WEBVTT 00:00:00.000 --> 00:00:06.461 align:middle line:84% [SQUEAKING] [RUSTLING] [CLICKING] 00:00:06.461 --> 00:00:17.210 align:middle line:90% 00:00:17.210 --> 00:00:20.390 align:middle line:84% GLENN ELLISON: So anyway, today's lecture, I was doing-- 00:00:20.390 --> 00:00:22.470 align:middle line:84% I did static competition last week. 00:00:22.470 --> 00:00:24.650 align:middle line:84% Today, my plan is I'm going to start with empirics 00:00:24.650 --> 00:00:26.278 align:middle line:90% of static competition. 00:00:26.278 --> 00:00:28.820 align:middle line:84% I'm going to fill in, in the middle, some extra theory that I 00:00:28.820 --> 00:00:32.023 align:middle line:84% didn't have time to do last Wednesday. 00:00:32.023 --> 00:00:33.440 align:middle line:84% And then the paper at the end, I'm 00:00:33.440 --> 00:00:35.690 align:middle line:84% going to do very, very quickly because I just was reading it 00:00:35.690 --> 00:00:37.398 align:middle line:84% over, making up the notes, and I realized 00:00:37.398 --> 00:00:41.480 align:middle line:84% that Tobias is going to be spending an awful lot of time 00:00:41.480 --> 00:00:44.570 align:middle line:84% next Monday explaining what the model is 00:00:44.570 --> 00:00:45.900 align:middle line:90% about and the estimation of it. 00:00:45.900 --> 00:00:47.540 align:middle line:84% And I figured that it'd be much better 00:00:47.540 --> 00:00:48.990 align:middle line:84% to wait until after he's done that 00:00:48.990 --> 00:00:51.290 align:middle line:84% and you can just look at my slides on it. 00:00:51.290 --> 00:00:53.700 align:middle line:84% And I'll just stick with basic comments. 00:00:53.700 --> 00:00:56.210 align:middle line:84% So anyway, I'm going to first cover 00:00:56.210 --> 00:00:58.170 align:middle line:84% one of my favorite papers in the course. 00:00:58.170 --> 00:01:01.670 align:middle line:84% So this is Bresnahan's paper, "Competition and Collusion 00:01:01.670 --> 00:01:06.260 align:middle line:84% in the American Automobile Industry: The 1955 Price War." 00:01:06.260 --> 00:01:08.960 align:middle line:84% I should say, in some sense, it's one of my favorite papers. 00:01:08.960 --> 00:01:11.210 align:middle line:84% I took the star off it this year because in some ways, 00:01:11.210 --> 00:01:13.610 align:middle line:84% I think as the paper gets more and more dated, 00:01:13.610 --> 00:01:16.200 align:middle line:84% you can appreciate it more by hearing about what it does than 00:01:16.200 --> 00:01:17.400 align:middle line:90% by actually reading it. 00:01:17.400 --> 00:01:19.860 align:middle line:84% There just many things that you read and you're like, wow, 00:01:19.860 --> 00:01:22.240 align:middle line:84% that sounds like a 40-year-old way to do something. 00:01:22.240 --> 00:01:25.440 align:middle line:84% And so I'll just leave out the 40-year-old parts that were 00:01:25.440 --> 00:01:28.350 align:middle line:84% a bit awkward and probably had some trouble publishing it 00:01:28.350 --> 00:01:30.910 align:middle line:84% to begin with, and I'll give you the main ideas. 00:01:30.910 --> 00:01:37.650 align:middle line:84% So this is a paper on the IO - Economic History boundary. 00:01:37.650 --> 00:01:44.350 align:middle line:84% The applied topic is what caused the 1955 surge in US auto sales, 00:01:44.350 --> 00:01:46.380 align:middle line:84% and it has this interesting hypothesis, 00:01:46.380 --> 00:01:49.920 align:middle line:84% which is that the 1955 surge in auto sales 00:01:49.920 --> 00:01:53.430 align:middle line:84% was due to a breakdown in collusion that, at that time, 00:01:53.430 --> 00:01:56.110 align:middle line:84% there were very few foreign cars sold in the United States. 00:01:56.110 --> 00:01:58.240 align:middle line:84% There were largely three American automakers. 00:01:58.240 --> 00:02:01.350 align:middle line:84% And his hypothesis is that the three US automakers were 00:02:01.350 --> 00:02:04.330 align:middle line:84% colluding to set monopoly prices, 00:02:04.330 --> 00:02:07.230 align:middle line:84% and then in 1955, the collusion broke down, 00:02:07.230 --> 00:02:09.600 align:middle line:84% prices dropped, many more cars were sold. 00:02:09.600 --> 00:02:12.550 align:middle line:84% Anyway, and then the firms got it figured out again. 00:02:12.550 --> 00:02:16.410 align:middle line:90% In 1956, prices went back up. 00:02:16.410 --> 00:02:20.300 align:middle line:84% The paper, besides having an interesting historical fact, 00:02:20.300 --> 00:02:24.080 align:middle line:84% it has been really influential as a model for future work. 00:02:24.080 --> 00:02:28.400 align:middle line:84% In some sense, I see this as the first structural empirical paper 00:02:28.400 --> 00:02:30.710 align:middle line:84% in IO, and I think that what you can think of 00:02:30.710 --> 00:02:34.880 align:middle line:84% is that really many future papers have picked up 00:02:34.880 --> 00:02:37.032 align:middle line:84% on Bresnahan's methodology and adopted this 00:02:37.032 --> 00:02:38.490 align:middle line:84% as a different way of doing things. 00:02:38.490 --> 00:02:41.720 align:middle line:84% So I will explain what it was that Bresnahan 00:02:41.720 --> 00:02:43.610 align:middle line:90% was trying to do in this paper. 00:02:43.610 --> 00:02:47.690 align:middle line:84% But first on the applied topic, you know, here's-- 00:02:47.690 --> 00:02:49.988 align:middle line:84% he starts with a simple table of car sales 00:02:49.988 --> 00:02:51.530 align:middle line:84% by year in the United States, and you 00:02:51.530 --> 00:02:56.210 align:middle line:84% get this striking finding, which is in 1954, 00:02:56.210 --> 00:02:58.770 align:middle line:84% US automakers produce 5.5 million cars. 00:02:58.770 --> 00:03:02.040 align:middle line:84% In 1955, they produce 7.9 million cars. 00:03:02.040 --> 00:03:06.500 align:middle line:84% In 1956, they produced 5.8 million cars. 00:03:06.500 --> 00:03:08.010 align:middle line:84% Cars are a big part of the economy, 00:03:08.010 --> 00:03:11.120 align:middle line:84% and the cars being produced going up by 40% in one year 00:03:11.120 --> 00:03:13.160 align:middle line:90% is a really striking fact. 00:03:13.160 --> 00:03:17.340 align:middle line:84% If you look at it, sales in dollars, it's still a big jump. 00:03:17.340 --> 00:03:20.580 align:middle line:84% The jump in dollars of car sales is not 00:03:20.580 --> 00:03:23.370 align:middle line:84% as big as the jump in unit car sales. 00:03:23.370 --> 00:03:25.530 align:middle line:84% In part, that's because prices were lower, 00:03:25.530 --> 00:03:30.270 align:middle line:84% and in part, that's because people did buy smaller cars. 00:03:30.270 --> 00:03:33.450 align:middle line:84% And then if you look at what was going on in the US economy, 00:03:33.450 --> 00:03:36.070 align:middle line:84% it was a boom time in the US economy. 00:03:36.070 --> 00:03:37.710 align:middle line:84% You can see non-auto durable goods 00:03:37.710 --> 00:03:40.680 align:middle line:84% spending went from 14.5 to 16 billion 00:03:40.680 --> 00:03:44.290 align:middle line:84% and then continued increasing to 17 billion the next year. 00:03:44.290 --> 00:03:46.560 align:middle line:84% So it was a good year for the US economy, 00:03:46.560 --> 00:03:50.280 align:middle line:84% but not a 40% larger increase in car sales 00:03:50.280 --> 00:03:51.420 align:middle line:90% boom in the US economy. 00:03:51.420 --> 00:03:59.190 align:middle line:84% There really are not 40% boom in car sales general macro shocks. 00:03:59.190 --> 00:04:01.890 align:middle line:84% You can find explanations in the popular press 00:04:01.890 --> 00:04:06.660 align:middle line:84% of what caused the 1955 thing as a cultural phenomenon. 00:04:06.660 --> 00:04:08.790 align:middle line:84% Here's a quote from Car Talk describing 00:04:08.790 --> 00:04:13.110 align:middle line:84% this, where they talk about how rock music and drive-in 00:04:13.110 --> 00:04:15.070 align:middle line:84% restaurants were becoming a thing 00:04:15.070 --> 00:04:20.180 align:middle line:84% and there were sleek new designs for cars and whatever. 00:04:20.180 --> 00:04:23.990 align:middle line:84% Apparently back in the 1950s, Paul Samuelson taught both 00:04:23.990 --> 00:04:25.895 align:middle line:90% [14.]121-- 00:04:25.895 --> 00:04:27.830 align:middle line:84% taught both in the micro sequence, 00:04:27.830 --> 00:04:31.340 align:middle line:84% [14.]121 through [14.]124, and in the econometric sequence, 00:04:31.340 --> 00:04:32.700 align:middle line:90% [14.]381, [14.]382. 00:04:32.700 --> 00:04:34.880 align:middle line:84% I think the numbers were still the same 50 years 00:04:34.880 --> 00:04:38.180 align:middle line:90% or 70 years ago, that is now. 00:04:38.180 --> 00:04:42.860 align:middle line:84% But Paul, apparently-- this was pre-14.192, 00:04:42.860 --> 00:04:46.070 align:middle line:84% but there was a required second year econometrics paper. 00:04:46.070 --> 00:04:48.770 align:middle line:84% And Paul Samuelson apparently was teaching the second year 00:04:48.770 --> 00:04:51.590 align:middle line:84% econometrics course, would say that he would flunk any paper 00:04:51.590 --> 00:04:55.280 align:middle line:84% that claimed it was providing an explanation for 1955 sales, 00:04:55.280 --> 00:04:58.118 align:middle line:84% where by "explanation" he meant you're regressing sales 00:04:58.118 --> 00:05:00.410 align:middle line:84% and a bunch of right-hand side variables like the price 00:05:00.410 --> 00:05:03.770 align:middle line:90% of steel and whatever. 00:05:03.770 --> 00:05:06.000 align:middle line:90% OK, so what were cars in 1955? 00:05:06.000 --> 00:05:08.100 align:middle line:84% So I've got some illustrations here. 00:05:08.100 --> 00:05:13.280 align:middle line:84% So 1955 Mercury Montclair, 1955 Chevy Bel Air, 00:05:13.280 --> 00:05:15.520 align:middle line:90% 1955 Ford Thunderbird. 00:05:15.520 --> 00:05:19.560 align:middle line:84% You can see the '55 Thunderbird is a very cute car. 00:05:19.560 --> 00:05:22.900 align:middle line:84% The other cars just look pretty generic to me. 00:05:22.900 --> 00:05:27.450 align:middle line:84% For contrast, here's the 1955 Chevy Bel Air versus the 1954 00:05:27.450 --> 00:05:29.880 align:middle line:84% Chevy Bel Air, and you can see that they 00:05:29.880 --> 00:05:32.490 align:middle line:84% seem to have taken away a little bit of this wheel thing 00:05:32.490 --> 00:05:36.000 align:middle line:84% that's stuck out here and they smoothed out the grille. 00:05:36.000 --> 00:05:38.620 align:middle line:84% Instead of these big things, they went to small things. 00:05:38.620 --> 00:05:41.280 align:middle line:84% You know, it fits with the sleeker car theory, 00:05:41.280 --> 00:05:45.073 align:middle line:84% but it doesn't look all that different to me. 00:05:45.073 --> 00:05:46.740 align:middle line:84% Perhaps someday-- perhaps today, someone 00:05:46.740 --> 00:05:48.330 align:middle line:84% could write a machine learning thing 00:05:48.330 --> 00:05:50.340 align:middle line:84% judging the sleekness of cars and seeing 00:05:50.340 --> 00:05:52.802 align:middle line:84% how much the sleekness of cars affects increased sales 00:05:52.802 --> 00:05:53.760 align:middle line:90% or something like that. 00:05:53.760 --> 00:05:54.302 align:middle line:90% I don't know. 00:05:54.302 --> 00:05:56.910 align:middle line:90% 00:05:56.910 --> 00:05:59.280 align:middle line:84% OK, anyway, so his hypothesis is that this 00:05:59.280 --> 00:06:02.165 align:middle line:84% was caused by a breakdown in collusion among the big three. 00:06:02.165 --> 00:06:05.040 align:middle line:90% 00:06:05.040 --> 00:06:08.183 align:middle line:84% In general, if you're writing a-- 00:06:08.183 --> 00:06:10.600 align:middle line:84% there are several types of empirical papers one can write. 00:06:10.600 --> 00:06:12.310 align:middle line:84% If you're writing a, here's a fact. 00:06:12.310 --> 00:06:14.940 align:middle line:84% I have an explanation for that fact. 00:06:14.940 --> 00:06:17.770 align:middle line:84% Generally, if the fact is well-known 00:06:17.770 --> 00:06:19.900 align:middle line:84% and you're proposing an explanation for it, 00:06:19.900 --> 00:06:21.550 align:middle line:84% there are a thousand explanations 00:06:21.550 --> 00:06:23.120 align:middle line:84% for any fact you can come up with. 00:06:23.120 --> 00:06:24.662 align:middle line:84% So generally, the way you try to make 00:06:24.662 --> 00:06:27.730 align:middle line:84% the case that your explanation is correct and convincing 00:06:27.730 --> 00:06:30.790 align:middle line:84% is to think of a lot of other predictions 00:06:30.790 --> 00:06:33.010 align:middle line:84% that go along with your explanation 00:06:33.010 --> 00:06:34.678 align:middle line:84% that people haven't talked about yet, 00:06:34.678 --> 00:06:36.970 align:middle line:84% and then show that here's a bunch of other implications 00:06:36.970 --> 00:06:37.512 align:middle line:90% of my theory. 00:06:37.512 --> 00:06:39.512 align:middle line:84% Now I'm going to look at the data in more detail 00:06:39.512 --> 00:06:41.500 align:middle line:84% and show these other implications of my theory 00:06:41.500 --> 00:06:43.660 align:middle line:90% are true as well. 00:06:43.660 --> 00:06:47.290 align:middle line:84% And that's what Bresnahan is trying to do. 00:06:47.290 --> 00:06:52.210 align:middle line:84% And his idea is that motivated by then recent literature 00:06:52.210 --> 00:06:53.830 align:middle line:84% on differentiated product competition 00:06:53.830 --> 00:06:57.880 align:middle line:84% was that if there was a shift from collusion to competition, 00:06:57.880 --> 00:07:01.210 align:middle line:84% it ought to have a different effect on different car models. 00:07:01.210 --> 00:07:04.900 align:middle line:84% In particular, car models that were close in product space 00:07:04.900 --> 00:07:09.700 align:middle line:84% to other car models would see a much bigger drop in price 00:07:09.700 --> 00:07:11.810 align:middle line:84% than car models that were very distinctive. 00:07:11.810 --> 00:07:16.460 align:middle line:84% So on my previous slide, you can see the Ford Thunderbird. 00:07:16.460 --> 00:07:18.740 align:middle line:90% It was an original car. 00:07:18.740 --> 00:07:21.260 align:middle line:84% It looks very different from those other cars. 00:07:21.260 --> 00:07:23.290 align:middle line:84% Maybe if you're colluding or in competition, 00:07:23.290 --> 00:07:25.415 align:middle line:84% it doesn't matter so much for the price of the Ford 00:07:25.415 --> 00:07:26.000 align:middle line:90% Thunderbird. 00:07:26.000 --> 00:07:28.030 align:middle line:84% You still have a lot of market power. 00:07:28.030 --> 00:07:29.650 align:middle line:84% But at least to my mind, if you're 00:07:29.650 --> 00:07:34.000 align:middle line:84% looking at the Mercury Montclair versus the Chevy Bel Air, 00:07:34.000 --> 00:07:36.070 align:middle line:84% I would have to spend some time learning 00:07:36.070 --> 00:07:38.240 align:middle line:84% to recognize the difference between those two 00:07:38.240 --> 00:07:40.400 align:middle line:84% if you were showing me pictures of them. 00:07:40.400 --> 00:07:42.250 align:middle line:84% And so you might think if there was 00:07:42.250 --> 00:07:46.480 align:middle line:84% Bertrand competition between Mercury and Chevy, 00:07:46.480 --> 00:07:49.540 align:middle line:84% the price of the Bel Air and the Montclair, 00:07:49.540 --> 00:07:51.880 align:middle line:84% they could be very tough competition 00:07:51.880 --> 00:07:54.640 align:middle line:84% and drive prices down towards cost. 00:07:54.640 --> 00:08:02.330 align:middle line:84% So just, Bresnahan has these pictures illustrating this. 00:08:02.330 --> 00:08:04.210 align:middle line:84% He supposes that, you know, suppose 00:08:04.210 --> 00:08:07.930 align:middle line:84% that this was just a pure one dimensional vertical competition 00:08:07.930 --> 00:08:09.980 align:middle line:84% model like I described last time. 00:08:09.980 --> 00:08:13.240 align:middle line:90% So we put quality on the x-axis. 00:08:13.240 --> 00:08:16.490 align:middle line:84% This solid line is supposed to represent cost. 00:08:16.490 --> 00:08:19.180 align:middle line:84% And then these are the prices of the different car 00:08:19.180 --> 00:08:23.510 align:middle line:84% models where General Motors and Ford are the two firms producing 00:08:23.510 --> 00:08:24.010 align:middle line:90% them. 00:08:24.010 --> 00:08:26.320 align:middle line:84% It is true that generally, General Motors 00:08:26.320 --> 00:08:28.390 align:middle line:84% tend to produce the bigger cars and Ford tended 00:08:28.390 --> 00:08:29.750 align:middle line:90% to produce the smaller cars. 00:08:29.750 --> 00:08:32.230 align:middle line:84% So GM, with its Cadillac division, 00:08:32.230 --> 00:08:34.690 align:middle line:84% had a lot of the upper end of the quality spectrum 00:08:34.690 --> 00:08:37.396 align:middle line:90% more to itself. 00:08:37.396 --> 00:08:41.460 align:middle line:84% And what he says is that in this vertical differentiation model, 00:08:41.460 --> 00:08:44.730 align:middle line:84% if you have models that are relatively far apart, 00:08:44.730 --> 00:08:47.190 align:middle line:84% like models four and five, if you 00:08:47.190 --> 00:08:51.240 align:middle line:84% view these as very different in quality space, 00:08:51.240 --> 00:08:54.720 align:middle line:84% then if these are the collusive prices 00:08:54.720 --> 00:08:56.520 align:middle line:84% and they start competing with each other, 00:08:56.520 --> 00:08:58.800 align:middle line:84% it may be that prices for four and five 00:08:58.800 --> 00:09:01.260 align:middle line:84% don't drop very much because they're still far enough 00:09:01.260 --> 00:09:04.560 align:middle line:84% apart that the marginal consumers are small 00:09:04.560 --> 00:09:07.350 align:middle line:84% relative to total demand so you don't do much to try to attract 00:09:07.350 --> 00:09:10.290 align:middle line:90% more marginal consumers. 00:09:10.290 --> 00:09:14.040 align:middle line:84% But on the other hand, if you look at models two and three 00:09:14.040 --> 00:09:15.750 align:middle line:90% that are very similar-- 00:09:15.750 --> 00:09:20.070 align:middle line:84% so this could be my Bel Air and Montclair. 00:09:20.070 --> 00:09:23.548 align:middle line:84% If they're colluding, they may charge fairly high prices. 00:09:23.548 --> 00:09:25.590 align:middle line:84% And then if they start competing with each other, 00:09:25.590 --> 00:09:28.200 align:middle line:84% prices may get driven very low because they're 00:09:28.200 --> 00:09:29.435 align:middle line:90% very close in product space. 00:09:29.435 --> 00:09:35.240 align:middle line:90% 00:09:35.240 --> 00:09:37.590 align:middle line:84% And so if you had that as an idea 00:09:37.590 --> 00:09:41.255 align:middle line:84% and you thought about how you could test that, Bresnahan could 00:09:41.255 --> 00:09:42.630 align:middle line:84% have taken this sort of, what I'd 00:09:42.630 --> 00:09:44.610 align:middle line:84% call the classic reduced form approach in IO, 00:09:44.610 --> 00:09:47.560 align:middle line:84% which is you make predictions, you test the predictions. 00:09:47.560 --> 00:09:50.070 align:middle line:84% And so what you could have looked for is, you know, 00:09:50.070 --> 00:09:53.970 align:middle line:84% are price drops from 1954 to 1955 00:09:53.970 --> 00:09:57.760 align:middle line:84% larger for car models that are closer in product space? 00:09:57.760 --> 00:09:59.940 align:middle line:84% And that would be an auxiliary prediction 00:09:59.940 --> 00:10:03.660 align:middle line:90% of a breakdown in collusion. 00:10:03.660 --> 00:10:08.580 align:middle line:84% But, you know, Bresnahan thought that this is really not all 00:10:08.580 --> 00:10:10.230 align:middle line:90% you could do with that theory. 00:10:10.230 --> 00:10:14.220 align:middle line:84% One thing is that the theory is more complicated than that. 00:10:14.220 --> 00:10:17.970 align:middle line:84% Like if you remember the vertical competition model, 00:10:17.970 --> 00:10:20.760 align:middle line:84% every firm does have a first order condition that's 00:10:20.760 --> 00:10:23.850 align:middle line:84% pinned down to the firm above and to the firm below, 00:10:23.850 --> 00:10:25.350 align:middle line:84% and those two first order conditions 00:10:25.350 --> 00:10:26.590 align:middle line:84% are going to pin down the prices. 00:10:26.590 --> 00:10:28.132 align:middle line:84% But those things are going to involve 00:10:28.132 --> 00:10:31.380 align:middle line:84% the prices of the next lower quality model and the model 00:10:31.380 --> 00:10:33.320 align:middle line:84% below that and the model below that. 00:10:33.320 --> 00:10:36.380 align:middle line:84% And so it is a system of n equations and n unknowns, 00:10:36.380 --> 00:10:39.730 align:middle line:84% and any change in the competition-- 00:10:39.730 --> 00:10:41.650 align:middle line:84% like the change in any one place is 00:10:41.650 --> 00:10:44.542 align:middle line:84% going to have ripple effects throughout the distribution 00:10:44.542 --> 00:10:46.750 align:middle line:84% and will change the prices of the higher price models 00:10:46.750 --> 00:10:50.047 align:middle line:84% even if they're not in very close competition. 00:10:50.047 --> 00:10:51.880 align:middle line:84% And then the other thing that Bresnahan said 00:10:51.880 --> 00:10:54.880 align:middle line:84% is that our theories are quantitative theories, not just 00:10:54.880 --> 00:10:55.880 align:middle line:90% qualitative theories. 00:10:55.880 --> 00:10:58.430 align:middle line:84% They don't just predict that prices will go down. 00:10:58.430 --> 00:11:01.570 align:middle line:84% They really are predicting how much the prices would go down, 00:11:01.570 --> 00:11:05.020 align:middle line:84% and how much the price would go down depend on things that we, 00:11:05.020 --> 00:11:06.790 align:middle line:90% in theory, could measure. 00:11:06.790 --> 00:11:09.970 align:middle line:84% And so really, to give a more complete test of the theory, 00:11:09.970 --> 00:11:12.930 align:middle line:84% we should really see whether the theory can quantitatively 00:11:12.930 --> 00:11:15.430 align:middle line:84% account for everything instead of just qualitatively account 00:11:15.430 --> 00:11:17.440 align:middle line:90% for everything. 00:11:17.440 --> 00:11:19.070 align:middle line:90% So what did Bresnahan do? 00:11:19.070 --> 00:11:21.610 align:middle line:84% And this is where the paper starts to look awfully primitive 00:11:21.610 --> 00:11:23.740 align:middle line:90% these days. 00:11:23.740 --> 00:11:27.700 align:middle line:84% He basically wrote down the standard vertical 00:11:27.700 --> 00:11:30.940 align:middle line:84% differentiation model that I described in lecture. 00:11:30.940 --> 00:11:36.580 align:middle line:84% So suppose there are n models with qualities v1 through vn, 00:11:36.580 --> 00:11:40.060 align:middle line:84% and suppose that the quality of a car 00:11:40.060 --> 00:11:43.900 align:middle line:84% is simply a function of observable characteristics. 00:11:43.900 --> 00:11:46.750 align:middle line:84% What function of observable characteristics? 00:11:46.750 --> 00:11:50.920 align:middle line:84% Basically it's just x times square root of x times beta, 00:11:50.920 --> 00:11:54.940 align:middle line:84% where x would contain things like the horsepower of the car, 00:11:54.940 --> 00:11:56.920 align:middle line:84% whether it had a V8 engine, the size 00:11:56.920 --> 00:11:59.110 align:middle line:84% in cubic inches, the weight of the car, 00:11:59.110 --> 00:12:02.890 align:middle line:84% whether it had air conditioning, things like that. 00:12:02.890 --> 00:12:05.050 align:middle line:84% He then assumes that there's, again, 00:12:05.050 --> 00:12:08.080 align:middle line:84% just like in the vertical differentiation model, 00:12:08.080 --> 00:12:11.500 align:middle line:84% there's a mass of consumers with types theta distributed 00:12:11.500 --> 00:12:14.140 align:middle line:84% uniform on some interval zero to vmax, 00:12:14.140 --> 00:12:18.170 align:middle line:84% and there's a density delta on that interval. 00:12:18.170 --> 00:12:20.450 align:middle line:84% And then he says that the utility of a type theta 00:12:20.450 --> 00:12:25.310 align:middle line:84% consumer from buying model i is going to be theta vi minus pi. 00:12:25.310 --> 00:12:27.320 align:middle line:90% So pure vertical model. 00:12:27.320 --> 00:12:30.980 align:middle line:84% And then he has the cost of producing q units at quality v 00:12:30.980 --> 00:12:34.460 align:middle line:84% is some constant or some function of v that's 00:12:34.460 --> 00:12:36.110 align:middle line:84% going to end up being irrelevant, 00:12:36.110 --> 00:12:39.880 align:middle line:90% and then q mu e to the v power. 00:12:39.880 --> 00:12:46.160 align:middle line:84% So costs, again-- v somehow is like the square root 00:12:46.160 --> 00:12:51.200 align:middle line:84% of the observable variables and then costs are exponentially 00:12:51.200 --> 00:12:54.290 align:middle line:84% increasing in v as you try to raise quality, 00:12:54.290 --> 00:12:56.180 align:middle line:90% and then there is-- 00:12:56.180 --> 00:12:58.310 align:middle line:90% it's a constant marginal cost. 00:12:58.310 --> 00:13:01.580 align:middle line:84% So mu times e to the v is the marginal cost 00:13:01.580 --> 00:13:02.815 align:middle line:90% of producing a single car. 00:13:02.815 --> 00:13:06.060 align:middle line:90% 00:13:06.060 --> 00:13:08.880 align:middle line:84% But then what Bresnahan's thought was, 00:13:08.880 --> 00:13:12.430 align:middle line:84% is that this is a model that I can fully specify. 00:13:12.430 --> 00:13:14.700 align:middle line:84% And if you fully specify the primitive-- if you fully 00:13:14.700 --> 00:13:17.760 align:middle line:84% specify the primitives, which is the production 00:13:17.760 --> 00:13:21.780 align:middle line:84% functions of the firms, the preferences of the consumers, 00:13:21.780 --> 00:13:25.440 align:middle line:84% and the model of firm behavior, then you 00:13:25.440 --> 00:13:27.540 align:middle line:84% can fully predict what's going to happen. 00:13:27.540 --> 00:13:30.930 align:middle line:84% So basically once you fix the parameters, beta mu 00:13:30.930 --> 00:13:33.510 align:middle line:84% and the production function and delta in vmax, 00:13:33.510 --> 00:13:37.110 align:middle line:84% in the consumer preferences, you can fully 00:13:37.110 --> 00:13:39.900 align:middle line:84% predict what's going to happen under any model of behavior. 00:13:39.900 --> 00:13:42.270 align:middle line:84% And so in particular, collusive pricing 00:13:42.270 --> 00:13:46.320 align:middle line:84% just becomes a multi-product monopoly model that we covered. 00:13:46.320 --> 00:13:50.520 align:middle line:84% Static competition would be another natural model, 00:13:50.520 --> 00:13:53.400 align:middle line:84% what people often apply to a situation like this. 00:13:53.400 --> 00:13:57.210 align:middle line:84% And then Bresnahan also considers two other models. 00:13:57.210 --> 00:14:01.260 align:middle line:84% One of these is like non-cooperative static 00:14:01.260 --> 00:14:03.010 align:middle line:84% competition between all products. 00:14:03.010 --> 00:14:05.220 align:middle line:84% So this is as if every Ford model 00:14:05.220 --> 00:14:08.380 align:middle line:84% is run by a different manager who's competing aggressively 00:14:08.380 --> 00:14:10.340 align:middle line:84% against every other Ford product. 00:14:10.340 --> 00:14:13.870 align:middle line:84% So every product in the market is 00:14:13.870 --> 00:14:17.860 align:middle line:84% run by a different manager who's profit maximizing solely 00:14:17.860 --> 00:14:19.220 align:middle line:90% based on that car. 00:14:19.220 --> 00:14:21.127 align:middle line:90% And then he's got a fourth one. 00:14:21.127 --> 00:14:23.710 align:middle line:84% This is one of those ones where it's based on somebody's paper 00:14:23.710 --> 00:14:25.900 align:middle line:84% from the 1970s, and I have a hard time-- 00:14:25.900 --> 00:14:27.460 align:middle line:84% it's sort of perfect competition. 00:14:27.460 --> 00:14:29.440 align:middle line:84% You have a hard time making out today what is that model 00:14:29.440 --> 00:14:31.232 align:middle line:84% and what would be the justification for it. 00:14:31.232 --> 00:14:32.740 align:middle line:84% But he's got four different models. 00:14:32.740 --> 00:14:34.615 align:middle line:84% He notes that under any of those four models, 00:14:34.615 --> 00:14:36.730 align:middle line:84% you can work out what should happen given 00:14:36.730 --> 00:14:40.450 align:middle line:84% any set of parameters, so then you can invert that relationship 00:14:40.450 --> 00:14:45.350 align:middle line:84% and say, given any one of these models 00:14:45.350 --> 00:14:47.720 align:middle line:84% and one of these models applied to any-- 00:14:47.720 --> 00:14:52.700 align:middle line:84% for any behavioral model in any year, 1954, '55, or '56, 00:14:52.700 --> 00:14:55.100 align:middle line:84% I can figure out what would be the parameters 00:14:55.100 --> 00:14:57.210 align:middle line:90% that best explain the data. 00:14:57.210 --> 00:15:02.020 align:middle line:84% And then Bresnahan's idea was for each model of firm behavior 00:15:02.020 --> 00:15:05.470 align:middle line:84% in each year, I figure out what parameters best explain the data 00:15:05.470 --> 00:15:07.770 align:middle line:84% and then ask how well does that model explain 00:15:07.770 --> 00:15:09.840 align:middle line:90% the data with those parameters? 00:15:09.840 --> 00:15:12.060 align:middle line:84% And then I pick out whichever model 00:15:12.060 --> 00:15:17.010 align:middle line:84% fits best in every year to say this is how the firms behaved 00:15:17.010 --> 00:15:17.660 align:middle line:90% in that year. 00:15:17.660 --> 00:15:20.226 align:middle line:90% 00:15:20.226 --> 00:15:23.160 align:middle line:84% And the way he does this is-- again, this seems archaic, 00:15:23.160 --> 00:15:27.960 align:middle line:84% but it's a set of non-nested hypothesis tests where he looks 00:15:27.960 --> 00:15:29.880 align:middle line:90% at how well the-- 00:15:29.880 --> 00:15:32.110 align:middle line:84% looks at the model, fits the model to the data, 00:15:32.110 --> 00:15:34.380 align:middle line:84% and then, in a likelihood ratio sense, says, 00:15:34.380 --> 00:15:37.980 align:middle line:84% do the predictions of the other model help explain the errors? 00:15:37.980 --> 00:15:39.690 align:middle line:84% Do the predictions of model two help 00:15:39.690 --> 00:15:41.650 align:middle line:90% explain the errors in model one? 00:15:41.650 --> 00:15:43.320 align:middle line:84% And if predictions of model two help 00:15:43.320 --> 00:15:47.370 align:middle line:84% explain the errors of model one, then model one 00:15:47.370 --> 00:15:49.230 align:middle line:84% is rejected by model two and model one 00:15:49.230 --> 00:15:50.740 align:middle line:90% is not the correct model. 00:15:50.740 --> 00:15:54.300 align:middle line:84% And so what he looks for is, is there one of these four 00:15:54.300 --> 00:15:56.220 align:middle line:84% models-- in each year, year by year, 00:15:56.220 --> 00:15:57.870 align:middle line:84% is there one of these four models 00:15:57.870 --> 00:16:01.736 align:middle line:84% that's not rejected by the other model? 00:16:01.736 --> 00:16:05.070 align:middle line:84% And then the other thing that Bresnahan really also pioneered, 00:16:05.070 --> 00:16:06.930 align:middle line:84% and this was a paper in his thesis, 00:16:06.930 --> 00:16:09.070 align:middle line:84% was the idea of a counterfactual. 00:16:09.070 --> 00:16:12.600 align:middle line:84% His thought was that if we do this 00:16:12.600 --> 00:16:15.150 align:middle line:84% and we figure out what is the correct model in each year, 00:16:15.150 --> 00:16:17.610 align:middle line:84% then I can do a counterfactual and predict 00:16:17.610 --> 00:16:19.990 align:middle line:84% what would have happened under any other model 00:16:19.990 --> 00:16:21.620 align:middle line:90% or any other government policy. 00:16:21.620 --> 00:16:24.760 align:middle line:84% I could tell you what if the government had had a 5% sales 00:16:24.760 --> 00:16:28.360 align:middle line:84% tax on cars, what would happen if the firms had colluded 00:16:28.360 --> 00:16:31.300 align:middle line:84% instead of competed with each other, what would have happened 00:16:31.300 --> 00:16:31.990 align:middle line:90% in-- 00:16:31.990 --> 00:16:34.960 align:middle line:84% any other thing change you can think of to the industry, 00:16:34.960 --> 00:16:38.590 align:middle line:84% you can predict what would have happened under his model, 00:16:38.590 --> 00:16:40.910 align:middle line:84% and it gives you the counterfactual prediction. 00:16:40.910 --> 00:16:45.890 align:middle line:90% 00:16:45.890 --> 00:16:50.060 align:middle line:84% OK, so what are the main applied results of the paper? 00:16:50.060 --> 00:16:55.760 align:middle line:84% And I always think of it like maybe someday, someone 00:16:55.760 --> 00:16:57.175 align:middle line:90% will rewrite this paper. 00:16:57.175 --> 00:16:58.550 align:middle line:84% Obviously this is primitive data, 00:16:58.550 --> 00:17:00.060 align:middle line:84% like minimal data from the 1950s. 00:17:00.060 --> 00:17:01.602 align:middle line:84% Maybe someone will rewrite this paper 00:17:01.602 --> 00:17:04.069 align:middle line:84% and say, from a more modern perspective, what 00:17:04.069 --> 00:17:05.339 align:middle line:90% do we think actually happened? 00:17:05.339 --> 00:17:07.770 align:middle line:90% But what does this analysis say? 00:17:07.770 --> 00:17:10.790 align:middle line:84% First one says that the collusive model fits the data 00:17:10.790 --> 00:17:15.890 align:middle line:84% best in 1954 and 1956, and the static competition model 00:17:15.890 --> 00:17:18.960 align:middle line:90% fits the data best in 1955. 00:17:18.960 --> 00:17:21.060 align:middle line:84% So that's his evidence for his main hypothesis 00:17:21.060 --> 00:17:23.339 align:middle line:84% that there was this breakdown in collusion 00:17:23.339 --> 00:17:26.810 align:middle line:84% and the collusive model fits the patterns best in '54, '56, 00:17:26.810 --> 00:17:29.640 align:middle line:90% the competitive model in '55. 00:17:29.640 --> 00:17:34.320 align:middle line:84% Further supporting that is his model has a bunch of parameters, 00:17:34.320 --> 00:17:36.270 align:middle line:84% the mu and the delta and the vmax, 00:17:36.270 --> 00:17:38.520 align:middle line:84% and especially the betas that talk about-- 00:17:38.520 --> 00:17:42.720 align:middle line:84% which relate to how valuable is horsepower, how valuable 00:17:42.720 --> 00:17:45.810 align:middle line:84% is air conditioning, things like that. 00:17:45.810 --> 00:17:50.010 align:middle line:84% What he notes is that if you estimate the collusive model 00:17:50.010 --> 00:17:53.850 align:middle line:84% in '54 and '56 and then estimate the static competition 00:17:53.850 --> 00:17:57.780 align:middle line:84% model in '55, the parameters like the betas 00:17:57.780 --> 00:18:01.930 align:middle line:84% turn out to be fairly similar between '54, '56, and '55. 00:18:01.930 --> 00:18:03.840 align:middle line:84% And that's further supporting evidence 00:18:03.840 --> 00:18:06.810 align:middle line:84% that consumer preferences towards horsepower 00:18:06.810 --> 00:18:09.500 align:middle line:84% should probably be fairly stable over time. 00:18:09.500 --> 00:18:13.650 align:middle line:84% And so what he notes is that if you do this model in these two 00:18:13.650 --> 00:18:15.660 align:middle line:84% years and the other model in '55, 00:18:15.660 --> 00:18:17.880 align:middle line:84% then the preference parameters are fairly 00:18:17.880 --> 00:18:20.140 align:middle line:90% similar across the years. 00:18:20.140 --> 00:18:22.900 align:middle line:84% Whereas if you try to fix any one model 00:18:22.900 --> 00:18:25.690 align:middle line:84% and use any one model to explain what happens, 00:18:25.690 --> 00:18:28.840 align:middle line:84% the parameters need to change a lot from '54 to '55 00:18:28.840 --> 00:18:31.090 align:middle line:90% and from '55 to '56. 00:18:31.090 --> 00:18:32.170 align:middle line:90% In some sense, you need-- 00:18:32.170 --> 00:18:34.150 align:middle line:84% prices did go down despite the fact 00:18:34.150 --> 00:18:36.400 align:middle line:84% that demand is going way up, and so you need something 00:18:36.400 --> 00:18:40.660 align:middle line:84% to say that suddenly cars got a lot cheaper-- 00:18:40.660 --> 00:18:42.040 align:middle line:84% cars got a lot cheaper to produce 00:18:42.040 --> 00:18:45.100 align:middle line:84% and people valued cars more than they used to value cars 00:18:45.100 --> 00:18:46.460 align:middle line:90% during that one year. 00:18:46.460 --> 00:18:50.260 align:middle line:84% Whereas if you use this change in behavior, 00:18:50.260 --> 00:18:53.140 align:middle line:84% then the parameter estimates are fairly stable across time. 00:18:53.140 --> 00:18:55.430 align:middle line:84% And so that's supporting evidence. 00:18:55.430 --> 00:18:58.280 align:middle line:90% 00:18:58.280 --> 00:19:00.590 align:middle line:84% Then what's in his counterfactual? 00:19:00.590 --> 00:19:03.530 align:middle line:84% The counterfactual he does is suppose 00:19:03.530 --> 00:19:06.380 align:middle line:84% firms had colluded in 1955 instead of 00:19:06.380 --> 00:19:09.380 align:middle line:84% engaging in differentiated product competition. 00:19:09.380 --> 00:19:14.510 align:middle line:84% And what he finds is that prices would have been 6% higher 00:19:14.510 --> 00:19:17.870 align:middle line:84% and demand would have been 11% lower. 00:19:17.870 --> 00:19:23.300 align:middle line:84% So this was a 40% jump in auto sales. 00:19:23.300 --> 00:19:25.860 align:middle line:84% His model is not fully explaining it. 00:19:25.860 --> 00:19:29.990 align:middle line:84% His model is saying, I'm explaining 00:19:29.990 --> 00:19:33.110 align:middle line:84% a little more than 25% of the boom in auto sales 00:19:33.110 --> 00:19:38.097 align:middle line:84% as coming from the change in collusion. 00:19:38.097 --> 00:19:39.680 align:middle line:84% So in some sense, he's already letting 00:19:39.680 --> 00:19:41.030 align:middle line:90% you know that his model-- 00:19:41.030 --> 00:19:42.830 align:middle line:84% at least the model as he's estimated-- 00:19:42.830 --> 00:19:45.353 align:middle line:84% isn't fully correct, because it's not fully accounting 00:19:45.353 --> 00:19:46.270 align:middle line:90% for the jump in sales. 00:19:46.270 --> 00:19:50.450 align:middle line:84% Now, there was a 10% jump in non-auto durable goods sales. 00:19:50.450 --> 00:19:54.440 align:middle line:84% You would have said just the background change in US economy 00:19:54.440 --> 00:19:58.470 align:middle line:84% income or change in income maybe could have explained 10% also. 00:19:58.470 --> 00:20:00.330 align:middle line:84% So he's saying his model is explaining 00:20:00.330 --> 00:20:04.230 align:middle line:90% part of what's going on. 00:20:04.230 --> 00:20:08.910 align:middle line:84% But it is a very nice use of the counterfactual 00:20:08.910 --> 00:20:14.040 align:middle line:84% to give us a sense of what would have happened in another world. 00:20:14.040 --> 00:20:17.430 align:middle line:84% You see every IO paper now has counterfactuals, 00:20:17.430 --> 00:20:18.665 align:middle line:90% which I think started here. 00:20:18.665 --> 00:20:23.030 align:middle line:90% 00:20:23.030 --> 00:20:26.563 align:middle line:84% As I said, if you read the paper, 00:20:26.563 --> 00:20:27.980 align:middle line:84% there are all kinds of things that 00:20:27.980 --> 00:20:30.680 align:middle line:84% are going to seem odd to you, like the adoption 00:20:30.680 --> 00:20:32.440 align:middle line:90% of the pure vertical model. 00:20:32.440 --> 00:20:38.862 align:middle line:84% Clearly, I don't think the Mercury Montclair 00:20:38.862 --> 00:20:40.820 align:middle line:84% and the Chevy Bel Air are so similar that there 00:20:40.820 --> 00:20:43.250 align:middle line:84% would be purely Bertrand competition between them. 00:20:43.250 --> 00:20:46.310 align:middle line:84% If you thought about doing that in today's car market, 00:20:46.310 --> 00:20:48.290 align:middle line:90% there are pickup trucks. 00:20:48.290 --> 00:20:49.940 align:middle line:90% There are convertibles. 00:20:49.940 --> 00:20:51.020 align:middle line:90% There are sports sedans. 00:20:51.020 --> 00:20:54.230 align:middle line:90% There are electric vehicles. 00:20:54.230 --> 00:20:57.080 align:middle line:84% If you try to put them on a one dimensional quality space, 00:20:57.080 --> 00:20:59.960 align:middle line:84% some of them, some pickup truck, is going to nearly coincide 00:20:59.960 --> 00:21:01.730 align:middle line:90% with a Honda Accord. 00:21:01.730 --> 00:21:04.820 align:middle line:84% Clearly, you need to have some kind 00:21:04.820 --> 00:21:06.560 align:middle line:84% of horizontal differentiation as well 00:21:06.560 --> 00:21:08.600 align:middle line:90% as vertical differentiation. 00:21:08.600 --> 00:21:11.420 align:middle line:90% 00:21:11.420 --> 00:21:13.880 align:middle line:84% Anyway, a number of aspects when you 00:21:13.880 --> 00:21:15.890 align:middle line:84% see how people do modern empirical 00:21:15.890 --> 00:21:17.520 align:middle line:90% work that look very different. 00:21:17.520 --> 00:21:22.260 align:middle line:84% But I think in some ways, the basic ideas still 00:21:22.260 --> 00:21:23.330 align:middle line:90% seem very spot on. 00:21:23.330 --> 00:21:24.747 align:middle line:84% Particularly, the one thing that I 00:21:24.747 --> 00:21:27.900 align:middle line:84% think he does better than many current models is 00:21:27.900 --> 00:21:30.780 align:middle line:84% he does take very seriously this idea of, 00:21:30.780 --> 00:21:33.090 align:middle line:84% I want this empirical-- this structural work 00:21:33.090 --> 00:21:35.940 align:middle line:84% only works if I choose the right framework, 00:21:35.940 --> 00:21:38.580 align:middle line:84% and he's going to come up with four different models 00:21:38.580 --> 00:21:40.380 align:middle line:84% and put a lot of effort into testing 00:21:40.380 --> 00:21:42.790 align:middle line:84% the four different structural models against each other. 00:21:42.790 --> 00:21:45.300 align:middle line:84% Whereas now very often it's like it's 00:21:45.300 --> 00:21:47.370 align:middle line:84% hard to estimate even any one model, so you see 00:21:47.370 --> 00:21:49.050 align:middle line:84% a lot of papers that have, here's my one model, 00:21:49.050 --> 00:21:50.520 align:middle line:84% and I can estimate that and I can 00:21:50.520 --> 00:21:52.800 align:middle line:84% talk about how well it fits by putting up a graph 00:21:52.800 --> 00:21:55.380 align:middle line:84% and showing some x's and y's and show the y's are 00:21:55.380 --> 00:21:56.470 align:middle line:90% close to the y-hats. 00:21:56.470 --> 00:22:01.920 align:middle line:84% But I think Bresnahan's doing a very nice job on that dimension. 00:22:01.920 --> 00:22:06.470 align:middle line:90% So any questions on Bresnahan? 00:22:06.470 --> 00:22:07.250 align:middle line:90% Yeah? 00:22:07.250 --> 00:22:10.457 align:middle line:84% AUDIENCE: Why would collusion break down 00:22:10.457 --> 00:22:13.040 align:middle line:84% for just one year and then all of a sudden, the next year it's 00:22:13.040 --> 00:22:13.977 align:middle line:90% back at it again? 00:22:13.977 --> 00:22:14.810 align:middle line:90% GLENN ELLISON: Yeah. 00:22:14.810 --> 00:22:18.910 align:middle line:90% 00:22:18.910 --> 00:22:21.935 align:middle line:90% You know, a little hard to say. 00:22:21.935 --> 00:22:24.860 align:middle line:84% I will, in a couple weeks, be discussing models 00:22:24.860 --> 00:22:26.960 align:middle line:90% of repeated game collusion. 00:22:26.960 --> 00:22:29.438 align:middle line:84% And in models of repeated game collusion, 00:22:29.438 --> 00:22:30.980 align:middle line:84% you do get this phenomenon of the way 00:22:30.980 --> 00:22:34.640 align:middle line:84% you support collusion is via price wars 00:22:34.640 --> 00:22:37.973 align:middle line:84% to punish somebody who cheats so that as if you're trying 00:22:37.973 --> 00:22:39.890 align:middle line:84% to collude, you think that someone is cheating 00:22:39.890 --> 00:22:43.400 align:middle line:84% on the collusive agreements and their dealers are charging-- 00:22:43.400 --> 00:22:45.950 align:middle line:84% you've all agreed that we're going to charge $7,000 00:22:45.950 --> 00:22:48.200 align:middle line:84% for our Bel Air and Montclair, and you think the other 00:22:48.200 --> 00:22:50.540 align:middle line:84% guy's dealers are actually giving consumers cuts 00:22:50.540 --> 00:22:54.290 align:middle line:84% and offering it for $6,500 instead of $7,000 even though 00:22:54.290 --> 00:22:55.610 align:middle line:90% the list price is $7,000. 00:22:55.610 --> 00:23:00.500 align:middle line:84% You might decide we're going to have to punish you 00:23:00.500 --> 00:23:03.490 align:middle line:84% for doing that and we're going to abandon colluding this year, 00:23:03.490 --> 00:23:05.240 align:middle line:84% and we're going to all learn how bad it is 00:23:05.240 --> 00:23:09.328 align:middle line:84% and then go back to cooperating the next year. 00:23:09.328 --> 00:23:11.370 align:middle line:84% You know, so you could tell some story like that. 00:23:11.370 --> 00:23:15.190 align:middle line:90% 00:23:15.190 --> 00:23:17.498 align:middle line:84% There's no discussion in the paper of why collusion 00:23:17.498 --> 00:23:18.540 align:middle line:90% would have brought down-- 00:23:18.540 --> 00:23:20.680 align:middle line:84% there's no narrative evidence at all 00:23:20.680 --> 00:23:23.480 align:middle line:84% of the executives talking to each other 00:23:23.480 --> 00:23:25.130 align:middle line:90% and having anything going wrong. 00:23:25.130 --> 00:23:26.470 align:middle line:90% But it is true that actually-- 00:23:26.470 --> 00:23:28.702 align:middle line:90% 00:23:28.702 --> 00:23:30.910 align:middle line:84% we're going to study several other papers on cartels. 00:23:30.910 --> 00:23:33.595 align:middle line:84% Almost all the cartels we study are finite lived. 00:23:33.595 --> 00:23:35.470 align:middle line:84% They go on, they're colluding with each other 00:23:35.470 --> 00:23:36.887 align:middle line:84% for some number of years, and then 00:23:36.887 --> 00:23:39.340 align:middle line:90% it breaks down at some point. 00:23:39.340 --> 00:23:42.370 align:middle line:84% In part because also it's illegal, but it's-- 00:23:42.370 --> 00:23:45.100 align:middle line:84% so people are afraid of getting caught or something goes wrong 00:23:45.100 --> 00:23:47.683 align:middle line:84% and they can't discuss it with each other and get things back. 00:23:47.683 --> 00:23:51.830 align:middle line:90% 00:23:51.830 --> 00:23:52.380 align:middle line:90% Anybody else? 00:23:52.380 --> 00:23:56.290 align:middle line:90% 00:23:56.290 --> 00:23:57.130 align:middle line:90% OK. 00:23:57.130 --> 00:23:58.908 align:middle line:84% So second thing I was going to do, then, 00:23:58.908 --> 00:24:01.450 align:middle line:84% I was going to jump ahead before I get to Miller and Weinberg 00:24:01.450 --> 00:24:05.470 align:middle line:84% and just do a half hour, 40 minutes on oligopoly price 00:24:05.470 --> 00:24:06.710 align:middle line:90% discrimination. 00:24:06.710 --> 00:24:11.590 align:middle line:84% And I think in Tirole's textbook and in many other places, 00:24:11.590 --> 00:24:13.090 align:middle line:84% price discrimination is typically 00:24:13.090 --> 00:24:15.610 align:middle line:90% taught as a monopoly topic. 00:24:15.610 --> 00:24:18.520 align:middle line:84% And I think that was because historically you 00:24:18.520 --> 00:24:21.520 align:middle line:84% had monopoly and perfect competition as the two models. 00:24:21.520 --> 00:24:24.470 align:middle line:84% With perfect competition, price equals marginal cost, 00:24:24.470 --> 00:24:26.270 align:middle line:90% you can't price discriminate. 00:24:26.270 --> 00:24:28.600 align:middle line:84% But with imperfect competition, price 00:24:28.600 --> 00:24:29.930 align:middle line:90% isn't equal to marginal cost. 00:24:29.930 --> 00:24:32.050 align:middle line:84% There's no reason why oligopolies 00:24:32.050 --> 00:24:34.240 align:middle line:84% can't price discriminate just like monopolies 00:24:34.240 --> 00:24:35.650 align:middle line:90% price discriminate. 00:24:35.650 --> 00:24:39.460 align:middle line:84% And particularly, there was an early paper in this, Borenstein 00:24:39.460 --> 00:24:43.540 align:middle line:84% and Rose, 1994, JPE, look at price dispersion 00:24:43.540 --> 00:24:45.140 align:middle line:90% in airline markets. 00:24:45.140 --> 00:24:49.570 align:middle line:84% And what they find is that if you look at airline city pair 00:24:49.570 --> 00:24:52.720 align:middle line:84% markets, like Boston to New York, New York to Phoenix, 00:24:52.720 --> 00:24:56.780 align:middle line:84% whatever, the markets that have more competition 00:24:56.780 --> 00:25:00.690 align:middle line:84% tend to have more price dispersion on the same plane. 00:25:00.690 --> 00:25:03.092 align:middle line:84% So if you look at a flight from Boston to New York, 00:25:03.092 --> 00:25:04.800 align:middle line:84% there are a lot of people on that flight. 00:25:04.800 --> 00:25:06.480 align:middle line:90% They all pay different prices. 00:25:06.480 --> 00:25:08.150 align:middle line:84% And the Borenstein-Rose finding is 00:25:08.150 --> 00:25:11.690 align:middle line:84% that the heterogeneity in prices on a flight 00:25:11.690 --> 00:25:14.240 align:middle line:84% is actually larger in the more competitive markets, 00:25:14.240 --> 00:25:17.120 align:middle line:84% not smaller in the more competitive markets. 00:25:17.120 --> 00:25:19.290 align:middle line:84% So that's an indication that it's not the case 00:25:19.290 --> 00:25:21.650 align:middle line:84% that competition that we often see 00:25:21.650 --> 00:25:25.340 align:middle line:84% makes it impossible to price discriminate. 00:25:25.340 --> 00:25:27.810 align:middle line:84% OK, so how would we model price discrimination? 00:25:27.810 --> 00:25:30.320 align:middle line:84% Well, let me first start with third degree price 00:25:30.320 --> 00:25:31.550 align:middle line:90% discrimination. 00:25:31.550 --> 00:25:36.330 align:middle line:84% So you have two firms competing in two separate markets, 00:25:36.330 --> 00:25:39.030 align:middle line:84% and suppose there's no cross market arbitrage. 00:25:39.030 --> 00:25:44.460 align:middle line:84% So this is two firms selling cell phone service in Boston 00:25:44.460 --> 00:25:46.310 align:middle line:84% to the same two firms selling-- or two firms 00:25:46.310 --> 00:25:48.200 align:middle line:84% selling cable TV service in Boston 00:25:48.200 --> 00:25:52.580 align:middle line:84% and selling cable TV service in Philadelphia. 00:25:52.580 --> 00:25:56.180 align:middle line:84% You can't get buy cable TV in Boston 00:25:56.180 --> 00:25:58.390 align:middle line:84% and then use it in Philadelphia or vice versa. 00:25:58.390 --> 00:26:02.270 align:middle line:84% You have to buy the service at your service address. 00:26:02.270 --> 00:26:04.850 align:middle line:84% And so you could have these same two firms competing 00:26:04.850 --> 00:26:07.058 align:middle line:84% and they're competing in these two different markets. 00:26:07.058 --> 00:26:09.466 align:middle line:90% 00:26:09.466 --> 00:26:13.130 align:middle line:84% So suppose that each of the markets has Hotelling 00:26:13.130 --> 00:26:17.120 align:middle line:84% preferences and the type theta consumers in market m get 00:26:17.120 --> 00:26:22.880 align:middle line:84% utility v minus p price of firm one in m minus tm theta if they 00:26:22.880 --> 00:26:27.540 align:middle line:84% buy from firm one and v minus p 2m minus tm1 minus theta if they 00:26:27.540 --> 00:26:28.640 align:middle line:90% buy it from firm two. 00:26:28.640 --> 00:26:32.253 align:middle line:84% So this is just competition on two lines. 00:26:32.253 --> 00:26:36.085 align:middle line:90% 00:26:36.085 --> 00:26:40.530 align:middle line:84% This is market one and it has some parameter t1, 00:26:40.530 --> 00:26:48.760 align:middle line:84% and this is market two and it has some parameter t2. 00:26:48.760 --> 00:26:51.800 align:middle line:90% I think I made t2 the high guys. 00:26:51.800 --> 00:26:53.488 align:middle line:90% OK, so you just have-- 00:26:53.488 --> 00:26:56.030 align:middle line:84% you're competing on this line, you're competing on this line. 00:26:56.030 --> 00:26:57.950 align:middle line:84% These markets are completely separate. 00:26:57.950 --> 00:27:01.750 align:middle line:90% 00:27:01.750 --> 00:27:05.240 align:middle line:84% Price discrimination is just like third degree in monopoly. 00:27:05.240 --> 00:27:07.280 align:middle line:84% It's just two totally separate markets. 00:27:07.280 --> 00:27:10.150 align:middle line:84% So if we have competition here, what's going to happen 00:27:10.150 --> 00:27:13.930 align:middle line:84% is you're going to get p star equals c plus t1. 00:27:13.930 --> 00:27:15.340 align:middle line:84% You have competition here, you're 00:27:15.340 --> 00:27:21.550 align:middle line:84% going to get p2 star is c plus t2. 00:27:21.550 --> 00:27:23.420 align:middle line:90% We just compete on a line here. 00:27:23.420 --> 00:27:25.990 align:middle line:84% We get Hotelling here, we compete on this line, 00:27:25.990 --> 00:27:28.000 align:middle line:90% we get Hotelling here. 00:27:28.000 --> 00:27:30.620 align:middle line:84% The prices are different across markets. 00:27:30.620 --> 00:27:33.910 align:middle line:84% We're discriminating against consumers in this market. 00:27:33.910 --> 00:27:36.050 align:middle line:84% The high types end up paying more. 00:27:36.050 --> 00:27:37.760 align:middle line:90% What are high types? 00:27:37.760 --> 00:27:41.560 align:middle line:84% A high type is somebody who lives in a market where there's 00:27:41.560 --> 00:27:43.870 align:middle line:90% more heterogeneity in taste. 00:27:43.870 --> 00:27:46.300 align:middle line:84% So if there's more heterogeneity in taste in market two 00:27:46.300 --> 00:27:48.760 align:middle line:84% than in market one, then people in market two 00:27:48.760 --> 00:27:51.730 align:middle line:84% end up paying more than people in market one. 00:27:51.730 --> 00:27:53.880 align:middle line:90% OK. 00:27:53.880 --> 00:27:57.030 align:middle line:84% The one policy question you get is, well, 00:27:57.030 --> 00:27:58.680 align:middle line:84% what would happen in a market like this 00:27:58.680 --> 00:28:02.850 align:middle line:90% if we banned people from-- 00:28:02.850 --> 00:28:07.050 align:middle line:84% what would happen if we ban people from price discriminating 00:28:07.050 --> 00:28:10.740 align:middle line:84% and say, you have to charge the same price in both markets? 00:28:10.740 --> 00:28:14.450 align:middle line:84% It turns out, in this model you just get an average markup. 00:28:14.450 --> 00:28:18.930 align:middle line:84% So if instead of being allowed to price separately, 00:28:18.930 --> 00:28:20.800 align:middle line:84% you force them to charge the same price, 00:28:20.800 --> 00:28:29.460 align:middle line:84% what you get is price equals c plus 2t1t2 over t1 plus t2. 00:28:29.460 --> 00:28:35.040 align:middle line:84% This is what mathematicians refer to as the harmonic 00:28:35.040 --> 00:28:41.970 align:middle line:90% mean of t1 and t2. 00:28:41.970 --> 00:28:46.700 align:middle line:84% And basic math fact is that if t1 is less than t2, 00:28:46.700 --> 00:28:55.880 align:middle line:84% then t1 is less than 2t1t2 over t1 plus t2, 00:28:55.880 --> 00:29:01.490 align:middle line:84% which is less than or equal to t1 plus t2 over 2 less than 00:29:01.490 --> 00:29:03.260 align:middle line:90% or equal to t2. 00:29:03.260 --> 00:29:07.940 align:middle line:84% This is called the arithmetic harmonic mean inequality. 00:29:07.940 --> 00:29:10.700 align:middle line:84% You can think of this as this is the reciprocal 00:29:10.700 --> 00:29:13.730 align:middle line:84% of the average of the reciprocals of t1 and t2. 00:29:13.730 --> 00:29:16.890 align:middle line:84% So it always lies between t1 and t2, 00:29:16.890 --> 00:29:20.100 align:middle line:84% and it is less than the arithmetic mean of t1 and t2. 00:29:20.100 --> 00:29:22.337 align:middle line:84% But just the way the math works, this 00:29:22.337 --> 00:29:24.170 align:middle line:84% is the price that you end up charging if you 00:29:24.170 --> 00:29:26.240 align:middle line:90% ban price discrimination. 00:29:26.240 --> 00:29:29.670 align:middle line:84% So what does banning price discrimination do? 00:29:29.670 --> 00:29:33.140 align:middle line:84% Typically, it makes the high types better off. 00:29:33.140 --> 00:29:38.490 align:middle line:84% Typically it makes the low types worse off. 00:29:38.490 --> 00:29:41.840 align:middle line:84% It does eliminate a misallocation of markets. 00:29:41.840 --> 00:29:44.300 align:middle line:84% In many models, if you have a high price in one market 00:29:44.300 --> 00:29:46.010 align:middle line:84% and a low price in another market, 00:29:46.010 --> 00:29:47.690 align:middle line:84% there are people in the high market who 00:29:47.690 --> 00:29:49.590 align:middle line:84% would like the good who aren't getting it 00:29:49.590 --> 00:29:51.410 align:middle line:84% and there are people in the low market who 00:29:51.410 --> 00:29:53.410 align:middle line:84% don't like the good as much who are getting it. 00:29:53.410 --> 00:29:55.910 align:middle line:84% That's a misallocation like there is in price discrimination 00:29:55.910 --> 00:29:56.960 align:middle line:90% models. 00:29:56.960 --> 00:30:00.650 align:middle line:84% So banning price discrimination eliminates that inefficiency. 00:30:00.650 --> 00:30:03.410 align:middle line:84% That inefficiency isn't present in Hotelling 00:30:03.410 --> 00:30:05.180 align:middle line:84% because in Hotelling everybody buys, 00:30:05.180 --> 00:30:08.260 align:middle line:90% at least if he is big enough. 00:30:08.260 --> 00:30:13.900 align:middle line:84% And then welfare can go either way. 00:30:13.900 --> 00:30:19.235 align:middle line:90% 00:30:19.235 --> 00:30:21.180 align:middle line:84% Here, again, there are no welfare changes. 00:30:21.180 --> 00:30:23.630 align:middle line:84% It's just a transfer because everybody buys-- 00:30:23.630 --> 00:30:25.340 align:middle line:84% all efficient transactions take place. 00:30:25.340 --> 00:30:28.220 align:middle line:84% First best is achieved in both models, 00:30:28.220 --> 00:30:29.820 align:middle line:90% so it's just a pure transfer. 00:30:29.820 --> 00:30:32.360 align:middle line:84% But in general, banning price discrimination 00:30:32.360 --> 00:30:35.630 align:middle line:84% can increase total social welfare 00:30:35.630 --> 00:30:37.450 align:middle line:84% or it can decrease total social welfare. 00:30:37.450 --> 00:30:40.800 align:middle line:90% 00:30:40.800 --> 00:30:42.090 align:middle line:90% OK. 00:30:42.090 --> 00:30:46.590 align:middle line:84% There's a set of papers in the 1980s noting that this is 00:30:46.590 --> 00:30:48.960 align:middle line:84% the typical effect of banning price discrimination 00:30:48.960 --> 00:30:52.860 align:middle line:84% in oligopoly markets, but the typical effects don't have 00:30:52.860 --> 00:30:54.360 align:middle line:90% to hold. 00:30:54.360 --> 00:30:57.390 align:middle line:84% So there's a paper by Thisse and Vives, AER '88, 00:30:57.390 --> 00:31:01.110 align:middle line:84% that makes the point that allowing price discrimination 00:31:01.110 --> 00:31:04.140 align:middle line:84% can make all consumers better off. 00:31:04.140 --> 00:31:08.010 align:middle line:84% So price discrimination can make everybody better off because 00:31:08.010 --> 00:31:10.530 align:middle line:84% of this effect that Thisse and Vives highlight. 00:31:10.530 --> 00:31:13.120 align:middle line:84% When you let firms price discriminate, 00:31:13.120 --> 00:31:16.980 align:middle line:84% it lets firms aggressively target a rival's 00:31:16.980 --> 00:31:22.260 align:middle line:84% natural consumers, and that can intensify price competition. 00:31:22.260 --> 00:31:25.500 align:middle line:84% So the simplest example that Thisse and Vives gave 00:31:25.500 --> 00:31:28.920 align:middle line:84% was consider again the Hotelling model. 00:31:28.920 --> 00:31:31.020 align:middle line:84% And I'm going to give the Hotelling model 00:31:31.020 --> 00:31:32.820 align:middle line:90% a different interpretation. 00:31:32.820 --> 00:31:38.370 align:middle line:84% So we have these two firms and they're competing, 00:31:38.370 --> 00:31:42.280 align:middle line:84% and the consumers have some types theta. 00:31:42.280 --> 00:31:48.550 align:middle line:84% And I'm going to assume that the firms can observe theta. 00:31:48.550 --> 00:31:53.380 align:middle line:84% So maybe these are two firms selling lumber to home builders 00:31:53.380 --> 00:31:56.130 align:middle line:84% and they know where the home builder's jobsite is, 00:31:56.130 --> 00:31:58.990 align:middle line:84% and so they know how long it's going to take the home 00:31:58.990 --> 00:32:01.930 align:middle line:84% builder to truck the lumber from their warehouse 00:32:01.930 --> 00:32:04.960 align:middle line:84% to his factory site so they know his theta. 00:32:04.960 --> 00:32:06.460 align:middle line:84% The theta is determining willingness 00:32:06.460 --> 00:32:10.900 align:middle line:84% to pay through some means that's known to the sellers. 00:32:10.900 --> 00:32:15.160 align:middle line:84% And so here, discrimination would 00:32:15.160 --> 00:32:18.610 align:middle line:84% be firms set theta dependent prices instead 00:32:18.610 --> 00:32:21.890 align:middle line:84% of theta independent prices because they observe theta. 00:32:21.890 --> 00:32:22.390 align:middle line:90% OK. 00:32:22.390 --> 00:32:24.400 align:middle line:84% Well, first, what happens in this model 00:32:24.400 --> 00:32:29.530 align:middle line:84% if firms are-- if they're not allowed to price discriminate, 00:32:29.530 --> 00:32:33.700 align:middle line:84% then you just get p star equals c plus t. 00:32:33.700 --> 00:32:37.342 align:middle line:84% Because if I know the locations, but I'm not 00:32:37.342 --> 00:32:38.800 align:middle line:84% allowed to charge different prices, 00:32:38.800 --> 00:32:42.730 align:middle line:84% then I just charge c plus t to everybody. 00:32:42.730 --> 00:32:47.390 align:middle line:84% What's going to happen if I now start discriminating? 00:32:47.390 --> 00:32:52.510 align:middle line:84% Well, if I'm firm one or if-- let's say I'm firm two 00:32:52.510 --> 00:32:55.600 align:middle line:84% and I'm trying to sell to a consumer over here. 00:32:55.600 --> 00:32:59.020 align:middle line:84% I know that that consumer prefers firm one, 00:32:59.020 --> 00:33:01.283 align:middle line:84% prefers firm one's product to my product. 00:33:01.283 --> 00:33:04.420 align:middle line:90% 00:33:04.420 --> 00:33:07.360 align:middle line:84% To get that consumer, I'm going to have to undercut firm one. 00:33:07.360 --> 00:33:10.630 align:middle line:90% 00:33:10.630 --> 00:33:12.250 align:middle line:84% In pricing to this consumer, there's 00:33:12.250 --> 00:33:14.500 align:middle line:90% no product differentiation left. 00:33:14.500 --> 00:33:17.410 align:middle line:84% We know this consumer has some value v1 for firm one's product 00:33:17.410 --> 00:33:19.880 align:middle line:84% and some value v2 for firm two's product, 00:33:19.880 --> 00:33:22.780 align:middle line:84% and we're just pricing to that consumer like in a Bertrand 00:33:22.780 --> 00:33:24.155 align:middle line:90% competition model. 00:33:24.155 --> 00:33:26.530 align:middle line:84% So what's going to happen is Bertrand competition is just 00:33:26.530 --> 00:33:37.540 align:middle line:84% going to force prices down until p2 star equals c and p1 star 00:33:37.540 --> 00:33:45.340 align:middle line:84% is going to be-- so p2 star of theta is c and p1 star of theta 00:33:45.340 --> 00:33:47.590 align:middle line:84% is going to be firm one's advantage, which 00:33:47.590 --> 00:33:51.630 align:middle line:84% is negative theta minus 1 minus theta. 00:33:51.630 --> 00:33:58.030 align:middle line:90% 00:33:58.030 --> 00:33:59.710 align:middle line:90% Minus 1 minus theta. 00:33:59.710 --> 00:34:03.470 align:middle line:90% So that's 1 minus c plus. 00:34:03.470 --> 00:34:07.090 align:middle line:90% So c plus 1 minus 2 theta. 00:34:07.090 --> 00:34:09.699 align:middle line:84% So what firm one is going to do is 00:34:09.699 --> 00:34:13.449 align:middle line:84% firm one is going to price at its advantage 00:34:13.449 --> 00:34:16.360 align:middle line:84% relative to firm two, and the consumer 00:34:16.360 --> 00:34:18.860 align:middle line:90% is going to buy from firm one. 00:34:18.860 --> 00:34:20.840 align:middle line:84% And so what you get if you-- over here I've 00:34:20.840 --> 00:34:24.590 align:middle line:84% tried to graph what is equilibrium price paid 00:34:24.590 --> 00:34:27.860 align:middle line:84% as a function of theta, and all these consumers over here 00:34:27.860 --> 00:34:29.093 align:middle line:90% buy from firm one. 00:34:29.093 --> 00:34:31.760 align:middle line:84% As you come to the middle of the line, at the middle of the line 00:34:31.760 --> 00:34:33.320 align:middle line:84% they're in pure Bertrand competition. 00:34:33.320 --> 00:34:35.278 align:middle line:84% The consumer has the same preferences for both, 00:34:35.278 --> 00:34:36.770 align:middle line:90% so they both price at c. 00:34:36.770 --> 00:34:39.929 align:middle line:84% And then you go back up here to c plus t at the end. 00:34:39.929 --> 00:34:42.449 align:middle line:90% So no discrimination. 00:34:42.449 --> 00:34:43.610 align:middle line:90% Everyone pays c plus t. 00:34:43.610 --> 00:34:46.190 align:middle line:90% 00:34:46.190 --> 00:34:49.040 align:middle line:84% With discrimination, the worst case is you pay c plus t 00:34:49.040 --> 00:34:50.810 align:middle line:84% if someone knows you're fully captive 00:34:50.810 --> 00:34:52.730 align:middle line:84% and like their product by t better, 00:34:52.730 --> 00:34:56.719 align:middle line:84% and then otherwise you pay c in the middle. 00:34:56.719 --> 00:34:57.590 align:middle line:90% OK. 00:34:57.590 --> 00:35:00.020 align:middle line:84% So all the consumers are better off, 00:35:00.020 --> 00:35:04.760 align:middle line:84% and they're better off because the rivals are cutting prices 00:35:04.760 --> 00:35:08.300 align:middle line:84% to target consumers who don't want to buy from them, 00:35:08.300 --> 00:35:12.200 align:middle line:84% and that targeting the rival's consumers is an intense-- 00:35:12.200 --> 00:35:15.140 align:middle line:84% allowing them to do that intensifies the competition, 00:35:15.140 --> 00:35:18.990 align:middle line:84% and that intensified competition cuts prices for everybody. 00:35:18.990 --> 00:35:21.270 align:middle line:84% So that's the thought that price discrimination can be 00:35:21.270 --> 00:35:24.490 align:middle line:90% a universal good for consumers. 00:35:24.490 --> 00:35:24.990 align:middle line:90% OK. 00:35:24.990 --> 00:35:27.905 align:middle line:90% 00:35:27.905 --> 00:35:33.720 align:middle line:84% A second insight, though, came, Ken Corts, a decade later, 00:35:33.720 --> 00:35:36.573 align:middle line:90% was that that's true. 00:35:36.573 --> 00:35:37.740 align:middle line:90% But there's a second effect. 00:35:37.740 --> 00:35:40.073 align:middle line:84% And price discrimination can actually make prices higher 00:35:40.073 --> 00:35:41.720 align:middle line:90% for all consumers. 00:35:41.720 --> 00:35:45.200 align:middle line:84% And it's the thought that price discrimination makes 00:35:45.200 --> 00:35:47.840 align:middle line:90% your pricing more flexible. 00:35:47.840 --> 00:35:50.510 align:middle line:84% One thing that lets you do is target your rival's consumers, 00:35:50.510 --> 00:35:52.230 align:middle line:90% and that's good for competition. 00:35:52.230 --> 00:35:54.350 align:middle line:84% The other thing it does is let you exploit 00:35:54.350 --> 00:35:56.900 align:middle line:90% people who are captive to you. 00:35:56.900 --> 00:36:02.300 align:middle line:84% And so Corts looked at this specification where if you think 00:36:02.300 --> 00:36:04.960 align:middle line:84% about it-- so over here, what I've tried to graph is-- 00:36:04.960 --> 00:36:07.700 align:middle line:90% 00:36:07.700 --> 00:36:13.730 align:middle line:84% what that's a graph of is what is v, the value that you 00:36:13.730 --> 00:36:16.130 align:middle line:90% get for buying from firm theta. 00:36:16.130 --> 00:36:19.340 align:middle line:84% And what he assumes is that you get value two 00:36:19.340 --> 00:36:21.170 align:middle line:90% from buying from a consumer-- 00:36:21.170 --> 00:36:23.920 align:middle line:84% so these are value they get buying from firm one. 00:36:23.920 --> 00:36:27.390 align:middle line:90% 00:36:27.390 --> 00:36:29.200 align:middle line:90% Your value goes like this. 00:36:29.200 --> 00:36:31.740 align:middle line:84% The farther your preferred type theta 00:36:31.740 --> 00:36:36.450 align:middle line:84% is from firm one's product, the lower 00:36:36.450 --> 00:36:39.570 align:middle line:90% is the utility you get from it. 00:36:39.570 --> 00:36:43.170 align:middle line:84% And then once the product is too far from what you want, 00:36:43.170 --> 00:36:45.810 align:middle line:90% your utility just drops to zero. 00:36:45.810 --> 00:36:48.670 align:middle line:84% So I have this decreasing distance from firm one. 00:36:48.670 --> 00:36:49.990 align:middle line:90% I like it less and less. 00:36:49.990 --> 00:36:51.907 align:middle line:84% And then once it gets to some point, I'm like, 00:36:51.907 --> 00:36:52.840 align:middle line:90% I can't use that. 00:36:52.840 --> 00:36:55.320 align:middle line:84% And then your utility drops to zero. 00:36:55.320 --> 00:36:58.230 align:middle line:84% So that's my utility buying from firm one, 00:36:58.230 --> 00:37:01.150 align:middle line:84% and my utility buying from firm two does the opposite. 00:37:01.150 --> 00:37:04.420 align:middle line:84% It's zero here and then it goes up like that. 00:37:04.420 --> 00:37:06.360 align:middle line:84% So these high theta consumers really 00:37:06.360 --> 00:37:10.100 align:middle line:84% like and get utility two from buying firm one's product. 00:37:10.100 --> 00:37:12.150 align:middle line:84% The guy in the middle gets utility one from both, 00:37:12.150 --> 00:37:13.270 align:middle line:90% and vice versa. 00:37:13.270 --> 00:37:15.690 align:middle line:84% So in some sense, I've eliminated 00:37:15.690 --> 00:37:19.050 align:middle line:84% that targeting your rival's competitors effect 00:37:19.050 --> 00:37:23.070 align:middle line:84% by saying that anyone who gets utility from one firm 00:37:23.070 --> 00:37:25.500 align:middle line:84% gets zero utility from the other firm. 00:37:25.500 --> 00:37:28.120 align:middle line:84% So letting them target your rival's consumers 00:37:28.120 --> 00:37:30.850 align:middle line:84% doesn't actually create any competition. 00:37:30.850 --> 00:37:34.420 align:middle line:84% So what would happen in this model? 00:37:34.420 --> 00:37:39.790 align:middle line:84% In this model, if I ban competition-- 00:37:39.790 --> 00:37:41.210 align:middle line:90% if I ban price discrimination. 00:37:41.210 --> 00:37:41.710 align:middle line:90% Sorry. 00:37:41.710 --> 00:37:43.930 align:middle line:84% If I ban price discrimination, firm one 00:37:43.930 --> 00:37:47.410 align:middle line:84% is essentially pricing as if it had a linear demand curve. 00:37:47.410 --> 00:37:49.220 align:middle line:84% It's a linear demand curve that cuts off, 00:37:49.220 --> 00:37:52.120 align:middle line:84% but if you remember, if you have a linear demand curve that 00:37:52.120 --> 00:37:53.800 align:middle line:84% looks like this, even if it continues, 00:37:53.800 --> 00:37:55.230 align:middle line:90% you still would price at one. 00:37:55.230 --> 00:37:57.890 align:middle line:90% 00:37:57.890 --> 00:38:00.020 align:middle line:84% If you ban price discrimination, firm one 00:38:00.020 --> 00:38:02.320 align:middle line:90% is going to set p star equals 1. 00:38:02.320 --> 00:38:04.250 align:middle line:84% Firm two is going to set p star equals 00:38:04.250 --> 00:38:07.220 align:middle line:84% 1, because they're both just pricing to the linear demand 00:38:07.220 --> 00:38:09.740 align:middle line:90% curve 2 minus p. 00:38:09.740 --> 00:38:12.620 align:middle line:84% Whereas if you allow them to discriminate, 00:38:12.620 --> 00:38:14.550 align:middle line:84% I do first degree price discrimination. 00:38:14.550 --> 00:38:16.670 align:middle line:84% This guy pays 2, this guy pays 1.9, 00:38:16.670 --> 00:38:19.950 align:middle line:84% this guy pays 1.8, and so on until this guy pays 1. 00:38:19.950 --> 00:38:22.820 align:middle line:84% So basically, I just raised prices to everyone 00:38:22.820 --> 00:38:25.940 align:middle line:84% because letting me price discriminate 00:38:25.940 --> 00:38:29.565 align:middle line:84% lets me exploit all-- lets me extract all of their surplus 00:38:29.565 --> 00:38:31.190 align:middle line:84% when I don't have competition, and here 00:38:31.190 --> 00:38:32.982 align:middle line:84% I've designed the oligopoly so there really 00:38:32.982 --> 00:38:34.340 align:middle line:90% isn't any competition. 00:38:34.340 --> 00:38:39.080 align:middle line:84% So I think Corts's basic insight was that there are these two 00:38:39.080 --> 00:38:42.170 align:middle line:84% effects and we shouldn't necessarily 00:38:42.170 --> 00:38:44.240 align:middle line:84% expect either of those two effects to dominate. 00:38:44.240 --> 00:38:46.220 align:middle line:84% Price discrimination can be good because it 00:38:46.220 --> 00:38:47.570 align:middle line:90% intensifies competition. 00:38:47.570 --> 00:38:50.210 align:middle line:84% It can be bad because it has more exploitation, 00:38:50.210 --> 00:38:52.550 align:middle line:84% and it's just going to matter on the distribution 00:38:52.550 --> 00:38:55.820 align:middle line:84% of relative preferences for firm one versus firm two's products, 00:38:55.820 --> 00:38:59.720 align:middle line:84% whether the exploitation effect dominates and makes prices go up 00:38:59.720 --> 00:39:01.610 align:middle line:84% or the competition effect dominates 00:39:01.610 --> 00:39:03.740 align:middle line:90% and makes prices go down. 00:39:03.740 --> 00:39:06.710 align:middle line:84% As I said, this Hotelling model, it's 00:39:06.710 --> 00:39:08.353 align:middle line:84% like it's an intermediate effect. 00:39:08.353 --> 00:39:10.020 align:middle line:84% It's like you charge intermediate prices 00:39:10.020 --> 00:39:13.515 align:middle line:84% and you lose the ability to compete and discriminate 00:39:13.515 --> 00:39:15.390 align:middle line:84% in a way that makes you just charge something 00:39:15.390 --> 00:39:17.973 align:middle line:84% in between what you would have charged the high and low types. 00:39:17.973 --> 00:39:22.425 align:middle line:90% 00:39:22.425 --> 00:39:24.190 align:middle line:90% Any questions on that argument? 00:39:24.190 --> 00:39:25.002 align:middle line:90% Yeah? 00:39:25.002 --> 00:39:26.460 align:middle line:84% AUDIENCE: He's just kind of arguing 00:39:26.460 --> 00:39:29.243 align:middle line:84% that if it's not actually an oligopoly, it's monopoly. 00:39:29.243 --> 00:39:30.660 align:middle line:84% GLENN ELLISON: Yeah, that's right. 00:39:30.660 --> 00:39:34.140 align:middle line:84% So he made it a monopoly, but then obviously you 00:39:34.140 --> 00:39:36.540 align:middle line:84% could have it be an oligopoly like this 00:39:36.540 --> 00:39:38.348 align:middle line:84% and you would get the same results, 00:39:38.348 --> 00:39:40.140 align:middle line:84% Like if it was an oligopoly where it really 00:39:40.140 --> 00:39:44.790 align:middle line:84% was an oligopoly but most of them got low utility from one 00:39:44.790 --> 00:39:47.070 align:middle line:84% and high from the other, then you would still 00:39:47.070 --> 00:39:48.720 align:middle line:90% get the same effect there. 00:39:48.720 --> 00:39:50.550 align:middle line:84% But he's arguing that it's basically 00:39:50.550 --> 00:39:52.050 align:middle line:84% it's adjacent to the monopoly model. 00:39:52.050 --> 00:39:53.830 align:middle line:84% In the monopoly model it goes this way, 00:39:53.830 --> 00:39:55.990 align:middle line:84% so oligopoly model is near that one, 00:39:55.990 --> 00:39:57.950 align:middle line:90% it's going to go the same way. 00:39:57.950 --> 00:39:58.450 align:middle line:90% Yeah. 00:39:58.450 --> 00:40:03.370 align:middle line:90% 00:40:03.370 --> 00:40:04.300 align:middle line:90% OK. 00:40:04.300 --> 00:40:07.970 align:middle line:84% So that was oligopoly third degree price discrimination. 00:40:07.970 --> 00:40:10.430 align:middle line:84% What about second degree price discrimination? 00:40:10.430 --> 00:40:14.350 align:middle line:84% And I do think that oligopoly second degree price 00:40:14.350 --> 00:40:17.710 align:middle line:84% discrimination is a really important topic because I think 00:40:17.710 --> 00:40:21.840 align:middle line:84% many, many products people buy, there 00:40:21.840 --> 00:40:23.640 align:middle line:84% is second degree discrimination there. 00:40:23.640 --> 00:40:24.730 align:middle line:90% You're buying phones. 00:40:24.730 --> 00:40:27.270 align:middle line:84% There is the iPhone 13 and the iPhone 14 00:40:27.270 --> 00:40:30.780 align:middle line:84% and the 128 megabyte and the 512 megabyte version. 00:40:30.780 --> 00:40:35.190 align:middle line:84% You're buying cable TV and it's cable TV package one, package 00:40:35.190 --> 00:40:37.230 align:middle line:84% two, package three, package four, 00:40:37.230 --> 00:40:39.360 align:middle line:90% all with increasing features. 00:40:39.360 --> 00:40:41.680 align:middle line:84% You have competing firms selling TVs. 00:40:41.680 --> 00:40:46.110 align:middle line:84% Everyone's got the 42-inch TV and the 49-inch TV 00:40:46.110 --> 00:40:48.810 align:middle line:84% and the 55-inch TV and the 62-inch TV, 00:40:48.810 --> 00:40:52.200 align:middle line:84% and each of those TVs has the 4k or doesn't have the 4k. 00:40:52.200 --> 00:40:56.340 align:middle line:84% But you generally see firms in oligopoly industries 00:40:56.340 --> 00:40:59.670 align:middle line:84% are selling multiple products, competing with multiple products 00:40:59.670 --> 00:41:00.320 align:middle line:90% of each other. 00:41:00.320 --> 00:41:03.450 align:middle line:90% 00:41:03.450 --> 00:41:05.700 align:middle line:84% So if you're talking about oligopoly second degree 00:41:05.700 --> 00:41:08.760 align:middle line:84% discrimination or the cars in Bresnahan's example, 00:41:08.760 --> 00:41:11.430 align:middle line:84% like it's Ford is producing a line of cars, 00:41:11.430 --> 00:41:12.970 align:middle line:90% GM is producing a line of cars. 00:41:12.970 --> 00:41:15.370 align:middle line:84% Those cars are competing against other models, 00:41:15.370 --> 00:41:17.580 align:middle line:84% and there's vertical constraints within the firm 00:41:17.580 --> 00:41:20.370 align:middle line:84% and there's also competition across the firms. 00:41:20.370 --> 00:41:25.770 align:middle line:84% So clearly you want to have vertical differentiation 00:41:25.770 --> 00:41:28.530 align:middle line:84% and you want to have horizontal differentiation. 00:41:28.530 --> 00:41:31.050 align:middle line:84% And then the substitution patterns, 00:41:31.050 --> 00:41:34.890 align:middle line:84% there's going to be cross firm substitution going on 00:41:34.890 --> 00:41:37.840 align:middle line:84% and there's going to be within firm substitution going on. 00:41:37.840 --> 00:41:40.500 align:middle line:84% And it just turns out it's hard to write down a nice tractable 00:41:40.500 --> 00:41:45.390 align:middle line:84% model that has the two dimensions of consumer tastes 00:41:45.390 --> 00:41:49.170 align:middle line:84% and has both substitution patterns mattering 00:41:49.170 --> 00:41:52.560 align:middle line:84% and the binding IC constraints and the binding 00:41:52.560 --> 00:41:54.990 align:middle line:90% cross firm constraints. 00:41:54.990 --> 00:41:57.210 align:middle line:84% So what the literatures tend to do 00:41:57.210 --> 00:41:59.420 align:middle line:84% is do a few different special cases. 00:41:59.420 --> 00:42:03.490 align:middle line:84% One thing it's done is to assume independent horizontal and 00:42:03.490 --> 00:42:04.600 align:middle line:90% vertical preferences. 00:42:04.600 --> 00:42:06.550 align:middle line:84% There's this observation that if you 00:42:06.550 --> 00:42:09.310 align:middle line:84% make horizontal and vertical preferences independent, 00:42:09.310 --> 00:42:12.250 align:middle line:84% you get a surprisingly tractable model that you can do even 00:42:12.250 --> 00:42:14.560 align:middle line:90% with a continuum of types. 00:42:14.560 --> 00:42:18.490 align:middle line:84% The other thing that people do is have discrete vertical types, 00:42:18.490 --> 00:42:21.760 align:middle line:84% and when you have discrete vertical types, 00:42:21.760 --> 00:42:24.940 align:middle line:84% you can sometimes make the IC constraints, the within firm IC 00:42:24.940 --> 00:42:26.770 align:middle line:84% constraints non-binding and eliminate 00:42:26.770 --> 00:42:29.830 align:middle line:84% the number of constraints you have to solve for. 00:42:29.830 --> 00:42:31.570 align:middle line:84% And then a third approach you sometimes 00:42:31.570 --> 00:42:33.370 align:middle line:84% see people do is actually just eliminate 00:42:33.370 --> 00:42:35.470 align:middle line:84% the vertical differentiation entirely and talk 00:42:35.470 --> 00:42:38.680 align:middle line:84% about second degree discrimination in a model that 00:42:38.680 --> 00:42:41.206 align:middle line:84% doesn't have vertical differentiation. 00:42:41.206 --> 00:42:44.080 align:middle line:90% 00:42:44.080 --> 00:42:45.340 align:middle line:90% So let me go-- 00:42:45.340 --> 00:42:49.680 align:middle line:84% I'll start with just an overview of what happens 00:42:49.680 --> 00:42:51.130 align:middle line:90% in this independent model. 00:42:51.130 --> 00:42:54.960 align:middle line:84% So Stole has a chapter of the handbook in IO 00:42:54.960 --> 00:42:59.340 align:middle line:84% that surveys this literature that's pretty big and seemingly 00:42:59.340 --> 00:43:00.070 align:middle line:90% general. 00:43:00.070 --> 00:43:01.980 align:middle line:90% You have these firms j. 00:43:01.980 --> 00:43:03.840 align:middle line:84% They can produce a range of qualities 00:43:03.840 --> 00:43:06.878 align:middle line:84% s at some constant marginal cost c of s. 00:43:06.878 --> 00:43:08.670 align:middle line:84% So typically they actually have a continuum 00:43:08.670 --> 00:43:12.090 align:middle line:84% of s's they could produce if they wanted to. 00:43:12.090 --> 00:43:16.470 align:middle line:84% Consumer i's utility from buying a good of quality s from firm j 00:43:16.470 --> 00:43:18.700 align:middle line:90% is theta i of s. 00:43:18.700 --> 00:43:20.440 align:middle line:84% So this is the vertical preference. 00:43:20.440 --> 00:43:23.850 align:middle line:84% High types are willing to pay more for quality. 00:43:23.850 --> 00:43:27.960 align:middle line:84% Minus the price plus the epsilon ij. 00:43:27.960 --> 00:43:33.950 align:middle line:84% So this is the logit-like error that we've had in our models. 00:43:33.950 --> 00:43:39.440 align:middle line:84% And then models assume that the theta i's and the epsilon ij's 00:43:39.440 --> 00:43:43.970 align:middle line:84% are all independent, and theta i is distributed 00:43:43.970 --> 00:43:47.372 align:middle line:84% f theta, epsilon ij is distributed f epsilon, 00:43:47.372 --> 00:43:48.830 align:middle line:84% and they're pretty general in terms 00:43:48.830 --> 00:43:51.080 align:middle line:84% of what distribution of theta is and what distribution 00:43:51.080 --> 00:43:52.950 align:middle line:90% of epsilons you have. 00:43:52.950 --> 00:43:59.920 align:middle line:84% So let me define s star of theta to be the efficient quality 00:43:59.920 --> 00:44:03.400 align:middle line:90% for a consumer of type-- 00:44:03.400 --> 00:44:05.890 align:middle line:84% that should say type theta to buy. 00:44:05.890 --> 00:44:06.940 align:middle line:90% So you have a consumer. 00:44:06.940 --> 00:44:09.050 align:middle line:84% That should be a theta instead of a v. 00:44:09.050 --> 00:44:12.700 align:middle line:84% So if you have a type theta consumer, 00:44:12.700 --> 00:44:16.420 align:middle line:84% they ought to be buying quality that maximizes theta times s 00:44:16.420 --> 00:44:17.250 align:middle line:90% minus c of s. 00:44:17.250 --> 00:44:19.630 align:middle line:84% So this is if the goods were sold at cost, 00:44:19.630 --> 00:44:20.633 align:middle line:90% what would they buy? 00:44:20.633 --> 00:44:22.425 align:middle line:84% That's the efficient thing to sell to them. 00:44:22.425 --> 00:44:25.415 align:middle line:90% 00:44:25.415 --> 00:44:28.550 align:middle line:84% So just an example of a model fitting this. 00:44:28.550 --> 00:44:31.540 align:middle line:84% Consider a Hotelling-like model, and suppose 00:44:31.540 --> 00:44:35.140 align:middle line:84% that epsilon i1 is distributed uniform on zero 00:44:35.140 --> 00:44:40.260 align:middle line:84% to t and epsilon i2 is just the negative of epsilon one. 00:44:40.260 --> 00:44:45.690 align:middle line:90% 00:44:45.690 --> 00:44:48.960 align:middle line:84% A Nash equilibrium of that model is just pjs is c of s 00:44:48.960 --> 00:44:51.450 align:middle line:90% plus t for all j and s. 00:44:51.450 --> 00:44:53.685 align:middle line:84% So if you imagine, it's as if we've got this-- 00:44:53.685 --> 00:44:58.910 align:middle line:90% 00:44:58.910 --> 00:45:02.395 align:middle line:84% I've got this continuum of Hotelling lines. 00:45:02.395 --> 00:45:07.420 align:middle line:90% 00:45:07.420 --> 00:45:09.590 align:middle line:84% So you have these different quality levels, 00:45:09.590 --> 00:45:12.913 align:middle line:90% s1, s2, s3, s4, s5. 00:45:12.913 --> 00:45:16.500 align:middle line:90% 00:45:16.500 --> 00:45:23.850 align:middle line:84% On each line, let's imagine I charge p star equals c of s. 00:45:23.850 --> 00:45:28.320 align:middle line:90% So quality of-- yeah. 00:45:28.320 --> 00:45:32.805 align:middle line:84% Price for quality s is c of s plus t. 00:45:32.805 --> 00:45:36.090 align:middle line:90% So imagine you had that pricing. 00:45:36.090 --> 00:45:40.980 align:middle line:84% That every quality I sell at the cost plus t. 00:45:40.980 --> 00:45:43.680 align:middle line:84% Then what are consumers going to do? 00:45:43.680 --> 00:45:47.440 align:middle line:84% The markups are the same for every quality level. 00:45:47.440 --> 00:45:49.770 align:middle line:84% So when I'm deciding which quality level to buy, 00:45:49.770 --> 00:45:53.430 align:middle line:84% I buy the efficient quality level. 00:45:53.430 --> 00:45:57.660 align:middle line:84% If I'm a type theta consumer, I buy s star of theta. 00:45:57.660 --> 00:46:00.690 align:middle line:84% So all the consumers for whom s3 is the optimal quality 00:46:00.690 --> 00:46:06.600 align:middle line:84% are like, OK, I'm either going to buy quality s3 from firm one 00:46:06.600 --> 00:46:08.710 align:middle line:84% or I'm going to buy quality three from s2, 00:46:08.710 --> 00:46:10.110 align:middle line:90% who do I buy from? 00:46:10.110 --> 00:46:12.570 align:middle line:90% Well, again, it depends on my-- 00:46:12.570 --> 00:46:17.460 align:middle line:84% now I'm calling this an epsilon i1 and an epsilon i2. 00:46:17.460 --> 00:46:22.560 align:middle line:84% But if my taste is I really like firm one's product, 00:46:22.560 --> 00:46:23.608 align:middle line:90% I buy from firm one. 00:46:23.608 --> 00:46:25.650 align:middle line:84% If my epsilon is I really buy firm two's product, 00:46:25.650 --> 00:46:27.160 align:middle line:90% I buy from firm two. 00:46:27.160 --> 00:46:30.030 align:middle line:90% And so it's just the-- 00:46:30.030 --> 00:46:32.580 align:middle line:84% as far as if the firm cuts its price, 00:46:32.580 --> 00:46:35.230 align:middle line:84% it's just like the competition on a line model. 00:46:35.230 --> 00:46:39.310 align:middle line:84% And if it cuts its price, it will attract more consumers 00:46:39.310 --> 00:46:40.340 align:middle line:90% from the other side. 00:46:40.340 --> 00:46:44.710 align:middle line:84% It would have this increase in demand one half plus p2 minus p1 00:46:44.710 --> 00:46:48.700 align:middle line:84% over 2t, but they're just basically competing on a line 00:46:48.700 --> 00:46:51.020 align:middle line:84% against the other firm for this quality level. 00:46:51.020 --> 00:46:53.680 align:middle line:84% And so the same Hotelling first order condition applies 00:46:53.680 --> 00:46:56.950 align:middle line:84% and they charge price c plus t on each line. 00:46:56.950 --> 00:46:58.630 align:middle line:84% And so if the firms are doing this 00:46:58.630 --> 00:47:00.940 align:middle line:84% and they're playing Hotelling on each line, 00:47:00.940 --> 00:47:03.100 align:middle line:84% they're pricing at c plus t on each line, 00:47:03.100 --> 00:47:06.830 align:middle line:84% all consumers do buy the best quality for them. 00:47:06.830 --> 00:47:09.670 align:middle line:84% And so the result here is that when 00:47:09.670 --> 00:47:12.280 align:middle line:84% you have competitive secondary discrimination, 00:47:12.280 --> 00:47:14.800 align:middle line:84% secondary discrimination goes away. 00:47:14.800 --> 00:47:17.740 align:middle line:84% The firms are not using different quality levels 00:47:17.740 --> 00:47:19.400 align:middle line:90% to discriminate among consumers. 00:47:19.400 --> 00:47:21.430 align:middle line:90% They just are charging-- 00:47:21.430 --> 00:47:24.070 align:middle line:84% it's not like the iPhone with 512 megabytes 00:47:24.070 --> 00:47:27.380 align:middle line:84% cost $300 more than the iPhone with 128 megabytes. 00:47:27.380 --> 00:47:29.260 align:middle line:84% They give them all at the same markup 00:47:29.260 --> 00:47:31.310 align:middle line:84% and all consumers buy the efficient quality. 00:47:31.310 --> 00:47:35.700 align:middle line:84% So all of that quality distortions 00:47:35.700 --> 00:47:37.340 align:middle line:84% in secondary discrimination goes away. 00:47:37.340 --> 00:47:41.160 align:middle line:90% 00:47:41.160 --> 00:47:45.230 align:middle line:84% And I gave this example in the Hotelling model, 00:47:45.230 --> 00:47:50.523 align:middle line:84% but Stole shows that this is really very general. 00:47:50.523 --> 00:47:51.940 align:middle line:84% So there's this very general thing 00:47:51.940 --> 00:47:55.300 align:middle line:84% of for f distributions of epsilon, 00:47:55.300 --> 00:47:57.820 align:middle line:84% satisfying some conditions, industries of epsilon 00:47:57.820 --> 00:48:00.280 align:middle line:84% satisfying some conditions, as long 00:48:00.280 --> 00:48:03.820 align:middle line:84% as the theta and epsilon are independent, 00:48:03.820 --> 00:48:05.710 align:middle line:90% this always happens. 00:48:05.710 --> 00:48:08.140 align:middle line:84% You just get competition on many lines 00:48:08.140 --> 00:48:10.750 align:middle line:84% and constant markups across goods. 00:48:10.750 --> 00:48:15.040 align:middle line:84% And I think the thing that I want to point out 00:48:15.040 --> 00:48:19.510 align:middle line:84% most about this model is that it seems very general, 00:48:19.510 --> 00:48:22.810 align:middle line:84% but it's got this hidden, special, and incorrect 00:48:22.810 --> 00:48:26.650 align:middle line:84% assumption in it, which is that the thetas and the epsilons 00:48:26.650 --> 00:48:31.880 align:middle line:90% are going to be independent. 00:48:31.880 --> 00:48:33.980 align:middle line:84% And a good example to think about this 00:48:33.980 --> 00:48:44.510 align:middle line:84% is think about you and a CEO of some company flying to Europe, 00:48:44.510 --> 00:48:46.880 align:middle line:84% and you're told that there are two flights available 00:48:46.880 --> 00:48:47.960 align:middle line:90% to go to Paris. 00:48:47.960 --> 00:48:50.375 align:middle line:84% There's a Delta flight or an Air France flight. 00:48:50.375 --> 00:48:53.930 align:middle line:90% 00:48:53.930 --> 00:48:58.070 align:middle line:84% You might say that some people like Air France better 00:48:58.070 --> 00:48:58.770 align:middle line:90% than Delta. 00:48:58.770 --> 00:49:00.933 align:middle line:84% Some people like Delta better than Air France. 00:49:00.933 --> 00:49:03.590 align:middle line:90% 00:49:03.590 --> 00:49:06.800 align:middle line:84% It may be that the preferences that students 00:49:06.800 --> 00:49:09.380 align:middle line:84% have on Delta versus Air France and the preferences 00:49:09.380 --> 00:49:11.660 align:middle line:84% that the CEOs have on Delta versus Air France 00:49:11.660 --> 00:49:13.050 align:middle line:90% are uncorrelated. 00:49:13.050 --> 00:49:14.660 align:middle line:84% That yeah, they're both-- like, you 00:49:14.660 --> 00:49:16.460 align:middle line:84% can model them as uniform on zero 00:49:16.460 --> 00:49:19.190 align:middle line:84% to t for their disutility for Air 00:49:19.190 --> 00:49:22.220 align:middle line:84% France relative to their ideal airline disutility for Delta. 00:49:22.220 --> 00:49:26.220 align:middle line:84% But what's uncorrelated is reasonable. 00:49:26.220 --> 00:49:29.660 align:middle line:84% Independent is not reasonable, because how much are you 00:49:29.660 --> 00:49:30.423 align:middle line:90% willing to pay? 00:49:30.423 --> 00:49:32.090 align:middle line:84% Those of you who really like Air France, 00:49:32.090 --> 00:49:34.800 align:middle line:84% how much extra are you willing to pay for to go on Air France? 00:49:34.800 --> 00:49:36.480 align:middle line:90% Maybe it's $20. 00:49:36.480 --> 00:49:37.560 align:middle line:90% Maybe it's $10. 00:49:37.560 --> 00:49:38.660 align:middle line:90% I don't know. 00:49:38.660 --> 00:49:43.250 align:middle line:84% The CEOs, they may pay $5,000 more to go on Air France 00:49:43.250 --> 00:49:46.040 align:middle line:84% relative to Delta if they like it, because for them, 00:49:46.040 --> 00:49:48.150 align:middle line:90% the $5,000 is rounding error. 00:49:48.150 --> 00:49:50.270 align:middle line:84% And so I think what's reasonable there 00:49:50.270 --> 00:49:54.880 align:middle line:84% is the uncorrelated preferences that the businessmen 00:49:54.880 --> 00:49:56.380 align:middle line:84% and the students may have the same-- 00:49:56.380 --> 00:49:58.890 align:middle line:90% 00:49:58.890 --> 00:50:04.200 align:middle line:84% well, same uncorrelated preferences for Delta versus Air 00:50:04.200 --> 00:50:07.540 align:middle line:84% France, but they wouldn't have independent preferences. 00:50:07.540 --> 00:50:13.980 align:middle line:84% What it really might look like is you'd have the student line-- 00:50:13.980 --> 00:50:19.380 align:middle line:84% you have the student line where t is very small, 00:50:19.380 --> 00:50:23.610 align:middle line:84% and then you have the businessman line 00:50:23.610 --> 00:50:26.430 align:middle line:90% where t is very, very big. 00:50:26.430 --> 00:50:29.497 align:middle line:84% So they have a higher vertical willingness to pay for the good, 00:50:29.497 --> 00:50:31.830 align:middle line:84% but then, in addition to the higher vertical willingness 00:50:31.830 --> 00:50:35.580 align:middle line:84% to pay, they just have much more intense horizontal preferences 00:50:35.580 --> 00:50:36.570 align:middle line:90% as measured in dollars. 00:50:36.570 --> 00:50:40.008 align:middle line:84% That's not covered by the independence case. 00:50:40.008 --> 00:50:42.300 align:middle line:84% Independence would say there should be one line up here 00:50:42.300 --> 00:50:44.100 align:middle line:84% like this where, again, they'll all 00:50:44.100 --> 00:50:47.717 align:middle line:84% shift from Air France to Delta for 20 bucks, which is just not 00:50:47.717 --> 00:50:48.425 align:middle line:90% going to be true. 00:50:48.425 --> 00:50:50.975 align:middle line:90% 00:50:50.975 --> 00:50:54.260 align:middle line:84% And it's the independence, in some sense, that results 00:50:54.260 --> 00:50:57.350 align:middle line:90% in the constant markups. 00:50:57.350 --> 00:51:00.080 align:middle line:84% So a second way people model this is just let's 00:51:00.080 --> 00:51:02.250 align:middle line:84% go with this discrete type space. 00:51:02.250 --> 00:51:08.160 align:middle line:84% So there's just two quantities, sl and sh. 00:51:08.160 --> 00:51:11.300 align:middle line:84% sl is the coach class seat where your knees are pressed 00:51:11.300 --> 00:51:13.970 align:middle line:84% into the seat in front of you and then sh 00:51:13.970 --> 00:51:17.360 align:middle line:84% is the lie flat bed with the nice TV 00:51:17.360 --> 00:51:19.760 align:middle line:84% and the champagne and glasses when you walk on the plane 00:51:19.760 --> 00:51:21.470 align:middle line:90% and everything else. 00:51:21.470 --> 00:51:24.530 align:middle line:84% And so what you can think of here 00:51:24.530 --> 00:51:27.050 align:middle line:84% is model this is I can model that 00:51:27.050 --> 00:51:32.200 align:middle line:84% as the vertical differentiation model where consumers have 00:51:32.200 --> 00:51:35.830 align:middle line:84% a discrete vertical type tl or th, 00:51:35.830 --> 00:51:40.350 align:middle line:84% and then they have horizontal types epsilon i1 and epsilon i2, 00:51:40.350 --> 00:51:42.340 align:middle line:84% just like in the Hotelling model. 00:51:42.340 --> 00:51:44.890 align:middle line:84% And let's instead then specify the utility. 00:51:44.890 --> 00:51:48.820 align:middle line:84% Instead of being theta s plus epsilon minus p plus epsilon, 00:51:48.820 --> 00:51:52.480 align:middle line:84% we'll have the theta multiply s minus epsilon. 00:51:52.480 --> 00:51:54.175 align:middle line:90% So that if you have-- 00:51:54.175 --> 00:51:58.450 align:middle line:90% 00:51:58.450 --> 00:52:00.670 align:middle line:84% everyone has horizontal preferences 00:52:00.670 --> 00:52:03.050 align:middle line:90% at the same intensity in utils. 00:52:03.050 --> 00:52:05.290 align:middle line:84% But when you convert the utils to dollars 00:52:05.290 --> 00:52:09.012 align:middle line:84% by multiplying by this theta, the preferences in dollar terms 00:52:09.012 --> 00:52:10.720 align:middle line:84% are much more intense for the businessmen 00:52:10.720 --> 00:52:12.490 align:middle line:90% than they are for the students. 00:52:12.490 --> 00:52:17.470 align:middle line:84% So the theta, the vertical type, multiplies both the quality 00:52:17.470 --> 00:52:23.050 align:middle line:84% and the relative preference of Delta, Air France. 00:52:23.050 --> 00:52:24.070 align:middle line:90% OK. 00:52:24.070 --> 00:52:26.500 align:middle line:84% So in this model, it's reasonable to assume 00:52:26.500 --> 00:52:29.870 align:middle line:84% that the thetas and the epsilons are now independent. 00:52:29.870 --> 00:52:32.650 align:middle line:84% So you have this willingness to pay for quality, 00:52:32.650 --> 00:52:36.160 align:middle line:84% whether it be quality of the nice seat or quality of the Air 00:52:36.160 --> 00:52:37.310 align:middle line:90% France food. 00:52:37.310 --> 00:52:40.300 align:middle line:84% And then you have an idiosyncratic taste 00:52:40.300 --> 00:52:43.160 align:middle line:84% for the Air France food versus the Delta food. 00:52:43.160 --> 00:52:44.360 align:middle line:90% OK. 00:52:44.360 --> 00:52:46.850 align:middle line:84% Anyway, so I'm going to assume for this numerical example 00:52:46.850 --> 00:52:49.460 align:middle line:84% that the thetas and the epsilons are independent 00:52:49.460 --> 00:52:53.150 align:middle line:84% and half the people are high types. 00:52:53.150 --> 00:52:55.020 align:middle line:90% So what happens here? 00:52:55.020 --> 00:52:56.270 align:middle line:90% Well, you know, simple fact. 00:52:56.270 --> 00:52:59.090 align:middle line:84% I've got conditions under which this is true, 00:52:59.090 --> 00:53:01.010 align:middle line:84% but under some set of conditions, 00:53:01.010 --> 00:53:05.780 align:middle line:84% what you get is if this is a line that has length tl 00:53:05.780 --> 00:53:10.250 align:middle line:84% and this is a line that has length th, 00:53:10.250 --> 00:53:14.600 align:middle line:84% you get p star equals c plus tl here 00:53:14.600 --> 00:53:18.070 align:middle line:84% and you get p star equals c plus th here. 00:53:18.070 --> 00:53:18.570 align:middle line:90% All right. 00:53:18.570 --> 00:53:22.830 align:middle line:90% So on the big line, you price-- 00:53:22.830 --> 00:53:24.427 align:middle line:84% with the big amount of differentiation 00:53:24.427 --> 00:53:26.510 align:middle line:84% on the little line, you price in the little amount 00:53:26.510 --> 00:53:31.070 align:middle line:84% of differentiation, and all the low types by the low quality 00:53:31.070 --> 00:53:34.430 align:middle line:84% product, all the high types by the high quality product. 00:53:34.430 --> 00:53:36.590 align:middle line:90% What has to be true for that? 00:53:36.590 --> 00:53:40.760 align:middle line:84% It has to be that the th times delta 00:53:40.760 --> 00:53:43.080 align:middle line:84% s is the willingness to pay of high types 00:53:43.080 --> 00:53:45.090 align:middle line:90% to get the high quality product. 00:53:45.090 --> 00:53:48.057 align:middle line:84% tl delta s is the willingness to pay of the low types 00:53:48.057 --> 00:53:49.390 align:middle line:90% to get the high quality product. 00:53:49.390 --> 00:53:52.800 align:middle line:84% And it just has to be that the markups that 00:53:52.800 --> 00:53:57.570 align:middle line:84% result from different degrees of differentiation 00:53:57.570 --> 00:53:59.928 align:middle line:84% are exactly between the willingness 00:53:59.928 --> 00:54:01.720 align:middle line:84% to pay of the high types and the low types. 00:54:01.720 --> 00:54:05.640 align:middle line:84% So here, just in this case, under those assumptions, 00:54:05.640 --> 00:54:08.485 align:middle line:84% the IC constraints are all non-binding. 00:54:08.485 --> 00:54:10.860 align:middle line:84% There are no consumers who are indifferent between buying 00:54:10.860 --> 00:54:12.940 align:middle line:84% first class ticket or a coach class ticket 00:54:12.940 --> 00:54:15.690 align:middle line:90% in this model I've written down. 00:54:15.690 --> 00:54:18.420 align:middle line:84% Students are just in competition on a line 00:54:18.420 --> 00:54:19.743 align:middle line:90% to buy the low class tickets. 00:54:19.743 --> 00:54:21.660 align:middle line:84% These are contrasted by the high cost tickets. 00:54:21.660 --> 00:54:23.790 align:middle line:90% It requires a coincidence. 00:54:23.790 --> 00:54:27.900 align:middle line:84% The coincidence is that th minus tl 00:54:27.900 --> 00:54:30.690 align:middle line:84% is in this range that makes both of the IC constraints 00:54:30.690 --> 00:54:33.600 align:middle line:84% non-binding and makes the businessman buy the business 00:54:33.600 --> 00:54:36.815 align:middle line:84% class tickets, the students buy the coach class tickets. 00:54:36.815 --> 00:54:41.670 align:middle line:90% 00:54:41.670 --> 00:54:42.750 align:middle line:90% I don't know. 00:54:42.750 --> 00:54:45.420 align:middle line:84% This is a perfectly reasonable model of secondary oligopoly 00:54:45.420 --> 00:54:47.070 align:middle line:90% price discrimination. 00:54:47.070 --> 00:54:52.420 align:middle line:84% Again, it does say that competition determines markups. 00:54:52.420 --> 00:54:55.410 align:middle line:84% So it is a very different mathematical structure 00:54:55.410 --> 00:54:58.290 align:middle line:90% from what you get in the-- 00:54:58.290 --> 00:54:59.970 align:middle line:84% it's a different mathematical structure 00:54:59.970 --> 00:55:04.535 align:middle line:84% from what you get in the second degree monopoly 00:55:04.535 --> 00:55:05.910 align:middle line:84% price discrimination model, where 00:55:05.910 --> 00:55:08.670 align:middle line:84% it's the IC constraints and the manipulating quality 00:55:08.670 --> 00:55:11.535 align:middle line:90% to do things. 00:55:11.535 --> 00:55:14.070 align:middle line:84% It is the case that you can still get-- 00:55:14.070 --> 00:55:15.430 align:middle line:90% you can think of you can get-- 00:55:15.430 --> 00:55:19.560 align:middle line:90% 00:55:19.560 --> 00:55:23.113 align:middle line:90% I set these sh and sl exogenous. 00:55:23.113 --> 00:55:25.530 align:middle line:84% It could be that you can get manipulation of the qualities 00:55:25.530 --> 00:55:27.240 align:middle line:90% here where firms charge-- 00:55:27.240 --> 00:55:30.120 align:middle line:84% choose a quality level sl that prevents the businessmen 00:55:30.120 --> 00:55:31.650 align:middle line:84% from taking-- in some sense, they 00:55:31.650 --> 00:55:34.410 align:middle line:84% choose how far apart to make these lines so that this price 00:55:34.410 --> 00:55:37.290 align:middle line:90% discrimination scheme holds up. 00:55:37.290 --> 00:55:41.220 align:middle line:84% If sl were too high, I increase sl at some point, 00:55:41.220 --> 00:55:42.840 align:middle line:84% the IR constraint would start binding 00:55:42.840 --> 00:55:45.450 align:middle line:84% and the high types would jump down and start buying 00:55:45.450 --> 00:55:47.570 align:middle line:90% the low quality good instead. 00:55:47.570 --> 00:55:52.510 align:middle line:90% 00:55:52.510 --> 00:55:54.280 align:middle line:84% And we say interesting about this model 00:55:54.280 --> 00:55:56.840 align:middle line:84% is this model is also a model of multiple equilibria. 00:55:56.840 --> 00:55:59.170 align:middle line:84% Sometimes this model has two equilibria, 00:55:59.170 --> 00:56:02.620 align:middle line:84% one where the firms discriminate and one where the firms just 00:56:02.620 --> 00:56:05.280 align:middle line:84% sell the high quality good to everybody. 00:56:05.280 --> 00:56:08.950 align:middle line:84% In some sense, there's just this complementarity. 00:56:08.950 --> 00:56:12.310 align:middle line:84% We know these things have upward sloping reaction functions where 00:56:12.310 --> 00:56:15.760 align:middle line:84% you tend to-- you only go one half to one, 00:56:15.760 --> 00:56:19.760 align:middle line:84% but as your rival raises their price, you raise your price. 00:56:19.760 --> 00:56:22.240 align:middle line:84% That same thing means if your rival is spreading its prices 00:56:22.240 --> 00:56:23.980 align:middle line:84% for the high and low quality good apart, 00:56:23.980 --> 00:56:26.783 align:middle line:84% you want to spread your prices apart. 00:56:26.783 --> 00:56:28.450 align:middle line:84% So you can actually have two equilibria, 00:56:28.450 --> 00:56:31.660 align:middle line:84% one where the low quality good doesn't exist in the firm. 00:56:31.660 --> 00:56:35.000 align:middle line:84% They just compete for everybody selling the high quality good, 00:56:35.000 --> 00:56:38.710 align:middle line:84% and then the other where this discrimination occurs. 00:56:38.710 --> 00:56:40.360 align:middle line:84% But the discriminatory equilibrium 00:56:40.360 --> 00:56:41.860 align:middle line:84% is better for the firms than the one 00:56:41.860 --> 00:56:43.370 align:middle line:90% where they don't discriminate. 00:56:43.370 --> 00:56:48.070 align:middle line:90% 00:56:48.070 --> 00:56:50.175 align:middle line:90% Any questions on that model? 00:56:50.175 --> 00:56:53.770 align:middle line:90% 00:56:53.770 --> 00:56:54.270 align:middle line:90% OK. 00:56:54.270 --> 00:56:56.550 align:middle line:84% So then the final theory topic I want to get to 00:56:56.550 --> 00:57:00.120 align:middle line:84% is a little bit on add-on pricing. 00:57:00.120 --> 00:57:04.650 align:middle line:84% So at first let me go back to Lal and Matutes 00:57:04.650 --> 00:57:09.780 align:middle line:90% Journal of Business, 1994. 00:57:09.780 --> 00:57:12.090 align:middle line:84% Lal and Matutes were thinking of this question 00:57:12.090 --> 00:57:19.210 align:middle line:84% of, why is it that turkeys are on sale on Thanksgiving? 00:57:19.210 --> 00:57:20.850 align:middle line:84% Wouldn't you think that the one day 00:57:20.850 --> 00:57:23.310 align:middle line:84% of the year when turkey demand is a hundred times higher 00:57:23.310 --> 00:57:25.470 align:middle line:84% than it is any other day of the year, turkey prices 00:57:25.470 --> 00:57:28.680 align:middle line:90% would be high instead of low? 00:57:28.680 --> 00:57:33.360 align:middle line:84% And let's think about the following model. 00:57:33.360 --> 00:57:36.720 align:middle line:84% And think of this as a model of people competing. 00:57:36.720 --> 00:57:41.640 align:middle line:84% This is competing supermarkets, and the supermarkets 00:57:41.640 --> 00:57:46.930 align:middle line:84% are selling two different products, v1 and v2, 00:57:46.930 --> 00:57:49.750 align:middle line:84% and consumers actually all want to buy both products. 00:57:49.750 --> 00:57:51.960 align:middle line:84% They want to buy the turkey and they 00:57:51.960 --> 00:57:56.640 align:middle line:84% want to buy the stuffing or the cranberry sauce or whatever. 00:57:56.640 --> 00:58:03.150 align:middle line:84% And so what my utility is, I get v1 plus v2 minus p11 minus p12 00:58:03.150 --> 00:58:04.810 align:middle line:90% minus p theta. 00:58:04.810 --> 00:58:08.010 align:middle line:84% If I am type theta and I drive to store one 00:58:08.010 --> 00:58:11.420 align:middle line:84% and I buy both products, they are at these prices. 00:58:11.420 --> 00:58:18.590 align:middle line:84% Or I can drive to firm two and again get v1 minus p21 plus v2 00:58:18.590 --> 00:58:22.310 align:middle line:84% minus p22 minus t times 1 minus theta 00:58:22.310 --> 00:58:26.560 align:middle line:84% if I drive and buy the products from firm two. 00:58:26.560 --> 00:58:32.260 align:middle line:84% And then I can, if I want, split my shopping 00:58:32.260 --> 00:58:36.430 align:middle line:84% and get v1 minus p11 by buying the turkey at store one 00:58:36.430 --> 00:58:39.250 align:middle line:84% and then get v2 minus p22 from buying the cranberry 00:58:39.250 --> 00:58:41.230 align:middle line:90% sauce at firm two. 00:58:41.230 --> 00:58:45.920 align:middle line:84% But then I have incurred both driving costs. 00:58:45.920 --> 00:58:48.890 align:middle line:84% And so Lal and Matutes consider this a multi-stage game 00:58:48.890 --> 00:58:52.160 align:middle line:84% where what firms do is imagine there were 00:58:52.160 --> 00:58:53.960 align:middle line:90% per product advertising costs. 00:58:53.960 --> 00:58:55.610 align:middle line:84% You don't want to advertise the price 00:58:55.610 --> 00:58:57.237 align:middle line:90% of every product in your store. 00:58:57.237 --> 00:58:58.820 align:middle line:84% That's clearly true with supermarkets. 00:58:58.820 --> 00:59:00.780 align:middle line:84% You've got 10,000 products there. 00:59:00.780 --> 00:59:02.460 align:middle line:90% You can't advertise all of them. 00:59:02.460 --> 00:59:05.720 align:middle line:84% So in the extreme case, suppose you choose and advertise 00:59:05.720 --> 00:59:09.230 align:middle line:84% a price of one good, and then after choosing 00:59:09.230 --> 00:59:12.050 align:middle line:84% the price of your one good, you choose an unadvertised price 00:59:12.050 --> 00:59:14.630 align:middle line:84% for the other good which is kept secret 00:59:14.630 --> 00:59:17.150 align:middle line:84% until consumers arrive at the store. 00:59:17.150 --> 00:59:21.350 align:middle line:84% Consumers drive to one firm, incurring costs either t theta 00:59:21.350 --> 00:59:23.160 align:middle line:90% or t1 minus theta. 00:59:23.160 --> 00:59:25.340 align:middle line:84% When they get there, they learn the price 00:59:25.340 --> 00:59:27.590 align:middle line:90% for the unadvertised good. 00:59:27.590 --> 00:59:29.792 align:middle line:84% And then consumers, after learning 00:59:29.792 --> 00:59:31.250 align:middle line:84% the price of the unadvertised good, 00:59:31.250 --> 00:59:33.080 align:middle line:84% they either buy one or both products 00:59:33.080 --> 00:59:36.710 align:middle line:84% there or they decide to drive to the other firm 00:59:36.710 --> 00:59:38.695 align:middle line:90% and buy it at firm two instead. 00:59:38.695 --> 00:59:41.790 align:middle line:90% 00:59:41.790 --> 00:59:44.070 align:middle line:90% Proposition first. 00:59:44.070 --> 00:59:46.470 align:middle line:84% In the above model, equilibrium prices 00:59:46.470 --> 00:59:51.750 align:middle line:84% satisfy p1 star plus p2 star is c1 plus c2 plus t. 00:59:51.750 --> 00:59:53.700 align:middle line:84% Again, this is just the Hotelling competition 00:59:53.700 --> 00:59:55.980 align:middle line:90% on a line model. 00:59:55.980 --> 00:59:59.180 align:middle line:84% In equilibrium, the price of the bundle is going to be c plus t. 00:59:59.180 --> 01:00:02.420 align:middle line:90% 01:00:02.420 --> 01:00:04.760 align:middle line:84% But the individual prices are not 01:00:04.760 --> 01:00:07.190 align:middle line:90% like c plus t over 2 for each. 01:00:07.190 --> 01:00:13.220 align:middle line:84% The individual prices are p2 star equals v2 and pj1 star 01:00:13.220 --> 01:00:16.490 align:middle line:90% is c1 plus c2 plus t minus v2. 01:00:16.490 --> 01:00:21.590 align:middle line:84% So basically cranberry sauce is always 01:00:21.590 --> 01:00:23.880 align:middle line:84% incredibly expensive on Thanksgiving. 01:00:23.880 --> 01:00:25.700 align:middle line:84% You know, you have this advertised product 01:00:25.700 --> 01:00:27.590 align:middle line:90% and the unadvertised product. 01:00:27.590 --> 01:00:31.040 align:middle line:84% You show up at the store to buy the unadvertised product 01:00:31.040 --> 01:00:34.270 align:middle line:84% and you find that, wow, that is exactly my full willingness 01:00:34.270 --> 01:00:34.770 align:middle line:90% to pay. 01:00:34.770 --> 01:00:38.030 align:middle line:84% I'm basically indifferent to walking out versus spending 01:00:38.030 --> 01:00:40.130 align:middle line:90% $11 buying cranberries. 01:00:40.130 --> 01:00:45.160 align:middle line:84% But in equilibrium, both firms are charging 01:00:45.160 --> 01:00:46.490 align:middle line:90% $11 for the cranberries. 01:00:46.490 --> 01:00:48.570 align:middle line:84% So there's the advertised price for the turkey. 01:00:48.570 --> 01:00:50.860 align:middle line:84% The advertised price is that turkey is a great deal 01:00:50.860 --> 01:00:53.213 align:middle line:84% and it just is a great deal reflecting how much they're 01:00:53.213 --> 01:00:54.880 align:middle line:84% going to rip you off for everything else 01:00:54.880 --> 01:00:57.130 align:middle line:84% you buy when you get into the store to buy the turkey, 01:00:57.130 --> 01:00:58.280 align:middle line:90% that's a good deal. 01:00:58.280 --> 01:01:01.270 align:middle line:84% And in equilibrium, it's all neutral. 01:01:01.270 --> 01:01:03.022 align:middle line:84% In equilibrium it's as if you advertised 01:01:03.022 --> 01:01:04.480 align:middle line:84% the prices for everything, we would 01:01:04.480 --> 01:01:06.320 align:middle line:90% get exactly the same outcome. 01:01:06.320 --> 01:01:08.420 align:middle line:90% It's c plus t pricing. 01:01:08.420 --> 01:01:12.820 align:middle line:84% It's just in a distorted way where one good is really cheap 01:01:12.820 --> 01:01:15.860 align:middle line:84% and then the other goods are expensive. 01:01:15.860 --> 01:01:19.750 align:middle line:84% And actually, there's actually a paper on Thanksgiving noting 01:01:19.750 --> 01:01:22.450 align:middle line:84% that, in fact, it is true that turkeys are cheapest 01:01:22.450 --> 01:01:24.530 align:middle line:84% on Thanksgiving than any other day of the year 01:01:24.530 --> 01:01:26.230 align:middle line:84% and everything else in the grocery store 01:01:26.230 --> 01:01:29.890 align:middle line:84% is, on average, more expensive than any other day of the year. 01:01:29.890 --> 01:01:33.610 align:middle line:84% This is what firms seem to do is, you know, 01:01:33.610 --> 01:01:35.705 align:middle line:90% turkeys are a loss leader. 01:01:35.705 --> 01:01:37.580 align:middle line:84% Actually, why are turkeys a good loss leader? 01:01:37.580 --> 01:01:40.570 align:middle line:84% Milk is often also a loss leader at convenience stores. 01:01:40.570 --> 01:01:42.700 align:middle line:84% Like a gallon of milk seems relatively cheap 01:01:42.700 --> 01:01:44.710 align:middle line:84% compared to a bag of potato chips. 01:01:44.710 --> 01:01:46.840 align:middle line:84% Turkey and milk are good loss leaders 01:01:46.840 --> 01:01:49.900 align:middle line:84% because you don't want to have a loss leader where people come in 01:01:49.900 --> 01:01:53.770 align:middle line:84% and stock up and buy 23 of the thing that's advertised 01:01:53.770 --> 01:01:57.370 align:middle line:84% as a great deal and then not buy anything else at your store 01:01:57.370 --> 01:01:59.292 align:middle line:84% and exploit you and save it up and live off 01:01:59.292 --> 01:02:00.500 align:middle line:90% that for the next six months. 01:02:00.500 --> 01:02:03.290 align:middle line:84% Like ramen noodles would be a terrible loss leader 01:02:03.290 --> 01:02:06.650 align:middle line:84% because you'd get the student comes in, buys 01:02:06.650 --> 01:02:09.690 align:middle line:84% the case of ramen noodles, and just leaves with just that. 01:02:09.690 --> 01:02:13.400 align:middle line:84% But turkeys, you really don't want more than one. 01:02:13.400 --> 01:02:16.290 align:middle line:84% Gallons of milk, you really don't want more than one. 01:02:16.290 --> 01:02:19.160 align:middle line:84% So something that you can give people a very good deal on, 01:02:19.160 --> 01:02:21.170 align:middle line:84% they buy exactly one, and then you 01:02:21.170 --> 01:02:22.650 align:middle line:84% rip them off for everything else, 01:02:22.650 --> 01:02:25.430 align:middle line:84% that's how you choose your loss leaders. 01:02:25.430 --> 01:02:26.240 align:middle line:90% OK. 01:02:26.240 --> 01:02:28.040 align:middle line:84% But, you know, the Lal and Matutes model 01:02:28.040 --> 01:02:32.180 align:middle line:84% says that this is how loss leaders work, but they 01:02:32.180 --> 01:02:34.440 align:middle line:90% are neutral economically. 01:02:34.440 --> 01:02:38.448 align:middle line:84% It's not like banning them would do anyone any good. 01:02:38.448 --> 01:02:40.490 align:middle line:84% They're completely neutral for the firm's profits 01:02:40.490 --> 01:02:42.500 align:middle line:84% and neutral for consumers, and they do 01:02:42.500 --> 01:02:44.250 align:middle line:90% economize on advertising costs. 01:02:44.250 --> 01:02:46.910 align:middle line:90% So in that sense, they're good. 01:02:46.910 --> 01:02:48.510 align:middle line:90% I put Carmen's picture up there. 01:02:48.510 --> 01:02:50.080 align:middle line:84% I respect-- Carmen, at one point, 01:02:50.080 --> 01:02:53.960 align:middle line:84% was a very successful economist and then 01:02:53.960 --> 01:02:55.960 align:middle line:84% had been a professor at Berkeley and dropped out 01:02:55.960 --> 01:02:57.100 align:middle line:90% and became a novelist. 01:02:57.100 --> 01:02:58.840 align:middle line:84% And actually is a successful novelist, 01:02:58.840 --> 01:03:01.060 align:middle line:84% although she writes in Spanish so I've never 01:03:01.060 --> 01:03:02.150 align:middle line:90% read any of her books. 01:03:02.150 --> 01:03:06.910 align:middle line:84% But anyway, won several prizes for her fiction 01:03:06.910 --> 01:03:11.435 align:middle line:84% writing after she decided to have a second career. 01:03:11.435 --> 01:03:11.935 align:middle line:90% OK. 01:03:11.935 --> 01:03:14.620 align:middle line:90% 01:03:14.620 --> 01:03:15.620 align:middle line:90% Add-on pricing. 01:03:15.620 --> 01:03:21.700 align:middle line:84% So add-on pricing thinks about basically this loss leader 01:03:21.700 --> 01:03:26.800 align:middle line:84% idea in a vertical competitive price discrimination model. 01:03:26.800 --> 01:03:30.310 align:middle line:84% So there are many cases where firms 01:03:30.310 --> 01:03:33.760 align:middle line:84% that are practicing second degree price discrimination 01:03:33.760 --> 01:03:36.470 align:middle line:84% do have non visible prices for add-ons. 01:03:36.470 --> 01:03:38.650 align:middle line:84% So for instance, you go to a hotel 01:03:38.650 --> 01:03:41.350 align:middle line:84% and it's very easy to find the price of a hotel 01:03:41.350 --> 01:03:44.150 align:middle line:84% online and compare prices of hotels. 01:03:44.150 --> 01:03:47.170 align:middle line:84% It's very hard to compare what's the price 01:03:47.170 --> 01:03:49.210 align:middle line:84% of the beer in the mini-bar or what's 01:03:49.210 --> 01:03:51.670 align:middle line:84% the price of the sparkling water in the mini-bar, what's 01:03:51.670 --> 01:03:53.810 align:middle line:84% the price of a croissant in the hotel lobby 01:03:53.810 --> 01:03:57.670 align:middle line:84% if I'm in a hurry to get somewhere. 01:03:57.670 --> 01:03:59.530 align:middle line:84% Rental cars, it's very easy to compare 01:03:59.530 --> 01:04:00.830 align:middle line:90% the price of the rental car. 01:04:00.830 --> 01:04:04.360 align:middle line:84% It's very hard to figure out what will I pay for the car seat 01:04:04.360 --> 01:04:07.300 align:middle line:84% if I need a car seat for my infant when I get there. 01:04:07.300 --> 01:04:08.170 align:middle line:90% I don't know. 01:04:08.170 --> 01:04:10.128 align:middle line:84% I guess you guys are too young to have infants. 01:04:10.128 --> 01:04:14.200 align:middle line:84% But it's striking to you that you go to rent a car and you can 01:04:14.200 --> 01:04:17.780 align:middle line:84% rent a $20,000 car for 19.95 a day, 01:04:17.780 --> 01:04:20.900 align:middle line:84% and then you want to rent a $37 car seat because you don't want 01:04:20.900 --> 01:04:24.080 align:middle line:84% to lug it on the plane, and it's like 13.95 a day. 01:04:24.080 --> 01:04:27.140 align:middle line:84% And you're like, you know, how can this be that they can rent 01:04:27.140 --> 01:04:30.860 align:middle line:84% me a $20,000 car for $20 and then the $37 car seat, 01:04:30.860 --> 01:04:35.030 align:middle line:84% they're charging me $13 a day or making me carry this enormous 01:04:35.030 --> 01:04:37.190 align:middle line:84% piece of plastic on the airplane. 01:04:37.190 --> 01:04:39.770 align:middle line:84% Obviously, it's this same add-on pricing thing 01:04:39.770 --> 01:04:42.320 align:middle line:84% where it's very easy to compare the prices of the cars 01:04:42.320 --> 01:04:44.700 align:middle line:90% and not of the car seats. 01:04:44.700 --> 01:04:47.580 align:middle line:84% Therefore, the car seat is priced at v2. 01:04:47.580 --> 01:04:50.730 align:middle line:84% What is your disutility from carrying a car seat on a plane? 01:04:50.730 --> 01:04:52.370 align:middle line:90% That's what they'll charge you. 01:04:52.370 --> 01:04:55.767 align:middle line:84% And then the cars are, in some sense, the loss leader. 01:04:55.767 --> 01:04:58.100 align:middle line:84% They make money selling the car seats and the rental car 01:04:58.100 --> 01:05:01.280 align:middle line:90% insurance. 01:05:01.280 --> 01:05:03.480 align:middle line:84% Inkjet printers-- I mentioned earlier that, 01:05:03.480 --> 01:05:05.480 align:middle line:84% again, the inkjet printer is very cheap, 01:05:05.480 --> 01:05:07.610 align:middle line:84% but they've invented this inkjet printer 01:05:07.610 --> 01:05:10.020 align:middle line:84% that has almost no ink in it when you buy it. 01:05:10.020 --> 01:05:12.620 align:middle line:84% And then it's got the starter cartridge, 01:05:12.620 --> 01:05:15.650 align:middle line:84% and then the regular cartridge of the inkjet printer 01:05:15.650 --> 01:05:18.410 align:middle line:84% is much more expensive, and that's the hidden feature 01:05:18.410 --> 01:05:22.670 align:middle line:90% when you buy the cartridge. 01:05:22.670 --> 01:05:27.050 align:middle line:84% To think about how this works in vertical competition, 01:05:27.050 --> 01:05:29.280 align:middle line:84% Lal and Matutes was purely additive competition. 01:05:29.280 --> 01:05:30.810 align:middle line:90% It wasn't a vertical model. 01:05:30.810 --> 01:05:33.560 align:middle line:84% So imagine you have firms advertise 01:05:33.560 --> 01:05:35.930 align:middle line:84% prices for their low quality goods like the rental car 01:05:35.930 --> 01:05:36.895 align:middle line:90% without a car seat. 01:05:36.895 --> 01:05:40.360 align:middle line:90% 01:05:40.360 --> 01:05:43.172 align:middle line:84% Costs sl and sh both have cost c, 01:05:43.172 --> 01:05:45.130 align:middle line:84% as if the car seat is essentially free for them 01:05:45.130 --> 01:05:47.650 align:middle line:90% to provide. 01:05:47.650 --> 01:05:52.880 align:middle line:84% Consumers incur some cost to visit j and learn the price. 01:05:52.880 --> 01:05:55.480 align:middle line:84% So consumers have to visit the firm 01:05:55.480 --> 01:05:58.785 align:middle line:84% to learn the price of the upgrades. 01:05:58.785 --> 01:06:00.910 align:middle line:84% You have to spend some time on the website learning 01:06:00.910 --> 01:06:03.700 align:middle line:84% what it costs to get a car seat added to your rental car, 01:06:03.700 --> 01:06:06.700 align:middle line:84% and then that would let you upgrade in quality. 01:06:06.700 --> 01:06:09.860 align:middle line:84% And then again, let's do the competition on two lines. 01:06:09.860 --> 01:06:14.450 align:middle line:84% So epsilon i1, epsilon i2, independent. 01:06:14.450 --> 01:06:18.250 align:middle line:90% Again, same thing, this setup. 01:06:18.250 --> 01:06:21.400 align:middle line:84% So this set up where what you do is imagine this set up 01:06:21.400 --> 01:06:24.220 align:middle line:84% where you're only advertising the price of the low quality 01:06:24.220 --> 01:06:28.180 align:middle line:84% good, and then consumers have to pay something after they arrive 01:06:28.180 --> 01:06:30.520 align:middle line:84% or after they've spent some time investigating your firm 01:06:30.520 --> 01:06:35.740 align:middle line:84% to learn what it cost to upgrade to the high quality good. 01:06:35.740 --> 01:06:38.530 align:middle line:90% What happens? 01:06:38.530 --> 01:06:43.000 align:middle line:84% Well, there's an equilibrium where, like in Lal and Matutes, 01:06:43.000 --> 01:06:48.850 align:middle line:84% the price of the upgrade is the full value for the upgrade 01:06:48.850 --> 01:06:50.550 align:middle line:90% for the high types. 01:06:50.550 --> 01:06:52.370 align:middle line:84% So the price of the upgrade is what 01:06:52.370 --> 01:06:57.950 align:middle line:84% are the businessmen willing to pay for the rental car seat. 01:06:57.950 --> 01:07:00.170 align:middle line:84% And then price of the low quality good 01:07:00.170 --> 01:07:04.580 align:middle line:84% ends up being this, which I won't get into. 01:07:04.580 --> 01:07:08.805 align:middle line:84% But main applied conclusions are that profits in this game 01:07:08.805 --> 01:07:10.430 align:middle line:84% are higher than the profits in the game 01:07:10.430 --> 01:07:12.328 align:middle line:84% where pl and ph are both visible. 01:07:12.328 --> 01:07:14.120 align:middle line:84% So in this competitive price discrimination 01:07:14.120 --> 01:07:20.970 align:middle line:84% model, add-on pricing or these loss leaders are not neutral. 01:07:20.970 --> 01:07:22.550 align:middle line:90% They raise prices. 01:07:22.550 --> 01:07:24.290 align:middle line:84% And not only do they raise prices, 01:07:24.290 --> 01:07:28.100 align:middle line:84% but they make all consumers worse off than in the game 01:07:28.100 --> 01:07:30.680 align:middle line:84% where the low quality good didn't exist. 01:07:30.680 --> 01:07:33.410 align:middle line:84% So if I banned the production of rental cars 01:07:33.410 --> 01:07:36.500 align:middle line:84% that don't come with a car seat, if you ask for it for free, 01:07:36.500 --> 01:07:38.270 align:middle line:84% all consumers would be better off. 01:07:38.270 --> 01:07:39.680 align:middle line:84% The low types would be better off 01:07:39.680 --> 01:07:43.025 align:middle line:84% and the high types would be better off. 01:07:43.025 --> 01:07:44.900 align:middle line:84% The high types are better off because they 01:07:44.900 --> 01:07:47.938 align:middle line:84% would pay a low price without this add-on pricing scheme. 01:07:47.938 --> 01:07:50.480 align:middle line:84% The low types are better off because they're getting a better 01:07:50.480 --> 01:07:53.300 align:middle line:84% quality good instead of having to carry their car 01:07:53.300 --> 01:07:56.540 align:middle line:90% seats on the plane. 01:07:56.540 --> 01:07:59.280 align:middle line:84% I'm short on time, so why don't I skip that? 01:07:59.280 --> 01:08:00.950 align:middle line:84% But let me just give you the intuition 01:08:00.950 --> 01:08:05.930 align:middle line:84% for why that add-on pricing becomes non-neutral. 01:08:05.930 --> 01:08:08.150 align:middle line:84% Think about reducing your price to p star 01:08:08.150 --> 01:08:11.870 align:middle line:84% minus dp in the add-on model and doing the same in the model 01:08:11.870 --> 01:08:14.450 align:middle line:90% with no low quality good. 01:08:14.450 --> 01:08:17.420 align:middle line:84% In both models, what you lose from a price cut 01:08:17.420 --> 01:08:21.080 align:middle line:84% is one half dp from charging the lower prices to the half 01:08:21.080 --> 01:08:23.120 align:middle line:90% the consumers you're serving. 01:08:23.120 --> 01:08:26.689 align:middle line:84% In both models, what you're gaining from the price 01:08:26.689 --> 01:08:30.470 align:middle line:84% cut is the number of customers gained times the per customer 01:08:30.470 --> 01:08:34.000 align:middle line:84% profit on the marginal consumer that you gain. 01:08:34.000 --> 01:08:38.560 align:middle line:84% And so because, in both models, if you tried to make-- 01:08:38.560 --> 01:08:41.050 align:middle line:84% in both models, the first order condition, 01:08:41.050 --> 01:08:43.750 align:middle line:90% this has to equal this. 01:08:43.750 --> 01:08:46.510 align:middle line:84% The one half dp is the same in both, the number of customers 01:08:46.510 --> 01:08:49.029 align:middle line:84% gained in both, therefore, the per customer profit 01:08:49.029 --> 01:08:52.050 align:middle line:84% on the marginal consumer is the same in both models. 01:08:52.050 --> 01:08:54.609 align:middle line:84% So in both models, the per customer profit 01:08:54.609 --> 01:08:59.500 align:middle line:84% you make on the marginal consumer is the same. 01:08:59.500 --> 01:09:04.670 align:middle line:84% But profits are based on profits on the average consumer, not 01:09:04.670 --> 01:09:06.500 align:middle line:90% the marginal consumer. 01:09:06.500 --> 01:09:08.720 align:middle line:84% And if you think about what goes on in a model 01:09:08.720 --> 01:09:12.770 align:middle line:84% like this, when I cut my price, who do I get by-- 01:09:12.770 --> 01:09:16.520 align:middle line:84% I offer my rental car for $1 less than everybody else's 01:09:16.520 --> 01:09:17.069 align:middle line:90% rental car. 01:09:17.069 --> 01:09:21.170 align:middle line:84% My rental car is 38.95 a day instead of 39.95 a day. 01:09:21.170 --> 01:09:22.939 align:middle line:90% I get all the cheapskates. 01:09:22.939 --> 01:09:25.340 align:middle line:84% And I get all the cheapskates where the low types 01:09:25.340 --> 01:09:27.710 align:middle line:84% were willing to carry their car seat on a plane 01:09:27.710 --> 01:09:31.279 align:middle line:84% and spend extra time waiting at the gate to get their car seat 01:09:31.279 --> 01:09:35.960 align:middle line:84% and carrying it through the airport to save the 13.95 a day. 01:09:35.960 --> 01:09:41.240 align:middle line:84% And so because the cheapskates are an adversely selected 01:09:41.240 --> 01:09:43.800 align:middle line:84% population relative to the full population, 01:09:43.800 --> 01:09:47.029 align:middle line:84% so when I cut my price in this add-on model, 01:09:47.029 --> 01:09:50.200 align:middle line:90% I get a bunch of cheapskates. 01:09:50.200 --> 01:09:53.033 align:middle line:84% If the marginal profit-- so in the add-on model, 01:09:53.033 --> 01:09:54.700 align:middle line:84% there's this big gap between the profits 01:09:54.700 --> 01:09:56.408 align:middle line:84% on the marginal consumers and the profits 01:09:56.408 --> 01:09:58.390 align:middle line:84% on the average consumers, whereas 01:09:58.390 --> 01:10:01.360 align:middle line:84% in the single good product, the marginal and average consumers 01:10:01.360 --> 01:10:03.890 align:middle line:90% are exactly the same. 01:10:03.890 --> 01:10:06.320 align:middle line:84% So what we've done is we've just created 01:10:06.320 --> 01:10:09.230 align:middle line:84% this gap between marginal consumers and average consumers, 01:10:09.230 --> 01:10:13.130 align:middle line:84% and that ends up raising equilibrium prices. 01:10:13.130 --> 01:10:16.250 align:middle line:84% Sometimes you've created an adverse selection problem, 01:10:16.250 --> 01:10:19.190 align:middle line:84% and that adverse selection problem 01:10:19.190 --> 01:10:21.890 align:middle line:84% makes it hard for firms to price compete against other firms, 01:10:21.890 --> 01:10:23.780 align:middle line:84% because if you cut prices, you just 01:10:23.780 --> 01:10:27.175 align:middle line:84% get cheapskates on whom you lose money renting the cars so cheap. 01:10:27.175 --> 01:10:30.410 align:middle line:90% 01:10:30.410 --> 01:10:31.890 align:middle line:90% Questions? 01:10:31.890 --> 01:10:32.390 align:middle line:90% No? 01:10:32.390 --> 01:10:34.340 align:middle line:90% OK. 01:10:34.340 --> 01:10:36.380 align:middle line:84% Leaves me with 10 minutes, and I said 01:10:36.380 --> 01:10:39.860 align:middle line:84% I would do this paper very quickly because this really 01:10:39.860 --> 01:10:42.620 align:middle line:84% follows very much the methodology Tobias is talking 01:10:42.620 --> 01:10:45.840 align:middle line:90% about on Monday of next week. 01:10:45.840 --> 01:10:51.440 align:middle line:84% So Miller and Weinberg is a very similar paper to Bresnahan, 01:10:51.440 --> 01:10:57.860 align:middle line:84% and it's a much more modern, well-done, frontier quality 01:10:57.860 --> 01:11:00.230 align:middle line:90% piece of applied work. 01:11:00.230 --> 01:11:04.370 align:middle line:84% The applied question is whether the rise in US beer 01:11:04.370 --> 01:11:08.210 align:middle line:84% prices following the 2008 joint venture of Miller and Coors 01:11:08.210 --> 01:11:11.450 align:middle line:84% can be explained as a consequence of just static price 01:11:11.450 --> 01:11:14.870 align:middle line:84% competition or if we need some degree of cooperation 01:11:14.870 --> 01:11:18.170 align:middle line:84% or collusion among the firms to understand what happens. 01:11:18.170 --> 01:11:20.360 align:middle line:84% And so what it does is it estimates 01:11:20.360 --> 01:11:25.940 align:middle line:84% a model that assumes that the firms were paying static-- 01:11:25.940 --> 01:11:27.860 align:middle line:84% matching with static differentiated product 01:11:27.860 --> 01:11:33.480 align:middle line:84% competition before the merger, and then after the merger, 01:11:33.480 --> 01:11:37.920 align:middle line:84% MillerCoors and Anheuser-Busch are partially cooperative. 01:11:37.920 --> 01:11:41.250 align:middle line:84% And then the partial cooperative model, what it does is-- 01:11:41.250 --> 01:11:43.270 align:middle line:84% Bresnahan had three or four models. 01:11:43.270 --> 01:11:45.040 align:middle line:90% They have a continuum of models. 01:11:45.040 --> 01:11:47.400 align:middle line:84% The continuum of models is assume 01:11:47.400 --> 01:11:50.190 align:middle line:84% it's as if each firm is maximizing their own profit 01:11:50.190 --> 01:11:53.230 align:middle line:84% plus kappa times the profit of the other firms. 01:11:53.230 --> 01:11:56.010 align:middle line:84% So kappa equals one would be full collusion, 01:11:56.010 --> 01:11:59.370 align:middle line:84% kappa equals zero would be static competition, 01:11:59.370 --> 01:12:03.450 align:middle line:84% and kappa in between is some degree of partial collusion 01:12:03.450 --> 01:12:05.970 align:middle line:90% between the firms. 01:12:05.970 --> 01:12:08.580 align:middle line:84% And then what they note is that in this model, 01:12:08.580 --> 01:12:10.920 align:middle line:84% with this continuum of models, you 01:12:10.920 --> 01:12:13.710 align:middle line:84% can test whether kappa equals zero to say, 01:12:13.710 --> 01:12:17.130 align:middle line:84% does the static competition model explain what's going on 01:12:17.130 --> 01:12:23.160 align:middle line:84% or do I need to have some degree of collusion starting 01:12:23.160 --> 01:12:26.430 align:middle line:84% after the MillerCoors merger to understand 01:12:26.430 --> 01:12:28.770 align:middle line:90% why prices are what they are? 01:12:28.770 --> 01:12:31.650 align:middle line:84% So anyway, I won't go into the details here. 01:12:31.650 --> 01:12:36.660 align:middle line:84% But background, I think even graduate students drink 01:12:36.660 --> 01:12:40.440 align:middle line:84% craft beer these days, but despite the incredible rise 01:12:40.440 --> 01:12:44.550 align:middle line:84% in craft beer, most Americans still drink really cheap beer. 01:12:44.550 --> 01:12:49.073 align:middle line:84% So most Americans drink Bud Light, Coors Light, Miller Lite. 01:12:49.073 --> 01:12:50.490 align:middle line:84% At least most sales because people 01:12:50.490 --> 01:12:53.700 align:middle line:90% buy Bud Light in 30 packs. 01:12:53.700 --> 01:12:58.170 align:middle line:84% But there's been tremendous consolidation of beer ownership 01:12:58.170 --> 01:13:00.700 align:middle line:90% in the past 20 years globally. 01:13:00.700 --> 01:13:02.790 align:middle line:84% So basically, in every country around the world, 01:13:02.790 --> 01:13:05.430 align:middle line:84% the same three companies sell all the beer. 01:13:05.430 --> 01:13:08.497 align:middle line:84% And so the same companies that sell Anheuser-Busch 01:13:08.497 --> 01:13:10.080 align:middle line:84% in the United States are also the ones 01:13:10.080 --> 01:13:13.590 align:middle line:84% that sell Beck's in Germany and sell Stella Artois in England 01:13:13.590 --> 01:13:14.440 align:middle line:90% and whatever. 01:13:14.440 --> 01:13:18.095 align:middle line:84% But these companies have put together portfolios of beers. 01:13:18.095 --> 01:13:21.100 align:middle line:90% 01:13:21.100 --> 01:13:25.120 align:middle line:84% Prior to the 2008 MillerCoors joint venture, 01:13:25.120 --> 01:13:29.450 align:middle line:84% Anheuser-Busch, InBev had a market share of about 35%, 01:13:29.450 --> 01:13:32.750 align:middle line:84% Miller had a market share of 18%, 01:13:32.750 --> 01:13:35.330 align:middle line:90% Coors had a market share of 11%. 01:13:35.330 --> 01:13:37.870 align:middle line:84% And again, those are market shares in the whole beer market. 01:13:37.870 --> 01:13:40.100 align:middle line:84% If you take away the premium beers, 01:13:40.100 --> 01:13:42.340 align:middle line:84% there are many submarkets in which those firms 01:13:42.340 --> 01:13:44.980 align:middle line:90% have much higher shares. 01:13:44.980 --> 01:13:49.480 align:middle line:84% And despite the fact that these were the overall market shares, 01:13:49.480 --> 01:13:53.950 align:middle line:84% Miller and Coors were, in 2008, allowed to form a joint venture 01:13:53.950 --> 01:13:56.650 align:middle line:90% and sell their beer together. 01:13:56.650 --> 01:14:00.868 align:middle line:84% So MillerCoors argued that, yes, it looks like the second 01:14:00.868 --> 01:14:02.410 align:middle line:84% and third biggest firms in the market 01:14:02.410 --> 01:14:06.550 align:middle line:84% are merging together to create a three firm competition to two 01:14:06.550 --> 01:14:09.970 align:middle line:84% firm competition, but this is going to be good for consumers 01:14:09.970 --> 01:14:12.822 align:middle line:84% because there's going to be enough cost savings that we're 01:14:12.822 --> 01:14:14.780 align:middle line:84% going to have an incentive to lower our prices. 01:14:14.780 --> 01:14:17.747 align:middle line:84% Particularly, Coors was mostly brewed in Colorado 01:14:17.747 --> 01:14:20.330 align:middle line:84% and then had to be shipped all across the United States, which 01:14:20.330 --> 01:14:21.630 align:middle line:90% is expensive. 01:14:21.630 --> 01:14:23.930 align:middle line:84% They're like, if only we could produce all the Coors 01:14:23.930 --> 01:14:26.097 align:middle line:84% beer in all those Miller plants that are distributed 01:14:26.097 --> 01:14:28.160 align:middle line:84% around the country and then share trucks 01:14:28.160 --> 01:14:30.800 align:middle line:84% shipping the Miller and Coors to all the beer distributors, 01:14:30.800 --> 01:14:33.170 align:middle line:84% then we'd save on the trucking costs, 01:14:33.170 --> 01:14:34.590 align:middle line:90% we'd save on the shipping costs. 01:14:34.590 --> 01:14:37.610 align:middle line:84% Costs will go down so much that consumers will be better off. 01:14:37.610 --> 01:14:42.350 align:middle line:84% And the justice department approved the merger. 01:14:42.350 --> 01:14:45.650 align:middle line:84% Anyway, you know, very beautiful graph 01:14:45.650 --> 01:14:47.990 align:middle line:84% just giving the basic thing of, did 01:14:47.990 --> 01:14:50.220 align:middle line:84% what the firms tell the justice department come true? 01:14:50.220 --> 01:14:51.595 align:middle line:84% Here are the graphs of the prices 01:14:51.595 --> 01:14:53.810 align:middle line:84% for Miller Lite, Bud Light, Coors Light 01:14:53.810 --> 01:14:55.770 align:middle line:84% leading up to the time of the merger. 01:14:55.770 --> 01:15:00.620 align:middle line:84% And you can see beer prices were trending down. 01:15:00.620 --> 01:15:04.130 align:middle line:84% After the merger, there was this temporary sale on Miller Lite. 01:15:04.130 --> 01:15:06.860 align:middle line:84% But then by a couple months after the merger, 01:15:06.860 --> 01:15:09.150 align:middle line:90% prices went up there. 01:15:09.150 --> 01:15:13.170 align:middle line:84% So there's just a very clear departure from trend, 01:15:13.170 --> 01:15:15.510 align:middle line:84% and the prices of Bud Light, Miller Lite, 01:15:15.510 --> 01:15:18.545 align:middle line:84% Coors Light jump after the MillerCoors merger. 01:15:18.545 --> 01:15:21.435 align:middle line:90% 01:15:21.435 --> 01:15:24.490 align:middle line:84% Their competitor's prices didn't do so much. 01:15:24.490 --> 01:15:27.000 align:middle line:84% So the two largest brands not sold 01:15:27.000 --> 01:15:30.570 align:middle line:84% by Anheuser-Busch, InBev, or MillerCoors 01:15:30.570 --> 01:15:33.450 align:middle line:90% are Heineken and Corona. 01:15:33.450 --> 01:15:37.930 align:middle line:84% Heineken and Corona prices, you don't notice-- 01:15:37.930 --> 01:15:40.620 align:middle line:84% maybe you look and see there's a little jump up here, 01:15:40.620 --> 01:15:43.380 align:middle line:84% but it seems like they're mostly on trend. 01:15:43.380 --> 01:15:45.720 align:middle line:84% So if we take that as a differences in differences thing 01:15:45.720 --> 01:15:47.520 align:middle line:84% to say there doesn't seem to be a difference in the cost 01:15:47.520 --> 01:15:49.650 align:middle line:84% of producing beer, if there's no difference 01:15:49.650 --> 01:15:51.930 align:middle line:84% in the cost of producing beer, what you get 01:15:51.930 --> 01:15:55.410 align:middle line:84% is that the 12 packs of the flagship brands 01:15:55.410 --> 01:15:58.000 align:middle line:90% went up by about 10%. 01:15:58.000 --> 01:16:00.370 align:middle line:84% Overall, prices went up by about 5%. 01:16:00.370 --> 01:16:02.790 align:middle line:90% Basically it's as if they-- 01:16:02.790 --> 01:16:05.070 align:middle line:84% if a 12 pack of Bud Light Costs $10, 01:16:05.070 --> 01:16:06.940 align:middle line:84% it's like they went from 10 to 11. 01:16:06.940 --> 01:16:09.292 align:middle line:90% The 20 packs, it seems like-- 01:16:09.292 --> 01:16:12.790 align:middle line:84% the 30 packs, they seem like they went from 18 to 19. 01:16:12.790 --> 01:16:15.720 align:middle line:84% But it's basically like prices go up by 5% on average, 01:16:15.720 --> 01:16:19.545 align:middle line:84% but 10% on the 12 pack of the standard flagship beers. 01:16:19.545 --> 01:16:22.540 align:middle line:90% 01:16:22.540 --> 01:16:25.090 align:middle line:90% So why does this happen? 01:16:25.090 --> 01:16:29.480 align:middle line:84% Again, there are three things going on here. 01:16:29.480 --> 01:16:32.770 align:middle line:84% There are-- costs are presumably going down 01:16:32.770 --> 01:16:36.370 align:middle line:84% because they're now brewing Coors all over the country. 01:16:36.370 --> 01:16:39.190 align:middle line:84% But MillerCoors has this unilateral incentive 01:16:39.190 --> 01:16:41.650 align:middle line:84% to raise its prices because it knows 01:16:41.650 --> 01:16:43.915 align:middle line:84% that if someone goes in to buy Coors Light 01:16:43.915 --> 01:16:45.790 align:middle line:84% and the price is now higher than they expect, 01:16:45.790 --> 01:16:47.540 align:middle line:84% they may switch to Miller Lite, and that's 01:16:47.540 --> 01:16:49.360 align:middle line:90% not a bad thing anymore. 01:16:49.360 --> 01:16:52.750 align:middle line:84% And then there's this possible collusion, third effect. 01:16:52.750 --> 01:16:54.710 align:middle line:90% So anyway, what do they do? 01:16:54.710 --> 01:16:58.790 align:middle line:84% It's random coefficients, logit demand. 01:16:58.790 --> 01:17:00.790 align:middle line:84% Tobias will tell you more about what that means. 01:17:00.790 --> 01:17:06.130 align:middle line:84% But basically, your utility from buying a brand j of beer 01:17:06.130 --> 01:17:07.840 align:middle line:84% if you live in region r at time t 01:17:07.840 --> 01:17:10.690 align:middle line:84% is characteristics of the beers times 01:17:10.690 --> 01:17:13.570 align:middle line:84% beta minus a disutility from paying 01:17:13.570 --> 01:17:16.580 align:middle line:84% for the beers plus a bunch of fixed effect. 01:17:16.580 --> 01:17:19.840 align:middle line:84% The brand fixed effect, the time fixed effect, 01:17:19.840 --> 01:17:22.210 align:middle line:84% an unobserved quality of the beer, 01:17:22.210 --> 01:17:26.860 align:middle line:84% and this idiosyncratic epsilon like in the Hotelling model. 01:17:26.860 --> 01:17:32.170 align:middle line:84% Some people like Coors relative to Budweiser just taste wise. 01:17:32.170 --> 01:17:34.090 align:middle line:84% What's different from the standard logit model 01:17:34.090 --> 01:17:37.210 align:middle line:84% is it's a random coefficients where the different people have 01:17:37.210 --> 01:17:39.850 align:middle line:84% different preferences over the characteristics, 01:17:39.850 --> 01:17:42.010 align:middle line:84% and some people are more price sensitive 01:17:42.010 --> 01:17:44.120 align:middle line:84% and some people are less price sensitive. 01:17:44.120 --> 01:17:46.240 align:middle line:84% So the random coefficients mean that not everyone 01:17:46.240 --> 01:17:48.760 align:middle line:90% has the same coefficients. 01:17:48.760 --> 01:17:51.310 align:middle line:84% Some people like hoppiness and some people 01:17:51.310 --> 01:17:52.370 align:middle line:90% don't like hoppiness. 01:17:52.370 --> 01:17:54.260 align:middle line:84% Some people dislike paying money. 01:17:54.260 --> 01:17:56.890 align:middle line:84% Other people don't care so much about how much they're paying. 01:17:56.890 --> 01:17:57.850 align:middle line:90% OK. 01:17:57.850 --> 01:17:59.600 align:middle line:90% They estimate this model. 01:17:59.600 --> 01:18:02.680 align:middle line:84% They have some instruments for price I will skip over, 01:18:02.680 --> 01:18:05.810 align:middle line:84% but they find a bunch of charts like this. 01:18:05.810 --> 01:18:09.570 align:middle line:84% So this chart says these are price elasticities that 01:18:09.570 --> 01:18:12.760 align:middle line:84% says if you raise the price of Budweiser by 1%, 01:18:12.760 --> 01:18:17.810 align:middle line:84% sales of Budweiser go down by 4.4%, or sales of Bud Light. 01:18:17.810 --> 01:18:19.550 align:middle line:84% But then people will buy other things. 01:18:19.550 --> 01:18:24.220 align:middle line:84% And so in particular, Budweiser sales will go up by 0.3%, 01:18:24.220 --> 01:18:27.190 align:middle line:84% Coors sales will go up by over 0.316%, 01:18:27.190 --> 01:18:30.220 align:middle line:84% Coors Light sales will go up by 0.351%, 01:18:30.220 --> 01:18:33.040 align:middle line:84% Corona sales will go up by 2.279%. 01:18:33.040 --> 01:18:35.010 align:middle line:84% We have this sort of substitution matrix 01:18:35.010 --> 01:18:36.810 align:middle line:90% they've estimated. 01:18:36.810 --> 01:18:39.700 align:middle line:84% And then they say it's a two-step estimation. 01:18:39.700 --> 01:18:42.790 align:middle line:84% First, we estimate demand to find all the elasticities. 01:18:42.790 --> 01:18:48.570 align:middle line:84% Then we say what would happen under the different models 01:18:48.570 --> 01:18:50.160 align:middle line:90% of behavior. 01:18:50.160 --> 01:18:52.098 align:middle line:90% And let me see. 01:18:52.098 --> 01:18:53.140 align:middle line:90% Two things I want to say. 01:18:53.140 --> 01:18:58.260 align:middle line:84% One thing they find is that they estimate which kappa model best 01:18:58.260 --> 01:19:01.680 align:middle line:84% fits the data, and they find that the post-period kappas 01:19:01.680 --> 01:19:05.080 align:middle line:84% of like 0.25 to 0.37 best fit the data. 01:19:05.080 --> 01:19:08.490 align:middle line:84% So it's as if the firms Budweiser and Miller are 01:19:08.490 --> 01:19:10.080 align:middle line:84% putting weight on each other's prices 01:19:10.080 --> 01:19:12.210 align:middle line:84% and starting to price higher than 01:19:12.210 --> 01:19:16.510 align:middle line:90% seems optimal after the merger. 01:19:16.510 --> 01:19:20.980 align:middle line:90% And here's the counterfactual. 01:19:20.980 --> 01:19:26.670 align:middle line:84% So this is what beer prices of-- so 12 packs were, like, $9 01:19:26.670 --> 01:19:28.350 align:middle line:90% or whatever before the merger. 01:19:28.350 --> 01:19:31.020 align:middle line:84% So we have this long-term trend of declining prices 01:19:31.020 --> 01:19:33.150 align:middle line:90% of 12 packs of cheap beer. 01:19:33.150 --> 01:19:35.520 align:middle line:84% They think absent the merger, it would 01:19:35.520 --> 01:19:40.500 align:middle line:84% have continued along this path with some rises and falls. 01:19:40.500 --> 01:19:42.390 align:middle line:84% I don't know if that's due to costs. 01:19:42.390 --> 01:19:44.350 align:middle line:84% But then we have three things happening. 01:19:44.350 --> 01:19:48.840 align:middle line:84% The first is this unilateral incentive of MillerCoors 01:19:48.840 --> 01:19:50.795 align:middle line:84% to raise its prices because it knows 01:19:50.795 --> 01:19:53.170 align:middle line:84% that customers it loses when it raises the price of Coors 01:19:53.170 --> 01:19:56.010 align:middle line:84% go to Miller, and that's no longer a bad thing. 01:19:56.010 --> 01:19:59.810 align:middle line:84% That would cause prices to go up for this much. 01:19:59.810 --> 01:20:03.390 align:middle line:84% But then there were savings from Coors being cheaper to produce, 01:20:03.390 --> 01:20:05.720 align:middle line:84% and they say that the Coors savings would 01:20:05.720 --> 01:20:07.280 align:middle line:90% have been enough to offset-- 01:20:07.280 --> 01:20:09.487 align:middle line:84% that Coors was not lying before the merger 01:20:09.487 --> 01:20:11.570 align:middle line:84% when they said that our costs are going to go down 01:20:11.570 --> 01:20:13.640 align:middle line:84% by enough that it's going to offset our incentive 01:20:13.640 --> 01:20:15.530 align:middle line:84% to raise prices, and they would say 01:20:15.530 --> 01:20:17.510 align:middle line:90% the cost savings are this big. 01:20:17.510 --> 01:20:20.240 align:middle line:84% So that would have caused this to happen. 01:20:20.240 --> 01:20:23.840 align:middle line:84% However, the third effect is that it 01:20:23.840 --> 01:20:26.330 align:middle line:84% looks like they're now coordinating their pricing, 01:20:26.330 --> 01:20:28.250 align:middle line:84% and that increase in collusion is 01:20:28.250 --> 01:20:31.670 align:middle line:84% enough to go from these triangles to this line up here. 01:20:31.670 --> 01:20:33.710 align:middle line:84% So they're arguing is that there are three 01:20:33.710 --> 01:20:36.310 align:middle line:90% roughly equally sized effect. 01:20:36.310 --> 01:20:40.060 align:middle line:84% The MillerCoors internalizing, it's 01:20:40.060 --> 01:20:43.360 align:middle line:84% doing multi-product monopoly pricing, the cost savings 01:20:43.360 --> 01:20:47.530 align:middle line:84% on producing Coors, and the Anheuser-Busch Miller 01:20:47.530 --> 01:20:49.510 align:middle line:90% cooperative behavior. 01:20:49.510 --> 01:20:52.420 align:middle line:84% All roughly equal sizes but combining together 01:20:52.420 --> 01:20:54.195 align:middle line:84% to cause this increase in the 12 packs. 01:20:54.195 --> 01:20:57.600 align:middle line:90% 01:20:57.600 --> 01:21:01.290 align:middle line:84% Anyway, I think, to appreciate-- the paper has some strengths 01:21:01.290 --> 01:21:02.560 align:middle line:90% and weaknesses or whatever. 01:21:02.560 --> 01:21:06.240 align:middle line:84% To appreciate those it's easier after Tobias's Monday lecture, 01:21:06.240 --> 01:21:07.980 align:middle line:84% so I figured I would just skip those 01:21:07.980 --> 01:21:12.210 align:middle line:84% and maybe you can go back and reread the paper after Tobias's 01:21:12.210 --> 01:21:15.383 align:middle line:84% talk next Monday and think about what 01:21:15.383 --> 01:21:17.550 align:middle line:84% do we think of the various aspects of the estimation 01:21:17.550 --> 01:21:19.150 align:middle line:90% strategy. 01:21:19.150 --> 01:21:24.430 align:middle line:84% But it is a Bresnahan-like idea of estimate multiple models, 01:21:24.430 --> 01:21:26.230 align:middle line:84% see which ones fit best, and that's 01:21:26.230 --> 01:21:30.030 align:middle line:84% how I determine how firms are behaving. 01:21:30.030 --> 01:21:30.530 align:middle line:90% OK. 01:21:30.530 --> 01:21:32.740 align:middle line:90% Any questions? 01:21:32.740 --> 01:21:34.610 align:middle line:90% OK, then we are done. 01:21:34.610 --> 01:21:37.310 align:middle line:90% I guess I get a week off. 01:21:37.310 --> 01:21:40.720 align:middle line:90% Tobias is going to be here. 01:21:40.720 --> 01:21:43.510 align:middle line:84% Gandhi and Nevo is published in whatever handbook 01:21:43.510 --> 01:21:45.710 align:middle line:84% of IO that's expensive and hard to find. 01:21:45.710 --> 01:21:47.830 align:middle line:84% But if you look at the same exact document 01:21:47.830 --> 01:21:52.840 align:middle line:84% as NBER working paper 29257, so if you look at that version, 01:21:52.840 --> 01:21:55.050 align:middle line:90% that's easy to find. 01:21:55.050 --> 01:22:07.000 align:middle line:90%