WEBVTT 00:00:00.000 --> 00:00:01.992 [SQUEAKING] 00:00:01.992 --> 00:00:03.486 [RUSTLING] 00:00:03.486 --> 00:00:04.980 [CLICKING] 00:00:11.377 --> 00:00:12.210 FRANK SCHILBACH: OK. 00:00:12.210 --> 00:00:13.770 So let me sort of just recap what 00:00:13.770 --> 00:00:16.140 we discussed last time fairly quickly, 00:00:16.140 --> 00:00:18.180 and then I want to move to empirical evidence 00:00:18.180 --> 00:00:19.710 on a variety of settings. 00:00:19.710 --> 00:00:22.050 We might not get through all of the slides. 00:00:22.050 --> 00:00:23.070 That's fine. 00:00:23.070 --> 00:00:27.370 In that case, we'll just discuss some of this in recitation. 00:00:27.370 --> 00:00:30.210 So let me sort of recap what Kahneman and Tversky, 00:00:30.210 --> 00:00:34.140 in their 1979 article, were proposing 00:00:34.140 --> 00:00:36.510 based on a bunch of experiments and empirical evidence 00:00:36.510 --> 00:00:38.610 that they had collected. 00:00:38.610 --> 00:00:42.840 So their theory, it was called prospect theory 00:00:42.840 --> 00:00:43.950 what they proposed. 00:00:43.950 --> 00:00:46.050 Versions of prospect theory is essentially 00:00:46.050 --> 00:00:49.500 versions of reference dependent utility, have been used 00:00:49.500 --> 00:00:52.608 or used prominently now in economics. 00:00:52.608 --> 00:00:54.150 So the first thing that they proposed 00:00:54.150 --> 00:00:57.550 was, what matters, what the carrier of utility is, 00:00:57.550 --> 00:01:00.390 is changes rather than levels. 00:01:00.390 --> 00:01:03.370 That's to say it doesn't matter for you necessarily how warm it 00:01:03.370 --> 00:01:03.870 is. 00:01:03.870 --> 00:01:06.370 It matters kind of what's the change of temperature compared 00:01:06.370 --> 00:01:07.140 to yesterday. 00:01:07.140 --> 00:01:08.640 It doesn't matter that much how much 00:01:08.640 --> 00:01:10.170 money people have in total. 00:01:10.170 --> 00:01:13.590 It matters how much that changes relative to what 00:01:13.590 --> 00:01:16.290 they had previously. 00:01:16.290 --> 00:01:19.750 More generally, what matters for people is essentially 00:01:19.750 --> 00:01:23.220 a certain consumption or the like relative 00:01:23.220 --> 00:01:27.240 to reference point as opposed to in absolute terms. 00:01:27.240 --> 00:01:31.350 Second, loss aversion, this is losses loom larger than gains. 00:01:31.350 --> 00:01:33.910 And we had some evidence of that in the last lecture. 00:01:33.910 --> 00:01:37.110 That's to say, if you lose some money or some consumption 00:01:37.110 --> 00:01:41.040 or anything else, grades, et cetera, 00:01:41.040 --> 00:01:43.350 a loss relative to either your status 00:01:43.350 --> 00:01:47.100 quo or to your expectation looms larger, is more important, 00:01:47.100 --> 00:01:50.010 hurts more than a gain of the same magnitude 00:01:50.010 --> 00:01:52.200 in the positive direction. 00:01:52.200 --> 00:01:54.660 And number three, which we talked about quickly, 00:01:54.660 --> 00:01:58.380 is what's referred to as diminishing sensitivity. 00:01:58.380 --> 00:02:01.470 That is people are risk averse in the gain domain, 00:02:01.470 --> 00:02:04.020 but risk loving in the loss domain. 00:02:04.020 --> 00:02:09.539 That's to say, for example, if you think about distance, time, 00:02:09.539 --> 00:02:13.380 chance, and the like, going from 0 to 1 or from 1 to 2 00:02:13.380 --> 00:02:15.810 or from 2 to 3 is more important for you 00:02:15.810 --> 00:02:20.400 than going from, like, 100 to 101, 101 to 102, and so on. 00:02:20.400 --> 00:02:23.700 Essentially, the further you go away 00:02:23.700 --> 00:02:27.120 from your status quo, your reference point, 00:02:27.120 --> 00:02:29.490 any marginal change is diminishing, right? 00:02:29.490 --> 00:02:33.310 And that's true for both the gain domain and the loss 00:02:33.310 --> 00:02:33.810 domain. 00:02:33.810 --> 00:02:36.720 So these are sort of the main three characteristics 00:02:36.720 --> 00:02:39.728 related to what Kahneman-Tversky called the value function. 00:02:39.728 --> 00:02:41.520 We can think of this as essentially version 00:02:41.520 --> 00:02:43.267 of the utility function. 00:02:43.267 --> 00:02:44.850 There's a fourth characteristic, which 00:02:44.850 --> 00:02:46.808 is probability weighting, which we're not going 00:02:46.808 --> 00:02:49.170 to talk about at least for now. 00:02:49.170 --> 00:02:51.870 Are there any questions on these three things so far? 00:02:57.840 --> 00:02:58.610 OK. 00:02:58.610 --> 00:03:04.310 So now, prospect theory is then what Kahneman and Tversky 00:03:04.310 --> 00:03:05.120 were proposing. 00:03:05.120 --> 00:03:07.010 They were essentially saying, instead of 00:03:07.010 --> 00:03:10.820 having concave utility, which I showed you last week, 00:03:10.820 --> 00:03:13.790 instead your utility function may look like this. 00:03:13.790 --> 00:03:16.580 And what are sort of the features of this utility 00:03:16.580 --> 00:03:17.360 function? 00:03:17.360 --> 00:03:19.913 What are the three key features that I just showed you? 00:03:19.913 --> 00:03:21.830 How are they showing up in this function here? 00:03:32.480 --> 00:03:34.040 Yes? 00:03:34.040 --> 00:03:35.873 AUDIENCE: C minus [INAUDIBLE] is the change? 00:03:35.873 --> 00:03:36.832 FRANK SCHILBACH: Right. 00:03:36.832 --> 00:03:38.610 So the carrier of the utility function-- 00:03:38.610 --> 00:03:41.030 so what we have here is c and r. 00:03:41.030 --> 00:03:42.082 c is like consumption. 00:03:42.082 --> 00:03:43.040 That could be anything. 00:03:43.040 --> 00:03:43.850 That could be apples. 00:03:43.850 --> 00:03:44.767 That could be bananas. 00:03:44.767 --> 00:03:47.990 That could be sort of how much money people have 00:03:47.990 --> 00:03:50.840 available to consume overall. 00:03:50.840 --> 00:03:53.480 The carrier of the utility function is not c itself. 00:03:53.480 --> 00:03:56.360 It's c minus r, so it's c of relative to some reference 00:03:56.360 --> 00:03:57.840 point r. 00:03:57.840 --> 00:03:59.240 That's exactly right. 00:03:59.240 --> 00:04:01.520 And if r is the status quo, if r is 00:04:01.520 --> 00:04:04.310 how much we have right now or the person has right now, 00:04:04.310 --> 00:04:07.430 c minus r is the change relative to the status quo. 00:04:07.430 --> 00:04:10.005 Notice that r-- we're going to talk about this a little bit-- 00:04:10.005 --> 00:04:11.130 could be also other things. 00:04:11.130 --> 00:04:13.500 It doesn't have to be necessarily the status quo. 00:04:13.500 --> 00:04:16.100 It could be also people's expectation or their goals 00:04:16.100 --> 00:04:18.380 or their aspirations for the future, right? 00:04:18.380 --> 00:04:20.089 The key part is, however, what matters 00:04:20.089 --> 00:04:21.980 is-- this is the first thing I was saying. 00:04:21.980 --> 00:04:24.050 It's changes rather than levels, changes 00:04:24.050 --> 00:04:26.960 relative to some reference point or consumption relative 00:04:26.960 --> 00:04:28.252 to some reference point. 00:04:28.252 --> 00:04:29.210 That's number one, yes. 00:04:32.090 --> 00:04:33.155 Yes? 00:04:33.155 --> 00:04:34.530 AUDIENCE: The curve flattens out, 00:04:34.530 --> 00:04:36.140 which is diminishing sensitivity. 00:04:36.140 --> 00:04:36.590 FRANK SCHILBACH: Exactly. 00:04:36.590 --> 00:04:38.670 The curve flattens out in both directions. 00:04:38.670 --> 00:04:40.670 That's diminishing sensitivity. 00:04:40.670 --> 00:04:43.340 It's essentially concave in the gain domain, 00:04:43.340 --> 00:04:46.220 in the domain where c is larger than r. 00:04:46.220 --> 00:04:48.020 And it's convex in the loss domain 00:04:48.020 --> 00:04:51.870 where c is smaller than r, right? 00:04:51.870 --> 00:04:54.920 And that's exactly the issue that essentially the first, 00:04:54.920 --> 00:05:00.050 the marginal change, going from, say, 1 to 2 00:05:00.050 --> 00:05:02.510 is relatively large compared to a marginal change going 00:05:02.510 --> 00:05:03.620 from 10 to 11. 00:05:03.620 --> 00:05:06.350 That's both going in the right direction in the gains domain 00:05:06.350 --> 00:05:08.450 and the left direction in the loss domain. 00:05:08.450 --> 00:05:09.420 Yes? 00:05:09.420 --> 00:05:12.960 AUDIENCE: For the loss aversion, the left side 00:05:12.960 --> 00:05:14.630 is steeper than the right side. 00:05:14.630 --> 00:05:15.672 FRANK SCHILBACH: Exactly. 00:05:15.672 --> 00:05:18.290 In particular, there's a kink in this function. 00:05:18.290 --> 00:05:25.100 When you look at 0, where c equals r, the gain and loss 00:05:25.100 --> 00:05:29.720 domain intersect, there we have a kink in the value function, 00:05:29.720 --> 00:05:32.510 in the utility function, which essentially exactly is the loss 00:05:32.510 --> 00:05:35.210 domain, the loss aversion which is like going to the right 00:05:35.210 --> 00:05:37.190 is less steep than going to the left. 00:05:37.190 --> 00:05:38.870 Put differently, if you lose, if you're 00:05:38.870 --> 00:05:40.820 sort of below the reference point, 00:05:40.820 --> 00:05:45.807 that's more painful than being the same amount of units to be 00:05:45.807 --> 00:05:46.890 above the reference point. 00:05:46.890 --> 00:05:47.640 That's number two. 00:05:47.640 --> 00:05:49.980 That's essentially exactly the loss aversion. 00:05:49.980 --> 00:05:52.390 So I just wrote down all of that. 00:05:52.390 --> 00:05:53.480 Again, let me repeat. 00:05:53.480 --> 00:05:55.880 Carrier of the utility changes relative to the reference 00:05:55.880 --> 00:05:56.720 point-- 00:05:56.720 --> 00:05:58.940 excuse me-- rather than levels. 00:05:58.940 --> 00:06:01.010 Second, there's loss aversion. 00:06:01.010 --> 00:06:02.750 There's a kink at 0 in this function. 00:06:02.750 --> 00:06:05.640 And three, there's diminishing sensitivity, 00:06:05.640 --> 00:06:08.270 which is concavity in gains and convexity in losses. 00:06:11.920 --> 00:06:13.090 Any questions on that? 00:06:15.810 --> 00:06:17.343 Now, a key question here is-- 00:06:17.343 --> 00:06:18.510 I want to sort of flag this. 00:06:18.510 --> 00:06:19.740 We're going to talk about this a little bit 00:06:19.740 --> 00:06:20.823 at the end of the lecture. 00:06:20.823 --> 00:06:23.323 We're going to talk to this a little bit in the next problem 00:06:23.323 --> 00:06:25.740 set is kind of like how is the reference point determined. 00:06:25.740 --> 00:06:27.540 And does it matter? 00:06:27.540 --> 00:06:29.550 As I said before, in Kahneman-Tversky's work, 00:06:29.550 --> 00:06:31.840 a lot of the reference point is the status quo. 00:06:31.840 --> 00:06:34.020 So they essentially postulated the status quo is 00:06:34.020 --> 00:06:36.510 what really matters originally. 00:06:36.510 --> 00:06:39.583 I think people have moved toward saying the reference point. 00:06:39.583 --> 00:06:42.000 And this is what Koszegi and Rabin and others have written 00:06:42.000 --> 00:06:44.430 down in their models of reference dependence utility 00:06:44.430 --> 00:06:49.110 and recent more economics work is what really matters 00:06:49.110 --> 00:06:50.440 is expectation. 00:06:50.440 --> 00:06:54.690 So what do you expect to consume or to have and so on? 00:06:54.690 --> 00:06:55.710 That matters. 00:06:55.710 --> 00:06:58.325 That sometimes coincides with the status quo. 00:06:58.325 --> 00:06:59.700 For example, if you have a house, 00:06:59.700 --> 00:07:01.367 the status quo is that you have a house. 00:07:01.367 --> 00:07:04.050 You probably assume that you have a house in the future. 00:07:04.050 --> 00:07:05.340 These things coincide. 00:07:05.340 --> 00:07:06.900 If you think about wages, et cetera, 00:07:06.900 --> 00:07:09.210 what seems to matter often is not so much 00:07:09.210 --> 00:07:11.370 what people's wages are right now. 00:07:11.370 --> 00:07:13.770 What matters is what wage gains and so on do 00:07:13.770 --> 00:07:15.720 they expect to receive in the future. 00:07:15.720 --> 00:07:19.800 And they evaluate their future, their outcomes in the future, 00:07:19.800 --> 00:07:22.740 not necessarily to the current wage, but rather 00:07:22.740 --> 00:07:25.690 what they expect they would get in the future. 00:07:25.690 --> 00:07:26.190 OK. 00:07:26.190 --> 00:07:27.810 And so here's an example. 00:07:27.810 --> 00:07:30.210 And you'll have some problems set questions. 00:07:30.210 --> 00:07:32.040 Again, this is a problem set three-- 00:07:32.040 --> 00:07:34.440 which is not posted yet, but will be-- 00:07:34.440 --> 00:07:37.290 of that kind, which is you might have reference 00:07:37.290 --> 00:07:39.060 dependent utility over-- 00:07:39.060 --> 00:07:40.680 this is just a very simple example-- 00:07:40.680 --> 00:07:41.760 shirts and money. 00:07:41.760 --> 00:07:43.780 So you have essentially two different domains. 00:07:43.780 --> 00:07:45.630 You have, essentially, losses and gains over 00:07:45.630 --> 00:07:47.040 both of those domains. 00:07:47.040 --> 00:07:48.810 You have a reference point, which is rs, 00:07:48.810 --> 00:07:50.227 is how many shirts you might have. 00:07:50.227 --> 00:07:53.100 You have a reference point, rm, how much money you might have. 00:07:53.100 --> 00:07:54.900 Now, what's important here is that, 00:07:54.900 --> 00:07:57.150 essentially, when you think about buying a shirt 00:07:57.150 --> 00:08:00.420 or selling a shirt, there's going 00:08:00.420 --> 00:08:03.547 to be two dimensions which you have to sort of consider. 00:08:03.547 --> 00:08:05.130 And when you think about the endowment 00:08:05.130 --> 00:08:07.500 effect of [INAUDIBLE] and so on, you 00:08:07.500 --> 00:08:11.830 have to consider not only the losses 00:08:11.830 --> 00:08:15.173 and gains and shirts, but also the losses and gains and money. 00:08:15.173 --> 00:08:17.340 So that's to say, if you're trying to buy something, 00:08:17.340 --> 00:08:18.840 you're going to get a gain in shirts 00:08:18.840 --> 00:08:21.060 relative to the reference point if it's unexpected. 00:08:21.060 --> 00:08:22.890 But you'll also have a loss in money. 00:08:22.890 --> 00:08:24.720 Similarly, if you're going to sell a shirt, 00:08:24.720 --> 00:08:28.650 you're going to have a loss in shirts, but a gain in money. 00:08:28.650 --> 00:08:30.390 And these things then interact. 00:08:30.390 --> 00:08:31.728 Now, what's the value function? 00:08:31.728 --> 00:08:33.270 The value function, as I said before, 00:08:33.270 --> 00:08:35.909 is usually concave in the gain domain 00:08:35.909 --> 00:08:38.679 and convex in the last domain. 00:08:38.679 --> 00:08:39.630 There's a kink at 0. 00:08:39.630 --> 00:08:41.640 It's steeper on the left than on the right. 00:08:41.640 --> 00:08:47.670 Usually, we think the relative slope is about 2, 2.5, OK? 00:08:47.670 --> 00:08:50.010 So one version of that would be this function 00:08:50.010 --> 00:08:50.940 that I wrote down. 00:08:50.940 --> 00:08:52.920 Again, there will be some problem 00:08:52.920 --> 00:08:56.080 set questions, et cetera, sort of to clarify that. 00:08:56.080 --> 00:08:58.050 But one key question here is, then 00:08:58.050 --> 00:08:59.370 what are the different domains? 00:08:59.370 --> 00:09:02.020 And that's kind of a question of mental accounting. 00:09:02.020 --> 00:09:04.235 We'll get to this in the second half of the course. 00:09:04.235 --> 00:09:05.610 The question's kind of like, what 00:09:05.610 --> 00:09:06.818 are the different categories? 00:09:06.818 --> 00:09:07.950 Do you have shirts? 00:09:07.950 --> 00:09:10.800 Do you have pants, sweaters? 00:09:10.800 --> 00:09:13.680 Or is it just for clothes overall? 00:09:13.680 --> 00:09:17.490 Or when you think about earnings and consumption, et cetera, 00:09:17.490 --> 00:09:18.510 is it daily consumption? 00:09:18.510 --> 00:09:19.560 Is it weekly consumption? 00:09:19.560 --> 00:09:21.190 Is it monthly consumption? 00:09:21.190 --> 00:09:24.150 So there's lots of questions on how to exactly specify 00:09:24.150 --> 00:09:25.560 this utility function. 00:09:25.560 --> 00:09:29.590 These questions are mostly unanswered in the literature. 00:09:29.590 --> 00:09:31.445 So for now, for us in our purposes, 00:09:31.445 --> 00:09:32.820 we're going to essentially assume 00:09:32.820 --> 00:09:35.490 there is a value function given and then 00:09:35.490 --> 00:09:36.900 sort of work with that. 00:09:39.560 --> 00:09:40.610 Any questions on that? 00:09:43.440 --> 00:09:45.120 OK, great. 00:09:45.120 --> 00:09:47.160 So now, we're going to talk about a number 00:09:47.160 --> 00:09:48.310 of different applications. 00:09:48.310 --> 00:09:51.480 We talked a little bit already about the endowment effect 00:09:51.480 --> 00:09:52.380 and about insurance. 00:09:52.380 --> 00:09:54.660 I'm going to skip this. 00:09:54.660 --> 00:09:58.565 There's some of this already in recitation in the problem sets. 00:09:58.565 --> 00:10:00.690 We're going to particularly talk about labor supply 00:10:00.690 --> 00:10:02.310 and employment decisions, essentially 00:10:02.310 --> 00:10:03.990 how much do people like to work. 00:10:03.990 --> 00:10:06.150 And is effort or people's work decisions, 00:10:06.150 --> 00:10:08.430 are they reference dependent? 00:10:08.430 --> 00:10:10.180 We're going to talk about finances, 00:10:10.180 --> 00:10:12.540 what was mentioned last time already about investment 00:10:12.540 --> 00:10:13.110 decisions. 00:10:13.110 --> 00:10:14.963 When do people sell and buy stocks? 00:10:14.963 --> 00:10:16.380 We're going to talk about housing. 00:10:16.380 --> 00:10:21.300 When do people decide to sell or buy a house and at what price? 00:10:21.300 --> 00:10:23.970 We're going to talk about sports, in particular marathon 00:10:23.970 --> 00:10:27.510 running and golf. 00:10:27.510 --> 00:10:29.280 There's some papers on domestic violence 00:10:29.280 --> 00:10:30.990 we're going to talk very briefly about. 00:10:30.990 --> 00:10:32.740 And then we talk a little bit about firms. 00:10:32.740 --> 00:10:35.790 How do firms think about pricing and so on? 00:10:35.790 --> 00:10:38.340 What is sort of the market response 00:10:38.340 --> 00:10:40.140 to reference dependence? 00:10:40.140 --> 00:10:42.690 That's to say, given that we know there's lots of reference 00:10:42.690 --> 00:10:45.900 dependence in the world, now, as a firm 00:10:45.900 --> 00:10:50.070 or treating other people, how should we think about, 00:10:50.070 --> 00:10:52.020 how does that affect, our own behavior 00:10:52.020 --> 00:10:54.750 or how maybe firms interact with us? 00:10:54.750 --> 00:10:55.710 OK. 00:10:55.710 --> 00:10:58.230 So let's start with labor supply. 00:10:58.230 --> 00:11:02.130 Let's start with a very simple example. 00:11:02.130 --> 00:11:04.650 Suppose there's a worker in the following situation. 00:11:04.650 --> 00:11:08.110 She can freely choose how many hours she works every day. 00:11:08.110 --> 00:11:12.330 And there are frequent temporary changes in her hourly wage. 00:11:12.330 --> 00:11:14.220 Now, there's different relationships 00:11:14.220 --> 00:11:16.620 between wages and hours per day. 00:11:16.620 --> 00:11:19.080 You could always work the same number of hours. 00:11:19.080 --> 00:11:21.960 You could work more hours on the days when wages are high. 00:11:21.960 --> 00:11:26.137 Or you could work fewer hours on days when the wages are high. 00:11:26.137 --> 00:11:27.720 What does sort of standard theory say? 00:11:27.720 --> 00:11:28.590 What should you do? 00:11:38.440 --> 00:11:39.130 Yes? 00:11:39.130 --> 00:11:39.700 Two? 00:11:39.700 --> 00:11:40.300 AUDIENCE: Two. 00:11:40.300 --> 00:11:42.190 FRANK SCHILBACH: Why is that? 00:11:42.190 --> 00:11:46.262 AUDIENCE: Because that maximizes your expected value for time. 00:11:46.262 --> 00:11:47.220 FRANK SCHILBACH: Right. 00:11:47.220 --> 00:11:51.070 So if hours are effortful, you usually don't like to work. 00:11:51.070 --> 00:11:54.370 You should work during the hours when 00:11:54.370 --> 00:11:56.558 your payment is the largest. 00:11:56.558 --> 00:11:57.100 That's right. 00:11:57.100 --> 00:11:59.920 So what about option number one? 00:11:59.920 --> 00:12:02.450 Why might option number one be optimal? 00:12:02.450 --> 00:12:04.225 So what you're saying is exactly right. 00:12:04.225 --> 00:12:06.442 It depends a little bit on something else. 00:12:06.442 --> 00:12:07.150 And what is that? 00:12:10.685 --> 00:12:12.310 Why might you choose number one anyway? 00:12:15.490 --> 00:12:16.970 AUDIENCE: Habit-forming is nice. 00:12:16.970 --> 00:12:18.640 FRANK SCHILBACH: Yeah, there could be sort of habits, 00:12:18.640 --> 00:12:19.130 exactly. 00:12:19.130 --> 00:12:21.047 Or it could just be that it's really effortful 00:12:21.047 --> 00:12:22.750 to work 12 hours in one day. 00:12:22.750 --> 00:12:26.180 Maybe there's kids at home, or maybe it's just really tedious 00:12:26.180 --> 00:12:26.680 to work. 00:12:26.680 --> 00:12:29.430 At some point, you just want to go home and do other stuff. 00:12:29.430 --> 00:12:31.900 So it could just be that working beyond, say, eight or nine 00:12:31.900 --> 00:12:33.435 hours per day is really tough to do. 00:12:33.435 --> 00:12:34.810 So then you might say, I'm always 00:12:34.810 --> 00:12:35.893 going to work eight hours. 00:12:35.893 --> 00:12:38.385 I'd love to work more, but it's difficult to do. 00:12:38.385 --> 00:12:39.760 So essentially, it's just to say, 00:12:39.760 --> 00:12:43.930 if the function of the effort as a function of hours is convex, 00:12:43.930 --> 00:12:46.845 then you might sort of keep it the same. 00:12:46.845 --> 00:12:48.220 Surely, what you don't want to do 00:12:48.220 --> 00:12:52.210 is number three, working fewer hours 00:12:52.210 --> 00:12:55.300 on days when wages are high unless effort costs are 00:12:55.300 --> 00:12:56.770 particularly high on those days. 00:12:56.770 --> 00:12:58.750 So it could be that it's really super hot. 00:12:58.750 --> 00:13:00.990 Or it could be it's tedious to work on those days. 00:13:00.990 --> 00:13:02.990 And you might say, then you don't want to do it. 00:13:02.990 --> 00:13:04.480 But assuming that's not the case, 00:13:04.480 --> 00:13:05.788 you kind of want to avoid this. 00:13:05.788 --> 00:13:07.330 Because for the same number of hours, 00:13:07.330 --> 00:13:10.710 you're going to make less money overall. 00:13:10.710 --> 00:13:13.010 And so here's a concrete example. 00:13:13.010 --> 00:13:15.280 Suppose wage is 5 hours an hour on day one 00:13:15.280 --> 00:13:17.190 and 10 hours a day on day two. 00:13:17.190 --> 00:13:18.700 So there's three strategies. 00:13:18.700 --> 00:13:20.170 You can work 8 hours on both days. 00:13:20.170 --> 00:13:22.960 You can work 6 hours on day one, 9 hours on day two, 00:13:22.960 --> 00:13:26.950 or the opposite, 9 hours on day one and 6 hours on day two. 00:13:26.950 --> 00:13:30.313 So if you do that, you can sort of calculate how much that is. 00:13:30.313 --> 00:13:32.230 You can essentially work 8 hours on both days. 00:13:32.230 --> 00:13:33.310 You get 120 hours. 00:13:33.310 --> 00:13:38.980 You can work 6 hours on days one and two, 00:13:38.980 --> 00:13:41.200 which makes $120 as well. 00:13:41.200 --> 00:13:45.130 Or you can essentially do the opposite, which makes you $105. 00:13:45.130 --> 00:13:46.510 This is what I was saying. 00:13:46.510 --> 00:13:49.180 Option three doesn't make a lot of sense 00:13:49.180 --> 00:13:50.950 unless effort costs are particularly 00:13:50.950 --> 00:13:53.350 high on certain days when the wages are high. 00:13:53.350 --> 00:13:55.730 We're assuming that away for now. 00:13:55.730 --> 00:13:56.688 And so now option two-- 00:13:56.688 --> 00:13:58.605 and this is what you were saying-- essentially 00:13:58.605 --> 00:13:59.770 saves you an hour overall. 00:13:59.770 --> 00:14:01.950 You work only 15 hours instead of 16 hours. 00:14:01.950 --> 00:14:04.270 And you make the same amount of money. 00:14:04.270 --> 00:14:10.360 Now, unless the ninth hour is extremely costly for you to do, 00:14:10.360 --> 00:14:12.410 you might not want to do that. 00:14:12.410 --> 00:14:14.200 OK. 00:14:14.200 --> 00:14:19.390 Now, why might you do something else instead? 00:14:19.390 --> 00:14:22.000 Or where does reference dependence come in here? 00:14:25.010 --> 00:14:26.100 Yes. 00:14:26.100 --> 00:14:29.105 AUDIENCE: I see on the high wage day you make more 00:14:29.105 --> 00:14:31.743 than the less wage day, so I feel like stopping maybe. 00:14:31.743 --> 00:14:33.410 FRANK SCHILBACH: And why would you stop? 00:14:33.410 --> 00:14:36.686 What's causing you to stop? 00:14:36.686 --> 00:14:40.900 AUDIENCE: [INAUDIBLE] wages higher relative to [INAUDIBLE].. 00:14:40.900 --> 00:14:41.920 FRANK SCHILBACH: Yeah. 00:14:41.920 --> 00:14:42.520 Yes. 00:14:42.520 --> 00:14:45.880 And so what are you evaluating? 00:14:45.880 --> 00:14:53.540 Or what's the-- or what happens, for example, 00:14:53.540 --> 00:14:58.563 if you work only 6 hours on a different day? 00:14:58.563 --> 00:15:00.730 How much do you make on the sixth hour day, I guess, 00:15:00.730 --> 00:15:04.960 which would be $30, right? 00:15:04.960 --> 00:15:09.448 And so what are you comparing that to, I guess? 00:15:09.448 --> 00:15:11.490 AUDIENCE: You're comparing it to the actual money 00:15:11.490 --> 00:15:12.952 you made at the end of the day. 00:15:12.952 --> 00:15:13.910 FRANK SCHILBACH: Right. 00:15:13.910 --> 00:15:16.213 But so suppose you, on average, want 00:15:16.213 --> 00:15:17.630 to make a certain amount of money, 00:15:17.630 --> 00:15:19.870 which is $50, $60 and so on. 00:15:19.870 --> 00:15:22.370 Now, if on some days you make a lot of money and on some day 00:15:22.370 --> 00:15:24.850 you make very little money, you might sort of 00:15:24.850 --> 00:15:27.082 evaluate that separately and say, on that day, 00:15:27.082 --> 00:15:28.540 I'm essentially in the loss domain. 00:15:28.540 --> 00:15:31.880 I'm below my target or below my expectation. 00:15:31.880 --> 00:15:34.840 And so if you evaluate your utility that way, 00:15:34.840 --> 00:15:37.360 you might sort of not want that because it feels essentially 00:15:37.360 --> 00:15:39.070 you're below a certain threshold. 00:15:39.070 --> 00:15:41.700 It feels kind of like a loss relative to your expectation 00:15:41.700 --> 00:15:43.810 and so on when you might be inclined to work 00:15:43.810 --> 00:15:45.760 a lot of hours. 00:15:45.760 --> 00:15:48.110 Instead, on the days when you make a lot of money, 00:15:48.110 --> 00:15:48.910 why might you stop? 00:15:52.310 --> 00:15:53.160 Yes? 00:15:53.160 --> 00:15:54.705 AUDIENCE: You might have a target 00:15:54.705 --> 00:15:57.080 that you expect to meet every day to cover your expenses. 00:15:57.080 --> 00:16:00.210 And if you feel like the work [INAUDIBLE] reach that target, 00:16:00.210 --> 00:16:02.142 you might [INAUDIBLE]. 00:16:02.142 --> 00:16:03.100 FRANK SCHILBACH: Right. 00:16:03.100 --> 00:16:04.740 So if you target, your reference point, 00:16:04.740 --> 00:16:07.470 is essentially a certain number, you 00:16:07.470 --> 00:16:09.750 might reach that target quickly because your wage 00:16:09.750 --> 00:16:11.232 is pretty high on that day. 00:16:11.232 --> 00:16:12.690 And once you reach that target, you 00:16:12.690 --> 00:16:14.640 might say, well, now utility function 00:16:14.640 --> 00:16:16.758 is relatively not very steep anymore 00:16:16.758 --> 00:16:18.300 relative to being in the loss domain. 00:16:18.300 --> 00:16:19.500 So it's flatter. 00:16:19.500 --> 00:16:21.840 So then you might just stop relatively soon 00:16:21.840 --> 00:16:25.590 because, essentially, any marginal earnings are not 00:16:25.590 --> 00:16:27.900 really valuable for you anymore. 00:16:27.900 --> 00:16:30.560 OK? 00:16:30.560 --> 00:16:32.180 Any questions on that or comments? 00:16:36.190 --> 00:16:39.372 So why do we want the wage changes to be temporary here? 00:16:39.372 --> 00:16:41.080 What's an issue here when you sort of try 00:16:41.080 --> 00:16:43.360 to look at this in the data? 00:16:43.360 --> 00:16:45.110 Suppose I had persistent wage changes. 00:16:45.110 --> 00:16:45.610 Yeah. 00:16:45.610 --> 00:16:48.713 AUDIENCE: So you can have a frame of reference? 00:16:48.713 --> 00:16:49.630 FRANK SCHILBACH: Yeah. 00:16:49.630 --> 00:16:51.630 You want to kind of keep the reference constant. 00:16:51.630 --> 00:16:54.423 So in some sense, if wage changes are permanent, 00:16:54.423 --> 00:16:56.090 then you get essentially income effects. 00:16:56.090 --> 00:16:57.610 You'll be a lot richer overall. 00:16:57.610 --> 00:16:59.465 If they're only temporary, essentially, 00:16:59.465 --> 00:17:01.840 you can sort of argue that, essentially, your expectation 00:17:01.840 --> 00:17:03.050 should be the same. 00:17:03.050 --> 00:17:05.567 And once you reach, once you have a lot more money, 00:17:05.567 --> 00:17:07.150 the neoclassical model should actually 00:17:07.150 --> 00:17:11.020 say that there should be income affects. 00:17:11.020 --> 00:17:12.910 Essentially, the neoclassical model says, 00:17:12.910 --> 00:17:15.190 you should essentially aggregate all of your income and say, 00:17:15.190 --> 00:17:17.023 it doesn't really matter whether you earn it 00:17:17.023 --> 00:17:19.119 on Monday, or Tuesday, Wednesday, or Thursday. 00:17:19.119 --> 00:17:21.640 You should look at how much are you earning overall 00:17:21.640 --> 00:17:24.609 depending whether you earn sufficiently much, 00:17:24.609 --> 00:17:27.130 you're going to work fewer or more hours. 00:17:27.130 --> 00:17:30.520 Now, if you earn a lot, because your wages doubles or whatever, 00:17:30.520 --> 00:17:32.230 you might actually work few hours 00:17:32.230 --> 00:17:34.655 not because you reach a reference point on a given day. 00:17:34.655 --> 00:17:36.280 It's just because you got a lot richer, 00:17:36.280 --> 00:17:37.697 and then you decide it's not worth 00:17:37.697 --> 00:17:38.837 for you to work that much. 00:17:38.837 --> 00:17:40.420 So we're kind of trying to avoid that. 00:17:40.420 --> 00:17:43.450 We're trying to have only temporary changes, which 00:17:43.450 --> 00:17:47.173 is to say, for given wealth overall on any given day, 00:17:47.173 --> 00:17:49.090 it shouldn't matter whether you earn the money 00:17:49.090 --> 00:17:50.630 on Monday or on Tuesday. 00:17:50.630 --> 00:17:52.575 So essentially, if the wage happens 00:17:52.575 --> 00:17:53.950 to be really high on Tuesday, you 00:17:53.950 --> 00:17:56.260 should be working more, earning more, on Tuesday 00:17:56.260 --> 00:17:58.055 as opposed to on Monday. 00:17:58.055 --> 00:18:00.430 Now, what I was saying is, if you're reference dependent, 00:18:00.430 --> 00:18:01.990 you might actually care about this. 00:18:01.990 --> 00:18:04.240 You might care about on Monday you didn't reach your target. 00:18:04.240 --> 00:18:05.740 Therefore, you want to work more. 00:18:05.740 --> 00:18:08.290 On Tuesday, you reached your target very quickly. 00:18:08.290 --> 00:18:12.280 And you work less even though the wage is actually 00:18:12.280 --> 00:18:13.810 really high. 00:18:13.810 --> 00:18:15.190 OK. 00:18:15.190 --> 00:18:20.170 Now, strategy one might be optimal even 00:18:20.170 --> 00:18:23.587 in the neoclassical model if effort costs are convex. 00:18:23.587 --> 00:18:25.420 This is what I was saying is, if it's really 00:18:25.420 --> 00:18:30.580 sort of costly for you to work 9 hours, you might say, 00:18:30.580 --> 00:18:31.697 I always work 8 hours. 00:18:31.697 --> 00:18:34.030 I want to have a certain amount of money for my children 00:18:34.030 --> 00:18:35.673 and so on. 00:18:35.673 --> 00:18:37.840 So the extra hour, if that's really, really painful, 00:18:37.840 --> 00:18:39.945 you might not want to do that. 00:18:39.945 --> 00:18:41.320 We don't think that's necessarily 00:18:41.320 --> 00:18:44.913 the case in so many situations. 00:18:44.913 --> 00:18:46.330 The question is, can we really say 00:18:46.330 --> 00:18:48.413 that strategy three-- that's the strategy of doing 00:18:48.413 --> 00:18:50.650 the opposite, of working essentially a lot of hours 00:18:50.650 --> 00:18:53.560 when wages are low and few hours when wages are high. 00:18:53.560 --> 00:18:55.610 Can we really say it's a mistake? 00:18:55.610 --> 00:18:58.120 Well, it depends on kind of whether the effort costs are 00:18:58.120 --> 00:18:59.450 correlated with wages. 00:18:59.450 --> 00:19:01.120 So if cab drivers, for example, make 00:19:01.120 --> 00:19:04.630 really high wages on some days and low wages on other days, 00:19:04.630 --> 00:19:09.700 it could just be on low wage day it's much less 00:19:09.700 --> 00:19:11.390 effortful to drive around. 00:19:11.390 --> 00:19:13.210 So you would do more hours. 00:19:13.210 --> 00:19:15.398 It turns out, when you actually ask cab drivers, 00:19:15.398 --> 00:19:16.690 they actually prefer busy days. 00:19:16.690 --> 00:19:18.640 They actually prefer it when there are customers as 00:19:18.640 --> 00:19:20.710 opposed to just driving around and looking for people. 00:19:20.710 --> 00:19:22.335 So we don't think that's actually true. 00:19:22.335 --> 00:19:25.750 But in principle, you would have to sort think about that. 00:19:25.750 --> 00:19:29.230 Now, there's a long literature starting with Camerer et al. 00:19:29.230 --> 00:19:31.570 On cab drivers. 00:19:31.570 --> 00:19:33.490 A lot of that essentially is pre-Uber, 00:19:33.490 --> 00:19:37.180 like collecting trip shifts from cab drivers. 00:19:37.180 --> 00:19:38.830 Now, there's lots of like Uber data 00:19:38.830 --> 00:19:41.303 that's essentially much more powerful in some ways. 00:19:41.303 --> 00:19:43.720 So there will be probably more papers using Uber and Lyft, 00:19:43.720 --> 00:19:46.720 et cetera, data on that. 00:19:46.720 --> 00:19:48.370 But what Camerer et al. did at the time 00:19:48.370 --> 00:19:51.790 was they essentially looked at typical cab 00:19:51.790 --> 00:19:54.670 drivers that rent their cab for 12 hour periods for a fixed 00:19:54.670 --> 00:19:55.540 fee. 00:19:55.540 --> 00:19:57.820 And within this 12 hour window, a driver 00:19:57.820 --> 00:19:59.260 can choose their hours freely. 00:19:59.260 --> 00:20:01.030 So you just get your cab that's not yours. 00:20:01.030 --> 00:20:01.960 You rent it for the day. 00:20:01.960 --> 00:20:04.043 And then you can essentially choose how many hours 00:20:04.043 --> 00:20:04.960 you want to work. 00:20:04.960 --> 00:20:08.230 And their wages, how much money they make in any given hour, 00:20:08.230 --> 00:20:11.226 varies by a lot. 00:20:11.226 --> 00:20:14.200 The weather varies, the subway breakdowns, conferences 00:20:14.200 --> 00:20:15.170 and so on and so forth. 00:20:15.170 --> 00:20:17.128 There's lots of variation in how much money you 00:20:17.128 --> 00:20:19.540 make on a given day. 00:20:19.540 --> 00:20:22.120 So then they have trip sheets that look at how long 00:20:22.120 --> 00:20:24.460 cab drivers work and their overall earnings. 00:20:24.460 --> 00:20:27.760 And so they can essentially back out the wages from each day 00:20:27.760 --> 00:20:30.880 and then look at like how much do people work on days 00:20:30.880 --> 00:20:35.230 when wages are high versus days when wages are low. 00:20:35.230 --> 00:20:37.280 And then they essentially find the basic finding. 00:20:37.280 --> 00:20:39.363 And that's a finding that's sort of been contested 00:20:39.363 --> 00:20:42.070 in the literature many people have found or confirmed 00:20:42.070 --> 00:20:42.850 subsequently. 00:20:42.850 --> 00:20:44.933 And others have not, but I think eventually people 00:20:44.933 --> 00:20:46.780 have sort of settled on this. 00:20:46.780 --> 00:20:49.360 Hours are negatively correlated with wages. 00:20:49.360 --> 00:20:52.810 So when wages in particular are unexpectedly high, 00:20:52.810 --> 00:20:55.600 cab drivers tend to work fewer hours. 00:20:55.600 --> 00:20:58.750 And again, this is not for permanent changes, 00:20:58.750 --> 00:21:00.160 but for transitory changes. 00:21:00.160 --> 00:21:03.250 Surprisingly, on a given day people get more money, 00:21:03.250 --> 00:21:07.390 then drivers work few hours. 00:21:07.390 --> 00:21:10.360 And that's very hard to explain for the neoclassical model. 00:21:10.360 --> 00:21:12.758 Because, essentially, you shouldn't care about 00:21:12.758 --> 00:21:14.800 whether you make the money on Monday, on Tuesday. 00:21:14.800 --> 00:21:17.500 As I said, you should care about how much money you make overall 00:21:17.500 --> 00:21:20.140 and how many hours you work. 00:21:20.140 --> 00:21:21.940 And so it's very hard to explain this. 00:21:21.940 --> 00:21:26.620 And sort of the explanation that Camerer et al. and others were 00:21:26.620 --> 00:21:28.750 testing or arguing is essentially 00:21:28.750 --> 00:21:31.552 this has to do with reference dependence. 00:21:31.552 --> 00:21:34.010 And so what's being evaluated in a reference dependent way? 00:21:34.010 --> 00:21:36.040 Or how do we think about this? 00:21:36.040 --> 00:21:36.540 Yeah. 00:21:36.540 --> 00:21:37.880 AUDIENCE: I have a question. 00:21:37.880 --> 00:21:38.810 FRANK SCHILBACH: Yeah. 00:21:38.810 --> 00:21:41.420 AUDIENCE: How did they rule out the possibility 00:21:41.420 --> 00:21:44.990 that maybe there is reverse causality maybe 00:21:44.990 --> 00:21:47.630 on a day where none of the cab drivers 00:21:47.630 --> 00:21:49.850 want to work that much because it's supply and demand 00:21:49.850 --> 00:21:50.725 that they just go on? 00:21:56.550 --> 00:21:59.640 FRANK SCHILBACH: So that's to say effort costs are high. 00:21:59.640 --> 00:22:02.040 So usually, it's to do with other drivers. 00:22:02.040 --> 00:22:03.772 So let me actually get through-- so let 00:22:03.772 --> 00:22:04.980 me defer this for one second. 00:22:04.980 --> 00:22:06.750 I have a slide on confounds. 00:22:06.750 --> 00:22:08.790 And then we can see the rest of your question 00:22:08.790 --> 00:22:10.510 and then get back to that. 00:22:10.510 --> 00:22:12.220 But that's a good question. 00:22:12.220 --> 00:22:17.292 So what is being evaluated in reference-dependent way? 00:22:17.292 --> 00:22:18.750 What are people looking at in terms 00:22:18.750 --> 00:22:23.610 of where's the reference point in the evaluation here? 00:22:23.610 --> 00:22:24.870 Yes. 00:22:24.870 --> 00:22:28.320 AUDIENCE: I guess, if you're a cab driver, you kind of say, 00:22:28.320 --> 00:22:30.510 oh, I want to make this much money today. 00:22:30.510 --> 00:22:32.700 And then you just kind of work until you 00:22:32.700 --> 00:22:35.250 feel like you've made enough, and then you just stop. 00:22:35.250 --> 00:22:36.040 FRANK SCHILBACH: Right, exactly. 00:22:36.040 --> 00:22:37.822 You have to pay your fixed fee for the day 00:22:37.822 --> 00:22:39.030 or for the month or whatever. 00:22:39.030 --> 00:22:41.783 But there's an implicit fixed fee for the day. 00:22:41.783 --> 00:22:43.950 So you kind of want to make at least that much money 00:22:43.950 --> 00:22:45.180 to make not a loss. 00:22:45.180 --> 00:22:47.820 You probably have some positive target in some way in saying, 00:22:47.820 --> 00:22:52.740 like, I want to make pay back for my fee plus 00:22:52.740 --> 00:22:55.500 you want to make some money for the day and minus 00:22:55.500 --> 00:22:56.820 sort of expenses. 00:22:56.820 --> 00:22:59.970 And once you reach that target, you are in the gain domain. 00:22:59.970 --> 00:23:02.940 Below that, you are in the loss domain, right? 00:23:02.940 --> 00:23:04.740 And so the daily income essentially-- 00:23:04.740 --> 00:23:08.190 and that's essentially sort of money after paying back 00:23:08.190 --> 00:23:09.660 the fee, but you could also get-- 00:23:09.660 --> 00:23:12.658 it's essentially what's being evaluated 00:23:12.658 --> 00:23:13.950 in the reference-dependent way. 00:23:13.950 --> 00:23:16.320 What's the reference point is some daily target 00:23:16.320 --> 00:23:17.340 that you have. 00:23:17.340 --> 00:23:20.160 Often, it's expectation and so on and sort of, 00:23:20.160 --> 00:23:22.708 essentially, how much you think you will make. 00:23:22.708 --> 00:23:24.750 And then what's the feature of the value function 00:23:24.750 --> 00:23:25.680 that explains the phenomenon? 00:23:25.680 --> 00:23:26.722 Well, it's loss aversion. 00:23:26.722 --> 00:23:29.580 If you're falling short of the target, essentially 00:23:29.580 --> 00:23:31.710 your marginal utility-- 00:23:31.710 --> 00:23:34.080 when you drive another hour or another trip, 00:23:34.080 --> 00:23:37.500 the marginal utility that you get, since the value function 00:23:37.500 --> 00:23:39.360 is very steep below the target-- 00:23:39.360 --> 00:23:40.470 is very high. 00:23:40.470 --> 00:23:42.510 Once you reach the target, it's very flat. 00:23:42.510 --> 00:23:45.555 And then essentially you tend to stop. 00:23:45.555 --> 00:23:46.430 Does that make sense? 00:23:46.430 --> 00:23:46.980 AUDIENCE: Yeah. 00:23:46.980 --> 00:23:48.120 FRANK SCHILBACH: OK, great. 00:23:48.120 --> 00:23:52.170 So the main takeaway is, so if drivers often 00:23:52.170 --> 00:23:54.630 stop at their daily income target, 00:23:54.630 --> 00:23:57.780 driver with a higher wage reach their targets faster. 00:23:57.780 --> 00:23:58.920 And they work fewer hours. 00:23:58.920 --> 00:24:04.690 Again, that's variation within drivers across days. 00:24:04.690 --> 00:24:07.530 And sort of there's lots of subsequent work and debate 00:24:07.530 --> 00:24:08.670 regarding this finding. 00:24:08.670 --> 00:24:10.330 The debate is still ongoing. 00:24:10.330 --> 00:24:13.230 For example, one recent paper looks at tips 00:24:13.230 --> 00:24:15.460 that drivers get unexpectedly. 00:24:15.460 --> 00:24:18.480 So sometimes drivers get large tips, sometimes get small tips. 00:24:18.480 --> 00:24:21.447 It depends a lot when in the day they receive the tip, 00:24:21.447 --> 00:24:23.280 if they receive it really early versus late, 00:24:23.280 --> 00:24:24.810 if they sort of get to their target. 00:24:24.810 --> 00:24:26.185 And essentially, the target seems 00:24:26.185 --> 00:24:27.300 to be adjusting over time. 00:24:27.300 --> 00:24:29.550 But overall-- and this is sort of getting a little bit 00:24:29.550 --> 00:24:30.725 at your question-- 00:24:33.230 --> 00:24:35.800 we think it's not sort of aggregate supply. 00:24:35.800 --> 00:24:37.470 But the debate is still ongoing. 00:24:37.470 --> 00:24:40.260 But overall, we sort of think that lots of labor supply 00:24:40.260 --> 00:24:43.230 decisions, when people have daily decisions of how 00:24:43.230 --> 00:24:45.750 many hours to work and their wages vary, 00:24:45.750 --> 00:24:49.800 depend on reference points and might sort of potentially 00:24:49.800 --> 00:24:51.780 be at least suboptimal. 00:24:51.780 --> 00:24:55.320 Or put differently, people could work fewer hours 00:24:55.320 --> 00:24:59.230 and try to sort of adjust their overall amounts. 00:24:59.230 --> 00:25:00.230 For a while, I did this. 00:25:00.230 --> 00:25:02.160 You can ask your Uber and Lyft drivers 00:25:02.160 --> 00:25:04.230 what they're doing and so on and see 00:25:04.230 --> 00:25:07.148 whether they're a reference-dependent driver, 00:25:07.148 --> 00:25:08.940 sort of where they have reference-dependent 00:25:08.940 --> 00:25:10.050 preferences. 00:25:10.050 --> 00:25:12.480 Now, what are sort of potential alternative hypotheses? 00:25:12.480 --> 00:25:15.780 One question, one issue, could be liquidity constraints. 00:25:15.780 --> 00:25:18.330 This is, for example, if you just don't have enough cash. 00:25:18.330 --> 00:25:20.755 If you have to pay back your fee for the day 00:25:20.755 --> 00:25:23.130 or for the next day, you might want to not work only very 00:25:23.130 --> 00:25:25.110 few hours on a given day. 00:25:25.110 --> 00:25:27.390 It turns out that drivers who own their own medallion 00:25:27.390 --> 00:25:29.430 exhibit the same patterns. 00:25:29.430 --> 00:25:30.940 There could be things like fatigue. 00:25:30.940 --> 00:25:33.600 Let's just say it's really tedious to work on certain days 00:25:33.600 --> 00:25:34.860 when wages are high. 00:25:34.860 --> 00:25:37.620 We don't think that's going on in part because drivers 00:25:37.620 --> 00:25:40.170 themselves, they say, it's actually easier 00:25:40.170 --> 00:25:41.890 to drive with more passengers. 00:25:41.890 --> 00:25:43.920 Again, recent papers, in fact, can also 00:25:43.920 --> 00:25:45.100 control for this and so on. 00:25:45.100 --> 00:25:47.462 So we don't think it's actually fatigue. 00:25:47.462 --> 00:25:49.170 And this is, I think, what you're saying. 00:25:49.170 --> 00:25:51.510 The last one is unobserved shocks, so some 00:25:51.510 --> 00:25:55.490 shocks that affect all drivers' labor supply at the same time. 00:25:55.490 --> 00:25:58.770 Example, there are some days in which all drivers get the flu. 00:25:58.770 --> 00:25:59.820 Fewer drivers will work. 00:25:59.820 --> 00:26:02.760 And those who will work work fewer hours. 00:26:02.760 --> 00:26:09.330 And those who work get higher wages. 00:26:09.330 --> 00:26:12.630 That's a little bit trickier for them to rule out. 00:26:12.630 --> 00:26:15.450 In part, there's other papers, other studies afterwards, 00:26:15.450 --> 00:26:17.310 that sort of try to get at this. 00:26:17.310 --> 00:26:23.040 Usually, yeah, for this specific data, that's hard to do. 00:26:23.040 --> 00:26:27.000 I think for the other data where people have essentially not 00:26:27.000 --> 00:26:28.542 just daily-- 00:26:28.542 --> 00:26:30.000 so this is essentially trip sheets. 00:26:30.000 --> 00:26:33.150 So what they're using mostly is daily wages overall. 00:26:33.150 --> 00:26:34.410 They don't even have the wage. 00:26:34.410 --> 00:26:40.207 They only have the overall earnings and then the hours. 00:26:40.207 --> 00:26:41.790 And then they sort of compute the wage 00:26:41.790 --> 00:26:44.220 dividing the two, which causes some other trouble. 00:26:44.220 --> 00:26:46.020 But once you have Uber data, once you 00:26:46.020 --> 00:26:48.180 have specific trips and particular also tips 00:26:48.180 --> 00:26:51.870 and so on, you can look at, on a given day when 00:26:51.870 --> 00:26:56.630 I get a high tip versus a low tip, 00:26:56.630 --> 00:26:58.380 you can predict, essentially, how much I'm 00:26:58.380 --> 00:26:59.588 going to earn on a given day. 00:26:59.588 --> 00:27:01.590 So suppose you predict that I'm going 00:27:01.590 --> 00:27:04.530 to earn $100 in a given day. 00:27:04.530 --> 00:27:07.110 Now, I have, say, $50 or $60. 00:27:07.110 --> 00:27:09.810 Now, I get a large trip and a tip and so on that 00:27:09.810 --> 00:27:11.050 gets me over that threshold. 00:27:11.050 --> 00:27:12.630 You can look at whether I stop. 00:27:12.630 --> 00:27:14.250 You can look at the exact same thing. 00:27:14.250 --> 00:27:17.530 When I have only $20 and I get a large tip, do I stop and so on? 00:27:17.530 --> 00:27:19.810 So you can essentially control for all of that 00:27:19.810 --> 00:27:22.110 and then do within driver comparisons, 00:27:22.110 --> 00:27:25.530 like for trips that happen to be large or small that you 00:27:25.530 --> 00:27:26.910 happen to get in a given hour. 00:27:26.910 --> 00:27:29.902 And then you can sort of deal with overall market conditions. 00:27:29.902 --> 00:27:31.360 I think even better, in the future, 00:27:31.360 --> 00:27:32.735 there will be sort of experiments 00:27:32.735 --> 00:27:37.230 and so on where you can look at when Uber and Lyft try 00:27:37.230 --> 00:27:39.930 to incentivize their workers and so just, 00:27:39.930 --> 00:27:44.012 say, essentially, sometimes pay people more and less randomly, 00:27:44.012 --> 00:27:45.720 sort of explicitly randomly, because they 00:27:45.720 --> 00:27:48.330 want to kind of learn about how to best incentivize 00:27:48.330 --> 00:27:49.195 their drivers. 00:27:49.195 --> 00:27:50.820 And then you can control for everything 00:27:50.820 --> 00:27:53.940 because it's explicitly random whether a certain driver gets 00:27:53.940 --> 00:27:58.800 high versus low wages or sort of trip fares. 00:27:58.800 --> 00:28:00.380 Yeah. 00:28:00.380 --> 00:28:03.030 OK. 00:28:03.030 --> 00:28:09.650 Any questions on the labor supply? 00:28:09.650 --> 00:28:11.457 Yes. 00:28:11.457 --> 00:28:13.790 AUDIENCE: I'm a bit confused about the unobserved shock. 00:28:13.790 --> 00:28:16.450 Because if you get a really large tip, 00:28:16.450 --> 00:28:20.205 that doesn't necessarily predict that the rest of the day you'd 00:28:20.205 --> 00:28:22.085 have higher wages if you kept working. 00:28:22.085 --> 00:28:23.793 [INAUDIBLE] you get this really high tip, 00:28:23.793 --> 00:28:25.960 and then you would be like, oh, today's a lucky day. 00:28:25.960 --> 00:28:26.960 I can stop early. 00:28:26.960 --> 00:28:28.460 But if I have worked more this day, 00:28:28.460 --> 00:28:31.492 I wouldn't necessarily be making continuously more than normal. 00:28:31.492 --> 00:28:32.450 FRANK SCHILBACH: Right. 00:28:32.450 --> 00:28:35.790 So in the unobserved, in the large tip example, 00:28:35.790 --> 00:28:37.310 the assumption is exactly-- 00:28:37.310 --> 00:28:38.660 or what they show in the paper. 00:28:38.660 --> 00:28:40.760 This is subsequent work, not this specific work. 00:28:40.760 --> 00:28:42.460 What they exactly show in this type of paper, 00:28:42.460 --> 00:28:43.793 these are sort of random events. 00:28:43.793 --> 00:28:45.770 In a sense, it's precisely not predictive 00:28:45.770 --> 00:28:48.290 of your future earnings. 00:28:48.290 --> 00:28:49.310 It's random. 00:28:49.310 --> 00:28:51.990 Now, one interpretation that you have is to say, 00:28:51.990 --> 00:28:54.380 well, it could be that you kind of have some expectations 00:28:54.380 --> 00:28:55.605 about your future earnings. 00:28:55.605 --> 00:28:57.230 They could be particularly high or low. 00:28:57.230 --> 00:29:00.180 It's like, I got my lucky day and so on and so forth. 00:29:00.180 --> 00:29:01.310 That is hard to rule out. 00:29:01.310 --> 00:29:04.190 In some sense, if you had rational expectations, 00:29:04.190 --> 00:29:06.310 that's hard to explain for the neoclassical model. 00:29:06.310 --> 00:29:08.390 What's harder to rule out is to say, you know, 00:29:08.390 --> 00:29:10.717 I now think I got my lucky draw for the day. 00:29:10.717 --> 00:29:11.300 And that's it. 00:29:11.300 --> 00:29:14.030 And I'm not going to get any lucky draw again. 00:29:14.030 --> 00:29:15.620 Let's just call it a day. 00:29:15.620 --> 00:29:17.160 That is hard to rule out. 00:29:17.160 --> 00:29:21.390 But what is hard to explain for the neoclassical model-- 00:29:21.390 --> 00:29:23.210 I think we can sort of reject-- is to say, 00:29:23.210 --> 00:29:25.040 if you just think this is a shock, which 00:29:25.040 --> 00:29:27.350 really in reality it is and you just happened 00:29:27.350 --> 00:29:32.180 to get a bunch of money on a given day just randomly, 00:29:32.180 --> 00:29:34.100 then the reference-dependent model can sort of 00:29:34.100 --> 00:29:35.480 explain why you stop early. 00:29:35.480 --> 00:29:38.960 Because it just gets you over the reference point. 00:29:38.960 --> 00:29:41.210 The neoclassical model should just say, 00:29:41.210 --> 00:29:43.057 depends on essentially how many hours 00:29:43.057 --> 00:29:44.390 you want to work on a given day. 00:29:44.390 --> 00:29:46.010 It's nothing to do with a target. 00:29:46.010 --> 00:29:47.270 Because essentially, again, it doesn't 00:29:47.270 --> 00:29:48.740 matter whether you make money on Monday, 00:29:48.740 --> 00:29:49.670 or Tuesday, or Wednesday. 00:29:49.670 --> 00:29:51.920 You should just care about the overall amount of money 00:29:51.920 --> 00:29:53.720 that you make overall. 00:29:53.720 --> 00:29:56.090 But that's a good question. 00:29:56.090 --> 00:29:56.870 OK. 00:29:56.870 --> 00:30:00.870 So now, a second paper is on the housing market. 00:30:00.870 --> 00:30:05.470 So what is the natural reference point for housing market? 00:30:05.470 --> 00:30:07.340 Or how do we think about housing prices? 00:30:07.340 --> 00:30:08.630 Or how do you think about-- 00:30:08.630 --> 00:30:10.700 I guess a few of you will own a house. 00:30:10.700 --> 00:30:13.400 But if you owned a house, how would 00:30:13.400 --> 00:30:15.950 you think about selling a house? 00:30:15.950 --> 00:30:19.070 What is an actual reference point? 00:30:19.070 --> 00:30:19.940 Yes? 00:30:19.940 --> 00:30:22.438 AUDIENCE: Whatever it was before or how [? changes ?] 00:30:22.438 --> 00:30:23.353 [INAUDIBLE]. 00:30:23.353 --> 00:30:24.770 FRANK SCHILBACH: Yeah, so exactly. 00:30:24.770 --> 00:30:27.080 So the previous purchase price, it 00:30:27.080 --> 00:30:29.630 turns out it's a very, very salient thing for owners. 00:30:29.630 --> 00:30:32.180 People really know how much they paid for their house. 00:30:32.180 --> 00:30:34.040 It's a huge expense in their life. 00:30:34.040 --> 00:30:37.245 They really remember it was like 300,00, 500,000, 00:30:37.245 --> 00:30:38.120 a million, et cetera. 00:30:38.120 --> 00:30:39.890 They know exactly how much they paid. 00:30:39.890 --> 00:30:43.640 And when you even ask people, the majority of people 00:30:43.640 --> 00:30:46.370 know exactly how much they paid for their home. 00:30:46.370 --> 00:30:48.140 Now, one claim is that loss aversion 00:30:48.140 --> 00:30:51.660 makes people unwilling to sell their houses at a loss. 00:30:51.660 --> 00:30:54.620 And so what they would then do is they ask essentially 00:30:54.620 --> 00:30:58.920 for higher prices, if a loss, relative to their purchase 00:30:58.920 --> 00:30:59.420 price. 00:30:59.420 --> 00:31:03.920 And let me just show you exactly what I mean by that. 00:31:03.920 --> 00:31:07.250 Genesove and Mayer have Boston condominium data 00:31:07.250 --> 00:31:11.720 from 1992 in 1997. 00:31:11.720 --> 00:31:14.660 Luckily for them, there's lots of variation or fluctuations 00:31:14.660 --> 00:31:16.285 in the housing market during that time. 00:31:16.285 --> 00:31:17.785 So the housing market went up a lot, 00:31:17.785 --> 00:31:19.730 and then it went down a lot and went up a lot. 00:31:19.730 --> 00:31:23.420 Now, suppose you have two sellers, A and B, who both want 00:31:23.420 --> 00:31:26.660 to sell their home in 1994. 00:31:26.660 --> 00:31:29.540 Now, what we can do is we can look at these two people. 00:31:29.540 --> 00:31:31.580 Suppose they have very similar houses 00:31:31.580 --> 00:31:34.220 based on observable characteristics and location 00:31:34.220 --> 00:31:35.280 and so on and so forth. 00:31:35.280 --> 00:31:37.880 So once they have really comparable houses, 00:31:37.880 --> 00:31:44.720 we can all look at seller A, purchased their home really 00:31:44.720 --> 00:31:45.975 early on, in 1989. 00:31:45.975 --> 00:31:48.350 That person happens to be quite unlucky because they just 00:31:48.350 --> 00:31:51.650 bought the house really at the peak of, I guess, 00:31:51.650 --> 00:31:53.150 the housing boom at the time. 00:31:53.150 --> 00:31:58.130 Or seller number B, who purchased in 1991, that person 00:31:58.130 --> 00:31:59.180 was relatively lucky. 00:31:59.180 --> 00:32:01.820 They bought essentially at a time 00:32:01.820 --> 00:32:05.510 when the housing market was relatively low and appreciated 00:32:05.510 --> 00:32:06.600 quite a bit. 00:32:06.600 --> 00:32:10.940 So now, we can look at these two people and ask the question, 00:32:10.940 --> 00:32:14.510 is seller A or seller B more likely to sell their house? 00:32:14.510 --> 00:32:17.720 And the hypothesis is that seller A will sort of 00:32:17.720 --> 00:32:19.100 view this as a loss. 00:32:19.100 --> 00:32:22.190 This seller will essentially just say I'm losing money here. 00:32:22.190 --> 00:32:23.390 I don't want to sell. 00:32:23.390 --> 00:32:24.800 And that seller might essentially 00:32:24.800 --> 00:32:26.990 ask for a higher listing price and wants 00:32:26.990 --> 00:32:29.240 make more money for this house because they don't want 00:32:29.240 --> 00:32:32.300 to make a loss on that sale. 00:32:32.300 --> 00:32:34.910 Well, seller B says, I'm actually gaining money anyway. 00:32:34.910 --> 00:32:37.287 So let's just sort of post whatever you think 00:32:37.287 --> 00:32:38.870 is actually the expected market price. 00:32:38.870 --> 00:32:41.420 I might be happy to sell at a lower price compared 00:32:41.420 --> 00:32:43.040 to seller A. 00:32:43.040 --> 00:32:43.550 OK. 00:32:43.550 --> 00:32:46.008 So one thing we can look at essentially are listing prices. 00:32:46.008 --> 00:32:48.050 How much do people want for the houses? 00:32:48.050 --> 00:32:51.630 The second thing they can look at is actual sales prices. 00:32:51.630 --> 00:32:54.170 Are you now selling this at a higher or lower price? 00:32:54.170 --> 00:32:56.270 And number three, we can look at how long 00:32:56.270 --> 00:32:57.930 is the house on the market. 00:32:57.930 --> 00:33:00.510 How long does it actually take to sell it? 00:33:00.510 --> 00:33:02.060 That's a quite costly thing to do 00:33:02.060 --> 00:33:04.143 to have your house on the market for quite a while 00:33:04.143 --> 00:33:07.160 because, essentially, often people then don't already 00:33:07.160 --> 00:33:08.090 move somewhere else. 00:33:08.090 --> 00:33:09.500 Or they can't really live in the house 00:33:09.500 --> 00:33:11.500 because they have to sort of show it and make it 00:33:11.500 --> 00:33:13.170 available for showings and so on. 00:33:13.170 --> 00:33:14.420 So it's a costly thing to do. 00:33:14.420 --> 00:33:16.670 You really don't want to have your house on the market 00:33:16.670 --> 00:33:18.860 for several months. 00:33:18.860 --> 00:33:20.480 But is sort of the broad idea clear 00:33:20.480 --> 00:33:21.605 of what we're trying to do? 00:33:25.440 --> 00:33:26.190 OK. 00:33:26.190 --> 00:33:29.833 So what predictions do we want to test? 00:33:29.833 --> 00:33:31.500 We want to test whether house owners are 00:33:31.500 --> 00:33:34.290 reluctant to sell their house when the current market price 00:33:34.290 --> 00:33:36.970 is below the purchase price. 00:33:36.970 --> 00:33:39.910 So the ideal specification is essentially the following. 00:33:39.910 --> 00:33:42.540 We look at the list price on the left-hand side. 00:33:42.540 --> 00:33:45.300 And then we are going to run a regression that looks at some 00:33:45.300 --> 00:33:48.270 constant-- that's just kind of time trends, et cetera-- 00:33:48.270 --> 00:33:51.270 plus a beta, which is the coefficient of interest. 00:33:51.270 --> 00:33:54.010 Or one coefficient of interest is of the actual market price. 00:33:54.010 --> 00:33:55.680 How much is the thing worth? 00:33:55.680 --> 00:33:59.610 And then delta on the loss is how much do you lose relative 00:33:59.610 --> 00:34:02.520 to your purchase price. 00:34:02.520 --> 00:34:05.087 Now, if people were not reference-dependent, 00:34:05.087 --> 00:34:05.920 what should we find? 00:34:05.920 --> 00:34:09.090 Or what would be the predictions? 00:34:09.090 --> 00:34:12.280 The neoclassical model, what should we find here? 00:34:12.280 --> 00:34:12.780 Yes. 00:34:12.780 --> 00:34:15.195 AUDIENCE: We should find that delta 0 [INAUDIBLE]?? 00:34:21.248 --> 00:34:22.290 FRANK SCHILBACH: Exactly. 00:34:22.290 --> 00:34:23.340 Delta should not matter. 00:34:23.340 --> 00:34:26.880 It shouldn't matter how much you lost or gained in that house. 00:34:26.880 --> 00:34:29.219 What should matter is what is the actual market value. 00:34:29.219 --> 00:34:32.250 You should essentially try to be willing to sell it not. 00:34:32.250 --> 00:34:34.949 Beta could be-- it doesn't have to be 1. 00:34:34.949 --> 00:34:37.813 It could be everybody pays above the actual market. 00:34:37.813 --> 00:34:39.480 There's some housing bubble or whatever. 00:34:39.480 --> 00:34:42.853 Beta could be 1.1 or whatever you might be. 00:34:42.853 --> 00:34:45.270 That depends on, essentially, sort of aggregate conditions 00:34:45.270 --> 00:34:46.170 and so on. 00:34:46.170 --> 00:34:48.300 But delta really should not matter. 00:34:48.300 --> 00:34:49.800 When I'm trying to sell you a house, 00:34:49.800 --> 00:34:52.350 you should ask, what's the actual market value? 00:34:52.350 --> 00:34:54.210 And I should, essentially, then based 00:34:54.210 --> 00:34:58.740 on that, put it on a market on that listing price. 00:34:58.740 --> 00:35:02.160 But what might be the case is that the loss actually matters. 00:35:02.160 --> 00:35:05.670 Now, one problem here is that the actual market value is not 00:35:05.670 --> 00:35:06.963 observable, right? 00:35:06.963 --> 00:35:08.880 So I don't actually know what the market value 00:35:08.880 --> 00:35:10.440 is because that's endogenous. 00:35:10.440 --> 00:35:12.410 That's part of the transaction. 00:35:12.410 --> 00:35:13.500 So what can I do instead? 00:35:13.500 --> 00:35:14.340 Or how do I do this? 00:35:21.620 --> 00:35:22.160 Yes. 00:35:22.160 --> 00:35:25.920 AUDIENCE: By asking people, would you take this trade? 00:35:25.920 --> 00:35:27.130 And then see what they say. 00:35:27.130 --> 00:35:27.890 [INAUDIBLE] 00:35:27.890 --> 00:35:28.190 FRANK SCHILBACH: Right. 00:35:28.190 --> 00:35:30.830 So you can look at actually the market outcome overall. 00:35:30.830 --> 00:35:32.690 You can look at who buys it and what's 00:35:32.690 --> 00:35:34.460 the actual purchase price. 00:35:34.460 --> 00:35:37.130 Now, that might also be endogenous to the listing 00:35:37.130 --> 00:35:37.760 price, right? 00:35:37.760 --> 00:35:40.093 So if you think you know that people who are loss averse 00:35:40.093 --> 00:35:43.280 are listing their houses too high relatively compared 00:35:43.280 --> 00:35:45.590 to what the market value actually is, 00:35:45.590 --> 00:35:47.910 it might actually be sold at a higher price. 00:35:47.910 --> 00:35:50.330 So that's hard to interpret. 00:35:50.330 --> 00:35:53.870 It could just be that, if you list it at a very high price 00:35:53.870 --> 00:35:56.870 and wait for a long time, you also sell it at a higher price. 00:35:56.870 --> 00:35:59.245 But it doesn't mean that that's actually worth that much. 00:35:59.245 --> 00:36:02.870 It just means that you happen to find a buyer who happens 00:36:02.870 --> 00:36:04.400 to be willing to pay a lot. 00:36:04.400 --> 00:36:07.248 And that's more likely when you wait for a long time, which 00:36:07.248 --> 00:36:09.290 is not necessarily optimal, because it's actually 00:36:09.290 --> 00:36:11.460 quite costly to do that. 00:36:11.460 --> 00:36:13.923 But you're saying something else which is asking them 00:36:13.923 --> 00:36:14.840 about what they think. 00:36:14.840 --> 00:36:16.850 But what are the characteristics that we could look at? 00:36:16.850 --> 00:36:18.380 Or what data could we look at? 00:36:22.740 --> 00:36:26.340 If you had Zillow data or data on essentially a bunch 00:36:26.340 --> 00:36:28.400 of these apps where you can look at houses, 00:36:28.400 --> 00:36:29.400 what data could you get? 00:36:29.400 --> 00:36:29.970 Yes. 00:36:29.970 --> 00:36:31.450 AUDIENCE: You'd look at the houses 00:36:31.450 --> 00:36:33.938 in the neighborhood, similar houses that sold recently. 00:36:33.938 --> 00:36:34.980 FRANK SCHILBACH: Exactly. 00:36:34.980 --> 00:36:36.485 What Redfin and Zillow, et cetera, 00:36:36.485 --> 00:36:37.860 do these days is essentially they 00:36:37.860 --> 00:36:41.610 have these algorithms that try to predict the actual market 00:36:41.610 --> 00:36:42.660 sales price. 00:36:42.660 --> 00:36:44.190 And what they tend to do essentially 00:36:44.190 --> 00:36:46.530 is look at, exactly as you say, surrounding 00:36:46.530 --> 00:36:50.747 houses that are similar in some characteristics. 00:36:50.747 --> 00:36:52.080 They look at the square footage. 00:36:52.080 --> 00:36:55.980 They look at the number of bathrooms and rooms in general. 00:36:55.980 --> 00:36:58.960 They look at sort of location and so on and so forth. 00:36:58.960 --> 00:37:03.330 And then you can predict how much-- 00:37:03.330 --> 00:37:05.760 and sort of they look at all the time trends and so on. 00:37:05.760 --> 00:37:07.350 How much is the actual market value? 00:37:07.350 --> 00:37:10.110 But essentially, fundamentally, it's a prediction exercise. 00:37:10.110 --> 00:37:13.330 You can try to predict what the actual value is. 00:37:13.330 --> 00:37:15.600 And that's exactly what Genesove and Mayer do. 00:37:15.600 --> 00:37:17.730 They do some more fancy things that 00:37:17.730 --> 00:37:21.850 try to get it other unobservable characteristics and so on. 00:37:21.850 --> 00:37:24.330 But the essence of this is exactly the prediction exercise 00:37:24.330 --> 00:37:27.527 where they say, let's just look at the characteristics 00:37:27.527 --> 00:37:28.110 of this house. 00:37:28.110 --> 00:37:31.290 Let's try to predict what the market value is and then 00:37:31.290 --> 00:37:32.948 look at the loss, which is essentially 00:37:32.948 --> 00:37:34.740 the difference between the previous selling 00:37:34.740 --> 00:37:36.810 price and the expected selling price 00:37:36.810 --> 00:37:39.690 truncated from 0 because, otherwise, it's a gain. 00:37:39.690 --> 00:37:41.670 And then we can look at does it really 00:37:41.670 --> 00:37:44.070 seem that, when people are in the loss domain, 00:37:44.070 --> 00:37:47.910 when their loss is positive, are they now selling their house 00:37:47.910 --> 00:37:51.320 or trying to sell their house at a higher price? 00:37:51.320 --> 00:37:52.170 Yes? 00:37:52.170 --> 00:37:53.920 AUDIENCE: How do you account for something 00:37:53.920 --> 00:37:56.250 like recent renovations since the last listing 00:37:56.250 --> 00:37:58.980 and relative to any other similar units 00:37:58.980 --> 00:38:03.422 or houses or [INAUDIBLE] [? nearby ?] [INAUDIBLE]?? 00:38:03.422 --> 00:38:04.380 FRANK SCHILBACH: Right. 00:38:04.380 --> 00:38:06.242 So that's tricky to do. 00:38:06.242 --> 00:38:08.200 And I don't think they have this in their data. 00:38:08.200 --> 00:38:09.540 So what you'd have to assume here, 00:38:09.540 --> 00:38:10.950 and that's perhaps reasonable, is 00:38:10.950 --> 00:38:13.470 to say that when you look at sort of this picture 00:38:13.470 --> 00:38:16.200 that seller A and seller B did not 00:38:16.200 --> 00:38:18.450 do differential renovation depending 00:38:18.450 --> 00:38:20.680 on when they bought the house. 00:38:20.680 --> 00:38:21.210 Right? 00:38:21.210 --> 00:38:24.210 So if you said, it's fine if people have done renovations. 00:38:24.210 --> 00:38:25.920 For example, this is what I was saying 00:38:25.920 --> 00:38:28.260 about the beta on the actual market value. 00:38:28.260 --> 00:38:30.900 If you systematically underestimate or overestimate 00:38:30.900 --> 00:38:33.720 how much people renovate and so on, or maybe the housing market 00:38:33.720 --> 00:38:37.140 is really hot or whatever, that's OK. 00:38:37.140 --> 00:38:42.580 The main issue is that can't be correlated with the loss here. 00:38:42.580 --> 00:38:45.600 So if it's the case that people who lost a bunch of money 00:38:45.600 --> 00:38:48.810 in terms of their housing prices sort of tanked, 00:38:48.810 --> 00:38:50.460 if those people have done more or less 00:38:50.460 --> 00:38:55.620 renovations, then you're in trouble with your estimates. 00:38:55.620 --> 00:38:56.370 Yeah. 00:38:56.370 --> 00:38:58.775 AUDIENCE: Is the expected selling price the estimation 00:38:58.775 --> 00:39:00.668 of actual [? work ?] value? 00:39:00.668 --> 00:39:01.710 FRANK SCHILBACH: Correct. 00:39:01.710 --> 00:39:02.280 AUDIENCE: OK. 00:39:02.280 --> 00:39:03.155 FRANK SCHILBACH: Yes. 00:39:05.160 --> 00:39:07.170 So now, what Genesove and Mayer find 00:39:07.170 --> 00:39:08.810 is-- and there's more detail to that. 00:39:08.810 --> 00:39:10.393 But essentially, the main finding is-- 00:39:10.393 --> 00:39:12.870 and that's a fairly solid one, there's a bunch of different 00:39:12.870 --> 00:39:14.610 specifications-- 00:39:14.610 --> 00:39:17.228 a 10% increase in a prospective loss. 00:39:17.228 --> 00:39:19.020 So if this loss coefficient, the difference 00:39:19.020 --> 00:39:23.670 between the expected selling price and the purchase price, 00:39:23.670 --> 00:39:26.730 if that's 10% higher, then essentially the list price 00:39:26.730 --> 00:39:31.770 is 2.5% to 3.5% higher. 00:39:31.770 --> 00:39:34.410 So people list the price, the house, at a higher price. 00:39:34.410 --> 00:39:35.880 If the house is at a loss compared 00:39:35.880 --> 00:39:38.220 to a similar house who's not at a loss 00:39:38.220 --> 00:39:41.400 or who's in the gain domain, these effects 00:39:41.400 --> 00:39:45.120 then translate into higher sales prices and a lower hazard 00:39:45.120 --> 00:39:46.140 rate of sale. 00:39:46.140 --> 00:39:49.480 That's to say people actually sell it at a higher price. 00:39:49.480 --> 00:39:52.650 So in fact, it's hard to say here what's optimal versus not. 00:39:52.650 --> 00:39:55.320 It could be that, overall, people are like, 00:39:55.320 --> 00:39:56.940 it's actually a good thing to do. 00:39:56.940 --> 00:39:59.370 That depends a lot on, essentially, your opportunity 00:39:59.370 --> 00:40:03.840 costs of money in terms how costly is it 00:40:03.840 --> 00:40:06.580 for you to keep the house on the market for a while. 00:40:06.580 --> 00:40:10.190 But essentially, people have a lower hazard right of sale. 00:40:10.190 --> 00:40:13.050 That's to say it just takes them a lot longer to sell the house. 00:40:13.050 --> 00:40:15.930 That tends to be very costly to do. 00:40:15.930 --> 00:40:19.163 But if you just otherwise would have your money in the bank 00:40:19.163 --> 00:40:20.580 and you have another place to stay 00:40:20.580 --> 00:40:24.820 and you don't really care, maybe then that's fine. 00:40:24.820 --> 00:40:28.620 But we surely have real effects in terms of people 00:40:28.620 --> 00:40:30.780 list their houses at higher prices. 00:40:30.780 --> 00:40:33.030 People sell them at somewhat higher prices. 00:40:33.030 --> 00:40:36.720 And people also keep them longer on the market. 00:40:36.720 --> 00:40:37.410 OK. 00:40:37.410 --> 00:40:38.280 Yeah. 00:40:38.280 --> 00:40:39.738 AUDIENCE: Is this the same analysis 00:40:39.738 --> 00:40:42.165 if it's gains on the value of the house? 00:40:44.748 --> 00:40:46.790 FRANK SCHILBACH: So essentially, what we're doing 00:40:46.790 --> 00:40:51.360 is we implicitly comparing losses versus gains, right? 00:40:51.360 --> 00:40:57.470 So the gains here would be in the actual market value 00:40:57.470 --> 00:40:58.900 already as it is. 00:41:02.358 --> 00:41:04.400 But essentially, what you're implicitly doing is, 00:41:04.400 --> 00:41:06.980 when comparing implicitly, this is what I was saying here. 00:41:06.980 --> 00:41:10.010 When you look at this picture, implicitly what we're doing 00:41:10.010 --> 00:41:13.700 is we look at people who are losing money compared to people 00:41:13.700 --> 00:41:15.080 who are gaining money. 00:41:15.080 --> 00:41:16.715 And explicitly or implicitly, we're 00:41:16.715 --> 00:41:18.350 asking, is the increase and the gain 00:41:18.350 --> 00:41:20.264 sort of predictive of that-- 00:41:20.264 --> 00:41:22.610 sorry, the increase in the losses predictive 00:41:22.610 --> 00:41:25.110 of your listing price? 00:41:25.110 --> 00:41:27.890 Now, it's hard to do this at the same time for gains 00:41:27.890 --> 00:41:30.770 because, essentially, an increase in the market price 00:41:30.770 --> 00:41:34.550 overall that's sort of collinear. 00:41:34.550 --> 00:41:37.120 In [INAUDIBLE],, essentially that's hard to separate. 00:41:37.120 --> 00:41:38.763 You could do the same analysis, and you 00:41:38.763 --> 00:41:40.430 wouldn't find that for the gains in part 00:41:40.430 --> 00:41:42.388 because essentially explicitly what we're doing 00:41:42.388 --> 00:41:46.080 is incurring losses to gains. 00:41:46.080 --> 00:41:46.580 OK. 00:41:48.960 --> 00:41:49.460 OK. 00:41:49.460 --> 00:41:51.210 So then there is another piece of evidence 00:41:51.210 --> 00:41:54.230 that sort of tells us perhaps this is not optimal. 00:41:54.230 --> 00:41:57.057 If you look at people who are owner-occupied compared 00:41:57.057 --> 00:41:58.640 to investors-- so there are people who 00:41:58.640 --> 00:41:59.840 essentially invest in houses. 00:41:59.840 --> 00:42:00.715 And they sell houses. 00:42:00.715 --> 00:42:02.930 And they sell lots of houses over time. 00:42:02.930 --> 00:42:06.410 Those people have much lower endowment effect, 00:42:06.410 --> 00:42:07.730 if you want, for houses. 00:42:07.730 --> 00:42:10.580 So they exhibit this behavior a lot less. 00:42:10.580 --> 00:42:12.290 But people who live in that house, who 00:42:12.290 --> 00:42:13.940 bought that house, for them it's mostly 00:42:13.940 --> 00:42:15.107 the only house they live in. 00:42:15.107 --> 00:42:16.760 It's the main purchase that they have. 00:42:16.760 --> 00:42:19.400 For them, it's essentially their house for which 00:42:19.400 --> 00:42:23.420 they don't want to make losses. 00:42:23.420 --> 00:42:25.340 For them, these effects are twice as large 00:42:25.340 --> 00:42:27.410 compared to the investors. 00:42:27.410 --> 00:42:30.200 And that's perhaps some evidence that this is not 00:42:30.200 --> 00:42:32.240 optimal behavior in a sense of sort 00:42:32.240 --> 00:42:33.830 of the professional investors. 00:42:33.830 --> 00:42:36.930 They presumably know pretty well how to price their houses. 00:42:36.930 --> 00:42:39.560 So if they do this behavior less, 00:42:39.560 --> 00:42:41.840 presumably that's a sign that there's 00:42:41.840 --> 00:42:46.160 some form of a mistake here going on. 00:42:46.160 --> 00:42:49.520 Second, there's some evidence-- and John List has some work 00:42:49.520 --> 00:42:50.780 on this overall-- 00:42:50.780 --> 00:42:53.360 to say that experience can mitigate 00:42:53.360 --> 00:42:54.980 reference dependent effects. 00:42:54.980 --> 00:42:57.200 Essentially, if you do a lot of trading, 00:42:57.200 --> 00:43:00.030 if you sell a lot of houses and so on, 00:43:00.030 --> 00:43:03.230 then you might still feel losses and gains, 00:43:03.230 --> 00:43:05.750 but you might sort of have a lot more experience with this. 00:43:05.750 --> 00:43:07.940 You kind of know that this is happening sometimes. 00:43:07.940 --> 00:43:10.250 And you might be less prone to these types of effects 00:43:10.250 --> 00:43:13.100 because you kind of know that you shouldn't be doing this. 00:43:13.100 --> 00:43:17.150 You shouldn't sort of have your feelings of losses and gains 00:43:17.150 --> 00:43:18.920 get in the way of making profits. 00:43:18.920 --> 00:43:20.337 So there are some people who would 00:43:20.337 --> 00:43:22.760 argue that this is very much consistent with markets 00:43:22.760 --> 00:43:24.430 over time or exposures to markets 00:43:24.430 --> 00:43:25.430 and several predictions. 00:43:25.430 --> 00:43:28.730 Experience makes some of these effects go away. 00:43:28.730 --> 00:43:30.830 And John List has some separate evidence 00:43:30.830 --> 00:43:33.530 on this on traders of cards and so on. 00:43:37.130 --> 00:43:37.703 OK. 00:43:37.703 --> 00:43:39.620 So now, next, we're going to talk a little bit 00:43:39.620 --> 00:43:42.310 about finance and stocks. 00:43:42.310 --> 00:43:44.540 So interestingly-- yeah. 00:43:44.540 --> 00:43:46.495 AUDIENCE: Sorry, on the previous study, 00:43:46.495 --> 00:43:48.746 why [? is it ?] said reference-dependent and not 00:43:48.746 --> 00:43:54.390 some informational effect, that I'm a home owner 00:43:54.390 --> 00:43:57.100 and I think that my house is worth this amount. 00:43:57.100 --> 00:43:59.210 And I just [INAUDIBLE] [? anchored ?] 00:43:59.210 --> 00:44:01.252 to believing that that's the amount. 00:44:01.252 --> 00:44:02.210 FRANK SCHILBACH: Right. 00:44:02.210 --> 00:44:07.860 So one thing you could say is that-- 00:44:07.860 --> 00:44:09.890 so what you always have to make the argument 00:44:09.890 --> 00:44:12.870 is people who are at losses compared to who are at gains. 00:44:12.870 --> 00:44:15.620 So you look at our investor A and B. 00:44:15.620 --> 00:44:20.540 They might have additional information on how much 00:44:20.540 --> 00:44:21.500 the house is worth. 00:44:21.500 --> 00:44:24.350 So it could be, for example, that seller A 00:44:24.350 --> 00:44:25.860 knows a lot about this house. 00:44:25.860 --> 00:44:29.510 It's very beautiful and so much light and this and that. 00:44:29.510 --> 00:44:31.623 And, therefore, they paid a lot. 00:44:31.623 --> 00:44:33.290 Therefore, they have private information 00:44:33.290 --> 00:44:34.560 that it's worth a lot. 00:44:34.560 --> 00:44:37.340 Therefore, they list it at a really high price. 00:44:37.340 --> 00:44:40.280 So there are some specification here that-- look at this. 00:44:40.280 --> 00:44:43.520 What you see is this is columns two, four, and six, 00:44:43.520 --> 00:44:46.760 which is the residual from the last sales price. 00:44:46.760 --> 00:44:47.720 What is that? 00:44:47.720 --> 00:44:50.540 That's essentially the difference between at the time 00:44:50.540 --> 00:44:52.758 when previously the house was sold, 00:44:52.758 --> 00:44:54.800 what was the prediction of the market price then, 00:44:54.800 --> 00:44:56.040 and how much was it sold. 00:44:56.040 --> 00:45:01.580 So it's kind of like, how much did you overpay, if you want, 00:45:01.580 --> 00:45:03.830 relative to what we expected at the time? 00:45:03.830 --> 00:45:06.230 Presumably, that's reflective and, again, 00:45:06.230 --> 00:45:07.950 using the same prediction method. 00:45:07.950 --> 00:45:08.900 So if you thought, you know, it's 00:45:08.900 --> 00:45:11.400 really beautiful and lots of windows and this and that, lots 00:45:11.400 --> 00:45:14.300 of light, and really quiet and so on, 00:45:14.300 --> 00:45:16.270 if you overpaid at the time, that 00:45:16.270 --> 00:45:18.500 should then sort of be predictive of the listing 00:45:18.500 --> 00:45:19.160 price. 00:45:19.160 --> 00:45:22.280 And sort of controlling for that then should make this go away. 00:45:22.280 --> 00:45:25.280 What you see, however, there's some of that perhaps going on. 00:45:25.280 --> 00:45:27.650 If you compare, for example, columns one and two, 00:45:27.650 --> 00:45:29.690 you see that the effect goes down a little bit. 00:45:29.690 --> 00:45:31.190 But it's still there quite a bit. 00:45:31.190 --> 00:45:34.280 This is why I was saying 25% to 35%. 00:45:34.280 --> 00:45:35.900 That is exactly right. 00:45:35.900 --> 00:45:36.800 That's a big concern. 00:45:36.800 --> 00:45:38.180 And there's a bunch of sort of robustness, 00:45:38.180 --> 00:45:40.070 et cetera, checks in this specific study. 00:45:40.070 --> 00:45:41.150 But that's exactly right. 00:45:41.150 --> 00:45:43.220 There could be sort of unobservable information 00:45:43.220 --> 00:45:46.670 that the owner might have about the house that is not 00:45:46.670 --> 00:45:48.540 in Zillow or in any sort of Redfin, 00:45:48.540 --> 00:45:51.410 et cetera, predictions that's available for the public. 00:45:51.410 --> 00:45:54.140 That's a great question, but I think it's, to the extent 00:45:54.140 --> 00:45:56.810 that that takes care of it, sort of the authors 00:45:56.810 --> 00:45:58.730 have thought about that. 00:45:58.730 --> 00:45:59.930 Yes. 00:45:59.930 --> 00:46:01.670 AUDIENCE: On the following slide, 00:46:01.670 --> 00:46:05.070 when you talk about the differences, 00:46:05.070 --> 00:46:07.160 how do you control for the selection bias 00:46:07.160 --> 00:46:11.930 about the people that may be the ones repeatedly selling houses 00:46:11.930 --> 00:46:14.090 exhibit this less and, therefore, stay 00:46:14.090 --> 00:46:17.460 in the market versus a change in those individuals' behavior? 00:46:20.152 --> 00:46:21.110 FRANK SCHILBACH: Right. 00:46:21.110 --> 00:46:22.110 That's a great question. 00:46:22.110 --> 00:46:24.740 So the question you're asking is essentially to say-- 00:46:24.740 --> 00:46:28.790 and it's, in fact, a great sort of segue 00:46:28.790 --> 00:46:30.830 into behavioral finance, which is to say, 00:46:30.830 --> 00:46:33.350 suppose there are some people who are really sophisticated. 00:46:33.350 --> 00:46:36.200 They don't have certain behavioral biases. 00:46:36.200 --> 00:46:37.790 Maybe they're not loss averse. 00:46:37.790 --> 00:46:39.830 That makes you a better investor, say. 00:46:39.830 --> 00:46:43.490 And, therefore, you stay in the market overall. 00:46:43.490 --> 00:46:45.770 I think, from this observation that I have here, 00:46:45.770 --> 00:46:53.810 I cannot tell you is it experience or is it selection. 00:46:53.810 --> 00:46:59.060 So the question kind of is, when people are investors, 00:46:59.060 --> 00:47:01.160 do the effects of reference-dependence of gains 00:47:01.160 --> 00:47:03.170 and losses go away over time? 00:47:03.170 --> 00:47:05.510 Essentially, maybe the first, second, third time 00:47:05.510 --> 00:47:08.690 they feel really a loss in terms of making a bad investment. 00:47:08.690 --> 00:47:11.900 But in house number 20, I'm just like, that's as usual. 00:47:11.900 --> 00:47:13.910 And I shouldn't sort of really care very much. 00:47:13.910 --> 00:47:15.440 Is it that this goes away? 00:47:15.440 --> 00:47:17.840 Or is it that the people who are particularly 00:47:17.840 --> 00:47:19.940 loss averse and sort of essentially 00:47:19.940 --> 00:47:21.710 engage in this type of behavior in terms 00:47:21.710 --> 00:47:26.510 of listing too high of a price for losses, 00:47:26.510 --> 00:47:29.090 these are sort of bad investor in the housing markets? 00:47:29.090 --> 00:47:32.070 And they sort of are essentially driven out of the market. 00:47:32.070 --> 00:47:34.070 So that specific setting I don't think we 00:47:34.070 --> 00:47:37.910 can necessarily account for that. 00:47:37.910 --> 00:47:41.120 I think in the studies by John List, 00:47:41.120 --> 00:47:44.900 it's very much people argue it's about experience. 00:47:44.900 --> 00:47:47.060 But again, there also some part could also 00:47:47.060 --> 00:47:51.200 be selection I think. 00:47:51.200 --> 00:47:56.520 So I think, in some sense, either way 00:47:56.520 --> 00:47:59.370 I think the evidence that the investors are doing 00:47:59.370 --> 00:48:01.440 this behavior less sort of tells us something 00:48:01.440 --> 00:48:04.920 about this is probably not at least financially 00:48:04.920 --> 00:48:08.460 optimal for you to engage in this type of behavior. 00:48:08.460 --> 00:48:09.840 It might be privately optimal. 00:48:09.840 --> 00:48:11.215 In some sense, if you really feel 00:48:11.215 --> 00:48:14.100 at selling your house at a loss, you 00:48:14.100 --> 00:48:16.710 should probably list it at a higher price 00:48:16.710 --> 00:48:19.050 because that sort of limits your losses. 00:48:19.050 --> 00:48:21.120 That's just what a utility function looks like. 00:48:21.120 --> 00:48:24.330 It's not necessarily suboptimal in the sense 00:48:24.330 --> 00:48:27.000 of how you feel afterwards. 00:48:27.000 --> 00:48:29.760 It might be suboptimal in terms of how much money you make 00:48:29.760 --> 00:48:32.610 or how much money you have eventually on how much 00:48:32.610 --> 00:48:34.950 you pay for keeping your house on the market 00:48:34.950 --> 00:48:38.380 and so on for an extended period of time. 00:48:38.380 --> 00:48:39.000 OK. 00:48:39.000 --> 00:48:41.250 So did that answer your question? 00:48:41.250 --> 00:48:42.690 Yeah, OK. 00:48:42.690 --> 00:48:44.760 So behavioral finance is an interesting field 00:48:44.760 --> 00:48:46.710 because, for quite a while, economists 00:48:46.710 --> 00:48:48.780 thought that sort of neoclassical assumptions 00:48:48.780 --> 00:48:51.930 are, in fact, most likely to hold in financial markets. 00:48:51.930 --> 00:48:54.450 And why is that? 00:48:54.450 --> 00:48:56.640 And some of this already I mentioned, but why 00:48:56.640 --> 00:49:01.160 are financial markets particular-- 00:49:01.160 --> 00:49:02.910 why might one think that financial markets 00:49:02.910 --> 00:49:04.450 are particularly efficient? 00:49:18.170 --> 00:49:18.740 Yes. 00:49:18.740 --> 00:49:20.990 AUDIENCE: Well, you might think that financial markets 00:49:20.990 --> 00:49:22.700 are very competitive. 00:49:22.700 --> 00:49:24.680 And so it's actually the ones who 00:49:24.680 --> 00:49:29.530 can get rid of their behavioral biases that benefit the most 00:49:29.530 --> 00:49:31.948 and stay within financial market. 00:49:31.948 --> 00:49:32.990 FRANK SCHILBACH: Exactly. 00:49:32.990 --> 00:49:36.680 So it's very much sort of the Chicago economics assumption 00:49:36.680 --> 00:49:41.400 is to say, so financial markets are extremely competitive. 00:49:41.400 --> 00:49:44.848 If I'm an investor who has various behavioral biases, 00:49:44.848 --> 00:49:46.640 presumably I'm going to lose some money one 00:49:46.640 --> 00:49:47.690 way or the other. 00:49:47.690 --> 00:49:49.940 Well, if markets are really competitive, 00:49:49.940 --> 00:49:52.580 in the long run I cannot stay in this market without sort 00:49:52.580 --> 00:49:55.820 of being driven out. 00:49:55.820 --> 00:49:57.390 So essentially, the market favors 00:49:57.390 --> 00:50:01.052 sort of results-oriented, rational, and selfish behavior. 00:50:01.052 --> 00:50:03.260 So people who are not rational and so on and so forth 00:50:03.260 --> 00:50:05.627 will be eliminated from the market eventually. 00:50:05.627 --> 00:50:06.710 There's two parts to that. 00:50:06.710 --> 00:50:08.300 That's true across firms. 00:50:08.300 --> 00:50:10.100 That's to say there are some firms 00:50:10.100 --> 00:50:11.330 sort of better than others. 00:50:11.330 --> 00:50:13.220 But also, within a firm, if I'm an investor 00:50:13.220 --> 00:50:15.890 and I'm sort of advising clients and the like-- and 00:50:15.890 --> 00:50:17.510 I'm sort of not very good at this. 00:50:17.510 --> 00:50:19.400 And essentially, I have certain people biases 00:50:19.400 --> 00:50:22.970 that are not optimal in terms of making money for people. 00:50:22.970 --> 00:50:24.530 Presumably, I will not be promoted. 00:50:24.530 --> 00:50:28.880 Presumably, I will be driven out of or fired from the company 00:50:28.880 --> 00:50:30.470 eventually. 00:50:30.470 --> 00:50:32.720 So surprisingly, in fact, finance 00:50:32.720 --> 00:50:34.940 became one of the most influential and most fruitful 00:50:34.940 --> 00:50:37.470 applications of the psychology in economics 00:50:37.470 --> 00:50:38.690 of behavioral economics. 00:50:38.690 --> 00:50:41.430 There's lots and lots of work in behavioral finance. 00:50:41.430 --> 00:50:44.030 The reason being perhaps not necessarily 00:50:44.030 --> 00:50:45.123 because people are-- 00:50:45.123 --> 00:50:46.790 so some people are presumably driven out 00:50:46.790 --> 00:50:48.860 of the market, but surely not everyone. 00:50:48.860 --> 00:50:51.530 But in particular, because there is a great data in finance. 00:50:51.530 --> 00:50:56.000 There's lots and lots of daily data in terms of things 00:50:56.000 --> 00:50:58.400 that you should be doing compared 00:50:58.400 --> 00:51:00.200 to what models would say. 00:51:00.200 --> 00:51:01.820 So it's a really great way of being 00:51:01.820 --> 00:51:04.220 able to test models or test essentially 00:51:04.220 --> 00:51:06.860 predictions of the neoclassical model or behavioral theories 00:51:06.860 --> 00:51:07.830 and so on. 00:51:07.830 --> 00:51:10.790 So that's why behavioral finance has been very influential 00:51:10.790 --> 00:51:12.920 because there's so much data available for people 00:51:12.920 --> 00:51:16.220 to, in fact, test theories. 00:51:16.220 --> 00:51:19.140 And then why is it that people are not entirely driven out? 00:51:19.140 --> 00:51:21.410 I think often the case is that, even if you're right, 00:51:21.410 --> 00:51:23.750 for example, even if you're right that you can predict-- 00:51:23.750 --> 00:51:28.050 and if you watch some movies on the financial crisis 00:51:28.050 --> 00:51:31.430 and so on, even if you're right about in the long run 00:51:31.430 --> 00:51:33.500 the market is going to tank, well, actually it's 00:51:33.500 --> 00:51:34.730 going to take a lot of money. 00:51:34.730 --> 00:51:37.100 And often, if everybody is wrong, 00:51:37.100 --> 00:51:39.520 prices will go up for quite a while. 00:51:39.520 --> 00:51:41.420 So it's not actually clear that, at least 00:51:41.420 --> 00:51:45.180 in the short, medium run, we'll be driven out of the market 00:51:45.180 --> 00:51:45.680 quickly. 00:51:45.680 --> 00:51:49.000 But that's sort of a separate topic. 00:51:49.000 --> 00:51:49.750 OK. 00:51:49.750 --> 00:51:52.968 So now, one reason why or one way 00:51:52.968 --> 00:51:54.760 in which reference-dependent behavior might 00:51:54.760 --> 00:51:57.250 be important in finance is people 00:51:57.250 --> 00:52:01.120 might be differentially likely to sell winners and hold 00:52:01.120 --> 00:52:03.910 on to losers, financial stocks. 00:52:03.910 --> 00:52:05.690 That was mentioned last time as well. 00:52:05.690 --> 00:52:09.160 So what Terry Odean did in 1997 is he had, in fact, 00:52:09.160 --> 00:52:12.160 brokerage accounts from the nationwide brokerage 00:52:12.160 --> 00:52:16.090 house, which had all trades and prices for, I guess, '87 00:52:16.090 --> 00:52:17.080 to '93. 00:52:17.080 --> 00:52:19.600 In some sense-- a little bit old fashioned in a sense of you 00:52:19.600 --> 00:52:22.150 shouldn't be an individual trading and so on. 00:52:22.150 --> 00:52:24.670 You should just hold the stock market or the S&P 500 00:52:24.670 --> 00:52:26.410 or some index funds and so on. 00:52:26.410 --> 00:52:28.750 Here, these are people who hold individual stocks 00:52:28.750 --> 00:52:31.870 and sell and buy them one by one. 00:52:31.870 --> 00:52:34.270 And so during each trading day, then what Odean can do 00:52:34.270 --> 00:52:36.910 is he can look at, evaluate, each stock in the portfolio 00:52:36.910 --> 00:52:41.137 and look at this is a loss relative to the purchase price. 00:52:41.137 --> 00:52:43.720 So you can look at, essentially, the portfolio and say there's 00:52:43.720 --> 00:52:46.210 losers and winners. 00:52:46.210 --> 00:52:48.230 He only has data on trading days. 00:52:48.230 --> 00:52:51.610 So you can look at are they losers and winners 00:52:51.610 --> 00:52:53.260 when they're being sold. 00:52:53.260 --> 00:52:58.210 And he can look at then at realized gains and realized 00:52:58.210 --> 00:52:58.880 losses. 00:52:58.880 --> 00:53:01.570 So if you sell a stock and it's essentially 00:53:01.570 --> 00:53:07.360 above the purchase price, he calls it a realized loss. 00:53:07.360 --> 00:53:09.490 Sorry, if it's a losing stock, it's sold. 00:53:09.490 --> 00:53:10.420 It's a realized loss. 00:53:10.420 --> 00:53:11.740 It's below the purchase price. 00:53:11.740 --> 00:53:13.210 If it's above the purchase price, 00:53:13.210 --> 00:53:15.430 it's going to be a realized gain. 00:53:15.430 --> 00:53:17.140 Now, one thing you could do is compare 00:53:17.140 --> 00:53:20.702 the number of realized losses to the number of realized gains. 00:53:20.702 --> 00:53:22.660 Does that work, or is that reference-dependent? 00:53:22.660 --> 00:53:24.130 So what did we learn from that? 00:53:40.770 --> 00:53:41.970 Yes. 00:53:41.970 --> 00:53:43.770 AUDIENCE: [INAUDIBLE] people [INAUDIBLE] 00:53:43.770 --> 00:53:46.200 people won't want to [INAUDIBLE] are losing stock 00:53:46.200 --> 00:53:48.867 because they're comparing to the price they bought [INAUDIBLE].. 00:53:48.867 --> 00:53:49.825 FRANK SCHILBACH: Right. 00:53:49.825 --> 00:53:51.482 So I could look at the realized gains 00:53:51.482 --> 00:53:53.940 and the number of realized gains and the number of realized 00:53:53.940 --> 00:53:54.870 losses. 00:53:54.870 --> 00:53:56.350 But what's the problem with that? 00:53:56.350 --> 00:53:58.747 That's exactly right, but what's the underlying, 00:53:58.747 --> 00:54:00.330 what's the problem with this approach? 00:54:05.802 --> 00:54:08.260 How does this depend on the stock market going up and down? 00:54:10.950 --> 00:54:11.590 Yes? 00:54:11.590 --> 00:54:15.034 AUDIENCE: You're not looking at the magnitude of those gains 00:54:15.034 --> 00:54:17.503 or losses? 00:54:17.503 --> 00:54:18.420 FRANK SCHILBACH: Yeah. 00:54:18.420 --> 00:54:20.040 So that's a separate issue. 00:54:20.040 --> 00:54:21.900 You could look at the magnitudes themselves. 00:54:21.900 --> 00:54:24.540 And you could look at, depending on where you are, 00:54:24.540 --> 00:54:25.980 how does that matter. 00:54:25.980 --> 00:54:28.980 But what about just a number of gains and losses? 00:54:28.980 --> 00:54:31.530 What if the stock market goes up a lot? 00:54:31.530 --> 00:54:34.081 What are people going to sell or [INAUDIBLE]?? 00:54:37.280 --> 00:54:38.010 Yeah. 00:54:38.010 --> 00:54:40.400 AUDIENCE: Well, one issue is that you don't know when 00:54:40.400 --> 00:54:43.100 or for how long something has been losing. 00:54:43.100 --> 00:54:45.725 So if it's been losing for long time 00:54:45.725 --> 00:54:48.420 or gaining for a long time, that might impact the [INAUDIBLE] 00:54:48.420 --> 00:54:50.057 hang onto it or sell it? 00:54:50.057 --> 00:54:51.640 FRANK SCHILBACH: So he does have that. 00:54:51.640 --> 00:54:54.010 I think I'm asking for something very basic, which 00:54:54.010 --> 00:54:56.020 is, if the stock market goes up a lot, 00:54:56.020 --> 00:54:57.220 you'll have lots of winners. 00:54:57.220 --> 00:54:59.220 So you're going to realize lots of-- if you just 00:54:59.220 --> 00:55:00.940 sell randomly, stocks, gains and losses, 00:55:00.940 --> 00:55:03.775 and you just don't care, you'll have much more realized gains 00:55:03.775 --> 00:55:06.400 compared to realized losses just because your stock market went 00:55:06.400 --> 00:55:07.240 up a lot. 00:55:07.240 --> 00:55:09.230 Similarly, if the stock market went down, 00:55:09.230 --> 00:55:12.040 you will find that people have way more realized losses 00:55:12.040 --> 00:55:13.487 compared to realize gains. 00:55:13.487 --> 00:55:15.820 And it looked like I'm really trying to sell the losers, 00:55:15.820 --> 00:55:17.695 but it's not I'm actually selling the losers. 00:55:17.695 --> 00:55:19.690 I just have a lot more losers. 00:55:19.690 --> 00:55:20.770 That's all I was asking. 00:55:20.770 --> 00:55:23.187 I think he has actually the information about the purchase 00:55:23.187 --> 00:55:23.740 prices. 00:55:23.740 --> 00:55:25.700 So what he then does is something very simple. 00:55:25.700 --> 00:55:27.730 It just looks at people's portfolios and says, 00:55:27.730 --> 00:55:29.470 how many losing stocks do you have? 00:55:29.470 --> 00:55:32.080 What's your propensity to sell the losing stocks, which 00:55:32.080 --> 00:55:36.220 is what you call the PLR, the Proportion of Losers Realized? 00:55:36.220 --> 00:55:38.230 The same he does for the PGR, which is 00:55:38.230 --> 00:55:39.903 a Proportion of Gains Realized. 00:55:39.903 --> 00:55:42.070 So he looks at each person when they sell something. 00:55:42.070 --> 00:55:44.070 They look at how many losing stocks do you have. 00:55:44.070 --> 00:55:45.670 How many winning stocks do you have? 00:55:45.670 --> 00:55:47.770 And then he looks at what's the probability of you 00:55:47.770 --> 00:55:52.340 selling any of those depending on they're winners or losers. 00:55:52.340 --> 00:55:54.040 So what the main finding then here 00:55:54.040 --> 00:55:57.490 is that the PGR, the Proportion of Gains Realized, 00:55:57.490 --> 00:55:59.920 is larger than the PLR. 00:55:59.920 --> 00:56:02.410 And that's to say that's what they call the disposition 00:56:02.410 --> 00:56:05.200 effect, which is a tendency to sell winners and hold on 00:56:05.200 --> 00:56:07.040 to losers. 00:56:07.040 --> 00:56:08.760 Does that make sense? 00:56:08.760 --> 00:56:09.300 OK. 00:56:09.300 --> 00:56:10.773 And so why do we care about that? 00:56:10.773 --> 00:56:11.940 Or why is this actually bad? 00:56:11.940 --> 00:56:14.465 Is it's necessarily suboptimal behavior? 00:56:14.465 --> 00:56:15.090 Why do we care? 00:56:26.870 --> 00:56:27.422 Yes. 00:56:27.422 --> 00:56:28.880 AUDIENCE: I think you had mentioned 00:56:28.880 --> 00:56:31.090 in a previous class [INAUDIBLE] sometimes it 00:56:31.090 --> 00:56:35.300 might be worthwhile to hold on to the winners 00:56:35.300 --> 00:56:39.236 or someone who's betting on [INAUDIBLE].. 00:56:39.236 --> 00:56:42.070 But probably shouldn't be an overall bias 00:56:42.070 --> 00:56:44.990 towards selling losers. 00:56:44.990 --> 00:56:47.700 And then probably [INAUDIBLE] overall effective strategy 00:56:47.700 --> 00:56:48.582 [INAUDIBLE]. 00:56:48.582 --> 00:56:49.540 FRANK SCHILBACH: Right. 00:56:49.540 --> 00:56:51.310 So it depends on, essentially, how well 00:56:51.310 --> 00:56:53.320 the winners and the losers are going to do. 00:56:53.320 --> 00:56:56.150 It turns out, so in principle, you would say, 00:56:56.150 --> 00:57:00.220 well, winners and losers should have the same expected 00:57:00.220 --> 00:57:03.280 return regardless of the winners and losers in the past. 00:57:03.280 --> 00:57:05.440 That's essentially the efficient market hypothesis, 00:57:05.440 --> 00:57:08.620 just to say past price changes should just not 00:57:08.620 --> 00:57:13.867 be predictive of future momentum or price changes overall. 00:57:13.867 --> 00:57:16.450 Because the current price should have incorporated essentially 00:57:16.450 --> 00:57:18.742 all information that's available at this point in time. 00:57:18.742 --> 00:57:21.325 So to that degree, it shouldn't matter actually what you sell. 00:57:21.325 --> 00:57:23.048 You could just randomly sell stuff. 00:57:23.048 --> 00:57:24.340 Now, you should not sell a lot. 00:57:24.340 --> 00:57:26.450 Usually, there's commissions involved in these trades. 00:57:26.450 --> 00:57:27.867 So essentially, you shouldn't sell 00:57:27.867 --> 00:57:30.783 anything that's essentially leading to over-trading. 00:57:30.783 --> 00:57:32.200 Odean has another paper that shows 00:57:32.200 --> 00:57:33.700 essentially, in particular, men tend 00:57:33.700 --> 00:57:36.280 to be overconfident in how good they are at trading. 00:57:36.280 --> 00:57:39.430 And they tend to over-trade, and that's really costly. 00:57:39.430 --> 00:57:44.380 It turns out that, in their specific period, 00:57:44.380 --> 00:57:45.850 in fact there's momentum. 00:57:45.850 --> 00:57:48.513 And that used to be the case quite a bit in that period 00:57:48.513 --> 00:57:50.680 of time, which essentially has to do with the winner 00:57:50.680 --> 00:57:52.480 is actually doing better than the losers 00:57:52.480 --> 00:57:54.430 by quite a big margin. 00:57:54.430 --> 00:57:57.230 That is to say, you should have actually did exactly 00:57:57.230 --> 00:57:57.980 then the opposite. 00:57:57.980 --> 00:57:59.813 If anything, you should have sold the losers 00:57:59.813 --> 00:58:02.650 and keep the winners because they are, in fact, making 00:58:02.650 --> 00:58:08.890 more money in the short and medium run in that period, OK? 00:58:08.890 --> 00:58:11.110 There's also the investors sell more losers 00:58:11.110 --> 00:58:12.277 and winners in December. 00:58:12.277 --> 00:58:13.360 This is what you see here. 00:58:13.360 --> 00:58:14.020 Why is that? 00:58:18.040 --> 00:58:18.540 Yes. 00:58:18.540 --> 00:58:19.530 AUDIENCE: Is it that you can count 00:58:19.530 --> 00:58:20.520 your losses toward your income? 00:58:20.520 --> 00:58:22.603 FRANK SCHILBACH: Exactly, this is for tax reasons. 00:58:22.603 --> 00:58:24.150 So that actually happens to-- 00:58:24.150 --> 00:58:26.310 Jim Poterba, who's in the economics department, 00:58:26.310 --> 00:58:28.020 actually did a paper on this. 00:58:28.020 --> 00:58:30.380 So there you can essentially realize losses, 00:58:30.380 --> 00:58:35.560 and that reduces your taxes overall. 00:58:35.560 --> 00:58:36.060 Exactly. 00:58:36.060 --> 00:58:37.768 But overall, essentially what's happening 00:58:37.768 --> 00:58:40.620 is that people do seem to engage in this behavior. 00:58:40.620 --> 00:58:42.150 In a pretty striking fashion, they 00:58:42.150 --> 00:58:46.500 seem to be losing quite a bit of money from that. 00:58:46.500 --> 00:58:47.190 OK. 00:58:47.190 --> 00:58:50.647 Let me mention at least the marathon running and perhaps 00:58:50.647 --> 00:58:52.230 the golf example, and then we're going 00:58:52.230 --> 00:58:55.390 to move towards prices and firms. 00:58:55.390 --> 00:58:57.045 How do firms react to these biases? 00:58:57.045 --> 00:58:58.920 But what I'm trying to do here is essentially 00:58:58.920 --> 00:59:00.660 show you a bunch of different settings. 00:59:00.660 --> 00:59:02.430 And essentially, if you look at different settings 00:59:02.430 --> 00:59:04.770 in the world, there's lots and lots of different settings 00:59:04.770 --> 00:59:06.510 where reference-dependence seems to matter. 00:59:06.510 --> 00:59:08.885 One way or the other, it seems to be important in shaping 00:59:08.885 --> 00:59:10.180 people's behavior. 00:59:10.180 --> 00:59:13.740 So this is a very nice paper about a marathon running 00:59:13.740 --> 00:59:14.610 finishing times. 00:59:14.610 --> 00:59:15.902 These are many, many marathons. 00:59:15.902 --> 00:59:18.970 They have lots of data on finishing times for people. 00:59:18.970 --> 00:59:21.030 So the law of large numbers would 00:59:21.030 --> 00:59:23.790 predict that, if you look at essentially people's finishing 00:59:23.790 --> 00:59:27.510 times, if people have different talents and so on and so forth, 00:59:27.510 --> 00:59:30.040 the finishing times should look something like this. 00:59:30.040 --> 00:59:30.540 OK. 00:59:30.540 --> 00:59:31.470 Some people are faster. 00:59:31.470 --> 00:59:32.428 Some people are slower. 00:59:32.428 --> 00:59:36.668 But overall they're should be some smooth distribution that 00:59:36.668 --> 00:59:39.210 essentially is like log normal or whatever you want it to be. 00:59:39.210 --> 00:59:42.070 But essentially, it should look something like this. 00:59:42.070 --> 00:59:43.950 So why might it not look like this? 00:59:43.950 --> 00:59:48.260 Or what might people do instead? 00:59:48.260 --> 00:59:49.100 Yes. 00:59:49.100 --> 00:59:52.080 AUDIENCE: They may say, I'll beat 4:30. 00:59:52.080 --> 00:59:54.770 And then you might see bunching at certain points, 00:59:54.770 --> 00:59:58.312 like 4 hours, 4:30, 5 hours, 5:30. 00:59:58.312 --> 00:59:59.270 FRANK SCHILBACH: Right. 00:59:59.270 --> 01:00:01.430 So exactly as you say, it might be 01:00:01.430 --> 01:00:03.140 that people have reference points 01:00:03.140 --> 01:00:05.420 not in terms of status quo here and the like. 01:00:05.420 --> 01:00:07.640 Reference points might be goals or aspirations. 01:00:07.640 --> 01:00:10.460 You might say, I really want to run the marathon in 4 hours 01:00:10.460 --> 01:00:13.380 or 4:30 or 4 hours if you want. 01:00:13.380 --> 01:00:16.130 And then what the marathon times actually look like 01:00:16.130 --> 01:00:17.295 is something like this. 01:00:17.295 --> 01:00:18.920 And in particular, it seems like people 01:00:18.920 --> 01:00:22.800 have lots of goals of reaching something like-- 01:00:22.800 --> 01:00:24.890 if you look at the distribution, what you see 01:00:24.890 --> 01:00:27.380 is exactly as you predict. 01:00:27.380 --> 01:00:29.840 If you look at the half hour or even 01:00:29.840 --> 01:00:33.020 the quarter hour sort of points, there's 01:00:33.020 --> 01:00:35.070 essentially bunching from below. 01:00:35.070 --> 01:00:37.220 So essentially, people seem to be, 01:00:37.220 --> 01:00:41.540 if they're at pace to finish at 4 hours and 1 minute, 01:00:41.540 --> 01:00:44.027 they try to sort of speed up and just make it to 3:59. 01:00:48.800 --> 01:00:52.100 So you see a bunch of bunching at the half hour slots. 01:00:52.100 --> 01:00:54.558 You see actually much less at 6 hours and 30 minutes 01:00:54.558 --> 01:00:55.100 or something. 01:00:55.100 --> 01:00:56.975 It seems that few people have actually goals. 01:00:56.975 --> 01:00:59.413 Once you run the marathon in 6 hours and 30 minutes, 01:00:59.413 --> 01:01:00.830 which is probably what I would do, 01:01:00.830 --> 01:01:02.872 you know, it doesn't really matter whether you're 01:01:02.872 --> 01:01:04.825 like 6:31 or 6:29. 01:01:04.825 --> 01:01:06.200 But there's very ambitious people 01:01:06.200 --> 01:01:10.280 who want to like finish in 4:30, 4 hours, or 3:30 and the like. 01:01:10.280 --> 01:01:15.030 And there's a bunch of bunching from below there. 01:01:15.030 --> 01:01:18.710 So you see the same for quarter hour times, a little bit less 01:01:18.710 --> 01:01:21.110 of that. 01:01:21.110 --> 01:01:23.360 And sort of that's consistent with the reference point 01:01:23.360 --> 01:01:25.733 here being a goal and aspiration. 01:01:25.733 --> 01:01:26.900 You want to reach 4 minutes. 01:01:26.900 --> 01:01:28.460 You want to brag to your friends and so on. 01:01:28.460 --> 01:01:30.050 And it's not so great if you do that and say, 01:01:30.050 --> 01:01:32.570 I finished in 4 hours and 1 minute as opposed to if you 01:01:32.570 --> 01:01:35.690 say I finished below 4 hours. 01:01:35.690 --> 01:01:40.130 Now, when you look at the effort at the end of the race, what 01:01:40.130 --> 01:01:40.880 would you expect? 01:01:40.880 --> 01:01:43.910 So what we have here on the x-axis is people. 01:01:43.910 --> 01:01:46.820 These are the 40 kilometer pace. 01:01:46.820 --> 01:01:50.010 These are in 30 second increments. 01:01:50.010 --> 01:01:57.860 So the marathon is 42.195 kilometers. 01:01:57.860 --> 01:02:00.770 So what I'm showing you here is, essentially, 01:02:00.770 --> 01:02:04.760 the 40 kilometer pace people are ranked or distributed 01:02:04.760 --> 01:02:08.310 by the 40 kilometer pace, the first 40 kilometers. 01:02:08.310 --> 01:02:10.160 Now, what you would expect for people-- 01:02:10.160 --> 01:02:12.740 so there are some people who were at pace 01:02:12.740 --> 01:02:17.715 to reach 3:55 and some people at pace to reach 4:05 and so on. 01:02:17.715 --> 01:02:19.340 What is it that you would expect people 01:02:19.340 --> 01:02:23.810 to do when you look at how fast people run towards the last two 01:02:23.810 --> 01:02:24.920 kilometers of the race? 01:02:31.380 --> 01:02:31.880 Yes? 01:02:31.880 --> 01:02:33.297 AUDIENCE: If they're really close, 01:02:33.297 --> 01:02:35.938 they're going [INAUDIBLE] especially hard. 01:02:35.938 --> 01:02:36.980 FRANK SCHILBACH: Exactly. 01:02:36.980 --> 01:02:38.022 And this is what you see. 01:02:38.022 --> 01:02:39.670 So low means you're running fast. 01:02:39.670 --> 01:02:42.740 This is minutes per kilometer I think 01:02:42.740 --> 01:02:44.570 or relative minutes per kilometer. 01:02:44.570 --> 01:02:46.460 So what you should expect is people 01:02:46.460 --> 01:02:50.330 who are just below the goal or people who are essentially just 01:02:50.330 --> 01:02:53.480 above the goal, in fact, these are the people who speed up, 01:02:53.480 --> 01:02:55.040 OK? 01:02:55.040 --> 01:02:56.540 And this is exactly what you sort of 01:02:56.540 --> 01:03:00.180 see is that people who are just below the 4 minute mark 01:03:00.180 --> 01:03:03.020 or some people who are just above, they speed up 01:03:03.020 --> 01:03:04.820 to just make it to that goal. 01:03:04.820 --> 01:03:06.320 And sort of Allen et al. 01:03:06.320 --> 01:03:08.270 Have a sort of analysis of this. 01:03:08.270 --> 01:03:12.050 Essentially, what they find is that everybody 01:03:12.050 --> 01:03:14.090 gets sort of slower towards the end of the race. 01:03:14.090 --> 01:03:16.280 But if you're sort of in reach of reaching 01:03:16.280 --> 01:03:20.180 the goal by reaching the time of 4 hours, 01:03:20.180 --> 01:03:23.750 you're going to slow down less or speed up a bit to just reach 01:03:23.750 --> 01:03:25.620 that specific goal, OK? 01:03:28.400 --> 01:03:32.120 Let me show you one more thing of sports, which is golf. 01:03:32.120 --> 01:03:33.410 So how does golf work? 01:03:33.410 --> 01:03:37.520 In case you don't know, you hit a ball with a club from a tee 01:03:37.520 --> 01:03:38.960 into a hole. 01:03:38.960 --> 01:03:43.610 The way this works is there's usually 4 rounds of 18 holes. 01:03:43.610 --> 01:03:48.830 There is very convex incentives in the golf tournament. 01:03:48.830 --> 01:03:50.600 So you get a bunch of money if you win, 01:03:50.600 --> 01:03:52.010 if you're sort of at the top. 01:03:52.010 --> 01:03:54.165 For an average performance, essentially you 01:03:54.165 --> 01:03:55.290 don't make that much money. 01:03:55.290 --> 01:03:58.430 I mean, you make good money, but the prize money 01:03:58.430 --> 01:04:02.010 is really in terms of when you do really well. 01:04:02.010 --> 01:04:03.230 So now, what is par? 01:04:03.230 --> 01:04:08.110 Par is how many strokes, many shots, do you need to-- 01:04:08.110 --> 01:04:10.280 how many shots a very good golfer should require 01:04:10.280 --> 01:04:13.880 to complete a given hole. 01:04:13.880 --> 01:04:17.840 So par is usually 3, 4, or 5 shots. 01:04:17.840 --> 01:04:20.280 And then eagle is 2 below par. 01:04:20.280 --> 01:04:22.700 So if you have a par 4 hole, if you do 2 shots, 01:04:22.700 --> 01:04:23.660 that's an eagle. 01:04:23.660 --> 01:04:26.120 If you do 3 shots in that case, that would be a birdie. 01:04:26.120 --> 01:04:27.110 4 would be par. 01:04:27.110 --> 01:04:29.030 Bogey would be 1 above par. 01:04:29.030 --> 01:04:32.900 And double bogey would be 2 above par, OK? 01:04:32.900 --> 01:04:36.380 So knowing all that, so what matters for golf 01:04:36.380 --> 01:04:38.600 at the end of a tournament is how many shots do you 01:04:38.600 --> 01:04:40.065 make in total. 01:04:40.065 --> 01:04:42.440 So how can we now look at reference-dependent preferences 01:04:42.440 --> 01:04:44.870 here in this setting? 01:05:02.090 --> 01:05:02.750 Yes. 01:05:02.750 --> 01:05:04.980 AUDIENCE: The reference is the par [INAUDIBLE].. 01:05:04.980 --> 01:05:05.855 FRANK SCHILBACH: Yes. 01:05:05.855 --> 01:05:06.980 The reference is the par. 01:05:06.980 --> 01:05:11.090 Now, suppose you are putting, which is at the end, 01:05:11.090 --> 01:05:12.398 you know, on the green. 01:05:12.398 --> 01:05:13.440 What are you going to do? 01:05:13.440 --> 01:05:15.005 What kinds of behaviors do we expect? 01:05:17.920 --> 01:05:18.765 Yes. 01:05:18.765 --> 01:05:19.890 AUDIENCE: Well, it depends. 01:05:19.890 --> 01:05:22.870 So if you're under par, then you might 01:05:22.870 --> 01:05:24.530 try-- like, you're shooting for birdie, 01:05:24.530 --> 01:05:26.530 or you're shooting for par and it's a long putt, 01:05:26.530 --> 01:05:29.290 you might try to make sure that you get it to the hole. 01:05:29.290 --> 01:05:32.740 Whereas, if you are already over par or double over par, 01:05:32.740 --> 01:05:34.650 you might try to play a little safer, 01:05:34.650 --> 01:05:35.992 make sure you're not way over. 01:05:35.992 --> 01:05:36.950 FRANK SCHILBACH: Right. 01:05:36.950 --> 01:05:39.490 So some of this is about risk preferences, 01:05:39.490 --> 01:05:41.673 how risky your shots are. 01:05:41.673 --> 01:05:43.090 Another way to think about this is 01:05:43.090 --> 01:05:45.130 kind of how much effort do you put in your shot. 01:05:45.130 --> 01:05:49.750 In some sense, to the extent that you can sort of allocate 01:05:49.750 --> 01:05:52.210 attention or really focus an effort, 01:05:52.210 --> 01:05:54.580 maybe that's sort of limited over the course of 18 holes 01:05:54.580 --> 01:05:59.830 and 4 rounds of each of those. 01:05:59.830 --> 01:06:03.250 You might sort of try particularly hard to do well 01:06:03.250 --> 01:06:06.190 on shots that make you reach par compared 01:06:06.190 --> 01:06:10.100 to shots that might get you a birdie or even better. 01:06:10.100 --> 01:06:12.610 And so this is exactly as you say. 01:06:12.610 --> 01:06:15.130 The fairly obvious reference point for each hole in golf 01:06:15.130 --> 01:06:16.570 is reach par. 01:06:16.570 --> 01:06:19.510 Importantly, it doesn't matter whether you 01:06:19.510 --> 01:06:23.020 have birdie, par, and bogie, versus par, par, par. 01:06:23.020 --> 01:06:26.050 Essentially, that gives you exactly the same amount 01:06:26.050 --> 01:06:27.400 of shots overall. 01:06:27.400 --> 01:06:32.110 But now what Pope and Schweitzer ask is the question, 01:06:32.110 --> 01:06:35.757 are putters more likely to make their par 01:06:35.757 --> 01:06:37.090 than their birdie points, right? 01:06:37.090 --> 01:06:42.340 So essentially, depending on are you at possibility of losing 01:06:42.340 --> 01:06:45.130 or at gaining, essentially if you're 01:06:45.130 --> 01:06:48.730 worried about losing par, are you going to behave differently 01:06:48.730 --> 01:06:51.520 compared to when you can make a birdie, which is in the gain 01:06:51.520 --> 01:06:52.780 domain potentially? 01:06:52.780 --> 01:06:55.240 The same you could say about bogies, where you're already 01:06:55.240 --> 01:06:57.615 in the loss domain because you're already doing terribly. 01:06:57.615 --> 01:07:00.310 And you're trying to avoid a double bogey. 01:07:00.310 --> 01:07:01.810 So now, what they find, essentially, 01:07:01.810 --> 01:07:04.830 is the par putts are much more likely to be made. 01:07:04.830 --> 01:07:06.910 There's 2 or 3 percentage points more likely 01:07:06.910 --> 01:07:09.040 to be made compared to equivalent birdie putts. 01:07:09.040 --> 01:07:11.380 The authors rule out a bunch of different explanation. 01:07:11.380 --> 01:07:13.780 It doesn't have to do with the heterogeneity of player 01:07:13.780 --> 01:07:14.680 ability. 01:07:14.680 --> 01:07:16.737 They even have sort of like GPS information 01:07:16.737 --> 01:07:18.820 of exactly where the ball is compared to the hole. 01:07:18.820 --> 01:07:20.278 And they do all sorts of comparison 01:07:20.278 --> 01:07:21.348 of how hard the shot is. 01:07:21.348 --> 01:07:23.140 They don't think it has to do with learning 01:07:23.140 --> 01:07:24.112 from earlier putts. 01:07:24.112 --> 01:07:25.570 And it seems to also not have to do 01:07:25.570 --> 01:07:26.910 with hole specific preferences. 01:07:26.910 --> 01:07:28.660 Some holes are really hard, and some holes 01:07:28.660 --> 01:07:31.520 are really simple and easy and so on. 01:07:31.520 --> 01:07:34.790 So it doesn't seem to do with any of that. 01:07:34.790 --> 01:07:37.690 So again, that's sort of another instance of reference-dependent 01:07:37.690 --> 01:07:40.850 that seems to be in quite a few settings. 01:07:40.850 --> 01:07:43.450 Let me sort of skip the Deal or No Deal TV show, which 01:07:43.450 --> 01:07:46.720 we can do briefly in recitation, and talk a little bit 01:07:46.720 --> 01:07:50.890 about prices and firms. 01:07:50.890 --> 01:07:54.160 So one fact that we see in the world 01:07:54.160 --> 01:07:56.770 is that demand often responds more strongly 01:07:56.770 --> 01:08:00.100 to price increases than to price decreases of frequently 01:08:00.100 --> 01:08:01.000 purchased items. 01:08:01.000 --> 01:08:02.920 And we already discussed this last time. 01:08:02.920 --> 01:08:06.730 That's to say, so usually people have a reference point 01:08:06.730 --> 01:08:08.530 in terms of either the price they pay 01:08:08.530 --> 01:08:11.000 or the expenditures on certain items. 01:08:11.000 --> 01:08:12.850 So now, as something becomes more expensive, 01:08:12.850 --> 01:08:14.308 what they tend to do is essentially 01:08:14.308 --> 01:08:16.720 reduce how much the, for example, gasoline or the like. 01:08:16.720 --> 01:08:20.229 They tend to then sort of just reduce their expenditures 01:08:20.229 --> 01:08:24.040 because they have a budget for, say, gasoline or certain items. 01:08:24.040 --> 01:08:25.600 And so what they tend to do is then 01:08:25.600 --> 01:08:27.580 the reference point is either the past price 01:08:27.580 --> 01:08:29.229 or the past expenditures. 01:08:29.229 --> 01:08:32.050 And people tend to sort of be loss averse 01:08:32.050 --> 01:08:39.970 then over their expenditures over the specific domain. 01:08:39.970 --> 01:08:42.080 Now, that then leads to-- 01:08:42.080 --> 01:08:44.229 so if a firm knows this, so essentially if you 01:08:44.229 --> 01:08:49.569 know that essentially people react a lot to price increases 01:08:49.569 --> 01:08:54.050 compared to price decreases, that leads to sticky prices, 01:08:54.050 --> 01:08:54.550 essentially. 01:08:54.550 --> 01:08:56.890 So raising your price above a past price 01:08:56.890 --> 01:08:59.470 is very costly because you lose a lot of customers 01:08:59.470 --> 01:09:00.680 from doing that. 01:09:00.680 --> 01:09:03.760 So now, lowering your price below the past price 01:09:03.760 --> 01:09:05.020 won't actually do very much. 01:09:05.020 --> 01:09:06.728 Essentially, you're not going to generate 01:09:06.728 --> 01:09:07.990 a lot more extra demand. 01:09:07.990 --> 01:09:10.060 Plus, raising the price in the future is costly. 01:09:10.060 --> 01:09:11.590 You know, essentially, once you lower the price, 01:09:11.590 --> 01:09:12.840 it's hard to get up any more. 01:09:12.840 --> 01:09:16.060 So for these frequently purchased items, 01:09:16.060 --> 01:09:17.890 you see a lot of price stickiness and price 01:09:17.890 --> 01:09:24.460 equalization across markets, across time and products. 01:09:24.460 --> 01:09:27.729 Now, that's sort of one thing about prices in the world 01:09:27.729 --> 01:09:29.352 that you would see. 01:09:29.352 --> 01:09:31.060 But more generally, I want to sort of ask 01:09:31.060 --> 01:09:32.859 the question in the last few minutes 01:09:32.859 --> 01:09:35.830 about, if you were a company, suppose 01:09:35.830 --> 01:09:39.130 you did an internship somewhere in the summer. 01:09:39.130 --> 01:09:41.895 You took behavioral economics. 01:09:41.895 --> 01:09:43.270 Hopefully, you learned something. 01:09:43.270 --> 01:09:49.383 But what can we learn about the firm policies and so on now 01:09:49.383 --> 01:09:50.800 that you know about loss aversion? 01:09:50.800 --> 01:09:52.467 What would you tell them they should do? 01:10:11.130 --> 01:10:11.730 Yes. 01:10:11.730 --> 01:10:14.310 AUDIENCE: You could start with having a really high price 01:10:14.310 --> 01:10:15.430 and then lowering them. 01:10:15.430 --> 01:10:18.270 Because people will feel like, oh, I'm 01:10:18.270 --> 01:10:19.542 getting [INAUDIBLE] deal. 01:10:19.542 --> 01:10:20.500 FRANK SCHILBACH: Right. 01:10:20.500 --> 01:10:23.010 So that was what previously was mentioned is I say, 01:10:23.010 --> 01:10:26.640 you could sort of, in particular when you introduce a price, 01:10:26.640 --> 01:10:29.278 new product and so on, you might want to start at a high price 01:10:29.278 --> 01:10:30.570 and then sort of lowering them. 01:10:30.570 --> 01:10:32.190 And people really like having deals, 01:10:32.190 --> 01:10:33.940 and they feel good about it. 01:10:33.940 --> 01:10:36.990 Notice that that doesn't work so well for products that you 01:10:36.990 --> 01:10:38.160 already have and so on. 01:10:38.160 --> 01:10:40.350 Because then essentially, once you start with a high price, 01:10:40.350 --> 01:10:41.350 people get really upset. 01:10:41.350 --> 01:10:43.180 And you will lose the customers. 01:10:43.180 --> 01:10:46.110 But once you introduce a new product, be it like an iPhone 01:10:46.110 --> 01:10:49.333 or be it like some whatever, new computer or whatever, 01:10:49.333 --> 01:10:51.750 it makes a lot of sense to start with a really high price. 01:10:51.750 --> 01:10:53.333 That sort of sets the reference point. 01:10:53.333 --> 01:10:55.110 And then people sort of feel like they 01:10:55.110 --> 01:10:57.092 get good deals overall. 01:10:57.092 --> 01:10:58.050 What else could you do? 01:10:58.050 --> 01:10:58.550 Yes. 01:10:58.550 --> 01:11:01.308 AUDIENCE: You could offer [INAUDIBLE] price. 01:11:01.308 --> 01:11:02.850 You could offer a temporary discount. 01:11:02.850 --> 01:11:04.770 That was you keep your reference point. 01:11:04.770 --> 01:11:06.900 [INAUDIBLE] offer discounts when you 01:11:06.900 --> 01:11:09.328 want to lower the price temporarily [INAUDIBLE].. 01:11:09.328 --> 01:11:10.370 FRANK SCHILBACH: Exactly. 01:11:10.370 --> 01:11:12.120 That's what a lot of companies tend to do. 01:11:12.120 --> 01:11:14.670 They tend to do, essentially, these special occasions 01:11:14.670 --> 01:11:17.502 with Black Friday and the like. 01:11:17.502 --> 01:11:18.960 Exactly as you say, it's temporary. 01:11:18.960 --> 01:11:22.520 It's not a thing that prices are permanently lower. 01:11:22.520 --> 01:11:25.040 It's just, right now, you get this great deal. 01:11:25.040 --> 01:11:26.990 And then things go back to normal. 01:11:26.990 --> 01:11:29.240 And somehow you have to hope that people don't sort of 01:11:29.240 --> 01:11:31.850 adjust their reference point towards the temporarily 01:11:31.850 --> 01:11:34.910 lowered price. 01:11:34.910 --> 01:11:35.450 What else? 01:11:35.450 --> 01:11:36.110 Yeah. 01:11:36.110 --> 01:11:38.490 AUDIENCE: Kind of down that line, you have free trials. 01:11:38.490 --> 01:11:39.860 So companies sends you-- 01:11:39.860 --> 01:11:40.735 FRANK SCHILBACH: Yes. 01:11:40.735 --> 01:11:43.040 AUDIENCE: --for [INAUDIBLE] and then take it away, 01:11:43.040 --> 01:11:45.900 people will feel more compelled to get it back 01:11:45.900 --> 01:11:47.440 because it's loss aversion. 01:11:47.440 --> 01:11:50.610 And that's what [? creates ?] [INAUDIBLE] people value it 01:11:50.610 --> 01:11:53.535 at higher price than otherwise. 01:11:53.535 --> 01:11:54.410 FRANK SCHILBACH: Yes. 01:11:54.410 --> 01:11:55.250 AUDIENCE: So [INAUDIBLE]. 01:11:55.250 --> 01:11:55.800 FRANK SCHILBACH: Yes. 01:11:55.800 --> 01:11:57.758 So if you shop online, you might have wondered. 01:11:57.758 --> 01:11:59.508 Lots of companies have this thing on like, 01:11:59.508 --> 01:12:01.038 oh, you can order whatever you want. 01:12:01.038 --> 01:12:02.330 Essentially, it's free returns. 01:12:02.330 --> 01:12:04.638 And you kind of wonder, that seems like a bad deal 01:12:04.638 --> 01:12:05.930 from the company's perspective. 01:12:05.930 --> 01:12:08.300 Because people buy a lot of stuff and send it back. 01:12:08.300 --> 01:12:09.685 The hope is exactly as you say. 01:12:09.685 --> 01:12:11.810 It might be just in part people like to experiment. 01:12:11.810 --> 01:12:13.160 And they like some stuff and not others. 01:12:13.160 --> 01:12:14.385 And that's worth doing it. 01:12:14.385 --> 01:12:16.260 But the hope, in particular, is to say, well, 01:12:16.260 --> 01:12:17.802 I want you to try it out for a while. 01:12:17.802 --> 01:12:19.460 You get used to it. 01:12:19.460 --> 01:12:20.930 Then the endowment effect kicks in. 01:12:20.930 --> 01:12:22.040 You value it more. 01:12:22.040 --> 01:12:23.660 It essentially becomes yours. 01:12:23.660 --> 01:12:28.700 And then essentially you become loss averse towards that. 01:12:28.700 --> 01:12:30.180 By the way, I should have mentioned 01:12:30.180 --> 01:12:33.335 this is really a fascinating book by Cialdini who 01:12:33.335 --> 01:12:37.760 was talking about the psychology of persuasion. 01:12:37.760 --> 01:12:40.310 He spend a lot of time with salespeople, 01:12:40.310 --> 01:12:42.410 in particular sort of car salespeople and so on, 01:12:42.410 --> 01:12:45.350 trying to learn what are they actually doing in markets. 01:12:45.350 --> 01:12:48.380 And he has these amazing stories of salespeople, what's 01:12:48.380 --> 01:12:50.300 all sorts of tricks they use. 01:12:50.300 --> 01:12:53.177 It's psychologically extremely rich and interesting in terms 01:12:53.177 --> 01:12:55.010 of just trying to understand what people do. 01:12:55.010 --> 01:12:59.390 And that ranges from things, once you purchase a car, 01:12:59.390 --> 01:13:01.190 they let you test drive in the car. 01:13:01.190 --> 01:13:02.030 And you sit in it. 01:13:02.030 --> 01:13:03.620 And it feels like yours. 01:13:03.620 --> 01:13:05.387 Or when you try to buy a house, then 01:13:05.387 --> 01:13:07.970 people would say, oh, you know, this will be your living room. 01:13:07.970 --> 01:13:09.887 And this is where your baby is going to sleep. 01:13:09.887 --> 01:13:12.590 And there's lots of sort of ways in which people 01:13:12.590 --> 01:13:14.870 make sort of something feel yours and really try 01:13:14.870 --> 01:13:19.940 to sort of get the endowment effect to kick in. 01:13:19.940 --> 01:13:21.800 If you were to work in an insurance firm, 01:13:21.800 --> 01:13:24.560 what would you do, somebody who offers insurance 01:13:24.560 --> 01:13:28.070 in one way or the other or products 01:13:28.070 --> 01:13:29.720 that offer insurance in some ways? 01:13:34.650 --> 01:13:35.193 Yes. 01:13:35.193 --> 01:13:37.860 AUDIENCE: We talked about this a little bit in a previous class. 01:13:37.860 --> 01:13:39.930 But this is why, I think, companies 01:13:39.930 --> 01:13:43.080 will sell things like Apple Care or operative warranties. 01:13:43.080 --> 01:13:45.360 Because they know they'll make money off of it 01:13:45.360 --> 01:13:46.868 because people tend to over-insure. 01:13:46.868 --> 01:13:47.910 FRANK SCHILBACH: Exactly. 01:13:47.910 --> 01:13:50.460 There's lots of different products 01:13:50.460 --> 01:13:52.470 where there's extended warranties and all sorts 01:13:52.470 --> 01:13:55.740 of things, like Apple care, et cetera, where people 01:13:55.740 --> 01:13:59.880 are very risk averse, what it looks like, presumably loss 01:13:59.880 --> 01:14:02.760 averse where, in fact, actually the claim 01:14:02.760 --> 01:14:04.230 rate tends to be very low. 01:14:04.230 --> 01:14:06.720 So you can make a lot of money with this by saying, yes, 01:14:06.720 --> 01:14:08.520 I'm going to exchange it and this and that. 01:14:08.520 --> 01:14:10.530 Because, essentially, it doesn't happen that 01:14:10.530 --> 01:14:13.477 often at the end of the day even if there's 01:14:13.477 --> 01:14:15.060 moral hazard or other issues of people 01:14:15.060 --> 01:14:19.600 not treating their stuff that well once they have insurance. 01:14:19.600 --> 01:14:26.070 What about wage setting, like when 01:14:26.070 --> 01:14:27.750 you set your employees' wages? 01:14:30.710 --> 01:14:31.210 Yeah. 01:14:31.210 --> 01:14:34.440 AUDIENCE: Could that be linked to mass unemployment 01:14:34.440 --> 01:14:37.720 in recessions because you know that, if you lower the wage, 01:14:37.720 --> 01:14:39.590 then morale will go down a lot? 01:14:39.590 --> 01:14:43.472 And you can actually decide just to kick them out? 01:14:43.472 --> 01:14:44.430 FRANK SCHILBACH: Right. 01:14:44.430 --> 01:14:44.930 Exactly. 01:14:44.930 --> 01:14:53.040 So there is a large literature on wage stickiness, 01:14:53.040 --> 01:14:57.990 essentially exactly as you say, where people are extremely 01:14:57.990 --> 01:14:59.910 reluctant to lower wages. 01:14:59.910 --> 01:15:02.250 Companies are very reluctant to lower wages. 01:15:02.250 --> 01:15:05.310 Because, essentially, workers are really unhappy. 01:15:05.310 --> 01:15:08.192 And these are often nominal wages or real wages. 01:15:08.192 --> 01:15:09.650 It doesn't really matter that much, 01:15:09.650 --> 01:15:11.130 but usually it's nominal wages. 01:15:11.130 --> 01:15:16.080 People really, really dislike nominal wage reductions. 01:15:16.080 --> 01:15:21.270 And so, now, in some times when companies would actually 01:15:21.270 --> 01:15:24.840 need to lower wages and sort of to be able to keep workers 01:15:24.840 --> 01:15:27.990 not to make losses, companies might rather sort of fire 01:15:27.990 --> 01:15:31.830 some workers rather than sort of lowering 01:15:31.830 --> 01:15:33.090 the wages for everybody. 01:15:33.090 --> 01:15:36.492 Because, essentially, the remaining workers, 01:15:36.492 --> 01:15:37.950 once you lower wages for everybody, 01:15:37.950 --> 01:15:39.090 everybody would be unhappy. 01:15:39.090 --> 01:15:41.760 If you just fire one worker, everybody else 01:15:41.760 --> 01:15:45.960 will be less unhappy than about their wage reductions. 01:15:45.960 --> 01:15:48.750 Similarly, firms are sort of reluctant to hire people 01:15:48.750 --> 01:15:51.510 at lower wages, if there's other people who make higher wages, 01:15:51.510 --> 01:15:54.155 because people really dislike weight dispersion. 01:15:54.155 --> 01:15:55.530 So essentially, overall, you want 01:15:55.530 --> 01:15:59.280 to avoid wage cuts as much as possible. 01:15:59.280 --> 01:16:01.190 And that leads to essentially then sort 01:16:01.190 --> 01:16:03.247 of macroeconomic inefficiencies. 01:16:03.247 --> 01:16:04.830 And people have argued, in particular, 01:16:04.830 --> 01:16:10.410 it leads to unemployment because, essentially, wages 01:16:10.410 --> 01:16:13.410 are not going down as much as they should in recessions. 01:16:13.410 --> 01:16:14.460 And that's bad for firms. 01:16:14.460 --> 01:16:16.200 And, therefore, they hire fewer workers 01:16:16.200 --> 01:16:18.983 or retain fewer workers overall. 01:16:18.983 --> 01:16:20.400 I think I mentioned, we mentioned, 01:16:20.400 --> 01:16:23.110 all of those kinds of things. 01:16:23.110 --> 01:16:26.833 So now, that's sort of what firms are doing. 01:16:26.833 --> 01:16:28.500 Now, another thing you might think about 01:16:28.500 --> 01:16:31.472 is what are you going to actually do in your real lives. 01:16:31.472 --> 01:16:33.930 I think there's many different things that you can actually 01:16:33.930 --> 01:16:37.093 think about is the framing of situations, for example, can 01:16:37.093 --> 01:16:38.010 make a big difference. 01:16:38.010 --> 01:16:39.720 If you present something to your friends, 01:16:39.720 --> 01:16:43.500 different options, whether you present that as gains or losses 01:16:43.500 --> 01:16:45.930 makes a big difference potentially. 01:16:45.930 --> 01:16:48.540 Managing people's expectations is really important. 01:16:48.540 --> 01:16:50.610 If there's some big goal that they could reach 01:16:50.610 --> 01:16:52.380 or some lower goal they could reach, 01:16:52.380 --> 01:16:54.030 if you sort of oversell the high chance 01:16:54.030 --> 01:16:56.048 of reaching some big goal, they might 01:16:56.048 --> 01:16:58.090 reach the other goal that's actually pretty good, 01:16:58.090 --> 01:17:01.020 might feel really disappointed about that, 01:17:01.020 --> 01:17:03.310 be it in job search or the like. 01:17:03.310 --> 01:17:04.810 So managing people's expectations, 01:17:04.810 --> 01:17:07.810 including your own expectations, seems really important. 01:17:07.810 --> 01:17:10.080 There's something about aggregating losses and gains, 01:17:10.080 --> 01:17:12.960 which essentially is to say, since people seem 01:17:12.960 --> 01:17:18.810 to be risk averse over gains-- 01:17:18.810 --> 01:17:22.170 so since value function is concave over gains and convex 01:17:22.170 --> 01:17:25.020 over losses, what you should do is 01:17:25.020 --> 01:17:29.790 potentially be really careful about when you give people 01:17:29.790 --> 01:17:34.320 a positive or negative feedback or bonuses and so on. 01:17:34.320 --> 01:17:37.950 You might want to sort of be careful whether you 01:17:37.950 --> 01:17:39.600 aggregate the losses or-- 01:17:39.600 --> 01:17:43.050 so aggregating losses makes sense because essentially it's 01:17:43.050 --> 01:17:45.480 convex in the loss domain, as opposed to you 01:17:45.480 --> 01:17:48.013 want to give small increments of gains overall. 01:17:48.013 --> 01:17:49.680 Now, one thing that I do want to mention 01:17:49.680 --> 01:17:51.480 is you want to be kind of very careful 01:17:51.480 --> 01:17:53.100 with loss-framed incentives. 01:17:53.100 --> 01:17:57.240 There's a company who was trying to do this. 01:17:57.240 --> 01:17:59.010 These are car manufacturers who were 01:17:59.010 --> 01:18:02.910 trying to give essentially their car dealers incentives, 01:18:02.910 --> 01:18:09.802 sort of targets for their sales of their cars. 01:18:09.802 --> 01:18:11.760 What essentially happened in the end of the day 01:18:11.760 --> 01:18:15.540 is there was a bunch of multitasking. 01:18:15.540 --> 01:18:17.280 The company or the car salespeople 01:18:17.280 --> 01:18:20.400 were essentially selling certain cars, but then not others 01:18:20.400 --> 01:18:21.900 and were essentially multitasking 01:18:21.900 --> 01:18:24.970 and then sort of reverting, sort of reallocating efforts 01:18:24.970 --> 01:18:27.570 to one thing versus the other, which then 01:18:27.570 --> 01:18:28.950 seemed like a really good idea. 01:18:28.950 --> 01:18:31.858 There was recently a valuation that sort of showed that that 01:18:31.858 --> 01:18:33.150 was actually a pretty bad idea. 01:18:33.150 --> 01:18:35.067 And the company would have lost a lot of money 01:18:35.067 --> 01:18:37.545 overall if this had been scaled. 01:18:37.545 --> 01:18:39.420 We're going to have this, the Deal or No Deal 01:18:39.420 --> 01:18:40.980 and this specific paper in recitation 01:18:40.980 --> 01:18:42.540 to tell you in more detail because I 01:18:42.540 --> 01:18:46.620 want to move towards social preferences next time. 01:18:46.620 --> 01:18:50.010 So as I said, next time we talk about social preferences. 01:18:50.010 --> 01:18:50.790 Bring your laptop. 01:18:50.790 --> 01:18:52.415 And I'll send you further instructions. 01:18:52.415 --> 01:18:54.020 Thank you.