So go ahead with that. 01:42:29.390 --> 01:42:31.350 align:middle line:90% AUDIENCE: That's good to know. 01:42:31.350 --> 01:42:33.110 align:middle line:90% Maybe that they're [INAUDIBLE]. 01:42:33.110 --> 01:42:36.497 align:middle line:90% 01:42:36.497 --> 01:42:38.330 align:middle line:84% HAZHIR RAHMANDAD: Status that is attributed? 01:42:38.330 --> 01:42:43.410 align:middle line:90% OK, good. 01:42:43.410 --> 01:42:46.070 align:middle line:90% 01:42:46.070 --> 01:42:50.690 align:middle line:90% Other sources of inertia? 01:42:50.690 --> 01:42:54.950 align:middle line:90% Anybody anywhere? 01:42:54.950 --> 01:42:58.070 align:middle line:84% AUDIENCE: Probably the dominance of speculative demand 01:42:58.070 --> 01:43:04.280 align:middle line:84% over value based demand, so that whole loop thing. 01:43:04.280 --> 01:43:07.760 align:middle line:84% HAZHIR RAHMANDAD: So where is the source 01:43:07.760 --> 01:43:10.568 align:middle line:84% of inertia within-- because there is this trade-off. 01:43:10.568 --> 01:43:12.360 align:middle line:84% AUDIENCE: The recent rates of price change. 01:43:12.360 --> 01:43:14.680 align:middle line:84% HAZHIR RAHMANDAD: OK, something about-- yes, 01:43:14.680 --> 01:43:20.180 align:middle line:90% perceptions of price change. 01:43:20.180 --> 01:43:20.680 align:middle line:90% Good. 01:43:20.680 --> 01:43:26.160 align:middle line:84% So that would be a space with clear inertia. 01:43:26.160 --> 01:43:30.260 align:middle line:84% AUDIENCE: Maybe a smaller perception inertia-- 01:43:30.260 --> 01:43:34.640 align:middle line:84% Perception of gains, net value gains of [INAUDIBLE]. 01:43:34.640 --> 01:43:44.820 align:middle line:84% HAZHIR RAHMANDAD: Yes, so winning expected, yep. 01:43:44.820 --> 01:43:47.920 align:middle line:90% 01:43:47.920 --> 01:43:48.680 align:middle line:90% Alexander? 01:43:48.680 --> 01:43:50.388 align:middle line:84% AUDIENCE: The perceptions of price change 01:43:50.388 --> 01:43:52.400 align:middle line:84% might actually differ between buyers 01:43:52.400 --> 01:43:54.260 align:middle line:90% of horses and the breeders. 01:43:54.260 --> 01:43:56.760 align:middle line:84% Maybe the breeders are much more risk averse and take longer 01:43:56.760 --> 01:43:58.600 align:middle line:90% to adjust their perception. 01:43:58.600 --> 01:43:59.920 align:middle line:90% HAZHIR RAHMANDAD: Interesting. 01:43:59.920 --> 01:44:08.560 align:middle line:90% So breeders versus buyers. 01:44:08.560 --> 01:44:10.766 align:middle line:90% OK. 01:44:10.766 --> 01:44:12.640 align:middle line:84% And this is actually getting us closer 01:44:12.640 --> 01:44:13.980 align:middle line:90% to the stocks, which is fine. 01:44:13.980 --> 01:44:16.480 align:middle line:84% There is really a continuum going from sources of inertia 01:44:16.480 --> 01:44:17.420 align:middle line:90% to the stocks. 01:44:17.420 --> 01:44:20.320 align:middle line:90% 01:44:20.320 --> 01:44:22.700 align:middle line:84% What else might be relevant here? 01:44:22.700 --> 01:44:25.800 align:middle line:90% 01:44:25.800 --> 01:44:26.400 align:middle line:90% Alban? 01:44:26.400 --> 01:44:28.880 align:middle line:84% AUDIENCE: There can have decision-making processes 01:44:28.880 --> 01:44:32.000 align:middle line:90% that also add inertia. 01:44:32.000 --> 01:44:35.400 align:middle line:84% HAZHIR RAHMANDAD: So the delays in decision making processes. 01:44:35.400 --> 01:44:38.360 align:middle line:90% 01:44:38.360 --> 01:44:40.880 align:middle line:84% Now, you could, again, do the decisions. 01:44:40.880 --> 01:44:44.500 align:middle line:84% Which decision we have in mind may have some differences. 01:44:44.500 --> 01:44:54.660 align:middle line:84% It's breeding decision, investment, leaving the market. 01:44:54.660 --> 01:44:57.820 align:middle line:90% So they are different. 01:44:57.820 --> 01:45:00.360 align:middle line:90% 01:45:00.360 --> 01:45:01.140 align:middle line:90% What else? 01:45:01.140 --> 01:45:12.840 align:middle line:90% 01:45:12.840 --> 01:45:13.680 align:middle line:90% I'm surprised. 01:45:13.680 --> 01:45:17.760 align:middle line:84% There is at least one very big source of inertia 01:45:17.760 --> 01:45:20.500 align:middle line:84% that we don't have on the screen at least. 01:45:20.500 --> 01:45:27.280 align:middle line:90% 01:45:27.280 --> 01:45:30.800 align:middle line:90% What is this market about? 01:45:30.800 --> 01:45:32.720 align:middle line:90% AUDIENCE: [INAUDIBLE] 01:45:32.720 --> 01:45:36.040 align:middle line:84% HAZHIR RAHMANDAD: Yeah, so we need something about horses. 01:45:36.040 --> 01:45:37.300 align:middle line:90% We have the maturity. 01:45:37.300 --> 01:45:38.340 align:middle line:90% That's good. 01:45:38.340 --> 01:45:42.380 align:middle line:90% So that's the calves growing up. 01:45:42.380 --> 01:45:45.600 align:middle line:84% But the horses that are active in the market, 01:45:45.600 --> 01:45:49.460 align:middle line:84% the current horses are a pretty big part of the dynamics, 01:45:49.460 --> 01:45:50.580 align:middle line:90% if you don't have that. 01:45:50.580 --> 01:45:58.120 align:middle line:90% 01:45:58.120 --> 01:46:00.600 align:middle line:84% AUDIENCE: [INAUDIBLE] from the previous explanation 01:46:00.600 --> 01:46:02.680 align:middle line:84% of the difference between sources of inertia 01:46:02.680 --> 01:46:05.200 align:middle line:84% and [INAUDIBLE] because all of these 01:46:05.200 --> 01:46:10.522 align:middle line:84% are almost very simple [INAUDIBLE] a broader idea. 01:46:10.522 --> 01:46:11.480 align:middle line:90% HAZHIR RAHMANDAD: Good. 01:46:11.480 --> 01:46:16.740 align:middle line:84% So it is OK to-- so right now, I have horses in the market. 01:46:16.740 --> 01:46:18.260 align:middle line:90% So that's a source of inertia. 01:46:18.260 --> 01:46:20.040 align:middle line:84% You can think of it as a single stock. 01:46:20.040 --> 01:46:23.520 align:middle line:84% But you could actually break it down based on age 01:46:23.520 --> 01:46:25.640 align:middle line:84% or based on the capability of the horse 01:46:25.640 --> 01:46:29.720 align:middle line:84% or based on-- so those decisions determine the exact stocks you 01:46:29.720 --> 01:46:30.560 align:middle line:90% would have. 01:46:30.560 --> 01:46:33.700 align:middle line:84% And that's where the difference becomes more clear. 01:46:33.700 --> 01:46:36.040 align:middle line:84% AUDIENCE: Is it because you could break down 01:46:36.040 --> 01:46:37.690 align:middle line:90% thoroughbred supply? 01:46:37.690 --> 01:46:38.940 align:middle line:90% HAZHIR RAHMANDAD: For example. 01:46:38.940 --> 01:46:41.148 align:middle line:84% AUDIENCE: Then it's allowed to be a source of inertia 01:46:41.148 --> 01:46:42.443 align:middle line:90% and not just a stock. 01:46:42.443 --> 01:46:43.360 align:middle line:90% HAZHIR RAHMANDAD: Yes. 01:46:43.360 --> 01:46:46.640 align:middle line:84% So thoroughbreds supply-- the word supply is a little bit 01:46:46.640 --> 01:46:47.880 align:middle line:90% ambiguous. 01:46:47.880 --> 01:46:52.860 align:middle line:84% Do you mean the horses that are now on the market for sales? 01:46:52.860 --> 01:46:55.720 align:middle line:84% Or is it actually all of the horses 01:46:55.720 --> 01:46:59.550 align:middle line:84% in the stock of all of the horses or a specific age group? 01:46:59.550 --> 01:47:03.802 align:middle line:84% So as such, I'm OK with thoroughbred supply-- somebody 01:47:03.802 --> 01:47:06.510 align:middle line:84% comes and tells me thoroughbreds supply is the source of inertia. 01:47:06.510 --> 01:47:07.830 align:middle line:90% I wouldn't say it is wrong. 01:47:07.830 --> 01:47:09.497 align:middle line:84% But I would say I like a little bit more 01:47:09.497 --> 01:47:12.010 align:middle line:84% clarity on what you have in mind. 01:47:12.010 --> 01:47:14.470 align:middle line:84% If you bring that as a stock, I would say, 01:47:14.470 --> 01:47:16.770 align:middle line:84% no, it's not going to work as a stock. 01:47:16.770 --> 01:47:18.430 align:middle line:90% It's going to be too confusing. 01:47:18.430 --> 01:47:20.090 align:middle line:90% The definition is not clear. 01:47:20.090 --> 01:47:23.190 align:middle line:84% We need a more clear definition to get it in place. 01:47:23.190 --> 01:47:26.330 align:middle line:90% 01:47:26.330 --> 01:47:27.410 align:middle line:90% Anything else? 01:47:27.410 --> 01:47:32.610 align:middle line:90% 01:47:32.610 --> 01:47:34.590 align:middle line:84% AUDIENCE: Can I ask a quick question? 01:47:34.590 --> 01:47:36.730 align:middle line:90% HAZHIR RAHMANDAD: Sure, please. 01:47:36.730 --> 01:47:37.410 align:middle line:90% AUDIENCE: Yep. 01:47:37.410 --> 01:47:40.910 align:middle line:84% So right now, we have a causal loop diagram. 01:47:40.910 --> 01:47:45.090 align:middle line:84% And that's what we're using to think about sources of inertia. 01:47:45.090 --> 01:47:49.490 align:middle line:84% But at what stage in the modeling process 01:47:49.490 --> 01:47:52.530 align:middle line:90% are you supposed to do this? 01:47:52.530 --> 01:47:55.110 align:middle line:84% Is it at the problem definition stage? 01:47:55.110 --> 01:47:57.170 align:middle line:84% Or would you have already thought 01:47:57.170 --> 01:48:00.110 align:middle line:84% about variables and causal loop diagrams? 01:48:00.110 --> 01:48:03.770 align:middle line:84% It seems to me that we should have thought about this before. 01:48:03.770 --> 01:48:05.690 align:middle line:84% HAZHIR RAHMANDAD: So in the stage 01:48:05.690 --> 01:48:07.830 align:middle line:84% that you are building causal loop diagrams, 01:48:07.830 --> 01:48:10.610 align:middle line:84% sometimes it is helpful to think about sources of inertia 01:48:10.610 --> 01:48:16.830 align:middle line:84% and stocks, especially when the problems are very physics based. 01:48:16.830 --> 01:48:20.370 align:middle line:84% So if I'm building a model of supply chains, 01:48:20.370 --> 01:48:23.570 align:middle line:84% I actually prefer to start with clear stocks 01:48:23.570 --> 01:48:25.670 align:middle line:90% that I'm thinking about. 01:48:25.670 --> 01:48:28.570 align:middle line:84% In my causal loop diagram, I may put stocks and flows 01:48:28.570 --> 01:48:31.290 align:middle line:84% in the causal loop diagram in the first place. 01:48:31.290 --> 01:48:35.130 align:middle line:84% For this problem, I don't see the stocks 01:48:35.130 --> 01:48:39.430 align:middle line:84% as indispensable for the early stage conceptualization. 01:48:39.430 --> 01:48:41.350 align:middle line:84% So in building the causal loop diagram, 01:48:41.350 --> 01:48:43.410 align:middle line:84% I'm very comfortable not worrying too much 01:48:43.410 --> 01:48:48.410 align:middle line:84% about the stocks yet and bringing them 01:48:48.410 --> 01:48:51.090 align:middle line:84% on when I get into the quantitative modeling 01:48:51.090 --> 01:48:52.690 align:middle line:84% or thinking about conceptualizing 01:48:52.690 --> 01:48:55.570 align:middle line:84% the quantitative model in the state. 01:48:55.570 --> 01:48:59.330 align:middle line:84% Here, we skipped all of the causal loop 01:48:59.330 --> 01:49:00.830 align:middle line:90% diagram, qualitative modeling. 01:49:00.830 --> 01:49:02.550 align:middle line:90% We didn't talk about boundary. 01:49:02.550 --> 01:49:05.730 align:middle line:84% We didn't talk about these mechanisms 01:49:05.730 --> 01:49:06.950 align:middle line:90% where did they come from. 01:49:06.950 --> 01:49:09.970 align:middle line:84% We took them for granted and started from that point. 01:49:09.970 --> 01:49:12.738 align:middle line:84% In an actual modeling exercise, you actually 01:49:12.738 --> 01:49:15.030 align:middle line:84% have to go through the whole process of building these. 01:49:15.030 --> 01:49:17.670 align:middle line:90% And it is not an easy process. 01:49:17.670 --> 01:49:20.210 align:middle line:84% But it is an easier process compared 01:49:20.210 --> 01:49:24.910 align:middle line:84% to building the quantitative model, in most cases. 01:49:24.910 --> 01:49:28.850 align:middle line:90% 01:49:28.850 --> 01:49:32.130 align:middle line:84% OK, any other sources of inertia? 01:49:32.130 --> 01:49:34.170 align:middle line:84% And we could actually start now shifting 01:49:34.170 --> 01:49:35.710 align:middle line:90% into discussing the stocks. 01:49:35.710 --> 01:49:38.510 align:middle line:84% So with these sources of inertia in mind-- 01:49:38.510 --> 01:49:43.130 align:middle line:84% and as we break them down, we could actually-- 01:49:43.130 --> 01:49:47.270 align:middle line:84% now, we go into stock variables, if you want. 01:49:47.270 --> 01:49:49.270 align:middle line:84% Now, for this, we need a clear audience. 01:49:49.270 --> 01:49:51.410 align:middle line:84% We can't continue discussing this 01:49:51.410 --> 01:49:53.530 align:middle line:84% without having an audience in mind. 01:49:53.530 --> 01:49:56.730 align:middle line:84% Let's focus, for now, on the theoretical model 01:49:56.730 --> 01:49:59.370 align:middle line:84% for a sociology journal, just in the interest 01:49:59.370 --> 01:50:03.970 align:middle line:84% of time and level of detail that we can manage right now. 01:50:03.970 --> 01:50:08.050 align:middle line:84% So with that, come up with the stocks. 01:50:08.050 --> 01:50:13.290 align:middle line:84% And we can add those stocks in the same diagram, 01:50:13.290 --> 01:50:14.390 align:middle line:90% in the same list. 01:50:14.390 --> 01:50:16.170 align:middle line:84% Basically, under each of these, we 01:50:16.170 --> 01:50:21.230 align:middle line:84% could come up with stocks that we think are helpful. 01:50:21.230 --> 01:50:24.570 align:middle line:84% So for horse maturity delay, what 01:50:24.570 --> 01:50:31.516 align:middle line:84% would be the right stock or stocks? 01:50:31.516 --> 01:50:32.892 align:middle line:90% AUDIENCE: [INAUDIBLE] 01:50:32.892 --> 01:50:34.350 align:middle line:84% HAZHIR RAHMANDAD: Can you speak up? 01:50:34.350 --> 01:50:35.490 align:middle line:90% Yeah, Maria? 01:50:35.490 --> 01:50:37.975 align:middle line:84% AUDIENCE: A population model for the horses 01:50:37.975 --> 01:50:38.850 align:middle line:90% HAZHIR RAHMANDAD: OK. 01:50:38.850 --> 01:50:45.110 align:middle line:84% AUDIENCE: Maybe we want to separate the females [INAUDIBLE] 01:50:45.110 --> 01:50:46.423 align:middle line:90% and then also by age. 01:50:46.423 --> 01:50:47.590 align:middle line:90% HAZHIR RAHMANDAD: Very good. 01:50:47.590 --> 01:50:51.340 align:middle line:84% So probably, you could actually use a stock and flow 01:50:51.340 --> 01:50:57.700 align:middle line:84% chain for horses that is separated by age-- 01:50:57.700 --> 01:51:00.900 align:middle line:90% that's pretty useful to have-- 01:51:00.900 --> 01:51:04.000 align:middle line:90% and by male versus female. 01:51:04.000 --> 01:51:06.240 align:middle line:84% So that's probably also relevant, 01:51:06.240 --> 01:51:09.220 align:middle line:84% especially on the breeding side of the story. 01:51:09.220 --> 01:51:12.280 align:middle line:84% I'm not sure if it matters as much about racing. 01:51:12.280 --> 01:51:15.840 align:middle line:84% I'm not an expert in the horse racing. 01:51:15.840 --> 01:51:19.913 align:middle line:90% 01:51:19.913 --> 01:51:22.080 align:middle line:84% So now, let's think about that stock and flow chain. 01:51:22.080 --> 01:51:25.000 align:middle line:84% I think that's useful because we are talking about horses. 01:51:25.000 --> 01:51:27.260 align:middle line:84% We need a stock and flow of horses. 01:51:27.260 --> 01:51:32.140 align:middle line:84% So what's a good level of breaking 01:51:32.140 --> 01:51:35.600 align:middle line:84% this down for the audience that we talked about? 01:51:35.600 --> 01:51:38.860 align:middle line:90% That is, a sociology journal. 01:51:38.860 --> 01:51:40.320 align:middle line:90% How would you break it down? 01:51:40.320 --> 01:51:43.460 align:middle line:90% 01:51:43.460 --> 01:51:45.060 align:middle line:90% Jason, what are you thinking? 01:51:45.060 --> 01:51:46.560 align:middle line:84% AUDIENCE: I don't know about horses, 01:51:46.560 --> 01:51:52.620 align:middle line:84% but one useful separation might be horses for racing, 01:51:52.620 --> 01:51:57.340 align:middle line:84% and then horses for breeding if they age into that 01:51:57.340 --> 01:52:00.660 align:middle line:90% or the other way around. 01:52:00.660 --> 01:52:04.740 align:middle line:84% I wouldn't go much more disaggregate 01:52:04.740 --> 01:52:11.500 align:middle line:84% than that for a general insights model. 01:52:11.500 --> 01:52:13.860 align:middle line:84% Even separating by gender might be too much, 01:52:13.860 --> 01:52:19.060 align:middle line:84% and then having to capture the pairing process, 01:52:19.060 --> 01:52:24.720 align:middle line:84% unless it's important for the dynamics of speculation 01:52:24.720 --> 01:52:26.700 align:middle line:90% we're trying to capture. 01:52:26.700 --> 01:52:28.620 align:middle line:84% So assuming gender isn't important for 01:52:28.620 --> 01:52:36.180 align:middle line:84% that, I would say just a one or two-stock aging chain 01:52:36.180 --> 01:52:37.893 align:middle line:90% might be enough. 01:52:37.893 --> 01:52:40.060 align:middle line:84% HAZHIR RAHMANDAD: Jake, you wanted to add something? 01:52:40.060 --> 01:52:41.727 align:middle line:84% AUDIENCE: I was just going to explicitly 01:52:41.727 --> 01:52:49.713 align:middle line:84% say [INAUDIBLE] juvenile stage [INAUDIBLE] retirement stage. 01:52:49.713 --> 01:52:50.880 align:middle line:90% HAZHIR RAHMANDAD: Very good. 01:52:50.880 --> 01:52:53.860 align:middle line:84% So basically, you are speculating 01:52:53.860 --> 01:52:57.580 align:middle line:84% three stocks-- juvenile, racing age, 01:52:57.580 --> 01:53:00.980 align:middle line:90% and retirement age of sorts. 01:53:00.980 --> 01:53:05.620 align:middle line:84% In fact, it's not obvious right now how much that retirement age 01:53:05.620 --> 01:53:08.020 align:middle line:84% stock is critical for the dynamics 01:53:08.020 --> 01:53:12.340 align:middle line:84% because it probably doesn't play a big role in the price 01:53:12.340 --> 01:53:14.460 align:middle line:90% speculation parts. 01:53:14.460 --> 01:53:17.560 align:middle line:84% It may play a role in the breeding part of the story. 01:53:17.560 --> 01:53:20.740 align:middle line:90% That's an interesting question. 01:53:20.740 --> 01:53:23.500 align:middle line:84% Where you put the thresholds, I don't know 01:53:23.500 --> 01:53:25.180 align:middle line:90% what's the right choice there. 01:53:25.180 --> 01:53:28.500 align:middle line:84% But I like the more aggregate version 01:53:28.500 --> 01:53:32.800 align:middle line:84% for a model for a theoretical sociology journal. 01:53:32.800 --> 01:53:35.620 align:middle line:84% You can't afford to have too many stocks. 01:53:35.620 --> 01:53:38.920 align:middle line:84% So if you can bring down the stocks of horses to two-- 01:53:38.920 --> 01:53:42.100 align:middle line:84% that is, horses that are juvenile 01:53:42.100 --> 01:53:46.220 align:middle line:84% versus horses that are in the racing age, adult 01:53:46.220 --> 01:53:49.480 align:middle line:84% or whatever you call it-- that's probably all you need. 01:53:49.480 --> 01:53:52.140 align:middle line:90% 01:53:52.140 --> 01:53:56.900 align:middle line:84% Male versus female, important for a more detailed model, 01:53:56.900 --> 01:53:58.660 align:middle line:90% not that important here. 01:53:58.660 --> 01:54:00.820 align:middle line:90% There is an issue. 01:54:00.820 --> 01:54:05.780 align:middle line:84% When you have some horse in the breeding stock, 01:54:05.780 --> 01:54:06.800 align:middle line:90% they are not racing. 01:54:06.800 --> 01:54:10.300 align:middle line:84% So there is an impact on that market. 01:54:10.300 --> 01:54:16.300 align:middle line:84% But that impact is in the months range, in terms of its time 01:54:16.300 --> 01:54:17.220 align:middle line:90% constant. 01:54:17.220 --> 01:54:21.340 align:middle line:84% So you could use a fraction of the total horses 01:54:21.340 --> 01:54:26.120 align:middle line:84% in their prime age that is allocated to breeding. 01:54:26.120 --> 01:54:29.840 align:middle line:84% That fraction, by definition, cannot be more than 50%. 01:54:29.840 --> 01:54:33.100 align:middle line:84% But then that allows you to capture 01:54:33.100 --> 01:54:36.597 align:middle line:84% what is the number of horses that are in the racing market 01:54:36.597 --> 01:54:38.180 align:middle line:84% and what are the number of horses that 01:54:38.180 --> 01:54:41.580 align:middle line:84% are used for breeding, which you do need for calculating 01:54:41.580 --> 01:54:46.310 align:middle line:84% breeding rates and the inflow into this stock and flow chain. 01:54:46.310 --> 01:54:49.590 align:middle line:84% So for that audience, you can see really 01:54:49.590 --> 01:54:53.430 align:middle line:84% a couple of stocks might be all you need. 01:54:53.430 --> 01:54:56.870 align:middle line:84% Maybe even somebody would go with a single stock. 01:54:56.870 --> 01:54:59.110 align:middle line:84% Because of the delays involved and the importance 01:54:59.110 --> 01:55:01.610 align:middle line:84% of those delays in the supply side, 01:55:01.610 --> 01:55:03.570 align:middle line:84% I would not go into a single stock. 01:55:03.570 --> 01:55:04.810 align:middle line:90% I'll keep a separate one. 01:55:04.810 --> 01:55:06.477 align:middle line:84% AUDIENCE: Just a question that popped up 01:55:06.477 --> 01:55:10.750 align:middle line:84% when you were talking about including a parameter to exclude 01:55:10.750 --> 01:55:13.230 align:middle line:90% some fractions for breeding-- 01:55:13.230 --> 01:55:15.350 align:middle line:84% what kind of thought process would 01:55:15.350 --> 01:55:17.190 align:middle line:84% go into deciding whether that's going 01:55:17.190 --> 01:55:19.708 align:middle line:84% to be a parameter or something you choose endogenously? 01:55:19.708 --> 01:55:21.250 align:middle line:84% HAZHIR RAHMANDAD: Very good question. 01:55:21.250 --> 01:55:24.510 align:middle line:84% So this is an architectural choice 01:55:24.510 --> 01:55:27.090 align:middle line:84% I'm now imposing here on my thinking. 01:55:27.090 --> 01:55:29.410 align:middle line:84% And that's not an obvious architectural choice. 01:55:29.410 --> 01:55:33.870 align:middle line:84% The choice is we could have a separate stock for horses 01:55:33.870 --> 01:55:40.250 align:middle line:84% that are in the breeding stock versus horses that are not. 01:55:40.250 --> 01:55:41.770 align:middle line:90% And then horses can go there. 01:55:41.770 --> 01:55:45.750 align:middle line:84% And once you're in the breeding stock-- this is for mares-- 01:55:45.750 --> 01:55:48.850 align:middle line:84% it would be a few months that you wouldn't be racing. 01:55:48.850 --> 01:55:54.710 align:middle line:84% And at the end of that, you can come back into the racing stock 01:55:54.710 --> 01:56:00.550 align:middle line:84% or remain and go back to the breeding stock again. 01:56:00.550 --> 01:56:08.950 align:middle line:84% The choice there for me is how fast those dynamics are. 01:56:08.950 --> 01:56:11.710 align:middle line:84% So if the time horizon we are looking at 01:56:11.710 --> 01:56:17.670 align:middle line:84% is about 20 years for this market to unfold completely, 01:56:17.670 --> 01:56:23.670 align:middle line:84% a few months' delays here is not as big of a deal. 01:56:23.670 --> 01:56:28.310 align:middle line:84% So I feel comfortable regulating that with an auxiliary variable 01:56:28.310 --> 01:56:30.690 align:middle line:84% that can go up and down relatively fast 01:56:30.690 --> 01:56:32.470 align:middle line:84% because that auxiliary variable-- that is, 01:56:32.470 --> 01:56:34.510 align:middle line:84% the breeding fraction-- is determined 01:56:34.510 --> 01:56:37.430 align:middle line:84% by a lot of other things about perceptions 01:56:37.430 --> 01:56:40.990 align:middle line:84% of whether, in future, it will be a good market 01:56:40.990 --> 01:56:42.390 align:middle line:90% to invest in and so on. 01:56:42.390 --> 01:56:52.550 align:middle line:84% So with that in mind, I am OK with making it determined 01:56:52.550 --> 01:56:55.470 align:middle line:84% by an auxiliary variable, keeping in mind that I have 01:56:55.470 --> 01:56:58.950 align:middle line:84% other delays in the perception and decision making process 01:56:58.950 --> 01:57:02.690 align:middle line:84% that can carry some of that inertia for me. 01:57:02.690 --> 01:57:05.170 align:middle line:84% So I don't need to explicitly capture that. 01:57:05.170 --> 01:57:07.770 align:middle line:84% So this is how you think about it. 01:57:07.770 --> 01:57:12.310 align:middle line:84% That is, OK, in this case, this auxiliary variable, yeah time 01:57:12.310 --> 01:57:14.410 align:middle line:84% constant of months is not irrelevant. 01:57:14.410 --> 01:57:15.930 align:middle line:90% It is potentially relevant. 01:57:15.930 --> 01:57:19.150 align:middle line:84% But it can already be captured in some other delays 01:57:19.150 --> 01:57:21.910 align:middle line:84% to some extent about how breeders 01:57:21.910 --> 01:57:23.690 align:middle line:90% update their perception. 01:57:23.690 --> 01:57:26.690 align:middle line:84% So their decision has its own inertia. 01:57:26.690 --> 01:57:28.290 align:middle line:84% I don't need a separate stock for it. 01:57:28.290 --> 01:57:30.970 align:middle line:84% That's how I'm thinking about it. 01:57:30.970 --> 01:57:31.470 align:middle line:90% Good. 01:57:31.470 --> 01:57:35.990 align:middle line:84% So that's for a stock and flow of horses. 01:57:35.990 --> 01:57:37.050 align:middle line:90% Other stocks? 01:57:37.050 --> 01:57:55.390 align:middle line:90% 01:57:55.390 --> 01:58:05.592 align:middle line:84% AUDIENCE: Expected price, yeah, over some expected future price. 01:58:05.592 --> 01:58:06.550 align:middle line:90% HAZHIR RAHMANDAD: Yeah. 01:58:06.550 --> 01:58:10.150 align:middle line:84% So basically, you need stocks for capturing price expectation 01:58:10.150 --> 01:58:11.550 align:middle line:90% formation. 01:58:11.550 --> 01:58:16.230 align:middle line:84% Now, if you are familiar with trend function or variants 01:58:16.230 --> 01:58:20.070 align:middle line:84% of that, you would even know exactly what chunks, what 01:58:20.070 --> 01:58:22.870 align:middle line:84% molecule will fit here in your modeling. 01:58:22.870 --> 01:58:24.450 align:middle line:84% And using that, you can actually say, 01:58:24.450 --> 01:58:26.010 align:middle line:90% OK, this is a trend function. 01:58:26.010 --> 01:58:29.070 align:middle line:84% It just tracks What's the rate of growth 01:58:29.070 --> 01:58:31.410 align:middle line:84% in the underlying concept-- in this case, 01:58:31.410 --> 01:58:33.350 align:middle line:84% the underlying concept being the price. 01:58:33.350 --> 01:58:36.880 align:middle line:84% And the rate of change in price, in this case, 01:58:36.880 --> 01:58:40.400 align:middle line:90% would be determined by-- 01:58:40.400 --> 01:58:43.960 align:middle line:84% I mean, that rate of change requires usually two stock 01:58:43.960 --> 01:58:46.640 align:middle line:90% variables to get it well. 01:58:46.640 --> 01:58:49.020 align:middle line:84% You could go with one stock variable as well. 01:58:49.020 --> 01:58:52.060 align:middle line:84% But then it is not the best, cleanest formulation. 01:58:52.060 --> 01:58:55.560 align:middle line:90% So that's a couple of stocks. 01:58:55.560 --> 01:58:58.320 align:middle line:84% But because it is a very standard formulation, 01:58:58.320 --> 01:59:01.300 align:middle line:84% you don't need to worry about explaining it to the audiences. 01:59:01.300 --> 01:59:05.880 align:middle line:84% So I'm not too worried about adding two stocks here 01:59:05.880 --> 01:59:08.300 align:middle line:84% for a trend function because you just say, 01:59:08.300 --> 01:59:10.600 align:middle line:84% I'm looking at the trend of this underlying thing 01:59:10.600 --> 01:59:14.520 align:middle line:84% and everybody follows what's happening. 01:59:14.520 --> 01:59:15.560 align:middle line:90% Good. 01:59:15.560 --> 01:59:19.808 align:middle line:84% Other things that might be relevant? 01:59:19.808 --> 01:59:20.800 align:middle line:90% AUDIENCE: [INAUDIBLE] 01:59:20.800 --> 01:59:22.675 align:middle line:84% HAZHIR RAHMANDAD: Something about the status. 01:59:22.675 --> 01:59:25.240 align:middle line:84% So these status perceptions, they have big lags. 01:59:25.240 --> 01:59:28.600 align:middle line:90% They move relatively slowly. 01:59:28.600 --> 01:59:31.840 align:middle line:84% So I would definitely have at least one stock and probably 01:59:31.840 --> 01:59:35.700 align:middle line:84% just one stock for this level of detail, not anymore, 01:59:35.700 --> 01:59:39.200 align:middle line:84% and a status value that is ascribed 01:59:39.200 --> 01:59:42.780 align:middle line:90% to having a thoroughbred horse. 01:59:42.780 --> 01:59:45.240 align:middle line:84% Now, the status, we need to be a little bit 01:59:45.240 --> 01:59:48.520 align:middle line:84% careful about this definition as well because it's not 01:59:48.520 --> 01:59:50.120 align:middle line:90% a very well defined thing. 01:59:50.120 --> 01:59:54.480 align:middle line:84% And later on, I'll need to use the status signal 01:59:54.480 --> 01:59:58.000 align:middle line:84% to calculate something about demand or drive demand 01:59:58.000 --> 01:59:59.580 align:middle line:90% or put a pressure on demand. 01:59:59.580 --> 02:00:04.520 align:middle line:90% 02:00:04.520 --> 02:00:07.100 align:middle line:84% For now, I would call it a single stock. 02:00:07.100 --> 02:00:10.380 align:middle line:84% But I also would have this worry in the back of my mind. 02:00:10.380 --> 02:00:13.880 align:middle line:84% How do I want to make this very qualitative, 02:00:13.880 --> 02:00:17.800 align:middle line:84% underdefined concept-- how do I turn it into something that 02:00:17.800 --> 02:00:19.920 align:middle line:90% plugs into my demand function? 02:00:19.920 --> 02:00:23.120 align:middle line:84% That, I don't have right now a solution for it, just by quick 02:00:23.120 --> 02:00:23.620 align:middle line:90% look at it. 02:00:23.620 --> 02:00:26.220 align:middle line:84% I need to think more clearly about the formulations. 02:00:26.220 --> 02:00:28.460 align:middle line:84% But that's not a problem to solve right now. 02:00:28.460 --> 02:00:33.080 align:middle line:84% It's OK to leave that a little bit ambiguous. 02:00:33.080 --> 02:00:35.680 align:middle line:84% AUDIENCE: So just going back to the idea of using stocks 02:00:35.680 --> 02:00:40.620 align:middle line:84% for the trend in price, just as a question on presentation, 02:00:40.620 --> 02:00:42.900 align:middle line:84% obviously sociologists, in this case, 02:00:42.900 --> 02:00:44.573 align:middle line:84% would be able to follow what a trend is. 02:00:44.573 --> 02:00:46.240 align:middle line:84% But they probably wouldn't know the math 02:00:46.240 --> 02:00:47.740 align:middle line:90% behind us calculating this. 02:00:47.740 --> 02:00:49.598 align:middle line:84% So would you put the math in the main paper? 02:00:49.598 --> 02:00:51.140 align:middle line:84% Or would you have it in the appendix? 02:00:51.140 --> 02:00:52.600 align:middle line:90% Would you just say [INAUDIBLE]? 02:00:52.600 --> 02:00:53.517 align:middle line:90% HAZHIR RAHMANDAD: Yes. 02:00:53.517 --> 02:00:57.400 align:middle line:84% So for this, I would say, trend function, Sterman 2000 02:00:57.400 --> 02:00:59.800 align:middle line:84% or whatever is the reference I want to use. 02:00:59.800 --> 02:01:01.960 align:middle line:84% And then I would have a technical appendix 02:01:01.960 --> 02:01:05.700 align:middle line:84% that is online and has all of the model equations. 02:01:05.700 --> 02:01:07.340 align:middle line:90% So it's fully documented. 02:01:07.340 --> 02:01:10.200 align:middle line:84% You also would include your actual simulation model 02:01:10.200 --> 02:01:12.140 align:middle line:90% with replication instructions. 02:01:12.140 --> 02:01:15.280 align:middle line:84% So that way, the people who are interested in the details 02:01:15.280 --> 02:01:16.900 align:middle line:84% can check it out and figure it out. 02:01:16.900 --> 02:01:19.000 align:middle line:84% But you are not wasting time and also 02:01:19.000 --> 02:01:21.320 align:middle line:84% creating more confusion and questions 02:01:21.320 --> 02:01:23.260 align:middle line:90% within the text of your paper. 02:01:23.260 --> 02:01:25.920 align:middle line:90% 02:01:25.920 --> 02:01:29.520 align:middle line:84% OK, so I think those actually are probably covering 02:01:29.520 --> 02:01:31.060 align:middle line:90% the most important pieces. 02:01:31.060 --> 02:01:33.920 align:middle line:84% There is a question on the price itself. 02:01:33.920 --> 02:01:36.760 align:middle line:84% That is, when I'm thinking of formulating price, 02:01:36.760 --> 02:01:39.480 align:middle line:84% the most standard formulation of price 02:01:39.480 --> 02:01:42.840 align:middle line:84% has a stock in it, which is basically 02:01:42.840 --> 02:01:45.040 align:middle line:90% its current price as a stock. 02:01:45.040 --> 02:01:47.280 align:middle line:84% And then the indicated price changes 02:01:47.280 --> 02:01:50.440 align:middle line:84% based on the pressures of supply and demand. 02:01:50.440 --> 02:01:54.200 align:middle line:84% And if you have more demand than currently there 02:01:54.200 --> 02:01:57.260 align:middle line:84% are horses-- and right now, I now know what is supply. 02:01:57.260 --> 02:01:59.400 align:middle line:84% I have the horses that are in the market 02:01:59.400 --> 02:02:03.240 align:middle line:84% and perhaps with an adjustment for the supply chain-- that 02:02:03.240 --> 02:02:07.740 align:middle line:84% is, the juvenile horses, the calves added to it. 02:02:07.740 --> 02:02:09.800 align:middle line:84% That gives me the expected supply 02:02:09.800 --> 02:02:14.700 align:middle line:84% in the marketplace that may put pressure on the price. 02:02:14.700 --> 02:02:16.580 align:middle line:84% And then the demand side is less clear. 02:02:16.580 --> 02:02:20.720 align:middle line:84% Right now, I don't have a clear formulation for that. 02:02:20.720 --> 02:02:27.560 align:middle line:84% But usually, to formulate that, I would also think about one 02:02:27.560 --> 02:02:29.650 align:middle line:84% of these anchor and adjustment types of things. 02:02:29.650 --> 02:02:32.510 align:middle line:84% That is, this demand is coming from what? 02:02:32.510 --> 02:02:36.790 align:middle line:84% From a lot of qualitative factors about status, 02:02:36.790 --> 02:02:40.070 align:middle line:84% about expected growth in the prices, and so on. 02:02:40.070 --> 02:02:43.890 align:middle line:84% So I can create a stock variable that 02:02:43.890 --> 02:02:46.930 align:middle line:90% is the anchor for the demand. 02:02:46.930 --> 02:02:50.050 align:middle line:84% And it adjusts towards these forces. 02:02:50.050 --> 02:02:53.810 align:middle line:84% That is, the status can push this demand up. 02:02:53.810 --> 02:02:58.090 align:middle line:84% The increasing trend in price can push this demand up, 02:02:58.090 --> 02:03:00.310 align:middle line:90% but also can bring it down. 02:03:00.310 --> 02:03:03.210 align:middle line:84% So that way, there would be another stock 02:03:03.210 --> 02:03:05.810 align:middle line:90% for modeling the demand. 02:03:05.810 --> 02:03:10.350 align:middle line:84% But that is a little bit less clear looking at this diagram. 02:03:10.350 --> 02:03:12.330 align:middle line:84% It is a choice about how I model demand 02:03:12.330 --> 02:03:15.650 align:middle line:90% that will determine that stock. 02:03:15.650 --> 02:03:18.530 align:middle line:84% So you may not know about it until you get 02:03:18.530 --> 02:03:19.830 align:middle line:90% into the modeling of demand. 02:03:19.830 --> 02:03:22.570 align:middle line:84% And you say, oh OK, the best way to model demand 02:03:22.570 --> 02:03:24.770 align:middle line:84% here would be an anchor and adjustment 02:03:24.770 --> 02:03:27.570 align:middle line:90% type of a formulation. 02:03:27.570 --> 02:03:29.890 align:middle line:84% AUDIENCE: What about something about the money 02:03:29.890 --> 02:03:33.853 align:middle line:84% available for investing in horses? 02:03:33.853 --> 02:03:34.770 align:middle line:90% HAZHIR RAHMANDAD: Yes. 02:03:34.770 --> 02:03:39.150 align:middle line:84% So the money available is actually a physical thing. 02:03:39.150 --> 02:03:41.610 align:middle line:84% It's a good candidate for a stock. 02:03:41.610 --> 02:03:44.210 align:middle line:84% Money has the features of stock variables. 02:03:44.210 --> 02:03:47.210 align:middle line:84% But in these dynamics, we should ask, 02:03:47.210 --> 02:03:48.750 align:middle line:90% does it make a big difference? 02:03:48.750 --> 02:03:50.930 align:middle line:84% And it would have made a difference 02:03:50.930 --> 02:03:53.010 align:middle line:84% if this was the only good that people 02:03:53.010 --> 02:03:55.270 align:middle line:90% were investing in and trading. 02:03:55.270 --> 02:03:58.850 align:middle line:84% And therefore, they might put all of their savings or not. 02:03:58.850 --> 02:04:01.350 align:middle line:84% But for the market we are looking at, 02:04:01.350 --> 02:04:04.290 align:middle line:84% the money that goes into the racing horses 02:04:04.290 --> 02:04:08.190 align:middle line:84% is tiny compared to the overall investments in this market. 02:04:08.190 --> 02:04:10.330 align:middle line:84% So I wouldn't be worried about money 02:04:10.330 --> 02:04:14.650 align:middle line:84% becoming a binding constraint in any of these interactions. 02:04:14.650 --> 02:04:16.190 align:middle line:90% So I would just leave that out. 02:04:16.190 --> 02:04:18.945 align:middle line:90% I'm not worried about that. 02:04:18.945 --> 02:04:21.570 align:middle line:84% AUDIENCE: I feel like, I mean, something that I was thinking of 02:04:21.570 --> 02:04:24.090 align:middle line:84% is, would you want to have a stock between people who 02:04:24.090 --> 02:04:27.810 align:middle line:84% are currently owners of horses and those 02:04:27.810 --> 02:04:30.450 align:middle line:84% who are not but are considering it? 02:04:30.450 --> 02:04:34.210 align:middle line:84% And then you have word of mouth effect 02:04:34.210 --> 02:04:37.610 align:middle line:84% that the status propagates to the people who 02:04:37.610 --> 02:04:38.360 align:middle line:90% don't have horses. 02:04:38.360 --> 02:04:40.193 align:middle line:84% HAZHIR RAHMANDAD: So that's a good question. 02:04:40.193 --> 02:04:42.130 align:middle line:84% The way I'm thinking about that is demand 02:04:42.130 --> 02:04:43.830 align:middle line:90% is essentially that population. 02:04:43.830 --> 02:04:45.930 align:middle line:84% So if I'm modeling demand as number 02:04:45.930 --> 02:04:49.050 align:middle line:84% of people who have horses or own horses 02:04:49.050 --> 02:04:52.090 align:middle line:84% or want to own horses or something like that, 02:04:52.090 --> 02:04:54.530 align:middle line:84% it will be-- because the horses that are owned 02:04:54.530 --> 02:04:56.750 align:middle line:84% are already in the stock and flow for ourselves. 02:04:56.750 --> 02:04:58.730 align:middle line:84% So we don't need a separate thing for it. 02:04:58.730 --> 02:05:01.650 align:middle line:84% The number of people who want to have horses 02:05:01.650 --> 02:05:04.870 align:middle line:90% becomes the demand function. 02:05:04.870 --> 02:05:07.530 align:middle line:84% Then there is the question do I want 02:05:07.530 --> 02:05:12.570 align:middle line:84% to capture the word of mouth type of dynamics here. 02:05:12.570 --> 02:05:15.570 align:middle line:84% I'm not seeing it in our diagram. 02:05:15.570 --> 02:05:17.610 align:middle line:84% There is a question of how fast people 02:05:17.610 --> 02:05:20.850 align:middle line:90% learn about the price changes. 02:05:20.850 --> 02:05:25.540 align:middle line:90% It's not hard to see that. 02:05:25.540 --> 02:05:28.240 align:middle line:84% Information is not the bottleneck in a sense here. 02:05:28.240 --> 02:05:32.400 align:middle line:84% That is, for people who might be into this, 02:05:32.400 --> 02:05:35.640 align:middle line:84% they probably will get their information if they want it. 02:05:35.640 --> 02:05:40.500 align:middle line:84% So the explicit word of mouth Bass diffusion type of thing 02:05:40.500 --> 02:05:43.740 align:middle line:84% is not, in my view, of high priority. 02:05:43.740 --> 02:05:45.380 align:middle line:84% It's actually one of those structures 02:05:45.380 --> 02:05:48.680 align:middle line:84% that drives a strong behavior inside the model. 02:05:48.680 --> 02:05:52.820 align:middle line:84% So if you want to put a Bass diffusion type of model 02:05:52.820 --> 02:05:54.940 align:middle line:84% into your formulations, really think hard 02:05:54.940 --> 02:05:57.980 align:middle line:84% because it will take over a whole structure in terms 02:05:57.980 --> 02:05:59.140 align:middle line:90% of dynamics. 02:05:59.140 --> 02:06:02.760 align:middle line:84% And if it is not the right fit for that problem, 02:06:02.760 --> 02:06:06.180 align:middle line:84% it will take your dynamics and your modeling in a direction 02:06:06.180 --> 02:06:07.840 align:middle line:90% that you may not actually want. 02:06:07.840 --> 02:06:12.440 align:middle line:84% So Bass diffusion, very solid, interesting classical model, 02:06:12.440 --> 02:06:17.140 align:middle line:84% but it is not a hammer to use for every nail. 02:06:17.140 --> 02:06:20.340 align:middle line:84% You need to be using it carefully. 02:06:20.340 --> 02:06:24.220 align:middle line:84% OK, so I think this is a good conversation for what time 02:06:24.220 --> 02:06:26.280 align:middle line:90% we have for the horse market. 02:06:26.280 --> 02:06:28.860 align:middle line:84% Let's switch to the next one, which 02:06:28.860 --> 02:06:38.300 align:middle line:84% is looking at congestion and transportation. 02:06:38.300 --> 02:06:44.900 align:middle line:84% So this model is looking at dynamics of congestion 02:06:44.900 --> 02:06:47.660 align:middle line:90% in an urban environment. 02:06:47.660 --> 02:06:50.580 align:middle line:84% So think of travel time as your measure 02:06:50.580 --> 02:06:53.700 align:middle line:84% of congestion, which impacts attractiveness 02:06:53.700 --> 02:06:58.940 align:middle line:84% of driving and pressure to reduce congestion. 02:06:58.940 --> 02:07:01.540 align:middle line:84% It is driven by the capacity of highways 02:07:01.540 --> 02:07:03.980 align:middle line:90% and the volume of traffic. 02:07:03.980 --> 02:07:06.940 align:middle line:84% And then these generate a variety of loops. 02:07:06.940 --> 02:07:09.600 align:middle line:84% On the supply side of highway capacity, 02:07:09.600 --> 02:07:11.060 align:middle line:84% the pressure to reduce congestion 02:07:11.060 --> 02:07:15.620 align:middle line:84% may lead to road construction, increases highway capacity. 02:07:15.620 --> 02:07:18.960 align:middle line:84% And then travel time impacts attractiveness of driving. 02:07:18.960 --> 02:07:21.480 align:middle line:84% So if driving becomes more attractive, 02:07:21.480 --> 02:07:23.320 align:middle line:90% people take more trips per day. 02:07:23.320 --> 02:07:27.220 align:middle line:90% 02:07:27.220 --> 02:07:29.500 align:middle line:90% They take longer trips. 02:07:29.500 --> 02:07:34.660 align:middle line:84% They may reduce their usage of public transit. 02:07:34.660 --> 02:07:36.360 align:middle line:90% And all of those have impacts. 02:07:36.360 --> 02:07:40.080 align:middle line:84% So the public transit ripple effects 02:07:40.080 --> 02:07:43.300 align:middle line:84% go through the revenue side, which 02:07:43.300 --> 02:07:46.820 align:middle line:84% may change the quality of the public transit 02:07:46.820 --> 02:07:51.460 align:middle line:90% network and its adequacy. 02:07:51.460 --> 02:07:55.020 align:middle line:84% And then on the other side, the traffic volume 02:07:55.020 --> 02:07:57.660 align:middle line:84% is driven by these factors that were affected 02:07:57.660 --> 02:08:00.900 align:middle line:84% by the attractiveness of driving, things like trips, 02:08:00.900 --> 02:08:03.860 align:middle line:84% number of trips, average length of trips, 02:08:03.860 --> 02:08:05.600 align:middle line:90% and cars that are in the region. 02:08:05.600 --> 02:08:10.340 align:middle line:84% So the buying of cars and is also captured in this diagram. 02:08:10.340 --> 02:08:13.900 align:middle line:84% Size of the region may also expand 02:08:13.900 --> 02:08:15.280 align:middle line:90% as you have more highways. 02:08:15.280 --> 02:08:18.020 align:middle line:84% And that can create interesting dynamics 02:08:18.020 --> 02:08:25.300 align:middle line:84% in terms of adequacy of public transit versus this people 02:08:25.300 --> 02:08:27.800 align:middle line:84% moving further out and therefore needing more cars. 02:08:27.800 --> 02:08:31.220 align:middle line:84% So those are the basic stories captured in the causal loop 02:08:31.220 --> 02:08:35.980 align:middle line:84% diagram-- again, coming from John's textbook from chapter 5. 02:08:35.980 --> 02:08:38.000 align:middle line:84% This one is discussed in detail in the text. 02:08:38.000 --> 02:08:39.480 align:middle line:84% So you can just read those as well. 02:08:39.480 --> 02:08:44.260 align:middle line:84% But I summarize those discussions in the assignment. 02:08:44.260 --> 02:08:48.100 align:middle line:84% So for this one, given the time, let's 02:08:48.100 --> 02:08:54.100 align:middle line:84% more quickly converge to the discussion of inertial forces. 02:08:54.100 --> 02:08:57.060 align:middle line:84% Specifically, again, we could go with different audiences 02:08:57.060 --> 02:08:59.820 align:middle line:84% and get very different answers to the questions 02:08:59.820 --> 02:09:03.860 align:middle line:84% of level of aggregation and relevant stocks. 02:09:03.860 --> 02:09:05.760 align:middle line:84% These are really important dynamics 02:09:05.760 --> 02:09:07.860 align:middle line:84% if you are building an operational model 02:09:07.860 --> 02:09:12.040 align:middle line:84% for managing transportation infrastructure in a city. 02:09:12.040 --> 02:09:14.060 align:middle line:84% So there are people building these types 02:09:14.060 --> 02:09:16.640 align:middle line:84% of models with really high levels of detail. 02:09:16.640 --> 02:09:21.870 align:middle line:84% They go into building models of public transit system 02:09:21.870 --> 02:09:25.150 align:middle line:84% that models the number of stations, all of the details 02:09:25.150 --> 02:09:31.270 align:middle line:84% of how many cars you have, and that you 02:09:31.270 --> 02:09:35.010 align:middle line:84% have models of highway capacity that are not a single capacity, 02:09:35.010 --> 02:09:38.150 align:middle line:84% but you actually have a network, and how that network creates 02:09:38.150 --> 02:09:44.030 align:middle line:84% traffic times, because it is really a very complex problem. 02:09:44.030 --> 02:09:46.390 align:middle line:84% So we could go into a lot of detail 02:09:46.390 --> 02:09:49.470 align:middle line:84% in that direction within the scope of this causal loop 02:09:49.470 --> 02:09:51.710 align:middle line:90% diagram. 02:09:51.710 --> 02:09:54.490 align:middle line:84% For most system dynamics types of models, 02:09:54.490 --> 02:09:57.190 align:middle line:84% that's too much detail if you are 02:09:57.190 --> 02:10:02.150 align:middle line:84% looking at dynamics that are longer term, in the order of 20, 02:10:02.150 --> 02:10:07.870 align:middle line:84% 30 years, which really shows off the trade off 02:10:07.870 --> 02:10:11.630 align:middle line:84% between public transit versus private transportation 02:10:11.630 --> 02:10:13.590 align:middle line:84% and the side effects of building roads 02:10:13.590 --> 02:10:15.450 align:middle line:84% and then having too much congestion. 02:10:15.450 --> 02:10:19.430 align:middle line:84% So those are the types of things that may be 02:10:19.430 --> 02:10:22.430 align:middle line:90% more interesting qualitatively. 02:10:22.430 --> 02:10:28.530 align:middle line:84% So that is a more theoretical high level strategic choice, 02:10:28.530 --> 02:10:30.530 align:middle line:84% which would be still relevant, for example, 02:10:30.530 --> 02:10:34.370 align:middle line:84% for a city mayor to think about their 10-year planning, 02:10:34.370 --> 02:10:39.950 align:middle line:84% but not for the transit commissioner for their next year 02:10:39.950 --> 02:10:40.450 align:middle line:90% planning. 02:10:40.450 --> 02:10:42.670 align:middle line:84% So these are two different audiences. 02:10:42.670 --> 02:10:44.950 align:middle line:84% If your audience is the city mayor 02:10:44.950 --> 02:10:50.430 align:middle line:84% in their strategic thinking or a transportation journal's 02:10:50.430 --> 02:10:53.690 align:middle line:84% high level theoretical conversations, 02:10:53.690 --> 02:10:58.970 align:middle line:84% what would be the right inertial concepts and stocks? 02:10:58.970 --> 02:11:01.890 align:middle line:84% So let's take five minutes, think about it independently. 02:11:01.890 --> 02:11:05.650 align:middle line:84% And then we will bring it back together in the class. 02:11:05.650 --> 02:11:10.210 align:middle line:90% 02:11:10.210 --> 02:11:10.710 align:middle line:90% OK. 02:11:10.710 --> 02:11:15.190 align:middle line:84% Let's go back online and get some input from folks online. 02:11:15.190 --> 02:11:16.990 align:middle line:84% Zakie, why don't you go first this time? 02:11:16.990 --> 02:11:23.756 align:middle line:90% 02:11:23.756 --> 02:11:30.350 align:middle line:84% AUDIENCE: [INAUDIBLE] about, for example, the time 02:11:30.350 --> 02:11:33.970 align:middle line:84% to build a new road to be a source of inertia 02:11:33.970 --> 02:11:39.270 align:middle line:84% because it will be slow, you need some time to do that. 02:11:39.270 --> 02:11:40.890 align:middle line:84% HAZHIR RAHMANDAD: So the travel time? 02:11:40.890 --> 02:11:42.057 align:middle line:90% The time you're on the road? 02:11:42.057 --> 02:11:42.630 align:middle line:90% AUDIENCE: No. 02:11:42.630 --> 02:11:47.208 align:middle line:84% No, I mean the time delay for building new roads. 02:11:47.208 --> 02:11:48.250 align:middle line:90% HAZHIR RAHMANDAD: Oh, OK. 02:11:48.250 --> 02:11:52.470 align:middle line:90% Yes, so construction delay. 02:11:52.470 --> 02:11:55.270 align:middle line:90% OK. 02:11:55.270 --> 02:12:00.110 align:middle line:84% Zakie, you started, but you didn't get a chance to continue. 02:12:00.110 --> 02:12:01.110 align:middle line:90% You're muted. 02:12:01.110 --> 02:12:04.610 align:middle line:84% Please unmute yourself first, yeah. 02:12:04.610 --> 02:12:18.115 align:middle line:90% AUDIENCE: Um, I don't-- 02:12:18.115 --> 02:12:20.490 align:middle line:84% HAZHIR RAHMANDAD: You don't have any good candidates yet? 02:12:20.490 --> 02:12:23.350 align:middle line:90% That's all right, no worries. 02:12:23.350 --> 02:12:25.830 align:middle line:84% Anne, do you want to share any candidates you have 02:12:25.830 --> 02:12:29.510 align:middle line:90% for sources of inertia here? 02:12:29.510 --> 02:12:34.230 align:middle line:84% AUDIENCE: I was thinking about cultural and social preferences 02:12:34.230 --> 02:12:38.210 align:middle line:84% for driving over taking public transport. 02:12:38.210 --> 02:12:41.150 align:middle line:90% 02:12:41.150 --> 02:12:44.890 align:middle line:84% HAZHIR RAHMANDAD: So that's preference for driving. 02:12:44.890 --> 02:12:49.190 align:middle line:84% So this is an interesting variable. 02:12:49.190 --> 02:12:52.730 align:middle line:84% So within a 20-30 year period, it can change. 02:12:52.730 --> 02:12:55.110 align:middle line:84% That is, the underlying preferences of people 02:12:55.110 --> 02:12:55.730 align:middle line:90% would change. 02:12:55.730 --> 02:12:58.910 align:middle line:84% So this would become an important stock. 02:12:58.910 --> 02:13:03.830 align:middle line:84% If we were doing the commissioners of transportation 02:13:03.830 --> 02:13:07.070 align:middle line:84% discussion, we may not even capture this as a stock 02:13:07.070 --> 02:13:09.230 align:middle line:84% because the time constant for this 02:13:09.230 --> 02:13:13.640 align:middle line:84% is in the order of several years for the underlying preference 02:13:13.640 --> 02:13:14.560 align:middle line:90% to change. 02:13:14.560 --> 02:13:18.360 align:middle line:84% And therefore, in a one to five-year planning period, 02:13:18.360 --> 02:13:21.840 align:middle line:84% you cannot count on enough cultural change to be 02:13:21.840 --> 02:13:24.180 align:middle line:84% meaningful, so you can use it as a parameter. 02:13:24.180 --> 02:13:27.760 align:middle line:84% So this is a way to think about what counts as a good stock 02:13:27.760 --> 02:13:29.640 align:middle line:84% or not or a good source of inertia 02:13:29.640 --> 02:13:31.860 align:middle line:84% really depends on the time horizon. 02:13:31.860 --> 02:13:34.080 align:middle line:90% So very good item to bring up. 02:13:34.080 --> 02:13:35.200 align:middle line:90% Great. 02:13:35.200 --> 02:13:37.520 align:middle line:90% Others? 02:13:37.520 --> 02:13:39.700 align:middle line:90% Anybody? 02:13:39.700 --> 02:13:41.200 align:middle line:84% AUDIENCE: I guess highway capacity. 02:13:41.200 --> 02:13:42.658 align:middle line:84% HAZHIR RAHMANDAD: Highway capacity. 02:13:42.658 --> 02:13:48.400 align:middle line:84% Yeah, so that's a pretty physical, clear stock 02:13:48.400 --> 02:13:49.780 align:middle line:90% relevant to the dynamics. 02:13:49.780 --> 02:13:51.400 align:middle line:90% So we do want to have that. 02:13:51.400 --> 02:13:52.840 align:middle line:84% AUDIENCE: Public transit capacity. 02:13:52.840 --> 02:13:54.923 align:middle line:84% HAZHIR RAHMANDAD: Public transit capacity as well. 02:13:54.923 --> 02:13:57.720 align:middle line:90% 02:13:57.720 --> 02:14:01.920 align:middle line:84% AUDIENCE: I would say the urban versus suburban population 02:14:01.920 --> 02:14:04.240 align:middle line:84% and the movement between the two. 02:14:04.240 --> 02:14:06.000 align:middle line:84% HAZHIR RAHMANDAD: So something about 02:14:06.000 --> 02:14:10.600 align:middle line:84% where the houses are-- that is, how much of the population 02:14:10.600 --> 02:14:13.620 align:middle line:90% is in the urban versus suburban. 02:14:13.620 --> 02:14:17.840 align:middle line:84% And if you are going with this high level of aggregation model, 02:14:17.840 --> 02:14:22.200 align:middle line:84% in terms of stocks, you may just stick to two-- 02:14:22.200 --> 02:14:26.540 align:middle line:84% urban residents versus suburban residents in households, 02:14:26.540 --> 02:14:27.980 align:middle line:90% but yeah. 02:14:27.980 --> 02:14:32.200 align:middle line:90% 02:14:32.200 --> 02:14:34.840 align:middle line:84% This is important because the preference 02:14:34.840 --> 02:14:39.740 align:middle line:84% to go to the suburban areas does depend to the highway capacity. 02:14:39.740 --> 02:14:41.920 align:middle line:84% It is an endogenous choice within the time 02:14:41.920 --> 02:14:44.260 align:middle line:90% horizon of a couple of decades. 02:14:44.260 --> 02:14:46.860 align:middle line:84% Again, for a three or four-year time horizon, 02:14:46.860 --> 02:14:48.460 align:middle line:90% it may not be as big a deal. 02:14:48.460 --> 02:14:50.400 align:middle line:84% But much of the transportation models 02:14:50.400 --> 02:14:52.520 align:middle line:84% don't explicitly model this feedback 02:14:52.520 --> 02:14:54.880 align:middle line:84% that people change their residence 02:14:54.880 --> 02:14:57.560 align:middle line:84% based on the transportation feedbacks. 02:14:57.560 --> 02:14:59.800 align:middle line:90% Good. 02:14:59.800 --> 02:15:01.300 align:middle line:84% AUDIENCE: Besides highway capacity, 02:15:01.300 --> 02:15:03.060 align:middle line:84% we have the number of cars on the road. 02:15:03.060 --> 02:15:04.810 align:middle line:84% HAZHIR RAHMANDAD: Absolutely, so something 02:15:04.810 --> 02:15:07.040 align:middle line:84% about cars on the road, car ownership. 02:15:07.040 --> 02:15:10.348 align:middle line:90% 02:15:10.348 --> 02:15:12.720 align:middle line:90% AUDIENCE: [INAUDIBLE] 02:15:12.720 --> 02:15:15.300 align:middle line:84% HAZHIR RAHMANDAD: So we have the capacity, 02:15:15.300 --> 02:15:16.620 align:middle line:90% but we don't have the rider. 02:15:16.620 --> 02:15:19.440 align:middle line:84% So there is this question of how the population is 02:15:19.440 --> 02:15:24.000 align:middle line:84% split between using public transit versus using 02:15:24.000 --> 02:15:26.120 align:middle line:90% private transportation. 02:15:26.120 --> 02:15:27.880 align:middle line:84% There is a question whether we want 02:15:27.880 --> 02:15:30.500 align:middle line:84% to think of that as a stock variable or not. 02:15:30.500 --> 02:15:31.920 align:middle line:90% It is not obvious. 02:15:31.920 --> 02:15:33.400 align:middle line:90% So why is it not obvious? 02:15:33.400 --> 02:15:36.200 align:middle line:84% As a physical thing, yes, people in the trains 02:15:36.200 --> 02:15:38.280 align:middle line:90% are obviously a stock variable. 02:15:38.280 --> 02:15:41.880 align:middle line:84% But people in the trains today, their time constant 02:15:41.880 --> 02:15:42.920 align:middle line:90% is in minutes. 02:15:42.920 --> 02:15:45.460 align:middle line:84% It's not relevant for our dynamics. 02:15:45.460 --> 02:15:48.720 align:middle line:84% So it could easily be an auxiliary variable. 02:15:48.720 --> 02:15:50.960 align:middle line:84% So the question is, what is the source 02:15:50.960 --> 02:15:53.520 align:middle line:84% of inertia that is more longer term 02:15:53.520 --> 02:15:58.080 align:middle line:84% in the context of this choice of public versus private transit? 02:15:58.080 --> 02:16:00.400 align:middle line:84% This is one of those places where you need 02:16:00.400 --> 02:16:02.380 align:middle line:90% to trace the source of inertia. 02:16:02.380 --> 02:16:05.920 align:middle line:84% Go always to the root of the source of inertia. 02:16:05.920 --> 02:16:08.040 align:middle line:84% So what is the source of inertia? 02:16:08.040 --> 02:16:10.640 align:middle line:84% One important source of inertia here 02:16:10.640 --> 02:16:15.360 align:middle line:84% is people's assessment perception 02:16:15.360 --> 02:16:19.360 align:middle line:84% of how good is public transit versus how 02:16:19.360 --> 02:16:23.400 align:middle line:90% good is private transit. 02:16:23.400 --> 02:16:28.200 align:middle line:84% So if I have perception of quality of public transit 02:16:28.200 --> 02:16:34.000 align:middle line:84% and perception of driving time, travel time as the major driver 02:16:34.000 --> 02:16:37.200 align:middle line:84% of congestion, if I have those two, 02:16:37.200 --> 02:16:41.840 align:middle line:84% I have already captured the core driver of inertia 02:16:41.840 --> 02:16:45.559 align:middle line:84% in how people change their mode-- 02:16:45.559 --> 02:16:50.160 align:middle line:84% their usage of different modes of transportation. 02:16:50.160 --> 02:16:53.760 align:middle line:84% So as such, I would be hesitant, for a simple model, 02:16:53.760 --> 02:16:58.000 align:middle line:84% to add another source of inertia for number 02:16:58.000 --> 02:17:02.020 align:middle line:84% of people who are using public transit, for example. 02:17:02.020 --> 02:17:03.520 align:middle line:84% This is actually a good conversation 02:17:03.520 --> 02:17:06.850 align:middle line:84% because this is not the most intuitive candidate stock. 02:17:06.850 --> 02:17:09.309 align:middle line:84% You would think of, OK, people on public transit, 02:17:09.309 --> 02:17:10.389 align:middle line:90% we're very intuitive. 02:17:10.389 --> 02:17:13.610 align:middle line:84% But it is not necessarily the right choice here. 02:17:13.610 --> 02:17:23.770 align:middle line:84% So let's add those perceived quality of public transit 02:17:23.770 --> 02:17:28.990 align:middle line:84% and congestion as important stocks. 02:17:28.990 --> 02:17:30.209 align:middle line:90% Sepideh 02:17:30.209 --> 02:17:34.690 align:middle line:84% AUDIENCE: I was wondering how you can measure the perceived 02:17:34.690 --> 02:17:36.263 align:middle line:90% quality of public transit. 02:17:36.263 --> 02:17:37.430 align:middle line:90% HAZHIR RAHMANDAD: Very good. 02:17:37.430 --> 02:17:42.410 align:middle line:84% So at this stage-- that is, deciding on good stock 02:17:42.410 --> 02:17:43.610 align:middle line:90% variables-- 02:17:43.610 --> 02:17:46.850 align:middle line:84% it is legitimate to not worry about it and say, 02:17:46.850 --> 02:17:48.150 align:middle line:90% I'll come back to it. 02:17:48.150 --> 02:17:49.750 align:middle line:90% But it is an important question. 02:17:49.750 --> 02:17:55.170 align:middle line:84% So thinking about this, I would think about some measure 02:17:55.170 --> 02:17:58.549 align:middle line:90% of time per mile of travel. 02:17:58.549 --> 02:18:02.209 align:middle line:84% That is, if I have a trip of typically one mile, 02:18:02.209 --> 02:18:05.129 align:middle line:84% what would be the time I need to be on the train 02:18:05.129 --> 02:18:06.950 align:middle line:90% and how much money it costs me. 02:18:06.950 --> 02:18:08.389 align:middle line:90% And the money part is easier. 02:18:08.389 --> 02:18:10.790 align:middle line:90% It's out there as a parameter. 02:18:10.790 --> 02:18:13.190 align:middle line:84% So I would focus on the expected time, 02:18:13.190 --> 02:18:17.450 align:middle line:84% which includes the waiting for the trips, the potential for-- 02:18:17.450 --> 02:18:20.110 align:middle line:84% and then that is not all of the quality. 02:18:20.110 --> 02:18:23.750 align:middle line:84% Then you may adjust that based on physical quality-- 02:18:23.750 --> 02:18:26.969 align:middle line:84% how clean it is, how safe it is, and so on. 02:18:26.969 --> 02:18:30.510 align:middle line:84% So I would incorporate all of it into a single measure. 02:18:30.510 --> 02:18:32.809 align:middle line:84% So it would be normalized travel time, 02:18:32.809 --> 02:18:36.382 align:middle line:84% normalized for the quality of my time spent there, 02:18:36.382 --> 02:18:37.549 align:middle line:90% something along those lines. 02:18:37.549 --> 02:18:40.990 align:middle line:84% AUDIENCE: So if for people who the audience of this model, 02:18:40.990 --> 02:18:46.330 align:middle line:84% maybe this is part of [INAUDIBLE] public transit 02:18:46.330 --> 02:18:51.652 align:middle line:84% cost or travel time is part of this perception for that. 02:18:51.652 --> 02:18:53.610 align:middle line:84% HAZHIR RAHMANDAD: Yes, so the perceived quality 02:18:53.610 --> 02:18:56.330 align:middle line:84% for a high level model should aggregate 02:18:56.330 --> 02:18:58.990 align:middle line:84% all of these components into a single construct. 02:18:58.990 --> 02:19:00.889 align:middle line:84% So that is why this conversation would 02:19:00.889 --> 02:19:05.410 align:middle line:84% be needed later on to how you operationalize it. 02:19:05.410 --> 02:19:07.809 align:middle line:84% AUDIENCE: I have a question of sorts. 02:19:07.809 --> 02:19:10.850 align:middle line:84% Is it reasonable to assume the public transit 02:19:10.850 --> 02:19:13.389 align:middle line:90% fare is some sort of stock? 02:19:13.389 --> 02:19:15.889 align:middle line:84% HAZHIR RAHMANDAD: It is, but we don't need to worry about it 02:19:15.889 --> 02:19:19.850 align:middle line:84% too much unless we are trying to have a formulation that 02:19:19.850 --> 02:19:23.650 align:middle line:84% says public transit fare responds endogenously 02:19:23.650 --> 02:19:26.770 align:middle line:84% to what is happening in the rest of the model, which 02:19:26.770 --> 02:19:27.469 align:middle line:90% is part of it. 02:19:27.469 --> 02:19:30.950 align:middle line:84% So the fare increase feedback loop is exactly capturing that. 02:19:30.950 --> 02:19:33.410 align:middle line:84% So if you want to endogenize public transit fare, 02:19:33.410 --> 02:19:36.870 align:middle line:84% because it takes time, it is a negotiation process and so on, 02:19:36.870 --> 02:19:40.010 align:middle line:84% and it may actually be something that takes months to a year, 02:19:40.010 --> 02:19:43.410 align:middle line:84% then, yes, then it would be a good stock candidate. 02:19:43.410 --> 02:19:44.090 align:middle line:90% Yeah? 02:19:44.090 --> 02:19:45.798 align:middle line:84% AUDIENCE: I don't know if it's important. 02:19:45.798 --> 02:19:49.050 align:middle line:84% But for example, also I was thinking about the time 02:19:49.050 --> 02:19:54.570 align:middle line:84% to improve the public transit capacity. 02:19:54.570 --> 02:20:00.410 align:middle line:84% HAZHIR RAHMANDAD: Yes, so the delays in PT construction delay 02:20:00.410 --> 02:20:03.050 align:middle line:90% needs something, exactly. 02:20:03.050 --> 02:20:03.570 align:middle line:90% Jake? 02:20:03.570 --> 02:20:04.650 align:middle line:84% AUDIENCE: I feel like there's something-- 02:20:04.650 --> 02:20:05.730 align:middle line:90% I don't know exactly-- 02:20:05.730 --> 02:20:09.810 align:middle line:84% but something about transit volume per person 02:20:09.810 --> 02:20:12.730 align:middle line:84% because we have trips per day and average trip 02:20:12.730 --> 02:20:14.973 align:middle line:84% length up there, but we're not capturing that. 02:20:14.973 --> 02:20:15.890 align:middle line:90% HAZHIR RAHMANDAD: Yes. 02:20:15.890 --> 02:20:21.030 align:middle line:84% So once we have determined the total demand for transportation, 02:20:21.030 --> 02:20:23.790 align:middle line:84% which you get from the population stocks-- 02:20:23.790 --> 02:20:25.730 align:middle line:84% so you need-- actually, those residents 02:20:25.730 --> 02:20:28.670 align:middle line:84% in suburban versus urban captures that. 02:20:28.670 --> 02:20:30.250 align:middle line:84% So you have the sources of inertia. 02:20:30.250 --> 02:20:33.930 align:middle line:84% Then you have the total demand for transit. 02:20:33.930 --> 02:20:37.730 align:middle line:84% Then you break it between public and private. 02:20:37.730 --> 02:20:40.030 align:middle line:84% So all of those can be done without a stock. 02:20:40.030 --> 02:20:42.730 align:middle line:90% 02:20:42.730 --> 02:20:44.470 align:middle line:90% So you have the demand in place. 02:20:44.470 --> 02:20:48.510 align:middle line:84% You have also the capacity, which we have a stock for. 02:20:48.510 --> 02:20:51.510 align:middle line:84% And so then you can see how well that demand is served. 02:20:51.510 --> 02:20:53.610 align:middle line:90% It creates also crowding. 02:20:53.610 --> 02:20:56.030 align:middle line:90% It creates your revenues. 02:20:56.030 --> 02:20:58.000 align:middle line:84% All of those will be endogenously created. 02:20:58.000 --> 02:21:01.080 align:middle line:84% So I don't think we actually are missing something there. 02:21:01.080 --> 02:21:02.498 align:middle line:90% AUDIENCE: [INAUDIBLE] 02:21:02.498 --> 02:21:04.040 align:middle line:84% HAZHIR RAHMANDAD: Demand, we have it. 02:21:04.040 --> 02:21:06.860 align:middle line:84% The stocks for demand are captured 02:21:06.860 --> 02:21:12.340 align:middle line:84% in the residents in that center of the city versus the suburbs. 02:21:12.340 --> 02:21:14.820 align:middle line:84% That's part of it, and then in their perceptions 02:21:14.820 --> 02:21:15.720 align:middle line:90% of the quality. 02:21:15.720 --> 02:21:19.500 align:middle line:84% So those together would allow you to calculate the demand 02:21:19.500 --> 02:21:20.360 align:middle line:90% explicitly. 02:21:20.360 --> 02:21:23.080 align:middle line:84% So that's how you go to the source of the inertia. 02:21:23.080 --> 02:21:25.860 align:middle line:90% Don't add more new variables. 02:21:25.860 --> 02:21:28.082 align:middle line:90% One more comment, and then-- 02:21:28.082 --> 02:21:29.540 align:middle line:84% AUDIENCE: There's a spatial element 02:21:29.540 --> 02:21:33.060 align:middle line:84% to this, which are hard to capture with our models. 02:21:33.060 --> 02:21:36.740 align:middle line:84% I don't know if we need it, but the geographic size 02:21:36.740 --> 02:21:41.200 align:middle line:90% of the catchment area. 02:21:41.200 --> 02:21:42.720 align:middle line:84% HAZHIR RAHMANDAD: Yeah, that's fair. 02:21:42.720 --> 02:21:45.500 align:middle line:84% And actually, that determines a bunch 02:21:45.500 --> 02:21:47.820 align:middle line:84% of parameters that translates how many people you 02:21:47.820 --> 02:21:50.860 align:middle line:84% have in suburban area into how much demand for traffic 02:21:50.860 --> 02:21:51.640 align:middle line:90% you have. 02:21:51.640 --> 02:21:55.000 align:middle line:84% So you could have the area of the city, in a sense, 02:21:55.000 --> 02:21:57.800 align:middle line:90% as a stock that expands. 02:21:57.800 --> 02:22:00.058 align:middle line:84% And then it changes those parameters endogenously 02:22:00.058 --> 02:22:02.600 align:middle line:84% through some function that is not trivial and it's nonlinear, 02:22:02.600 --> 02:22:04.200 align:middle line:90% but it can be done. 02:22:04.200 --> 02:22:04.700 align:middle line:90% Good. 02:22:04.700 --> 02:22:07.540 align:middle line:84% I think this is kind of a nice conversation about thinking 02:22:07.540 --> 02:22:08.720 align:middle line:90% about stocks. 02:22:08.720 --> 02:22:11.100 align:middle line:84% And hopefully, you can get some of these ideas 02:22:11.100 --> 02:22:12.520 align:middle line:90% into the next steps. 02:22:12.520 --> 02:22:17.340 align:middle line:84% So next, week what we are doing is-- 02:22:17.340 --> 02:22:20.060 align:middle line:84% let me just quickly go through that-- 02:22:20.060 --> 02:22:23.120 align:middle line:90% is extending the SIR model. 02:22:23.120 --> 02:22:27.520 align:middle line:84% So SIR model, well known, I hope everybody is familiar with it. 02:22:27.520 --> 02:22:31.180 align:middle line:84% It is susceptible infectious recovered or removed. 02:22:31.180 --> 02:22:34.380 align:middle line:90% It goes back to the 1920s. 02:22:34.380 --> 02:22:36.620 align:middle line:84% Simple but very powerful model, also 02:22:36.620 --> 02:22:39.580 align:middle line:84% overlaps with the Bass diffusion model in many ways. 02:22:39.580 --> 02:22:42.020 align:middle line:84% You're adding three things-- behavioral responses 02:22:42.020 --> 02:22:45.500 align:middle line:84% to risk-- that is, you have an ongoing epidemic. 02:22:45.500 --> 02:22:48.460 align:middle line:84% More death, as a result of the epidemic, 02:22:48.460 --> 02:22:51.200 align:middle line:84% leads to people staying home, reducing their interactions, 02:22:51.200 --> 02:22:55.980 align:middle line:84% becoming more cautious in terms of their interactions. 02:22:55.980 --> 02:22:59.940 align:middle line:84% Waning of immunity-- not waving, it should be waning. 02:22:59.940 --> 02:23:01.540 align:middle line:84% Waning of immunity-- that is, people 02:23:01.540 --> 02:23:04.400 align:middle line:84% lose immunity after some time, when they get the disease. 02:23:04.400 --> 02:23:06.480 align:middle line:84% So once you have recovered from the disease, 02:23:06.480 --> 02:23:09.720 align:middle line:84% you wouldn't remain immune forever for many diseases. 02:23:09.720 --> 02:23:11.920 align:middle line:84% For some, you remain immune almost forever. 02:23:11.920 --> 02:23:14.300 align:middle line:84% But for COVID, for example, you don't. 02:23:14.300 --> 02:23:16.400 align:middle line:84% So it's good to have a case in mind. 02:23:16.400 --> 02:23:18.740 align:middle line:84% In this case, COVID-19 is a good example 02:23:18.740 --> 02:23:22.100 align:middle line:84% to keep in mind to inform your decision making. 02:23:22.100 --> 02:23:25.660 align:middle line:84% And then vaccination, how would you capture vaccination? 02:23:25.660 --> 02:23:30.620 align:middle line:84% So there is more details on how to capture 02:23:30.620 --> 02:23:33.980 align:middle line:84% these or the assumptions you need to make 02:23:33.980 --> 02:23:35.720 align:middle line:90% and those judgment calls. 02:23:35.720 --> 02:23:38.660 align:middle line:84% So basically, you go through adding these three. 02:23:38.660 --> 02:23:40.340 align:middle line:84% And these are your three iterations-- 02:23:40.340 --> 02:23:43.740 align:middle line:84% add the first behavioral response, then 02:23:43.740 --> 02:23:46.860 align:middle line:84% the waning of immunity, then the vaccination. 02:23:46.860 --> 02:23:49.740 align:middle line:84% In fact, for the part with the waning of immunity, 02:23:49.740 --> 02:23:51.320 align:middle line:84% if you want to challenge yourself, 02:23:51.320 --> 02:23:54.480 align:middle line:84% try to put that model not only in equilibrium, 02:23:54.480 --> 02:23:57.620 align:middle line:84% but also calculate some of the equilibrium 02:23:57.620 --> 02:24:01.280 align:middle line:84% features of the model if you felt comfortable. 02:24:01.280 --> 02:24:03.420 align:middle line:84% And it depends on how you formulate 02:24:03.420 --> 02:24:05.000 align:middle line:90% the behavioral response. 02:24:05.000 --> 02:24:07.560 align:middle line:84% Depending on the nonlinearity that you capture there, 02:24:07.560 --> 02:24:09.540 align:middle line:84% you may or may not be able to do that 02:24:09.540 --> 02:24:12.140 align:middle line:90% with the analytical equilibrium. 02:24:12.140 --> 02:24:16.680 align:middle line:84% And then in terms of stocks and flows, 02:24:16.680 --> 02:24:18.840 align:middle line:84% the SIR parts are pretty well established. 02:24:18.840 --> 02:24:22.540 align:middle line:84% So there is no other stock that you need for that. 02:24:22.540 --> 02:24:26.420 align:middle line:84% But when you think about waning of immunity and other features 02:24:26.420 --> 02:24:30.780 align:middle line:84% that are introduced, there are additional architectural 02:24:30.780 --> 02:24:31.300 align:middle line:90% choices. 02:24:31.300 --> 02:24:33.780 align:middle line:90% And how can you keep it simple? 02:24:33.780 --> 02:24:36.760 align:middle line:84% So for the audience in this case, again, 02:24:36.760 --> 02:24:38.860 align:middle line:84% this is a model that you could actually 02:24:38.860 --> 02:24:43.220 align:middle line:84% turn into a policy oriented or decision support oriented model 02:24:43.220 --> 02:24:45.580 align:middle line:84% with more details or keep it fairly 02:24:45.580 --> 02:24:52.150 align:middle line:84% aggregate for a more conceptual, insight based model. 02:24:52.150 --> 02:24:53.170 align:middle line:90% Go with the latter. 02:24:53.170 --> 02:24:56.870 align:middle line:84% Keep it as aggregate as you can, as long 02:24:56.870 --> 02:25:00.910 align:middle line:84% as you can address those requirements that are discussed 02:25:00.910 --> 02:25:04.390 align:middle line:90% in the exercise description. 02:25:04.390 --> 02:25:07.090 align:middle line:84% So I think that's all we have time for today, 02:25:07.090 --> 02:25:08.810 align:middle line:84% except that if there are any questions, 02:25:08.810 --> 02:25:11.270 align:middle line:84% especially on minor clarifications 02:25:11.270 --> 02:25:16.630 align:middle line:84% or administrative items, let's get those addressed. 02:25:16.630 --> 02:25:21.930 align:middle line:84% And I think a couple of you still don't have video releases. 02:25:21.930 --> 02:25:23.870 align:middle line:84% If you can just sign those or let 02:25:23.870 --> 02:25:26.630 align:middle line:84% me know if you have a concern, we can figure that out. 02:25:26.630 --> 02:25:27.490 align:middle line:90% That would be great. 02:25:27.490 --> 02:25:30.950 align:middle line:90% And that's that. 02:25:30.950 --> 02:25:32.010 align:middle line:90% Any questions? 02:25:32.010 --> 02:25:35.670 align:middle line:90% 02:25:35.670 --> 02:25:38.230 align:middle line:90% OK. 02:25:38.230 --> 02:25:40.270 align:middle line:90% So then thank you, folks. 02:25:40.270 --> 02:25:42.650 align:middle line:84% Folks online, we'll see you next week. 02:25:42.650 --> 02:25:45.140 align:middle line:90% And have fun. 02:25:45.140 --> 02:25:50.000 align:middle line:90%