WEBVTT 00:00:00.000 --> 00:00:02.430 align:middle line:90% [SQUEAKING] 00:00:02.430 --> 00:00:03.888 align:middle line:90% [RUSTLING] 00:00:03.888 --> 00:00:06.318 align:middle line:90% [CLICKING] 00:00:06.318 --> 00:00:16.477 align:middle line:90% 00:00:16.477 --> 00:00:18.060 align:middle line:84% GLENN ELLISON: So bounded rationality. 00:00:18.060 --> 00:00:21.937 align:middle line:84% So we start that, you know, there's really, 00:00:21.937 --> 00:00:24.020 align:middle line:84% if you go back and look through the IO literature, 00:00:24.020 --> 00:00:26.420 align:middle line:84% there's this very, very long history of studying 00:00:26.420 --> 00:00:29.960 align:middle line:90% bounded rationality. 00:00:29.960 --> 00:00:32.840 align:middle line:84% In some sense, it starts at the same time 00:00:32.840 --> 00:00:35.780 align:middle line:84% that Samuelson was having of revolution in going 00:00:35.780 --> 00:00:39.080 align:middle line:84% towards rational economics, the entire time there 00:00:39.080 --> 00:00:42.200 align:middle line:84% was active debate about whether we should be modeling 00:00:42.200 --> 00:00:44.130 align:middle line:90% things rationally or not. 00:00:44.130 --> 00:00:47.000 align:middle line:84% And if you look at the early IO with bounded rationality, 00:00:47.000 --> 00:00:48.800 align:middle line:84% actually the difference from today 00:00:48.800 --> 00:00:52.310 align:middle line:84% is there's an awful lot of IO with bounded rationality 00:00:52.310 --> 00:00:53.440 align:middle line:90% on the firm side. 00:00:53.440 --> 00:00:56.780 align:middle line:84% There's a lot of skepticism of this idea of firms 00:00:56.780 --> 00:00:59.660 align:middle line:84% as sophisticated, profit-maximizing entities 00:00:59.660 --> 00:01:01.260 align:middle line:90% that just always do what's best. 00:01:01.260 --> 00:01:06.470 align:middle line:84% And so in the 1940s, there was this debate of should we think 00:01:06.470 --> 00:01:09.050 align:middle line:84% of firms as maximizing profits or should we think of firms 00:01:09.050 --> 00:01:13.430 align:middle line:84% as trying to get a pretty good profit that's secure and not 00:01:13.430 --> 00:01:14.730 align:middle line:90% trying to maximize profits? 00:01:14.730 --> 00:01:17.490 align:middle line:90% 00:01:17.490 --> 00:01:20.760 align:middle line:84% There was also a lot of literature about this team 00:01:20.760 --> 00:01:27.120 align:middle line:84% theory or Simon about firms have to process information that's 00:01:27.120 --> 00:01:31.050 align:middle line:84% really complicated, just like firms have to manufacture goods, 00:01:31.050 --> 00:01:33.100 align:middle line:84% firms have to deal with information. 00:01:33.100 --> 00:01:36.010 align:middle line:84% They have bounded information processing capabilities. 00:01:36.010 --> 00:01:38.170 align:middle line:84% They organize themselves to do this, 00:01:38.170 --> 00:01:40.920 align:middle line:84% and there are costs of trying to set the right price 00:01:40.920 --> 00:01:42.922 align:middle line:90% or costs of-- 00:01:42.922 --> 00:01:45.130 align:middle line:84% just like there are costs of trying to produce goods, 00:01:45.130 --> 00:01:48.240 align:middle line:84% and we should take that those costs into account. 00:01:48.240 --> 00:01:50.370 align:middle line:84% I'm actually going to talk a little bit more 00:01:50.370 --> 00:01:52.330 align:middle line:84% about Smallwood and Conlisk on Wednesday, 00:01:52.330 --> 00:01:55.140 align:middle line:84% but Smallwood and Conlisk were very clear arguing, 00:01:55.140 --> 00:01:57.450 align:middle line:84% when first rational consumers were 00:01:57.450 --> 00:02:00.930 align:middle line:84% coming in, that they felt rational consumers were much 00:02:00.930 --> 00:02:02.920 align:middle line:84% less plausible than irrational consumers 00:02:02.920 --> 00:02:05.910 align:middle line:84% and we shouldn't go all the way to assuming that consumers 00:02:05.910 --> 00:02:11.009 align:middle line:84% are these hyper sophisticated, rational people. 00:02:11.009 --> 00:02:14.220 align:middle line:84% More recent literature has really, over the last decades, 00:02:14.220 --> 00:02:17.930 align:middle line:84% has focused on abandon the irrational firms 00:02:17.930 --> 00:02:20.390 align:middle line:84% and gone with hyper sophisticated firms, 00:02:20.390 --> 00:02:24.380 align:middle line:84% but firms exploiting boundedly rational customers. 00:02:24.380 --> 00:02:26.960 align:middle line:84% That's a big literature and there's a thought of, 00:02:26.960 --> 00:02:29.360 align:middle line:84% you know, this is why we want to have a consumer 00:02:29.360 --> 00:02:31.640 align:middle line:84% financial protection bureau or other agencies, 00:02:31.640 --> 00:02:35.270 align:middle line:84% because firms are going to try to exploit irrational consumers 00:02:35.270 --> 00:02:38.210 align:middle line:90% and that creates inefficiencies. 00:02:38.210 --> 00:02:40.190 align:middle line:84% These inefficiencies can be large because-- 00:02:40.190 --> 00:02:44.330 align:middle line:84% like normal inefficiencies are triangles of price are too high, 00:02:44.330 --> 00:02:46.760 align:middle line:84% consumers who have a value a bit above cost 00:02:46.760 --> 00:02:48.280 align:middle line:90% don't get to purchase things. 00:02:48.280 --> 00:02:50.030 align:middle line:84% If you put in boundedly rational consumers 00:02:50.030 --> 00:02:52.040 align:middle line:84% and consumers are buying things they don't want 00:02:52.040 --> 00:02:54.320 align:middle line:84% or paying much more than something is worth to them, 00:02:54.320 --> 00:02:56.540 align:middle line:84% you can get bigger inefficiencies. 00:02:56.540 --> 00:02:59.150 align:middle line:84% And I would say there's three different approaches here 00:02:59.150 --> 00:03:01.850 align:middle line:90% that you see taken. 00:03:01.850 --> 00:03:04.820 align:middle line:84% One is the rule of thumb approach, 00:03:04.820 --> 00:03:07.580 align:middle line:84% following Smallwood and Conlisk, which just says, 00:03:07.580 --> 00:03:09.410 align:middle line:84% rather than modeling consumers as rational, 00:03:09.410 --> 00:03:13.010 align:middle line:84% let's just say what we think is a reasonable model for what 00:03:13.010 --> 00:03:14.930 align:middle line:84% consumers would actually do in a situation 00:03:14.930 --> 00:03:17.480 align:middle line:84% like a consumer walks into a restaurant. 00:03:17.480 --> 00:03:21.200 align:middle line:84% Do they really have a utility for every different thing 00:03:21.200 --> 00:03:23.583 align:middle line:84% on the menu and have beliefs about how much they're 00:03:23.583 --> 00:03:25.250 align:middle line:84% going to like everything they don't have 00:03:25.250 --> 00:03:26.250 align:middle line:90% and rationally maximize? 00:03:26.250 --> 00:03:28.563 align:middle line:84% Or do people just look at it quickly, 00:03:28.563 --> 00:03:30.230 align:middle line:84% pick out something they think they like, 00:03:30.230 --> 00:03:32.810 align:middle line:90% and buy it, or whatever. 00:03:32.810 --> 00:03:34.890 align:middle line:84% The explicit bounded rationality. 00:03:34.890 --> 00:03:37.430 align:middle line:84% In some sense, search theory is a model 00:03:37.430 --> 00:03:41.000 align:middle line:84% of explicit bounded rationality that consumers may be rational 00:03:41.000 --> 00:03:44.400 align:middle line:84% but have computational or other informational costs, 00:03:44.400 --> 00:03:46.850 align:middle line:84% and they do this combination of doing 00:03:46.850 --> 00:03:50.750 align:middle line:84% the best thing they can given their computational costs. 00:03:50.750 --> 00:03:54.420 align:middle line:84% Search is one example, but there are many others. 00:03:54.420 --> 00:03:57.740 align:middle line:84% And then the third branch is the psychology and economics branch 00:03:57.740 --> 00:04:00.290 align:middle line:84% where this is building off the kinds of things 00:04:00.290 --> 00:04:04.230 align:middle line:84% you would see in a behavioral economics class and modeling 00:04:04.230 --> 00:04:06.810 align:middle line:84% consumers as having behavioral biases 00:04:06.810 --> 00:04:09.600 align:middle line:84% and thinking about how firms exploit those. 00:04:09.600 --> 00:04:13.150 align:middle line:84% You know, this has been the most active branch in recent years, 00:04:13.150 --> 00:04:15.090 align:middle line:84% and so today's lecture, I'm pretty much-- 00:04:15.090 --> 00:04:17.673 align:middle line:84% unless I get to the paper at the end I think I won't have time 00:04:17.673 --> 00:04:20.700 align:middle line:84% for, today is pretty much going to be about these psychology 00:04:20.700 --> 00:04:23.890 align:middle line:90% and economics papers. 00:04:23.890 --> 00:04:24.490 align:middle line:90% What goes on. 00:04:24.490 --> 00:04:28.210 align:middle line:84% If you take a behavioral economics class? 00:04:28.210 --> 00:04:30.270 align:middle line:84% You know, so over the last 25 years, 00:04:30.270 --> 00:04:33.280 align:middle line:84% there's been this explosion of work in behavioral economics. 00:04:33.280 --> 00:04:35.940 align:middle line:84% And the way behavioral economics classes normally go 00:04:35.940 --> 00:04:39.330 align:middle line:84% is you talk about and read psychological studies that 00:04:39.330 --> 00:04:43.170 align:middle line:84% talk about how consumers do things that are not rational, 00:04:43.170 --> 00:04:46.680 align:middle line:84% and then you discuss models that people 00:04:46.680 --> 00:04:50.400 align:middle line:84% have written to model behavior that's 00:04:50.400 --> 00:04:52.470 align:middle line:84% kind of close to rational but not quite 00:04:52.470 --> 00:04:55.990 align:middle line:84% rational in a way that would explain the psychological bias. 00:04:55.990 --> 00:04:59.040 align:middle line:84% I mean, I guess Drew does a little bit of this in 123. 00:04:59.040 --> 00:05:03.180 align:middle line:90% But it's you take a model that-- 00:05:03.180 --> 00:05:05.850 align:middle line:84% take a standard model and then you make some small tweak to it 00:05:05.850 --> 00:05:08.800 align:middle line:84% that would allow consumers to behave in some different way, 00:05:08.800 --> 00:05:12.630 align:middle line:84% and then get a utility-like model that may have one 00:05:12.630 --> 00:05:14.890 align:middle line:84% or two extra free parameters that 00:05:14.890 --> 00:05:18.210 align:middle line:84% could display this is nearly rational but with these two 00:05:18.210 --> 00:05:19.960 align:middle line:84% extra parameters describing their behavior 00:05:19.960 --> 00:05:23.892 align:middle line:84% instead of behaving fully rationally. 00:05:23.892 --> 00:05:27.880 align:middle line:84% And there are many, many ways in the behavioral economics 00:05:27.880 --> 00:05:29.440 align:middle line:84% literature in which consumers are 00:05:29.440 --> 00:05:32.500 align:middle line:84% shown to behave not rationally, and there 00:05:32.500 --> 00:05:36.020 align:middle line:84% are rational-like models for most of them. 00:05:36.020 --> 00:05:38.130 align:middle line:90% So why don't I-- 00:05:38.130 --> 00:05:41.590 align:middle line:84% and, you know, the idea of how this behavioral economics 00:05:41.590 --> 00:05:43.990 align:middle line:84% literature could have an influence is you write down 00:05:43.990 --> 00:05:46.318 align:middle line:84% a nearly rational model, and then 00:05:46.318 --> 00:05:48.610 align:middle line:84% any model that you could solve with rational consumers, 00:05:48.610 --> 00:05:50.693 align:middle line:84% you can solve with these nearly rational consumers 00:05:50.693 --> 00:05:52.732 align:middle line:84% because you have a well-specified thing 00:05:52.732 --> 00:05:54.190 align:middle line:84% for their behavior which isn't just 00:05:54.190 --> 00:05:56.290 align:middle line:90% maximizing utility function. 00:05:56.290 --> 00:06:00.400 align:middle line:90% So, for instance, loss aversion. 00:06:00.400 --> 00:06:03.730 align:middle line:84% In financial problems, what do consumers do? 00:06:03.730 --> 00:06:07.486 align:middle line:90% They're taking some action a. 00:06:07.486 --> 00:06:10.340 align:middle line:84% And in the rational model, if they take an action a, 00:06:10.340 --> 00:06:13.640 align:middle line:84% they'll have some wealth w of a which is a random variable, 00:06:13.640 --> 00:06:15.212 align:middle line:84% and then they have a utility that's 00:06:15.212 --> 00:06:16.670 align:middle line:84% a function of that wealth, and then 00:06:16.670 --> 00:06:19.460 align:middle line:84% they maximize the expected utility of the wealth 00:06:19.460 --> 00:06:22.160 align:middle line:90% after taking the action. 00:06:22.160 --> 00:06:25.280 align:middle line:84% In loss averse problems, what is assumed 00:06:25.280 --> 00:06:29.000 align:middle line:84% is that consumers have a reference point, w0, 00:06:29.000 --> 00:06:31.040 align:middle line:84% which could be their current wealth, 00:06:31.040 --> 00:06:35.090 align:middle line:84% and they try to maximize some expected function 00:06:35.090 --> 00:06:37.540 align:middle line:90% of w of a minus w of 0. 00:06:37.540 --> 00:06:40.190 align:middle line:84% So you have utility not for wealth 00:06:40.190 --> 00:06:45.860 align:middle line:84% but for changes in wealth, and zero is a, somehow, 00:06:45.860 --> 00:06:48.600 align:middle line:90% point that's special. 00:06:48.600 --> 00:06:50.918 align:middle line:84% And the simplest version of this is 00:06:50.918 --> 00:06:52.460 align:middle line:84% that you have a utility function that 00:06:52.460 --> 00:07:01.490 align:middle line:84% looks like this where you care much more about losing money 00:07:01.490 --> 00:07:05.090 align:middle line:84% than you care about making money, and somehow losing money 00:07:05.090 --> 00:07:06.050 align:middle line:90% is painful to you. 00:07:06.050 --> 00:07:07.280 align:middle line:90% Because if you ask-- 00:07:07.280 --> 00:07:10.820 align:middle line:84% there are all these experiments, if you ask students which-- 00:07:10.820 --> 00:07:16.580 align:middle line:84% would you a 50/50 coin flip, win $101, lose $100? 00:07:16.580 --> 00:07:20.390 align:middle line:84% And almost everyone says, no, I don't want to do the 50/50 coin 00:07:20.390 --> 00:07:23.820 align:middle line:84% flip, win 101, lose 100, or something like that. 00:07:23.820 --> 00:07:27.728 align:middle line:84% It's not enough of a gain to make it worth losing $100. 00:07:27.728 --> 00:07:30.020 align:middle line:84% Matt Rabin has this very beautiful paper on calibrating 00:07:30.020 --> 00:07:34.070 align:middle line:84% preferences that says it's just completely inconceivable 00:07:34.070 --> 00:07:38.540 align:middle line:84% for a rational consumer to say no to a 50/50, 101, 100 gamble, 00:07:38.540 --> 00:07:41.780 align:middle line:84% because if you're going to turn down a 50/50, 101, 100 gamble, 00:07:41.780 --> 00:07:47.450 align:middle line:84% you would also turn down a lose 5,000, win a trillion gamble. 00:07:47.450 --> 00:07:49.520 align:middle line:84% Because in some sense, to turn down that gamble, 00:07:49.520 --> 00:07:53.970 align:middle line:84% you've got to have a very concave utility function 00:07:53.970 --> 00:07:55.930 align:middle line:84% and it's got to just keep getting concave. 00:07:55.930 --> 00:07:59.340 align:middle line:84% And if you don't like small gambles, you can't-- 00:07:59.340 --> 00:08:02.070 align:middle line:84% obviously you all take million dollar gambles with your lives, 00:08:02.070 --> 00:08:04.470 align:middle line:84% like when you choose your careers or whatever. 00:08:04.470 --> 00:08:06.840 align:middle line:84% Somehow people treat big stakes gambles 00:08:06.840 --> 00:08:09.390 align:middle line:84% very different from small stakes gambles. 00:08:09.390 --> 00:08:11.940 align:middle line:84% This loss aversion is one way to explain 00:08:11.940 --> 00:08:14.128 align:middle line:90% why people might do that. 00:08:14.128 --> 00:08:16.170 align:middle line:84% People often-- I guess, more sophisticated models 00:08:16.170 --> 00:08:20.850 align:middle line:84% do something more like this as a utility function. 00:08:20.850 --> 00:08:23.697 align:middle line:84% But the whole point is that it does have a kink here where 00:08:23.697 --> 00:08:26.030 align:middle line:84% the slope to the left and the slope to the right differ. 00:08:26.030 --> 00:08:28.770 align:middle line:90% 00:08:28.770 --> 00:08:30.270 align:middle line:84% When you have multiple goods, you 00:08:30.270 --> 00:08:32.900 align:middle line:84% have to assume, like, how do we deal with like, OK, 00:08:32.900 --> 00:08:34.150 align:middle line:90% this is the utility of wealth. 00:08:34.150 --> 00:08:35.733 align:middle line:84% How would we deal with multiple goods? 00:08:35.733 --> 00:08:36.750 align:middle line:90% And there are these-- 00:08:36.750 --> 00:08:38.460 align:middle line:84% again, these classic experiments, 00:08:38.460 --> 00:08:42.900 align:middle line:84% like you bring people into a lab and you have them play some game 00:08:42.900 --> 00:08:45.420 align:middle line:84% and you say, as a reward for doing this, 00:08:45.420 --> 00:08:49.740 align:middle line:84% you get an MIT Department of Economics water bottle. 00:08:49.740 --> 00:08:52.320 align:middle line:84% And then you let them have the MIT Economics water 00:08:52.320 --> 00:08:54.487 align:middle line:84% bottle for a few minutes and then say, oh, you know, 00:08:54.487 --> 00:08:57.150 align:middle line:90% actually, I also have these-- 00:08:57.150 --> 00:08:59.020 align:middle line:84% I also have these Lindt chocolate bars. 00:08:59.020 --> 00:09:02.640 align:middle line:84% You can have the Lindt chocolate bar instead of the water bottle, 00:09:02.640 --> 00:09:05.350 align:middle line:84% you know, if you want to trade or whatever. 00:09:05.350 --> 00:09:08.140 align:middle line:84% And then everyone says, no, no, I'll keep the water bottle. 00:09:08.140 --> 00:09:09.520 align:middle line:84% And then you run the experiment the opposite way. 00:09:09.520 --> 00:09:11.062 align:middle line:84% You give them the Lindt chocolate bar 00:09:11.062 --> 00:09:14.280 align:middle line:84% and say, oh, we also have this MIT water bottle you could have 00:09:14.280 --> 00:09:14.805 align:middle line:90% or whatever. 00:09:14.805 --> 00:09:16.930 align:middle line:84% Nobody trades a chocolate bar for the water bottle. 00:09:16.930 --> 00:09:19.470 align:middle line:84% And so the idea there is that how 00:09:19.470 --> 00:09:22.270 align:middle line:84% do we explain that people just get 00:09:22.270 --> 00:09:23.952 align:middle line:84% attached to things that they have. 00:09:23.952 --> 00:09:25.410 align:middle line:84% The loss aversion, what you've done 00:09:25.410 --> 00:09:28.850 align:middle line:84% is you have two different functions like this. 00:09:28.850 --> 00:09:32.440 align:middle line:84% I have my water bottle utility function 00:09:32.440 --> 00:09:35.220 align:middle line:84% and my chocolate bar utility function. 00:09:35.220 --> 00:09:36.970 align:middle line:84% And once you've given me the chocolate bar 00:09:36.970 --> 00:09:38.590 align:middle line:84% and I've had it for a few minutes, 00:09:38.590 --> 00:09:41.600 align:middle line:84% then it feels like a loss to give up the chocolate bar 00:09:41.600 --> 00:09:43.685 align:middle line:84% and it would be a win to get a water bottle. 00:09:43.685 --> 00:09:45.310 align:middle line:84% But you know, a loss of a chocolate bar 00:09:45.310 --> 00:09:47.393 align:middle line:84% and a win to a water bottle, that you wouldn't do, 00:09:47.393 --> 00:09:51.920 align:middle line:84% and the same vice versa once you've changed the endowments. 00:09:51.920 --> 00:09:53.270 align:middle line:90% And I think this is-- 00:09:53.270 --> 00:09:55.700 align:middle line:84% people do think that this is how you think about-- 00:09:55.700 --> 00:09:58.220 align:middle line:84% in some sense, this is one explanation 00:09:58.220 --> 00:10:02.180 align:middle line:84% for how stores try to lure people in with 00:10:02.180 --> 00:10:04.220 align:middle line:84% and then have high last minute charges 00:10:04.220 --> 00:10:07.070 align:middle line:84% or hidden charges is once you've gotten used to the fact 00:10:07.070 --> 00:10:09.140 align:middle line:84% that I'm going to own this TV, it's 00:10:09.140 --> 00:10:12.020 align:middle line:84% then hard to walk away from the TV 00:10:12.020 --> 00:10:14.330 align:middle line:84% because it's now a loss of one in the TV dimension 00:10:14.330 --> 00:10:16.163 align:middle line:84% because for the half an hour you thought you 00:10:16.163 --> 00:10:18.830 align:middle line:90% had that nice TV or something. 00:10:18.830 --> 00:10:21.110 align:middle line:90% OK. 00:10:21.110 --> 00:10:22.950 align:middle line:90% Fairness preferences. 00:10:22.950 --> 00:10:26.480 align:middle line:84% So obviously there's nothing in utility function 00:10:26.480 --> 00:10:29.510 align:middle line:84% that says that utility is only based on your own wealth, 00:10:29.510 --> 00:10:31.710 align:middle line:84% not on everybody else's wealth in the entire world 00:10:31.710 --> 00:10:34.100 align:middle line:90% and people can be altruistic. 00:10:34.100 --> 00:10:38.330 align:middle line:84% But I think when we're doing normal-- 00:10:38.330 --> 00:10:40.130 align:middle line:84% normally when we're doing most topics, 00:10:40.130 --> 00:10:43.952 align:middle line:84% we do stick with preferences that are non-altruistic 00:10:43.952 --> 00:10:45.410 align:middle line:84% and just everyone's preferences are 00:10:45.410 --> 00:10:48.410 align:middle line:90% based on their own consumption. 00:10:48.410 --> 00:10:50.540 align:middle line:84% There's a number of different models 00:10:50.540 --> 00:10:54.690 align:middle line:84% out there of preferences that do have some other regarding 00:10:54.690 --> 00:10:55.510 align:middle line:90% nature to them. 00:10:55.510 --> 00:10:59.280 align:middle line:84% So for instance, Fehr-Schmidt is the most popular model 00:10:59.280 --> 00:11:01.590 align:middle line:84% of fairness in the behavioral literature. 00:11:01.590 --> 00:11:04.170 align:middle line:84% What consumers do in a Fehr-Schmidt model 00:11:04.170 --> 00:11:08.375 align:middle line:84% is they are trying to maximize their own payoff. 00:11:08.375 --> 00:11:10.500 align:middle line:84% So I'm going to write pi here for the payoff rather 00:11:10.500 --> 00:11:11.620 align:middle line:90% than the utility. 00:11:11.620 --> 00:11:14.190 align:middle line:84% They're trying to maximize their own payoff, 00:11:14.190 --> 00:11:16.980 align:middle line:84% but they dislike when other people are getting 00:11:16.980 --> 00:11:19.320 align:middle line:90% more than them in a game. 00:11:19.320 --> 00:11:24.090 align:middle line:84% And so we subtract alpha times the fact 00:11:24.090 --> 00:11:26.430 align:middle line:84% that the extent to which other people's payoffs 00:11:26.430 --> 00:11:30.090 align:middle line:90% exceed your payoff. 00:11:30.090 --> 00:11:33.180 align:middle line:84% Fehr-Schmidt also assume that consumers 00:11:33.180 --> 00:11:35.740 align:middle line:84% dislike when other people earn less than them. 00:11:35.740 --> 00:11:39.360 align:middle line:84% They are other-regarding, and so they put weight beta 00:11:39.360 --> 00:11:43.470 align:middle line:84% on getting more money than other people in the game are getting. 00:11:43.470 --> 00:11:47.370 align:middle line:84% But Fehr-Schmidt assume that most people have 00:11:47.370 --> 00:11:49.620 align:middle line:84% a sense of fairness, but the fairness is more acute 00:11:49.620 --> 00:11:52.210 align:middle line:84% when they're the ones being treated unfairly, 00:11:52.210 --> 00:11:53.940 align:middle line:84% and so alpha is bigger than beta. 00:11:53.940 --> 00:11:57.870 align:middle line:84% Then you put more weight on me getting the small share 00:11:57.870 --> 00:12:00.810 align:middle line:84% than on other people getting a small share, although I 00:12:00.810 --> 00:12:02.760 align:middle line:90% don't like that either. 00:12:02.760 --> 00:12:06.960 align:middle line:84% And then again, you start with any utility model, 00:12:06.960 --> 00:12:08.820 align:middle line:84% and then you could do the Fehr-Schmidt model 00:12:08.820 --> 00:12:11.400 align:middle line:84% with two extra parameters alpha and beta, 00:12:11.400 --> 00:12:12.960 align:middle line:84% and then you can think about what 00:12:12.960 --> 00:12:15.640 align:middle line:84% happens in that model as a function of alpha and beta. 00:12:15.640 --> 00:12:18.280 align:middle line:84% So in some sense it becomes just a plug and play thing. 00:12:18.280 --> 00:12:24.413 align:middle line:84% Take a standard model, put in the alpha beta preferences, 00:12:24.413 --> 00:12:25.830 align:middle line:84% and then you can resolve the model 00:12:25.830 --> 00:12:28.038 align:middle line:84% and see what happens as a function of alpha and beta. 00:12:28.038 --> 00:12:31.485 align:middle line:84% And you can calibrate alpha and beta to what's in the data. 00:12:31.485 --> 00:12:35.125 align:middle line:90% 00:12:35.125 --> 00:12:38.470 align:middle line:84% Most of the examples I picked out to do today in IO, 00:12:38.470 --> 00:12:41.600 align:middle line:84% I'm doing what are called quasi-hyperbolic preferences. 00:12:41.600 --> 00:12:45.190 align:middle line:84% So in preferences with quasi-hyperbolic discounting, 00:12:45.190 --> 00:12:48.700 align:middle line:84% consumers are maximizing, at each point in time, 00:12:48.700 --> 00:12:52.030 align:middle line:84% the utility of the consumption they get today, 00:12:52.030 --> 00:12:57.980 align:middle line:84% plus beta times a discounted sum of future preferences. 00:12:57.980 --> 00:13:00.970 align:middle line:84% So everyone always cares about today 00:13:00.970 --> 00:13:03.590 align:middle line:84% more than they care about all future periods, 00:13:03.590 --> 00:13:07.930 align:middle line:84% and so your trade-off between should I 00:13:07.930 --> 00:13:10.778 align:middle line:84% clean up my apartment 11 days from now or 12 days from now, 00:13:10.778 --> 00:13:12.820 align:middle line:84% you're completely indifferent between whether you 00:13:12.820 --> 00:13:15.770 align:middle line:84% spend a day cleaning up 11 days from now or 12 days from now. 00:13:15.770 --> 00:13:18.140 align:middle line:84% But should I do it today versus tomorrow? 00:13:18.140 --> 00:13:20.270 align:middle line:84% You really want to put it off to tomorrow? 00:13:20.270 --> 00:13:23.650 align:middle line:84% So there's some much bigger emphasis discounting 00:13:23.650 --> 00:13:25.930 align:middle line:84% between today and all future times 00:13:25.930 --> 00:13:29.930 align:middle line:84% than there is between future times that come later. 00:13:29.930 --> 00:13:32.740 align:middle line:84% So the discounting between any two future periods 00:13:32.740 --> 00:13:37.470 align:middle line:84% is delta, but between the current period and-- 00:13:37.470 --> 00:13:41.200 align:middle line:84% you put weight one on today, beta delta on tomorrow, 00:13:41.200 --> 00:13:43.700 align:middle line:84% beta delta squared on the next day 00:13:43.700 --> 00:13:49.210 align:middle line:84% so that the ratio is beta delta here and only delta here. 00:13:49.210 --> 00:13:51.050 align:middle line:84% Again, so people have preferences. 00:13:51.050 --> 00:13:52.570 align:middle line:84% Like I would like to go on a diet, 00:13:52.570 --> 00:13:55.210 align:middle line:84% but I would like to go on a diet starting tomorrow. 00:13:55.210 --> 00:13:56.710 align:middle line:84% Or I would like to save more money, 00:13:56.710 --> 00:14:00.880 align:middle line:84% but I'd like to save more money after I buy this one TV I really 00:14:00.880 --> 00:14:03.040 align:middle line:84% like and then I'm going to start saving my money, 00:14:03.040 --> 00:14:04.960 align:middle line:90% or something like that. 00:14:04.960 --> 00:14:09.190 align:middle line:84% And then there are variants of this model. 00:14:09.190 --> 00:14:12.370 align:middle line:84% Sophisticated consumers recognize-- 00:14:12.370 --> 00:14:14.950 align:middle line:84% sophisticated hyperbolic consumers recognize 00:14:14.950 --> 00:14:18.580 align:middle line:84% that they have these preferences and they 00:14:18.580 --> 00:14:22.210 align:middle line:84% want to do things to commit themselves to not behave 00:14:22.210 --> 00:14:24.070 align:middle line:84% in a hyperbolic way, or they know 00:14:24.070 --> 00:14:26.480 align:middle line:84% that they need to commit their future selves to things 00:14:26.480 --> 00:14:28.000 align:middle line:84% so their future selves will actually do what they 00:14:28.000 --> 00:14:29.530 align:middle line:90% want their future selves to do. 00:14:29.530 --> 00:14:32.510 align:middle line:84% So it's like, I am going to put something 00:14:32.510 --> 00:14:35.530 align:middle line:84% on my phone that shuts me off of social media for some number 00:14:35.530 --> 00:14:38.560 align:middle line:84% of hours to commit that future lazy self of mine 00:14:38.560 --> 00:14:41.860 align:middle line:84% to actually get work done and not spend time 00:14:41.860 --> 00:14:44.680 align:middle line:90% scrolling through social media. 00:14:44.680 --> 00:14:46.860 align:middle line:90% The other ones are-- 00:14:46.860 --> 00:14:49.000 align:middle line:90% other models, the naive model. 00:14:49.000 --> 00:14:51.645 align:middle line:84% The naive model, you're hyperbolic 00:14:51.645 --> 00:14:53.770 align:middle line:84% and you don't recognize that your future selves are 00:14:53.770 --> 00:14:57.370 align:middle line:84% hyperbolic and you just think, I have willpower. 00:14:57.370 --> 00:14:58.580 align:middle line:90% These are my preferences. 00:14:58.580 --> 00:14:59.950 align:middle line:84% It's perfectly reasonable for me to say 00:14:59.950 --> 00:15:02.200 align:middle line:84% I'm going to sit here and scroll through social media, 00:15:02.200 --> 00:15:05.210 align:middle line:84% and then tomorrow my future self will not scroll. 00:15:05.210 --> 00:15:06.940 align:middle line:84% I'll just make the plan that this week I 00:15:06.940 --> 00:15:08.350 align:middle line:90% get six hours of social media. 00:15:08.350 --> 00:15:10.030 align:middle line:84% I'll use it all up this afternoon, 00:15:10.030 --> 00:15:13.310 align:middle line:84% and then my future self will not do any social media tomorrow. 00:15:13.310 --> 00:15:16.540 align:middle line:84% And that's my plan for this week. 00:15:16.540 --> 00:15:21.430 align:middle line:84% And then the partially naive model, 00:15:21.430 --> 00:15:23.470 align:middle line:84% an extension to go from a two parameter 00:15:23.470 --> 00:15:26.380 align:middle line:84% to a three parameter hyperbolic extension 00:15:26.380 --> 00:15:29.810 align:middle line:90% is I am a beta delta consumer. 00:15:29.810 --> 00:15:34.180 align:middle line:84% I recognize that my future self is also hyperbolic, 00:15:34.180 --> 00:15:37.540 align:middle line:84% but I think my future self is like a beta hat delta 00:15:37.540 --> 00:15:40.550 align:middle line:84% consumer with beta hat greater than beta. 00:15:40.550 --> 00:15:43.545 align:middle line:90% So I just underestimate. 00:15:43.545 --> 00:15:47.140 align:middle line:84% I recognize that I'm hyperbolic, but I 00:15:47.140 --> 00:15:49.810 align:middle line:84% underestimate the degree to which my future self is hyper 00:15:49.810 --> 00:15:52.531 align:middle line:90% hyperbolic. 00:15:52.531 --> 00:15:54.325 align:middle line:84% And again, if you think about the delta 00:15:54.325 --> 00:15:56.200 align:middle line:84% as already being a parameter we already have, 00:15:56.200 --> 00:15:58.690 align:middle line:84% again, maybe just beta and beta hat 00:15:58.690 --> 00:16:01.620 align:middle line:84% are two parameters you add to the standard rational model. 00:16:01.620 --> 00:16:04.750 align:middle line:90% 00:16:04.750 --> 00:16:06.160 align:middle line:90% Other common examples. 00:16:06.160 --> 00:16:08.950 align:middle line:84% I won't talk about IO examples today with them. 00:16:08.950 --> 00:16:12.400 align:middle line:90% Imperfect Bayesian updating. 00:16:12.400 --> 00:16:14.095 align:middle line:84% Eyster and Rabin proposed this model. 00:16:14.095 --> 00:16:15.970 align:middle line:84% There's a lot of evidence that people are not 00:16:15.970 --> 00:16:17.390 align:middle line:90% so good at Bayesian updating. 00:16:17.390 --> 00:16:19.640 align:middle line:84% They propose this model of imperfect Bayesian updating 00:16:19.640 --> 00:16:22.780 align:middle line:84% that's just the simplest one in the world. 00:16:22.780 --> 00:16:26.410 align:middle line:84% Your posterior on theta after observing 00:16:26.410 --> 00:16:30.910 align:middle line:84% some signal x is some constant times your prior plus 1 00:16:30.910 --> 00:16:37.040 align:middle line:84% minus chi times what the correct preference is, correct Bayesian 00:16:37.040 --> 00:16:37.540 align:middle line:90% updating is. 00:16:37.540 --> 00:16:42.070 align:middle line:84% So people just under-update when they get new information. 00:16:42.070 --> 00:16:43.913 align:middle line:84% They tend to be overly confident, 00:16:43.913 --> 00:16:44.913 align:middle line:90% stick with their priors. 00:16:44.913 --> 00:16:48.040 align:middle line:90% 00:16:48.040 --> 00:16:50.440 align:middle line:90% Overconfidence bias. 00:16:50.440 --> 00:16:53.110 align:middle line:84% Overconfidence bias is people think 00:16:53.110 --> 00:16:55.750 align:middle line:84% they are better than they are at many, many things. 00:16:55.750 --> 00:17:01.720 align:middle line:84% A classic example is ask a class how many miles is it from Boston 00:17:01.720 --> 00:17:07.105 align:middle line:84% to Buenos Aires and write down your best guess for Boston 00:17:07.105 --> 00:17:08.359 align:middle line:90% to Buenos Aires. 00:17:08.359 --> 00:17:11.349 align:middle line:84% And now write down a 50% confidence interval 00:17:11.349 --> 00:17:12.950 align:middle line:90% or a 90% confidence interval. 00:17:12.950 --> 00:17:15.208 align:middle line:90% 00:17:15.208 --> 00:17:17.500 align:middle line:84% You admit you don't know how far it is to Buenos Aires, 00:17:17.500 --> 00:17:21.430 align:middle line:84% but write down an interval you think there's a 50% chance 00:17:21.430 --> 00:17:23.470 align:middle line:84% that the distance from Boston to Buenos Aires 00:17:23.470 --> 00:17:25.339 align:middle line:84% is within the interval you write down. 00:17:25.339 --> 00:17:28.150 align:middle line:84% And overwhelmingly, I don't know, is it, like, 00:17:28.150 --> 00:17:29.530 align:middle line:90% 50% of the people-- 00:17:29.530 --> 00:17:31.780 align:middle line:84% like if you ask for 90% confidence interval, 00:17:31.780 --> 00:17:34.300 align:middle line:84% 50% of the correct answers are within the 90%-- 00:17:34.300 --> 00:17:37.120 align:middle line:84% 50% of the time, the 90% confidence interval 00:17:37.120 --> 00:17:38.380 align:middle line:90% contains the answer. 00:17:38.380 --> 00:17:40.810 align:middle line:84% 30% of time, the 50% confidence interval 00:17:40.810 --> 00:17:42.700 align:middle line:90% contains the correct answer. 00:17:42.700 --> 00:17:45.580 align:middle line:84% It's just people think they are better at many things 00:17:45.580 --> 00:17:48.490 align:middle line:90% than they are. 00:17:48.490 --> 00:17:50.270 align:middle line:90% Self-serving bias. 00:17:50.270 --> 00:17:52.982 align:middle line:84% This is the idea that we all like to think-- 00:17:52.982 --> 00:17:54.940 align:middle line:84% this is that we all think we're good at things. 00:17:54.940 --> 00:17:57.740 align:middle line:84% Self-serving bias is we all like to think we're good at things. 00:17:57.740 --> 00:18:00.520 align:middle line:84% So you make a decision, and then when 00:18:00.520 --> 00:18:02.740 align:middle line:84% you're called on to make a second decision, 00:18:02.740 --> 00:18:05.110 align:middle line:84% you make a second decision in a way that makes 00:18:05.110 --> 00:18:08.470 align:middle line:90% your first decision correct. 00:18:08.470 --> 00:18:11.020 align:middle line:84% That is so instead of like admitting-- 00:18:11.020 --> 00:18:14.670 align:middle line:90% 00:18:14.670 --> 00:18:15.532 align:middle line:90% you make some plan. 00:18:15.532 --> 00:18:16.740 align:middle line:90% You're going to go somewhere. 00:18:16.740 --> 00:18:20.188 align:middle line:84% And then you get there and realize, wow, 00:18:20.188 --> 00:18:22.480 align:middle line:84% I spent half an hour on the train to get to this place. 00:18:22.480 --> 00:18:25.080 align:middle line:84% This restaurant looks way more expensive than I wanted 00:18:25.080 --> 00:18:26.880 align:middle line:90% or I don't want to be here. 00:18:26.880 --> 00:18:27.990 align:middle line:90% Food looks bad. 00:18:27.990 --> 00:18:31.290 align:middle line:84% But once you're there, you go ahead and eat there anyway, 00:18:31.290 --> 00:18:34.800 align:middle line:84% because if you don't eat there, you're 00:18:34.800 --> 00:18:37.650 align:middle line:84% admitting that I made a bad decision in deciding 00:18:37.650 --> 00:18:40.150 align:middle line:84% to pick out this restaurant and come here for half an hour. 00:18:40.150 --> 00:18:42.317 align:middle line:84% And then you want to make my first decision correct. 00:18:42.317 --> 00:18:44.192 align:middle line:84% The only way my first decision can be correct 00:18:44.192 --> 00:18:45.930 align:middle line:84% is if I then carry out my second decision 00:18:45.930 --> 00:18:47.573 align:middle line:90% to go ahead and eat there. 00:18:47.573 --> 00:18:48.990 align:middle line:84% Then maybe you don't like the food 00:18:48.990 --> 00:18:52.140 align:middle line:84% and you're like, oh, wow that was unknowable, so I just 00:18:52.140 --> 00:18:53.650 align:middle line:90% had a bad draw or whatever. 00:18:53.650 --> 00:18:55.540 align:middle line:84% But it wasn't an ex-ante bad decision. 00:18:55.540 --> 00:18:58.092 align:middle line:84% But the idea of self-serving is we always-- 00:18:58.092 --> 00:18:59.550 align:middle line:84% we follow these paths in life where 00:18:59.550 --> 00:19:04.020 align:middle line:84% we just keep taking decisions, and each step we take 00:19:04.020 --> 00:19:06.720 align:middle line:84% a next step that's consistent with the earlier steps 00:19:06.720 --> 00:19:09.540 align:middle line:84% being right, rather than admitting your first step 00:19:09.540 --> 00:19:12.195 align:middle line:84% was wrong and I should have just admitted my first step was 00:19:12.195 --> 00:19:14.820 align:middle line:84% wrong and said, yeah, I made a bad decision there and gone back 00:19:14.820 --> 00:19:18.060 align:middle line:90% and start over or something. 00:19:18.060 --> 00:19:21.790 align:middle line:84% OK, anyway, so many, many things. 00:19:21.790 --> 00:19:24.720 align:middle line:84% And in some ways, I think this is a-- 00:19:24.720 --> 00:19:26.850 align:middle line:84% field has expanded enough in the last decade 00:19:26.850 --> 00:19:30.720 align:middle line:84% that it's no longer as true, but in some ways, 00:19:30.720 --> 00:19:33.390 align:middle line:84% once you have this view of-- and this is sometimes 00:19:33.390 --> 00:19:35.790 align:middle line:84% why behavioral economics does so well 00:19:35.790 --> 00:19:39.420 align:middle line:84% in every field of economics, because if you think 00:19:39.420 --> 00:19:41.790 align:middle line:84% about it, before behavioral came into fashion, 00:19:41.790 --> 00:19:43.560 align:middle line:84% what had Tirole done in his textbook? 00:19:43.560 --> 00:19:46.320 align:middle line:84% He worked out with rational consumers what 00:19:46.320 --> 00:19:48.607 align:middle line:84% happens with monopoly pricing and price discrimination 00:19:48.607 --> 00:19:50.940 align:middle line:84% and selling durable goods and horizontal differentiation 00:19:50.940 --> 00:19:53.592 align:middle line:84% and vertical differentiation and search costs and so on. 00:19:53.592 --> 00:19:55.800 align:middle line:84% You basically take the entire syllabus of this course 00:19:55.800 --> 00:19:58.590 align:middle line:84% and it's all rational consumers and firms 00:19:58.590 --> 00:20:00.570 align:middle line:90% dealing with this situation. 00:20:00.570 --> 00:20:03.630 align:middle line:84% And then if you think about from the behavioral perspective, 00:20:03.630 --> 00:20:07.810 align:middle line:84% well, rationality is just one incorrect model of consumers, 00:20:07.810 --> 00:20:12.360 align:middle line:84% and there are actually this many or more 00:20:12.360 --> 00:20:13.920 align:middle line:84% other models of consumers that would 00:20:13.920 --> 00:20:15.670 align:middle line:90% be better models of consumers. 00:20:15.670 --> 00:20:17.190 align:middle line:84% In some sense, what we need to do 00:20:17.190 --> 00:20:22.390 align:middle line:84% is take the entire field of IO as it existed in the year 2005 00:20:22.390 --> 00:20:24.972 align:middle line:84% and multiply it out into a matrix. 00:20:24.972 --> 00:20:26.680 align:middle line:84% What happens in every one of these models 00:20:26.680 --> 00:20:30.755 align:middle line:84% with quasi-hyperbolic consumers, with loss averse consumers, 00:20:30.755 --> 00:20:32.380 align:middle line:84% with consumers who care about fairness, 00:20:32.380 --> 00:20:35.680 align:middle line:84% with overconfident consumers, and so on? 00:20:35.680 --> 00:20:39.020 align:middle line:84% It seems like, how can I think of a paper to write? 00:20:39.020 --> 00:20:41.410 align:middle line:84% But in some sense, you can take every paper that 00:20:41.410 --> 00:20:45.850 align:middle line:84% exists in the literature and write 14 new papers about it, 00:20:45.850 --> 00:20:48.070 align:middle line:84% and so it just opens up this massive field. 00:20:48.070 --> 00:20:52.840 align:middle line:84% And again, is-- what do I have here? 00:20:52.840 --> 00:20:54.700 align:middle line:90% Yeah. 00:20:54.700 --> 00:20:57.050 align:middle line:84% Are all of those good papers to write? 00:20:57.050 --> 00:20:58.540 align:middle line:90% You know, probably not. 00:20:58.540 --> 00:21:04.160 align:middle line:84% It may be that maybe you try to do horizontal differentiation, 00:21:04.160 --> 00:21:05.813 align:middle line:84% you put in quasi-hyperbolic consumers 00:21:05.813 --> 00:21:07.480 align:middle line:84% and then you realize, wait, the hoteling 00:21:07.480 --> 00:21:09.080 align:middle line:90% model is a static model. 00:21:09.080 --> 00:21:10.400 align:middle line:90% It's all in one period. 00:21:10.400 --> 00:21:13.900 align:middle line:84% Therefore, hyperbolic has nothing to do with it. 00:21:13.900 --> 00:21:17.920 align:middle line:84% Certainly loss averse consumers in a horizontal differentiation 00:21:17.920 --> 00:21:20.350 align:middle line:84% model matter and consumers could care about fairness 00:21:20.350 --> 00:21:24.310 align:middle line:84% and consumers could be overconfident in their thoughts 00:21:24.310 --> 00:21:27.340 align:middle line:84% about their own preferences or whatever. 00:21:27.340 --> 00:21:29.710 align:middle line:84% But in some ways, there are many, 00:21:29.710 --> 00:21:32.510 align:middle line:84% many of these boxes that have something there, 00:21:32.510 --> 00:21:34.360 align:middle line:84% and in some ways, I think what you 00:21:34.360 --> 00:21:38.170 align:middle line:84% want is there's this opportunity to explore them and think 00:21:38.170 --> 00:21:43.810 align:middle line:84% about which of these preferences might be relevant to which 00:21:43.810 --> 00:21:47.230 align:middle line:84% of these models we looked at and think about what is there 00:21:47.230 --> 00:21:48.500 align:middle line:90% interesting to say. 00:21:48.500 --> 00:21:51.490 align:middle line:84% And I think what a lot of what people are interested in 00:21:51.490 --> 00:21:55.330 align:middle line:84% is if you can tell stories where I make some small change 00:21:55.330 --> 00:21:57.370 align:middle line:84% to the existing model and now I get 00:21:57.370 --> 00:22:00.640 align:middle line:84% a big change in the outcomes, so I don't 00:22:00.640 --> 00:22:02.828 align:middle line:90% need an enormous irrationality. 00:22:02.828 --> 00:22:04.870 align:middle line:84% In some sense, the original model we started with 00:22:04.870 --> 00:22:06.420 align:middle line:84% was very sensitive to rationality. 00:22:06.420 --> 00:22:08.962 align:middle line:84% When you put in a small change, it makes a big difference. 00:22:08.962 --> 00:22:11.170 align:middle line:84% And you could certainly imagine many of these things, 00:22:11.170 --> 00:22:14.230 align:middle line:84% like, it could make a big difference in durable goods 00:22:14.230 --> 00:22:18.550 align:middle line:84% models if we had quasi-hyperbolic preferences. 00:22:18.550 --> 00:22:19.220 align:middle line:90% Take an example. 00:22:19.220 --> 00:22:20.803 align:middle line:84% I guess you probably didn't get there, 00:22:20.803 --> 00:22:27.800 align:middle line:84% but when I talked about my add-on pricing paper about 00:22:27.800 --> 00:22:32.030 align:middle line:84% do firms want to post an unadvertised-- post 00:22:32.030 --> 00:22:35.430 align:middle line:84% an advertised price that doesn't include some add-on 00:22:35.430 --> 00:22:37.430 align:middle line:84% and then hold up consumers for the full monopoly 00:22:37.430 --> 00:22:40.847 align:middle line:84% price of the add-on after consumers show up at the firm. 00:22:40.847 --> 00:22:42.680 align:middle line:84% Back in the back half of that paper, there's 00:22:42.680 --> 00:22:45.890 align:middle line:84% a section that says in a game where the two firms are 00:22:45.890 --> 00:22:49.670 align:middle line:84% simultaneously deciding whether to advertise the base good only 00:22:49.670 --> 00:22:52.285 align:middle line:84% or the base good and an add-on, which would they do? 00:22:52.285 --> 00:22:53.660 align:middle line:84% And the answer is the firms don't 00:22:53.660 --> 00:22:55.785 align:middle line:84% want-- it's not individually rational in that model 00:22:55.785 --> 00:22:58.310 align:middle line:84% for the firms to do add-on pricing. 00:22:58.310 --> 00:23:02.095 align:middle line:84% However, if you add a small number of behavioral consumers 00:23:02.095 --> 00:23:04.220 align:middle line:84% who don't realize how expensive the add-on is going 00:23:04.220 --> 00:23:06.050 align:middle line:84% to be until they get to the point of sale, 00:23:06.050 --> 00:23:07.530 align:middle line:90% then it can make sense. 00:23:07.530 --> 00:23:10.430 align:middle line:84% And so it's just an example where you have a small change 00:23:10.430 --> 00:23:11.930 align:middle line:84% to the model, like a small fraction 00:23:11.930 --> 00:23:17.841 align:middle line:84% of the irrational consumers can have a big change on the model. 00:23:17.841 --> 00:23:23.780 align:middle line:84% So I'm going to start with just, you know, very early paper 00:23:23.780 --> 00:23:25.710 align:middle line:90% bringing out some basic ideas. 00:23:25.710 --> 00:23:29.630 align:middle line:84% So this is Della Vigna and Malmendier 00:23:29.630 --> 00:23:33.530 align:middle line:84% thinking about how a monopolist would design products or sales 00:23:33.530 --> 00:23:37.430 align:middle line:84% mechanisms to exploit hyperbolic consumers. 00:23:37.430 --> 00:23:41.000 align:middle line:84% And this is a multi-stage game motivated by health clubs 00:23:41.000 --> 00:23:46.150 align:middle line:84% where the idea is that everybody wishes they went to the gym more 00:23:46.150 --> 00:23:49.240 align:middle line:90% than they actually do. 00:23:49.240 --> 00:23:51.170 align:middle line:90% And so here's their model. 00:23:51.170 --> 00:23:54.470 align:middle line:84% So firms offer consumers a two part tariff. 00:23:54.470 --> 00:23:57.730 align:middle line:84% You pay l, which is a sign-up fee to join the gym. 00:23:57.730 --> 00:24:01.400 align:middle line:84% And then if you visit x times, you pay p times x. 00:24:01.400 --> 00:24:04.930 align:middle line:84% So every time you come to the gym, you pay p to go there. 00:24:04.930 --> 00:24:07.870 align:middle line:84% Consumers have these beta delta hyperbolic preferences 00:24:07.870 --> 00:24:10.660 align:middle line:84% and they say yes or no, I will pay my upfront fee l 00:24:10.660 --> 00:24:12.670 align:middle line:90% to join the gym. 00:24:12.670 --> 00:24:16.335 align:middle line:84% And then in their model there's uncertainty. 00:24:16.335 --> 00:24:19.100 align:middle line:84% You know you should join the gym and that it's good for you 00:24:19.100 --> 00:24:20.830 align:middle line:84% and you don't remember how unpleasant 00:24:20.830 --> 00:24:22.218 align:middle line:90% it is to go to the gym. 00:24:22.218 --> 00:24:24.010 align:middle line:84% Like, how long does it take to actually get 00:24:24.010 --> 00:24:29.230 align:middle line:84% there, how unpleasant is the exercising there, 00:24:29.230 --> 00:24:31.540 align:middle line:84% how gross is the showers at the gym, 00:24:31.540 --> 00:24:33.580 align:middle line:84% and whatever, how much are you going 00:24:33.580 --> 00:24:36.250 align:middle line:84% to dislike exercising once you get there. 00:24:36.250 --> 00:24:38.470 align:middle line:84% And so after you go to the gym, you 00:24:38.470 --> 00:24:41.710 align:middle line:84% learn your disutility d that you incur 00:24:41.710 --> 00:24:44.540 align:middle line:90% every time you go to the gym. 00:24:44.540 --> 00:24:48.440 align:middle line:84% And then you then choose how many times am I going 00:24:48.440 --> 00:24:49.700 align:middle line:90% to go to the gym this month. 00:24:49.700 --> 00:24:53.180 align:middle line:84% To simplify the algebra, they do either zero or one, 00:24:53.180 --> 00:24:55.823 align:middle line:84% but obviously we're thinking of one 00:24:55.823 --> 00:24:58.490 align:middle line:84% as being go to the gym every day and zero as never go to the gym 00:24:58.490 --> 00:24:59.600 align:middle line:90% or something like that. 00:24:59.600 --> 00:25:03.590 align:middle line:84% And then you incur disutility d plus p times x 00:25:03.590 --> 00:25:05.780 align:middle line:90% when you go to the gym. 00:25:05.780 --> 00:25:11.120 align:middle line:84% And then you get a delayed health benefit x times b. 00:25:11.120 --> 00:25:13.460 align:middle line:84% So you do feel better after you've been to the gym. 00:25:13.460 --> 00:25:16.910 align:middle line:84% You're happy that you went yesterday, and that's x times b. 00:25:16.910 --> 00:25:17.480 align:middle line:90% OK? 00:25:17.480 --> 00:25:20.780 align:middle line:84% And so the hyperbolic preferences come in here 00:25:20.780 --> 00:25:27.020 align:middle line:84% where the consumer at t equals one who's hyperbolic 00:25:27.020 --> 00:25:29.390 align:middle line:84% wants his future selves to go to the gym 00:25:29.390 --> 00:25:35.000 align:middle line:84% because he treats the disutility and the benefit as both having 00:25:35.000 --> 00:25:37.552 align:middle line:84% roughly equal weight because they're just both future things, 00:25:37.552 --> 00:25:39.260 align:middle line:84% and he thinks the future benefit is worth 00:25:39.260 --> 00:25:42.200 align:middle line:90% the future disutility, probably. 00:25:42.200 --> 00:25:45.170 align:middle line:84% But then the consumer who's actually here making 00:25:45.170 --> 00:25:47.480 align:middle line:84% the do I go to the gym today or not decision, 00:25:47.480 --> 00:25:50.420 align:middle line:84% that consumer is treating the disutility 00:25:50.420 --> 00:25:53.510 align:middle line:84% and the money paid as a present thing and the health benefit 00:25:53.510 --> 00:25:56.930 align:middle line:84% as a future thing, so that lazy future self is tempted 00:25:56.930 --> 00:26:00.080 align:middle line:84% to not go today and say, OK, the tomorrow 00:26:00.080 --> 00:26:02.650 align:middle line:84% self will be the one to go to the gym, not the today self. 00:26:02.650 --> 00:26:05.720 align:middle line:90% 00:26:05.720 --> 00:26:10.700 align:middle line:84% So what are some of the observations? 00:26:10.700 --> 00:26:15.620 align:middle line:84% First monopoly observation is that in this model, 00:26:15.620 --> 00:26:22.310 align:middle line:84% we get distorted pricing where p is distorted down. 00:26:22.310 --> 00:26:25.610 align:middle line:84% In their paper, they have a costs, marginal costs, 00:26:25.610 --> 00:26:27.570 align:middle line:84% c times x of having people go to the gym. 00:26:27.570 --> 00:26:30.000 align:middle line:84% I mean, c may be small, but the gym does-- 00:26:30.000 --> 00:26:31.610 align:middle line:90% they have to wash your towel. 00:26:31.610 --> 00:26:33.090 align:middle line:84% They may wipe down the equipment. 00:26:33.090 --> 00:26:35.690 align:middle line:84% They have some capacity levels, so they incur some small costs 00:26:35.690 --> 00:26:36.950 align:middle line:90% c when you come. 00:26:36.950 --> 00:26:42.980 align:middle line:84% But so the main distortion we get is we get p less than c. 00:26:42.980 --> 00:26:45.810 align:middle line:84% This can be going to the gym is free. 00:26:45.810 --> 00:26:48.060 align:middle line:84% You don't pay when you show up on your gym membership. 00:26:48.060 --> 00:26:51.260 align:middle line:84% Or it could be that p is even negative, that you don't pay 00:26:51.260 --> 00:26:53.570 align:middle line:84% and they give you a free juice or they give you 00:26:53.570 --> 00:26:56.570 align:middle line:84% some other things for free that they don't charge you 00:26:56.570 --> 00:27:00.720 align:middle line:90% for when you're at the gym. 00:27:00.720 --> 00:27:01.290 align:middle line:90% Why? 00:27:01.290 --> 00:27:02.800 align:middle line:90% Why do they distort p down? 00:27:02.800 --> 00:27:04.100 align:middle line:90% There are two reasons. 00:27:04.100 --> 00:27:07.470 align:middle line:84% First is let's think about the sophisticated hyperbolic model 00:27:07.470 --> 00:27:10.770 align:middle line:84% where this consumer knows that his future self is 00:27:10.770 --> 00:27:12.810 align:middle line:90% going to be lazy. 00:27:12.810 --> 00:27:15.660 align:middle line:84% So that means he knows there's going to be a problem of, 00:27:15.660 --> 00:27:17.080 align:middle line:84% I sign up for the gym membership. 00:27:17.080 --> 00:27:20.910 align:middle line:84% This guy is not going to go, and therefore, I'm 00:27:20.910 --> 00:27:22.650 align:middle line:84% not going to get the health benefit. 00:27:22.650 --> 00:27:25.500 align:middle line:84% So what happens is the sophisticated consumer 00:27:25.500 --> 00:27:29.890 align:middle line:84% wants to commit his future self to actually go to the gym. 00:27:29.890 --> 00:27:32.740 align:middle line:84% And how does he commit his future self to go to the gym? 00:27:32.740 --> 00:27:36.720 align:middle line:90% He makes the price. 00:27:36.720 --> 00:27:39.240 align:middle line:84% If you offer a contract where the p is negative-- 00:27:39.240 --> 00:27:41.970 align:middle line:84% there's all kinds of free stuff offered at the gym and they 00:27:41.970 --> 00:27:43.350 align:middle line:90% don't charge you any money-- 00:27:43.350 --> 00:27:45.727 align:middle line:84% then you're thinking, that lazy future self of mine 00:27:45.727 --> 00:27:47.310 align:middle line:84% is going to go to the gym because he's 00:27:47.310 --> 00:27:50.700 align:middle line:84% going to want to get the negative $10 fee for going 00:27:50.700 --> 00:27:52.710 align:middle line:84% to the gym, and then that's going 00:27:52.710 --> 00:27:56.490 align:middle line:84% to make him willing to incur the disutility to get the benefit. 00:27:56.490 --> 00:27:59.280 align:middle line:84% And so the sophisticated consumers 00:27:59.280 --> 00:28:02.320 align:middle line:84% want to have a contract-- and they're 00:28:02.320 --> 00:28:04.510 align:middle line:84% willing to pay a lot of money up front 00:28:04.510 --> 00:28:07.180 align:middle line:84% to get the contract with the low p that's 00:28:07.180 --> 00:28:10.400 align:middle line:84% going to force their future self to go to the gym. 00:28:10.400 --> 00:28:15.290 align:middle line:84% So that's the sophisticated argument. 00:28:15.290 --> 00:28:17.330 align:middle line:84% If consumers are naive hyperbolics 00:28:17.330 --> 00:28:19.430 align:middle line:84% instead of sophisticated hyperbolics, 00:28:19.430 --> 00:28:21.650 align:middle line:90% you also want them-- 00:28:21.650 --> 00:28:24.170 align:middle line:84% you also want to do p less than c. 00:28:24.170 --> 00:28:28.280 align:middle line:84% And the idea here is that the naive hyperbolic consumers 00:28:28.280 --> 00:28:30.610 align:middle line:84% misestimate how often they go to the gym. 00:28:30.610 --> 00:28:35.540 align:middle line:84% They say, OK, you know, I know b is bigger than d, therefore, 00:28:35.540 --> 00:28:39.417 align:middle line:84% if b is bigger-- if I know b is bigger than d and the price 00:28:39.417 --> 00:28:41.000 align:middle line:84% is zero, my future self is going to go 00:28:41.000 --> 00:28:42.970 align:middle line:90% to the gym every single day. 00:28:42.970 --> 00:28:45.470 align:middle line:84% And then what you can say is, so if you have this person who 00:28:45.470 --> 00:28:50.090 align:middle line:84% thinks they're going to go every-- so suppose that we have 00:28:50.090 --> 00:28:54.530 align:middle line:84% the benefit b is bigger than the disutility d, 00:28:54.530 --> 00:28:58.800 align:middle line:84% but the benefit b is less than beta times delta. 00:28:58.800 --> 00:28:59.300 align:middle line:90% OK? 00:28:59.300 --> 00:29:06.010 align:middle line:84% So this is the case where the future self thinks 00:29:06.010 --> 00:29:08.920 align:middle line:90% they're going to go to the-- 00:29:08.920 --> 00:29:09.940 align:middle line:90% Sorry. 00:29:09.940 --> 00:29:13.150 align:middle line:90% This is d, not delta. 00:29:13.150 --> 00:29:15.650 align:middle line:84% The future self thinks they're going to go to the gym, 00:29:15.650 --> 00:29:17.650 align:middle line:84% but they're not actually going to go to the gym. 00:29:17.650 --> 00:29:20.500 align:middle line:84% Because you think that they're going to be rational and trade 00:29:20.500 --> 00:29:24.820 align:middle line:84% off benefit versus disutility, but they're actually 00:29:24.820 --> 00:29:28.630 align:middle line:84% going to trade off disutility versus beta times benefit. 00:29:28.630 --> 00:29:31.210 align:middle line:90% So this is-- oops. 00:29:31.210 --> 00:29:35.830 align:middle line:84% So beta times b is less than the disutility. 00:29:35.830 --> 00:29:36.637 align:middle line:90% OK? 00:29:36.637 --> 00:29:37.970 align:middle line:90% Well, so here's what you can do. 00:29:37.970 --> 00:29:41.255 align:middle line:84% You can tell that consumer, OK, look. 00:29:41.255 --> 00:29:42.880 align:middle line:84% What we're going to do is go to the gym 00:29:42.880 --> 00:29:45.310 align:middle line:84% and instead of charging you cost every time you go, 00:29:45.310 --> 00:29:49.780 align:middle line:84% I'm going to give you $10 less than cost every time you go. 00:29:49.780 --> 00:29:52.000 align:middle line:90% And so you're going to save-- 00:29:52.000 --> 00:29:54.940 align:middle line:84% you're going to earn $300 in these bonus payments 00:29:54.940 --> 00:29:56.890 align:middle line:84% for going to the gym every day, and so 00:29:56.890 --> 00:29:59.560 align:middle line:84% for that, for the low, low price of $299 00:29:59.560 --> 00:30:01.773 align:middle line:84% a month, I'm going to give you that $10 every day. 00:30:01.773 --> 00:30:03.190 align:middle line:84% And the naive hyperbolic consumers 00:30:03.190 --> 00:30:05.360 align:middle line:84% are like, wow, that's an awesome deal. 00:30:05.360 --> 00:30:06.880 align:middle line:84% I have a negative gym membership, 00:30:06.880 --> 00:30:10.670 align:middle line:84% so they only pay 299 up front and then I get $300 back. 00:30:10.670 --> 00:30:15.617 align:middle line:84% But then obviously, you know, you're never going to actually-- 00:30:15.617 --> 00:30:17.700 align:middle line:84% if the negative 10 is not enough to get you to go, 00:30:17.700 --> 00:30:19.050 align:middle line:84% you're never actually going to get to go. 00:30:19.050 --> 00:30:20.633 align:middle line:84% You've paid 299 up front for something 00:30:20.633 --> 00:30:21.980 align:middle line:90% that was worth zero to you. 00:30:21.980 --> 00:30:27.920 align:middle line:84% And because you're going to overestimate your gym usage, 00:30:27.920 --> 00:30:29.270 align:middle line:90% you then end up-- 00:30:29.270 --> 00:30:31.770 align:middle line:84% you want to charge them with that kind of scheme. 00:30:31.770 --> 00:30:35.030 align:middle line:90% 00:30:35.030 --> 00:30:36.380 align:middle line:90% OK. 00:30:36.380 --> 00:30:39.770 align:middle line:84% Third observation here is in this sophisticated model, 00:30:39.770 --> 00:30:42.490 align:middle line:90% the p is socially efficient. 00:30:42.490 --> 00:30:44.070 align:middle line:90% There isn't actually distortion. 00:30:44.070 --> 00:30:46.430 align:middle line:84% You're charging them less than cost, 00:30:46.430 --> 00:30:48.173 align:middle line:84% but the price that you charge them, 00:30:48.173 --> 00:30:50.090 align:middle line:84% even though p is less-- we normally think of p 00:30:50.090 --> 00:30:53.003 align:middle line:84% equals c as what gives us full social efficiency. 00:30:53.003 --> 00:30:54.920 align:middle line:84% Here it's actually p less than c that gives us 00:30:54.920 --> 00:30:58.400 align:middle line:90% full social efficiency. 00:30:58.400 --> 00:31:00.240 align:middle line:90% Do I have that on here? 00:31:00.240 --> 00:31:00.740 align:middle line:90% Yeah. 00:31:00.740 --> 00:31:01.750 align:middle line:90% Why is that? 00:31:01.750 --> 00:31:05.090 align:middle line:90% 00:31:05.090 --> 00:31:07.490 align:middle line:84% You know, in some ways this is like back 00:31:07.490 --> 00:31:09.425 align:middle line:90% to lecture one of this class. 00:31:09.425 --> 00:31:11.300 align:middle line:84% If you remember in lecture one of this class, 00:31:11.300 --> 00:31:15.050 align:middle line:84% I had if you have a firm choosing product quality, 00:31:15.050 --> 00:31:17.995 align:middle line:84% they choose the optimal quality for the marginal consumer. 00:31:17.995 --> 00:31:21.108 align:middle line:90% 00:31:21.108 --> 00:31:23.190 align:middle line:84% In the Della Vigna-Malmendier paper, 00:31:23.190 --> 00:31:25.320 align:middle line:84% all the consumers are ex-ante identical, 00:31:25.320 --> 00:31:28.090 align:middle line:84% so the marginal consumer is the average consumer. 00:31:28.090 --> 00:31:30.240 align:middle line:84% Therefore, what you do is you always choose 00:31:30.240 --> 00:31:32.370 align:middle line:84% the optimal product quality and just extract 00:31:32.370 --> 00:31:34.805 align:middle line:90% all of that with a fixed fee. 00:31:34.805 --> 00:31:36.220 align:middle line:90% That's what this model is here. 00:31:36.220 --> 00:31:39.765 align:middle line:84% It's like p is just a dimension of quality. 00:31:39.765 --> 00:31:43.740 align:middle line:84% A lower p gym contract, a lower per visit contract, 00:31:43.740 --> 00:31:48.510 align:middle line:84% is a contract that has a higher quality, 00:31:48.510 --> 00:31:52.140 align:middle line:84% and therefore, you just choose the optimal quality 00:31:52.140 --> 00:31:54.770 align:middle line:84% for the consumers and then you extract all the fixed fee. 00:31:54.770 --> 00:31:56.880 align:middle line:84% It's fully efficient from the point 00:31:56.880 --> 00:31:59.280 align:middle line:84% of view of the sophisticated hyperbolic consumer who 00:31:59.280 --> 00:32:01.770 align:middle line:90% is the one buying it. 00:32:01.770 --> 00:32:09.750 align:middle line:84% And more generally, that theorem from the first class 00:32:09.750 --> 00:32:14.825 align:middle line:84% was that it's efficient in terms of maximizing the sum of-- 00:32:14.825 --> 00:32:16.200 align:middle line:84% it's socially efficient, in terms 00:32:16.200 --> 00:32:22.135 align:middle line:84% of maximizing the sum of profits of the firm 00:32:22.135 --> 00:32:23.525 align:middle line:90% and the utility of the consumer. 00:32:23.525 --> 00:32:26.150 align:middle line:84% But it wasn't really the utility of the consumer that mattered. 00:32:26.150 --> 00:32:28.280 align:middle line:84% It was the willingness to pay of the consumer. 00:32:28.280 --> 00:32:30.700 align:middle line:84% And so if you make the willingness to pay different 00:32:30.700 --> 00:32:34.030 align:middle line:84% from the utility, then a monopolist 00:32:34.030 --> 00:32:35.650 align:middle line:84% serving a homogeneous population is 00:32:35.650 --> 00:32:39.850 align:middle line:84% going to choose the quality that maximizes the sum of profits 00:32:39.850 --> 00:32:41.042 align:middle line:90% and willingness to pay. 00:32:41.042 --> 00:32:43.000 align:middle line:84% And so it's going to be efficient in that sense 00:32:43.000 --> 00:32:45.620 align:middle line:84% of willingness to pay as utility. 00:32:45.620 --> 00:32:48.040 align:middle line:84% And so that's what we have in the sophisticated hyperbolic 00:32:48.040 --> 00:32:48.590 align:middle line:90% consumer. 00:32:48.590 --> 00:32:53.950 align:middle line:84% You're always going to maximize the utility that you get. 00:32:53.950 --> 00:32:56.830 align:middle line:90% And this is going to-- 00:32:56.830 --> 00:32:58.360 align:middle line:84% in some ways, a general lesson here 00:32:58.360 --> 00:33:01.270 align:middle line:84% is that it's not like competition 00:33:01.270 --> 00:33:05.440 align:middle line:84% would force you to eliminate these distorting consumers. 00:33:05.440 --> 00:33:08.290 align:middle line:84% Competition forces you to exploit behavioral consumers 00:33:08.290 --> 00:33:11.080 align:middle line:84% because competition forces you to design a product 00:33:11.080 --> 00:33:14.050 align:middle line:84% quality or a contract that maximizes willingness 00:33:14.050 --> 00:33:16.730 align:middle line:84% to pay and then give all the rents back to the consumers. 00:33:16.730 --> 00:33:19.210 align:middle line:84% So whether it's monopoly or perfect competition, 00:33:19.210 --> 00:33:22.240 align:middle line:84% firms will always exploit behavioral consumers 00:33:22.240 --> 00:33:24.010 align:middle line:84% because they're just always trying 00:33:24.010 --> 00:33:25.745 align:middle line:90% to maximize willingness to pay. 00:33:25.745 --> 00:33:28.120 align:middle line:84% And it doesn't matter that willingness to pay is utility. 00:33:28.120 --> 00:33:30.190 align:middle line:84% You're just trying to maximize willingness to pay. 00:33:30.190 --> 00:33:32.190 align:middle line:84% If willingness to pay and utility are different, 00:33:32.190 --> 00:33:34.980 align:middle line:84% you maximize willingness to pay, which is exploiting people. 00:33:34.980 --> 00:33:39.880 align:middle line:90% 00:33:39.880 --> 00:33:44.950 align:middle line:84% So they have one of the first behavioral empirical papers 00:33:44.950 --> 00:33:51.670 align:middle line:84% trying to analyze this idea in a data set on health club 00:33:51.670 --> 00:33:55.540 align:middle line:84% memberships in the Boston area, and the idea 00:33:55.540 --> 00:33:58.720 align:middle line:90% is it's basically a reduced-- 00:33:58.720 --> 00:34:00.390 align:middle line:84% one of these reduced form papers where 00:34:00.390 --> 00:34:01.990 align:middle line:84% we try to say, what are predictions 00:34:01.990 --> 00:34:03.490 align:middle line:90% of the rational model? 00:34:03.490 --> 00:34:07.180 align:middle line:84% What would be not necessarily predictions 00:34:07.180 --> 00:34:09.280 align:middle line:84% of the hyperbolic model, but directions 00:34:09.280 --> 00:34:12.190 align:middle line:84% in which we think pricing of hyperbolic consumers 00:34:12.190 --> 00:34:14.440 align:middle line:84% would differ from pricing with rational consumers? 00:34:14.440 --> 00:34:17.020 align:middle line:84% And what they're trying to do is reject the rational model 00:34:17.020 --> 00:34:19.570 align:middle line:84% of health club consumers and show 00:34:19.570 --> 00:34:21.550 align:middle line:84% that some kind of hyperbolic-- it's 00:34:21.550 --> 00:34:24.250 align:middle line:84% departing from that hyperbolic direction. 00:34:24.250 --> 00:34:29.620 align:middle line:84% So anyway, the data set has 7,000 members at three Boston 00:34:29.620 --> 00:34:31.067 align:middle line:90% area health clubs. 00:34:31.067 --> 00:34:32.650 align:middle line:84% A lot of what they have is just a data 00:34:32.650 --> 00:34:34.310 align:middle line:84% set on the usage of these health clubs, 00:34:34.310 --> 00:34:35.679 align:middle line:84% and then they do also have a survey data 00:34:35.679 --> 00:34:37.387 align:middle line:84% where they stood outside the health clubs 00:34:37.387 --> 00:34:40.370 align:middle line:84% and just asked people survey questions about how much they 00:34:40.370 --> 00:34:41.989 align:middle line:90% use the gym and stuff. 00:34:41.989 --> 00:34:45.739 align:middle line:84% So first, inconsistent with the fully rational model. 00:34:45.739 --> 00:34:48.600 align:middle line:84% Customers who sign up for the gym just pay too much. 00:34:48.600 --> 00:34:51.800 align:middle line:90% 00:34:51.800 --> 00:34:55.909 align:middle line:84% One of their plans their health club offered was a $70 a month 00:34:55.909 --> 00:34:58.190 align:middle line:90% unlimited gym pass. 00:34:58.190 --> 00:34:59.160 align:middle line:90% You just pay $70. 00:34:59.160 --> 00:35:01.250 align:middle line:84% You can go to the gym as many times as you like. 00:35:01.250 --> 00:35:04.490 align:middle line:84% Those who sign up for the $70 a month plan 00:35:04.490 --> 00:35:07.230 align:middle line:84% averaged visiting 4.3 times a month, 00:35:07.230 --> 00:35:10.160 align:middle line:84% and so they were paying, like, $17 every time 00:35:10.160 --> 00:35:11.113 align:middle line:90% they went to the gym. 00:35:11.113 --> 00:35:13.280 align:middle line:84% The health club also had a plan where you could just 00:35:13.280 --> 00:35:18.650 align:middle line:84% get a coupon book, which was 10 coupons for $100 where you just 00:35:18.650 --> 00:35:21.080 align:middle line:90% pay $10 on a per visit basis. 00:35:21.080 --> 00:35:25.650 align:middle line:84% Clearly it's not fully rational model. 00:35:25.650 --> 00:35:27.578 align:middle line:84% You should not be paying $70 a month 00:35:27.578 --> 00:35:29.370 align:middle line:84% for a gym you're going to go to four times. 00:35:29.370 --> 00:35:32.420 align:middle line:84% You should just buy the book that gives you the 10 visits 00:35:32.420 --> 00:35:33.693 align:middle line:90% and just use those coupons. 00:35:33.693 --> 00:35:36.470 align:middle line:90% 00:35:36.470 --> 00:35:39.440 align:middle line:84% So either the consumers are misestimating how often they're 00:35:39.440 --> 00:35:43.790 align:middle line:84% going or the consumers value the commitment 00:35:43.790 --> 00:35:46.085 align:middle line:84% and are buying the gym membership because if they 00:35:46.085 --> 00:35:48.710 align:middle line:84% didn't buy the gym membership, they would go zero times instead 00:35:48.710 --> 00:35:51.590 align:middle line:84% of four times if they were actually going to pay $10 00:35:51.590 --> 00:35:52.740 align:middle line:90% every time they went. 00:35:52.740 --> 00:35:54.920 align:middle line:84% So we don't know which of those, but both 00:35:54.920 --> 00:35:56.930 align:middle line:84% of those naive and sophisticated hyperbolic 00:35:56.930 --> 00:35:59.480 align:middle line:84% models predict that consumers should be in this situation 00:35:59.480 --> 00:36:04.460 align:middle line:84% where they're paying too much ex-post for the gym visits. 00:36:04.460 --> 00:36:10.730 align:middle line:84% Second thing of showing that consumers are not fully rational 00:36:10.730 --> 00:36:15.530 align:middle line:84% is this company had two types of plans. 00:36:15.530 --> 00:36:19.280 align:middle line:84% Besides the daily, the per visit pass, they had a monthly pass 00:36:19.280 --> 00:36:22.430 align:middle line:90% and they had an annual pass. 00:36:22.430 --> 00:36:25.160 align:middle line:84% The monthly pass might have been like, you know-- 00:36:25.160 --> 00:36:26.300 align:middle line:90% I forget the numbers. 00:36:26.300 --> 00:36:27.713 align:middle line:90% $70 a month. 00:36:27.713 --> 00:36:29.630 align:middle line:84% But then you also may have had a choice of you 00:36:29.630 --> 00:36:34.850 align:middle line:84% could also pay upfront $700 and get a full year membership, 00:36:34.850 --> 00:36:39.290 align:middle line:84% paying less per month if you're gung ho enough on the gym 00:36:39.290 --> 00:36:42.200 align:middle line:90% that you would pay $700 fee. 00:36:42.200 --> 00:36:47.540 align:middle line:84% And they note that people who pay the $700 fee up front 00:36:47.540 --> 00:36:50.820 align:middle line:84% do go to the gym more than the people who pay the $70 a month. 00:36:50.820 --> 00:36:53.240 align:middle line:84% So it's as if the people who buy the annual pass 00:36:53.240 --> 00:36:55.370 align:middle line:84% know that they are people who go to the gym 00:36:55.370 --> 00:36:57.470 align:middle line:84% and they're willing to pay the $700 00:36:57.470 --> 00:37:00.590 align:middle line:84% check up front to get the annual membership. 00:37:00.590 --> 00:37:02.630 align:middle line:84% But then one of the interesting observations 00:37:02.630 --> 00:37:04.850 align:middle line:84% is that the monthly plan, they charge 00:37:04.850 --> 00:37:07.100 align:middle line:84% to your credit card every month until you do something 00:37:07.100 --> 00:37:08.330 align:middle line:90% to get them to stop. 00:37:08.330 --> 00:37:10.850 align:middle line:84% The $700 plan, you actually have to-- it 00:37:10.850 --> 00:37:13.700 align:middle line:84% didn't charge $700 to your credit card a year from now. 00:37:13.700 --> 00:37:15.950 align:middle line:84% You had to actually go in and sign up 00:37:15.950 --> 00:37:17.750 align:middle line:90% for another annual membership. 00:37:17.750 --> 00:37:20.990 align:middle line:84% And what they find is if you look 13 months later, 00:37:20.990 --> 00:37:23.210 align:middle line:84% people who signed up for the $70 plan 00:37:23.210 --> 00:37:25.580 align:middle line:84% are more likely to be in the gym than the people who 00:37:25.580 --> 00:37:27.980 align:middle line:90% signed up for the $700 plan. 00:37:27.980 --> 00:37:30.047 align:middle line:84% And it's because these people just 00:37:30.047 --> 00:37:31.880 align:middle line:84% got auto renewed if they didn't do anything, 00:37:31.880 --> 00:37:35.060 align:middle line:84% whereas these people had to go in and actually decide, 00:37:35.060 --> 00:37:38.335 align:middle line:84% yes, affirmatively, I do want to still be signed up for the gym. 00:37:38.335 --> 00:37:40.710 align:middle line:84% And so it's sometimes telling that sometimes these people 00:37:40.710 --> 00:37:42.010 align:middle line:90% aren't paying attention. 00:37:42.010 --> 00:37:45.420 align:middle line:84% Another statistic about this was that on average, their consumers 00:37:45.420 --> 00:37:50.490 align:middle line:84% spend $187 after the very last time they ever went to the gym. 00:37:50.490 --> 00:37:52.470 align:middle line:84% So if you look at people who quit the gym, 00:37:52.470 --> 00:37:57.390 align:middle line:84% they've paid $187 since the very last time that they visited. 00:37:57.390 --> 00:37:59.826 align:middle line:84% This is saying that people aren't paying attention-- 00:37:59.826 --> 00:38:02.332 align:middle line:90% 00:38:02.332 --> 00:38:04.040 align:middle line:84% aren't fully paying attention and the gym 00:38:04.040 --> 00:38:06.770 align:middle line:84% is designing a payment scheme with the auto renewal 00:38:06.770 --> 00:38:09.600 align:middle line:84% on your credit card to exploit that. 00:38:09.600 --> 00:38:10.860 align:middle line:90% OK. 00:38:10.860 --> 00:38:14.460 align:middle line:84% Other observations that suggest the partially naive model 00:38:14.460 --> 00:38:19.410 align:middle line:84% may be better than the fully rational explanation for this. 00:38:19.410 --> 00:38:24.090 align:middle line:84% You ask people in surveys, how often will you go to the gym? 00:38:24.090 --> 00:38:26.040 align:middle line:84% And people say that they're going 00:38:26.040 --> 00:38:29.080 align:middle line:84% to go to the gym more than twice as much as they actually go. 00:38:29.080 --> 00:38:31.287 align:middle line:84% So you ask the people who have-- you 00:38:31.287 --> 00:38:33.870 align:middle line:84% ask people who have signed up for the plans, how many times do 00:38:33.870 --> 00:38:35.078 align:middle line:90% you think you're going to go? 00:38:35.078 --> 00:38:37.470 align:middle line:90% And they get that wrong. 00:38:37.470 --> 00:38:41.970 align:middle line:84% And then also, there's a correlation 00:38:41.970 --> 00:38:44.700 align:middle line:84% between lower usage of the gym and taking a longer 00:38:44.700 --> 00:38:47.080 align:middle line:84% time to cancel after you stop going to the gym. 00:38:47.080 --> 00:38:48.957 align:middle line:84% And so this is a sense that there 00:38:48.957 --> 00:38:50.790 align:middle line:84% seem to be some people who are more rational 00:38:50.790 --> 00:38:52.360 align:middle line:84% and some people who are less rational, 00:38:52.360 --> 00:38:53.735 align:middle line:84% and then the less rational people 00:38:53.735 --> 00:38:56.340 align:middle line:84% do multiple things that seem irrational together, 00:38:56.340 --> 00:38:59.850 align:middle line:84% like the not canceling and the not going so much. 00:38:59.850 --> 00:39:06.008 align:middle line:84% Again, they don't estimate a specific behavioral model 00:39:06.008 --> 00:39:08.300 align:middle line:84% and show that specific behavioral model can rationalize 00:39:08.300 --> 00:39:10.800 align:middle line:84% all the things, but it's saying that the rational model just 00:39:10.800 --> 00:39:12.800 align:middle line:90% doesn't work. 00:39:12.800 --> 00:39:13.300 align:middle line:90% OK. 00:39:13.300 --> 00:39:16.060 align:middle line:90% 00:39:16.060 --> 00:39:16.560 align:middle line:90% OK. 00:39:16.560 --> 00:39:18.750 align:middle line:84% So I thought next thing I would do 00:39:18.750 --> 00:39:23.700 align:middle line:84% is cover in more detail one theoretical paper 00:39:23.700 --> 00:39:27.340 align:middle line:84% to just give you more of a sense of how these papers work. 00:39:27.340 --> 00:39:29.460 align:middle line:84% So this is a paper by Heidhues and Koszegi 00:39:29.460 --> 00:39:33.150 align:middle line:90% about credit markets. 00:39:33.150 --> 00:39:38.640 align:middle line:84% And I think there's a feeling that many people have 00:39:38.640 --> 00:39:39.580 align:middle line:90% too much debt. 00:39:39.580 --> 00:39:41.497 align:middle line:84% In the last financial crisis, there was a view 00:39:41.497 --> 00:39:42.880 align:middle line:90% that people have too much debt. 00:39:42.880 --> 00:39:46.170 align:middle line:84% There's also an observation that many firms offer credit card 00:39:46.170 --> 00:39:49.440 align:middle line:84% deals that look like things like sign up for our credit card, 00:39:49.440 --> 00:39:53.280 align:middle line:84% pay 0.9 interest for the first six months 00:39:53.280 --> 00:39:57.010 align:middle line:84% to pay off your other debts, and then it goes up to 19.8% 00:39:57.010 --> 00:39:59.260 align:middle line:84% And there's this use of these teaser rates, 00:39:59.260 --> 00:40:01.718 align:middle line:84% and then people think that, are those teaser rates 00:40:01.718 --> 00:40:03.510 align:middle line:84% designed to get people to take on more debt 00:40:03.510 --> 00:40:06.720 align:middle line:84% than they should take on or more debt than they want to take on? 00:40:06.720 --> 00:40:10.530 align:middle line:84% And so that's the idea of the Heidhues-Koszegi paper is 00:40:10.530 --> 00:40:17.450 align:middle line:84% to think about, would quasi-hyperbolic consumers-- 00:40:17.450 --> 00:40:20.540 align:middle line:84% would the market exploit quasi-hyperbolic consumers 00:40:20.540 --> 00:40:23.990 align:middle line:84% in a way that creates them taking on too much debt 00:40:23.990 --> 00:40:26.360 align:middle line:84% and would it explain why it is we 00:40:26.360 --> 00:40:32.110 align:middle line:84% see that these contracts that have initial upfront teaser 00:40:32.110 --> 00:40:36.510 align:middle line:84% rates followed by higher rates later on? 00:40:36.510 --> 00:40:38.140 align:middle line:90% So in their model-- 00:40:38.140 --> 00:40:42.310 align:middle line:84% and this becomes-- it's a nice paper in that it's 00:40:42.310 --> 00:40:45.640 align:middle line:84% a paper that shows there's this very, very sharp discontinuity 00:40:45.640 --> 00:40:51.460 align:middle line:84% where, with fully sophisticated quasi-hyperbolic consumers, 00:40:51.460 --> 00:40:54.670 align:middle line:84% one thing happens and then you make even an epsilon difference 00:40:54.670 --> 00:40:58.820 align:middle line:84% between their self-recognition and full self-recognition, 00:40:58.820 --> 00:41:00.730 align:middle line:84% you get a big discrete jump in behavior 00:41:00.730 --> 00:41:05.230 align:middle line:84% with an epsilon change in the behavioral assumption. 00:41:05.230 --> 00:41:06.940 align:middle line:84% So anyway, we have this agent, they 00:41:06.940 --> 00:41:09.448 align:middle line:84% have the beta, beta hat one preferences. 00:41:09.448 --> 00:41:10.990 align:middle line:84% So just we're eliminating discounting 00:41:10.990 --> 00:41:12.820 align:middle line:90% because t equals three. 00:41:12.820 --> 00:41:15.340 align:middle line:84% So the consumer is going to borrow some money at t 00:41:15.340 --> 00:41:17.440 align:middle line:90% equals zero. 00:41:17.440 --> 00:41:21.220 align:middle line:84% Think of this as taking out a car loan. 00:41:21.220 --> 00:41:23.590 align:middle line:84% I'm borrowing some money c to buy a car, 00:41:23.590 --> 00:41:25.870 align:middle line:84% and then I buy a durable good and I 00:41:25.870 --> 00:41:31.450 align:middle line:84% get utility from the durable good at t equals one. 00:41:31.450 --> 00:41:34.060 align:middle line:84% So there's some sense there's a time 00:41:34.060 --> 00:41:37.390 align:middle line:84% zero where I pay the money and then a time one 00:41:37.390 --> 00:41:40.008 align:middle line:84% where I enjoy the durable good, and then time two, 00:41:40.008 --> 00:41:41.800 align:middle line:84% I could still be enjoying the durable good. 00:41:41.800 --> 00:41:43.870 align:middle line:84% They do that the durable good is already worn out 00:41:43.870 --> 00:41:45.537 align:middle line:84% and I still just have my debt for my car 00:41:45.537 --> 00:41:47.080 align:middle line:90% after I've already sold my car. 00:41:47.080 --> 00:41:49.640 align:middle line:84% But anyway, there's I borrowed money. 00:41:49.640 --> 00:41:52.180 align:middle line:84% I get utility of t equals one from my durable good. 00:41:52.180 --> 00:41:55.540 align:middle line:84% At t equals one I'm making my car payments, and then at t 00:41:55.540 --> 00:41:57.930 align:middle line:84% equals two, I'm still making an additional car payment. 00:41:57.930 --> 00:42:01.160 align:middle line:90% 00:42:01.160 --> 00:42:03.080 align:middle line:84% So they have this conflict of interest 00:42:03.080 --> 00:42:07.760 align:middle line:84% between how the different selves will view the car payments 00:42:07.760 --> 00:42:09.540 align:middle line:90% versus the enjoying the car. 00:42:09.540 --> 00:42:11.960 align:middle line:84% So from the point of view of the time zero self, which 00:42:11.960 --> 00:42:15.075 align:middle line:84% is before he's going to enjoy the car 00:42:15.075 --> 00:42:16.950 align:middle line:84% and before he's going to make the repayments, 00:42:16.950 --> 00:42:22.910 align:middle line:84% time zero self's preferences are c minus k of q minus k of r. 00:42:22.910 --> 00:42:29.870 align:middle line:84% So there's this convex function about your repayment. 00:42:29.870 --> 00:42:33.050 align:middle line:84% You have this convex disutility of a repayment 00:42:33.050 --> 00:42:35.630 align:middle line:84% because the more you have to repay in your monthly car 00:42:35.630 --> 00:42:37.880 align:middle line:84% payment, the more you're cutting back on other things, 00:42:37.880 --> 00:42:41.000 align:middle line:84% and the more things you cut back on increasing disutility 00:42:41.000 --> 00:42:42.470 align:middle line:84% for cutting back on food as you're 00:42:42.470 --> 00:42:44.525 align:middle line:90% cutting back more and more. 00:42:44.525 --> 00:42:46.580 align:middle line:84% And so the time zero self utility 00:42:46.580 --> 00:42:49.430 align:middle line:84% is c is the enjoyment of the car minus-- 00:42:49.430 --> 00:42:51.383 align:middle line:84% all these things would be multiplied by beta 00:42:51.383 --> 00:42:52.800 align:middle line:84% because they're all in the future, 00:42:52.800 --> 00:42:55.350 align:middle line:84% but we ignore that fact because they're all in the future. 00:42:55.350 --> 00:42:57.850 align:middle line:84% So from the time zero self, these things are all comparable, 00:42:57.850 --> 00:43:00.975 align:middle line:84% and it's c minus k of q minus k of r. 00:43:00.975 --> 00:43:03.690 align:middle line:90% 00:43:03.690 --> 00:43:06.550 align:middle line:84% From the time one's perspective self, 00:43:06.550 --> 00:43:11.470 align:middle line:84% time one cares about the car and the repayment made at t 00:43:11.470 --> 00:43:12.560 align:middle line:90% equals one. 00:43:12.560 --> 00:43:15.693 align:middle line:84% By the time one self discounts the pain 00:43:15.693 --> 00:43:17.110 align:middle line:84% he's going to suffer repaying at t 00:43:17.110 --> 00:43:19.540 align:middle line:84% equals two because that's a future thing. 00:43:19.540 --> 00:43:22.540 align:middle line:84% And then by the time it gets to the time two self, the time two 00:43:22.540 --> 00:43:25.240 align:middle line:84% self dislikes the repayment that the time two 00:43:25.240 --> 00:43:26.535 align:middle line:90% self is being forced to make. 00:43:26.535 --> 00:43:31.630 align:middle line:90% 00:43:31.630 --> 00:43:33.610 align:middle line:84% We're going to assume that the credit market is 00:43:33.610 --> 00:43:34.970 align:middle line:90% perfectly competitive. 00:43:34.970 --> 00:43:36.940 align:middle line:90% It has zero interest rates. 00:43:36.940 --> 00:43:39.970 align:middle line:84% And firms can offer these option contracts 00:43:39.970 --> 00:43:42.940 align:middle line:84% where you pay, you get some amount of c 00:43:42.940 --> 00:43:45.370 align:middle line:84% is given to you up front, and then 00:43:45.370 --> 00:43:49.180 align:middle line:84% you have these repayment plans you can choose among, 00:43:49.180 --> 00:43:52.787 align:middle line:84% q sub s, r sub s, and at t equals one 00:43:52.787 --> 00:43:54.620 align:middle line:84% you choose how you're how you want to repay. 00:43:54.620 --> 00:43:56.620 align:middle line:84% Do you want to make the immediate repayment at t 00:43:56.620 --> 00:43:57.230 align:middle line:90% equals one? 00:43:57.230 --> 00:43:59.890 align:middle line:84% Do you want to delay your payment and spread it out over t 00:43:59.890 --> 00:44:01.460 align:middle line:90% equals one and t equals two? 00:44:01.460 --> 00:44:04.790 align:middle line:84% Or maybe they have many, many different possible plans. 00:44:04.790 --> 00:44:05.290 align:middle line:90% OK. 00:44:05.290 --> 00:44:07.930 align:middle line:90% 00:44:07.930 --> 00:44:09.920 align:middle line:84% Suppose we have a rational consumer. 00:44:09.920 --> 00:44:11.900 align:middle line:84% What does a rational consumer do? 00:44:11.900 --> 00:44:15.340 align:middle line:84% Well, the rational consumer, you're just maximizing this, 00:44:15.340 --> 00:44:17.860 align:middle line:84% and there's going to be a break even constraint 00:44:17.860 --> 00:44:19.960 align:middle line:84% in the competitive market that is going to have 00:44:19.960 --> 00:44:22.930 align:middle line:90% to be c equals q plus r. 00:44:22.930 --> 00:44:25.720 align:middle line:84% The competitive interest lenders are 00:44:25.720 --> 00:44:29.250 align:middle line:84% going to always offer you the full amount of the repayments 00:44:29.250 --> 00:44:31.410 align:middle line:84% as your upfront price of the loan. 00:44:31.410 --> 00:44:35.270 align:middle line:84% So we're just trying to maximize this. 00:44:35.270 --> 00:44:41.120 align:middle line:84% So we're trying to maximize c minus k of q 00:44:41.120 --> 00:44:47.590 align:middle line:84% minus k of r subject to the zero profit constraint. 00:44:47.590 --> 00:44:48.775 align:middle line:90% OK. 00:44:48.775 --> 00:44:52.660 align:middle line:90% What's the solution to that? 00:44:52.660 --> 00:44:54.100 align:middle line:90% Lagrangians are all ones. 00:44:54.100 --> 00:44:55.630 align:middle line:84% You differentiate with respect to q, 00:44:55.630 --> 00:44:58.570 align:middle line:84% and the repayments are going to have 00:44:58.570 --> 00:45:04.270 align:middle line:84% to satisfy the marginal disutility of repaying a dollar 00:45:04.270 --> 00:45:08.380 align:middle line:84% when you have a payment of q is equal to one 00:45:08.380 --> 00:45:13.830 align:middle line:84% and the marginal value of repaying a dollar 00:45:13.830 --> 00:45:15.970 align:middle line:84% when you're paying r is equal to one. 00:45:15.970 --> 00:45:21.750 align:middle line:84% And so you repay evenly, setting these repayment rates 00:45:21.750 --> 00:45:24.610 align:middle line:84% so that the disutility of paying that dollar is one, 00:45:24.610 --> 00:45:26.790 align:middle line:84% and then you set c equal to q plus r. 00:45:26.790 --> 00:45:29.460 align:middle line:90% 00:45:29.460 --> 00:45:32.910 align:middle line:84% Anyway, the k prime at q equals one pins down 00:45:32.910 --> 00:45:34.920 align:middle line:84% how big the loan should be, and then 00:45:34.920 --> 00:45:38.535 align:middle line:84% you just repay evenly over the two periods. 00:45:38.535 --> 00:45:41.580 align:middle line:84% And the rational loan contract doesn't have 00:45:41.580 --> 00:45:42.850 align:middle line:90% to be some complicated thing. 00:45:42.850 --> 00:45:46.320 align:middle line:84% The rational loan contract just says, you know, here you go. 00:45:46.320 --> 00:45:47.320 align:middle line:90% You get no choices. 00:45:47.320 --> 00:45:50.610 align:middle line:84% You get c first best, and then you make those payments 00:45:50.610 --> 00:45:55.380 align:middle line:90% q first best, r first best. 00:45:55.380 --> 00:45:57.035 align:middle line:84% There's exactly one repayment option. 00:45:57.035 --> 00:45:58.160 align:middle line:90% That's what you have to do. 00:45:58.160 --> 00:46:05.880 align:middle line:90% 00:46:05.880 --> 00:46:08.190 align:middle line:90% OK. 00:46:08.190 --> 00:46:13.680 align:middle line:84% Imagine that the consumers are sophisticated. 00:46:13.680 --> 00:46:15.880 align:middle line:84% So you have sophisticated hyperbolic. 00:46:15.880 --> 00:46:19.080 align:middle line:84% So the consumers are hyperbolic, but they're sophisticated so 00:46:19.080 --> 00:46:20.790 align:middle line:84% that they have beta delta preferences, 00:46:20.790 --> 00:46:24.080 align:middle line:84% but they recognize their future self as beta delta preference. 00:46:24.080 --> 00:46:29.910 align:middle line:84% Then the time zero consumer, though hyperbolic, 00:46:29.910 --> 00:46:32.230 align:middle line:84% the time zero consumer here is not getting the benefit. 00:46:32.230 --> 00:46:33.780 align:middle line:84% It's the time one and two consumers 00:46:33.780 --> 00:46:35.370 align:middle line:90% who get the benefit of the car. 00:46:35.370 --> 00:46:40.200 align:middle line:84% So the time zero consumer has essentially rational preferences 00:46:40.200 --> 00:46:42.930 align:middle line:84% over what he wants the future selves to do. 00:46:42.930 --> 00:46:47.250 align:middle line:84% So in a sophisticated hyperbolic model, 00:46:47.250 --> 00:46:50.070 align:middle line:84% a sophisticated hyperbolic model, 00:46:50.070 --> 00:46:54.090 align:middle line:84% the monopolist is still going to offer the rational contract. 00:46:54.090 --> 00:46:56.070 align:middle line:84% And so you're going to commit your future self 00:46:56.070 --> 00:46:58.080 align:middle line:84% to make those two even payments, because that's 00:46:58.080 --> 00:47:00.840 align:middle line:84% what is best for the future selves 00:47:00.840 --> 00:47:03.540 align:middle line:84% to do from the point of view of the time zero purchaser. 00:47:03.540 --> 00:47:08.540 align:middle line:84% So in some sense, a lending market 00:47:08.540 --> 00:47:11.060 align:middle line:84% sells commitment to the time zero consumers who 00:47:11.060 --> 00:47:13.130 align:middle line:84% want commitment and lets them overcome 00:47:13.130 --> 00:47:15.830 align:middle line:90% their hyperbolic preferences. 00:47:15.830 --> 00:47:20.210 align:middle line:84% Again, they get their first best. 00:47:20.210 --> 00:47:23.300 align:middle line:84% We get the first best buy just committed to the contract 00:47:23.300 --> 00:47:24.087 align:middle line:90% is the first best. 00:47:24.087 --> 00:47:25.045 align:middle line:90% That's the only option. 00:47:25.045 --> 00:47:28.310 align:middle line:90% 00:47:28.310 --> 00:47:30.800 align:middle line:84% Then the interesting theorem is what 00:47:30.800 --> 00:47:34.370 align:middle line:84% happens when beta hat is different from beta. 00:47:34.370 --> 00:47:37.010 align:middle line:84% The result is that when beta hat is greater than beta, 00:47:37.010 --> 00:47:40.830 align:middle line:84% you think you're even epsilon less hyperbolic than you are. 00:47:40.830 --> 00:47:43.310 align:middle line:84% There's a discontinuity in the contract. 00:47:43.310 --> 00:47:48.020 align:middle line:84% And instead of offering just this straightforward no option 00:47:48.020 --> 00:47:52.610 align:middle line:84% contract, firms now offer an option contract. 00:47:52.610 --> 00:47:56.000 align:middle line:84% The contract looks like you get some amount c up front, 00:47:56.000 --> 00:47:58.500 align:middle line:84% and then you have two repayment options. 00:47:58.500 --> 00:48:01.640 align:middle line:84% One is to repay everything at t equals one 00:48:01.640 --> 00:48:03.920 align:middle line:84% and get a very good deal repaying everything at t 00:48:03.920 --> 00:48:07.130 align:middle line:84% equals one, and then the other one is 00:48:07.130 --> 00:48:10.460 align:middle line:84% to spread your repayments out over t equals one 00:48:10.460 --> 00:48:14.030 align:middle line:84% and t equals two, and make an even larger repayment at t 00:48:14.030 --> 00:48:17.600 align:middle line:84% equals two than you make at t equals one. 00:48:17.600 --> 00:48:20.930 align:middle line:84% So you can think of it as like you have these-- 00:48:20.930 --> 00:48:22.715 align:middle line:84% you know, they give you money up front. 00:48:22.715 --> 00:48:25.460 align:middle line:90% 00:48:25.460 --> 00:48:28.850 align:middle line:84% Let's suppose they give you $30,000 up front and then they 00:48:28.850 --> 00:48:31.740 align:middle line:84% have these two options for repayments. 00:48:31.740 --> 00:48:37.190 align:middle line:84% One is to repay 25,000 at t equals one and zero at t equals 00:48:37.190 --> 00:48:42.440 align:middle line:84% two, and the other is to repay 15,000 at t equals one 00:48:42.440 --> 00:48:50.800 align:middle line:84% and 25,000 or 20,000 at t equals two. 00:48:50.800 --> 00:48:51.300 align:middle line:90% OK? 00:48:51.300 --> 00:48:54.570 align:middle line:84% So the first best in my simple example would be they loan you 00:48:54.570 --> 00:49:01.660 align:middle line:84% 30,000 and you repay 15,000 in each of the two future periods. 00:49:01.660 --> 00:49:04.680 align:middle line:84% That's what the first efficient thing would be in this model. 00:49:04.680 --> 00:49:07.600 align:middle line:84% But instead what they do is they actually loan you-- 00:49:07.600 --> 00:49:14.790 align:middle line:84% actually, they do following, they loan you 35,000 and they 00:49:14.790 --> 00:49:15.960 align:middle line:90% give you two options. 00:49:15.960 --> 00:49:19.950 align:middle line:84% Either repay 25,000 at t equals one and nothing at t equals two, 00:49:19.950 --> 00:49:26.920 align:middle line:84% or repay 15,000 at t equals one and 25,000 at t equals two. 00:49:26.920 --> 00:49:31.720 align:middle line:84% And so the contract has two different distortions. 00:49:31.720 --> 00:49:36.490 align:middle line:84% One distortion is that the consumers end up 00:49:36.490 --> 00:49:40.870 align:middle line:84% taking on more debt than they want-- 00:49:40.870 --> 00:49:42.550 align:middle line:84% take on more debt than they want, 00:49:42.550 --> 00:49:44.830 align:middle line:84% more debt than they should in a rational contract 00:49:44.830 --> 00:49:48.630 align:middle line:84% and get a bigger loan and buy a more expensive car. 00:49:48.630 --> 00:49:50.460 align:middle line:84% And then the second distortion is 00:49:50.460 --> 00:49:54.840 align:middle line:84% that they end up choosing the wrong repayment option, where 00:49:54.840 --> 00:49:56.492 align:middle line:84% you give them this option, they think 00:49:56.492 --> 00:49:57.700 align:middle line:90% they're going to choose this. 00:49:57.700 --> 00:49:58.870 align:middle line:90% Wow, that's a great deal. 00:49:58.870 --> 00:50:00.870 align:middle line:84% If I just pay everything back next period, 00:50:00.870 --> 00:50:03.640 align:middle line:84% I'm getting an awesome deal repaying my car loan. 00:50:03.640 --> 00:50:06.720 align:middle line:84% But then in fact, once it gets the next period and trying 00:50:06.720 --> 00:50:10.670 align:middle line:84% to come up with 25,000 all in one year is too painful, 00:50:10.670 --> 00:50:12.510 align:middle line:84% you end up spreading it out and saying, OK, 00:50:12.510 --> 00:50:18.228 align:middle line:84% I'll pay 15,000 today and then 25,000 tomorrow. 00:50:18.228 --> 00:50:21.780 align:middle line:84% So we get these two different distortions, the distortion 00:50:21.780 --> 00:50:25.860 align:middle line:84% of loans that are too big and loan payments that 00:50:25.860 --> 00:50:27.480 align:middle line:84% are distorted with a balloon payment 00:50:27.480 --> 00:50:30.510 align:middle line:84% at the end that's larger than the payment in the beginning, 00:50:30.510 --> 00:50:33.900 align:middle line:84% whereas spreading the payments evenly is what would minimize 00:50:33.900 --> 00:50:37.500 align:middle line:90% the customer's disutility. 00:50:37.500 --> 00:50:38.000 align:middle line:90% OK. 00:50:38.000 --> 00:50:40.890 align:middle line:84% And, you know, it's a model with perfect competition. 00:50:40.890 --> 00:50:43.880 align:middle line:84% So with perfect competition, the lenders-- 00:50:43.880 --> 00:50:47.240 align:middle line:84% actually, so I should have done this with 20. 00:50:47.240 --> 00:50:50.300 align:middle line:84% In this model, the lenders, zero profits. 00:50:50.300 --> 00:50:52.020 align:middle line:84% You get zero profits for the lenders, 00:50:52.020 --> 00:50:53.960 align:middle line:84% so the amount that they borrow is 00:50:53.960 --> 00:50:56.040 align:middle line:90% exactly equal to the repayment. 00:50:56.040 --> 00:50:58.670 align:middle line:84% So it's not like lenders make any extra-- lenders don't 00:50:58.670 --> 00:50:59.460 align:middle line:90% make any money. 00:50:59.460 --> 00:51:02.000 align:middle line:84% It's just utility is being destroyed 00:51:02.000 --> 00:51:06.470 align:middle line:84% by having consumers get a larger loan than is rational for them 00:51:06.470 --> 00:51:12.110 align:middle line:84% to get and pay the money back unevenly 00:51:12.110 --> 00:51:14.990 align:middle line:84% in a way that makes them suffer more disutility 00:51:14.990 --> 00:51:16.393 align:middle line:90% from the unequal payments. 00:51:16.393 --> 00:51:23.220 align:middle line:90% 00:51:23.220 --> 00:51:27.245 align:middle line:84% So questions on the finding of what they're trying to show? 00:51:27.245 --> 00:51:30.550 align:middle line:90% 00:51:30.550 --> 00:51:31.050 align:middle line:90% OK. 00:51:31.050 --> 00:51:33.717 align:middle line:84% So what I thought I would do is I would actually just go through 00:51:33.717 --> 00:51:38.920 align:middle line:84% the-- it's a standard type of mechanism design argument. 00:51:38.920 --> 00:51:41.230 align:middle line:84% It's pretty simple, so I thought I would do that. 00:51:41.230 --> 00:51:44.040 align:middle line:84% So first way to think about analyzing a model like this 00:51:44.040 --> 00:51:49.950 align:middle line:84% is we can restrict ourselves to contracts 00:51:49.950 --> 00:51:53.460 align:middle line:84% that have just two options, because all that really matters 00:51:53.460 --> 00:51:55.620 align:middle line:84% is what does the consumer think they're 00:51:55.620 --> 00:51:58.680 align:middle line:84% going to repay when they sign the contract 00:51:58.680 --> 00:52:00.333 align:middle line:90% and what do they actually repay. 00:52:00.333 --> 00:52:03.000 align:middle line:84% So there's one option that needs to be there, which is what they 00:52:03.000 --> 00:52:03.880 align:middle line:90% think they're going to pay. 00:52:03.880 --> 00:52:05.980 align:middle line:84% There's another one, what are they actually going to pay? 00:52:05.980 --> 00:52:07.840 align:middle line:84% You can put 10 different options in there. 00:52:07.840 --> 00:52:09.755 align:middle line:90% The other eight don't matter. 00:52:09.755 --> 00:52:12.090 align:middle line:84% OK, so we can think of a general class of contracts 00:52:12.090 --> 00:52:14.370 align:middle line:84% as being how much money does the lenders 00:52:14.370 --> 00:52:17.513 align:middle line:84% give them up front, what's the repayment option 00:52:17.513 --> 00:52:19.680 align:middle line:84% that they think they're going to take when they sign 00:52:19.680 --> 00:52:21.930 align:middle line:84% the contract, and then what's the repayment 00:52:21.930 --> 00:52:24.568 align:middle line:84% option that they're actually going to take. 00:52:24.568 --> 00:52:25.068 align:middle line:90% OK. 00:52:25.068 --> 00:52:27.990 align:middle line:90% 00:52:27.990 --> 00:52:30.690 align:middle line:84% The lenders are competing to make the most attractive offers, 00:52:30.690 --> 00:52:32.850 align:middle line:84% which means it must be that the offer that 00:52:32.850 --> 00:52:36.540 align:middle line:84% gets made and accepted in equilibrium solves this problem. 00:52:36.540 --> 00:52:39.480 align:middle line:84% We're maximizing the utility that consumers 00:52:39.480 --> 00:52:44.050 align:middle line:84% think you're giving them subject to three constraints. 00:52:44.050 --> 00:52:46.770 align:middle line:84% The first is the individual rationality constraint 00:52:46.770 --> 00:52:48.510 align:middle line:84% of the lenders, which is the lenders 00:52:48.510 --> 00:52:52.040 align:middle line:84% have to earn zero profits, so the amount 00:52:52.040 --> 00:52:53.790 align:middle line:84% that they give the consumer up front can't 00:52:53.790 --> 00:52:55.665 align:middle line:84% be any bigger than the repayment that they're 00:52:55.665 --> 00:52:59.950 align:middle line:84% going to get in equilibrium, which is q and r. 00:52:59.950 --> 00:53:02.860 align:middle line:84% There's the IC constraint for the time zero self, 00:53:02.860 --> 00:53:04.520 align:middle line:90% the one making the purchase. 00:53:04.520 --> 00:53:07.210 align:middle line:84% The time zero self has to think that they're 00:53:07.210 --> 00:53:10.550 align:middle line:90% going to repay q hat r hat. 00:53:10.550 --> 00:53:16.320 align:middle line:84% And so it must be that the disutility that you 00:53:16.320 --> 00:53:19.440 align:middle line:84% get from paying q hat r hat, which is k of q hat 00:53:19.440 --> 00:53:22.980 align:middle line:84% plus beta hat k of r hat, that's how you think your future self 00:53:22.980 --> 00:53:27.990 align:middle line:84% is going to be hyperbolic, that has to be less than k of q 00:53:27.990 --> 00:53:29.200 align:middle line:90% plus beta hat k of r. 00:53:29.200 --> 00:53:33.300 align:middle line:84% So you think, wow, this one is such a bad-- 00:53:33.300 --> 00:53:35.190 align:middle line:84% IC zero says this is such a bad deal 00:53:35.190 --> 00:53:39.060 align:middle line:84% relative to this one that my future self will pay this one. 00:53:39.060 --> 00:53:42.600 align:middle line:84% And then it must also be that when we get to t equals one, 00:53:42.600 --> 00:53:44.580 align:middle line:84% from the t equals one self, when the t 00:53:44.580 --> 00:53:46.740 align:middle line:84% equals one self is comparing these, 00:53:46.740 --> 00:53:49.240 align:middle line:84% the t equals one self says, wow, this is painful. 00:53:49.240 --> 00:53:51.030 align:middle line:84% I'm going to choose this one instead. 00:53:51.030 --> 00:53:54.730 align:middle line:84% So those are the two IC constraints. 00:53:54.730 --> 00:53:56.670 align:middle line:84% The t equals zero self has to say, 00:53:56.670 --> 00:53:59.820 align:middle line:84% I think I'm going to take the q hat r hat option, 00:53:59.820 --> 00:54:03.210 align:middle line:84% and then the t equals one self has to say, 00:54:03.210 --> 00:54:05.313 align:middle line:84% that q hat r hat option is painful. 00:54:05.313 --> 00:54:06.980 align:middle line:84% I'm going to take the qr option instead. 00:54:06.980 --> 00:54:09.590 align:middle line:90% 00:54:09.590 --> 00:54:10.310 align:middle line:90% OK? 00:54:10.310 --> 00:54:14.660 align:middle line:84% So we have this fairly simple problem 00:54:14.660 --> 00:54:21.530 align:middle line:84% that initially has five unknowns, c, q, r, q hat, r hat. 00:54:21.530 --> 00:54:24.560 align:middle line:84% And as you often do in mechanism design problems, 00:54:24.560 --> 00:54:28.910 align:middle line:84% the way to solve these things is to start by eliminating as many 00:54:28.910 --> 00:54:31.760 align:middle line:84% of the constraints as you can or turning the inequality 00:54:31.760 --> 00:54:33.440 align:middle line:84% constraints into equality constraints 00:54:33.440 --> 00:54:35.480 align:middle line:90% and eliminating variables. 00:54:35.480 --> 00:54:37.520 align:middle line:84% The first variable I'm going to note 00:54:37.520 --> 00:54:41.130 align:middle line:84% is the IR constraint is always going to be binding. 00:54:41.130 --> 00:54:44.120 align:middle line:84% You're always going to set c equals q plus r. 00:54:44.120 --> 00:54:45.830 align:middle line:84% You're going to set c equals q plus r, 00:54:45.830 --> 00:54:48.680 align:middle line:84% because giving the consumer more money 00:54:48.680 --> 00:54:52.020 align:middle line:84% up front makes the consumer happier, 00:54:52.020 --> 00:54:54.840 align:middle line:84% and this is the only constraint where c appears. 00:54:54.840 --> 00:54:59.990 align:middle line:84% So clearly we can really drop that out and just say whatever q 00:54:59.990 --> 00:55:02.155 align:middle line:84% and r we choose, c is going to be q plus r. 00:55:02.155 --> 00:55:05.440 align:middle line:90% 00:55:05.440 --> 00:55:06.010 align:middle line:90% OK. 00:55:06.010 --> 00:55:11.350 align:middle line:84% Second observation is that this one also 00:55:11.350 --> 00:55:13.300 align:middle line:90% has to hold with equality. 00:55:13.300 --> 00:55:16.840 align:middle line:84% If this constraint was non-binding, 00:55:16.840 --> 00:55:19.570 align:middle line:90% I can make q and r bigger. 00:55:19.570 --> 00:55:22.180 align:middle line:84% And if I make q and r bigger this constraint, if this 00:55:22.180 --> 00:55:24.010 align:middle line:84% was non-binding, I'd make q and r bigger, 00:55:24.010 --> 00:55:25.880 align:middle line:90% this would still be satisfied. 00:55:25.880 --> 00:55:28.142 align:middle line:84% This would be satisfied, this is satisfied, 00:55:28.142 --> 00:55:29.350 align:middle line:90% this doesn't involve q and r. 00:55:29.350 --> 00:55:33.070 align:middle line:84% So I'm always going to want to make q and r as big as I can 00:55:33.070 --> 00:55:35.870 align:middle line:84% until this constraint gets satisfied. 00:55:35.870 --> 00:55:39.940 align:middle line:84% So what limits these repayments is that the consumer at t 00:55:39.940 --> 00:55:42.730 align:middle line:84% equals one has to actually be willing to make them. 00:55:42.730 --> 00:55:45.906 align:middle line:84% So anyway, so this one holds with equality. 00:55:45.906 --> 00:55:48.650 align:middle line:90% 00:55:48.650 --> 00:55:52.820 align:middle line:84% So once you know that this one has an equal sign in it, 00:55:52.820 --> 00:55:57.020 align:middle line:84% think about what does this constraint equivalent to when 00:55:57.020 --> 00:55:58.347 align:middle line:90% this one holds with equality? 00:55:58.347 --> 00:55:59.930 align:middle line:84% Well, if this one holds with equality, 00:55:59.930 --> 00:56:03.080 align:middle line:84% I can just subtract the two left side-- the left side here 00:56:03.080 --> 00:56:05.760 align:middle line:84% from this, subtract the right side here from this, 00:56:05.760 --> 00:56:07.760 align:middle line:90% and both things are still true. 00:56:07.760 --> 00:56:11.090 align:middle line:84% And so the top inequality still holds, and these things cancel, 00:56:11.090 --> 00:56:12.690 align:middle line:90% and these things cancel. 00:56:12.690 --> 00:56:14.900 align:middle line:84% And so what I find is that it must be true 00:56:14.900 --> 00:56:20.900 align:middle line:84% that beta hat minus beta k of r, the right side, this minus this, 00:56:20.900 --> 00:56:26.750 align:middle line:84% is bigger than beta hat, minus beta times k of r hat. 00:56:26.750 --> 00:56:29.390 align:middle line:90% So the IC constraint. 00:56:29.390 --> 00:56:31.550 align:middle line:84% So once this holds with equality, 00:56:31.550 --> 00:56:35.360 align:middle line:84% this constraint is simply equivalent to r hat 00:56:35.360 --> 00:56:36.325 align:middle line:90% being less than r. 00:56:36.325 --> 00:56:38.740 align:middle line:90% 00:56:38.740 --> 00:56:39.240 align:middle line:90% OK? 00:56:39.240 --> 00:56:42.520 align:middle line:84% So IC zero is just this holds with equality, 00:56:42.520 --> 00:56:44.610 align:middle line:84% this holds with equality, this is 00:56:44.610 --> 00:56:46.850 align:middle line:90% equivalent to r hat less than r. 00:56:46.850 --> 00:56:50.570 align:middle line:90% 00:56:50.570 --> 00:56:53.990 align:middle line:90% So what's my simplified problem? 00:56:53.990 --> 00:56:59.090 align:middle line:84% I'm maximizing c equals q plus r minus k of q hat minus k 00:56:59.090 --> 00:57:02.510 align:middle line:84% of r hat, and my two constraints are 00:57:02.510 --> 00:57:06.320 align:middle line:84% that r hat has to be less than or equal to r, 00:57:06.320 --> 00:57:11.130 align:middle line:84% and this bottom one, the IC one, holds with equality. 00:57:11.130 --> 00:57:19.228 align:middle line:90% 00:57:19.228 --> 00:57:22.300 align:middle line:84% So again, to simplify this problem, 00:57:22.300 --> 00:57:27.670 align:middle line:84% I now take this expression for what k of q hat is, 00:57:27.670 --> 00:57:29.780 align:middle line:90% and I plug this in here. 00:57:29.780 --> 00:57:33.580 align:middle line:84% So this is k of q hat equals this plus this minus this. 00:57:33.580 --> 00:57:35.980 align:middle line:84% And so when I do that, the equation 00:57:35.980 --> 00:57:46.780 align:middle line:84% becomes q plus r minus this, so minus k of q minus beta k of r, 00:57:46.780 --> 00:57:50.180 align:middle line:84% and then I get a minus 1 minus beta k of r hat. 00:57:50.180 --> 00:57:52.600 align:middle line:84% So I maximizing that subject to the constraint 00:57:52.600 --> 00:57:55.155 align:middle line:84% that r hat is less than or equal to r. 00:57:55.155 --> 00:57:57.840 align:middle line:90% 00:57:57.840 --> 00:58:00.000 align:middle line:84% I, again, look at this expression 00:58:00.000 --> 00:58:01.950 align:middle line:84% and I recognize that the r hat only 00:58:01.950 --> 00:58:04.470 align:middle line:84% appears one way, one place, and the r hat 00:58:04.470 --> 00:58:07.200 align:middle line:84% has got a positive constant times 00:58:07.200 --> 00:58:09.040 align:middle line:90% an increasing function of r hat. 00:58:09.040 --> 00:58:11.970 align:middle line:84% So if my only constraint is that r hat equals r, 00:58:11.970 --> 00:58:13.865 align:middle line:90% I want to set r equals zero. 00:58:13.865 --> 00:58:16.580 align:middle line:90% 00:58:16.580 --> 00:58:18.400 align:middle line:84% So that's where the zero comes from. 00:58:18.400 --> 00:58:21.080 align:middle line:90% 00:58:21.080 --> 00:58:25.100 align:middle line:84% And then I'm left with a very simple problem, 00:58:25.100 --> 00:58:29.510 align:middle line:84% which is just I'm trying to maximize what's left. 00:58:29.510 --> 00:58:42.868 align:middle line:84% I'm trying to maximize q plus r minus k of q minus beta k of r. 00:58:42.868 --> 00:58:44.660 align:middle line:84% And so then just the first order conditions 00:58:44.660 --> 00:58:47.870 align:middle line:84% say you set k prime at q equal to 1 00:58:47.870 --> 00:58:51.490 align:middle line:84% and you set k prime at r equals to 1 over beta. 00:58:51.490 --> 00:58:55.915 align:middle line:90% 00:58:55.915 --> 00:58:58.530 align:middle line:84% And so it's the k prime at q equals 1 that 00:58:58.530 --> 00:59:01.680 align:middle line:90% makes q the efficient level. 00:59:01.680 --> 00:59:03.900 align:middle line:90% Good, so I did get that right. 00:59:03.900 --> 00:59:05.880 align:middle line:84% If the efficient level is 15,000, 00:59:05.880 --> 00:59:11.130 align:middle line:84% you ask for the efficient level at that period, 00:59:11.130 --> 00:59:13.660 align:middle line:84% but then k prime and r is 1 over beta, 00:59:13.660 --> 00:59:18.360 align:middle line:84% so you ask for discretely more money in this preference. 00:59:18.360 --> 00:59:22.185 align:middle line:90% And so the model was that-- 00:59:22.185 --> 00:59:25.410 align:middle line:84% with the beta delta hyperbolic consumers, 00:59:25.410 --> 00:59:26.980 align:middle line:90% you didn't do any distorting. 00:59:26.980 --> 00:59:30.960 align:middle line:84% But now with beta, beta hat delta, with beta hat 00:59:30.960 --> 00:59:34.260 align:middle line:84% arbitrarily close to beta, you get this discrete jump 00:59:34.260 --> 00:59:37.470 align:middle line:84% reflecting the 1 over beta discounting 00:59:37.470 --> 00:59:39.600 align:middle line:90% versus the beta discounting. 00:59:39.600 --> 00:59:42.210 align:middle line:84% And there's just no way that you can come in as a lender 00:59:42.210 --> 00:59:43.710 align:middle line:84% and compete in this market and not 00:59:43.710 --> 00:59:47.360 align:middle line:84% exploit the consumers because this is 00:59:47.360 --> 00:59:48.610 align:middle line:90% what the consumers are seeing. 00:59:48.610 --> 00:59:51.120 align:middle line:84% The consumers are seeing this borrow 35,000, 00:59:51.120 --> 00:59:53.710 align:middle line:90% repay 25,000 contract. 00:59:53.710 --> 00:59:56.000 align:middle line:84% They think that's what they're going to do. 00:59:56.000 --> 00:59:58.613 align:middle line:84% I can't come in and say, no, look, take a reasonable size 00:59:58.613 --> 01:00:01.030 align:middle line:84% loan, pay it out evenly, because they're going to be like, 01:00:01.030 --> 01:00:03.980 align:middle line:84% no, I would like to get the nicer car 01:00:03.980 --> 01:00:07.910 align:middle line:84% and it's a huge financial win for me to take this teaser deal. 01:00:07.910 --> 01:00:10.090 align:middle line:84% Therefore, you can't compete with that teaser deal 01:00:10.090 --> 01:00:12.325 align:middle line:84% with the regular deal, because if you're not 01:00:12.325 --> 01:00:14.200 align:middle line:84% exploiting the consumers, you can't make them 01:00:14.200 --> 01:00:15.575 align:middle line:84% that outlandish promise of you're 01:00:15.575 --> 01:00:19.520 align:middle line:84% going to get all this money and repay this little amount. 01:00:19.520 --> 01:00:20.730 align:middle line:90% Next paper I want to do. 01:00:20.730 --> 01:00:23.395 align:middle line:84% I guess I won't get through all of these today, but. 01:00:23.395 --> 01:00:26.340 align:middle line:90% 01:00:26.340 --> 01:00:28.410 align:middle line:84% So this is a paper by Michael Grubb, 01:00:28.410 --> 01:00:33.394 align:middle line:84% "Selling to Overconfident Consumers," which has a-- 01:00:33.394 --> 01:00:36.480 align:middle line:84% it's based on an overconfidence bias. 01:00:36.480 --> 01:00:39.030 align:middle line:84% People think they know more about their future self 01:00:39.030 --> 01:00:40.410 align:middle line:90% than they do. 01:00:40.410 --> 01:00:44.483 align:middle line:84% And what he's trying to explain is 01:00:44.483 --> 01:00:46.650 align:middle line:84% one of the things I did early on in monopoly pricing 01:00:46.650 --> 01:00:47.820 align:middle line:90% is two part tariffs. 01:00:47.820 --> 01:00:51.090 align:middle line:84% Like at Disneyland, you pay $100 to get into the park 01:00:51.090 --> 01:00:53.730 align:middle line:84% and then every ride is free, or you 01:00:53.730 --> 01:00:57.120 align:middle line:84% have something where you pay some upfront cost 01:00:57.120 --> 01:00:59.340 align:middle line:84% and then you pay a low cost per unit 01:00:59.340 --> 01:01:00.900 align:middle line:84% and they extract all your surplus 01:01:00.900 --> 01:01:04.475 align:middle line:84% you're getting from price equals cost in a fixed fee up front. 01:01:04.475 --> 01:01:07.662 align:middle line:90% 01:01:07.662 --> 01:01:10.730 align:middle line:84% What Grubb noticed is that there are actually 01:01:10.730 --> 01:01:12.770 align:middle line:84% a lot of applications where what we seem to see 01:01:12.770 --> 01:01:17.570 align:middle line:84% is not two part tariffs, but three part tariffs. 01:01:17.570 --> 01:01:20.600 align:middle line:84% So the three part tariff is this payment scheme 01:01:20.600 --> 01:01:24.320 align:middle line:84% where you have an upfront payment for the good, 01:01:24.320 --> 01:01:26.960 align:middle line:84% and then some number of units of the good 01:01:26.960 --> 01:01:31.460 align:middle line:84% sold at below cost, and then additional units 01:01:31.460 --> 01:01:34.340 align:middle line:84% beyond that sold at an above cost price, 01:01:34.340 --> 01:01:36.310 align:middle line:84% and often a very, very high price. 01:01:36.310 --> 01:01:38.600 align:middle line:90% So examples he gives. 01:01:38.600 --> 01:01:41.240 align:middle line:84% Some of them, like the credit card example, 01:01:41.240 --> 01:01:44.090 align:middle line:84% you pay some upfront fee to transfer your money 01:01:44.090 --> 01:01:46.870 align:middle line:90% to the new credit card. 01:01:46.870 --> 01:01:49.565 align:middle line:84% They'll give you six months interest free, 01:01:49.565 --> 01:01:51.190 align:middle line:84% and then after the six months are over, 01:01:51.190 --> 01:01:54.650 align:middle line:84% they're going to start charging you 19.8%. 01:01:54.650 --> 01:01:59.030 align:middle line:84% Or you lease a car, you pay $12,000 for the three-year car 01:01:59.030 --> 01:02:00.792 align:middle line:90% lease. 01:02:00.792 --> 01:02:02.750 align:middle line:84% Car leases-- I don't know if you've done these. 01:02:02.750 --> 01:02:04.880 align:middle line:84% Car leases do have mileage limits on them. 01:02:04.880 --> 01:02:06.950 align:middle line:84% Like, because you lease the car for three years, 01:02:06.950 --> 01:02:09.570 align:middle line:84% you get 36,000 miles included free. 01:02:09.570 --> 01:02:11.990 align:middle line:84% They don't charge you for the wear and tear on the car. 01:02:11.990 --> 01:02:16.370 align:middle line:84% But then if you drive the car more than 36,000 miles, 01:02:16.370 --> 01:02:19.550 align:middle line:90% you pay an extra $0.29 a mile. 01:02:19.550 --> 01:02:21.770 align:middle line:84% Actually, maybe for you guys, renting U-Haul trucks 01:02:21.770 --> 01:02:22.470 align:middle line:90% works this way. 01:02:22.470 --> 01:02:24.762 align:middle line:84% I don't know if you've ever rented a U-Haul truck where 01:02:24.762 --> 01:02:29.900 align:middle line:84% it's like you rent the truck and it's 99.99 for the truck, 01:02:29.900 --> 01:02:33.150 align:middle line:84% up to 150 miles of use of the truck is free. 01:02:33.150 --> 01:02:36.770 align:middle line:84% If you go over to 150 miles on the truck, 01:02:36.770 --> 01:02:40.520 align:middle line:84% it's like $0.79 a mile or something for every mile 01:02:40.520 --> 01:02:42.920 align:middle line:84% that you charge, and you run up this enormous over charge 01:02:42.920 --> 01:02:45.890 align:middle line:90% on the U-Haul trucks. 01:02:45.890 --> 01:02:50.990 align:middle line:84% Cell phones back in the day, you paid 39.99 a month. 01:02:50.990 --> 01:02:53.940 align:middle line:84% You would have two gigabytes of data free. 01:02:53.940 --> 01:02:56.810 align:middle line:84% Once you used up your two gigabytes of usage, 01:02:56.810 --> 01:02:58.670 align:middle line:84% they start charging you some extra amount 01:02:58.670 --> 01:02:59.637 align:middle line:90% of money per megabyte. 01:02:59.637 --> 01:03:01.220 align:middle line:84% That doesn't sound so big, but then it 01:03:01.220 --> 01:03:04.310 align:middle line:84% becomes an enormous charge for the overage on your data 01:03:04.310 --> 01:03:05.310 align:middle line:90% overage. 01:03:05.310 --> 01:03:08.790 align:middle line:90% 01:03:08.790 --> 01:03:11.250 align:middle line:84% Grubb suggests that why do we see these-- 01:03:11.250 --> 01:03:12.780 align:middle line:84% we know why we would see a two part 01:03:12.780 --> 01:03:15.220 align:middle line:84% tariff, because it's efficient to price things at cost. 01:03:15.220 --> 01:03:17.410 align:middle line:84% Why would we see a three part tariff? 01:03:17.410 --> 01:03:21.060 align:middle line:84% And Grubb's argument is that we see three part tariffs 01:03:21.060 --> 01:03:23.610 align:middle line:84% because it's an exploitation of consumers 01:03:23.610 --> 01:03:26.070 align:middle line:84% who think they're better than they are at predicting 01:03:26.070 --> 01:03:27.245 align:middle line:90% their future consumption. 01:03:27.245 --> 01:03:28.620 align:middle line:84% Just like you think you're better 01:03:28.620 --> 01:03:31.470 align:middle line:84% than you are at predicting how far Buenos Aires is, 01:03:31.470 --> 01:03:33.210 align:middle line:84% you think you're better than you are 01:03:33.210 --> 01:03:35.760 align:middle line:84% at predicting how many gigabytes of data 01:03:35.760 --> 01:03:38.100 align:middle line:84% or how many miles am I going to drive the U-Haul truck 01:03:38.100 --> 01:03:41.500 align:middle line:90% or something like that. 01:03:41.500 --> 01:03:46.330 align:middle line:84% So he gives the simple example about where the three part 01:03:46.330 --> 01:03:50.110 align:middle line:84% tariffs would come from, which is this straightforward 01:03:50.110 --> 01:03:51.350 align:middle line:90% numerical thing. 01:03:51.350 --> 01:03:54.250 align:middle line:84% So suppose firms have fixed costs 01:03:54.250 --> 01:03:58.270 align:middle line:84% of $50 of serving the consumer and then 01:03:58.270 --> 01:04:02.110 align:middle line:84% they have marginal costs of $0.05 per unit 01:04:02.110 --> 01:04:03.110 align:middle line:90% of serving the consumer. 01:04:03.110 --> 01:04:06.790 align:middle line:84% So this could be cost of providing you 01:04:06.790 --> 01:04:09.070 align:middle line:84% with a cell phone and then marginal cost per megabyte 01:04:09.070 --> 01:04:11.350 align:middle line:84% or cost of renting you the car and then 01:04:11.350 --> 01:04:14.900 align:middle line:84% cost per mile of having you drive the car. 01:04:14.900 --> 01:04:17.810 align:middle line:84% And assume that the firm can offer a non-linear tariff 01:04:17.810 --> 01:04:19.760 align:middle line:90% t of x. 01:04:19.760 --> 01:04:23.900 align:middle line:84% Assume that consumers get benefits of $0.45 per unit 01:04:23.900 --> 01:04:26.540 align:middle line:84% for theta units, where theta is either 01:04:26.540 --> 01:04:30.700 align:middle line:84% 100, 400, or 700, and all three are equally likely. 01:04:30.700 --> 01:04:33.205 align:middle line:90% 01:04:33.205 --> 01:04:35.510 align:middle line:84% Either I get my phone and then I'm 01:04:35.510 --> 01:04:38.828 align:middle line:84% going to have some number of megabytes of data I want to use, 01:04:38.828 --> 01:04:41.120 align:middle line:84% and I don't know which of those three it's going to be, 01:04:41.120 --> 01:04:42.662 align:middle line:84% but whatever it is, I'm going to have 01:04:42.662 --> 01:04:45.580 align:middle line:84% just some inherent need for the good that I want to use those. 01:04:45.580 --> 01:04:48.750 align:middle line:84% I'm going to get $0.45 per unit using theta units, 01:04:48.750 --> 01:04:51.200 align:middle line:84% and then I'm not going to want more than theta units. 01:04:51.200 --> 01:04:53.360 align:middle line:84% Or the U-Haul rental truck, I'm either 01:04:53.360 --> 01:04:57.060 align:middle line:84% going to want to drive it 100 miles, 400 miles, or 700 miles. 01:04:57.060 --> 01:04:59.310 align:middle line:84% I don't know which of those three it's going to be. 01:04:59.310 --> 01:05:01.520 align:middle line:84% However many I get, I get this high value 01:05:01.520 --> 01:05:03.680 align:middle line:84% for using it for those miles and then no value 01:05:03.680 --> 01:05:05.360 align:middle line:90% for using it beyond that. 01:05:05.360 --> 01:05:08.900 align:middle line:84% So anyway, all three are considered 01:05:08.900 --> 01:05:11.360 align:middle line:84% equally likely by the consumer when they purchase, 01:05:11.360 --> 01:05:13.878 align:middle line:84% and then after they purchase, they learn theta 01:05:13.878 --> 01:05:16.170 align:middle line:84% and then they choose how much to actually use the thing 01:05:16.170 --> 01:05:19.170 align:middle line:90% that they've purchased. 01:05:19.170 --> 01:05:19.680 align:middle line:90% OK. 01:05:19.680 --> 01:05:22.590 align:middle line:84% What would happen with rational consumers? 01:05:22.590 --> 01:05:26.460 align:middle line:84% With rational consumers, we do a two part tariff. 01:05:26.460 --> 01:05:30.120 align:middle line:84% The cost of providing the units to the seller is $0.05. 01:05:30.120 --> 01:05:33.120 align:middle line:84% So you charge consumers $0.05 a unit, 01:05:33.120 --> 01:05:36.000 align:middle line:84% and then you extract all the ex-ante surplus 01:05:36.000 --> 01:05:37.390 align:middle line:90% through a fixed fee. 01:05:37.390 --> 01:05:39.180 align:middle line:84% And if consumers are going to, on average, 01:05:39.180 --> 01:05:46.590 align:middle line:84% use 400 units and they're getting $0.40 per unit 01:05:46.590 --> 01:05:50.350 align:middle line:84% of surplus, 400 times $0.40 is $160. 01:05:50.350 --> 01:05:56.290 align:middle line:84% So the rational consumer plan, you would do 160 plus $0.05. 01:05:56.290 --> 01:05:58.140 align:middle line:84% With perfectly competitive firms, 01:05:58.140 --> 01:06:00.240 align:middle line:84% you'd, again, use the $0.5 fee and you'd just 01:06:00.240 --> 01:06:03.390 align:middle line:84% be charging the 50 up front to charge cost. 01:06:03.390 --> 01:06:06.485 align:middle line:84% And this is socially efficient pricing. 01:06:06.485 --> 01:06:11.110 align:middle line:90% 01:06:11.110 --> 01:06:16.790 align:middle line:84% What happens with the monopolist Grubb's observation 01:06:16.790 --> 01:06:18.970 align:middle line:84% on what happens with the monopolist is 01:06:18.970 --> 01:06:22.420 align:middle line:84% a monopolist in this problem would use a three part tariff. 01:06:22.420 --> 01:06:23.950 align:middle line:90% They would charge you-- 01:06:23.950 --> 01:06:26.510 align:middle line:84% instead of $160 plus $0.05 a minute, 01:06:26.510 --> 01:06:29.840 align:middle line:90% they'd charge you $180 up front. 01:06:29.840 --> 01:06:35.150 align:middle line:84% The first 400 units would be free, and then units beyond 400, 01:06:35.150 --> 01:06:37.160 align:middle line:90% they would charge you $0.45. 01:06:37.160 --> 01:06:41.430 align:middle line:84% So we just get this very high, steep three part tariff. 01:06:41.430 --> 01:06:45.110 align:middle line:84% We start high fixed fee, no cost for 400 units, 01:06:45.110 --> 01:06:50.190 align:middle line:84% and then steeply increasing price beyond 400. 01:06:50.190 --> 01:06:51.660 align:middle line:90% OK. 01:06:51.660 --> 01:06:55.230 align:middle line:90% Why do we do this? 01:06:55.230 --> 01:06:59.280 align:middle line:84% First, why do you charge nothing for the first 400 units? 01:06:59.280 --> 01:07:00.570 align:middle line:90% Oh, sorry. 01:07:00.570 --> 01:07:01.670 align:middle line:90% I didn't say this. 01:07:01.670 --> 01:07:03.420 align:middle line:84% He's going to analyze this model where you 01:07:03.420 --> 01:07:05.160 align:middle line:90% have overconfident consumers. 01:07:05.160 --> 01:07:06.660 align:middle line:84% The starkest model of overconfidence 01:07:06.660 --> 01:07:08.940 align:middle line:84% is the consumers think they're going to need 01:07:08.940 --> 01:07:11.320 align:middle line:90% 400 units with probability one. 01:07:11.320 --> 01:07:13.830 align:middle line:84% So they don't realize that they might only need 100 01:07:13.830 --> 01:07:15.600 align:middle line:90% or they might need 700. 01:07:15.600 --> 01:07:18.570 align:middle line:84% So they think they're going to use 400 with probability one. 01:07:18.570 --> 01:07:19.080 align:middle line:90% OK. 01:07:19.080 --> 01:07:22.480 align:middle line:84% Why do I charge zero for the first 400 units? 01:07:22.480 --> 01:07:27.030 align:middle line:84% Why do I do 180 and then zero for the first 400 units? 01:07:27.030 --> 01:07:35.040 align:middle line:84% Why don't I do 160 up front and $0.05 for the first 400 units 01:07:35.040 --> 01:07:38.120 align:middle line:84% so you pay 180 if you use 400 but less if you don't? 01:07:38.120 --> 01:07:44.060 align:middle line:84% Well, the point is if I'm going to charge you a positive amount, 01:07:44.060 --> 01:07:47.660 align:middle line:84% like $0.05 for the first 400 units, 01:07:47.660 --> 01:07:51.290 align:middle line:84% I'm only going to charge you 160 up front instead of 180 01:07:51.290 --> 01:07:52.580 align:middle line:90% up front. 01:07:52.580 --> 01:07:55.790 align:middle line:84% And then when you actually use 100 units, 01:07:55.790 --> 01:07:58.940 align:middle line:84% you're going to end up paying less. 01:07:58.940 --> 01:08:05.420 align:middle line:84% So you might use 100, you might use 400, you might use 700. 01:08:05.420 --> 01:08:10.730 align:middle line:84% So if I used a plan where I charged you a per unit 01:08:10.730 --> 01:08:15.260 align:middle line:84% fee for the first 400 units, I have to give you 01:08:15.260 --> 01:08:18.350 align:middle line:90% the full discount of-- 01:08:18.350 --> 01:08:23.540 align:middle line:84% I have to give you 400 delta p back in the fixed fee. 01:08:23.540 --> 01:08:26.770 align:middle line:84% And then I give you 400 delta p back in the fixed fee, 01:08:26.770 --> 01:08:30.560 align:middle line:84% but then I only get 300 p in extra revenue 01:08:30.560 --> 01:08:34.100 align:middle line:84% because sometimes you get a better 01:08:34.100 --> 01:08:35.660 align:middle line:90% deal than you're expecting. 01:08:35.660 --> 01:08:37.910 align:middle line:84% Since you don't expect the deal, you're 01:08:37.910 --> 01:08:41.569 align:middle line:84% not going to pay me up front for that, that option of there's 01:08:41.569 --> 01:08:44.510 align:middle line:84% some chance I get to save and only 01:08:44.510 --> 01:08:47.430 align:middle line:90% pay you for 100 instead of 400. 01:08:47.430 --> 01:08:51.290 align:middle line:84% And so because the consumers aren't willing to pay you, 01:08:51.290 --> 01:08:53.359 align:middle line:84% it's always better to charge them 01:08:53.359 --> 01:08:55.972 align:middle line:84% the full price even for the units they're not going-- 01:08:55.972 --> 01:08:57.680 align:middle line:84% this is like exploiting the consumers who 01:08:57.680 --> 01:08:59.930 align:middle line:84% think they're going to go to the gym every single day. 01:08:59.930 --> 01:09:03.698 align:middle line:84% You always want to charge them that amount up front, and then 01:09:03.698 --> 01:09:06.240 align:middle line:84% say, OK, I'm giving you this great deal where you get the 400 01:09:06.240 --> 01:09:07.946 align:middle line:84% units for free, and then you're going 01:09:07.946 --> 01:09:09.779 align:middle line:84% to pay me in advance for all those 400 units 01:09:09.779 --> 01:09:10.987 align:middle line:90% that you're getting for free. 01:09:10.987 --> 01:09:13.520 align:middle line:90% 01:09:13.520 --> 01:09:19.490 align:middle line:84% And then why do firms charge for $0.45 for units beyond 400? 01:09:19.490 --> 01:09:21.490 align:middle line:84% Well, the consumers are sure they're never going 01:09:21.490 --> 01:09:24.007 align:middle line:90% to use any units beyond 400. 01:09:24.007 --> 01:09:26.590 align:middle line:84% So if you offer them a discount, you can have units beyond 400 01:09:26.590 --> 01:09:29.620 align:middle line:84% for $0.40, they're going to pay you nothing up front for that 01:09:29.620 --> 01:09:31.930 align:middle line:84% discount because they think they're not going to use 01:09:31.930 --> 01:09:33.260 align:middle line:90% the extra units. 01:09:33.260 --> 01:09:35.843 align:middle line:84% And so if they think they're not going to use the extra units, 01:09:35.843 --> 01:09:38.609 align:middle line:84% they won't pay anything for the lower price and the extra units, 01:09:38.609 --> 01:09:40.720 align:middle line:84% so you might as well keep raising the price 01:09:40.720 --> 01:09:44.240 align:middle line:84% on the extra units until they get to $0.45. 01:09:44.240 --> 01:09:46.383 align:middle line:84% You don't go beyond 45 because then ex-post, 01:09:46.383 --> 01:09:47.800 align:middle line:84% they're not even going to buy them 01:09:47.800 --> 01:09:49.840 align:middle line:84% and they're just going to not drive the rental truck 01:09:49.840 --> 01:09:51.135 align:middle line:90% for the extra miles they need. 01:09:51.135 --> 01:09:54.229 align:middle line:90% 01:09:54.229 --> 01:09:57.870 align:middle line:84% So what we get is this three part tariff. 01:09:57.870 --> 01:10:00.530 align:middle line:84% And if you had perfectly competitive firms, 01:10:00.530 --> 01:10:02.480 align:middle line:84% perfectly competitive firms would-- again, 01:10:02.480 --> 01:10:04.730 align:middle line:84% they're forced to exploit the consumers, 01:10:04.730 --> 01:10:07.040 align:middle line:84% and so they would do something like that. 01:10:07.040 --> 01:10:09.140 align:middle line:84% So you would give all the surplus back 01:10:09.140 --> 01:10:11.700 align:middle line:84% to the consumers with a really low fixed fee. 01:10:11.700 --> 01:10:14.180 align:middle line:84% But still, what makes it most attractive to consumers 01:10:14.180 --> 01:10:17.960 align:middle line:84% is to design the package so that they think 01:10:17.960 --> 01:10:19.550 align:middle line:90% they're going to pay the least. 01:10:19.550 --> 01:10:22.130 align:middle line:84% Which is give them the big discount up front, 01:10:22.130 --> 01:10:24.500 align:middle line:84% tell them the 400 are included, and tell them 01:10:24.500 --> 01:10:28.200 align:middle line:84% we are going to charge you these high over charges. 01:10:28.200 --> 01:10:29.950 align:middle line:84% But you think you're not going to pay them 01:10:29.950 --> 01:10:31.470 align:middle line:90% so they don't care about that. 01:10:31.470 --> 01:10:34.760 align:middle line:84% And so this is what lets the competitive sellers offer 01:10:34.760 --> 01:10:35.950 align:middle line:90% the best possible package. 01:10:35.950 --> 01:10:42.110 align:middle line:90% 01:10:42.110 --> 01:10:42.770 align:middle line:90% Yeah? 01:10:42.770 --> 01:10:44.870 align:middle line:84% AUDIENCE: How would you tell this sort of behavior 01:10:44.870 --> 01:10:48.230 align:middle line:84% out from something like a increasing [? call ?] structure? 01:10:48.230 --> 01:10:49.700 align:middle line:90% GLENN ELLISON: Yeah. 01:10:49.700 --> 01:10:54.680 align:middle line:84% You know, obviously, I think in some of these examples, like-- 01:10:54.680 --> 01:10:57.600 align:middle line:84% some examples that he gives are better than others on that. 01:10:57.600 --> 01:11:03.050 align:middle line:84% So for instance, the cell phone minute plan. 01:11:03.050 --> 01:11:07.620 align:middle line:84% The megabytes are system wide, not customer wide. 01:11:07.620 --> 01:11:12.917 align:middle line:84% So it can't be that you using more megabytes beyond-- 01:11:12.917 --> 01:11:14.750 align:middle line:84% there, you know that their cost per megabyte 01:11:14.750 --> 01:11:17.210 align:middle line:84% has got to be linear because it's 01:11:17.210 --> 01:11:18.600 align:middle line:90% averaged across all consumers. 01:11:18.600 --> 01:11:20.660 align:middle line:84% It's a network capacity constraint. 01:11:20.660 --> 01:11:24.170 align:middle line:84% For ones like the car lease, you could say, is there 01:11:24.170 --> 01:11:26.660 align:middle line:84% something about the wear and tear 01:11:26.660 --> 01:11:31.490 align:middle line:84% on the car that increases as the car is driven more miles in? 01:11:31.490 --> 01:11:34.460 align:middle line:84% Is there some increasing wear and tear 01:11:34.460 --> 01:11:37.490 align:middle line:84% on the car or the rental truck because it goes more miles? 01:11:37.490 --> 01:11:40.890 align:middle line:84% Yeah, that one you would have to have something to tell apart. 01:11:40.890 --> 01:11:43.520 align:middle line:84% Again, I think the credit card one, 01:11:43.520 --> 01:11:47.150 align:middle line:84% it's going to be harder for a discontinuity in months 01:11:47.150 --> 01:11:48.890 align:middle line:90% or whatever. 01:11:48.890 --> 01:11:51.680 align:middle line:84% What Grubb does have in his paper 01:11:51.680 --> 01:11:56.480 align:middle line:90% is he has an empirical section. 01:11:56.480 --> 01:12:00.320 align:middle line:84% We say, yeah, what did Grubb do to get this paper? 01:12:00.320 --> 01:12:03.470 align:middle line:84% Obviously the story I told you is fairly simple. 01:12:03.470 --> 01:12:05.640 align:middle line:84% How do you turn this into an AER paper? 01:12:05.640 --> 01:12:08.090 align:middle line:84% This is like your idea you come to me with 01:12:08.090 --> 01:12:10.220 align:middle line:84% and you explain it and it's like, OK, 01:12:10.220 --> 01:12:11.990 align:middle line:90% that's a nice, cute example. 01:12:11.990 --> 01:12:14.600 align:middle line:90% That's not a paper in the AER. 01:12:14.600 --> 01:12:18.770 align:middle line:84% He has a much more general model where he has-- 01:12:18.770 --> 01:12:20.030 align:middle line:90% you have a theta. 01:12:20.030 --> 01:12:24.440 align:middle line:90% The theta has some distribution. 01:12:24.440 --> 01:12:29.570 align:middle line:84% Your consumers have some belief about theta f star. 01:12:29.570 --> 01:12:34.250 align:middle line:84% When f star differs from f, it's kind 01:12:34.250 --> 01:12:38.510 align:middle line:84% of mean preserving spread or single crossing condition. 01:12:38.510 --> 01:12:39.600 align:middle line:90% What happens? 01:12:39.600 --> 01:12:42.960 align:middle line:84% He goes out and solves the model and says, OK, this simple model 01:12:42.960 --> 01:12:44.960 align:middle line:84% I had had this three part tariff, 01:12:44.960 --> 01:12:47.300 align:middle line:84% but I do this general model and what I show 01:12:47.300 --> 01:12:51.002 align:middle line:84% is that the optimal solution is not like a three part-- not 01:12:51.002 --> 01:12:52.460 align:middle line:84% exactly a three part tariff, but it 01:12:52.460 --> 01:12:56.330 align:middle line:84% is something that looks kind of like a three part tariff. 01:12:56.330 --> 01:12:57.860 align:middle line:84% But then the second thing he does 01:12:57.860 --> 01:13:00.513 align:middle line:84% is he has this data set on cell phone users. 01:13:00.513 --> 01:13:05.000 align:middle line:90% 01:13:05.000 --> 01:13:07.730 align:middle line:84% I don't think he says in the paper, but I think this was-- 01:13:07.730 --> 01:13:10.340 align:middle line:84% there was a time when cell phones were new, 01:13:10.340 --> 01:13:15.140 align:middle line:84% when universities provided cell phone plans to their students 01:13:15.140 --> 01:13:17.390 align:middle line:84% and you could get your cell phone not through Verizon, 01:13:17.390 --> 01:13:18.680 align:middle line:90% but through MIT. 01:13:18.680 --> 01:13:21.020 align:middle line:84% You'd get an MIT cell phone and MIT 01:13:21.020 --> 01:13:22.820 align:middle line:84% would negotiate a deal with Verizon 01:13:22.820 --> 01:13:26.510 align:middle line:84% to offer discounted cell phone service for its students. 01:13:26.510 --> 01:13:28.588 align:middle line:84% I take it that these were probably Stanford-- 01:13:28.588 --> 01:13:29.630 align:middle line:90% Doesn't say in the paper. 01:13:29.630 --> 01:13:31.963 align:middle line:84% I take it these are probably Stanford student cell phone 01:13:31.963 --> 01:13:32.600 align:middle line:90% plans. 01:13:32.600 --> 01:13:35.540 align:middle line:84% But they had four cell phone plans, 01:13:35.540 --> 01:13:41.000 align:middle line:84% one where you get this amount, just a flat fee per minute. 01:13:41.000 --> 01:13:45.060 align:middle line:84% Small upfront charge, and then a flat fee per minute used. 01:13:45.060 --> 01:13:49.160 align:middle line:84% And then you had plans that had different numbers of included 01:13:49.160 --> 01:13:51.950 align:middle line:84% usage times like, at the time it might 01:13:51.950 --> 01:13:55.320 align:middle line:84% have been 100 minutes of usage or 400 minutes of usage 01:13:55.320 --> 01:13:58.050 align:middle line:90% or 700 minutes of usage. 01:13:58.050 --> 01:14:01.000 align:middle line:84% And so each of these plans was a three part tariff. 01:14:01.000 --> 01:14:02.380 align:middle line:90% We paid some higher upfront fee. 01:14:02.380 --> 01:14:03.660 align:middle line:90% You got the 100 minutes. 01:14:03.660 --> 01:14:05.040 align:middle line:84% And then beyond that, you started 01:14:05.040 --> 01:14:06.960 align:middle line:90% to pay these overage charges. 01:14:06.960 --> 01:14:11.190 align:middle line:84% So anyway, this unnamed university offering four plans 01:14:11.190 --> 01:14:12.500 align:middle line:90% had a structure like that. 01:14:12.500 --> 01:14:15.240 align:middle line:90% 01:14:15.240 --> 01:14:19.770 align:middle line:84% He got data on what plans people signed up for 01:14:19.770 --> 01:14:24.480 align:middle line:84% and how much usage they actually made. 01:14:24.480 --> 01:14:25.840 align:middle line:90% So what are some of the results? 01:14:25.840 --> 01:14:28.740 align:middle line:84% So plan one customers-- let's suppose 01:14:28.740 --> 01:14:33.660 align:middle line:84% that this is plan one, one of the most popular plans. 01:14:33.660 --> 01:14:37.560 align:middle line:84% What he finds is that customers on that plan, 01:14:37.560 --> 01:14:42.120 align:middle line:84% 56% of the time they underutilized their plan 01:14:42.120 --> 01:14:46.320 align:middle line:84% and used so few minutes that they would have been better 01:14:46.320 --> 01:14:49.110 align:middle line:90% off on the per minute plan. 01:14:49.110 --> 01:14:52.620 align:middle line:84% 16% of the time they ran up overages 01:14:52.620 --> 01:14:57.390 align:middle line:84% and were on this part of the tariff, and only 28% of the time 01:14:57.390 --> 01:15:00.240 align:middle line:84% were in of sweet spot where plan one is actually better 01:15:00.240 --> 01:15:03.570 align:middle line:90% than a regular monthly plan. 01:15:03.570 --> 01:15:08.370 align:middle line:84% And so when he estimates what the consumers would 01:15:08.370 --> 01:15:11.460 align:middle line:84% have paid had they just all bought the linear plan, 01:15:11.460 --> 01:15:15.768 align:middle line:84% he finds that, on average, the saving was 42%. 01:15:15.768 --> 01:15:18.060 align:middle line:84% Now that doesn't mean that any individual consumer made 01:15:18.060 --> 01:15:19.500 align:middle line:84% a mistake, because each individual consumer, 01:15:19.500 --> 01:15:21.042 align:middle line:84% you don't know what their belief was. 01:15:21.042 --> 01:15:23.550 align:middle line:84% But certainly if, on average, all the consumers 01:15:23.550 --> 01:15:27.510 align:middle line:84% would have saved 40% by just buying the flat rate plan, 01:15:27.510 --> 01:15:32.430 align:middle line:84% then some consumers must be irrational buying these plans 01:15:32.430 --> 01:15:34.500 align:middle line:84% and must be getting exploited by them. 01:15:34.500 --> 01:15:37.330 align:middle line:84% And then similar calculation for the plan two customers. 01:15:37.330 --> 01:15:43.530 align:middle line:84% Again, there even more of them didn't hit this point. 01:15:43.530 --> 01:15:45.330 align:middle line:84% Again, there's a small percentage who 01:15:45.330 --> 01:15:47.190 align:middle line:84% hit the overage range, but the overage 01:15:47.190 --> 01:15:50.940 align:middle line:84% range, they pay so much more that they account for-- 01:15:50.940 --> 01:15:53.427 align:middle line:84% they're paying two and a half times as much 01:15:53.427 --> 01:15:55.510 align:middle line:84% as they would pay had they gotten the linear plan. 01:15:55.510 --> 01:15:59.940 align:middle line:84% So their overage customers are way up here. 01:15:59.940 --> 01:16:04.985 align:middle line:84% And the overages account for 23% of salary revenues. 01:16:04.985 --> 01:16:10.930 align:middle line:90% 01:16:10.930 --> 01:16:15.610 align:middle line:84% He's got something in the paper arguing, 01:16:15.610 --> 01:16:18.320 align:middle line:84% could this be due to price discrimination 01:16:18.320 --> 01:16:19.820 align:middle line:84% rather than to exploiting consumers? 01:16:19.820 --> 01:16:22.403 align:middle line:84% And then he has some conditions on the distribution they would 01:16:22.403 --> 01:16:24.071 align:middle line:84% need for this to be price discrimination 01:16:24.071 --> 01:16:25.852 align:middle line:84% and not exploiting irrational consumers. 01:16:25.852 --> 01:16:28.310 align:middle line:84% And he argues, no, this is exploiting irrational consumers. 01:16:28.310 --> 01:16:31.210 align:middle line:90% 01:16:31.210 --> 01:16:31.783 align:middle line:90% OK. 01:16:31.783 --> 01:16:33.200 align:middle line:84% I guess I'm out of time for today. 01:16:33.200 --> 01:16:40.690 align:middle line:84% So anyway, I have slides on here on a couple other papers. 01:16:40.690 --> 01:16:43.240 align:middle line:84% One I think is very cute is this is just 01:16:43.240 --> 01:16:47.958 align:middle line:84% another version of exploiting the hyperbolic consumers, 01:16:47.958 --> 01:16:49.750 align:middle line:84% noting that another real constraint you get 01:16:49.750 --> 01:16:52.420 align:middle line:84% with hyperbolic consumers is that you can exploit 01:16:52.420 --> 01:16:55.420 align:middle line:84% them even more if they think they're 01:16:55.420 --> 01:16:57.260 align:middle line:90% going to switch tomorrow. 01:16:57.260 --> 01:16:59.470 align:middle line:84% And so a constraint that hyperbolic consumers have 01:16:59.470 --> 01:17:03.160 align:middle line:84% is not just that I think I'm going to go to the gym more 01:17:03.160 --> 01:17:05.567 align:middle line:84% than I am, but I think I'm going to cancel. 01:17:05.567 --> 01:17:07.900 align:middle line:84% As long as I have the option to cancel my gym membership 01:17:07.900 --> 01:17:10.330 align:middle line:84% tomorrow if I determine I don't like the gym, 01:17:10.330 --> 01:17:11.840 align:middle line:84% I can always just cancel it tomorrow 01:17:11.840 --> 01:17:13.840 align:middle line:84% so I might as well stay signed up today. 01:17:13.840 --> 01:17:17.020 align:middle line:84% And noting that this sort of friendly option of letting 01:17:17.020 --> 01:17:20.357 align:middle line:84% consumers cancel at any time can be a very effective way 01:17:20.357 --> 01:17:21.940 align:middle line:84% to get them to pay much more than they 01:17:21.940 --> 01:17:25.330 align:middle line:84% would pay if you didn't have that extra option in there. 01:17:25.330 --> 01:17:26.290 align:middle line:90% OK? 01:17:26.290 --> 01:17:28.960 align:middle line:84% And then, yeah, I think that's it. 01:17:28.960 --> 01:17:31.870 align:middle line:84% That's all I'll talk about for today. 01:17:31.870 --> 01:17:49.000 align:middle line:90%