WEBVTT 00:00:00.000 --> 00:00:03.409 align:middle line:84% [SQUEAKING] [RUSTLING] [CLICKING] 00:00:03.409 --> 00:00:17.570 align:middle line:90% 00:00:17.570 --> 00:00:23.860 align:middle line:84% GLENN ELLISON: Today I'm going to do some empirical literature 00:00:23.860 --> 00:00:24.360 align:middle line:90% on entry. 00:00:24.360 --> 00:00:28.640 align:middle line:84% And I wanted to start with this really classic paper 00:00:28.640 --> 00:00:31.100 align:middle line:90% by Bresnahan and Reiss. 00:00:31.100 --> 00:00:34.760 align:middle line:84% And Bresnahan and Reiss-- it's actually-- it's an entry paper, 00:00:34.760 --> 00:00:37.500 align:middle line:84% but it's really a static competition paper. 00:00:37.500 --> 00:00:41.420 align:middle line:84% So the research question here is, how empirically 00:00:41.420 --> 00:00:43.700 align:middle line:84% does level of markups and competition 00:00:43.700 --> 00:00:46.130 align:middle line:84% change when you have more firms in the market? 00:00:46.130 --> 00:00:47.930 align:middle line:90% Is it true that-- 00:00:47.930 --> 00:00:50.030 align:middle line:84% I guess maybe it is the entry graph. 00:00:50.030 --> 00:00:52.147 align:middle line:84% What does this graph look like when-- 00:00:52.147 --> 00:00:56.260 align:middle line:90% 00:00:56.260 --> 00:00:58.460 align:middle line:84% what does the price/cost margin look 00:00:58.460 --> 00:01:00.680 align:middle line:90% like as the number of firms-- 00:01:00.680 --> 00:01:05.060 align:middle line:84% or their profits look like as the number of firms enters. 00:01:05.060 --> 00:01:08.300 align:middle line:84% And Bresnahan and Reiss are starting from this classic NEIO 00:01:08.300 --> 00:01:12.260 align:middle line:84% approach of thinking that it's very hard to estimate profits 00:01:12.260 --> 00:01:15.140 align:middle line:84% because it's very hard to estimate costs. 00:01:15.140 --> 00:01:17.970 align:middle line:84% In many businesses, there are lots of intangible costs. 00:01:17.970 --> 00:01:20.710 align:middle line:84% And you can't just [INAUDIBLE] to get that. 00:01:20.710 --> 00:01:23.445 align:middle line:84% So we want to use something that's readily observable, 00:01:23.445 --> 00:01:26.710 align:middle line:84% and the idea is that entry is very clearly observable, 00:01:26.710 --> 00:01:31.020 align:middle line:84% whereas the [INAUDIBLE] costs observed are not. 00:01:31.020 --> 00:01:32.700 align:middle line:84% The other idea in Bresnahan and Reiss 00:01:32.700 --> 00:01:35.910 align:middle line:84% is that in IO, one of the challenges is always, 00:01:35.910 --> 00:01:38.220 align:middle line:84% how do you get a cross enough-- a big enough cross 00:01:38.220 --> 00:01:42.810 align:middle line:84% section of observations to run the significant statistical 00:01:42.810 --> 00:01:43.690 align:middle line:90% estimates? 00:01:43.690 --> 00:01:46.170 align:middle line:84% And Bresnahan and Reiss had this idea 00:01:46.170 --> 00:01:48.880 align:middle line:84% of looking at isolated small towns. 00:01:48.880 --> 00:01:50.715 align:middle line:84% And so if you look at the United States, 00:01:50.715 --> 00:01:52.590 align:middle line:84% one thing that's nice about the United States 00:01:52.590 --> 00:01:54.340 align:middle line:90% is it's very, very big. 00:01:54.340 --> 00:01:55.980 align:middle line:84% And there are many parts of it where 00:01:55.980 --> 00:01:57.580 align:middle line:90% there aren't a lot of people. 00:01:57.580 --> 00:02:00.720 align:middle line:84% And so that gives you a lot of towns where there are towns, 00:02:00.720 --> 00:02:02.980 align:middle line:84% and then there's very little around them. 00:02:02.980 --> 00:02:05.760 align:middle line:90% And so if you think about-- 00:02:05.760 --> 00:02:07.890 align:middle line:84% you go search around the United States, 00:02:07.890 --> 00:02:10.720 align:middle line:84% it's a big enough country that you can-- if you want towns-- 00:02:10.720 --> 00:02:12.360 align:middle line:84% and in Bresnahan and Reiss's case, 00:02:12.360 --> 00:02:13.730 align:middle line:90% what they're looking for is-- 00:02:13.730 --> 00:02:16.710 align:middle line:90% I forget what the cut-offs are. 00:02:16.710 --> 00:02:19.220 align:middle line:84% You have towns that have at least-- 00:02:19.220 --> 00:02:23.340 align:middle line:84% you have cities that are isolated in the sense of-- 00:02:23.340 --> 00:02:27.640 align:middle line:84% I think they're all 100,000 miles from any city of 100,000-- 00:02:27.640 --> 00:02:30.670 align:middle line:84% they're 100 miles from any city of 100,000 people. 00:02:30.670 --> 00:02:33.840 align:middle line:84% And they are at least 20 miles from any city of 1,000 people. 00:02:33.840 --> 00:02:35.340 align:middle line:84% So these are all towns that are just 00:02:35.340 --> 00:02:38.100 align:middle line:84% completely isolated from anything else that you would-- 00:02:38.100 --> 00:02:40.027 align:middle line:90% any other population centers. 00:02:40.027 --> 00:02:41.860 align:middle line:84% And they note that there are a lot of them-- 00:02:41.860 --> 00:02:45.690 align:middle line:90% and, in fact, I think they-- 00:02:45.690 --> 00:02:48.300 align:middle line:84% Peter tells stories of driving around and wanting 00:02:48.300 --> 00:02:51.210 align:middle line:84% to verify the data by just driving around 00:02:51.210 --> 00:02:53.970 align:middle line:84% to all of these small towns and then asking them, 00:02:53.970 --> 00:02:55.840 align:middle line:84% where can I get a plumber in this town? 00:02:55.840 --> 00:02:57.215 align:middle line:84% Where can I get-- where can I buy 00:02:57.215 --> 00:02:59.100 align:middle line:90% tires for my car or whatever? 00:02:59.100 --> 00:03:00.150 align:middle line:90% And so what they did is-- 00:03:00.150 --> 00:03:01.590 align:middle line:84% most of the data collection was done 00:03:01.590 --> 00:03:02.798 align:middle line:90% through business directories. 00:03:02.798 --> 00:03:05.040 align:middle line:84% They basically just looked at a US map, 00:03:05.040 --> 00:03:06.972 align:middle line:84% found places that were completely isolated. 00:03:06.972 --> 00:03:08.430 align:middle line:84% And then we used to have this thing 00:03:08.430 --> 00:03:11.400 align:middle line:84% called yellow pages, which was like a phone directory, where 00:03:11.400 --> 00:03:13.960 align:middle line:84% you would look up phone numbers in a book. 00:03:13.960 --> 00:03:16.640 align:middle line:84% And so people would use these books to look up phone numbers. 00:03:16.640 --> 00:03:19.650 align:middle line:84% They would find out who was advertising as a plumber in each 00:03:19.650 --> 00:03:20.490 align:middle line:90% of these towns. 00:03:20.490 --> 00:03:23.560 align:middle line:84% They also use just business directories. 00:03:23.560 --> 00:03:26.280 align:middle line:84% And then they visited them to see, 00:03:26.280 --> 00:03:28.680 align:middle line:84% was this data being accurately collected? 00:03:28.680 --> 00:03:31.392 align:middle line:90% 00:03:31.392 --> 00:03:34.920 align:middle line:84% The paper-- it's a single cross-section. 00:03:34.920 --> 00:03:38.820 align:middle line:84% They don't observe firms entering in or exiting. 00:03:38.820 --> 00:03:42.200 align:middle line:84% But what they're going to infer is that sometime prior 00:03:42.200 --> 00:03:44.480 align:middle line:84% to the point at which they're collecting the data, 00:03:44.480 --> 00:03:48.710 align:middle line:90% this firm's-- 00:03:48.710 --> 00:03:51.570 align:middle line:84% classic game is firms choose in/out. 00:03:51.570 --> 00:03:57.645 align:middle line:84% And then firms choose prices or quantity star of n-- 00:03:57.645 --> 00:03:59.170 align:middle line:90% q star of n. 00:03:59.170 --> 00:04:02.930 align:middle line:90% 00:04:02.930 --> 00:04:04.240 align:middle line:90% They paid a fixed cost. 00:04:04.240 --> 00:04:05.990 align:middle line:84% They assume that this game has been played 00:04:05.990 --> 00:04:07.080 align:middle line:90% at some point in the past. 00:04:07.080 --> 00:04:09.290 align:middle line:84% And so what we're observing in each market is, 00:04:09.290 --> 00:04:12.520 align:middle line:84% what's the equilibrium number of firms in that market? 00:04:12.520 --> 00:04:16.640 align:middle line:90% 00:04:16.640 --> 00:04:18.920 align:middle line:84% The basic thing the paper then tries to do 00:04:18.920 --> 00:04:21.440 align:middle line:84% is say, if we can observe the number of firms 00:04:21.440 --> 00:04:23.870 align:middle line:84% in a large cross-section of markets, that's 00:04:23.870 --> 00:04:25.940 align:middle line:84% going to give us a means to estimate profit 00:04:25.940 --> 00:04:31.980 align:middle line:84% levels without any data on prices or on costs or on demand. 00:04:31.980 --> 00:04:35.133 align:middle line:90% And so consider of basic model. 00:04:35.133 --> 00:04:36.300 align:middle line:90% You have the entry decision. 00:04:36.300 --> 00:04:37.860 align:middle line:84% You have a firm that has fixed costs, 00:04:37.860 --> 00:04:41.170 align:middle line:84% F. There's going to be symmetric competition. 00:04:41.170 --> 00:04:45.060 align:middle line:84% So if there are N firms in a market with S people, 00:04:45.060 --> 00:04:48.510 align:middle line:84% you're going to get S over N-- so the total number of customers 00:04:48.510 --> 00:04:53.700 align:middle line:84% divided by the number of firms times pN minus c D of pN. 00:04:53.700 --> 00:04:56.730 align:middle line:84% And so they're going to call that's S over N times pi N. 00:04:56.730 --> 00:05:02.040 align:middle line:84% So pi N is the per population profit-- 00:05:02.040 --> 00:05:05.910 align:middle line:84% or per population served profit in this market. 00:05:05.910 --> 00:05:09.227 align:middle line:84% It reflects both sales you make to each customer-- there's 00:05:09.227 --> 00:05:11.310 align:middle line:84% obviously only some fraction of the consumers that 00:05:11.310 --> 00:05:12.160 align:middle line:90% are buying from you. 00:05:12.160 --> 00:05:15.060 align:middle line:84% And so this is profit per person that's 00:05:15.060 --> 00:05:16.690 align:middle line:90% in your part of the market. 00:05:16.690 --> 00:05:20.430 align:middle line:84% So if you have a town, and roughly, it's divided like this, 00:05:20.430 --> 00:05:23.140 align:middle line:84% and each firm gets one quarter of the market, 00:05:23.140 --> 00:05:27.300 align:middle line:84% it's your per customer profit for the market that you serve. 00:05:27.300 --> 00:05:29.500 align:middle line:84% We're going to have N firms in equilibrium 00:05:29.500 --> 00:05:33.070 align:middle line:84% if S over N times pi N is bigger than the fixed cost. 00:05:33.070 --> 00:05:36.010 align:middle line:84% But S over N plus 1 times pi N plus 1 00:05:36.010 --> 00:05:37.570 align:middle line:84% would be less than the fixed cost. 00:05:37.570 --> 00:05:39.120 align:middle line:84% And obviously, this is decreasing-- 00:05:39.120 --> 00:05:42.340 align:middle line:84% this is smaller than this, both because you're 00:05:42.340 --> 00:05:45.520 align:middle line:84% dividing the city into more fractions, 00:05:45.520 --> 00:05:47.680 align:middle line:84% and the per customer profit goes down 00:05:47.680 --> 00:05:50.670 align:middle line:90% because the price goes down. 00:05:50.670 --> 00:05:53.100 align:middle line:84% And what we're going to say is what asset would 00:05:53.100 --> 00:05:56.790 align:middle line:90% be the minimum size in which-- 00:05:56.790 --> 00:06:00.810 align:middle line:84% so, for instance, let's put N on this axis, 00:06:00.810 --> 00:06:03.120 align:middle line:84% and then you can put the number of doctors 00:06:03.120 --> 00:06:05.485 align:middle line:90% in a town on this axis. 00:06:05.485 --> 00:06:07.860 align:middle line:84% And what you would observe, if you could draw this graph, 00:06:07.860 --> 00:06:12.600 align:middle line:84% is towns that have no-- that are very small don't have a doctor. 00:06:12.600 --> 00:06:14.280 align:middle line:84% And then there's some threshold-- 00:06:14.280 --> 00:06:16.230 align:middle line:90% maybe it's like 800 people-- 00:06:16.230 --> 00:06:19.830 align:middle line:84% where now there's a doctor in the town. 00:06:19.830 --> 00:06:27.490 align:middle line:84% And then some other threshold where 2,000 people were. 00:06:27.490 --> 00:06:30.255 align:middle line:84% Now you have to have two doctors, and then 00:06:30.255 --> 00:06:32.713 align:middle line:84% some other threshold where you tend to have three doctors-- 00:06:32.713 --> 00:06:36.690 align:middle line:90% 00:06:36.690 --> 00:06:39.500 align:middle line:90% 2, 3. 00:06:39.500 --> 00:06:43.010 align:middle line:84% And so the idea is that we can estimate these cut-off-- 00:06:43.010 --> 00:06:44.370 align:middle line:90% we can estimate these curves. 00:06:44.370 --> 00:06:47.640 align:middle line:84% Obviously, in practice, you're going to see a lot of zeros. 00:06:47.640 --> 00:06:51.230 align:middle line:84% You may see some doctors even in very small towns. 00:06:51.230 --> 00:06:53.660 align:middle line:84% You'll still see some towns without doctors here, 00:06:53.660 --> 00:06:58.190 align:middle line:84% but you'll see a lot of 1s, and maybe you'll even see some 2s. 00:06:58.190 --> 00:07:01.010 align:middle line:84% But there's going to be, in general, some curve like this 00:07:01.010 --> 00:07:03.860 align:middle line:84% where this is-- typically, you need 00:07:03.860 --> 00:07:09.410 align:middle line:84% to have S1 customers to have one doctor. 00:07:09.410 --> 00:07:11.730 align:middle line:90% You need S2 to have two doctors. 00:07:11.730 --> 00:07:16.730 align:middle line:84% Now, here, you need S3 to have three doctors. 00:07:16.730 --> 00:07:18.590 align:middle line:84% And so what they want to do is estimate 00:07:18.590 --> 00:07:20.030 align:middle line:90% what does this look like? 00:07:20.030 --> 00:07:23.060 align:middle line:84% How many people do you need to have before the first doctor 00:07:23.060 --> 00:07:23.992 align:middle line:90% comes in? 00:07:23.992 --> 00:07:26.450 align:middle line:84% How many do you need to have before the second doctor comes 00:07:26.450 --> 00:07:26.810 align:middle line:90% in? 00:07:26.810 --> 00:07:29.227 align:middle line:84% How many do you need to have before the third doctor comes 00:07:29.227 --> 00:07:31.500 align:middle line:90% in? 00:07:31.500 --> 00:07:34.597 align:middle line:84% Possibly you can estimate this just by plotting the cities 00:07:34.597 --> 00:07:36.930 align:middle line:84% with the population on the x-axis, the number of doctors 00:07:36.930 --> 00:07:38.460 align:middle line:90% on the y-axis. 00:07:38.460 --> 00:07:40.680 align:middle line:84% Fit a curve that looks like this through them-- 00:07:40.680 --> 00:07:42.200 align:middle line:90% see where the jump points are. 00:07:42.200 --> 00:07:45.210 align:middle line:90% 00:07:45.210 --> 00:07:49.470 align:middle line:84% So if you assume that the fixed costs don't vary with N, 00:07:49.470 --> 00:07:55.680 align:middle line:84% then SN over N is going to be pi N times pi N is going to be F. 00:07:55.680 --> 00:08:00.630 align:middle line:84% And also, SN plus 1 over N plus 1, pi N plus 1 is going to be F. 00:08:00.630 --> 00:08:03.420 align:middle line:90% So both at this point-- 00:08:03.420 --> 00:08:05.273 align:middle line:84% at this point you know the second doctor 00:08:05.273 --> 00:08:07.440 align:middle line:84% is just indifferent to entering or not because if it 00:08:07.440 --> 00:08:08.732 align:middle line:90% was smaller, he wouldn't enter. 00:08:08.732 --> 00:08:10.530 align:middle line:84% If it was bigger, he would enter. 00:08:10.530 --> 00:08:15.670 align:middle line:84% Here, where the third doctor is in different entering or not. 00:08:15.670 --> 00:08:19.800 align:middle line:84% So both of these, the profit that you get if you do come in, 00:08:19.800 --> 00:08:22.650 align:middle line:90% is exactly F. 00:08:22.650 --> 00:08:26.490 align:middle line:84% And so if you just set this equal to this, 00:08:26.490 --> 00:08:28.440 align:middle line:90% and then you divide the-- 00:08:28.440 --> 00:08:30.280 align:middle line:84% bring the pi N down here in the denominator, 00:08:30.280 --> 00:08:33.820 align:middle line:84% you get that the profit-- with N plus 1 firms divided 00:08:33.820 --> 00:08:36.190 align:middle line:84% by the profit of N firms is N plus 1 00:08:36.190 --> 00:08:39.880 align:middle line:90% over N, SN over SN plus 1. 00:08:39.880 --> 00:08:47.050 align:middle line:84% You could also write that as pi N plus 1 over pi N 00:08:47.050 --> 00:08:54.840 align:middle line:84% equals SN over N divided by SN plus 1 over N plus 1. 00:08:54.840 --> 00:08:59.250 align:middle line:90% 00:08:59.250 --> 00:09:04.740 align:middle line:84% So the idea here is that imagine it takes 1,000 people 00:09:04.740 --> 00:09:14.910 align:middle line:84% in a market to have one doctor, and S2 is 4,000. 00:09:14.910 --> 00:09:16.050 align:middle line:90% Then this is-- 00:09:16.050 --> 00:09:18.310 align:middle line:90% S1 over 1 is 1,000. 00:09:18.310 --> 00:09:21.300 align:middle line:84% You need 1,000 patients in order to be worthwhile to open 00:09:21.300 --> 00:09:22.740 align:middle line:90% a doctor's office. 00:09:22.740 --> 00:09:26.430 align:middle line:84% But then here, S2 over 2 is 2,000. 00:09:26.430 --> 00:09:28.893 align:middle line:84% So that would say you need 2,000 people to make it work-- 00:09:28.893 --> 00:09:31.560 align:middle line:84% you need 2,000 patients for the new doctor to make it worthwhile 00:09:31.560 --> 00:09:33.600 align:middle line:84% to have a second doctor's office. 00:09:33.600 --> 00:09:38.400 align:middle line:84% And from that, we infer that it must be that pi 2 is just 1/2 00:09:38.400 --> 00:09:40.350 align:middle line:90% of pi 1. 00:09:40.350 --> 00:09:44.760 align:middle line:84% So this ratio of this to this is how much profits per person 00:09:44.760 --> 00:09:47.923 align:middle line:84% must have declined for the second doctor to not come in 00:09:47.923 --> 00:09:50.340 align:middle line:84% until the second doctor is going to get many more patients 00:09:50.340 --> 00:09:51.640 align:middle line:90% than the first doctor got. 00:09:51.640 --> 00:09:55.190 align:middle line:90% 00:09:55.190 --> 00:09:57.830 align:middle line:84% And so anyway-- so that's the really powerful idea 00:09:57.830 --> 00:09:58.980 align:middle line:90% of this paper. 00:09:58.980 --> 00:10:02.690 align:middle line:84% We don't have to have any data at all on prices or quantities. 00:10:02.690 --> 00:10:07.580 align:middle line:84% And yet, we can tell how the ratio of profits 00:10:07.580 --> 00:10:10.640 align:middle line:84% declines as more firms enter the market by just looking 00:10:10.640 --> 00:10:13.310 align:middle line:84% at how many more customers have to be there for the second firm 00:10:13.310 --> 00:10:14.430 align:middle line:90% to want to come in. 00:10:14.430 --> 00:10:17.130 align:middle line:90% 00:10:17.130 --> 00:10:20.460 align:middle line:90% Question about that basic idea? 00:10:20.460 --> 00:10:21.720 align:middle line:90% OK. 00:10:21.720 --> 00:10:24.750 align:middle line:84% The previous slide suggested what Bresnahan and Reiss 00:10:24.750 --> 00:10:29.700 align:middle line:84% were going to do is put population on the x-axis, number 00:10:29.700 --> 00:10:31.320 align:middle line:84% of doctors on the y-axis, and then 00:10:31.320 --> 00:10:35.160 align:middle line:84% just draw a curve like this that best fit the data. 00:10:35.160 --> 00:10:38.700 align:middle line:84% That's not exactly what they do because they 00:10:38.700 --> 00:10:41.730 align:middle line:84% want to have a lot of covariates in the regression that are also 00:10:41.730 --> 00:10:43.270 align:middle line:90% affecting all of the elements. 00:10:43.270 --> 00:10:46.200 align:middle line:84% So basically, they set up the models. 00:10:46.200 --> 00:10:50.370 align:middle line:84% First, they have something they call the effect of population. 00:10:50.370 --> 00:10:56.550 align:middle line:84% The effect of population in city I is its actual population 00:10:56.550 --> 00:10:59.330 align:middle line:90% plus some X-- 00:10:59.330 --> 00:11:03.810 align:middle line:84% some explanatory variables, X, times some coefficients, gamma. 00:11:03.810 --> 00:11:06.570 align:middle line:84% Those X's, they use for things like the rural population 00:11:06.570 --> 00:11:09.030 align:middle line:84% within 5 miles of the city, the rural population 00:11:09.030 --> 00:11:12.875 align:middle line:84% within 10 miles of the city, the growth rate of the population. 00:11:12.875 --> 00:11:14.250 align:middle line:84% I forget what else goes in there. 00:11:14.250 --> 00:11:16.860 align:middle line:84% But they have the actual population of the city 00:11:16.860 --> 00:11:21.000 align:middle line:84% plus some controls for nearby population. 00:11:21.000 --> 00:11:24.240 align:middle line:84% Per consumer profits-- again, the idea 00:11:24.240 --> 00:11:26.670 align:middle line:84% is that you would have per consumer profits, alpha 00:11:26.670 --> 00:11:28.020 align:middle line:90% 1 under a monopoly. 00:11:28.020 --> 00:11:33.340 align:middle line:84% Profits alpha 1 under the monopoly. 00:11:33.340 --> 00:11:37.620 align:middle line:84% Alpha 2 would be duopoly profits. 00:11:37.620 --> 00:11:40.405 align:middle line:90% Alpha 3 would be 3 firms. 00:11:40.405 --> 00:11:42.780 align:middle line:84% But then you're going to have other variables that affect 00:11:42.780 --> 00:11:44.320 align:middle line:90% the profitability of a market. 00:11:44.320 --> 00:11:48.960 align:middle line:84% Those variables can include the per capita income in the city, 00:11:48.960 --> 00:11:50.040 align:middle line:90% the age distribution. 00:11:50.040 --> 00:11:53.100 align:middle line:84% Maybe higher income cities you earn more profits per customer 00:11:53.100 --> 00:11:56.340 align:middle line:90% or things like that. 00:11:56.340 --> 00:11:59.623 align:middle line:84% The fixed costs-- again, it's mostly a fixed cost. 00:11:59.623 --> 00:12:02.040 align:middle line:84% They do allow there to be covariates that affect the fixed 00:12:02.040 --> 00:12:03.310 align:middle line:90% costs across cities. 00:12:03.310 --> 00:12:04.920 align:middle line:84% So, for instance, cities that have 00:12:04.920 --> 00:12:06.870 align:middle line:84% more valuable agricultural land, it 00:12:06.870 --> 00:12:11.220 align:middle line:84% may be more expensive to open up a tire dealership than in places 00:12:11.220 --> 00:12:12.660 align:middle line:90% with cheaper agricultural land. 00:12:12.660 --> 00:12:15.930 align:middle line:84% Because the land that you need to buy to set up your tire 00:12:15.930 --> 00:12:16.920 align:middle line:90% shop costs more. 00:12:16.920 --> 00:12:22.210 align:middle line:84% So they have a fixed cost to also vary with some covariates. 00:12:22.210 --> 00:12:24.000 align:middle line:84% And then the basic thought is that-- so I 00:12:24.000 --> 00:12:25.000 align:middle line:90% have all these elements. 00:12:25.000 --> 00:12:30.700 align:middle line:84% I've redefined Si, pi iN, and Fi to be their covariates. 00:12:30.700 --> 00:12:35.340 align:middle line:84% And then my model is if there were N firms in the market, 00:12:35.340 --> 00:12:41.730 align:middle line:84% the profits would be Si over N, pi iN minus Fi plus epsilon i, 00:12:41.730 --> 00:12:44.520 align:middle line:84% where epsilon i is just a normally distributed 00:12:44.520 --> 00:12:47.040 align:middle line:90% random variable. 00:12:47.040 --> 00:12:50.730 align:middle line:84% And so in this model, when are we going to observe in firms? 00:12:50.730 --> 00:12:56.640 align:middle line:84% We're going to observe in firms if Si N-- 00:12:56.640 --> 00:13:06.060 align:middle line:84% sorry-- Si over N minus Fi plus epsilon i is greater than 0. 00:13:06.060 --> 00:13:15.000 align:middle line:84% And Si over n plus 1, pi iN plus 1 minus Fi plus epsilon i equals 00:13:15.000 --> 00:13:17.940 align:middle line:90% 0. 00:13:17.940 --> 00:13:20.100 align:middle line:84% So we have these two things that have 00:13:20.100 --> 00:13:24.990 align:middle line:84% to both be true for there to be exactly N firms in the market. 00:13:24.990 --> 00:13:26.700 align:middle line:84% But you could write the first one 00:13:26.700 --> 00:13:28.680 align:middle line:84% as epsilon is bigger than something, 00:13:28.680 --> 00:13:31.860 align:middle line:84% and the second one is epsilon is less than something. 00:13:31.860 --> 00:13:35.280 align:middle line:84% So if you view epsilon as a random variable here, 00:13:35.280 --> 00:13:38.580 align:middle line:84% a random shock to fixed costs, we're 00:13:38.580 --> 00:13:40.590 align:middle line:84% going to observe N firms whenever 00:13:40.590 --> 00:13:42.150 align:middle line:90% epsilon is in this interval. 00:13:42.150 --> 00:13:44.310 align:middle line:90% It's between this and this. 00:13:44.310 --> 00:13:49.950 align:middle line:84% And so given the param-- given all the data and the parameters, 00:13:49.950 --> 00:13:52.110 align:middle line:84% you can compute the likelihood of this 00:13:52.110 --> 00:13:55.350 align:middle line:84% as just the normal CDF evaluated at this point 00:13:55.350 --> 00:13:58.360 align:middle line:84% minus the normal CDF evaluated at this point. 00:13:58.360 --> 00:14:01.290 align:middle line:84% And so then you can estimate all the parameters-- 00:14:01.290 --> 00:14:07.270 align:middle line:84% gamma, beta, alpha, delta, gamma-- 00:14:07.270 --> 00:14:09.450 align:middle line:90% this one was lambda. 00:14:09.450 --> 00:14:13.560 align:middle line:84% You can estimate all of those by maximum likelihood. 00:14:13.560 --> 00:14:17.700 align:middle line:84% And that's the basic estimation technique in the paper. 00:14:17.700 --> 00:14:21.150 align:middle line:84% People-- I know we still teach ordered probit models 00:14:21.150 --> 00:14:21.912 align:middle line:90% in econometrics. 00:14:21.912 --> 00:14:24.370 align:middle line:84% This is what people refer to as ordered probit model, where 00:14:24.370 --> 00:14:29.980 align:middle line:84% you have discrete data produced by the discretization 00:14:29.980 --> 00:14:31.910 align:middle line:84% of something that has a normal error in it. 00:14:31.910 --> 00:14:35.580 align:middle line:90% 00:14:35.580 --> 00:14:39.150 align:middle line:84% And obviously, the main things of interest here coming out 00:14:39.150 --> 00:14:42.420 align:middle line:90% of this estimation are alpha 1-- 00:14:42.420 --> 00:14:43.990 align:middle line:84% these were the objects of interest. 00:14:43.990 --> 00:14:46.866 align:middle line:84% What is the per customer profits under monopoly? 00:14:46.866 --> 00:14:51.990 align:middle line:90% 00:14:51.990 --> 00:14:53.400 align:middle line:84% What are the per customer profits 00:14:53.400 --> 00:14:55.470 align:middle line:84% under monopoly, duopoly, whatever? 00:14:55.470 --> 00:14:57.780 align:middle line:84% But somehow in the paper, they really 00:14:57.780 --> 00:15:00.340 align:middle line:84% liked the idea of the S's as the thresholds. 00:15:00.340 --> 00:15:02.970 align:middle line:84% And so they basically estimate these things. 00:15:02.970 --> 00:15:04.650 align:middle line:84% And then they convert from all the-- 00:15:04.650 --> 00:15:06.210 align:middle line:84% they convert from all the parameters 00:15:06.210 --> 00:15:10.260 align:middle line:84% back to what is the mean S1, mean S2, mean S3, 00:15:10.260 --> 00:15:12.870 align:middle line:84% holding all the covariates at their sample means. 00:15:12.870 --> 00:15:16.680 align:middle line:84% And then they do a lot of the presentation just talking about 00:15:16.680 --> 00:15:17.585 align:middle line:90% what are the S's? 00:15:17.585 --> 00:15:21.320 align:middle line:90% 00:15:21.320 --> 00:15:21.820 align:middle line:90% OK. 00:15:21.820 --> 00:15:24.340 align:middle line:84% Anyway, so data samples-- as I said, 00:15:24.340 --> 00:15:26.050 align:middle line:84% they've got data on a whole bunch 00:15:26.050 --> 00:15:31.370 align:middle line:84% of businesses gathered from 202 small towns in Western US. 00:15:31.370 --> 00:15:32.630 align:middle line:90% These are small places. 00:15:32.630 --> 00:15:36.310 align:middle line:90% The mean population is 3,740. 00:15:36.310 --> 00:15:39.190 align:middle line:84% And, as I said, they're all at least 20 miles away from any 00:15:39.190 --> 00:15:44.214 align:middle line:84% place that has 1,000 people in it. 00:15:44.214 --> 00:15:47.300 align:middle line:84% I went to college with a guy from Wyoming 00:15:47.300 --> 00:15:51.372 align:middle line:84% who said he was 90 miles from the nearest movie theater. 00:15:51.372 --> 00:15:53.830 align:middle line:84% So there are parts of the United States that are like that. 00:15:53.830 --> 00:15:57.290 align:middle line:90% 00:15:57.290 --> 00:16:04.430 align:middle line:84% Here is the data sample showing the full set of occupation-- 00:16:04.430 --> 00:16:06.350 align:middle line:84% or businesses they started with and the number 00:16:06.350 --> 00:16:08.240 align:middle line:90% of firms they get. 00:16:08.240 --> 00:16:11.580 align:middle line:84% They end up focusing on the ones on the top part of the table. 00:16:11.580 --> 00:16:12.650 align:middle line:90% There are some paper-- 00:16:12.650 --> 00:16:14.270 align:middle line:84% some results in papers about things 00:16:14.270 --> 00:16:15.720 align:middle line:90% farther down in the paper. 00:16:15.720 --> 00:16:17.420 align:middle line:90% But what they wanted-- 00:16:17.420 --> 00:16:19.130 align:middle line:84% in some sense, if you want to talk 00:16:19.130 --> 00:16:20.780 align:middle line:84% about how the competition changes 00:16:20.780 --> 00:16:22.780 align:middle line:84% with the number of firms, you want 00:16:22.780 --> 00:16:24.530 align:middle line:84% there to be a significant number of cities 00:16:24.530 --> 00:16:29.190 align:middle line:84% that have zero firms, 1, 2, 3, 4, or whatever. 00:16:29.190 --> 00:16:33.500 align:middle line:84% And a lot of these businesses, it turns out, you never get-- 00:16:33.500 --> 00:16:36.380 align:middle line:84% a lot of these businesses are apparently 00:16:36.380 --> 00:16:39.440 align:middle line:84% too big businesses to have many of them in the towns 00:16:39.440 --> 00:16:40.940 align:middle line:90% that they're working with. 00:16:40.940 --> 00:16:42.730 align:middle line:84% So they don't get much beyond N equals 2. 00:16:42.730 --> 00:16:44.640 align:middle line:84% But if you look at the druggist-- 00:16:44.640 --> 00:16:47.790 align:middle line:90% I assume this means pharmacies-- 00:16:47.790 --> 00:16:51.100 align:middle line:84% they're very relatively rare to have no pharmacy. 00:16:51.100 --> 00:16:55.980 align:middle line:84% But then you do get 1, 2, 3, and some 4s and 5s and beyond. 00:16:55.980 --> 00:16:58.860 align:middle line:84% Same things with the doctors, dentists, plumbers, tire 00:16:58.860 --> 00:17:00.135 align:middle line:90% dealers is-- 00:17:00.135 --> 00:17:01.740 align:middle line:84% in some sense, they focus on these 00:17:01.740 --> 00:17:04.260 align:middle line:84% because they all have at least some number of markets with 3 00:17:04.260 --> 00:17:08.180 align:middle line:84% and some number of markets with 4. 00:17:08.180 --> 00:17:11.839 align:middle line:84% There was also a question of how well-defined are these markets? 00:17:11.839 --> 00:17:17.450 align:middle line:84% Like, the sum of these things, like the barbers or whatever-- 00:17:17.450 --> 00:17:19.635 align:middle line:84% people who are not barbers can cut people's hair. 00:17:19.635 --> 00:17:21.260 align:middle line:84% And so they're sort of-- it gets harder 00:17:21.260 --> 00:17:24.619 align:middle line:84% to count how many barbers or beauticians there are in a town 00:17:24.619 --> 00:17:26.990 align:middle line:84% because it's just somebody who works somewhere else-- 00:17:26.990 --> 00:17:27.795 align:middle line:90% also cuts hair. 00:17:27.795 --> 00:17:29.920 align:middle line:84% People know they can go to that person or whatever. 00:17:29.920 --> 00:17:31.500 align:middle line:90% So it's hard to-- 00:17:31.500 --> 00:17:34.700 align:middle line:84% whereas, things like dentists, I guess, 00:17:34.700 --> 00:17:37.480 align:middle line:84% are easier to identify that someone is a dentist. 00:17:37.480 --> 00:17:39.320 align:middle line:84% You don't get just guys who come over 00:17:39.320 --> 00:17:41.650 align:middle line:84% and drill holes in your teeth on weekends. 00:17:41.650 --> 00:17:45.380 align:middle line:90% 00:17:45.380 --> 00:17:47.910 align:middle line:84% So what do the entry thresholds look like? 00:17:47.910 --> 00:17:53.690 align:middle line:84% So, again, the numbers I put up on the board over there are-- 00:17:53.690 --> 00:17:57.050 align:middle line:84% most of these businesses tend to start to exist in towns, 00:17:57.050 --> 00:17:59.510 align:middle line:84% and they have fewer than 1,000 people. 00:17:59.510 --> 00:18:02.520 align:middle line:84% So the doctors-- the cutoff is 880 people. 00:18:02.520 --> 00:18:06.020 align:middle line:84% The dentist-- the cut-off is 710 people. 00:18:06.020 --> 00:18:08.680 align:middle line:90% The druggist-- it's 530 people. 00:18:08.680 --> 00:18:10.853 align:middle line:84% The tire dealers, it's 490 people. 00:18:10.853 --> 00:18:12.520 align:middle line:84% You need to be bigger to have a plumber. 00:18:12.520 --> 00:18:17.850 align:middle line:84% That's over 1,000-- 1,400 people. 00:18:17.850 --> 00:18:21.450 align:middle line:84% But then if you look at-- let's just look at-- the druggists 00:18:21.450 --> 00:18:23.350 align:middle line:84% is one of the most stark examples. 00:18:23.350 --> 00:18:26.430 align:middle line:84% It seems like you need 500 people to have one. 00:18:26.430 --> 00:18:28.930 align:middle line:84% You need 2,000 people to have two. 00:18:28.930 --> 00:18:32.610 align:middle line:84% So that's when you have 1,000 customers per drugstore. 00:18:32.610 --> 00:18:36.180 align:middle line:90% To have three, you need 5,000. 00:18:36.180 --> 00:18:40.080 align:middle line:84% So that's like 1,600 people per druggist. 00:18:40.080 --> 00:18:45.750 align:middle line:84% And when they put the ratios over here and make it clear, 00:18:45.750 --> 00:18:50.280 align:middle line:84% the S2 over S1 ratio-- this is how many customers per firm you 00:18:50.280 --> 00:18:51.640 align:middle line:90% need to have to have-- 00:18:51.640 --> 00:19:00.128 align:middle line:84% so this is-- they use little sn to be capital SN over N, which 00:19:00.128 --> 00:19:02.670 align:middle line:84% is always tough for when you're putting things on slides that 00:19:02.670 --> 00:19:03.280 align:middle line:90% are small. 00:19:03.280 --> 00:19:07.730 align:middle line:84% So anyway, these are the ratio of the per customer-- 00:19:07.730 --> 00:19:09.655 align:middle line:84% the per firm number of customers, 00:19:09.655 --> 00:19:11.030 align:middle line:84% which you can think of as this is 00:19:11.030 --> 00:19:14.370 align:middle line:84% the thing that's supposed to estimate pi N over pi N plus 1. 00:19:14.370 --> 00:19:20.090 align:middle line:84% So this is supposed to be pi 1 over pi 2. 00:19:20.090 --> 00:19:22.490 align:middle line:84% And so what this is saying is that 00:19:22.490 --> 00:19:24.830 align:middle line:84% in many of these industries, profits 00:19:24.830 --> 00:19:28.400 align:middle line:84% are dropping almost in half when you have-- you have this when 00:19:28.400 --> 00:19:31.100 align:middle line:84% you have a second firm, or per customer profits dropping almost 00:19:31.100 --> 00:19:31.600 align:middle line:90% in half. 00:19:31.600 --> 00:19:34.700 align:middle line:90% It's like profits-- divide-- 00:19:34.700 --> 00:19:37.080 align:middle line:84% you divide it by either 1.9 or by 1.8. 00:19:37.080 --> 00:19:39.290 align:middle line:84% So that's like a 45%, 50% drop in profit 00:19:39.290 --> 00:19:41.360 align:middle line:90% when the second firm comes in. 00:19:41.360 --> 00:19:43.910 align:middle line:84% The one exception appears to be plumbers, 00:19:43.910 --> 00:19:46.410 align:middle line:84% where we don't get the first plumber for a while. 00:19:46.410 --> 00:19:52.100 align:middle line:84% And so it seems that the second plumber is also serving 50. 00:19:52.100 --> 00:19:54.080 align:middle line:84% You get one plumber when you have 1,400 people 00:19:54.080 --> 00:19:56.413 align:middle line:84% for the plumber to serve, a second plumber when you have 00:19:56.413 --> 00:19:59.840 align:middle line:90% 1,500. 00:19:59.840 --> 00:20:03.500 align:middle line:84% If you look at what happens when you go-- so one firm to two-- 00:20:03.500 --> 00:20:05.160 align:middle line:90% it's a big drop in profit. 00:20:05.160 --> 00:20:09.870 align:middle line:84% When you go from two firms to three, it varies by industry. 00:20:09.870 --> 00:20:13.350 align:middle line:84% But most of them, other than the plumbers, 00:20:13.350 --> 00:20:18.910 align:middle line:84% we have a 1 point-- a 10% drop, 30% drop, 60% drop or whatever. 00:20:18.910 --> 00:20:21.910 align:middle line:84% So it, again, is a-- or 1 over 1.6. 00:20:21.910 --> 00:20:28.170 align:middle line:84% So, again, you get substantial drops often going from 2 to 3. 00:20:28.170 --> 00:20:32.280 align:middle line:84% By the time you're out here, though, the S5 over S4 numbers, 00:20:32.280 --> 00:20:34.303 align:middle line:84% here, it seems like competition-- 00:20:34.303 --> 00:20:35.970 align:middle line:84% whatever competition you're going to get 00:20:35.970 --> 00:20:37.980 align:middle line:84% is all done by the time you have four firms. 00:20:37.980 --> 00:20:40.680 align:middle line:84% Like four firms are getting you just as much competition 00:20:40.680 --> 00:20:43.740 align:middle line:90% as five firms is giving you. 00:20:43.740 --> 00:20:49.260 align:middle line:84% So the conclusion here of how basics of competition 00:20:49.260 --> 00:20:51.870 align:middle line:84% seems to work is there's a big difference between monopoly 00:20:51.870 --> 00:20:53.220 align:middle line:90% and two firms. 00:20:53.220 --> 00:20:56.850 align:middle line:84% It seems like prices keep going down from 2 to 3. 00:20:56.850 --> 00:20:59.760 align:middle line:84% And then it seems like we asymptote pretty quickly 00:20:59.760 --> 00:21:03.760 align:middle line:84% to things are not getting any more competitive. 00:21:03.760 --> 00:21:07.530 align:middle line:90% 00:21:07.530 --> 00:21:08.590 align:middle line:90% Any questions? 00:21:08.590 --> 00:21:13.545 align:middle line:90% 00:21:13.545 --> 00:21:20.420 align:middle line:84% I mean, I think it's a fantastic paper in that this idea of entry 00:21:20.420 --> 00:21:22.070 align:middle line:84% and telling us about profit levels 00:21:22.070 --> 00:21:24.230 align:middle line:84% and that being an optimality condition we can use 00:21:24.230 --> 00:21:27.020 align:middle line:84% has just been tremendously influential. 00:21:27.020 --> 00:21:28.620 align:middle line:84% As far as the applied results go, 00:21:28.620 --> 00:21:32.240 align:middle line:84% I would say the biggest difficulty with this 00:21:32.240 --> 00:21:34.430 align:middle line:84% about-- with regard to this as an applied result 00:21:34.430 --> 00:21:37.100 align:middle line:84% is this model is interpreting what 00:21:37.100 --> 00:21:41.840 align:middle line:84% we're seeing as the equilibrium N star of this game. 00:21:41.840 --> 00:21:44.090 align:middle line:84% So the equilibrium N star of that game 00:21:44.090 --> 00:21:46.372 align:middle line:84% implies in part that everybody who is there 00:21:46.372 --> 00:21:47.330 align:middle line:90% is happy they're there. 00:21:47.330 --> 00:21:49.802 align:middle line:84% And everyone who's not there is happy they're not there. 00:21:49.802 --> 00:21:51.260 align:middle line:84% I think the happy they're not there 00:21:51.260 --> 00:21:54.130 align:middle line:90% part I think is very clear. 00:21:54.130 --> 00:21:57.770 align:middle line:84% The question is, what do we make of all these doctors in towns 00:21:57.770 --> 00:21:59.030 align:middle line:90% of 800 people? 00:21:59.030 --> 00:22:01.340 align:middle line:84% Is the doctor in a town of 800 people 00:22:01.340 --> 00:22:04.070 align:middle line:84% because the doctor is happy serving 800 patients? 00:22:04.070 --> 00:22:06.740 align:middle line:84% Or is the doctor in a town of 800 people because it was a town 00:22:06.740 --> 00:22:11.550 align:middle line:84% of 1,500 people and a thriving farming town in the 1950s when 00:22:11.550 --> 00:22:12.550 align:middle line:90% the doctor moved there. 00:22:12.550 --> 00:22:14.430 align:middle line:84% And the doctor is now 75 years old 00:22:14.430 --> 00:22:17.160 align:middle line:84% and doesn't want to leave the patients the doctor has 00:22:17.160 --> 00:22:20.160 align:middle line:84% had for 40 years, even though the number of patients 00:22:20.160 --> 00:22:24.750 align:middle line:84% is dwindling relative to what it was when the doctor got there. 00:22:24.750 --> 00:22:28.470 align:middle line:84% And if you think about it, if there's this sort of-- you pay 00:22:28.470 --> 00:22:32.250 align:middle line:84% this sunk cost to enter, and you don't get your sunk cost back 00:22:32.250 --> 00:22:36.133 align:middle line:84% when you leave, it could be that people enter-- 00:22:36.133 --> 00:22:37.800 align:middle line:84% the entry-- the things that we're seeing 00:22:37.800 --> 00:22:39.130 align:middle line:90% are not entry thresholds. 00:22:39.130 --> 00:22:43.350 align:middle line:84% They're exit-- we're seeing exit thresholds. 00:22:43.350 --> 00:22:48.410 align:middle line:84% And the exit thresholds might not be counting all fixed costs. 00:22:48.410 --> 00:22:51.270 align:middle line:90% 00:22:51.270 --> 00:22:55.230 align:middle line:84% Another potential issue is just that counting firms 00:22:55.230 --> 00:22:59.480 align:middle line:84% is more difficult in their application than in others. 00:22:59.480 --> 00:23:01.980 align:middle line:84% How do you count someone who does some plumbing on the side, 00:23:01.980 --> 00:23:03.563 align:middle line:84% or someone who does some hair cutting? 00:23:03.563 --> 00:23:05.790 align:middle line:84% So I think the ones in the licensed professions, 00:23:05.790 --> 00:23:09.460 align:middle line:84% like the dentists and doctors, are probably better with that. 00:23:09.460 --> 00:23:12.120 align:middle line:84% But the ones that are more like the businesses we normally 00:23:12.120 --> 00:23:16.200 align:middle line:84% think about price competition operating in, some of those 00:23:16.200 --> 00:23:17.730 align:middle line:84% have this issue of it's going to be 00:23:17.730 --> 00:23:21.870 align:middle line:84% hard to identify businesses, and ensure that we're [INAUDIBLE]. 00:23:21.870 --> 00:23:23.700 align:middle line:84% For instance, the result on plumbers 00:23:23.700 --> 00:23:27.158 align:middle line:84% could be that you don't get a lot of market power 00:23:27.158 --> 00:23:29.700 align:middle line:84% with one plumber because there's always somebody else in town 00:23:29.700 --> 00:23:32.220 align:middle line:84% who you can call who just owns wrenches and can 00:23:32.220 --> 00:23:35.037 align:middle line:84% come over and fix stuff for you, even if they're not 00:23:35.037 --> 00:23:35.870 align:middle line:90% officially plumbers. 00:23:35.870 --> 00:23:39.310 align:middle line:90% 00:23:39.310 --> 00:23:40.282 align:middle line:90% Questions? 00:23:40.282 --> 00:23:43.000 align:middle line:90% 00:23:43.000 --> 00:23:43.940 align:middle line:90% OK. 00:23:43.940 --> 00:23:49.220 align:middle line:84% So then the next step I want to go to was Berry and Waldfogel. 00:23:49.220 --> 00:23:51.950 align:middle line:84% So this is very much following up 00:23:51.950 --> 00:23:54.825 align:middle line:84% on of theory of efficiency of entry I did last class. 00:23:54.825 --> 00:23:57.270 align:middle line:84% And so, as I said, there are reasons 00:23:57.270 --> 00:23:59.270 align:middle line:84% why entry is going to be higher or lower than we 00:23:59.270 --> 00:24:00.210 align:middle line:90% would like it to be. 00:24:00.210 --> 00:24:01.627 align:middle line:84% And Berry and Waldfogel are trying 00:24:01.627 --> 00:24:03.530 align:middle line:90% to say, let's pick an industry. 00:24:03.530 --> 00:24:05.940 align:middle line:84% Let's say in this industry, how does it work out? 00:24:05.940 --> 00:24:08.700 align:middle line:84% Are there more firms than we would like or fewer firms? 00:24:08.700 --> 00:24:11.270 align:middle line:84% How big are the welfare losses from the number of firms 00:24:11.270 --> 00:24:12.620 align:middle line:90% that we observe? 00:24:12.620 --> 00:24:18.340 align:middle line:84% The industry that they pick is commercial radio. 00:24:18.340 --> 00:24:21.110 align:middle line:84% And I think the argument is that there's-- 00:24:21.110 --> 00:24:25.350 align:middle line:84% you need a license to open a radio station. 00:24:25.350 --> 00:24:28.400 align:middle line:84% There are some places where you might-- where there's a-- 00:24:28.400 --> 00:24:30.733 align:middle line:84% you can't open a radio station and broadcast on any band 00:24:30.733 --> 00:24:32.650 align:middle line:84% because you'd be interfering with the stations 00:24:32.650 --> 00:24:33.390 align:middle line:90% right around you. 00:24:33.390 --> 00:24:35.570 align:middle line:84% But in most cities in the United States, 00:24:35.570 --> 00:24:37.580 align:middle line:84% there are bands that are available. 00:24:37.580 --> 00:24:41.010 align:middle line:84% There's obviously a fixed cost of applying for the government 00:24:41.010 --> 00:24:43.840 align:middle line:84% license and opening up and just putting up a transmission tower. 00:24:43.840 --> 00:24:44.610 align:middle line:90% But that's fine. 00:24:44.610 --> 00:24:46.953 align:middle line:90% That's exactly all the Fs. 00:24:46.953 --> 00:24:48.870 align:middle line:84% And so this is an example of an industry where 00:24:48.870 --> 00:24:51.363 align:middle line:84% we think there is a substantial F. Because to have a radio 00:24:51.363 --> 00:24:52.530 align:middle line:90% station, you need equipment. 00:24:52.530 --> 00:24:54.210 align:middle line:90% You need people there. 00:24:54.210 --> 00:24:58.140 align:middle line:84% But then we can think about how are the-- 00:24:58.140 --> 00:25:00.810 align:middle line:90% how is the prices affected. 00:25:00.810 --> 00:25:03.473 align:middle line:90% Now, radio is free. 00:25:03.473 --> 00:25:04.890 align:middle line:84% So in some sense, the price you're 00:25:04.890 --> 00:25:07.590 align:middle line:84% used to thinking about for radio is zero. 00:25:07.590 --> 00:25:10.900 align:middle line:90% But radio is a platform market. 00:25:10.900 --> 00:25:13.480 align:middle line:84% So there are two types of customers at radio stations. 00:25:13.480 --> 00:25:15.930 align:middle line:84% There are listeners who typically pay zero to listen. 00:25:15.930 --> 00:25:18.150 align:middle line:84% But then there are advertisers who pay money 00:25:18.150 --> 00:25:20.070 align:middle line:90% to put commercials on. 00:25:20.070 --> 00:25:23.610 align:middle line:90% And so the standard way that-- 00:25:23.610 --> 00:25:25.560 align:middle line:84% because the price is free, the standard way 00:25:25.560 --> 00:25:29.580 align:middle line:84% that the FTC or antitrust regulators tend to treat radio 00:25:29.580 --> 00:25:33.040 align:middle line:84% is to focus on the business of producing advertising. 00:25:33.040 --> 00:25:35.250 align:middle line:84% So they think of as a radio station is-- 00:25:35.250 --> 00:25:39.220 align:middle line:84% they have this technology where they buy rights to music. 00:25:39.220 --> 00:25:40.200 align:middle line:90% They play that music. 00:25:40.200 --> 00:25:42.840 align:middle line:84% That music produces ears of people. 00:25:42.840 --> 00:25:45.180 align:middle line:84% And then they sell those ears of people 00:25:45.180 --> 00:25:47.610 align:middle line:84% to advertisers who want to advertise to them. 00:25:47.610 --> 00:25:51.030 align:middle line:84% So when we're thinking about price in this market, 00:25:51.030 --> 00:25:54.150 align:middle line:84% we should be thinking about price of advertisers-- 00:25:54.150 --> 00:25:59.820 align:middle line:84% price of-- think of the price as the price of a listener, 00:25:59.820 --> 00:26:02.850 align:middle line:84% and think of the firms as having a technology for producing 00:26:02.850 --> 00:26:03.620 align:middle line:90% listeners. 00:26:03.620 --> 00:26:05.370 align:middle line:84% So this technology for producing listeners 00:26:05.370 --> 00:26:08.400 align:middle line:84% involves something about radio music quality 00:26:08.400 --> 00:26:09.930 align:middle line:90% or content quality. 00:26:09.930 --> 00:26:12.090 align:middle line:84% And then that produces people that they then 00:26:12.090 --> 00:26:13.900 align:middle line:90% sell to the advertisers. 00:26:13.900 --> 00:26:16.230 align:middle line:84% So the advertisers are the consumers. 00:26:16.230 --> 00:26:17.720 align:middle line:90% The people are ignored. 00:26:17.720 --> 00:26:21.480 align:middle line:84% And one thing you might notice, if that's your view of radio, 00:26:21.480 --> 00:26:22.500 align:middle line:90% you'd be like, wait. 00:26:22.500 --> 00:26:25.230 align:middle line:84% Wouldn't that mean like whenever two radio stations are proposing 00:26:25.230 --> 00:26:27.720 align:middle line:84% merging, what would be considered bad 00:26:27.720 --> 00:26:29.820 align:middle line:84% is if by merging the radio stations 00:26:29.820 --> 00:26:32.040 align:middle line:84% are going to raise the price of advertising 00:26:32.040 --> 00:26:36.450 align:middle line:84% and have fewer commercials per hour and play music instead. 00:26:36.450 --> 00:26:39.632 align:middle line:84% And that actually tends to be how 00:26:39.632 --> 00:26:41.340 align:middle line:84% the US antitrust authorities have largely 00:26:41.340 --> 00:26:42.490 align:middle line:90% conceptualized radio. 00:26:42.490 --> 00:26:43.948 align:middle line:84% That is, if there's something going 00:26:43.948 --> 00:26:46.980 align:middle line:84% on that's going to result in fewer commercials per hour 00:26:46.980 --> 00:26:49.833 align:middle line:84% and more music or more news, then they're skeptical of it. 00:26:49.833 --> 00:26:51.500 align:middle line:84% If it's something that's going to result 00:26:51.500 --> 00:26:53.040 align:middle line:84% in even more commercials per hour, 00:26:53.040 --> 00:26:56.310 align:middle line:84% that's good because that's stopping the deadweight loss 00:26:56.310 --> 00:26:58.830 align:middle line:90% on the advertiser's side. 00:26:58.830 --> 00:27:02.800 align:middle line:84% Obviously, Berry and Waldfogel mostly adopt that. 00:27:02.800 --> 00:27:05.153 align:middle line:84% They are clearly well aware that radio is also 00:27:05.153 --> 00:27:07.320 align:middle line:84% something people listen to, and they talk about that 00:27:07.320 --> 00:27:09.750 align:middle line:84% at the end when all the analysis is done, 00:27:09.750 --> 00:27:12.920 align:middle line:84% thinking about radio stations as things that produce advertising 00:27:12.920 --> 00:27:14.056 align:middle line:90% to sell to advertisers. 00:27:14.056 --> 00:27:18.880 align:middle line:90% 00:27:18.880 --> 00:27:21.318 align:middle line:90% So we've got 135 US markets. 00:27:21.318 --> 00:27:23.860 align:middle line:84% We're trying to compare with what would be socially efficient 00:27:23.860 --> 00:27:26.140 align:middle line:84% and talk about the welfare losses. 00:27:26.140 --> 00:27:29.500 align:middle line:84% We're going to focus on the radio stations 00:27:29.500 --> 00:27:34.010 align:middle line:84% as firms that have advertisers as customers. 00:27:34.010 --> 00:27:35.740 align:middle line:90% They're going to assume-- 00:27:35.740 --> 00:27:38.990 align:middle line:84% they're starting with the very first model of entry that I did. 00:27:38.990 --> 00:27:41.410 align:middle line:84% So they're assuming that the ears that are being produced 00:27:41.410 --> 00:27:43.150 align:middle line:90% are a homogeneous good. 00:27:43.150 --> 00:27:46.150 align:middle line:84% So every listener is worth exactly the same amount 00:27:46.150 --> 00:27:48.640 align:middle line:90% to every advertiser. 00:27:48.640 --> 00:27:55.360 align:middle line:84% And, in practice, obviously, there are going to be stations 00:27:55.360 --> 00:27:58.270 align:middle line:84% playing music from the 1950s, where they have lots of ads 00:27:58.270 --> 00:28:00.430 align:middle line:90% for reverse mortgages. 00:28:00.430 --> 00:28:04.930 align:middle line:84% There are going to be stations playing 00:28:04.930 --> 00:28:09.970 align:middle line:84% much more up-to-date music that are advertising concert tickets. 00:28:09.970 --> 00:28:12.970 align:middle line:84% Somehow they just they just abstract away from that. 00:28:12.970 --> 00:28:15.070 align:middle line:84% It's just-- they're just going to go to listeners 00:28:15.070 --> 00:28:16.220 align:middle line:90% are a homogeneous effect. 00:28:16.220 --> 00:28:18.470 align:middle line:84% There's just a downward-sloping demand 00:28:18.470 --> 00:28:22.820 align:middle line:84% curve for the more listeners you have, the price-- 00:28:22.820 --> 00:28:26.210 align:middle line:84% or the more listeners you-- the more commercial slots 00:28:26.210 --> 00:28:30.410 align:middle line:84% you have to sell, lowers the price per commercial slot. 00:28:30.410 --> 00:28:32.470 align:middle line:84% And that's how that market is going to work. 00:28:32.470 --> 00:28:35.660 align:middle line:84% Anyway, so the-- what's also good 00:28:35.660 --> 00:28:39.290 align:middle line:84% here is they have purely fixed costs and no marginal costs 00:28:39.290 --> 00:28:40.680 align:middle line:90% of producing listeners. 00:28:40.680 --> 00:28:42.470 align:middle line:84% So you get even simpler welfare function 00:28:42.470 --> 00:28:43.970 align:middle line:84% than the one we talked about before. 00:28:43.970 --> 00:28:47.750 align:middle line:84% So the welfare-- if you have-- and radio stations, 00:28:47.750 --> 00:28:49.500 align:middle line:84% it's going to be the integral from 0. 00:28:49.500 --> 00:28:51.708 align:middle line:84% So L star of N is going to be the number of listeners 00:28:51.708 --> 00:28:53.810 align:middle line:84% per station when you have N stations. 00:28:53.810 --> 00:28:58.590 align:middle line:84% So it's just the integral from 0 to N L star of N. P of x dx-- 00:28:58.590 --> 00:29:01.220 align:middle line:84% the willingness to pay for the very first-- for each listener 00:29:01.220 --> 00:29:03.303 align:middle line:84% integrated out over the total number of listeners. 00:29:03.303 --> 00:29:06.230 align:middle line:84% So this is the advertiser willingness 00:29:06.230 --> 00:29:08.060 align:middle line:90% to pay for listeners. 00:29:08.060 --> 00:29:10.670 align:middle line:84% And this is the gross surplus, is just 00:29:10.670 --> 00:29:15.500 align:middle line:84% the area under that curve, minus the fixed costs 00:29:15.500 --> 00:29:18.200 align:middle line:84% that are incurred in generating all those listeners. 00:29:18.200 --> 00:29:19.990 align:middle line:84% And so the things that we need to estimate 00:29:19.990 --> 00:29:23.410 align:middle line:84% are this per firm listening function L 00:29:23.410 --> 00:29:27.790 align:middle line:84% star of N. We need the advertiser inverse demand 00:29:27.790 --> 00:29:32.170 align:middle line:84% function, L of N. And we need the fixed cost F. So 00:29:32.170 --> 00:29:34.420 align:middle line:84% to evaluate welfare in this perspective, 00:29:34.420 --> 00:29:36.405 align:middle line:84% you just need to estimate those three objects. 00:29:36.405 --> 00:29:41.283 align:middle line:90% 00:29:41.283 --> 00:29:42.200 align:middle line:90% How are we doing this? 00:29:42.200 --> 00:29:44.500 align:middle line:84% So the first [INAUDIBLE] of listening, this 00:29:44.500 --> 00:29:49.570 align:middle line:84% is where you get the sense of how much NEIO people like 00:29:49.570 --> 00:29:52.150 align:middle line:84% discrete choice models and discrete-- discrete choice 00:29:52.150 --> 00:29:53.950 align:middle line:84% models of estimating things because they 00:29:53.950 --> 00:29:55.610 align:middle line:90% don't have any price variation. 00:29:55.610 --> 00:29:57.777 align:middle line:84% They're not actually trying to estimate the consumer 00:29:57.777 --> 00:29:58.450 align:middle line:90% preferences. 00:29:58.450 --> 00:30:00.783 align:middle line:84% Despite not trying to estimate the consumer preferences, 00:30:00.783 --> 00:30:04.300 align:middle line:84% they're still using a model with consumers having discrete choice 00:30:04.300 --> 00:30:07.360 align:middle line:84% preferences, choosing among stations or the outside good 00:30:07.360 --> 00:30:10.360 align:middle line:84% to generate their listening function. 00:30:10.360 --> 00:30:15.910 align:middle line:84% So this is-- people know what nested logit models are? 00:30:15.910 --> 00:30:17.470 align:middle line:90% So, OK. 00:30:17.470 --> 00:30:20.830 align:middle line:84% So anyway, this is a-- what we call a nested logit model. 00:30:20.830 --> 00:30:23.650 align:middle line:84% Do you remember I sort of-- we went with the standard logit 00:30:23.650 --> 00:30:25.420 align:middle line:84% model, the utility the consumer i 00:30:25.420 --> 00:30:29.320 align:middle line:84% gets from listening to station j is delta j, 00:30:29.320 --> 00:30:32.710 align:middle line:84% which is like the quality of station j plus epsilon ij, which 00:30:32.710 --> 00:30:34.225 align:middle line:90% is like the logit type. 00:30:34.225 --> 00:30:37.210 align:middle line:90% 00:30:37.210 --> 00:30:42.480 align:middle line:84% In the nested logic model, you add a second shock. 00:30:42.480 --> 00:30:45.320 align:middle line:84% So the epsilon ij is the shock that's 00:30:45.320 --> 00:30:46.925 align:middle line:90% independent across stations. 00:30:46.925 --> 00:30:48.300 align:middle line:84% I like the music on this station. 00:30:48.300 --> 00:30:49.758 align:middle line:84% I don't like music on this station. 00:30:49.758 --> 00:30:53.210 align:middle line:84% That's additive to my preferences. 00:30:53.210 --> 00:30:56.840 align:middle line:84% But then there's also a common shock, vi. 00:30:56.840 --> 00:30:59.840 align:middle line:84% And the vi is a random variable that's 00:30:59.840 --> 00:31:01.850 align:middle line:84% drawn randomly at the level of the customer, 00:31:01.850 --> 00:31:03.150 align:middle line:90% not at the station. 00:31:03.150 --> 00:31:06.620 align:middle line:84% And it is their taste for music as a whole versus not listening 00:31:06.620 --> 00:31:07.820 align:middle line:90% to the radio. 00:31:07.820 --> 00:31:09.680 align:middle line:90% So we have these two shocks-- 00:31:09.680 --> 00:31:15.740 align:middle line:84% one common shock that customer i gets regardless of which station 00:31:15.740 --> 00:31:16.790 align:middle line:90% they listen to. 00:31:16.790 --> 00:31:20.270 align:middle line:84% And then one set of-- and then a set of idiosyncratic shocks 00:31:20.270 --> 00:31:23.270 align:middle line:84% that are independent across stations that you get if 00:31:23.270 --> 00:31:26.780 align:middle line:84% and only if you listen to that particular station. 00:31:26.780 --> 00:31:29.120 align:middle line:84% And the idea here is by having a sigma that 00:31:29.120 --> 00:31:31.470 align:middle line:90% can vary between 0 and 1-- 00:31:31.470 --> 00:31:36.140 align:middle line:84% so sigma equals 0 is the pure logit model, where 00:31:36.140 --> 00:31:39.980 align:middle line:84% when you add another station, it draws customers 00:31:39.980 --> 00:31:42.050 align:middle line:84% equally from all of the other stations 00:31:42.050 --> 00:31:43.760 align:middle line:90% and from the nonlisteners. 00:31:43.760 --> 00:31:46.310 align:middle line:84% And in radio, nonlisteners are most people. 00:31:46.310 --> 00:31:49.550 align:middle line:84% In most half hours, most people in the United States 00:31:49.550 --> 00:31:51.118 align:middle line:90% are not listening to the radio. 00:31:51.118 --> 00:31:52.410 align:middle line:90% It's actually surprisingly few. 00:31:52.410 --> 00:31:54.710 align:middle line:84% It's about-- I think it's like 85%. 00:31:54.710 --> 00:31:58.340 align:middle line:84% So typical half hour, 85% of people in United States 00:31:58.340 --> 00:32:00.290 align:middle line:90% are not listening to the radio. 00:32:00.290 --> 00:32:03.230 align:middle line:84% But so then if you have the pure logit, that 00:32:03.230 --> 00:32:06.740 align:middle line:84% means when a new station attracts someone, 00:32:06.740 --> 00:32:08.900 align:middle line:84% it's 85% chance that they're expanding the market 00:32:08.900 --> 00:32:11.150 align:middle line:84% and choosing someone who wasn't listening to the radio 00:32:11.150 --> 00:32:12.470 align:middle line:90% to begin with. 00:32:12.470 --> 00:32:17.130 align:middle line:84% The other extreme, sigma equals 1, this shock goes away. 00:32:17.130 --> 00:32:19.400 align:middle line:90% It's all the common shock. 00:32:19.400 --> 00:32:21.650 align:middle line:84% And when it's all the common shock, 00:32:21.650 --> 00:32:24.170 align:middle line:90% then it's 100% business dealing. 00:32:24.170 --> 00:32:26.810 align:middle line:84% If you only have common shocks, and the idiosyncratic means are 00:32:26.810 --> 00:32:29.300 align:middle line:84% essentially zero, you add a new station, 00:32:29.300 --> 00:32:31.800 align:middle line:84% it's just going to pull people away from the other stations. 00:32:31.800 --> 00:32:33.175 align:middle line:84% And it's only the people who have 00:32:33.175 --> 00:32:36.410 align:middle line:84% the vi's that are big enough to overcome delta that will listen 00:32:36.410 --> 00:32:40.470 align:middle line:90% to radio regardless. 00:32:40.470 --> 00:32:43.042 align:middle line:84% Anyway, so this-- they have this model. 00:32:43.042 --> 00:32:44.500 align:middle line:84% It's got this free parameter sigma. 00:32:44.500 --> 00:32:46.290 align:middle line:84% And that's going to let them say how important business 00:32:46.290 --> 00:32:46.790 align:middle line:90% dealing is. 00:32:46.790 --> 00:32:48.540 align:middle line:84% And it's very important to have that 00:32:48.540 --> 00:32:49.923 align:middle line:90% because a business dealing-- 00:32:49.923 --> 00:32:51.840 align:middle line:84% we know that there's a lot of business dealing 00:32:51.840 --> 00:32:52.993 align:middle line:90% if there's too much entry. 00:32:52.993 --> 00:32:54.660 align:middle line:84% If there's very little business dealing, 00:32:54.660 --> 00:32:59.700 align:middle line:84% then we might have insufficient entry. 00:32:59.700 --> 00:33:01.950 align:middle line:90% If you have a totally-- 00:33:01.950 --> 00:33:03.532 align:middle line:84% if this model was totally symmetric, 00:33:03.532 --> 00:33:06.240 align:middle line:84% if all the deltas were the same, if all the stations were exactly 00:33:06.240 --> 00:33:12.300 align:middle line:84% the same quality, then the nested logit model N l star of N 00:33:12.300 --> 00:33:14.790 align:middle line:84% would be N to the 1 minus sigma power over e 00:33:14.790 --> 00:33:17.500 align:middle line:84% to the minus delta plus the N to the 1 minus sigma. 00:33:17.500 --> 00:33:21.150 align:middle line:84% So again, for sigma equals 1, total listenership 00:33:21.150 --> 00:33:23.640 align:middle line:84% is just completely fixed at 1 over 1 00:33:23.640 --> 00:33:25.890 align:middle line:84% plus e to the minus delta, regardless 00:33:25.890 --> 00:33:27.430 align:middle line:90% of the number of consumers. 00:33:27.430 --> 00:33:34.500 align:middle line:84% But then when sigma goes to zero, this thing is like-- 00:33:34.500 --> 00:33:37.980 align:middle line:84% it's asymptoting towards 1, but it's going up initially 00:33:37.980 --> 00:33:39.830 align:middle line:90% linearly then. 00:33:39.830 --> 00:33:44.050 align:middle line:84% Because it's like N to the first over this plus something small. 00:33:44.050 --> 00:33:45.995 align:middle line:90% So it's going up again. 00:33:45.995 --> 00:33:53.720 align:middle line:90% 00:33:53.720 --> 00:33:55.670 align:middle line:90% Anyway, so they estimate this-- 00:33:55.670 --> 00:33:58.760 align:middle line:84% we're estimating this using the data on how many people listen 00:33:58.760 --> 00:34:01.700 align:middle line:84% as a function of N. And they have an IV approach 00:34:01.700 --> 00:34:04.400 align:middle line:90% where they use population. 00:34:04.400 --> 00:34:07.910 align:middle line:84% You might overesti-- underestimate the effect of N 00:34:07.910 --> 00:34:10.219 align:middle line:84% on markets because you'll see a lot of-- 00:34:10.219 --> 00:34:11.900 align:middle line:84% or overestimate it because you'd see 00:34:11.900 --> 00:34:14.120 align:middle line:84% a lot of stations entering the markets that 00:34:14.120 --> 00:34:15.290 align:middle line:90% have a lot of demand. 00:34:15.290 --> 00:34:17.623 align:middle line:84% Like, there are some cities where people like the radio, 00:34:17.623 --> 00:34:20.540 align:middle line:84% like Nashville, and some like Boston where they don't, then 00:34:20.540 --> 00:34:23.270 align:middle line:84% you might get more firms in Nashville than in Boston 00:34:23.270 --> 00:34:25.190 align:middle line:84% because they're meeting the demand. 00:34:25.190 --> 00:34:28.040 align:middle line:84% And so the instrument for the number of firms, 00:34:28.040 --> 00:34:30.230 align:middle line:84% they use the population as an instrument 00:34:30.230 --> 00:34:31.230 align:middle line:90% for the number of firms. 00:34:31.230 --> 00:34:34.340 align:middle line:84% And so they're sometimes looking at how does listening 00:34:34.340 --> 00:34:40.135 align:middle line:84% compare in cities with a lot of people versus a few people. 00:34:40.135 --> 00:34:41.760 align:middle line:84% Then the second thing they do is just-- 00:34:41.760 --> 00:34:44.639 align:middle line:84% they want to estimate this downward-sloping advertising 00:34:44.639 --> 00:34:46.770 align:middle line:90% demand curve. 00:34:46.770 --> 00:34:49.590 align:middle line:84% They're going to assume that the advertising prices depend 00:34:49.590 --> 00:34:53.040 align:middle line:84% on the share of people listening to the radio and market k. 00:34:53.040 --> 00:34:59.160 align:middle line:84% And so it's just the log of price of advertising in market k 00:34:59.160 --> 00:35:04.660 align:middle line:84% is equal to some characteristics of the city, plus something-- 00:35:04.660 --> 00:35:07.560 align:middle line:84% if you have a downward-sloping demand in the number 00:35:07.560 --> 00:35:08.820 align:middle line:90% of listeners-- 00:35:08.820 --> 00:35:11.730 align:middle line:84% listener hours that there are plus omega k. 00:35:11.730 --> 00:35:15.970 align:middle line:84% So, again, the idea is the more listener hours you have, 00:35:15.970 --> 00:35:17.970 align:middle line:84% the lower is the willingness to pay per listener 00:35:17.970 --> 00:35:22.800 align:middle line:90% hour for the commercials. 00:35:22.800 --> 00:35:25.410 align:middle line:90% And then the fixed costs. 00:35:25.410 --> 00:35:29.640 align:middle line:84% In some sense, this is like Bresnahan and Reiss kind 00:35:29.640 --> 00:35:31.790 align:middle line:90% of approach, that you just-- 00:35:31.790 --> 00:35:34.680 align:middle line:84% you are trying to rationalize the number of radio stations 00:35:34.680 --> 00:35:35.490 align:middle line:90% in every market. 00:35:35.490 --> 00:35:38.220 align:middle line:84% And so you rationalize it by assuming 00:35:38.220 --> 00:35:40.830 align:middle line:90% that there are these errors. 00:35:40.830 --> 00:35:42.420 align:middle line:84% In this case, they do errors drawn 00:35:42.420 --> 00:35:44.580 align:middle line:84% from a log-normal distribution instead of 00:35:44.580 --> 00:35:46.900 align:middle line:90% from a normal distribution. 00:35:46.900 --> 00:35:49.710 align:middle line:90% And they do the-- 00:35:49.710 --> 00:35:51.360 align:middle line:84% estimate two parameters-- the mean 00:35:51.360 --> 00:35:55.170 align:middle line:84% and variance of the log-normal distribution. 00:35:55.170 --> 00:36:00.150 align:middle line:84% But anyway, that's just like Bresnahan and Reiss. 00:36:00.150 --> 00:36:04.720 align:middle line:84% And the data have information about 3,285 stations, 00:36:04.720 --> 00:36:06.610 align:middle line:90% 135 markets. 00:36:06.610 --> 00:36:08.640 align:middle line:84% So they have individual station characteristics 00:36:08.640 --> 00:36:11.430 align:middle line:84% that they're using in some places. 00:36:11.430 --> 00:36:14.430 align:middle line:84% These regressions, like the advertising demand regressions, 00:36:14.430 --> 00:36:15.720 align:middle line:90% those are just run-- 00:36:15.720 --> 00:36:18.400 align:middle line:84% their advertising data is purely at the market level. 00:36:18.400 --> 00:36:20.940 align:middle line:84% It's not at the station market level. 00:36:20.940 --> 00:36:24.710 align:middle line:84% So there, they're just using market level advertising prices 00:36:24.710 --> 00:36:27.130 align:middle line:84% they get from some consulting firm that 00:36:27.130 --> 00:36:28.940 align:middle line:90% provides advertising prices. 00:36:28.940 --> 00:36:32.200 align:middle line:84% I think it actually comes from the revenues and rate. 00:36:32.200 --> 00:36:35.320 align:middle line:84% It comes with radio stations advertising-- 00:36:35.320 --> 00:36:38.710 align:middle line:90% reported advertising revenues. 00:36:38.710 --> 00:36:41.470 align:middle line:84% The raw data, which is going to tell you 00:36:41.470 --> 00:36:45.280 align:middle line:84% a lot about what the results come out, is here in the paper. 00:36:45.280 --> 00:36:48.130 align:middle line:84% So this graph is as the number of stations 00:36:48.130 --> 00:36:51.172 align:middle line:84% in a market increases, how does listening go up? 00:36:51.172 --> 00:36:52.630 align:middle line:84% And you might say, OK, I definitely 00:36:52.630 --> 00:36:54.400 align:middle line:90% see that positive slope-- 00:36:54.400 --> 00:36:57.310 align:middle line:84% more radio stations, more listening. 00:36:57.310 --> 00:37:00.400 align:middle line:84% But if you look at this graph, this scale 00:37:00.400 --> 00:37:04.070 align:middle line:84% is really running from 0.1 to 0.18, or 0.12 to 0.18. 00:37:04.070 --> 00:37:06.540 align:middle line:84% So when you have 15 radio stations, 00:37:06.540 --> 00:37:09.040 align:middle line:84% it's looking like you're getting 14% of the people listening 00:37:09.040 --> 00:37:10.810 align:middle line:90% to the radio. 00:37:10.810 --> 00:37:15.140 align:middle line:84% When you have 45 radio stations, you're getting like 15 or 16. 00:37:15.140 --> 00:37:19.060 align:middle line:84% So we're tripling the number of radio stations in the market, 00:37:19.060 --> 00:37:23.860 align:middle line:84% and listening is going from 14% to 16% or something like that. 00:37:23.860 --> 00:37:28.330 align:middle line:84% So this is saying that it does seem like radio station entry is 00:37:28.330 --> 00:37:30.400 align:middle line:90% mostly business stealing. 00:37:30.400 --> 00:37:33.250 align:middle line:84% And, again, we think that this graph, if you drew the OLS, 00:37:33.250 --> 00:37:37.150 align:middle line:84% might be overstated how much market expansion there is. 00:37:37.150 --> 00:37:40.060 align:middle line:84% So when you're on the IV, it's not surprising that it says, 00:37:40.060 --> 00:37:44.080 align:middle line:84% yes, there's an awful lot of business dealing here. 00:37:44.080 --> 00:37:45.430 align:middle line:90% And then-- OK, yeah. 00:37:45.430 --> 00:37:50.830 align:middle line:84% This is just showing you where the instrument comes from. 00:37:50.830 --> 00:37:53.770 align:middle line:84% That is, it is true that the cities that are small, 00:37:53.770 --> 00:37:55.870 align:middle line:84% the cities of a few hundred thousand, 00:37:55.870 --> 00:37:57.980 align:middle line:84% tend to have only 10 or 15 radio stations. 00:37:57.980 --> 00:37:59.560 align:middle line:84% And as markets get bigger, you're 00:37:59.560 --> 00:38:00.860 align:middle line:90% getting more radio stations. 00:38:00.860 --> 00:38:02.625 align:middle line:90% And so that's the first stage. 00:38:02.625 --> 00:38:05.650 align:middle line:90% 00:38:05.650 --> 00:38:07.633 align:middle line:90% And then the advertising prices. 00:38:07.633 --> 00:38:11.350 align:middle line:90% 00:38:11.350 --> 00:38:13.600 align:middle line:84% As the listening share increases-- 00:38:13.600 --> 00:38:19.510 align:middle line:84% this is per listener hour advertising price or advertising 00:38:19.510 --> 00:38:20.470 align:middle line:90% revenue-- 00:38:20.470 --> 00:38:23.530 align:middle line:84% advertising revenues per listener hour 00:38:23.530 --> 00:38:29.687 align:middle line:84% are clearly declining as the listening share is going up. 00:38:29.687 --> 00:38:32.270 align:middle line:84% And the model for this would be there are some things that you 00:38:32.270 --> 00:38:33.870 align:middle line:90% need to advertise on the radio. 00:38:33.870 --> 00:38:37.340 align:middle line:84% And so if there's very few radio commercial slots available, 00:38:37.340 --> 00:38:38.485 align:middle line:90% we pay a lot for them. 00:38:38.485 --> 00:38:39.860 align:middle line:84% And then beyond that, you're just 00:38:39.860 --> 00:38:42.530 align:middle line:84% left with playing radio ads for things that you 00:38:42.530 --> 00:38:44.155 align:middle line:90% could advertise anywhere else. 00:38:44.155 --> 00:38:45.530 align:middle line:84% And the advertiser-- there's just 00:38:45.530 --> 00:38:47.630 align:middle line:84% fewer advertisers who really want those ads [INAUDIBLE] 00:38:47.630 --> 00:38:49.020 align:middle line:90% the prices after the decline. 00:38:49.020 --> 00:38:55.100 align:middle line:90% 00:38:55.100 --> 00:38:56.310 align:middle line:90% So what do they get? 00:38:56.310 --> 00:38:58.670 align:middle line:84% So anyway, they estimate sigma close to 1, entries 00:38:58.670 --> 00:39:00.000 align:middle line:90% mostly business stealing. 00:39:00.000 --> 00:39:02.900 align:middle line:84% Therefore, there's going to be too much entry. 00:39:02.900 --> 00:39:07.840 align:middle line:84% Almost shocking results here on how much excessive entry there 00:39:07.840 --> 00:39:08.340 align:middle line:90% is. 00:39:08.340 --> 00:39:14.150 align:middle line:84% So welfare under free entry is $5 billion. 00:39:14.150 --> 00:39:18.350 align:middle line:84% A social planner could increase social welfare by almost 40% 00:39:18.350 --> 00:39:20.870 align:middle line:84% by restricting the number of radio stations. 00:39:20.870 --> 00:39:25.040 align:middle line:84% If we even had a monopolist-- so there was just one firm that 00:39:25.040 --> 00:39:27.133 align:middle line:84% had a unique license to operate all radio 00:39:27.133 --> 00:39:29.550 align:middle line:84% stations in each market, and that monopolist could operate 00:39:29.550 --> 00:39:31.860 align:middle line:84% as many stations as it wanted, it 00:39:31.860 --> 00:39:34.320 align:middle line:84% would say we would do almost as well under monopolist as we 00:39:34.320 --> 00:39:36.060 align:middle line:90% do under the social planner. 00:39:36.060 --> 00:39:39.420 align:middle line:84% So the estimates imply that we want very few radio stations. 00:39:39.420 --> 00:39:41.170 align:middle line:84% A monopolist wouldn't open just one. 00:39:41.170 --> 00:39:46.140 align:middle line:84% A monopolist would open three or four or five 00:39:46.140 --> 00:39:47.200 align:middle line:90% stations per market. 00:39:47.200 --> 00:39:49.530 align:middle line:84% But in some sense, the model is saying 00:39:49.530 --> 00:39:51.780 align:middle line:84% that there's just an awful lot of wasteful duplication 00:39:51.780 --> 00:39:54.270 align:middle line:84% of fixed costs, so we could get by with many fewer radio 00:39:54.270 --> 00:39:56.220 align:middle line:90% stations. 00:39:56.220 --> 00:40:03.720 align:middle line:84% So in their data set, there are 2,509 within metro stations. 00:40:03.720 --> 00:40:06.870 align:middle line:84% And they basically estimate that we have four times as many 00:40:06.870 --> 00:40:14.850 align:middle line:84% as we want, that optimal would be to only have 649 stations. 00:40:14.850 --> 00:40:17.820 align:middle line:84% What would happen in this world where 3/4 of the radio stations 00:40:17.820 --> 00:40:19.680 align:middle line:90% disappeared? 00:40:19.680 --> 00:40:23.920 align:middle line:84% Advertising price would go from 277 to 336. 00:40:23.920 --> 00:40:28.560 align:middle line:84% So it goes up, but only moderately. 00:40:28.560 --> 00:40:33.460 align:middle line:84% Listening share would go from 12.9% of the population to 9.3%. 00:40:33.460 --> 00:40:37.800 align:middle line:84% So it is true that more than a quarter of all radio listening 00:40:37.800 --> 00:40:40.440 align:middle line:84% would stop when all those stations disappeared. 00:40:40.440 --> 00:40:43.590 align:middle line:84% But the big social welfare gain is the fixed costs 00:40:43.590 --> 00:40:47.590 align:middle line:84% of operating all these radio stations drops from $5 billion 00:40:47.590 --> 00:40:49.440 align:middle line:90% to $1 billion. 00:40:49.440 --> 00:40:51.840 align:middle line:84% So I just have to say, this does make you 00:40:51.840 --> 00:40:53.340 align:middle line:90% wonder a little bit about this. 00:40:53.340 --> 00:40:57.060 align:middle line:84% Does this seem like some Marvel superhero movie where 00:40:57.060 --> 00:40:58.530 align:middle line:84% there's this evil guy who's like, 00:40:58.530 --> 00:41:02.070 align:middle line:84% I'm going to destroy 3/4 of all the people on Earth, 00:41:02.070 --> 00:41:04.200 align:middle line:84% and then everyone's going to be better off somehow. 00:41:04.200 --> 00:41:08.730 align:middle line:84% So these guys, they're very well [INAUDIBLE] and destroy 3/4 00:41:08.730 --> 00:41:09.830 align:middle line:90% of all the radio stations. 00:41:09.830 --> 00:41:11.580 align:middle line:84% And yes, fewer people are going to listen, 00:41:11.580 --> 00:41:13.290 align:middle line:84% but we're going to save all this money, 00:41:13.290 --> 00:41:16.140 align:middle line:84% and we're all going to be better off here. 00:41:16.140 --> 00:41:18.120 align:middle line:84% One thing they note is that it is true 00:41:18.120 --> 00:41:20.280 align:middle line:84% that the advertising listening share goes down 00:41:20.280 --> 00:41:22.520 align:middle line:84% from 12.9% to 9.3% Again, remember, 00:41:22.520 --> 00:41:24.270 align:middle line:84% this is a calculation where there are only 00:41:24.270 --> 00:41:26.830 align:middle line:84% two types of people in the world. 00:41:26.830 --> 00:41:30.540 align:middle line:84% There are radio stations, and there are advertisers. 00:41:30.540 --> 00:41:32.430 align:middle line:84% And the people are just intermediaries 00:41:32.430 --> 00:41:35.520 align:middle line:84% who are just being used to generate ad revenue. 00:41:35.520 --> 00:41:39.540 align:middle line:84% They ask, what happened-- how important would people have 00:41:39.540 --> 00:41:43.140 align:middle line:84% to be and people's enjoyment of radio have to be in order to not 00:41:43.140 --> 00:41:45.460 align:middle line:84% want to kill 3/4 of the radio stations? 00:41:45.460 --> 00:41:48.700 align:middle line:84% And my take on the number is that it's really kind of small. 00:41:48.700 --> 00:41:52.373 align:middle line:84% So the advertiser revenue per listener hour, 00:41:52.373 --> 00:41:53.790 align:middle line:84% if I'm reading the data correctly, 00:41:53.790 --> 00:41:56.455 align:middle line:84% is about 4.2 cents per listener hour. 00:41:56.455 --> 00:41:58.830 align:middle line:84% That is, if you think about it, like you listen to radio, 00:41:58.830 --> 00:42:00.420 align:middle line:84% how often you actually buy anything 00:42:00.420 --> 00:42:01.690 align:middle line:90% that you hear in a radio ad? 00:42:01.690 --> 00:42:03.690 align:middle line:84% How much would someone be willing to pay to have 00:42:03.690 --> 00:42:04.732 align:middle line:90% you hear those radio ads? 00:42:04.732 --> 00:42:06.960 align:middle line:90% The answer is almost nothing. 00:42:06.960 --> 00:42:11.040 align:middle line:84% So to offset all these losses, you would need consumer surplus 00:42:11.040 --> 00:42:16.290 align:middle line:84% to be about 13 and 1/2 cents per listener hour to say we 00:42:16.290 --> 00:42:18.780 align:middle line:84% don't want to kill 3/4 of the radio stations in the United 00:42:18.780 --> 00:42:19.300 align:middle line:90% States. 00:42:19.300 --> 00:42:23.040 align:middle line:84% And again, I-- again, I don't know how much 00:42:23.040 --> 00:42:25.363 align:middle line:84% people are willing to pay for listening to the radio. 00:42:25.363 --> 00:42:28.030 align:middle line:84% It is true that satellite radio doesn't get that many customers. 00:42:28.030 --> 00:42:29.040 align:middle line:84% And I don't know how much Spotify 00:42:29.040 --> 00:42:30.210 align:middle line:90% charges per listener hour. 00:42:30.210 --> 00:42:32.772 align:middle line:84% But 13 and 1/2 cents is not such a big number. 00:42:32.772 --> 00:42:34.230 align:middle line:84% And so, in some sense, you wouldn't 00:42:34.230 --> 00:42:36.630 align:middle line:84% need customers to be all that important 00:42:36.630 --> 00:42:39.130 align:middle line:84% and not want to get rid of all the firms in this way. 00:42:39.130 --> 00:42:42.610 align:middle line:90% 00:42:42.610 --> 00:42:45.700 align:middle line:90% Any questions on that? 00:42:45.700 --> 00:42:47.710 align:middle line:84% Oh, these are just-- yeah, anyway. 00:42:47.710 --> 00:42:48.800 align:middle line:90% Had the graphs up there. 00:42:48.800 --> 00:42:49.300 align:middle line:90% So yeah. 00:42:49.300 --> 00:42:51.700 align:middle line:84% So the monopolist different numbers-- so like, monopolist 00:42:51.700 --> 00:42:52.900 align:middle line:84% is nearly as good in their market. 00:42:52.900 --> 00:42:54.400 align:middle line:84% The monopolist gets all the way down 00:42:54.400 --> 00:42:58.130 align:middle line:84% to just 341 radio stations in 135 cities. 00:42:58.130 --> 00:43:01.540 align:middle line:84% So we're going down literally less than three radio stations 00:43:01.540 --> 00:43:02.830 align:middle line:90% per city. 00:43:02.830 --> 00:43:06.100 align:middle line:84% And again, the listening share dropped substantially 00:43:06.100 --> 00:43:09.370 align:middle line:90% under monopoly. 00:43:09.370 --> 00:43:12.670 align:middle line:84% So some comments I wanted to make. 00:43:12.670 --> 00:43:16.270 align:middle line:84% In some sense, I didn't need to actually do this paper-- 00:43:16.270 --> 00:43:17.620 align:middle line:90% do the empirics of this paper. 00:43:17.620 --> 00:43:21.880 align:middle line:84% This is-- as I said, they model this as a homogeneous good. 00:43:21.880 --> 00:43:23.920 align:middle line:84% I already gave that theorem that once when 00:43:23.920 --> 00:43:26.390 align:middle line:84% you have a homogeneous good, entry is always excessive. 00:43:26.390 --> 00:43:28.480 align:middle line:84% So whatever parameters, sigma and gamma 00:43:28.480 --> 00:43:31.300 align:middle line:84% and whatever they've estimated, we 00:43:31.300 --> 00:43:32.800 align:middle line:84% know that this model is going to say 00:43:32.800 --> 00:43:35.080 align:middle line:84% there's too much entry if the advertisers-- 00:43:35.080 --> 00:43:39.915 align:middle line:84% if the people are homogeneous goods being sold to advertisers. 00:43:39.915 --> 00:43:41.290 align:middle line:84% It's definitely true that they're 00:43:41.290 --> 00:43:44.360 align:middle line:84% estimating how excessive it is by estimating that sigma 00:43:44.360 --> 00:43:44.870 align:middle line:90% parameter. 00:43:44.870 --> 00:43:47.210 align:middle line:84% And if that sigma parameter had to not be zero, 00:43:47.210 --> 00:43:49.440 align:middle line:84% they would have gotten a much less excessive entry. 00:43:49.440 --> 00:43:53.225 align:middle line:84% But the model is always saying that entry is excessive 00:43:53.225 --> 00:43:55.100 align:middle line:84% and doesn't have estimates of the things that 00:43:55.100 --> 00:43:58.970 align:middle line:84% will let you say there's insufficient entry. 00:43:58.970 --> 00:44:00.480 align:middle line:90% What would those things be? 00:44:00.480 --> 00:44:03.020 align:middle line:84% For instance, if you could estimate 00:44:03.020 --> 00:44:05.100 align:middle line:84% how differentiated are the goods-- 00:44:05.100 --> 00:44:07.880 align:middle line:84% are people who listen to classical music versus top 40 00:44:07.880 --> 00:44:10.770 align:middle line:84% music valued differently by different advertisers? 00:44:10.770 --> 00:44:13.590 align:middle line:84% Do the advertisers gain from match quality? 00:44:13.590 --> 00:44:15.140 align:middle line:84% That would be one thing leading us 00:44:15.140 --> 00:44:17.665 align:middle line:84% to want more stations besides the consumer surplus. 00:44:17.665 --> 00:44:20.870 align:middle line:90% 00:44:20.870 --> 00:44:24.890 align:middle line:84% The paper also is also very limited in letting 00:44:24.890 --> 00:44:27.440 align:middle line:90% quality vary across stations. 00:44:27.440 --> 00:44:30.290 align:middle line:84% I think it is true that there are big stations that 00:44:30.290 --> 00:44:31.590 align:middle line:90% have very big fixed costs. 00:44:31.590 --> 00:44:33.410 align:middle line:84% There are a lot of college radio stations 00:44:33.410 --> 00:44:37.620 align:middle line:84% that have very small fixed costs and very low-powered antennas. 00:44:37.620 --> 00:44:42.200 align:middle line:84% And you imagine that it's not the case that we cut out 3/4 00:44:42.200 --> 00:44:44.430 align:middle line:84% of stations, we cut out 3/4 of the fixed costs. 00:44:44.430 --> 00:44:47.097 align:middle line:84% It may be that all those little stations that no one's listening 00:44:47.097 --> 00:44:48.890 align:middle line:84% to also are there because they all 00:44:48.890 --> 00:44:52.790 align:middle line:84% have very little fixed costs, and we would like to do that. 00:44:52.790 --> 00:44:55.610 align:middle line:84% The other thing that I think they don't have the data to do, 00:44:55.610 --> 00:44:57.380 align:middle line:90% but it would be-- 00:44:57.380 --> 00:44:59.300 align:middle line:84% I think a big question in these media markets 00:44:59.300 --> 00:45:01.850 align:middle line:84% is, what is the number of advertising minutes per hour, 00:45:01.850 --> 00:45:04.820 align:middle line:84% and how is that optimal relative to consumer preferences 00:45:04.820 --> 00:45:06.480 align:middle line:90% and advertiser preferences? 00:45:06.480 --> 00:45:12.740 align:middle line:84% And it would be very nice to have more on-- 00:45:12.740 --> 00:45:16.310 align:middle line:84% know more about how do advertisers choose 00:45:16.310 --> 00:45:17.860 align:middle line:90% the number of minutes per-- 00:45:17.860 --> 00:45:20.360 align:middle line:84% how do radio stations choose the number of minutes per hour, 00:45:20.360 --> 00:45:23.010 align:middle line:84% and what does the trade-off on that look like 00:45:23.010 --> 00:45:26.330 align:middle line:84% And you can imagine that we don't get price variation 00:45:26.330 --> 00:45:29.060 align:middle line:84% on the consumer side, but you do get minutes of advertising 00:45:29.060 --> 00:45:30.480 align:middle line:90% variation on the consumer side. 00:45:30.480 --> 00:45:32.600 align:middle line:84% And so you could imagine that exploiting 00:45:32.600 --> 00:45:35.180 align:middle line:84% that minute of advertising variation 00:45:35.180 --> 00:45:41.690 align:middle line:84% can be an interesting way to get better at what are the-- what's 00:45:41.690 --> 00:45:44.080 align:middle line:84% the consumer surplus of the consumers who are listening 00:45:44.080 --> 00:45:45.560 align:middle line:90% to the free radio? 00:45:45.560 --> 00:45:47.990 align:middle line:84% And I should say there are a number of papers 00:45:47.990 --> 00:45:50.360 align:middle line:84% that do have advertising data to try to address 00:45:50.360 --> 00:45:51.360 align:middle line:90% some of these questions. 00:45:51.360 --> 00:45:52.985 align:middle line:84% But I think it's still an area where we 00:45:52.985 --> 00:45:55.940 align:middle line:90% could do some interesting work. 00:45:55.940 --> 00:45:58.260 align:middle line:84% So then the final thing I want to do 00:45:58.260 --> 00:46:02.240 align:middle line:84% is a couple of papers that are, in some sense, 00:46:02.240 --> 00:46:05.210 align:middle line:84% arguing that often, these entry models are 00:46:05.210 --> 00:46:07.940 align:middle line:90% doing the wrong model of entry. 00:46:07.940 --> 00:46:12.350 align:middle line:84% Most of the models of entry that I discussed in Tirole's textbook 00:46:12.350 --> 00:46:13.830 align:middle line:90% are of this variety. 00:46:13.830 --> 00:46:17.330 align:middle line:84% We have N for-- we have some large population of firms, 00:46:17.330 --> 00:46:20.220 align:middle line:84% and they're all choosing in or out, paying a fixed cost. 00:46:20.220 --> 00:46:21.740 align:middle line:84% And then once they enter, there's 00:46:21.740 --> 00:46:24.378 align:middle line:84% this symmetric competition between them. 00:46:24.378 --> 00:46:25.670 align:middle line:90% I think we know, though, that-- 00:46:25.670 --> 00:46:27.410 align:middle line:84% I've got a radio is an example-- that it's not 00:46:27.410 --> 00:46:28.410 align:middle line:90% symmetric competition. 00:46:28.410 --> 00:46:31.880 align:middle line:84% There are some big firms, and there are some small firms. 00:46:31.880 --> 00:46:37.640 align:middle line:84% And what's in these next two papers is thinking about, 00:46:37.640 --> 00:46:39.500 align:middle line:90% even when you have products-- 00:46:39.500 --> 00:46:41.702 align:middle line:84% Like in radio, we think there are high-quality, 00:46:41.702 --> 00:46:43.910 align:middle line:84% or there's popular music stations that lots of people 00:46:43.910 --> 00:46:44.410 align:middle line:90% listen to. 00:46:44.410 --> 00:46:46.010 align:middle line:84% And then there are esoteric stations 00:46:46.010 --> 00:46:48.290 align:middle line:84% that not a lot of people listen to with low power. 00:46:48.290 --> 00:46:50.090 align:middle line:84% Even when you look at products that 00:46:50.090 --> 00:46:52.850 align:middle line:84% seem like they should be homogeneous, 00:46:52.850 --> 00:46:54.830 align:middle line:84% it seems like symmetry of competition 00:46:54.830 --> 00:46:56.765 align:middle line:90% is not a correct answer. 00:46:56.765 --> 00:46:58.640 align:middle line:84% And so these papers on entry that are, again, 00:46:58.640 --> 00:47:00.182 align:middle line:84% also papers really about competition, 00:47:00.182 --> 00:47:03.650 align:middle line:84% are arguing that this model of entry and then 00:47:03.650 --> 00:47:05.270 align:middle line:84% symmetric competition seems like it 00:47:05.270 --> 00:47:06.750 align:middle line:90% doesn't fit the data very well. 00:47:06.750 --> 00:47:09.830 align:middle line:84% So anyway, here's an example of homogeneous product-- 00:47:09.830 --> 00:47:10.610 align:middle line:90% mayonnaise. 00:47:10.610 --> 00:47:12.560 align:middle line:84% I've heard that if you do blind taste tests, 00:47:12.560 --> 00:47:15.240 align:middle line:84% people have a very hard time telling one brand of mayonnaise 00:47:15.240 --> 00:47:16.865 align:middle line:84% apart from another brand of mayonnaise. 00:47:16.865 --> 00:47:18.865 align:middle line:84% Shouldn't mayonnaise just be a homogeneous good, 00:47:18.865 --> 00:47:20.630 align:middle line:84% where there's a quantity of mayonnaise, 00:47:20.630 --> 00:47:22.130 align:middle line:84% market-clearing price in mayonnaise. 00:47:22.130 --> 00:47:24.422 align:middle line:84% Anyway, it turns out empirically that that's actually-- 00:47:24.422 --> 00:47:25.100 align:middle line:90% that's not true. 00:47:25.100 --> 00:47:27.170 align:middle line:84% Well, it's not true that mayonnaise doesn't all 00:47:27.170 --> 00:47:27.900 align:middle line:90% taste the same. 00:47:27.900 --> 00:47:30.815 align:middle line:84% It's not-- what's not true is that people treat mayonnaise 00:47:30.815 --> 00:47:32.815 align:middle line:84% like a homogeneous product when they're deciding 00:47:32.815 --> 00:47:35.990 align:middle line:90% what to buy in the supermarket. 00:47:35.990 --> 00:47:38.780 align:middle line:84% So, in particular, one of the things that's-- 00:47:38.780 --> 00:47:41.400 align:middle line:90% 00:47:41.400 --> 00:47:44.040 align:middle line:84% so there's this paper by Golder and Tellis 00:47:44.040 --> 00:47:46.740 align:middle line:84% that's a few decades old now, noting 00:47:46.740 --> 00:47:49.890 align:middle line:84% that there are many, many leading brands that 00:47:49.890 --> 00:47:52.290 align:middle line:84% have been leaders for a very, very long time. 00:47:52.290 --> 00:47:54.480 align:middle line:84% Like, when did Coca-Cola become the leading cola 00:47:54.480 --> 00:47:55.840 align:middle line:90% in the United States? 00:47:55.840 --> 00:47:58.050 align:middle line:84% It's 100 and something years ago. 00:47:58.050 --> 00:47:59.970 align:middle line:84% And like many of the products you can 00:47:59.970 --> 00:48:02.760 align:middle line:90% think of have been the same-- 00:48:02.760 --> 00:48:05.100 align:middle line:84% the same leader has been there for decades 00:48:05.100 --> 00:48:06.390 align:middle line:90% and decades and decades. 00:48:06.390 --> 00:48:11.490 align:middle line:84% And Eveready batteries or whatever, 00:48:11.490 --> 00:48:13.500 align:middle line:90% like the Pampers diapers. 00:48:13.500 --> 00:48:15.270 align:middle line:84% These same firms just persist for decade 00:48:15.270 --> 00:48:18.270 align:middle line:90% after decade after decade. 00:48:18.270 --> 00:48:21.390 align:middle line:84% The interesting fact that Bronnenberg, Dhar, and Dubé were 00:48:21.390 --> 00:48:23.687 align:middle line:84% bringing out in their paper is that not only is it true 00:48:23.687 --> 00:48:25.895 align:middle line:84% that in many consumer goods there are market leaders, 00:48:25.895 --> 00:48:30.170 align:middle line:84% but those market leaders differ in similar markets. 00:48:30.170 --> 00:48:33.410 align:middle line:84% So this is the mayonnaise graph from their paper. 00:48:33.410 --> 00:48:38.390 align:middle line:84% So there are two main mayonnaise firms in the United States, 00:48:38.390 --> 00:48:40.860 align:middle line:84% at least in the-- probably in the South. 00:48:40.860 --> 00:48:43.780 align:middle line:84% There's Kraft, and there's Unilever. 00:48:43.780 --> 00:48:48.070 align:middle line:84% And Kraft produces Kraft mayonnaise. 00:48:48.070 --> 00:48:50.362 align:middle line:84% Unilever produces two brands, depending 00:48:50.362 --> 00:48:51.320 align:middle line:90% on what city you're in. 00:48:51.320 --> 00:48:53.990 align:middle line:84% They produce something called Hellmann's Real mayonnaise. 00:48:53.990 --> 00:48:56.510 align:middle line:84% They also produce something called Best Foods mayonnaise. 00:48:56.510 --> 00:48:59.500 align:middle line:84% You can tell from the jars that Hellmann's and Best Foods 00:48:59.500 --> 00:49:00.830 align:middle line:90% mayonnaises are-- 00:49:00.830 --> 00:49:03.470 align:middle line:84% they're the same mayonnaise produced by Unilever. 00:49:03.470 --> 00:49:05.167 align:middle line:84% Just in some cities, they sell it 00:49:05.167 --> 00:49:07.000 align:middle line:84% as Hellmann's, and some cities, they sell it 00:49:07.000 --> 00:49:09.340 align:middle line:90% as Best Foods mayonnaise. 00:49:09.340 --> 00:49:12.430 align:middle line:84% But then what's striking is this is data on-- 00:49:12.430 --> 00:49:14.260 align:middle line:84% over the course of half a year, I 00:49:14.260 --> 00:49:17.290 align:middle line:84% think-- on what's the market share by week 00:49:17.290 --> 00:49:22.840 align:middle line:84% of Kraft mayonnaise versus Unilever mayonnaise in Denver? 00:49:22.840 --> 00:49:26.920 align:middle line:84% And you can see in Denver, 65%, 70% of people 00:49:26.920 --> 00:49:31.120 align:middle line:84% buy Kraft mayonnaise, and 15% to 20% of people 00:49:31.120 --> 00:49:33.490 align:middle line:90% buy the Unilever mayonnaise. 00:49:33.490 --> 00:49:35.330 align:middle line:90% And then you go to Los Angeles. 00:49:35.330 --> 00:49:37.930 align:middle line:84% And in Los Angeles, only 25% of the people 00:49:37.930 --> 00:49:41.680 align:middle line:84% are buying Kraft mayonnaise, and 60%, 00:49:41.680 --> 00:49:44.080 align:middle line:84% I think it's the Best Foods mayonnaise sold 00:49:44.080 --> 00:49:45.340 align:middle line:90% on the West Coast. 00:49:45.340 --> 00:49:49.840 align:middle line:84% And so it's striking that like, in some sense, in Denver, 00:49:49.840 --> 00:49:52.270 align:middle line:84% Kraft is the clear mayonnaise leader. 00:49:52.270 --> 00:49:58.160 align:middle line:84% And in Los Angeles, Unilever is the clear mayonnaise leader. 00:49:58.160 --> 00:50:00.310 align:middle line:84% And so there are these different market leaders, 00:50:00.310 --> 00:50:02.350 align:middle line:84% and they've been different leaders. 00:50:02.350 --> 00:50:05.110 align:middle line:84% And it seems quite stable that there's a leading 00:50:05.110 --> 00:50:06.440 align:middle line:90% mayonnaise that everyone buys. 00:50:06.440 --> 00:50:10.420 align:middle line:84% And then there's a second mayonnaise that some people buy. 00:50:10.420 --> 00:50:12.800 align:middle line:84% They give other examples in the paper. 00:50:12.800 --> 00:50:15.310 align:middle line:84% If you look at ground coffee, I think 00:50:15.310 --> 00:50:17.650 align:middle line:84% of as another fairly undifferentiated product. 00:50:17.650 --> 00:50:20.700 align:middle line:84% I don't know if you guys are wealthy enough 00:50:20.700 --> 00:50:23.670 align:middle line:84% that you don't drink frozen-ground coffee, 00:50:23.670 --> 00:50:25.350 align:middle line:84% and you go to Starbucks or whatever. 00:50:25.350 --> 00:50:27.880 align:middle line:84% But you buy the frozen-ground coffee in a can-- 00:50:27.880 --> 00:50:30.090 align:middle line:84% I don't know how different the different cans are. 00:50:30.090 --> 00:50:33.690 align:middle line:84% But anyway, Folgers' share ranges from 16% in New York 00:50:33.690 --> 00:50:36.750 align:middle line:90% to 59% in Des Moines, Iowa. 00:50:36.750 --> 00:50:39.480 align:middle line:84% Maxwell House's share is only 4% in Seattle, 00:50:39.480 --> 00:50:41.830 align:middle line:90% but it's 46% in Pittsburgh. 00:50:41.830 --> 00:50:45.305 align:middle line:84% And, again, these shares seem to be persistent in the data. 00:50:45.305 --> 00:50:49.150 align:middle line:90% 00:50:49.150 --> 00:50:53.350 align:middle line:84% So what Bronnenberg, Dhar, and Dubé do to talk about this 00:50:53.350 --> 00:51:00.400 align:middle line:84% is they just investigate this as being a really, really long time 00:51:00.400 --> 00:51:04.540 align:middle line:84% span echo of what happened when these products were introduced. 00:51:04.540 --> 00:51:07.270 align:middle line:84% And so their idea is that in consumer 00:51:07.270 --> 00:51:11.230 align:middle line:84% packaged goods markets, people just-- 00:51:11.230 --> 00:51:14.350 align:middle line:84% somehow there's some model where people learn to buy a brand. 00:51:14.350 --> 00:51:15.305 align:middle line:90% They buy that brand. 00:51:15.305 --> 00:51:16.930 align:middle line:84% And either they buy that brand and they 00:51:16.930 --> 00:51:19.210 align:middle line:84% get used to exactly what that brand tastes like, 00:51:19.210 --> 00:51:21.620 align:middle line:84% and that's what they think mayonnaise should taste like. 00:51:21.620 --> 00:51:24.740 align:middle line:84% Or they just know that I buy Kraft mayonnaise. 00:51:24.740 --> 00:51:26.270 align:middle line:84% My mother bought Kraft mayonnaise. 00:51:26.270 --> 00:51:28.215 align:middle line:84% Her mother bought Kraft mayonnaise before her. 00:51:28.215 --> 00:51:29.590 align:middle line:84% When you go to mayonnaise, you've 00:51:29.590 --> 00:51:31.173 align:middle line:84% seen your mother buy Kraft mayonnaise. 00:51:31.173 --> 00:51:32.650 align:middle line:90% You buy Kraft mayonnaise. 00:51:32.650 --> 00:51:35.500 align:middle line:84% So anyway, that's the story that they're thinking of, 00:51:35.500 --> 00:51:39.250 align:middle line:84% is there's just something about people forming tastes. 00:51:39.250 --> 00:51:42.515 align:middle line:84% It could also be something that's less-- 00:51:42.515 --> 00:51:44.640 align:middle line:84% it could also be something that's less taste based. 00:51:44.640 --> 00:51:48.110 align:middle line:84% It could also be there are some advantages to dominance. 00:51:48.110 --> 00:51:50.270 align:middle line:84% Like, if you're the dominant mayonnaise, 00:51:50.270 --> 00:51:54.140 align:middle line:84% you can go to the supermarket and tell Albertsons, 00:51:54.140 --> 00:51:56.232 align:middle line:84% I want my mayonnaise at eye level. 00:51:56.232 --> 00:51:58.190 align:middle line:84% I want you to put that Unilever mayonnaise down 00:51:58.190 --> 00:51:59.360 align:middle line:90% on the bottom shelf. 00:51:59.360 --> 00:52:01.130 align:middle line:84% And that somehow the dominant mayonnaise 00:52:01.130 --> 00:52:02.407 align:middle line:90% gets put at eye level. 00:52:02.407 --> 00:52:04.490 align:middle line:84% The second mayonnaise gets put on the lower shelf. 00:52:04.490 --> 00:52:07.157 align:middle line:84% And then people are just always seeing the eye level mayonnaise, 00:52:07.157 --> 00:52:08.690 align:middle line:90% and that's what they buy. 00:52:08.690 --> 00:52:09.890 align:middle line:90% And so there's just some-- 00:52:09.890 --> 00:52:15.170 align:middle line:84% sometimes this is a story of dominance producing dominance 00:52:15.170 --> 00:52:19.640 align:middle line:84% and some kind of market power being produced by large market 00:52:19.640 --> 00:52:20.930 align:middle line:90% share. 00:52:20.930 --> 00:52:25.130 align:middle line:84% But what they've done is they go back to-- 00:52:25.130 --> 00:52:26.880 align:middle line:90% they go back 100 years. 00:52:26.880 --> 00:52:29.480 align:middle line:84% And so all of these products, like the Hellmann's mayonnaise 00:52:29.480 --> 00:52:35.000 align:middle line:84% or the Kraft mayonnaise, they find where those products were 00:52:35.000 --> 00:52:39.290 align:middle line:84% first introduced in the early 20th century, 00:52:39.290 --> 00:52:42.270 align:middle line:84% and sometimes in the 19th century. 00:52:42.270 --> 00:52:43.980 align:middle line:90% And then they run-- 00:52:43.980 --> 00:52:45.250 align:middle line:90% they do two things. 00:52:45.250 --> 00:52:49.980 align:middle line:84% So this is their main analysis where they have 40 brands which 00:52:49.980 --> 00:52:52.620 align:middle line:90% they could find these origins. 00:52:52.620 --> 00:52:56.010 align:middle line:84% And they regress the market share 00:52:56.010 --> 00:52:59.010 align:middle line:84% that brand i is getting in city m. 00:52:59.010 --> 00:53:01.440 align:middle line:90% Forget about the c's. 00:53:01.440 --> 00:53:03.040 align:middle line:90% It's a constant. 00:53:03.040 --> 00:53:06.180 align:middle line:84% So Hellmann's is better than Kraft. 00:53:06.180 --> 00:53:10.800 align:middle line:84% Plus these coefficients based on your distance 00:53:10.800 --> 00:53:12.010 align:middle line:90% to the origin city. 00:53:12.010 --> 00:53:13.470 align:middle line:90% Are you the origin city? 00:53:13.470 --> 00:53:15.190 align:middle line:84% Are you 50 miles from the origin city? 00:53:15.190 --> 00:53:19.060 align:middle line:84% Are you 100 miles, 200 miles, 1,000 miles? 00:53:19.060 --> 00:53:21.000 align:middle line:84% The thought being that they don't 00:53:21.000 --> 00:53:23.160 align:middle line:84% know when all these mayonnaises got 00:53:23.160 --> 00:53:24.460 align:middle line:90% to all these different places. 00:53:24.460 --> 00:53:28.536 align:middle line:84% But their idea would be that there's-- 00:53:28.536 --> 00:53:32.160 align:middle line:90% if we look at United States. 00:53:32.160 --> 00:53:35.160 align:middle line:84% So anyway, I have a map of the United States. 00:53:35.160 --> 00:53:37.650 align:middle line:84% And then there's just like Kraft mayonnaise gets 00:53:37.650 --> 00:53:40.642 align:middle line:90% invented here in the year 1895. 00:53:40.642 --> 00:53:42.600 align:middle line:84% And then their thought is that it probably just 00:53:42.600 --> 00:53:43.720 align:middle line:90% spread like this. 00:53:43.720 --> 00:53:45.880 align:middle line:90% And so by 1910, it is here. 00:53:45.880 --> 00:53:47.880 align:middle line:90% And by 1930, it was here. 00:53:47.880 --> 00:53:50.430 align:middle line:84% And so we're seeing the Kraft spreading, 00:53:50.430 --> 00:53:52.740 align:middle line:84% and we're seeing Hellmann's spreading. 00:53:52.740 --> 00:53:57.780 align:middle line:84% And so how far you are from where it started 00:53:57.780 --> 00:54:00.150 align:middle line:84% is going to be how many other mayonnaises were there 00:54:00.150 --> 00:54:00.970 align:middle line:90% before you. 00:54:00.970 --> 00:54:02.987 align:middle line:84% And so if you were near the origin, 00:54:02.987 --> 00:54:05.070 align:middle line:84% you were there and got to establish the early lead 00:54:05.070 --> 00:54:06.270 align:middle line:90% and now dominate. 00:54:06.270 --> 00:54:10.265 align:middle line:84% If you're far from the origin, by the time this mayonnaise gets 00:54:10.265 --> 00:54:11.640 align:middle line:84% to California, people were eating 00:54:11.640 --> 00:54:14.990 align:middle line:84% several other mayonnaises, and so it's stayed behind. 00:54:14.990 --> 00:54:18.060 align:middle line:90% 00:54:18.060 --> 00:54:19.920 align:middle line:84% So that's the main analysis because they 00:54:19.920 --> 00:54:21.733 align:middle line:84% were able to do this for 40 products. 00:54:21.733 --> 00:54:24.150 align:middle line:84% The second thing they did is there were six products where 00:54:24.150 --> 00:54:26.940 align:middle line:84% they were actually able to identify 00:54:26.940 --> 00:54:29.667 align:middle line:84% at the city-by-city level, when did Miller beer 00:54:29.667 --> 00:54:30.375 align:middle line:90% get to Milwaukee? 00:54:30.375 --> 00:54:32.202 align:middle line:84% And when did Miller beer get to Chicago? 00:54:32.202 --> 00:54:33.910 align:middle line:84% When did Miller beer get to Indianapolis? 00:54:33.910 --> 00:54:35.560 align:middle line:90% When did it get to Philadelphia? 00:54:35.560 --> 00:54:37.290 align:middle line:90% When did it get to Miami? 00:54:37.290 --> 00:54:39.900 align:middle line:84% So they actually were able to track when each of the brands 00:54:39.900 --> 00:54:41.860 align:middle line:84% entered all the different cities. 00:54:41.860 --> 00:54:45.810 align:middle line:84% And then they can regress city-specific market shares 00:54:45.810 --> 00:54:48.250 align:middle line:90% on dummies for entry order. 00:54:48.250 --> 00:54:50.370 align:middle line:84% Were you there before your rival? 00:54:50.370 --> 00:54:54.820 align:middle line:84% Which mayonnaise got to Phoenix first is the regression. 00:54:54.820 --> 00:54:58.680 align:middle line:90% 00:54:58.680 --> 00:54:59.180 align:middle line:90% Anyway. 00:54:59.180 --> 00:55:03.300 align:middle line:84% So here's what their data sample looks like. 00:55:03.300 --> 00:55:06.570 align:middle line:90% So again, this is 40 products. 00:55:06.570 --> 00:55:08.270 align:middle line:84% And so for each of these 40 products-- 00:55:08.270 --> 00:55:11.270 align:middle line:84% let's go to mayonnaise down at the bottom-- 00:55:11.270 --> 00:55:12.880 align:middle line:90% mayonnaise at the very bottom. 00:55:12.880 --> 00:55:15.620 align:middle line:84% Kraft mayonnaise-- oh, I was just about right. 00:55:15.620 --> 00:55:17.840 align:middle line:84% Kraft mayonnaise started in Salem, Illinois. 00:55:17.840 --> 00:55:21.020 align:middle line:84% And it started Salem, Illinois in 1931. 00:55:21.020 --> 00:55:23.510 align:middle line:84% And Unilever mayonnaise, which is Hellmann's 00:55:23.510 --> 00:55:24.750 align:middle line:90% on the East Coast-- 00:55:24.750 --> 00:55:29.270 align:middle line:84% Hellmann's mayonnaise started in New York City in the year 1905. 00:55:29.270 --> 00:55:31.550 align:middle line:84% And so what they're looking at is 00:55:31.550 --> 00:55:36.590 align:middle line:84% how do how does the share of Unilever versus Kraft 00:55:36.590 --> 00:55:41.140 align:middle line:84% depend on your distance relative to these two points of origin? 00:55:41.140 --> 00:55:46.490 align:middle line:90% 00:55:46.490 --> 00:55:46.990 align:middle line:90% OK. 00:55:46.990 --> 00:55:49.300 align:middle line:84% So distance to the origin analysis 00:55:49.300 --> 00:55:54.610 align:middle line:84% finds that current market shares are 18 percentage points higher 00:55:54.610 --> 00:55:57.310 align:middle line:84% in the closest cities than in the most distant cities. 00:55:57.310 --> 00:56:01.698 align:middle line:84% So the mean share of these products is like 22%. 00:56:01.698 --> 00:56:03.240 align:middle line:84% So this is saying that you get like-- 00:56:03.240 --> 00:56:07.070 align:middle line:84% you can think of the 18 difference as being-- 00:56:07.070 --> 00:56:07.805 align:middle line:90% you get 31%. 00:56:07.805 --> 00:56:12.630 align:middle line:90% 00:56:12.630 --> 00:56:13.130 align:middle line:90% Yeah. 00:56:13.130 --> 00:56:15.825 align:middle line:90% 00:56:15.825 --> 00:56:16.540 align:middle line:90% Let's see. 00:56:16.540 --> 00:56:19.600 align:middle line:84% So you get like 31% in the closest cities, 00:56:19.600 --> 00:56:24.040 align:middle line:84% and you get 13% in the farthest away cities. 00:56:24.040 --> 00:56:28.000 align:middle line:84% So your market share is just dramatically higher 00:56:28.000 --> 00:56:29.830 align:middle line:84% in the cities near where you started-- 00:56:29.830 --> 00:56:32.500 align:middle line:84% right where you started than cities farther away 00:56:32.500 --> 00:56:33.550 align:middle line:90% where you started. 00:56:33.550 --> 00:56:37.960 align:middle line:84% We get this big effect for your origin city. 00:56:37.960 --> 00:56:42.500 align:middle line:84% And then it's declining as we get to 250 miles, 500 miles, 00:56:42.500 --> 00:56:44.360 align:middle line:90% 1,000 miles. 00:56:44.360 --> 00:56:47.047 align:middle line:84% It then seems to be fairly flat, and then it goes down a lot 00:56:47.047 --> 00:56:49.380 align:middle line:84% when you get to the opposite coast of the United States. 00:56:49.380 --> 00:56:52.670 align:middle line:84% So it's like East Coast products are selling noticeably 00:56:52.670 --> 00:56:53.875 align:middle line:90% less on the West Coast. 00:56:53.875 --> 00:56:55.250 align:middle line:84% West Coast products are selling-- 00:56:55.250 --> 00:56:57.625 align:middle line:84% or West Coast products are selling notably-- considerably 00:56:57.625 --> 00:56:59.660 align:middle line:90% less on the East Coast. 00:56:59.660 --> 00:57:02.990 align:middle line:84% So you may think that with our politics, 00:57:02.990 --> 00:57:06.143 align:middle line:84% the coasts are actually similar to each other, 00:57:06.143 --> 00:57:08.310 align:middle line:84% and it's the middle of the country that's different. 00:57:08.310 --> 00:57:10.610 align:middle line:84% I think this graph is saying, no, it's actually the-- 00:57:10.610 --> 00:57:11.693 align:middle line:90% it really is the distance. 00:57:11.693 --> 00:57:14.235 align:middle line:84% It's like there are your East Coast mayonnaises and your West 00:57:14.235 --> 00:57:15.050 align:middle line:90% Coast mayonnaises. 00:57:15.050 --> 00:57:17.490 align:middle line:84% There are East Coast bagels and West Coast bagels. 00:57:17.490 --> 00:57:21.500 align:middle line:84% And the market shares are most different on the opposite coast 00:57:21.500 --> 00:57:24.510 align:middle line:90% of the country. 00:57:24.510 --> 00:57:28.320 align:middle line:84% And then the second part of the analysis 00:57:28.320 --> 00:57:32.352 align:middle line:84% about just is your market share higher or low-- 00:57:32.352 --> 00:57:33.810 align:middle line:84% higher in the cities that you reach 00:57:33.810 --> 00:57:37.500 align:middle line:84% first versus second relative to your main biggest rival? 00:57:37.500 --> 00:57:40.860 align:middle line:84% And these numbers are more moderate. 00:57:40.860 --> 00:57:42.260 align:middle line:90% They range across products. 00:57:42.260 --> 00:57:45.240 align:middle line:84% So like the Budweiser versus Miller beer, 00:57:45.240 --> 00:57:49.350 align:middle line:84% there's not that much difference between Bud and Miller market 00:57:49.350 --> 00:57:52.140 align:middle line:90% shares in the United States. 00:57:52.140 --> 00:57:54.480 align:middle line:84% On average, whoever got there first 00:57:54.480 --> 00:57:57.870 align:middle line:84% is doing a 1.3 percentage point better 00:57:57.870 --> 00:58:00.330 align:middle line:90% than whoever got there second. 00:58:00.330 --> 00:58:03.690 align:middle line:84% In the mayonnaise example, who got there first versus second 00:58:03.690 --> 00:58:06.500 align:middle line:90% is 6.3 percentage points. 00:58:06.500 --> 00:58:14.694 align:middle line:90% 00:58:14.694 --> 00:58:15.690 align:middle line:90% Sure. 00:58:15.690 --> 00:58:18.090 align:middle line:84% So it is, I think, fairly strong evidence, 00:58:18.090 --> 00:58:20.550 align:middle line:84% especially that first part of there 00:58:20.550 --> 00:58:21.970 align:middle line:90% just are these asymmetries. 00:58:21.970 --> 00:58:24.130 align:middle line:84% And remember, these are events 100 years ago. 00:58:24.130 --> 00:58:28.530 align:middle line:84% And so it does seem like there's very, very long-lasting market 00:58:28.530 --> 00:58:30.400 align:middle line:90% leadership advantages. 00:58:30.400 --> 00:58:34.187 align:middle line:84% And if we're thinking about entrance, 00:58:34.187 --> 00:58:36.270 align:middle line:84% it's not like you enter, and one year later you're 00:58:36.270 --> 00:58:37.437 align:middle line:90% going to be on even footing. 00:58:37.437 --> 00:58:40.020 align:middle line:84% It's like you're going to enter, and 100 years later, you're 00:58:40.020 --> 00:58:43.500 align:middle line:84% still going to have a fraction of the share 00:58:43.500 --> 00:58:45.708 align:middle line:84% that the leading firm had 100 years ago on something 00:58:45.708 --> 00:58:47.250 align:middle line:84% like mayonnaise, where it's clearly-- 00:58:47.250 --> 00:58:48.900 align:middle line:84% it's not like their mayonnaise has just been better 00:58:48.900 --> 00:58:49.665 align:middle line:90% over the years. 00:58:49.665 --> 00:58:53.600 align:middle line:90% 00:58:53.600 --> 00:58:54.100 align:middle line:90% OK. 00:58:54.100 --> 00:58:57.220 align:middle line:84% So then the final paper I was going to cover today-- 00:58:57.220 --> 00:59:00.010 align:middle line:90% Bronnenberg, Dubé, Gentzkow. 00:59:00.010 --> 00:59:03.520 align:middle line:84% So this is really digging into where 00:59:03.520 --> 00:59:06.280 align:middle line:84% do long-lived advantages come from? 00:59:06.280 --> 00:59:08.470 align:middle line:84% They draw two main distinctions for where-- 00:59:08.470 --> 00:59:10.390 align:middle line:84% I think two main types of stories. 00:59:10.390 --> 00:59:12.850 align:middle line:84% One is what they will call demand side stories 00:59:12.850 --> 00:59:15.170 align:middle line:84% or preference side-- preference-based stories. 00:59:15.170 --> 00:59:16.960 align:middle line:84% Their preference-based stories are 00:59:16.960 --> 00:59:20.620 align:middle line:84% people develop tastes for products 00:59:20.620 --> 00:59:22.540 align:middle line:90% that they consume growing up. 00:59:22.540 --> 00:59:25.000 align:middle line:84% Like, I'm just used to this product. 00:59:25.000 --> 00:59:26.680 align:middle line:84% I like the taste of this product. 00:59:26.680 --> 00:59:29.050 align:middle line:84% Or I've just learned from my parents 00:59:29.050 --> 00:59:31.210 align:middle line:84% that this is the right product to buy. 00:59:31.210 --> 00:59:35.290 align:middle line:84% I just-- I have preferences about what this product is. 00:59:35.290 --> 00:59:37.210 align:middle line:84% The supply side stories are the ones 00:59:37.210 --> 00:59:42.040 align:middle line:84% I talked about about things like shelf space or advertising, 00:59:42.040 --> 00:59:45.460 align:middle line:84% where if you're dominant, you get the premier shelf space. 00:59:45.460 --> 00:59:48.920 align:middle line:84% Or if you're dominant, it pays you to put on TV commercials. 00:59:48.920 --> 00:59:50.780 align:middle line:84% You put on TV commercials that reinforce 00:59:50.780 --> 00:59:55.190 align:middle line:84% the people's love for your type-- for your product. 00:59:55.190 --> 00:59:59.450 align:middle line:84% And so what Bronnenberg, Dubé, and Gentzkow want to do is say, 00:59:59.450 --> 01:00:01.760 align:middle line:84% does it look like demand side, or does it 01:00:01.760 --> 01:00:03.860 align:middle line:90% look like supply side? 01:00:03.860 --> 01:00:07.010 align:middle line:84% And they have this sort of very thoughtful idea, which is, how 01:00:07.010 --> 01:00:11.000 align:middle line:84% about to investigate this, let's look at people when they move. 01:00:11.000 --> 01:00:14.000 align:middle line:84% And so like many of you have moved to Boston. 01:00:14.000 --> 01:00:16.610 align:middle line:84% And the question is, once you've moved to Boston, 01:00:16.610 --> 01:00:20.505 align:middle line:84% do you guys buy now the Boston mayonnaise brand? 01:00:20.505 --> 01:00:22.130 align:middle line:84% Or do you guys buy the mayonnaise brand 01:00:22.130 --> 01:00:23.780 align:middle line:90% that was big where you grew up? 01:00:23.780 --> 01:00:25.310 align:middle line:90% So many of you probably-- 01:00:25.310 --> 01:00:29.300 align:middle line:84% I think-- I'd have to guess that Boston is a Hellmann's 01:00:29.300 --> 01:00:32.990 align:middle line:84% mayonnaise city, given that, again, Hellmann's 01:00:32.990 --> 01:00:35.690 align:middle line:84% started in New York, and Kraft started in Illinois. 01:00:35.690 --> 01:00:38.330 align:middle line:84% Again-- and there was a 6 percentage point difference. 01:00:38.330 --> 01:00:40.970 align:middle line:84% But anyway, I would guess that Hellmann's is the Boston 01:00:40.970 --> 01:00:41.490 align:middle line:90% mayonnaise. 01:00:41.490 --> 01:00:44.360 align:middle line:84% And the question is, do you guys, when you move here-- 01:00:44.360 --> 01:00:45.230 align:middle line:90% I don't know. 01:00:45.230 --> 01:00:47.660 align:middle line:84% I guess not that many of you are from the Midwest either. 01:00:47.660 --> 01:00:50.570 align:middle line:84% But when you guys moved here, did you 01:00:50.570 --> 01:00:52.010 align:middle line:84% start buying Hellmann's mayonnaise 01:00:52.010 --> 01:00:53.390 align:middle line:84% and give up your Kraft mayonnaise? 01:00:53.390 --> 01:00:55.598 align:middle line:84% Or do you continue to buy the Kraft mayonnaise, which 01:00:55.598 --> 01:00:57.980 align:middle line:84% is totally available in Star Market and Whole Foods 01:00:57.980 --> 01:00:58.620 align:middle line:90% and whatever. 01:00:58.620 --> 01:01:02.240 align:middle line:84% Well, I'm not sure if it's in Whole Foods. 01:01:02.240 --> 01:01:04.222 align:middle line:90% But anyway, that's their idea. 01:01:04.222 --> 01:01:06.180 align:middle line:84% And you might think, well, that's a great idea. 01:01:06.180 --> 01:01:08.310 align:middle line:90% How would I ever find this out? 01:01:08.310 --> 01:01:12.287 align:middle line:84% And this is where, again, you get the recognition 01:01:12.287 --> 01:01:14.120 align:middle line:84% of the importance in our profession of being 01:01:14.120 --> 01:01:17.180 align:middle line:84% very well-funded and well-connected sometimes. 01:01:17.180 --> 01:01:23.210 align:middle line:84% So what-- University of Chicago has a long-standing relationship 01:01:23.210 --> 01:01:24.630 align:middle line:90% with Nielsen's. 01:01:24.630 --> 01:01:27.470 align:middle line:90% 01:01:27.470 --> 01:01:29.300 align:middle line:84% Firms like Nielsen, besides doing 01:01:29.300 --> 01:01:30.710 align:middle line:84% things for people they have relationships with, 01:01:30.710 --> 01:01:32.127 align:middle line:84% also do things for people who have 01:01:32.127 --> 01:01:33.540 align:middle line:90% money to pay them to do things. 01:01:33.540 --> 01:01:36.440 align:middle line:84% And so what Bronnenberg, Dubé, and Gentzkow 01:01:36.440 --> 01:01:39.110 align:middle line:90% got Nielsen to do is-- 01:01:39.110 --> 01:01:41.300 align:middle line:84% have I talked about Nielsen Homescan Panel? 01:01:41.300 --> 01:01:42.170 align:middle line:90% OK. 01:01:42.170 --> 01:01:43.970 align:middle line:84% Nielsen Homescan Panel-- if you want it-- 01:01:43.970 --> 01:01:46.910 align:middle line:84% it's actually-- a lot of this data is available for free 01:01:46.910 --> 01:01:50.870 align:middle line:84% through the University of Chicago website. 01:01:50.870 --> 01:01:52.460 align:middle line:84% But anyway, Nielsen Homescan Panel 01:01:52.460 --> 01:01:55.040 align:middle line:84% is they've got a panel of thousands of people 01:01:55.040 --> 01:01:56.465 align:middle line:90% across the United States. 01:01:56.465 --> 01:01:58.340 align:middle line:84% And what these thousands of people do is they 01:01:58.340 --> 01:02:00.200 align:middle line:90% all have a part-time job. 01:02:00.200 --> 01:02:04.130 align:middle line:84% Their part-time job is recording everything that they buy, 01:02:04.130 --> 01:02:06.470 align:middle line:84% and they're employed by Nielsen to do that. 01:02:06.470 --> 01:02:09.582 align:middle line:84% And so they're given like a supermarket scanner. 01:02:09.582 --> 01:02:10.790 align:middle line:90% So you go to the supermarket. 01:02:10.790 --> 01:02:11.418 align:middle line:90% You buy stuff. 01:02:11.418 --> 01:02:13.460 align:middle line:84% You scan it all in the scanner when you're there. 01:02:13.460 --> 01:02:16.043 align:middle line:84% And then you bring it home, and as you take it out of the bag, 01:02:16.043 --> 01:02:17.900 align:middle line:90% you scan it again. 01:02:17.900 --> 01:02:20.330 align:middle line:84% And you just scan it in so that Nielsen knows everything 01:02:20.330 --> 01:02:21.230 align:middle line:90% that you bought. 01:02:21.230 --> 01:02:23.490 align:middle line:84% And you're supposed to do this everywhere you go. 01:02:23.490 --> 01:02:25.940 align:middle line:84% You go to Target, you come back, you scan it. 01:02:25.940 --> 01:02:26.887 align:middle line:90% You go to the CVS. 01:02:26.887 --> 01:02:27.470 align:middle line:90% You come back. 01:02:27.470 --> 01:02:28.010 align:middle line:90% You scan it. 01:02:28.010 --> 01:02:28.970 align:middle line:90% You go to the supermarket. 01:02:28.970 --> 01:02:29.553 align:middle line:90% You come back. 01:02:29.553 --> 01:02:30.150 align:middle line:90% You scan it. 01:02:30.150 --> 01:02:33.440 align:middle line:84% You basically just are home scans. 01:02:33.440 --> 01:02:35.360 align:middle line:84% What these people do is they buy stuff, 01:02:35.360 --> 01:02:37.470 align:middle line:84% and then they just record everything they got. 01:02:37.470 --> 01:02:40.290 align:middle line:84% And then they're also recording where they got it. 01:02:40.290 --> 01:02:44.150 align:middle line:84% And so because Nielsen is already in the Star Market, 01:02:44.150 --> 01:02:47.360 align:middle line:84% and Nielsen is already in the Whole Foods, you bring it home. 01:02:47.360 --> 01:02:48.020 align:middle line:90% You scan it. 01:02:48.020 --> 01:02:49.610 align:middle line:84% They know what you paid for it because they 01:02:49.610 --> 01:02:50.870 align:middle line:84% know where you bought it, and they know what 01:02:50.870 --> 01:02:52.470 align:middle line:90% that store is charging for it. 01:02:52.470 --> 01:02:56.570 align:middle line:84% So you get this sort of data set of what 01:02:56.570 --> 01:03:01.250 align:middle line:84% is it that people at the individual level are purchasing. 01:03:01.250 --> 01:03:05.630 align:middle line:84% And you can also get aggregate market shares out of this. 01:03:05.630 --> 01:03:08.980 align:middle line:84% Nielsen Homescan Panel did not tell them-- 01:03:08.980 --> 01:03:10.730 align:middle line:84% it tells you where the people are located, 01:03:10.730 --> 01:03:12.000 align:middle line:90% that this person is in Boston. 01:03:12.000 --> 01:03:16.160 align:middle line:84% It doesn't tell you that they grew up in Illinois 01:03:16.160 --> 01:03:19.670 align:middle line:84% and moved to Boston when they went to college at age of 19. 01:03:19.670 --> 01:03:22.220 align:middle line:84% So what they got Nielsen to do is 01:03:22.220 --> 01:03:25.070 align:middle line:84% survey the members of the Homescan Panel and ask them. 01:03:25.070 --> 01:03:28.710 align:middle line:90% So what state were you born in? 01:03:28.710 --> 01:03:31.910 align:middle line:84% When did you leave your state of birth? 01:03:31.910 --> 01:03:38.300 align:middle line:84% When do you-- how long have you been in your current location? 01:03:38.300 --> 01:03:40.910 align:middle line:84% And ask you other questions like, what's your gender, 01:03:40.910 --> 01:03:44.923 align:middle line:84% or what's your highest level of education completed, and so on. 01:03:44.923 --> 01:03:46.340 align:middle line:84% Normally, you would think you send 01:03:46.340 --> 01:03:47.900 align:middle line:84% a survey like that to people, it's very hard 01:03:47.900 --> 01:03:48.930 align:middle line:90% to get them to respond. 01:03:48.930 --> 01:03:51.300 align:middle line:84% But since this survey was sponsored 01:03:51.300 --> 01:03:54.870 align:middle line:84% by Homescan, who is an employer of all the Homescan people, 01:03:54.870 --> 01:03:58.800 align:middle line:84% they got a 65% response rate on their survey 01:03:58.800 --> 01:03:59.993 align:middle line:90% asking this information. 01:03:59.993 --> 01:04:02.160 align:middle line:84% They're able to merge that information together with 01:04:02.160 --> 01:04:03.990 align:middle line:84% the people's purchasing patterns, 01:04:03.990 --> 01:04:08.400 align:middle line:84% and they get this remarkable data set of 80,000 individuals 01:04:08.400 --> 01:04:11.100 align:middle line:90% living in 49,000 households. 01:04:11.100 --> 01:04:13.770 align:middle line:84% And they know their life history of moving, 01:04:13.770 --> 01:04:17.172 align:middle line:84% and then they can examine what those people purchased. 01:04:17.172 --> 01:04:20.720 align:middle line:84% And I think it's an interesting contrast with Bresnahan 01:04:20.720 --> 01:04:22.255 align:middle line:90% and Reiss, where we have like-- 01:04:22.255 --> 01:04:23.630 align:middle line:84% we're opening up the phone books, 01:04:23.630 --> 01:04:25.963 align:middle line:84% and we're counting there are three dentists or whatever, 01:04:25.963 --> 01:04:29.690 align:middle line:84% and that's for 200 different cities versus this 80,000 01:04:29.690 --> 01:04:32.900 align:middle line:84% people, you've got basically their shopping lives laid out 01:04:32.900 --> 01:04:35.335 align:middle line:84% in front of you, as well as some of their life history. 01:04:35.335 --> 01:04:40.760 align:middle line:90% 01:04:40.760 --> 01:04:47.150 align:middle line:84% Nielsen classifies UPC codes by product category and brand. 01:04:47.150 --> 01:04:52.200 align:middle line:84% In this paper, they've picked 238 different modules. 01:04:52.200 --> 01:04:54.470 align:middle line:90% So a module could be mayonnaise. 01:04:54.470 --> 01:04:56.980 align:middle line:90% So they pick 238 modules. 01:04:56.980 --> 01:05:00.650 align:middle line:84% They pick modules where they have at least 5,000 households 01:05:00.650 --> 01:05:05.030 align:middle line:84% purchasing that item and where the top two brands are owned 01:05:05.030 --> 01:05:06.600 align:middle line:90% by different firms. 01:05:06.600 --> 01:05:09.560 align:middle line:84% So, for instance, in the cream cheese module, 01:05:09.560 --> 01:05:12.420 align:middle line:84% the number one cream cheese is Philadelphia cream cheese. 01:05:12.420 --> 01:05:15.210 align:middle line:84% The number two cream cheese is Philadelphia Light cream cheese. 01:05:15.210 --> 01:05:17.080 align:middle line:84% So they would throw out cream cheese. 01:05:17.080 --> 01:05:18.930 align:middle line:84% But in mayonnaise, number one mayonnaise 01:05:18.930 --> 01:05:20.640 align:middle line:90% is Unilever mayonnaise. 01:05:20.640 --> 01:05:22.570 align:middle line:84% Number two mayonnaise is Kraft mayonnaise. 01:05:22.570 --> 01:05:26.300 align:middle line:90% They keep mayonnaise. 01:05:26.300 --> 01:05:28.010 align:middle line:84% Anyway, the current market share data 01:05:28.010 --> 01:05:31.160 align:middle line:84% is also supplemented with historical data 01:05:31.160 --> 01:05:33.740 align:middle line:84% at the state level that's obtained from some published 01:05:33.740 --> 01:05:36.860 align:middle line:84% surveys that asked people about, what brand of mayonnaise 01:05:36.860 --> 01:05:37.830 align:middle line:90% do you normally buy? 01:05:37.830 --> 01:05:41.210 align:middle line:84% And they found these archival things from the '40s up 01:05:41.210 --> 01:05:42.920 align:middle line:90% to the '60s. 01:05:42.920 --> 01:05:45.230 align:middle line:84% I won't talk so much about that part, though. 01:05:45.230 --> 01:05:47.315 align:middle line:84% Actually, they also do advertising data. 01:05:47.315 --> 01:05:49.190 align:middle line:84% I'm not going to talk so much about what they 01:05:49.190 --> 01:05:51.710 align:middle line:90% do with the advertising data. 01:05:51.710 --> 01:05:54.420 align:middle line:84% But anyway, so here is part of their product list. 01:05:54.420 --> 01:05:57.500 align:middle line:84% And, again, it's 238 things long. 01:05:57.500 --> 01:06:01.603 align:middle line:90% But so all of these are-- 01:06:01.603 --> 01:06:02.270 align:middle line:90% well, let's see. 01:06:02.270 --> 01:06:03.895 align:middle line:84% Down here at the bottom, I've got beer. 01:06:03.895 --> 01:06:05.420 align:middle line:90% I've talked about beer before. 01:06:05.420 --> 01:06:09.740 align:middle line:84% So the number one beer is Budweiser. 01:06:09.740 --> 01:06:13.100 align:middle line:90% The number two beer is Miller. 01:06:13.100 --> 01:06:16.790 align:middle line:84% What they do is they record the aggregate purchase share. 01:06:16.790 --> 01:06:21.200 align:middle line:84% So aggregate share here means share of product one. 01:06:21.200 --> 01:06:24.540 align:middle line:84% Aggregate purchase share is share of Budweiser 01:06:24.540 --> 01:06:27.280 align:middle line:84% and the fraction people buying Budweiser or Miller. 01:06:27.280 --> 01:06:31.830 align:middle line:84% So 64% of the people who buy Budweiser or Miller buy Miller. 01:06:31.830 --> 01:06:35.860 align:middle line:84% And then, again, these are the cross-state standard deviation. 01:06:35.860 --> 01:06:37.163 align:middle line:90% So it's 64 on average. 01:06:37.163 --> 01:06:38.830 align:middle line:84% But in different states, it's different. 01:06:38.830 --> 01:06:43.470 align:middle line:84% So it's 64 plus or minus 19 for Budweiser versus Miller. 01:06:43.470 --> 01:06:48.180 align:middle line:84% Some of the products, like of bread, 01:06:48.180 --> 01:06:50.950 align:middle line:84% Nature's Own versus Sara Lee Soft and Smooth, 01:06:50.950 --> 01:06:54.276 align:middle line:84% it's like 50/50 between those two plus or minus 32. 01:06:54.276 --> 01:06:59.160 align:middle line:90% 01:06:59.160 --> 01:07:01.480 align:middle line:84% Some of them there's very little variation. 01:07:01.480 --> 01:07:03.900 align:middle line:84% So antacids-- Prilosec and Rolaids 01:07:03.900 --> 01:07:06.030 align:middle line:90% are the top two antacids. 01:07:06.030 --> 01:07:10.050 align:middle line:84% Prilosec is a 71% share in that comparison. 01:07:10.050 --> 01:07:11.790 align:middle line:84% It's a more effective drug, I guess. 01:07:11.790 --> 01:07:14.040 align:middle line:90% And it's 71 plus or minus 8. 01:07:14.040 --> 01:07:19.370 align:middle line:90% 01:07:19.370 --> 01:07:20.770 align:middle line:84% I look at some of these, and they 01:07:20.770 --> 01:07:23.530 align:middle line:84% seem like things that are very similar products. 01:07:23.530 --> 01:07:27.220 align:middle line:84% Like, I think I would be shocked if Comet abrasive cleaning 01:07:27.220 --> 01:07:29.440 align:middle line:84% powder and Ajax abrasive cleaning powder 01:07:29.440 --> 01:07:30.500 align:middle line:90% are very different. 01:07:30.500 --> 01:07:32.150 align:middle line:84% Some of them, like the bakery products, 01:07:32.150 --> 01:07:34.110 align:middle line:84% maybe it seems like they're not as homogeneous 01:07:34.110 --> 01:07:35.760 align:middle line:90% as some of the other things. 01:07:35.760 --> 01:07:40.820 align:middle line:90% 01:07:40.820 --> 01:07:43.830 align:middle line:84% So what does shopper's homescan-- 01:07:43.830 --> 01:07:46.590 align:middle line:84% or these homescan shoppers in the United States look like? 01:07:46.590 --> 01:07:49.590 align:middle line:84% I don't know how this compares to the US population as a whole. 01:07:49.590 --> 01:07:51.560 align:middle line:90% Someone must know that. 01:07:51.560 --> 01:07:58.460 align:middle line:84% About 27% of the households in their sample are migrants. 01:07:58.460 --> 01:08:00.808 align:middle line:84% I guess, obviously, almost everybody 01:08:00.808 --> 01:08:02.600 align:middle line:84% I know is a migrant, which I guess tells me 01:08:02.600 --> 01:08:03.920 align:middle line:90% that migrants are unusual-- 01:08:03.920 --> 01:08:07.850 align:middle line:84% that we're somewhat unusual relative to the US population. 01:08:07.850 --> 01:08:12.260 align:middle line:84% But 27% of the households were born in a different state 01:08:12.260 --> 01:08:13.910 align:middle line:90% where they now reside. 01:08:13.910 --> 01:08:17.270 align:middle line:84% 16% were born in a different census region-- so conditional 01:08:17.270 --> 01:08:19.010 align:middle line:90% on having moved states. 01:08:19.010 --> 01:08:23.000 align:middle line:84% Many people have moved fairly far. 01:08:23.000 --> 01:08:27.703 align:middle line:84% Obviously, this is-- you'll notice people don't really 01:08:27.703 --> 01:08:28.620 align:middle line:90% move to the Northeast. 01:08:28.620 --> 01:08:29.995 align:middle line:84% People are born in the Northeast, 01:08:29.995 --> 01:08:31.529 align:middle line:84% and then they stay in the Northeast. 01:08:31.529 --> 01:08:33.290 align:middle line:84% Very few people, especially from the West, 01:08:33.290 --> 01:08:36.740 align:middle line:84% move to the Northeast, given the weather. 01:08:36.740 --> 01:08:42.090 align:middle line:84% More people in the South move to the South from the Northeast 01:08:42.090 --> 01:08:43.020 align:middle line:90% or from the Midwest. 01:08:43.020 --> 01:08:45.870 align:middle line:90% 01:08:45.870 --> 01:08:50.700 align:middle line:84% Ages-- the people in their sample tend to be in their 40s, 01:08:50.700 --> 01:08:52.590 align:middle line:90% 50s, 60s. 01:08:52.590 --> 01:08:56.490 align:middle line:84% You get very few people in their 20s who want to have a scanner 01:08:56.490 --> 01:09:00.600 align:middle line:84% and scan everything they buy at the grocery store. 01:09:00.600 --> 01:09:02.100 align:middle line:84% Age leaving state of birth-- there's 01:09:02.100 --> 01:09:03.675 align:middle line:84% an awful lot of variation in that. 01:09:03.675 --> 01:09:06.300 align:middle line:84% You get a lot of people who left before they were 10 years old, 01:09:06.300 --> 01:09:10.350 align:middle line:84% a lot of people who move in their 20s, not surprisingly. 01:09:10.350 --> 01:09:14.760 align:middle line:84% And then you do get people who also move in their 60s when 01:09:14.760 --> 01:09:16.960 align:middle line:90% they're retiring. 01:09:16.960 --> 01:09:19.279 align:middle line:90% So what's the main analysis? 01:09:19.279 --> 01:09:22.229 align:middle line:84% The main thing is to ask, let's look at these people 01:09:22.229 --> 01:09:26.399 align:middle line:84% and ask when they move, how far do they move from their birth 01:09:26.399 --> 01:09:31.170 align:middle line:84% place purchasing patterns to their destination purchasing 01:09:31.170 --> 01:09:32.189 align:middle line:90% patterns? 01:09:32.189 --> 01:09:36.420 align:middle line:84% And so, they write mu sj for the share 01:09:36.420 --> 01:09:40.380 align:middle line:84% of people who never move-- the share of nonmigrants in state 01:09:40.380 --> 01:09:42.810 align:middle line:90% s who buy product j. 01:09:42.810 --> 01:09:46.020 align:middle line:84% And then mu s prime j is the people in state 01:09:46.020 --> 01:09:47.930 align:middle line:90% s prime who never move-- 01:09:47.930 --> 01:09:50.470 align:middle line:84% what fraction of them buy product j? 01:09:50.470 --> 01:09:52.470 align:middle line:84% And so what I'm doing is I'm thinking of someone 01:09:52.470 --> 01:09:56.940 align:middle line:84% who moves from state s to state s prime and saying, 01:09:56.940 --> 01:10:00.810 align:middle line:84% how does their purchasing compare to-- 01:10:00.810 --> 01:10:05.710 align:middle line:84% how far in the direction of s prime has their purchasing-- 01:10:05.710 --> 01:10:09.090 align:middle line:84% how far along the way from mu s to mu s prime has 01:10:09.090 --> 01:10:13.710 align:middle line:84% their purchasing moved, assuming they started at mu sj when they 01:10:13.710 --> 01:10:14.730 align:middle line:90% were in-- 01:10:14.730 --> 01:10:18.450 align:middle line:84% before they moved 20 years ago when I don't observe them? 01:10:18.450 --> 01:10:20.970 align:middle line:90% So the question is, if this-- 01:10:20.970 --> 01:10:23.370 align:middle line:90% if you still purchase-- 01:10:23.370 --> 01:10:29.110 align:middle line:84% if these migrants still purchase exactly what they purchased-- 01:10:29.110 --> 01:10:31.257 align:middle line:84% well, if they were representative of the population 01:10:31.257 --> 01:10:33.840 align:middle line:84% in state s, and still purchase what they purchased before they 01:10:33.840 --> 01:10:36.810 align:middle line:90% left, beta would be 0. 01:10:36.810 --> 01:10:40.650 align:middle line:84% If they all purchase are exactly representative 01:10:40.650 --> 01:10:43.300 align:middle line:84% of the nonmigrants in their new home and purchase 01:10:43.300 --> 01:10:47.070 align:middle line:84% what people in the new home purchase, beta would be 1. 01:10:47.070 --> 01:10:49.620 align:middle line:84% So beta is like, what fraction of the way 01:10:49.620 --> 01:10:53.760 align:middle line:84% have people moved from looking like where they were born people 01:10:53.760 --> 01:10:56.670 align:middle line:90% to where they now live people? 01:10:56.670 --> 01:11:00.510 align:middle line:84% And we're going to basically have a regression estimating 01:11:00.510 --> 01:11:08.940 align:middle line:84% beta ij as a function of the age that you were when you moved 01:11:08.940 --> 01:11:11.912 align:middle line:84% and the time that you've been in your new location. 01:11:11.912 --> 01:11:14.370 align:middle line:84% So you might think if you moved when you were one year old, 01:11:14.370 --> 01:11:16.420 align:middle line:84% you've been your location for 60 years, 01:11:16.420 --> 01:11:19.178 align:middle line:84% you ought to look like people who grew up in that location 01:11:19.178 --> 01:11:20.970 align:middle line:84% where you moved when you were one year old, 01:11:20.970 --> 01:11:22.637 align:middle line:84% other than maybe your parents taught you 01:11:22.637 --> 01:11:24.630 align:middle line:90% about what mayonnaise to buy. 01:11:24.630 --> 01:11:28.680 align:middle line:84% Whereas, if you're 60 years old, and you moved at age-- 01:11:28.680 --> 01:11:31.230 align:middle line:84% if you moved at age 60, and you've been there 01:11:31.230 --> 01:11:33.950 align:middle line:84% for six months, you might expect that those people would 01:11:33.950 --> 01:11:35.700 align:middle line:84% look more like the origin people, at least 01:11:35.700 --> 01:11:37.188 align:middle line:90% on the taste side. 01:11:37.188 --> 01:11:40.370 align:middle line:90% 01:11:40.370 --> 01:11:42.620 align:middle line:84% And then they do this via weighted least squares. 01:11:42.620 --> 01:11:44.120 align:middle line:84% Because if you think about it like-- 01:11:44.120 --> 01:11:47.150 align:middle line:84% if mu s and mu s prime are almost identical-- we're 01:11:47.150 --> 01:11:49.230 align:middle line:84% getting almost no information from that market, 01:11:49.230 --> 01:11:51.047 align:middle line:84% and it's where mu s and mu s prime 01:11:51.047 --> 01:11:53.130 align:middle line:84% are very different that we're getting information, 01:11:53.130 --> 01:11:55.220 align:middle line:84% and so they do a weighted least squares estimate 01:11:55.220 --> 01:11:59.090 align:middle line:84% weighting the behavior of people who made moves 01:11:59.090 --> 01:12:01.957 align:middle line:90% to more different places. 01:12:01.957 --> 01:12:04.040 align:middle line:84% Anyway, so they've got a few figures that they do. 01:12:04.040 --> 01:12:05.340 align:middle line:90% They've got this one. 01:12:05.340 --> 01:12:09.440 align:middle line:84% Sometimes they're estimating this coefficient separately. 01:12:09.440 --> 01:12:11.030 align:middle line:90% Again, you've got 80,000 people. 01:12:11.030 --> 01:12:13.880 align:middle line:84% You've got a lot of people who move at age 15 01:12:13.880 --> 01:12:16.050 align:middle line:84% and have been there for 35 years. 01:12:16.050 --> 01:12:18.260 align:middle line:84% So they have separate coefficients 01:12:18.260 --> 01:12:25.850 align:middle line:84% for every age at move and every year since move. 01:12:25.850 --> 01:12:28.710 align:middle line:84% I find this one hard to see, though, in three dimensions. 01:12:28.710 --> 01:12:32.040 align:middle line:84% It's hard to see how high these things are versus the slope. 01:12:32.040 --> 01:12:35.190 align:middle line:84% So easier to read is these ones here. 01:12:35.190 --> 01:12:39.030 align:middle line:84% They just go-- they collapse it down to one dimension. 01:12:39.030 --> 01:12:43.320 align:middle line:84% And what this is saying is that an awful lot of it 01:12:43.320 --> 01:12:45.660 align:middle line:90% seems to be supply side. 01:12:45.660 --> 01:12:49.140 align:middle line:84% Because if you look at people who have just moved-- 01:12:49.140 --> 01:12:53.040 align:middle line:84% so it's zero to four years since their move, 01:12:53.040 --> 01:12:57.600 align:middle line:84% they're immediately like 60-something percent of the way 01:12:57.600 --> 01:13:00.570 align:middle line:90% to the new purchasing patterns. 01:13:00.570 --> 01:13:03.300 align:middle line:84% When you look at people who moved 40 or 50 years ago, 01:13:03.300 --> 01:13:06.960 align:middle line:84% they are like 80% or 90% of the way to the purchasing patterns. 01:13:06.960 --> 01:13:10.470 align:middle line:84% So it seems like there's a big supply side element that 01:13:10.470 --> 01:13:13.320 align:middle line:84% makes you immediately on moving here to Boston, 01:13:13.320 --> 01:13:15.660 align:middle line:84% you start buying Hellman's mayonnaise. 01:13:15.660 --> 01:13:18.340 align:middle line:84% But there is something that's longer term. 01:13:18.340 --> 01:13:21.390 align:middle line:84% And so this could be that advertising playing 01:13:21.390 --> 01:13:25.900 align:middle line:84% on you over 40 years, or this could just be your tastes adapt. 01:13:25.900 --> 01:13:26.942 align:middle line:90% You go to friends houses. 01:13:26.942 --> 01:13:28.317 align:middle line:84% They serve Hellmann's mayonnaise. 01:13:28.317 --> 01:13:30.010 align:middle line:84% You think, wow, that Hellmann's is good. 01:13:30.010 --> 01:13:31.920 align:middle line:90% I should buy that myself. 01:13:31.920 --> 01:13:35.550 align:middle line:84% So there's adaptation over time, but a lot of it 01:13:35.550 --> 01:13:39.360 align:middle line:90% happens on day one. 01:13:39.360 --> 01:13:41.850 align:middle line:84% And then here's the age at move graph. 01:13:41.850 --> 01:13:44.607 align:middle line:84% The people who moved when they were children-- obviously, 01:13:44.607 --> 01:13:46.440 align:middle line:84% the people who moved when they were children 01:13:46.440 --> 01:13:50.357 align:middle line:84% have already had to have lived for quite 01:13:50.357 --> 01:13:52.440 align:middle line:84% a while in the new location because they were only 01:13:52.440 --> 01:13:53.620 align:middle line:90% surveying adults. 01:13:53.620 --> 01:13:56.430 align:middle line:84% But the people who moved at children-- 01:13:56.430 --> 01:13:57.570 align:middle line:90% you moved at age one. 01:13:57.570 --> 01:13:59.960 align:middle line:84% You do retain something of your birth mayonnaise. 01:13:59.960 --> 01:14:02.830 align:middle line:84% So this is you're getting something about mayonnaise 01:14:02.830 --> 01:14:04.120 align:middle line:90% from your parents. 01:14:04.120 --> 01:14:07.600 align:middle line:84% Because it can't be that you moved at age one, therefore, 01:14:07.600 --> 01:14:13.000 align:middle line:84% you remember the mayonnaise of your babyhood. 01:14:13.000 --> 01:14:14.750 align:middle line:84% So there is something that you're getting. 01:14:14.750 --> 01:14:19.870 align:middle line:84% It's only 0.9, but then it declines in the age it moves. 01:14:19.870 --> 01:14:24.970 align:middle line:84% But it stays that even people who moved in their 60s, 01:14:24.970 --> 01:14:27.640 align:middle line:84% they're mostly adjusting to the new mayonnaise. 01:14:27.640 --> 01:14:33.920 align:middle line:90% 01:14:33.920 --> 01:14:34.420 align:middle line:90% OK. 01:14:34.420 --> 01:14:35.590 align:middle line:90% So what do I have here? 01:14:35.590 --> 01:14:37.370 align:middle line:90% What does it say? 01:14:37.370 --> 01:14:40.090 align:middle line:84% So anyway, people instantly move 60% of the way. 01:14:40.090 --> 01:14:45.880 align:middle line:84% This suggests that it is a lot of supply side stuff. 01:14:45.880 --> 01:14:47.170 align:middle line:90% Yeah. 01:14:47.170 --> 01:14:49.040 align:middle line:84% People become more similar over time. 01:14:49.040 --> 01:14:50.290 align:middle line:90% I said that. 01:14:50.290 --> 01:14:52.590 align:middle line:84% Betas are related to age and move. 01:14:52.590 --> 01:14:56.440 align:middle line:84% So age 60-plus, you're still getting a big movement toward 01:14:56.440 --> 01:14:58.985 align:middle line:90% new things. 01:14:58.985 --> 01:15:01.310 align:middle line:84% Oh, number 4-- data on premove migrants. 01:15:01.310 --> 01:15:03.250 align:middle line:84% Again, there's this thing of you guys 01:15:03.250 --> 01:15:08.800 align:middle line:84% moved to Boston because you're not the typical people where 01:15:08.800 --> 01:15:09.820 align:middle line:90% you grew up. 01:15:09.820 --> 01:15:11.680 align:middle line:84% You probably don't drink the same coffee 01:15:11.680 --> 01:15:15.710 align:middle line:84% that people where you grew up drink. 01:15:15.710 --> 01:15:17.380 align:middle line:84% You probably don't drink the same beer 01:15:17.380 --> 01:15:19.420 align:middle line:84% that people where you grew up drink. 01:15:19.420 --> 01:15:21.962 align:middle line:84% You probably don't-- maybe you don't have the same mayonnaise 01:15:21.962 --> 01:15:23.650 align:middle line:90% like people you grew up. 01:15:23.650 --> 01:15:25.960 align:middle line:84% So the question is, is this in part-- 01:15:25.960 --> 01:15:30.455 align:middle line:84% is this like 60% figure because the people-- 01:15:30.455 --> 01:15:32.080 align:middle line:84% remember, most of the analysis are only 01:15:32.080 --> 01:15:33.730 align:middle line:84% observing where people are today. 01:15:33.730 --> 01:15:36.010 align:middle line:84% So the 60% figure is that people living in Boston, 01:15:36.010 --> 01:15:38.422 align:middle line:90% maybe it's that like-- 01:15:38.422 --> 01:15:39.630 align:middle line:90% it's like, you are different. 01:15:39.630 --> 01:15:41.330 align:middle line:84% Therefore, you get your coffee at Starbucks. 01:15:41.330 --> 01:15:43.040 align:middle line:84% Whereas, people you grew up with didn't 01:15:43.040 --> 01:15:44.780 align:middle line:84% go to Starbucks or something like that. 01:15:44.780 --> 01:15:47.690 align:middle line:84% And you moved to Boston because you were like the Starbucks type 01:15:47.690 --> 01:15:49.820 align:middle line:90% person or something. 01:15:49.820 --> 01:15:51.050 align:middle line:90% They have evidence, though-- 01:15:51.050 --> 01:15:54.620 align:middle line:84% they actually look at the month-by-month data. 01:15:54.620 --> 01:15:57.140 align:middle line:84% They don't-- can't track you precisely, but they're ask-- 01:15:57.140 --> 01:16:02.510 align:middle line:84% they did their survey in October of 2007. 01:16:02.510 --> 01:16:05.120 align:middle line:90% And then-- no, sorry. 01:16:05.120 --> 01:16:07.640 align:middle line:84% They did their survey in September of 2008. 01:16:07.640 --> 01:16:08.605 align:middle line:90% And they asked people-- 01:16:08.605 --> 01:16:09.980 align:middle line:84% like, they asked people, how long 01:16:09.980 --> 01:16:11.150 align:middle line:84% have you been in your current state? 01:16:11.150 --> 01:16:11.850 align:middle line:90% And what are the answers? 01:16:11.850 --> 01:16:12.892 align:middle line:90% The answers are integers. 01:16:12.892 --> 01:16:16.810 align:middle line:84% So it's less than 1, 1 to 2, 2 to 3, whatever. 01:16:16.810 --> 01:16:18.770 align:middle line:84% So all the people who said less than-- 01:16:18.770 --> 01:16:21.440 align:middle line:84% I'm living in some state in September 2008. 01:16:21.440 --> 01:16:23.240 align:middle line:90% I've been here less than a year. 01:16:23.240 --> 01:16:27.440 align:middle line:84% They know that they moved sometime between October of 2007 01:16:27.440 --> 01:16:29.010 align:middle line:90% and September of 2008. 01:16:29.010 --> 01:16:31.100 align:middle line:84% They didn't ask for the month of the move. 01:16:31.100 --> 01:16:32.570 align:middle line:84% But what they can then do is say, 01:16:32.570 --> 01:16:35.960 align:middle line:84% look at these people who moved here within the past year, 01:16:35.960 --> 01:16:39.920 align:middle line:84% and look at what mayonnaise did they eat before they moved. 01:16:39.920 --> 01:16:42.560 align:middle line:84% And what they find is that before they 01:16:42.560 --> 01:16:47.660 align:middle line:84% moved over this year, it's looking to me 01:16:47.660 --> 01:16:50.240 align:middle line:84% like the people were only like 10% of the way 01:16:50.240 --> 01:16:51.660 align:middle line:90% to their new destination. 01:16:51.660 --> 01:16:53.330 align:middle line:90% So maybe they were more worldly. 01:16:53.330 --> 01:16:56.570 align:middle line:84% Maybe they already had more Boston kind of tastes. 01:16:56.570 --> 01:17:00.380 align:middle line:84% But they're only maybe 10% Boston taste people. 01:17:00.380 --> 01:17:03.290 align:middle line:84% And then you look at them as they move through the year. 01:17:03.290 --> 01:17:05.120 align:middle line:90% And by the time they moved-- 01:17:05.120 --> 01:17:08.680 align:middle line:84% like that 60%, 1 to 4 years isn't 60% 0 to 4 years. 01:17:08.680 --> 01:17:13.442 align:middle line:84% That's 60% the within the first 12 months after you get there. 01:17:13.442 --> 01:17:15.650 align:middle line:84% So some of these people have only been here one month 01:17:15.650 --> 01:17:17.630 align:middle line:84% and are already buying the new mayonnaise. 01:17:17.630 --> 01:17:19.650 align:middle line:84% Some of them have been here six or nine months. 01:17:19.650 --> 01:17:23.870 align:middle line:84% But it seems that this 65% thing looks 01:17:23.870 --> 01:17:25.860 align:middle line:90% like a causal effect of moving. 01:17:25.860 --> 01:17:26.930 align:middle line:90% And it looks like-- 01:17:26.930 --> 01:17:29.790 align:middle line:84% or maybe 80% of it's a causal effect of moving. 01:17:29.790 --> 01:17:31.550 align:middle line:84% Some part of it is like a preexisting, 01:17:31.550 --> 01:17:33.330 align:middle line:90% underlying heterogeneity. 01:17:33.330 --> 01:17:35.393 align:middle line:84% But then there's this big, big one. 01:17:35.393 --> 01:17:41.320 align:middle line:90% 01:17:41.320 --> 01:17:43.300 align:middle line:90% And, again, it's just-- 01:17:43.300 --> 01:17:46.360 align:middle line:84% they're just-- 80,000 is a good-- 01:17:46.360 --> 01:17:48.680 align:middle line:84% 80,000 data points is a good thing. 01:17:48.680 --> 01:17:52.660 align:middle line:84% And 80,000 data points times 238 products is a good thing. 01:17:52.660 --> 01:17:56.207 align:middle line:84% And just-- this is vantage of data scale, 01:17:56.207 --> 01:17:58.790 align:middle line:84% as you can do these things that would be very, very hard to do 01:17:58.790 --> 01:18:00.772 align:middle line:90% with any small [INAUDIBLE]. 01:18:00.772 --> 01:18:06.370 align:middle line:90% 01:18:06.370 --> 01:18:09.180 align:middle line:90% Questions on that? 01:18:09.180 --> 01:18:12.110 align:middle line:90% So that is what I got today. 01:18:12.110 --> 01:18:18.100 align:middle line:84% So next week, again, is theory Monday, empirics Wednesday. 01:18:18.100 --> 01:18:21.400 align:middle line:84% Next week, the theory topic is strategic investment. 01:18:21.400 --> 01:18:24.910 align:middle line:84% Again, this is going to be a very Tirole textbook 01:18:24.910 --> 01:18:26.430 align:middle line:90% kind of lecture on Monday. 01:18:26.430 --> 01:18:29.140 align:middle line:84% If you look at chapter 8 of the Tirole's book. 01:18:29.140 --> 01:18:30.770 align:middle line:84% It's the longest chapter in the book. 01:18:30.770 --> 01:18:33.190 align:middle line:84% So I will just try to go over the basics 01:18:33.190 --> 01:18:34.320 align:middle line:90% of strategic investment. 01:18:34.320 --> 01:18:36.220 align:middle line:84% And strategic investment is about thinking 01:18:36.220 --> 01:18:38.650 align:middle line:84% of things firms do to affect entry in markets. 01:18:38.650 --> 01:18:41.860 align:middle line:84% Because if entry really is the primary determinant of markups, 01:18:41.860 --> 01:18:45.130 align:middle line:84% really should want to affect entry and keep firms out 01:18:45.130 --> 01:18:46.622 align:middle line:90% of markets [INAUDIBLE]. 01:18:46.622 --> 01:18:49.270 align:middle line:90% 01:18:49.270 --> 01:18:51.150 align:middle line:90% See you on Monday. 01:18:51.150 --> 01:19:03.000 align:middle line:90%