WEBVTT 00:00:00.000 --> 00:00:01.960 align:middle line:90% [SQUEAKING] 00:00:01.960 --> 00:00:04.410 align:middle line:90% [RUSTLING] 00:00:04.410 --> 00:00:07.350 align:middle line:90% [CLICKING] 00:00:07.350 --> 00:00:18.078 align:middle line:90% 00:00:18.078 --> 00:00:19.870 align:middle line:84% BEN OLKEN: OK, let's get started, everyone. 00:00:19.870 --> 00:00:22.150 align:middle line:84% All right, so just to remind you guys where we were, 00:00:22.150 --> 00:00:23.830 align:middle line:90% we're talking about land issues. 00:00:23.830 --> 00:00:28.750 align:middle line:84% And in the previous lecture, we talked about moral hazard 00:00:28.750 --> 00:00:31.810 align:middle line:84% issues and how do we think about moral hazard in land. 00:00:31.810 --> 00:00:34.540 align:middle line:84% And the second piece of the land puzzle 00:00:34.540 --> 00:00:36.520 align:middle line:84% that I want to talk about today is 00:00:36.520 --> 00:00:41.028 align:middle line:84% about land titling and the impacts of land titling. 00:00:41.028 --> 00:00:42.820 align:middle line:84% So that's going to be the main thing today. 00:00:42.820 --> 00:00:48.130 align:middle line:84% So there's a few other things I just wanted to finish up, 00:00:48.130 --> 00:00:53.360 align:middle line:84% on land, on the moral hazard stuff first. 00:00:53.360 --> 00:00:55.630 align:middle line:84% So I think we basically finished the-- we finished 00:00:55.630 --> 00:00:57.905 align:middle line:84% the discussion of the experimental paper. 00:00:57.905 --> 00:00:59.530 align:middle line:84% The last thing I just wanted to mention 00:00:59.530 --> 00:01:03.250 align:middle line:84% was if you think about the theoretical models 00:01:03.250 --> 00:01:07.570 align:middle line:84% that I outlined last time, we talked about both moral hazard 00:01:07.570 --> 00:01:09.880 align:middle line:84% models and limited liability models, 00:01:09.880 --> 00:01:14.020 align:middle line:84% insurance kind of motives and limited liability challenges. 00:01:14.020 --> 00:01:19.210 align:middle line:84% And you might want to distinguish between those two. 00:01:19.210 --> 00:01:22.750 align:middle line:84% And here, I don't know of a super great paper 00:01:22.750 --> 00:01:23.540 align:middle line:90% that does that. 00:01:23.540 --> 00:01:25.120 align:middle line:84% So I just wanted to mention this because it 00:01:25.120 --> 00:01:26.980 align:middle line:84% seems like a natural thing that you'd like to distinguish 00:01:26.980 --> 00:01:27.970 align:middle line:90% between those two. 00:01:27.970 --> 00:01:30.400 align:middle line:84% There's a paper by Laffont and Matoussi 00:01:30.400 --> 00:01:31.630 align:middle line:90% that thinks about this. 00:01:31.630 --> 00:01:34.240 align:middle line:84% This is, as far as I know, the original limited liability 00:01:34.240 --> 00:01:36.513 align:middle line:90% sharecropping kind of paper. 00:01:36.513 --> 00:01:38.680 align:middle line:84% When you say original, it's always a risk of someone 00:01:38.680 --> 00:01:39.760 align:middle line:84% to find something from before that. 00:01:39.760 --> 00:01:40.802 align:middle line:90% That's the one I know of. 00:01:40.802 --> 00:01:45.135 align:middle line:90% 00:01:45.135 --> 00:01:49.900 align:middle line:84% And in their paper, the way they think about this, 00:01:49.900 --> 00:01:54.440 align:middle line:84% they think of this idea of working capital. 00:01:54.440 --> 00:01:57.370 align:middle line:84% So working capital is like money that you 00:01:57.370 --> 00:02:02.290 align:middle line:84% have available to you in the short run for buying things, 00:02:02.290 --> 00:02:07.660 align:middle line:84% like paying your rent or buying your fertilizer inputs, 00:02:07.660 --> 00:02:09.800 align:middle line:90% or whatever. 00:02:09.800 --> 00:02:12.090 align:middle line:84% And their idea is that that working capital 00:02:12.090 --> 00:02:14.590 align:middle line:84% is the thing that's going to determine the limited liability 00:02:14.590 --> 00:02:17.290 align:middle line:84% constraint for how much you pay upfront. 00:02:17.290 --> 00:02:19.570 align:middle line:84% So in their model they, basically say, 00:02:19.570 --> 00:02:22.390 align:middle line:84% instead of saying l has to be greater than 0, 00:02:22.390 --> 00:02:25.180 align:middle line:84% they put in this constraint that the amount that you pay-- 00:02:25.180 --> 00:02:27.100 align:middle line:84% this l is the payment in the low state, 00:02:27.100 --> 00:02:29.100 align:middle line:84% if you remember from the model we had last time. 00:02:29.100 --> 00:02:33.833 align:middle line:84% So l greater than K, then the higher K, the more the contract 00:02:33.833 --> 00:02:35.500 align:middle line:84% looks like a rental contract rather than 00:02:35.500 --> 00:02:38.780 align:middle line:90% a sharecropping contract. 00:02:38.780 --> 00:02:41.830 align:middle line:84% So that's fine, whereas in a risk aversion model, 00:02:41.830 --> 00:02:45.370 align:middle line:84% then what should matter is your total wealth, not 00:02:45.370 --> 00:02:49.140 align:middle line:84% just the amount of liquid assets you have. 00:02:49.140 --> 00:02:52.510 align:middle line:84% So that's the test they use to argue for limited liability. 00:02:52.510 --> 00:02:56.650 align:middle line:84% So what do you think of this as a test-- 00:02:56.650 --> 00:02:57.940 align:middle line:90% good idea, bad idea? 00:02:57.940 --> 00:02:59.420 align:middle line:90% What do you think? 00:02:59.420 --> 00:03:00.860 align:middle line:90% It's plausible, not plausible? 00:03:00.860 --> 00:03:13.670 align:middle line:90% 00:03:13.670 --> 00:03:14.385 align:middle line:90% Yeah, Aaron. 00:03:14.385 --> 00:03:16.010 align:middle line:84% AUDIENCE: Wouldn't you mean that wealth 00:03:16.010 --> 00:03:18.233 align:middle line:84% is pretty highly correlated with working capital? 00:03:18.233 --> 00:03:18.900 align:middle line:90% BEN OLKEN: Yeah. 00:03:18.900 --> 00:03:20.400 align:middle line:84% Suppose you could measure them both. 00:03:20.400 --> 00:03:22.200 align:middle line:90% 00:03:22.200 --> 00:03:23.325 align:middle line:90% AUDIENCE: I mean, another-- 00:03:23.325 --> 00:03:26.370 align:middle line:90% 00:03:26.370 --> 00:03:30.250 align:middle line:84% don't we also think that risk aversion is decreasing overall, 00:03:30.250 --> 00:03:31.230 align:middle line:90% so that-- 00:03:31.230 --> 00:03:33.750 align:middle line:84% BEN OLKEN: That's-- yeah, that's the idea, exactly right. 00:03:33.750 --> 00:03:35.800 align:middle line:84% So risk aversion would be decreasing. 00:03:35.800 --> 00:03:38.100 align:middle line:84% so the idea of the test is risk aversion should be-- 00:03:38.100 --> 00:03:40.322 align:middle line:84% if is a risk aversion model, then 00:03:40.322 --> 00:03:42.280 align:middle line:84% the determinants of whatever-- of sharecropping 00:03:42.280 --> 00:03:44.850 align:middle line:84% or whatever-- should be your total wealth. 00:03:44.850 --> 00:03:48.646 align:middle line:84% Because in a risk aversion model, your-- 00:03:48.646 --> 00:03:51.417 align:middle line:90% 00:03:51.417 --> 00:03:54.000 align:middle line:84% well I'm not going to write it down because I'll get it wrong. 00:03:54.000 --> 00:03:55.800 align:middle line:84% But your utility is a function of your total consumption. 00:03:55.800 --> 00:03:57.900 align:middle line:84% Your total consumption is a function of your total wealth. 00:03:57.900 --> 00:03:58.830 align:middle line:84% And you can use your total wealth 00:03:58.830 --> 00:04:00.288 align:middle line:84% to smooth your consumption choices. 00:04:00.288 --> 00:04:03.158 align:middle line:84% So that's kind of what's driving risk aversion. 00:04:03.158 --> 00:04:05.700 align:middle line:84% That's why risk aversion should be a function of total wealth 00:04:05.700 --> 00:04:07.710 align:middle line:84% whereas the limited liability is that you 00:04:07.710 --> 00:04:10.136 align:middle line:90% have cash that you need to pay. 00:04:10.136 --> 00:04:11.570 align:middle line:90% AUDIENCE: [INAUDIBLE]. 00:04:11.570 --> 00:04:12.170 align:middle line:90% BEN OLKEN: So yes, exactly. 00:04:12.170 --> 00:04:14.090 align:middle line:84% The idea is if it's a function of total wealth, 00:04:14.090 --> 00:04:15.632 align:middle line:84% it should be the risk aversion model. 00:04:15.632 --> 00:04:17.487 align:middle line:84% If it's a function of the limited liability, 00:04:17.487 --> 00:04:18.695 align:middle line:90% they're going to argue it's-- 00:04:18.695 --> 00:04:20.120 align:middle line:84% sorry, if it's a function of working capital, 00:04:20.120 --> 00:04:22.453 align:middle line:84% they're going to argue it's the limited liability model. 00:04:22.453 --> 00:04:25.230 align:middle line:90% 00:04:25.230 --> 00:04:26.650 align:middle line:90% So what else do you think-- 00:04:26.650 --> 00:04:29.315 align:middle line:90% good idea, not good idea? 00:04:29.315 --> 00:04:42.767 align:middle line:90% 00:04:42.767 --> 00:04:45.100 align:middle line:84% If I were to put up the results, is this is a good test? 00:04:45.100 --> 00:04:45.830 align:middle line:90% Yeah. 00:04:45.830 --> 00:04:47.582 align:middle line:84% AUDIENCE: I just asked is it clear 00:04:47.582 --> 00:04:49.747 align:middle line:90% those two things are different? 00:04:49.747 --> 00:04:55.620 align:middle line:84% So if someone [INAUDIBLE] household [INAUDIBLE]?? 00:04:55.620 --> 00:04:57.150 align:middle line:84% BEN OLKEN: That's true, but again, 00:04:57.150 --> 00:04:58.500 align:middle line:84% that's a related question to Aaron's. 00:04:58.500 --> 00:04:59.970 align:middle line:84% So suppose I can run a regression. 00:04:59.970 --> 00:05:03.270 align:middle line:84% Suppose I get data, and I run sharecropping 00:05:03.270 --> 00:05:11.170 align:middle line:84% is equal to alpha K plus beta W plus epsilon, 00:05:11.170 --> 00:05:13.305 align:middle line:90% where this is working capital. 00:05:13.305 --> 00:05:17.050 align:middle line:90% 00:05:17.050 --> 00:05:19.290 align:middle line:84% And this thing is your total wealth. 00:05:19.290 --> 00:05:23.500 align:middle line:90% 00:05:23.500 --> 00:05:25.083 align:middle line:84% So to run this regression, number one, 00:05:25.083 --> 00:05:27.458 align:middle line:84% I have to be able to measure these two things separately. 00:05:27.458 --> 00:05:29.480 align:middle line:84% And number two, I have to have some variation. 00:05:29.480 --> 00:05:31.438 align:middle line:84% So you're right-- for some households, 00:05:31.438 --> 00:05:32.980 align:middle line:84% they'll be the same, but those aren't 00:05:32.980 --> 00:05:34.813 align:middle line:84% the ones we're going to identify this thing. 00:05:34.813 --> 00:05:37.248 align:middle line:84% I mean, if everyone looked like that, 00:05:37.248 --> 00:05:39.290 align:middle line:84% then you wouldn't be able to run this regression. 00:05:39.290 --> 00:05:40.720 align:middle line:84% So this is going to be identified off 00:05:40.720 --> 00:05:42.820 align:middle line:84% of some households that have different amounts of working 00:05:42.820 --> 00:05:44.320 align:middle line:84% capital and some households with different amounts of wealth. 00:05:44.320 --> 00:05:46.427 align:middle line:84% So yes, some of them will have none of anything, 00:05:46.427 --> 00:05:48.760 align:middle line:84% but then some must have some wealth and not much working 00:05:48.760 --> 00:05:49.900 align:middle line:84% capital, and some must have some working 00:05:49.900 --> 00:05:50.830 align:middle line:90% capital and not much wealth. 00:05:50.830 --> 00:05:52.100 align:middle line:84% That's what can identify this regression. 00:05:52.100 --> 00:05:53.683 align:middle line:84% So suppose you can run this regression 00:05:53.683 --> 00:05:56.715 align:middle line:84% and you get some coefficients alpha and beta. 00:05:56.715 --> 00:05:58.090 align:middle line:84% What do you think that tells you? 00:05:58.090 --> 00:06:00.490 align:middle line:84% Or is a good test or not a good test? 00:06:00.490 --> 00:06:09.270 align:middle line:90% 00:06:09.270 --> 00:06:11.400 align:middle line:84% Yeah, [INAUDIBLE] what do you think? 00:06:11.400 --> 00:06:13.830 align:middle line:84% AUDIENCE: Why do we think that what capital is 00:06:13.830 --> 00:06:16.250 align:middle line:84% this [INAUDIBLE] measure [INAUDIBLE].. 00:06:16.250 --> 00:06:17.970 align:middle line:90% BEN OLKEN: Sorry, you guys-- 00:06:17.970 --> 00:06:19.970 align:middle line:84% AUDIENCE: Sorry, why do we think working capital 00:06:19.970 --> 00:06:21.440 align:middle line:90% is the limit of-- 00:06:21.440 --> 00:06:24.380 align:middle line:84% couldn't someone be taken to court, 00:06:24.380 --> 00:06:26.367 align:middle line:90% get total wealth extracted? 00:06:26.367 --> 00:06:27.450 align:middle line:90% BEN OLKEN: Yeah, so right. 00:06:27.450 --> 00:06:30.360 align:middle line:84% So number one is it could go the other way. 00:06:30.360 --> 00:06:33.810 align:middle line:84% So could be if you have wealth, you could go to court 00:06:33.810 --> 00:06:34.650 align:middle line:90% and get the wealth. 00:06:34.650 --> 00:06:35.325 align:middle line:90% You could. 00:06:35.325 --> 00:06:43.840 align:middle line:90% 00:06:43.840 --> 00:06:44.717 align:middle line:90% So that's one reason. 00:06:44.717 --> 00:06:45.550 align:middle line:90% So I agree with you. 00:06:45.550 --> 00:06:47.710 align:middle line:84% So certainly in a well-functioning legal system, 00:06:47.710 --> 00:06:49.780 align:middle line:84% absolutely, they go after your total assets. 00:06:49.780 --> 00:06:51.905 align:middle line:84% They can make you sell illiquid assets or whatever. 00:06:51.905 --> 00:06:53.225 align:middle line:90% And so that's one thing. 00:06:53.225 --> 00:06:53.725 align:middle line:90% What else? 00:06:53.725 --> 00:07:01.790 align:middle line:90% 00:07:01.790 --> 00:07:02.613 align:middle line:90% Yeah, [INAUDIBLE]. 00:07:02.613 --> 00:07:04.280 align:middle line:84% AUDIENCE: They are naturally [INAUDIBLE] 00:07:04.280 --> 00:07:07.000 align:middle line:84% and I would think [INAUDIBLE] wealth 00:07:07.000 --> 00:07:11.520 align:middle line:84% is measured with much more error than [INAUDIBLE].. 00:07:11.520 --> 00:07:15.600 align:middle line:84% BEN OLKEN: Yeah, so actually, I'm not sure I agree with that. 00:07:15.600 --> 00:07:18.013 align:middle line:84% But you're right that there could be measurement error. 00:07:18.013 --> 00:07:19.930 align:middle line:84% They're correlated in their measurement error. 00:07:19.930 --> 00:07:21.540 align:middle line:84% So I agree that's going to be a problem for interpreting 00:07:21.540 --> 00:07:22.240 align:middle line:90% this regression. 00:07:22.240 --> 00:07:24.448 align:middle line:84% So that's all kind of related the measurement issues. 00:07:24.448 --> 00:07:26.650 align:middle line:84% But the reason I'm not sure I agree is actually, 00:07:26.650 --> 00:07:28.650 align:middle line:84% your total wealth, if most of it's in your land, 00:07:28.650 --> 00:07:29.940 align:middle line:84% that might be relatively easy to value 00:07:29.940 --> 00:07:32.190 align:middle line:84% compared to actually adding up your working capital. 00:07:32.190 --> 00:07:35.303 align:middle line:90% It may be hard to find that. 00:07:35.303 --> 00:07:37.470 align:middle line:84% So it's not obvious to me which is a better measure. 00:07:37.470 --> 00:07:39.750 align:middle line:84% AUDIENCE: Because we are computing for both, 00:07:39.750 --> 00:07:40.740 align:middle line:90% they are adding both. 00:07:40.740 --> 00:07:41.740 align:middle line:90% BEN OLKEN: Yes, I agree. 00:07:41.740 --> 00:07:43.115 align:middle line:84% So if you have measurement error, 00:07:43.115 --> 00:07:45.900 align:middle line:84% if one thing is measured poorly and they're correlated, 00:07:45.900 --> 00:07:48.220 align:middle line:84% that could be a problem in this regression. 00:07:48.220 --> 00:07:49.600 align:middle line:90% Yeah. 00:07:49.600 --> 00:07:51.790 align:middle line:84% AUDIENCE: I don't have the [INAUDIBLE] to that, 00:07:51.790 --> 00:07:54.088 align:middle line:84% but I feel like having no working capital would 00:07:54.088 --> 00:07:55.720 align:middle line:90% make [INAUDIBLE]. 00:07:55.720 --> 00:07:57.040 align:middle line:90% BEN OLKEN: Yeah, so why? 00:07:57.040 --> 00:07:58.040 align:middle line:90% Tell me more about that. 00:07:58.040 --> 00:07:59.373 align:middle line:90% AUDIENCE: Well, it's just like-- 00:07:59.373 --> 00:08:04.600 align:middle line:90% 00:08:04.600 --> 00:08:08.693 align:middle line:90% it just seems like-- 00:08:08.693 --> 00:08:10.360 align:middle line:84% I don't know, it just seems like there's 00:08:10.360 --> 00:08:12.423 align:middle line:90% this direct relationship. 00:08:12.423 --> 00:08:13.090 align:middle line:90% BEN OLKEN: Yeah. 00:08:13.090 --> 00:08:15.395 align:middle line:90% 00:08:15.395 --> 00:08:16.770 align:middle line:84% AUDIENCE: I don't know, is that-- 00:08:16.770 --> 00:08:23.790 align:middle line:90% 00:08:23.790 --> 00:08:25.540 align:middle line:84% BEN OLKEN: Yes, that's exactly-- so that's 00:08:25.540 --> 00:08:27.700 align:middle line:84% the second thing I think is a challenge with this regression, 00:08:27.700 --> 00:08:27.910 align:middle line:90% actually. 00:08:27.910 --> 00:08:28.618 align:middle line:90% I agree with you. 00:08:28.618 --> 00:08:34.220 align:middle line:84% So if you think about, if you take any Behavioral Economics, 00:08:34.220 --> 00:08:36.880 align:middle line:84% you'll see that people make all kinds of risk-aversion 00:08:36.880 --> 00:08:39.460 align:middle line:84% decisions based on things that don't look like-- 00:08:39.460 --> 00:08:42.580 align:middle line:84% the standard economic model says you should be making-- 00:08:42.580 --> 00:08:44.260 align:middle line:84% what should matter is your total wealth 00:08:44.260 --> 00:08:46.510 align:middle line:84% because that's going to affect the degree you can kind 00:08:46.510 --> 00:08:48.790 align:middle line:84% of smooth your consumption from period to period, 00:08:48.790 --> 00:08:51.227 align:middle line:84% and/or you're directly risk averse over your total wealth. 00:08:51.227 --> 00:08:53.560 align:middle line:84% But we know people make all kinds of decisions that look 00:08:53.560 --> 00:08:55.780 align:middle line:90% like they're not like that. 00:08:55.780 --> 00:08:58.660 align:middle line:84% So people think about cash on hand. 00:08:58.660 --> 00:09:04.390 align:middle line:84% They make decisions over much smaller lotteries. 00:09:04.390 --> 00:09:05.890 align:middle line:84% So it's, I think, very easy to think 00:09:05.890 --> 00:09:08.190 align:middle line:84% that if I have some cash on hand that's very liquid, 00:09:08.190 --> 00:09:09.940 align:middle line:84% and then a bunch of stuff that's illiquid, 00:09:09.940 --> 00:09:11.690 align:middle line:84% I might be thinking about the liquid piece 00:09:11.690 --> 00:09:12.700 align:middle line:90% when I'm making my-- 00:09:12.700 --> 00:09:14.230 align:middle line:84% thinking about my risk aversion decisions, 00:09:14.230 --> 00:09:16.210 align:middle line:84% even though our good economic model says that that 00:09:16.210 --> 00:09:17.140 align:middle line:90% shouldn't be the case. 00:09:17.140 --> 00:09:19.780 align:middle line:84% So I think that's exactly the second set of concerns here. 00:09:19.780 --> 00:09:21.880 align:middle line:84% There's also a non-behavioral economic story 00:09:21.880 --> 00:09:23.380 align:middle line:84% to that, which is just like if there 00:09:23.380 --> 00:09:27.100 align:middle line:84% are costs of taking the non-working capital 00:09:27.100 --> 00:09:28.750 align:middle line:84% piece of my wealth that's not liquid-- 00:09:28.750 --> 00:09:30.430 align:middle line:90% like I own a lot of land. 00:09:30.430 --> 00:09:31.930 align:middle line:84% Yes, I own that, but actually, there 00:09:31.930 --> 00:09:35.050 align:middle line:84% could be some costs of using that for consumption smoothing. 00:09:35.050 --> 00:09:37.030 align:middle line:84% And even a straightforward economic model 00:09:37.030 --> 00:09:39.632 align:middle line:84% might say that actually, the working capital is more liquid 00:09:39.632 --> 00:09:41.590 align:middle line:84% and therefore, I going to be more risk averse-- 00:09:41.590 --> 00:09:43.840 align:middle line:84% my risk aversion will be determined by that piece, not 00:09:43.840 --> 00:09:44.632 align:middle line:90% the overall wealth. 00:09:44.632 --> 00:09:48.250 align:middle line:84% So I think both of these are a challenge with this test. 00:09:48.250 --> 00:09:50.860 align:middle line:90% 00:09:50.860 --> 00:09:54.250 align:middle line:84% They find that the working capital is more 00:09:54.250 --> 00:09:55.545 align:middle line:90% predictive than total wealth. 00:09:55.545 --> 00:09:57.670 align:middle line:84% And they argue that's an evidence of sharecropping. 00:09:57.670 --> 00:10:00.558 align:middle line:84% But I think these are some reasons why you might not think 00:10:00.558 --> 00:10:01.850 align:middle line:90% this is the final word on that. 00:10:01.850 --> 00:10:04.142 align:middle line:84% And I just wanted to think that through because I think 00:10:04.142 --> 00:10:08.110 align:middle line:84% that that's like- this test is an example of taking the theory 00:10:08.110 --> 00:10:11.412 align:middle line:90% super duper seriously. 00:10:11.412 --> 00:10:13.870 align:middle line:84% But I think for the reasons that we talked about, actually, 00:10:13.870 --> 00:10:15.250 align:middle line:90% it's not totally obvious. 00:10:15.250 --> 00:10:19.140 align:middle line:84% So that's all I want to say about that. 00:10:19.140 --> 00:10:22.220 align:middle line:90% Other questions? 00:10:22.220 --> 00:10:23.873 align:middle line:90% All right. 00:10:23.873 --> 00:10:25.790 align:middle line:84% And the final thing I wanted to say about this 00:10:25.790 --> 00:10:30.150 align:middle line:84% is, as I mentioned, I think that there 00:10:30.150 --> 00:10:33.690 align:middle line:84% are important implications of these two different models 00:10:33.690 --> 00:10:35.610 align:middle line:90% for land reform. 00:10:35.610 --> 00:10:42.720 align:middle line:84% So as we talked about, under a limited liability, 00:10:42.720 --> 00:10:45.160 align:middle line:84% this could have a really big long-run effect. 00:10:45.160 --> 00:10:49.260 align:middle line:84% The big long effect is it relaxes kind of your liability 00:10:49.260 --> 00:10:52.470 align:middle line:90% because you don't pay anybody. 00:10:52.470 --> 00:10:57.105 align:middle line:90% Under risk aversion, it would-- 00:10:57.105 --> 00:10:58.480 align:middle line:84% you might have less risk aversion 00:10:58.480 --> 00:11:00.640 align:middle line:84% because you're not wealthier because you own the land. 00:11:00.640 --> 00:11:02.015 align:middle line:84% So on a pure risk aversion story, 00:11:02.015 --> 00:11:03.700 align:middle line:84% you would be a little bit richer. 00:11:03.700 --> 00:11:06.550 align:middle line:84% But on the other hand, if you're still risk averse, 00:11:06.550 --> 00:11:09.310 align:middle line:84% sharecropping might, again, kind of emerging endogenously 00:11:09.310 --> 00:11:11.230 align:middle line:84% as a way of insuring some of the risk 00:11:11.230 --> 00:11:12.850 align:middle line:90% that you're still faced against. 00:11:12.850 --> 00:11:15.940 align:middle line:84% So that is a prediction that would 00:11:15.940 --> 00:11:17.890 align:middle line:84% be really interesting to see what happens. 00:11:17.890 --> 00:11:22.390 align:middle line:84% Like if we do a land reform, does this kind of institution 00:11:22.390 --> 00:11:25.960 align:middle line:84% emerge endogenously again or not? 00:11:25.960 --> 00:11:27.280 align:middle line:90% And how does that happen? 00:11:27.280 --> 00:11:29.530 align:middle line:84% And that's not a paper that I think 00:11:29.530 --> 00:11:36.380 align:middle line:90% that there's a great paper on. 00:11:36.380 --> 00:11:39.140 align:middle line:84% More generally, I think that if you 00:11:39.140 --> 00:11:46.250 align:middle line:84% think about the paper we saw about moral hazard, 00:11:46.250 --> 00:11:48.350 align:middle line:84% another prediction of a land reform 00:11:48.350 --> 00:11:50.660 align:middle line:84% is that it should lead to substantial productivity 00:11:50.660 --> 00:11:53.030 align:middle line:90% improvements. 00:11:53.030 --> 00:11:55.190 align:middle line:84% So this one this one in some sense 00:11:55.190 --> 00:11:58.070 align:middle line:84% is about how do we share kind of what's going on. 00:11:58.070 --> 00:12:00.500 align:middle line:84% So several things-- number one is land reform should make 00:12:00.500 --> 00:12:01.190 align:middle line:90% the-- 00:12:01.190 --> 00:12:03.830 align:middle line:84% if you're just giving the land, if you're just 00:12:03.830 --> 00:12:06.020 align:middle line:84% redistributing it, that should make people richer. 00:12:06.020 --> 00:12:08.450 align:middle line:84% Number two, this is unclear what's 00:12:08.450 --> 00:12:10.310 align:middle line:84% going to happen to the form of the contract. 00:12:10.310 --> 00:12:13.370 align:middle line:84% Number three, if there's moral hazard going on 00:12:13.370 --> 00:12:16.213 align:middle line:84% and you don't end up back in the share arrangement, 00:12:16.213 --> 00:12:17.630 align:middle line:84% then you might think there's going 00:12:17.630 --> 00:12:20.790 align:middle line:84% to be substantial increases in output. 00:12:20.790 --> 00:12:22.897 align:middle line:84% Yeah, Kyle, do you have a question? 00:12:22.897 --> 00:12:24.980 align:middle line:84% It's a little hard to tell people without the math 00:12:24.980 --> 00:12:25.740 align:middle line:90% sometimes. 00:12:25.740 --> 00:12:27.920 align:middle line:84% So it would be really nice, I think, 00:12:27.920 --> 00:12:31.370 align:middle line:84% to understand what the implications of land reform 00:12:31.370 --> 00:12:32.430 align:middle line:90% are. 00:12:32.430 --> 00:12:38.110 align:middle line:84% And my own view is that I feel like there 00:12:38.110 --> 00:12:42.160 align:middle line:84% is an awesome land reform paper waiting to be written. 00:12:42.160 --> 00:12:46.397 align:middle line:84% And in particular-- first of all, 00:12:46.397 --> 00:12:48.730 align:middle line:84% there are a lot more failed land reforms successful land 00:12:48.730 --> 00:12:49.300 align:middle line:90% reforms. 00:12:49.300 --> 00:12:51.880 align:middle line:84% There was a huge amount of political pushback 00:12:51.880 --> 00:12:56.410 align:middle line:84% against land reform for, I think, obvious reasons. 00:12:56.410 --> 00:12:59.170 align:middle line:84% The people who were going to get their land taken away from them 00:12:59.170 --> 00:13:01.240 align:middle line:90% were A, upset, and B, wealthy. 00:13:01.240 --> 00:13:03.280 align:middle line:84% And so it's not totally surprising they 00:13:03.280 --> 00:13:06.110 align:middle line:90% were able to push back. 00:13:06.110 --> 00:13:08.330 align:middle line:90% So a lot of them didn't work. 00:13:08.330 --> 00:13:09.580 align:middle line:90% But some of them did. 00:13:09.580 --> 00:13:11.410 align:middle line:90% It was a few. 00:13:11.410 --> 00:13:13.340 align:middle line:84% At least I think there were some examples. 00:13:13.340 --> 00:13:16.053 align:middle line:84% And so some of them had sharp discontinuities 00:13:16.053 --> 00:13:17.470 align:middle line:84% in the amount of land people could 00:13:17.470 --> 00:13:20.980 align:middle line:84% retain, like up to 1 hectare, up to 5 hectares, or whatever. 00:13:20.980 --> 00:13:23.620 align:middle line:84% So it feels like you should be able to-- 00:13:23.620 --> 00:13:25.840 align:middle line:84% if you had the right data, I could 00:13:25.840 --> 00:13:28.810 align:middle line:84% imagine you could match your pre-period data 00:13:28.810 --> 00:13:31.270 align:middle line:84% on what land parcels were and you could somehow 00:13:31.270 --> 00:13:32.230 align:middle line:90% match everything. 00:13:32.230 --> 00:13:34.428 align:middle line:84% You could find discontinuity in your data. 00:13:34.428 --> 00:13:35.470 align:middle line:90% You could try to do that. 00:13:35.470 --> 00:13:40.500 align:middle line:90% 00:13:40.500 --> 00:13:42.048 align:middle line:84% And then you can watch what happened, 00:13:42.048 --> 00:13:43.590 align:middle line:84% both to the people who got wealthier, 00:13:43.590 --> 00:13:45.540 align:middle line:84% to the land-- did it get more productive-- to the contract 00:13:45.540 --> 00:13:46.415 align:middle line:90% form, all that stuff. 00:13:46.415 --> 00:13:53.710 align:middle line:84% So there's a paper by Kitamura in Japan, 00:13:53.710 --> 00:13:56.550 align:middle line:84% which tries something similar to this but without the microdata. 00:13:56.550 --> 00:13:58.940 align:middle line:84% They have to aggregate things up. 00:13:58.940 --> 00:14:01.002 align:middle line:84% There was a paper by Joonas Tuhkuri, who's 00:14:01.002 --> 00:14:02.460 align:middle line:84% a grad student actually [INAUDIBLE] 00:14:02.460 --> 00:14:04.320 align:middle line:84% market tried at some point to study in Finland. 00:14:04.320 --> 00:14:05.680 align:middle line:84% For some reason, it didn't work out. 00:14:05.680 --> 00:14:06.570 align:middle line:90% I don't remember why. 00:14:06.570 --> 00:14:09.870 align:middle line:84% I, working with a grad student a number of years ago, 00:14:09.870 --> 00:14:14.250 align:middle line:84% spent a summer, or the student spent a summer working with me, 00:14:14.250 --> 00:14:17.130 align:middle line:84% looking around for examples all around the world of examples 00:14:17.130 --> 00:14:19.650 align:middle line:84% where we had the-- could find the discontinuity, 00:14:19.650 --> 00:14:22.440 align:middle line:84% and the landform that happened, and the data. 00:14:22.440 --> 00:14:24.062 align:middle line:90% And we didn't come up with any. 00:14:24.062 --> 00:14:25.770 align:middle line:84% But I'm just going to leave this out here 00:14:25.770 --> 00:14:30.960 align:middle line:84% as something I would like to see in a paper. 00:14:30.960 --> 00:14:33.203 align:middle line:84% Because an important part of a PhD class 00:14:33.203 --> 00:14:35.620 align:middle line:84% is not just showing you the papers that have been written, 00:14:35.620 --> 00:14:37.050 align:middle line:84% but mentioning papers that haven't been written yet. 00:14:37.050 --> 00:14:38.883 align:middle line:84% Because that's kind of, of course, the goal. 00:14:38.883 --> 00:14:45.490 align:middle line:84% So anyway, some related studies-- 00:14:45.490 --> 00:14:47.770 align:middle line:84% here are two that came out relatively recently 00:14:47.770 --> 00:14:49.150 align:middle line:90% that are related. 00:14:49.150 --> 00:14:53.320 align:middle line:84% There's also a paper by Abhijit and coauthors 00:14:53.320 --> 00:14:59.810 align:middle line:84% looking at a tenancy reform in West Bengal, which basically, 00:14:59.810 --> 00:15:03.220 align:middle line:84% as I understand it, gave existing sharecropped people 00:15:03.220 --> 00:15:06.142 align:middle line:84% greater protections and greater rights, but didn't actually 00:15:06.142 --> 00:15:08.350 align:middle line:84% change, totally go all the way to the farm ownership. 00:15:08.350 --> 00:15:11.140 align:middle line:84% But that is another one you can look at. 00:15:11.140 --> 00:15:14.740 align:middle line:84% The paper by Galan, which is a study of Columbia, which 00:15:14.740 --> 00:15:21.898 align:middle line:90% looks at people who were-- 00:15:21.898 --> 00:15:24.190 align:middle line:84% there was this program that provided land for the poor. 00:15:24.190 --> 00:15:26.790 align:middle line:90% So this is closer. 00:15:26.790 --> 00:15:31.085 align:middle line:84% I don't think it's quite the same, for a variety of reasons, 00:15:31.085 --> 00:15:33.210 align:middle line:84% but I'll mention this one as one you could look at. 00:15:33.210 --> 00:15:35.580 align:middle line:84% There's also a paper by Montero, which 00:15:35.580 --> 00:15:41.310 align:middle line:84% is an RD that haciendas, large haciendas, 00:15:41.310 --> 00:15:43.060 align:middle line:84% were reorganized into worker cooperatives. 00:15:43.060 --> 00:15:44.050 align:middle line:84% But of course, worker cooperatives 00:15:44.050 --> 00:15:46.180 align:middle line:84% are not the same as divvying everything up individually 00:15:46.180 --> 00:15:47.222 align:middle line:90% to the individual worker. 00:15:47.222 --> 00:15:49.780 align:middle line:84% So these papers are closer and interested. 00:15:49.780 --> 00:15:52.960 align:middle line:84% And I would recommend these plus the West [INAUDIBLE] Bengal 00:15:52.960 --> 00:15:55.080 align:middle line:84% paper, all of these are in the West Bengal paper, 00:15:55.080 --> 00:15:57.650 align:middle line:84% if you're interested in this topic. 00:15:57.650 --> 00:16:01.815 align:middle line:84% OK, that's what I want to say on moral hazard and land reform 00:16:01.815 --> 00:16:03.190 align:middle line:84% before I move on to land titling. 00:16:03.190 --> 00:16:04.570 align:middle line:90% Any last questions or comments? 00:16:04.570 --> 00:16:09.160 align:middle line:90% 00:16:09.160 --> 00:16:11.663 align:middle line:84% So the discussion thus far-- yeah. 00:16:11.663 --> 00:16:12.830 align:middle line:90% AUDIENCE: I have a question. 00:16:12.830 --> 00:16:16.030 align:middle line:84% So if in any of these letter forms, what you're doing 00:16:16.030 --> 00:16:18.900 align:middle line:84% is you're giving property rights to some people. 00:16:18.900 --> 00:16:21.840 align:middle line:84% Are they then required to farm that land, 00:16:21.840 --> 00:16:23.610 align:middle line:84% or can they become landlords themselves? 00:16:23.610 --> 00:16:24.860 align:middle line:90% BEN OLKEN: I think it depends. 00:16:24.860 --> 00:16:26.852 align:middle line:90% 00:16:26.852 --> 00:16:28.560 align:middle line:84% But I'm going to talk about some of this. 00:16:28.560 --> 00:16:29.250 align:middle line:90% Yeah, I think it would depend. 00:16:29.250 --> 00:16:30.570 align:middle line:84% That would be one of the things you decide. 00:16:30.570 --> 00:16:32.653 align:middle line:84% In principle-- but that's actually kind of related 00:16:32.653 --> 00:16:38.365 align:middle line:84% to some of these other questions, like or can 00:16:38.365 --> 00:16:40.590 align:middle line:84% they choose to become sharecroppers again? 00:16:40.590 --> 00:16:43.830 align:middle line:84% Could they choose to sell the land, pocket the money, 00:16:43.830 --> 00:16:45.345 align:middle line:90% or exit farming entirely? 00:16:45.345 --> 00:16:47.970 align:middle line:84% Maybe the other thing they could do is they could sell the land 00:16:47.970 --> 00:16:49.710 align:middle line:90% and move to the city. 00:16:49.710 --> 00:16:51.810 align:middle line:84% So there's lots-- if you have that kind 00:16:51.810 --> 00:16:57.180 align:middle line:84% of complete ownership, you can do what you want with it. 00:16:57.180 --> 00:16:59.120 align:middle line:84% But in many cases, but in some cases, no. 00:16:59.120 --> 00:17:05.119 align:middle line:84% And actually, that understanding of whether or not-- 00:17:05.119 --> 00:17:07.910 align:middle line:84% whether you get the land and it's fully sellable 00:17:07.910 --> 00:17:10.609 align:middle line:84% and transferable versus you get the land 00:17:10.609 --> 00:17:13.273 align:middle line:84% but you can't get rid of it-- has huge implications. 00:17:13.273 --> 00:17:14.690 align:middle line:84% Because you might decide you don't 00:17:14.690 --> 00:17:16.128 align:middle line:90% want to be a farmer anymore. 00:17:16.128 --> 00:17:18.170 align:middle line:84% And so the efficient thing to do is sell the farm 00:17:18.170 --> 00:17:18.829 align:middle line:90% and move to the city. 00:17:18.829 --> 00:17:20.329 align:middle line:84% But if you're tied to the land, then you 00:17:20.329 --> 00:17:21.270 align:middle line:90% can be stuck with that. 00:17:21.270 --> 00:17:23.728 align:middle line:84% And I'm going to talk about some related issues in the land 00:17:23.728 --> 00:17:26.400 align:middle line:84% titling section right now, actually. 00:17:26.400 --> 00:17:27.900 align:middle line:84% So the discussion thus far has been 00:17:27.900 --> 00:17:29.942 align:middle line:84% about contract type, essentially, own versus rent 00:17:29.942 --> 00:17:32.220 align:middle line:84% and the implications for moral hazard. 00:17:32.220 --> 00:17:35.693 align:middle line:84% But what I want to talk about now is this idea that-- 00:17:35.693 --> 00:17:37.110 align:middle line:84% this is kind of related, actually, 00:17:37.110 --> 00:17:39.540 align:middle line:84% to this point-- conditional on owning, 00:17:39.540 --> 00:17:42.990 align:middle line:84% there are different degrees of ownership. 00:17:42.990 --> 00:17:46.620 align:middle line:84% And those can matter in really important ways. 00:17:46.620 --> 00:17:49.860 align:middle line:84% And in particular, many people in many developing countries 00:17:49.860 --> 00:17:54.000 align:middle line:84% do not have well-protected formal titles to their land. 00:17:54.000 --> 00:17:58.520 align:middle line:84% So when you think of a land title, 00:17:58.520 --> 00:18:01.160 align:middle line:84% you probably think of a deed that says, this is my land, 00:18:01.160 --> 00:18:03.320 align:middle line:84% and I can, if I want to sell the house, 00:18:03.320 --> 00:18:05.117 align:middle line:84% or sell it, or mortgage it, or whatever-- 00:18:05.117 --> 00:18:06.950 align:middle line:84% that piece of paper saying this is the land, 00:18:06.950 --> 00:18:09.830 align:middle line:84% this is my title to the land, is really important. 00:18:09.830 --> 00:18:15.137 align:middle line:84% And so if you don't have formal title to your land-- 00:18:15.137 --> 00:18:17.720 align:middle line:84% maybe you own it, but it's kind of some informal understanding 00:18:17.720 --> 00:18:19.558 align:middle line:84% that you own it-- that can create 00:18:19.558 --> 00:18:20.600 align:middle line:90% all kinds of distortions. 00:18:20.600 --> 00:18:24.240 align:middle line:84% For example, it might have reduced investment. 00:18:24.240 --> 00:18:28.280 align:middle line:84% So for example, if I don't know that I own the land forever, 00:18:28.280 --> 00:18:30.788 align:middle line:84% then I might not want to put a lot of fixed investments 00:18:30.788 --> 00:18:32.330 align:middle line:84% into my land because I can be worried 00:18:32.330 --> 00:18:34.340 align:middle line:84% that I'm going to get expropriated and lose 00:18:34.340 --> 00:18:37.250 align:middle line:90% the returns of that investment. 00:18:37.250 --> 00:18:40.310 align:middle line:84% I may have to put in costly effort 00:18:40.310 --> 00:18:42.440 align:middle line:90% to prevent expropriation. 00:18:42.440 --> 00:18:46.092 align:middle line:84% So for example, I may have to stick around. 00:18:46.092 --> 00:18:48.050 align:middle line:84% If I own the land and I have clear title to it, 00:18:48.050 --> 00:18:51.830 align:middle line:84% I can go work in the city and be confident 00:18:51.830 --> 00:18:54.350 align:middle line:84% that my land will still be there and still be owned to me. 00:18:54.350 --> 00:18:56.042 align:middle line:84% If my title is not great, I may actually 00:18:56.042 --> 00:18:57.500 align:middle line:84% have to be there to prevent someone 00:18:57.500 --> 00:18:59.760 align:middle line:90% else from coming and taking it. 00:18:59.760 --> 00:19:01.218 align:middle line:84% So there are things you have to do. 00:19:01.218 --> 00:19:02.843 align:middle line:84% And we're going to talk about all this. 00:19:02.843 --> 00:19:04.070 align:middle line:90% This is an overview. 00:19:04.070 --> 00:19:05.720 align:middle line:90% Another is misallocation. 00:19:05.720 --> 00:19:09.290 align:middle line:84% so if you think about the Coase theorem, which you 00:19:09.290 --> 00:19:13.070 align:middle line:84% guys have heard of the Coase theorem, yes, no? 00:19:13.070 --> 00:19:15.070 align:middle line:84% I always assume that everyone-- sometimes 00:19:15.070 --> 00:19:16.528 align:middle line:84% I assume everyone knows everything. 00:19:16.528 --> 00:19:19.370 align:middle line:84% Of course, then what would be the point of having the class? 00:19:19.370 --> 00:19:21.832 align:middle line:84% So in the Coase theorem, it basically 00:19:21.832 --> 00:19:24.290 align:middle line:84% says if we have good, formal, transferable property rights, 00:19:24.290 --> 00:19:28.400 align:middle line:84% then the property should end up being 00:19:28.400 --> 00:19:32.000 align:middle line:84% owned by the person who's going to the most efficient owner. 00:19:32.000 --> 00:19:34.785 align:middle line:84% And if we have a transfer-- if we 00:19:34.785 --> 00:19:37.160 align:middle line:84% have full ownership in markets, we'll be able to do that. 00:19:37.160 --> 00:19:39.535 align:middle line:84% Because if I have a higher value of the land than you do, 00:19:39.535 --> 00:19:41.545 align:middle line:84% I'll offer to buy the land for more, 00:19:41.545 --> 00:19:42.920 align:middle line:84% and eventually, the land will get 00:19:42.920 --> 00:19:47.548 align:middle line:84% allocated to people with the highest value of the land. 00:19:47.548 --> 00:19:49.090 align:middle line:84% But if I can't sell the land, then we 00:19:49.090 --> 00:19:51.040 align:middle line:84% could end up in a world with misallocation, and people 00:19:51.040 --> 00:19:52.750 align:middle line:84% who are not particularly good farmers 00:19:52.750 --> 00:19:54.880 align:middle line:84% or whatever may end up owning the land because they 00:19:54.880 --> 00:19:56.922 align:middle line:84% can't sell it to people who'll be better farmers. 00:19:56.922 --> 00:20:00.230 align:middle line:90% 00:20:00.230 --> 00:20:04.820 align:middle line:84% And the fourth one is that I can't either sell or mortgage 00:20:04.820 --> 00:20:09.440 align:middle line:84% the property or use it for other transactions. 00:20:09.440 --> 00:20:10.535 align:middle line:90% So if you ever-- 00:20:10.535 --> 00:20:13.160 align:middle line:90% 00:20:13.160 --> 00:20:15.560 align:middle line:84% I suspect none of you have bought a house yet. 00:20:15.560 --> 00:20:20.570 align:middle line:84% But when you do someday, you will probably get a mortgage. 00:20:20.570 --> 00:20:23.192 align:middle line:84% And the mortgage basically says, I pledge-- 00:20:23.192 --> 00:20:24.900 align:middle line:84% there's actually two documents in the US, 00:20:24.900 --> 00:20:26.420 align:middle line:90% at least if you buy a house. 00:20:26.420 --> 00:20:29.313 align:middle line:84% One is a loan saying, which is actually, 00:20:29.313 --> 00:20:31.730 align:middle line:84% as I discovered when I bought a house, the short document, 00:20:31.730 --> 00:20:34.820 align:middle line:84% which basically says, I owe the bank X $1,000. 00:20:34.820 --> 00:20:36.380 align:middle line:84% Then the more complicated document 00:20:36.380 --> 00:20:38.000 align:middle line:84% is the mortgage, which basically says, 00:20:38.000 --> 00:20:41.630 align:middle line:84% and if I don't pay the loan, you can take my house, 00:20:41.630 --> 00:20:46.520 align:middle line:84% and that I'm pledging the house as collateral for the loan. 00:20:46.520 --> 00:20:50.090 align:middle line:84% And the idea is that the ability to have good titles 00:20:50.090 --> 00:20:52.550 align:middle line:84% is important because then you can use them 00:20:52.550 --> 00:20:54.380 align:middle line:90% as collateral for loans. 00:20:54.380 --> 00:20:56.780 align:middle line:84% And it can relax your credit constraints. 00:20:56.780 --> 00:21:02.090 align:middle line:84% And there was a big argument by De Soto, in I guess, 00:21:02.090 --> 00:21:04.160 align:middle line:84% I think, the late '80s or early '90s 00:21:04.160 --> 00:21:06.530 align:middle line:84% or so, which was basically arguing 00:21:06.530 --> 00:21:10.490 align:middle line:84% that the lack of formal title was creating a lot of credit 00:21:10.490 --> 00:21:14.390 align:middle line:84% constraints and putting limitations on people's ability 00:21:14.390 --> 00:21:16.700 align:middle line:90% to borrow. 00:21:16.700 --> 00:21:19.250 align:middle line:84% So what I want to talk about now is 00:21:19.250 --> 00:21:21.380 align:middle line:84% what do we know about all this stuff. 00:21:21.380 --> 00:21:24.380 align:middle line:84% And in some sense, De Soto was saying this is a huge problem, 00:21:24.380 --> 00:21:26.082 align:middle line:84% and then there was all this evidence, 00:21:26.082 --> 00:21:27.290 align:middle line:90% work trying to understand it. 00:21:27.290 --> 00:21:30.770 align:middle line:84% And you'll see where I think we are on some of this stuff. 00:21:30.770 --> 00:21:31.430 align:middle line:90% Questions? 00:21:31.430 --> 00:21:33.000 align:middle line:90% Overview? 00:21:33.000 --> 00:21:33.530 align:middle line:90% Yeah. 00:21:33.530 --> 00:21:34.030 align:middle line:90% Good. 00:21:34.030 --> 00:21:37.040 align:middle line:90% 00:21:37.040 --> 00:21:41.900 align:middle line:84% So as an aside, let me just say that the institutions 00:21:41.900 --> 00:21:45.410 align:middle line:84% of property rights are complicated. 00:21:45.410 --> 00:21:50.610 align:middle line:84% And by this, I'm going to give you an example as follows. 00:21:50.610 --> 00:21:55.200 align:middle line:84% So suppose you were buying a house, 00:21:55.200 --> 00:21:57.510 align:middle line:84% and someone offered you in exchange 00:21:57.510 --> 00:22:00.320 align:middle line:84% for all the thousands of dollars what's 00:22:00.320 --> 00:22:02.900 align:middle line:90% called a "quitclaim" deed. 00:22:02.900 --> 00:22:06.700 align:middle line:84% A quitclaim deed-- this is a definition that you can get 00:22:06.700 --> 00:22:07.870 align:middle line:90% from Wikipedia-- 00:22:07.870 --> 00:22:10.420 align:middle line:84% a quitclaim deed says the following-- the owner 00:22:10.420 --> 00:22:12.100 align:middle line:84% or grantor of this deed terminates 00:22:12.100 --> 00:22:15.640 align:middle line:84% or quits any right or claim to the property, 00:22:15.640 --> 00:22:17.440 align:middle line:84% thereby allowing the right or the claim 00:22:17.440 --> 00:22:20.560 align:middle line:84% to transfer to the person you're granting it to. 00:22:20.560 --> 00:22:22.390 align:middle line:84% But unlike most other property deeds, 00:22:22.390 --> 00:22:27.190 align:middle line:84% a quitclaim deed contains no title covenant 00:22:27.190 --> 00:22:29.960 align:middle line:84% and thus, does not offer any guarantee or warranty 00:22:29.960 --> 00:22:31.210 align:middle line:90% as to the status of the title. 00:22:31.210 --> 00:22:34.240 align:middle line:84% So the grantee is entitled only to whatever interest 00:22:34.240 --> 00:22:37.090 align:middle line:84% the grantor actually may or may not 00:22:37.090 --> 00:22:39.340 align:middle line:84% possess at the time of the transfer. 00:22:39.340 --> 00:22:42.040 align:middle line:84% So this means that the person issuing this deed 00:22:42.040 --> 00:22:47.170 align:middle line:84% does not guarantee that he or she actually owns the property, 00:22:47.170 --> 00:22:51.180 align:middle line:84% or if she does, that no one else has a claim on the property. 00:22:51.180 --> 00:22:53.160 align:middle line:84% It's just saying-- this deed basically 00:22:53.160 --> 00:22:56.490 align:middle line:84% says, if I happen to own anything 00:22:56.490 --> 00:23:02.120 align:middle line:84% in this property, whatever I may or may not own, it's yours. 00:23:02.120 --> 00:23:04.440 align:middle line:90% So would you buy that house? 00:23:04.440 --> 00:23:05.990 align:middle line:90% What do you think? 00:23:05.990 --> 00:23:07.910 align:middle line:90% No. 00:23:07.910 --> 00:23:08.947 align:middle line:90% Anyone else? 00:23:08.947 --> 00:23:11.530 align:middle line:84% AUDIENCE: Well, if you have some mechanism to verify what they 00:23:11.530 --> 00:23:14.162 align:middle line:84% [INAUDIBLE] very simple registry, then that's kind of-- 00:23:14.162 --> 00:23:14.870 align:middle line:90% BEN OLKEN: Right. 00:23:14.870 --> 00:23:18.600 align:middle line:90% 00:23:18.600 --> 00:23:20.750 align:middle line:84% Like Salome, when I first heard about this, 00:23:20.750 --> 00:23:22.000 align:middle line:90% I thought this sounded insane. 00:23:22.000 --> 00:23:24.720 align:middle line:90% 00:23:24.720 --> 00:23:26.560 align:middle line:90% Why would I possibly do this? 00:23:26.560 --> 00:23:28.895 align:middle line:84% Turns out every deed, essentially 95% 00:23:28.895 --> 00:23:30.520 align:middle line:84% of deeds in the State of Massachusetts, 00:23:30.520 --> 00:23:34.740 align:middle line:84% including my house, come with this form of deed. 00:23:34.740 --> 00:23:38.700 align:middle line:90% And why is that? 00:23:38.700 --> 00:23:40.360 align:middle line:90% And why does this possibly work? 00:23:40.360 --> 00:23:41.490 align:middle line:84% So I think, as you're saying, Rebecca, 00:23:41.490 --> 00:23:43.198 align:middle line:84% it's not the quitclaim deed in isolation. 00:23:43.198 --> 00:23:47.160 align:middle line:84% So what basically happens is we here 00:23:47.160 --> 00:23:50.880 align:middle line:84% in Massachusetts have very good land records, 00:23:50.880 --> 00:23:52.900 align:middle line:90% but not perfect land records. 00:23:52.900 --> 00:23:57.060 align:middle line:84% So when you buy the quitclaim deed, you can go, 00:23:57.060 --> 00:23:58.830 align:middle line:84% or your lawyer can go, and search-- 00:23:58.830 --> 00:24:02.138 align:middle line:84% go to the registry of deeds and search the history of quitclaim 00:24:02.138 --> 00:24:03.180 align:middle line:90% deeds back to your house. 00:24:03.180 --> 00:24:06.442 align:middle line:84% And they can say, well, so-and-so 00:24:06.442 --> 00:24:08.400 align:middle line:84% has a quitclaim deed from some previous person, 00:24:08.400 --> 00:24:09.760 align:middle line:84% previously so-and-so, and so forth. 00:24:09.760 --> 00:24:11.427 align:middle line:84% So it looks like maybe there's something 00:24:11.427 --> 00:24:14.790 align:middle line:84% going on there that looks-- and that gives you 00:24:14.790 --> 00:24:16.770 align:middle line:84% reasonably good assurance that the person who's 00:24:16.770 --> 00:24:20.610 align:middle line:84% giving you the deed probably owns the house. 00:24:20.610 --> 00:24:25.825 align:middle line:84% And if not, you also buy title insurance. 00:24:25.825 --> 00:24:27.450 align:middle line:84% So when you buy a house, at least here, 00:24:27.450 --> 00:24:30.060 align:middle line:84% you have to buy insurance against the fact 00:24:30.060 --> 00:24:32.580 align:middle line:84% that maybe they don't actually own the house. 00:24:32.580 --> 00:24:34.230 align:middle line:84% And it's a small enough risk because 00:24:34.230 --> 00:24:37.980 align:middle line:84% of the pretty good information that actually, that insurance 00:24:37.980 --> 00:24:38.952 align:middle line:90% is not-- 00:24:38.952 --> 00:24:41.160 align:middle line:84% I don't think it's particularly competitively priced, 00:24:41.160 --> 00:24:42.160 align:middle line:90% I will say. 00:24:42.160 --> 00:24:44.978 align:middle line:84% But I think that insurance is not that expensive compared 00:24:44.978 --> 00:24:46.020 align:middle line:90% to the cost of the house. 00:24:46.020 --> 00:24:47.320 align:middle line:84% Because as you were saying, Rebecca, 00:24:47.320 --> 00:24:48.790 align:middle line:84% they have pretty good information. 00:24:48.790 --> 00:24:51.860 align:middle line:90% So this is a pretty small risk. 00:24:51.860 --> 00:24:56.165 align:middle line:84% So the only point I'm making is that actually-- 00:24:56.165 --> 00:24:57.540 align:middle line:84% first of all, I love this example 00:24:57.540 --> 00:24:59.165 align:middle line:84% because I still can't get over the fact 00:24:59.165 --> 00:25:01.230 align:middle line:84% that this is how this actually works. 00:25:01.230 --> 00:25:05.700 align:middle line:84% But I think the broader point here is it's not just-- 00:25:05.700 --> 00:25:08.070 align:middle line:84% you have to think about the whole set of institutions 00:25:08.070 --> 00:25:11.700 align:middle line:84% in their entirety and how all the different pieces of them 00:25:11.700 --> 00:25:12.420 align:middle line:90% work together. 00:25:12.420 --> 00:25:18.030 align:middle line:84% This not super great deed works reasonably well 00:25:18.030 --> 00:25:19.950 align:middle line:84% as part of a three-part process where 00:25:19.950 --> 00:25:22.530 align:middle line:84% you have these reasonable deeds, very good records, 00:25:22.530 --> 00:25:24.580 align:middle line:84% and title insurance for the third piece of it. 00:25:24.580 --> 00:25:31.260 align:middle line:84% And so I think the institutional details are really important. 00:25:31.260 --> 00:25:35.930 align:middle line:84% And if you try to understand them, 00:25:35.930 --> 00:25:37.600 align:middle line:84% they're important for understanding them 00:25:37.600 --> 00:25:38.940 align:middle line:84% if you're ever going to do anything in this context. 00:25:38.940 --> 00:25:41.023 align:middle line:84% Another thing which, by the way, is also kind of-- 00:25:41.023 --> 00:25:43.717 align:middle line:90% 00:25:43.717 --> 00:25:45.800 align:middle line:84% let me make one other general institutional point. 00:25:45.800 --> 00:25:51.940 align:middle line:84% You may think that why would there be any possible downside 00:25:51.940 --> 00:25:53.230 align:middle line:90% to land titling? 00:25:53.230 --> 00:25:55.090 align:middle line:84% It seems like of course, we should 00:25:55.090 --> 00:25:57.050 align:middle line:90% issue really good titles. 00:25:57.050 --> 00:26:01.780 align:middle line:84% However, let me just point out that the process 00:26:01.780 --> 00:26:03.970 align:middle line:84% of actually going through and titling land 00:26:03.970 --> 00:26:05.600 align:middle line:90% is very, very complicated. 00:26:05.600 --> 00:26:08.560 align:middle line:84% So if you're in a world with informal boundaries 00:26:08.560 --> 00:26:11.050 align:middle line:84% or whatever, and you're going to title them, 00:26:11.050 --> 00:26:12.520 align:middle line:84% everyone understands that whatever 00:26:12.520 --> 00:26:14.290 align:middle line:84% titles get marked down and delineated 00:26:14.290 --> 00:26:17.650 align:middle line:84% are going to be actual land border definitions now 00:26:17.650 --> 00:26:19.080 align:middle line:90% and forever. 00:26:19.080 --> 00:26:25.550 align:middle line:84% And so every possible dispute over every possible border 00:26:25.550 --> 00:26:27.650 align:middle line:84% has to be adjudicated and settled 00:26:27.650 --> 00:26:29.538 align:middle line:90% before the titling is done. 00:26:29.538 --> 00:26:31.580 align:middle line:84% So you might imagine that there's lots of borders 00:26:31.580 --> 00:26:36.030 align:middle line:84% that we don't actually need to really worry about. 00:26:36.030 --> 00:26:37.500 align:middle line:84% It's kind of like we have our land, 00:26:37.500 --> 00:26:38.910 align:middle line:84% and there's kind of some forests in between, 00:26:38.910 --> 00:26:41.220 align:middle line:84% and we kind of know roughly where the thing is. 00:26:41.220 --> 00:26:46.320 align:middle line:84% And that's fine for a lot for us because you and I don't 00:26:46.320 --> 00:26:48.900 align:middle line:84% actually particularly care whether the border is 00:26:48.900 --> 00:26:50.922 align:middle line:84% on this side of the table or a foot over here 00:26:50.922 --> 00:26:53.130 align:middle line:84% on this side of the table, because it's approximately 00:26:53.130 --> 00:26:54.780 align:middle line:90% the border and we're good. 00:26:54.780 --> 00:26:55.810 align:middle line:90% It seems like it's fine. 00:26:55.810 --> 00:26:57.893 align:middle line:84% But if the government is going to come on and say, 00:26:57.893 --> 00:27:00.390 align:middle line:84% this is the answer, every single one of those disputes 00:27:00.390 --> 00:27:02.310 align:middle line:90% has to be adjudicated. 00:27:02.310 --> 00:27:04.200 align:middle line:84% So in some sense, you move from a world 00:27:04.200 --> 00:27:07.110 align:middle line:84% where disputes are only adjudicated 00:27:07.110 --> 00:27:10.393 align:middle line:84% in the small set of cases where they actually matter to a world 00:27:10.393 --> 00:27:12.810 align:middle line:84% where you have to actually adjudicate everything up front. 00:27:12.810 --> 00:27:13.860 align:middle line:90% And that could be costly. 00:27:13.860 --> 00:27:15.777 align:middle line:84% I'm not saying, by the way, it's not worth it. 00:27:15.777 --> 00:27:19.680 align:middle line:84% I'm just saying it's worth thinking those things through. 00:27:19.680 --> 00:27:21.540 align:middle line:84% There are other issues that, if you're 00:27:21.540 --> 00:27:23.050 align:middle line:84% interested in this stuff, get kind of complicated. 00:27:23.050 --> 00:27:24.210 align:middle line:84% For example, another reason why we 00:27:24.210 --> 00:27:25.963 align:middle line:84% have these crazy deeds in Massachusetts 00:27:25.963 --> 00:27:27.630 align:middle line:84% is there's a common law principle called 00:27:27.630 --> 00:27:29.400 align:middle line:84% adverse possession, which basically means 00:27:29.400 --> 00:27:33.240 align:middle line:84% if I'm using land for a certain number of years, 00:27:33.240 --> 00:27:36.983 align:middle line:84% and no one has explicitly given me permission to use it, 00:27:36.983 --> 00:27:38.400 align:middle line:84% no one's stopped me from using it. 00:27:38.400 --> 00:27:40.430 align:middle line:84% I just kind of do it and nobody says anything. 00:27:40.430 --> 00:27:42.180 align:middle line:84% After a certain number of years, that land 00:27:42.180 --> 00:27:43.170 align:middle line:90% actually becomes mine. 00:27:43.170 --> 00:27:45.437 align:middle line:90% That's a common law principle. 00:27:45.437 --> 00:27:47.770 align:middle line:84% And so actually, another reason we have the crazy deeds, 00:27:47.770 --> 00:27:50.500 align:middle line:84% we don't know actually what's happened with all that usage 00:27:50.500 --> 00:27:51.000 align:middle line:90% stuff. 00:27:51.000 --> 00:27:53.670 align:middle line:84% And in fact, there would be reasonable reasons 00:27:53.670 --> 00:27:57.990 align:middle line:84% to have those kinds of things to formally recognize things that 00:27:57.990 --> 00:27:59.740 align:middle line:90% have been happening over time. 00:27:59.740 --> 00:28:02.160 align:middle line:84% So my only point is I actually think 00:28:02.160 --> 00:28:06.630 align:middle line:84% that the details of how do you do all this titling 00:28:06.630 --> 00:28:08.580 align:middle line:84% are probably a lot more complicated 00:28:08.580 --> 00:28:11.378 align:middle line:84% than you might at first think, of oh, it's your house. 00:28:11.378 --> 00:28:12.420 align:middle line:90% I sent you a little deed. 00:28:12.420 --> 00:28:13.378 align:middle line:90% Write down, we're done. 00:28:13.378 --> 00:28:15.295 align:middle line:84% The actual process of doing it is interesting. 00:28:15.295 --> 00:28:17.880 align:middle line:84% And I suspect there's probably some interesting and good 00:28:17.880 --> 00:28:21.643 align:middle line:84% research to be done on the details of how do you do that. 00:28:21.643 --> 00:28:23.310 align:middle line:84% There could be really interesting papers 00:28:23.310 --> 00:28:27.085 align:middle line:90% on the adjudication process. 00:28:27.085 --> 00:28:29.460 align:middle line:84% I, at one point, wanted to do a paper on the adjudication 00:28:29.460 --> 00:28:30.090 align:middle line:90% process. 00:28:30.090 --> 00:28:32.550 align:middle line:84% And how do you actually get communities to agree 00:28:32.550 --> 00:28:34.750 align:middle line:90% on what the titles are? 00:28:34.750 --> 00:28:38.010 align:middle line:84% And maybe if-- how does the political environment 00:28:38.010 --> 00:28:40.010 align:middle line:84% affect kind of that adjudication process? 00:28:40.010 --> 00:28:42.010 align:middle line:84% I think there's lots of interesting stuff there. 00:28:42.010 --> 00:28:45.300 align:middle line:84% So I just wanted to mention that before I talk about it a bit. 00:28:45.300 --> 00:28:46.170 align:middle line:90% Yeah, kind of, 00:28:46.170 --> 00:28:47.878 align:middle line:84% AUDIENCE: When you're thinking about this 00:28:47.878 --> 00:28:51.100 align:middle line:84% and the risk of the limited liability stuff, 00:28:51.100 --> 00:28:54.390 align:middle line:84% is it like that we are sure that land is being used 00:28:54.390 --> 00:28:56.890 align:middle line:84% inefficiently in developing countries, 00:28:56.890 --> 00:28:58.990 align:middle line:84% and we're trying to figure out why that is? 00:28:58.990 --> 00:29:01.760 align:middle line:84% Or is it like there are these problems? 00:29:01.760 --> 00:29:04.280 align:middle line:84% BEN OLKEN: I'm going to talk a little bit about that later. 00:29:04.280 --> 00:29:16.030 align:middle line:84% So let's come back to that at the end of this lecture. 00:29:16.030 --> 00:29:17.530 align:middle line:84% Let me see if some of the papers I'm 00:29:17.530 --> 00:29:19.160 align:middle line:84% going to answer solve your question-- 00:29:19.160 --> 00:29:20.410 align:middle line:90% answer some of your questions. 00:29:20.410 --> 00:29:21.952 align:middle line:84% Because there are at least a few that 00:29:21.952 --> 00:29:23.740 align:middle line:90% are going to show inefficiency. 00:29:23.740 --> 00:29:27.060 align:middle line:90% Yeah, question? 00:29:27.060 --> 00:29:32.343 align:middle line:84% All right, so now, I want to talk about the four-- 00:29:32.343 --> 00:29:34.260 align:middle line:84% after my little aside-- the four institutions, 00:29:34.260 --> 00:29:36.390 align:middle line:84% the four little pieces, four pieces here. 00:29:36.390 --> 00:29:38.100 align:middle line:84% What are the impacts on investment, 00:29:38.100 --> 00:29:41.790 align:middle line:84% on effort to prevent expropriation, misallocation-- 00:29:41.790 --> 00:29:44.830 align:middle line:84% which is a bit kind of related to-- 00:29:44.830 --> 00:29:48.150 align:middle line:84% sorry, on inability for collateral. 00:29:48.150 --> 00:29:50.240 align:middle line:84% I think all of these, actually in some sense, 00:29:50.240 --> 00:29:52.740 align:middle line:84% are trying to answer versions of the efficiency consequences 00:29:52.740 --> 00:29:54.848 align:middle line:90% question. 00:29:54.848 --> 00:29:56.890 align:middle line:84% Only one of them is like the wrong people own it. 00:29:56.890 --> 00:29:58.830 align:middle line:84% But all of these are saying on net, 00:29:58.830 --> 00:30:01.420 align:middle line:84% does all this stuff have inefficiency consequences? 00:30:01.420 --> 00:30:04.270 align:middle line:84% Not necessarily because of moral hazard, but in general. 00:30:04.270 --> 00:30:07.273 align:middle line:90% 00:30:07.273 --> 00:30:08.940 align:middle line:84% And I think the answer to your question, 00:30:08.940 --> 00:30:10.732 align:middle line:84% I think the point of that [INAUDIBLE] paper 00:30:10.732 --> 00:30:14.360 align:middle line:84% that we read last time was that does show the efficiency 00:30:14.360 --> 00:30:16.670 align:middle line:84% consequences of the sharecropping 00:30:16.670 --> 00:30:18.585 align:middle line:90% arrangements associated. 00:30:18.585 --> 00:30:19.710 align:middle line:90% That was kind of the point. 00:30:19.710 --> 00:30:23.310 align:middle line:90% 00:30:23.310 --> 00:30:28.060 align:middle line:84% So the first thing I want to talk about is investment 00:30:28.060 --> 00:30:31.030 align:middle line:84% and how does investment change with better property rights. 00:30:31.030 --> 00:30:34.530 align:middle line:84% So Goldstein law, we're going to study a land demarcation 00:30:34.530 --> 00:30:37.370 align:middle line:90% paper in Benin. 00:30:37.370 --> 00:30:43.290 align:middle line:84% Now note-- this is not the titling piece. 00:30:43.290 --> 00:30:46.252 align:middle line:90% This is the demarcation piece. 00:30:46.252 --> 00:30:48.710 align:middle line:84% So I don't think they actually then fully title everything, 00:30:48.710 --> 00:30:54.080 align:middle line:84% but they're just going through where exactly are the borders 00:30:54.080 --> 00:30:56.570 align:middle line:84% piece, and let's all agree on that. 00:30:56.570 --> 00:30:58.010 align:middle line:84% But you can think of this as part 00:30:58.010 --> 00:31:01.490 align:middle line:84% of the process of moving towards clarifying property rights. 00:31:01.490 --> 00:31:02.990 align:middle line:84% And there's an expectation this will 00:31:02.990 --> 00:31:05.118 align:middle line:84% form the basis of a future property titling, 00:31:05.118 --> 00:31:07.660 align:middle line:84% but they haven't actually done that at the time of the study. 00:31:07.660 --> 00:31:11.230 align:middle line:84% Now, what I want to talk about for a sec here 00:31:11.230 --> 00:31:12.940 align:middle line:90% is this experimental design. 00:31:12.940 --> 00:31:14.733 align:middle line:84% So take a minute and have a look at this. 00:31:14.733 --> 00:31:16.650 align:middle line:84% And if you can't read it, hopefully, you still 00:31:16.650 --> 00:31:18.525 align:middle line:84% have your handout from last time because this 00:31:18.525 --> 00:31:21.240 align:middle line:90% is reproduced in the handout. 00:31:21.240 --> 00:31:24.810 align:middle line:84% I want to compare the experimental design 00:31:24.810 --> 00:31:27.360 align:middle line:84% they did on the left with the experimental design 00:31:27.360 --> 00:31:29.850 align:middle line:84% that one could have done on the right. 00:31:29.850 --> 00:31:32.802 align:middle line:84% And I will put as an aside, I put this 00:31:32.802 --> 00:31:35.010 align:middle line:84% up there because I put this once as an exam question, 00:31:35.010 --> 00:31:36.670 align:middle line:84% and I felt like people did not fully know the answer. 00:31:36.670 --> 00:31:38.337 align:middle line:84% So I decided it should go in the lecture 00:31:38.337 --> 00:31:40.615 align:middle line:90% rather than on the exam. 00:31:40.615 --> 00:31:41.490 align:middle line:90% So what do you think? 00:31:41.490 --> 00:31:44.770 align:middle line:84% What's the difference between these two experimental designs? 00:31:44.770 --> 00:31:48.142 align:middle line:84% So what's different about these two experimental designs? 00:31:48.142 --> 00:31:49.350 align:middle line:90% It's a randomized experiment. 00:31:49.350 --> 00:31:52.440 align:middle line:84% We're going to randomize these two different-- 00:31:52.440 --> 00:31:53.820 align:middle line:84% we're going to randomize villages 00:31:53.820 --> 00:31:58.440 align:middle line:84% into the getting this kind of land titling treatment or not. 00:31:58.440 --> 00:32:00.845 align:middle line:90% And let's forget about this. 00:32:00.845 --> 00:32:02.970 align:middle line:84% I don't remember what this EMICoV thing is exactly. 00:32:02.970 --> 00:32:03.970 align:middle line:90% Let's forget about this. 00:32:03.970 --> 00:32:10.960 align:middle line:84% I think it's actually not essential to this, not central. 00:32:10.960 --> 00:32:13.120 align:middle line:84% So what's the difference between these two designs? 00:32:13.120 --> 00:32:17.980 align:middle line:90% 00:32:17.980 --> 00:32:19.435 align:middle line:90% OK, go ahead, Christine. 00:32:19.435 --> 00:32:21.190 align:middle line:84% I haven't heard from you in a while. 00:32:21.190 --> 00:32:24.190 align:middle line:84% AUDIENCE: In the left hand, you're 00:32:24.190 --> 00:32:27.150 align:middle line:84% working only with villages that are interested. 00:32:27.150 --> 00:32:30.720 align:middle line:84% And on the right hand, you are working 00:32:30.720 --> 00:32:34.020 align:middle line:84% with villages that are both interested and not interested. 00:32:34.020 --> 00:32:37.290 align:middle line:90% 00:32:37.290 --> 00:32:38.220 align:middle line:90% BEN OLKEN: No. 00:32:38.220 --> 00:32:39.060 align:middle line:90% Well, not really. 00:32:39.060 --> 00:32:41.850 align:middle line:84% So what do you mean, working with? 00:32:41.850 --> 00:32:45.830 align:middle line:84% AUDIENCE: I mean, [INAUDIBLE] people 00:32:45.830 --> 00:32:49.247 align:middle line:84% only from a subset of people that self-selected and be 00:32:49.247 --> 00:32:50.330 align:middle line:90% interested in this design. 00:32:50.330 --> 00:32:50.890 align:middle line:90% BEN OLKEN: On the left one. 00:32:50.890 --> 00:32:51.680 align:middle line:90% AUDIENCE: Yeah. 00:32:51.680 --> 00:32:55.700 align:middle line:84% And on the right hand, you have controls that could be either 00:32:55.700 --> 00:32:57.650 align:middle line:90% inclusive or not. 00:32:57.650 --> 00:33:04.370 align:middle line:84% And [INAUDIBLE] people that are interested or not interested. 00:33:04.370 --> 00:33:06.770 align:middle line:90% BEN OLKEN: Correct. 00:33:06.770 --> 00:33:07.310 align:middle line:90% Yes. 00:33:07.310 --> 00:33:08.810 align:middle line:84% I think that is the main difference. 00:33:08.810 --> 00:33:12.050 align:middle line:90% [INAUDIBLE] Same thing? 00:33:12.050 --> 00:33:14.391 align:middle line:84% AUDIENCE: Oh, I guess yeah, you might only 00:33:14.391 --> 00:33:17.734 align:middle line:84% be interested in the population of villages that are 00:33:17.734 --> 00:33:21.250 align:middle line:90% interested and informed anyway. 00:33:21.250 --> 00:33:25.144 align:middle line:84% So if you're looking to find a treatment [INAUDIBLE].. 00:33:25.144 --> 00:33:28.622 align:middle line:90% 00:33:28.622 --> 00:33:30.830 align:middle line:84% BEN OLKEN: Yeah, that would be interested population. 00:33:30.830 --> 00:33:32.330 align:middle line:90% Yeah, Pablo, what do you think? 00:33:32.330 --> 00:33:35.270 align:middle line:84% AUDIENCE: The last piece is the eligibility part. 00:33:35.270 --> 00:33:43.450 align:middle line:84% So on the left-hand side, there are eligible villages 00:33:43.450 --> 00:33:46.930 align:middle line:84% that are interested that are stratified into treatment 00:33:46.930 --> 00:33:48.370 align:middle line:90% and control groups. 00:33:48.370 --> 00:33:51.310 align:middle line:84% Whereas on the left-hand side, once 00:33:51.310 --> 00:33:53.052 align:middle line:84% you're in the treatment group, if you're 00:33:53.052 --> 00:33:55.510 align:middle line:84% interested and eligible, you're always going to be treated. 00:33:55.510 --> 00:33:56.260 align:middle line:90% BEN OLKEN: [INAUDIBLE]. 00:33:56.260 --> 00:33:58.052 align:middle line:84% AUDIENCE: You're always going to be treated 00:33:58.052 --> 00:34:00.910 align:middle line:84% if you are interested and eligible, and [INAUDIBLE] 00:34:00.910 --> 00:34:03.140 align:middle line:90% treatment [INAUDIBLE]. 00:34:03.140 --> 00:34:05.240 align:middle line:84% BEN OLKEN: Yeah, so how we deal with interest 00:34:05.240 --> 00:34:06.230 align:middle line:84% and how we deal with eligibility, 00:34:06.230 --> 00:34:08.397 align:middle line:84% we deal with them at different stages of this design 00:34:08.397 --> 00:34:09.030 align:middle line:90% and before-- 00:34:09.030 --> 00:34:11.558 align:middle line:84% so now my next question-- that's the key difference. 00:34:11.558 --> 00:34:12.350 align:middle line:90% So which is better? 00:34:12.350 --> 00:34:13.475 align:middle line:90% Which is the better design? 00:34:13.475 --> 00:34:18.260 align:middle line:90% 00:34:18.260 --> 00:34:20.514 align:middle line:84% AUDIENCE: The one on the left seems better. 00:34:20.514 --> 00:34:22.139 align:middle line:84% BEN OLKEN: Sorry, what were you saying? 00:34:22.139 --> 00:34:24.625 align:middle line:84% AUDIENCE: Doesn't that depend on what effect 00:34:24.625 --> 00:34:25.750 align:middle line:90% you are trying to estimate. 00:34:25.750 --> 00:34:27.639 align:middle line:84% Because treatment group is the same. 00:34:27.639 --> 00:34:30.110 align:middle line:84% Control groups will be different. 00:34:30.110 --> 00:34:31.750 align:middle line:84% BEN OLKEN: So Ahmed says maybe it 00:34:31.750 --> 00:34:33.790 align:middle line:84% depends on what you're going to estimate. 00:34:33.790 --> 00:34:36.940 align:middle line:84% [INAUDIBLE],, why do you think the left one's better? 00:34:36.940 --> 00:34:40.210 align:middle line:90% AUDIENCE: So you are-- 00:34:40.210 --> 00:34:44.170 align:middle line:84% I mean, the classification of knowing what types of people 00:34:44.170 --> 00:34:47.350 align:middle line:84% are in the control group, you don't know because you're not 00:34:47.350 --> 00:34:49.300 align:middle line:90% measuring this. 00:34:49.300 --> 00:34:51.429 align:middle line:84% BEN OLKEN: Right, you don't know who's over here. 00:34:51.429 --> 00:34:52.846 align:middle line:84% AUDIENCE: And also, you're getting 00:34:52.846 --> 00:34:56.985 align:middle line:84% pretty intense selection effects on the right side. 00:34:56.985 --> 00:34:59.500 align:middle line:84% BEN OLKEN: OK, so when you say selection effects, 00:34:59.500 --> 00:35:00.495 align:middle line:90% what do you mean? 00:35:00.495 --> 00:35:01.870 align:middle line:84% So does this give you the right-- 00:35:01.870 --> 00:35:06.520 align:middle line:90% 00:35:06.520 --> 00:35:10.283 align:middle line:84% are both these designs internally valid? 00:35:10.283 --> 00:35:11.116 align:middle line:90% AUDIENCE: I suppose. 00:35:11.116 --> 00:35:12.742 align:middle line:84% BEN OLKEN: Let me ask you a question. 00:35:12.742 --> 00:35:14.200 align:middle line:84% So do these designs differ in terms 00:35:14.200 --> 00:35:15.880 align:middle line:84% of their internal validity or external validity? 00:35:15.880 --> 00:35:17.505 align:middle line:84% Let me ask those more simple questions. 00:35:17.505 --> 00:35:19.458 align:middle line:90% 00:35:19.458 --> 00:35:21.000 align:middle line:84% Sorry, Patrick, you had your hand up. 00:35:21.000 --> 00:35:21.673 align:middle line:90% I'm sorry. 00:35:21.673 --> 00:35:23.340 align:middle line:84% AUDIENCE: I mean, it was just addressing 00:35:23.340 --> 00:35:24.290 align:middle line:90% which one was better. 00:35:24.290 --> 00:35:25.165 align:middle line:90% BEN OLKEN: I'm sorry? 00:35:25.165 --> 00:35:27.380 align:middle line:84% AUDIENCE: It was just addressing which one is better. 00:35:27.380 --> 00:35:28.268 align:middle line:90% BEN OLKEN: Which one? 00:35:28.268 --> 00:35:30.060 align:middle line:84% AUDIENCE: I think if power is not an issue, 00:35:30.060 --> 00:35:31.670 align:middle line:90% I'd say the right-hand side. 00:35:31.670 --> 00:35:33.770 align:middle line:84% BEN OLKEN: But we live in a finite world, 00:35:33.770 --> 00:35:34.853 align:middle line:90% so which one do you think? 00:35:34.853 --> 00:35:36.933 align:middle line:90% 00:35:36.933 --> 00:35:37.850 align:middle line:90% We have finite number. 00:35:37.850 --> 00:35:40.310 align:middle line:84% We've got a finite number of villages. 00:35:40.310 --> 00:35:42.890 align:middle line:84% Let's suppose we have 600 villages over here, 00:35:42.890 --> 00:35:44.820 align:middle line:84% or 1,000 villages, finite number of villages. 00:35:44.820 --> 00:35:53.100 align:middle line:90% 00:35:53.100 --> 00:35:54.460 align:middle line:90% Sorry, what were you saying? 00:35:54.460 --> 00:35:56.252 align:middle line:84% AUDIENCE: Probably for the right-hand side, 00:35:56.252 --> 00:35:58.790 align:middle line:84% are we going to just calculate average [INAUDIBLE]?? 00:35:58.790 --> 00:36:00.500 align:middle line:84% BEN OLKEN: Sorry, you guys speak louder. 00:36:00.500 --> 00:36:01.560 align:middle line:84% AUDIENCE: For the right-hand side, 00:36:01.560 --> 00:36:03.810 align:middle line:84% are we just going to calculate the average differences 00:36:03.810 --> 00:36:05.587 align:middle line:84% between the treatment and control? 00:36:05.587 --> 00:36:07.420 align:middle line:84% Or are we also [INAUDIBLE] the control group 00:36:07.420 --> 00:36:09.920 align:middle line:90% actually [INAUDIBLE]? 00:36:09.920 --> 00:36:11.920 align:middle line:84% BEN OLKEN: So how are we going to estimate this? 00:36:11.920 --> 00:36:24.090 align:middle line:84% So on the left-hand one, we can run y equals alpha 00:36:24.090 --> 00:36:28.620 align:middle line:90% treat plus epsilon. 00:36:28.620 --> 00:36:33.930 align:middle line:84% And on the right-hand one, we could run an IV regression. 00:36:33.930 --> 00:36:36.270 align:middle line:84% Because not everyone on the right-hand one-- 00:36:36.270 --> 00:36:37.830 align:middle line:84% not everyone lotteried into treatment 00:36:37.830 --> 00:36:38.997 align:middle line:90% will ultimately get treated. 00:36:38.997 --> 00:36:41.580 align:middle line:90% So that's a nice case to run IV. 00:36:41.580 --> 00:36:48.900 align:middle line:84% So we can run IV where we can run y equals alpha treat-- 00:36:48.900 --> 00:36:56.870 align:middle line:84% alpha get the treat, get treatment, plus epsilon. 00:36:56.870 --> 00:37:00.260 align:middle line:84% And we can use the lottery as an instrument. 00:37:00.260 --> 00:37:08.010 align:middle line:90% 00:37:08.010 --> 00:37:14.460 align:middle line:84% So my first question is, are these different 00:37:14.460 --> 00:37:16.397 align:middle line:90% in terms of internal validity? 00:37:16.397 --> 00:37:18.480 align:middle line:84% So I think the things we want to think about are-- 00:37:18.480 --> 00:37:22.730 align:middle line:90% 00:37:22.730 --> 00:37:25.760 align:middle line:84% one is, are these internally valid? 00:37:25.760 --> 00:37:30.690 align:middle line:90% 00:37:30.690 --> 00:37:34.230 align:middle line:84% The second is external validity or what 00:37:34.230 --> 00:37:41.950 align:middle line:90% LATE does alpha estimate? 00:37:41.950 --> 00:37:45.950 align:middle line:90% 00:37:45.950 --> 00:37:49.190 align:middle line:90% And three is what's the power? 00:37:49.190 --> 00:37:52.840 align:middle line:90% What's the system power here? 00:37:52.840 --> 00:37:53.636 align:middle line:90% Yeah. 00:37:53.636 --> 00:37:54.628 align:middle line:90% AUDIENCE: [INAUDIBLE]? 00:37:54.628 --> 00:37:59.100 align:middle line:90% 00:37:59.100 --> 00:38:00.220 align:middle line:90% BEN OLKEN: Who knows? 00:38:00.220 --> 00:38:01.890 align:middle line:90% Let's just say it's something. 00:38:01.890 --> 00:38:03.330 align:middle line:90% I don't remember. 00:38:03.330 --> 00:38:05.430 align:middle line:84% Let's say it's something that you can only 00:38:05.430 --> 00:38:09.570 align:middle line:84% investigate if you go to the village, and do something, 00:38:09.570 --> 00:38:12.510 align:middle line:84% and talk to people, and do some stuff. 00:38:12.510 --> 00:38:21.300 align:middle line:84% So say it requires some digging around to figure out. 00:38:21.300 --> 00:38:22.290 align:middle line:90% Yeah, Reika. 00:38:22.290 --> 00:38:25.110 align:middle line:84% AUDIENCE: So I think on two, it kind of 00:38:25.110 --> 00:38:28.620 align:middle line:84% is contingent on whether you think that this policy could 00:38:28.620 --> 00:38:31.230 align:middle line:84% be implemented in the event that a village wasn't interested, 00:38:31.230 --> 00:38:33.605 align:middle line:84% and also what drives the village to be interested or not. 00:38:33.605 --> 00:38:36.750 align:middle line:84% So if you think that it's just going 00:38:36.750 --> 00:38:39.060 align:middle line:84% to be categorically impossible to ever implement 00:38:39.060 --> 00:38:42.180 align:middle line:84% this in a population where people are not 00:38:42.180 --> 00:38:43.950 align:middle line:84% interested in taking up the treatment, 00:38:43.950 --> 00:38:48.660 align:middle line:84% then it's not interesting to you. 00:38:48.660 --> 00:38:51.227 align:middle line:84% What would measuring the effect in that population? 00:38:51.227 --> 00:38:53.310 align:middle line:84% Because it's like you can't implement that policy. 00:38:53.310 --> 00:38:54.227 align:middle line:90% BEN OLKEN: So I agree. 00:38:54.227 --> 00:38:57.868 align:middle line:84% So if you never implement this policy in villages 00:38:57.868 --> 00:39:00.160 align:middle line:84% that we're interested in, then LATE is not interesting, 00:39:00.160 --> 00:39:03.260 align:middle line:84% the LATE of what would be the average human factor 00:39:03.260 --> 00:39:05.280 align:middle line:84% in the entire population is not a relevant LATE. 00:39:05.280 --> 00:39:06.810 align:middle line:84% Well, you're saying something else? 00:39:06.810 --> 00:39:07.980 align:middle line:90% AUDIENCE: Oh, yeah. 00:39:07.980 --> 00:39:09.980 align:middle line:84% BEN OLKEN: But we have another question, though. 00:39:09.980 --> 00:39:12.486 align:middle line:84% So does the left-hand side and the right-hand side-- 00:39:12.486 --> 00:39:14.653 align:middle line:84% do they estimate the same LATE or differently LATEs? 00:39:14.653 --> 00:39:19.090 align:middle line:90% 00:39:19.090 --> 00:39:25.513 align:middle line:84% AUDIENCE: Are you [INAUDIBLE] the population of-- 00:39:25.513 --> 00:39:27.680 align:middle line:84% are you including when villages say that they're not 00:39:27.680 --> 00:39:30.758 align:middle line:90% interested in the treatment? 00:39:30.758 --> 00:39:31.425 align:middle line:90% BEN OLKEN: Here? 00:39:31.425 --> 00:39:33.780 align:middle line:84% AUDIENCE: On the right-hand side [INAUDIBLE] control group. 00:39:33.780 --> 00:39:34.905 align:middle line:90% BEN OLKEN: Yeah, of course. 00:39:34.905 --> 00:39:38.110 align:middle line:84% So how do we estimate-- how do we run this regression? 00:39:38.110 --> 00:39:39.820 align:middle line:84% So how do we run this regression? 00:39:39.820 --> 00:39:43.530 align:middle line:84% So in this regression, let's forget 00:39:43.530 --> 00:39:46.040 align:middle line:84% about the eligibles, whatever, just 00:39:46.040 --> 00:39:47.750 align:middle line:90% to make the life more simple. 00:39:47.750 --> 00:39:51.930 align:middle line:84% I take these people over here, only these people, 00:39:51.930 --> 00:39:55.730 align:middle line:84% and run a regression on treatment and control. 00:39:55.730 --> 00:40:00.900 align:middle line:84% In this one, I have to condition on what I randomize. 00:40:00.900 --> 00:40:04.290 align:middle line:84% So I take all these people and I compare them 00:40:04.290 --> 00:40:09.630 align:middle line:84% to all these people, so I have data, so I do all that. 00:40:09.630 --> 00:40:12.570 align:middle line:90% So that's using the regression. 00:40:12.570 --> 00:40:19.590 align:middle line:84% AUDIENCE: [INAUDIBLE] So they're not measured in the same way? 00:40:19.590 --> 00:40:21.590 align:middle line:84% BEN OLKEN: They're not measured in the same way. 00:40:21.590 --> 00:40:22.370 align:middle line:90% Why? 00:40:22.370 --> 00:40:25.820 align:middle line:84% AUDIENCE: Because first, you've cleaved into-- 00:40:25.820 --> 00:40:31.460 align:middle line:84% in the only villages that are interested and informed, 00:40:31.460 --> 00:40:34.070 align:middle line:84% and then your treated versus control populations [INAUDIBLE] 00:40:34.070 --> 00:40:36.570 align:middle line:84% conditional on being interested and informing. 00:40:36.570 --> 00:40:40.560 align:middle line:84% In the other case, you're looking at all villages, 00:40:40.560 --> 00:40:44.520 align:middle line:84% and then you're just inducing an incentive 00:40:44.520 --> 00:40:48.170 align:middle line:84% to get treated, and comparing that against a control 00:40:48.170 --> 00:40:53.063 align:middle line:84% population that includes people who are not interested. 00:40:53.063 --> 00:40:53.730 align:middle line:90% BEN OLKEN: Yeah. 00:40:53.730 --> 00:40:55.920 align:middle line:84% So it is definitely the case, the sample 00:40:55.920 --> 00:40:57.600 align:middle line:90% here is very different. 00:40:57.600 --> 00:41:01.980 align:middle line:84% This includes all the people who are 00:41:01.980 --> 00:41:04.250 align:middle line:90% interested or not interested. 00:41:04.250 --> 00:41:08.178 align:middle line:90% Does anyone disagree? 00:41:08.178 --> 00:41:08.720 align:middle line:90% You disagree? 00:41:08.720 --> 00:41:09.830 align:middle line:90% Yes, no? 00:41:09.830 --> 00:41:10.790 align:middle line:90% Yeah, Paolo. 00:41:10.790 --> 00:41:13.430 align:middle line:84% AUDIENCE: But then if you do like the classic IV 00:41:13.430 --> 00:41:16.160 align:middle line:84% thing of how many of the people actually took it up, 00:41:16.160 --> 00:41:21.020 align:middle line:84% then you get the LATE exactly for the people who took it up. 00:41:21.020 --> 00:41:22.670 align:middle line:90% BEN OLKEN: Yeah, so say more. 00:41:22.670 --> 00:41:26.240 align:middle line:84% AUDIENCE: So if you take the average from-- 00:41:26.240 --> 00:41:28.010 align:middle line:84% for the alternative design, the difference 00:41:28.010 --> 00:41:30.470 align:middle line:84% between the left branch and the right branch, and then 00:41:30.470 --> 00:41:32.330 align:middle line:84% you divide it by the percent of people 00:41:32.330 --> 00:41:35.090 align:middle line:84% who took up the treatment, then you 00:41:35.090 --> 00:41:38.768 align:middle line:84% would get the coefficient for the treatment effect 00:41:38.768 --> 00:41:40.310 align:middle line:84% for the people who decided to take up 00:41:40.310 --> 00:41:41.972 align:middle line:90% the program, or the villages. 00:41:41.972 --> 00:41:42.680 align:middle line:90% BEN OLKEN: Right. 00:41:42.680 --> 00:41:46.700 align:middle line:84% So actually, I agree with you, Paolo. 00:41:46.700 --> 00:41:48.770 align:middle line:84% So they are different samples, but they're going 00:41:48.770 --> 00:41:51.270 align:middle line:90% to get you the exact same LATE. 00:41:51.270 --> 00:41:54.080 align:middle line:84% So why are they going to get you the exact same LATE? 00:41:54.080 --> 00:41:55.850 align:middle line:84% Well, there's this control group, 00:41:55.850 --> 00:41:57.620 align:middle line:84% and I agree with you-- that's going to have lots of people 00:41:57.620 --> 00:41:59.000 align:middle line:90% who aren't interested in there. 00:41:59.000 --> 00:42:03.380 align:middle line:84% But none of those people are going to take it up if offered. 00:42:03.380 --> 00:42:06.540 align:middle line:84% So we've got a bunch of controls who don't do anything. 00:42:06.540 --> 00:42:09.708 align:middle line:84% So who are the people who are in this treatment group-- and you 00:42:09.708 --> 00:42:11.750 align:middle line:84% compare everybody over here, everybody over here. 00:42:11.750 --> 00:42:13.550 align:middle line:84% Who are the ones who are going to get the treatment? 00:42:13.550 --> 00:42:15.800 align:middle line:84% Well, the only ones who are going to get the treatment 00:42:15.800 --> 00:42:19.370 align:middle line:84% are the ones who are interested and eligible. 00:42:19.370 --> 00:42:24.020 align:middle line:84% So over here, we're comparing a change 00:42:24.020 --> 00:42:28.900 align:middle line:84% for the interested and eligible group, and no change 00:42:28.900 --> 00:42:31.670 align:middle line:84% for the not interested or not eligible group, 00:42:31.670 --> 00:42:34.580 align:middle line:84% to some kind of weighted average with the same proportions 00:42:34.580 --> 00:42:36.440 align:middle line:84% of interested and eligible people, 00:42:36.440 --> 00:42:37.947 align:middle line:90% but with no treatment effect. 00:42:37.947 --> 00:42:40.280 align:middle line:84% So the only difference between this group and this group 00:42:40.280 --> 00:42:42.020 align:middle line:84% is the interested and eligible people 00:42:42.020 --> 00:42:46.430 align:middle line:84% get the treatment and the interested and eligible 00:42:46.430 --> 00:42:48.050 align:middle line:84% controls do not get the treatment. 00:42:48.050 --> 00:42:51.040 align:middle line:90% Otherwise, these are the same. 00:42:51.040 --> 00:42:54.212 align:middle line:84% And over here, it's exactly the same set of people. 00:42:54.212 --> 00:42:55.920 align:middle line:84% We're taking the interested and eligible, 00:42:55.920 --> 00:42:58.503 align:middle line:84% and we're comparing the ones who get the treatment to the ones 00:42:58.503 --> 00:42:59.200 align:middle line:90% who don't. 00:42:59.200 --> 00:43:02.942 align:middle line:90% Now, we have a bunch of people. 00:43:02.942 --> 00:43:04.650 align:middle line:84% In this sample, we have a bunch of people 00:43:04.650 --> 00:43:07.308 align:middle line:84% who are in the treatment group who are not eligible and are 00:43:07.308 --> 00:43:08.850 align:middle line:84% not interested who don't get treated, 00:43:08.850 --> 00:43:10.980 align:middle line:84% and with the analogous people in the control. 00:43:10.980 --> 00:43:12.472 align:middle line:84% And that's why we have to use IV. 00:43:12.472 --> 00:43:14.430 align:middle line:84% If you compare just the reduced form estimates, 00:43:14.430 --> 00:43:15.930 align:middle line:84% they would be different because you've got all these people who 00:43:15.930 --> 00:43:16.980 align:middle line:90% aren't doing anything. 00:43:16.980 --> 00:43:22.510 align:middle line:84% But the relevant set of compliers is exactly the same. 00:43:22.510 --> 00:43:23.010 align:middle line:90% Yes. 00:43:23.010 --> 00:43:24.520 align:middle line:84% AUDIENCE: So then is the difference-- 00:43:24.520 --> 00:43:26.687 align:middle line:84% BEN OLKEN: Oh, I'm sorry, let me say one last thing. 00:43:26.687 --> 00:43:29.790 align:middle line:84% Under the assumption that being offered the treatment 00:43:29.790 --> 00:43:32.460 align:middle line:84% and turning it down, like over here, 00:43:32.460 --> 00:43:35.190 align:middle line:84% is no different than being asked over here. 00:43:35.190 --> 00:43:37.350 align:middle line:84% So under that assumption, they are identical. 00:43:37.350 --> 00:43:38.395 align:middle line:90% Sorry, go ahead. 00:43:38.395 --> 00:43:40.020 align:middle line:84% AUDIENCE: OK, so then is the difference 00:43:40.020 --> 00:43:42.780 align:middle line:84% that on the right-hand side, you're 00:43:42.780 --> 00:43:45.780 align:middle line:84% ex ante throwing away a bunch of people 00:43:45.780 --> 00:43:49.150 align:middle line:84% that could have otherwise been in your estimate? 00:43:49.150 --> 00:43:51.210 align:middle line:90% BEN OLKEN: Useless people. 00:43:51.210 --> 00:43:53.980 align:middle line:84% You're throwing away useless people on the right-hand side. 00:43:53.980 --> 00:43:59.310 align:middle line:84% So you're-- I mean, by useless, I mean, obviously-- 00:43:59.310 --> 00:44:01.780 align:middle line:84% sorry-- useless for the point of view 00:44:01.780 --> 00:44:03.030 align:middle line:90% of estimating this regression. 00:44:03.030 --> 00:44:06.360 align:middle line:84% They have many wonderful uses in the world, obviously, 00:44:06.360 --> 00:44:08.490 align:middle line:84% but for the point of view of this regression, 00:44:08.490 --> 00:44:10.560 align:middle line:90% they are just nuisance. 00:44:10.560 --> 00:44:14.010 align:middle line:84% They are people who are in your treatment group 00:44:14.010 --> 00:44:17.640 align:middle line:84% and in your control group, but who are never takers. 00:44:17.640 --> 00:44:20.610 align:middle line:84% So they're never-- they're just in-- but the problem is 00:44:20.610 --> 00:44:22.680 align:middle line:84% in this design, in the control group, 00:44:22.680 --> 00:44:25.170 align:middle line:84% you don't know who the never-takers are. 00:44:25.170 --> 00:44:26.700 align:middle line:90% So we can't drop them. 00:44:26.700 --> 00:44:28.620 align:middle line:84% We have to include them in the treatment group and the control 00:44:28.620 --> 00:44:28.920 align:middle line:90% group. 00:44:28.920 --> 00:44:30.100 align:middle line:90% Sorry what were you saying? 00:44:30.100 --> 00:44:34.770 align:middle line:84% So yes, so what this design allows you to do 00:44:34.770 --> 00:44:37.620 align:middle line:84% is figure out who the never-takers are 00:44:37.620 --> 00:44:39.460 align:middle line:90% and drop them from the sample. 00:44:39.460 --> 00:44:42.570 align:middle line:84% And in this design, you can't do that because I don't know who-- 00:44:42.570 --> 00:44:45.720 align:middle line:84% I know who the never-takers are in the treatment arm, 00:44:45.720 --> 00:44:48.210 align:middle line:84% but I don't know who the comparable never-takers are 00:44:48.210 --> 00:44:49.002 align:middle line:90% in the control arm. 00:44:49.002 --> 00:44:51.627 align:middle line:84% So I have to include all of them in the treatment group and all 00:44:51.627 --> 00:44:54.180 align:middle line:84% of them in the control group for them to be comparable. 00:44:54.180 --> 00:44:54.720 align:middle line:90% Yeah, Aaron. 00:44:54.720 --> 00:44:56.678 align:middle line:84% AUDIENCE: There's one question on this, though, 00:44:56.678 --> 00:44:59.790 align:middle line:84% is, from the standpoint of experimental design 00:44:59.790 --> 00:45:04.183 align:middle line:84% and thinking ex ante, on the right-hand side, 00:45:04.183 --> 00:45:05.850 align:middle line:84% we're doing this randomization before we 00:45:05.850 --> 00:45:08.017 align:middle line:84% talk to any villagers about their level of interest. 00:45:08.017 --> 00:45:12.090 align:middle line:84% So there's a risk that we do this treatment assignment, 00:45:12.090 --> 00:45:14.780 align:middle line:84% and then we go, and there's two villages that are actually 00:45:14.780 --> 00:45:16.140 align:middle line:90% interested in taking that. 00:45:16.140 --> 00:45:17.070 align:middle line:84% BEN OLKEN: Which one, the right-hand side? 00:45:17.070 --> 00:45:18.320 align:middle line:90% AUDIENCE: The right-hand side. 00:45:18.320 --> 00:45:21.660 align:middle line:84% So that's just, I guess separate from this issue 00:45:21.660 --> 00:45:23.040 align:middle line:90% of who are never-takers. 00:45:23.040 --> 00:45:25.500 align:middle line:84% I guess this falls into the bucket of power, 00:45:25.500 --> 00:45:29.160 align:middle line:84% that you're running the risk that, based on how you've 00:45:29.160 --> 00:45:32.990 align:middle line:84% constrained yourself from the start of the experiment, 00:45:32.990 --> 00:45:33.490 align:middle line:90% you might-- 00:45:33.490 --> 00:45:35.570 align:middle line:84% BEN OLKEN: You might not have enough people. 00:45:35.570 --> 00:45:36.260 align:middle line:90% Yes. 00:45:36.260 --> 00:45:36.915 align:middle line:90% Yes. 00:45:36.915 --> 00:45:38.540 align:middle line:84% AUDIENCE: Whereas that's not an issue-- 00:45:38.540 --> 00:45:39.090 align:middle line:90% BEN OLKEN: That's also true. 00:45:39.090 --> 00:45:40.507 align:middle line:84% AUDIENCE: --on the left-hand side. 00:45:40.507 --> 00:45:42.260 align:middle line:90% BEN OLKEN: Yeah. 00:45:42.260 --> 00:45:43.490 align:middle line:90% So is everyone agreed? 00:45:43.490 --> 00:45:44.990 align:middle line:90% Yeah, [INAUDIBLE]. 00:45:44.990 --> 00:45:48.110 align:middle line:84% AUDIENCE: Just like [INAUDIBLE] maybe 00:45:48.110 --> 00:45:52.278 align:middle line:84% like in the right-hand side, maybe 00:45:52.278 --> 00:45:54.320 align:middle line:84% it makes sense because it would be more expensive 00:45:54.320 --> 00:45:56.880 align:middle line:84% and you're including people that you shouldn't include. 00:45:56.880 --> 00:45:59.430 align:middle line:84% But if you're doing the left-hand side, 00:45:59.430 --> 00:46:02.430 align:middle line:84% maybe you should be careful to see how many people [INAUDIBLE] 00:46:02.430 --> 00:46:05.070 align:middle line:84% because maybe giving different [INAUDIBLE] 00:46:05.070 --> 00:46:07.865 align:middle line:84% was only for people that were interested in the [INAUDIBLE].. 00:46:07.865 --> 00:46:08.740 align:middle line:90% BEN OLKEN: No, sorry. 00:46:08.740 --> 00:46:09.960 align:middle line:84% I'm sorry, I don't understand what you're saying. 00:46:09.960 --> 00:46:10.380 align:middle line:90% So sorry. 00:46:10.380 --> 00:46:10.950 align:middle line:90% Say it again. 00:46:10.950 --> 00:46:13.780 align:middle line:84% AUDIENCE: I guess there's no external validity. 00:46:13.780 --> 00:46:17.040 align:middle line:84% And you have to be careful with the first version 00:46:17.040 --> 00:46:19.590 align:middle line:84% because maybe not enough people would be interested. 00:46:19.590 --> 00:46:22.910 align:middle line:84% BEN OLKEN: No, external validity is the same. 00:46:22.910 --> 00:46:24.620 align:middle line:84% So external validity is identical here. 00:46:24.620 --> 00:46:25.970 align:middle line:84% That's what I was saying, you get the same LATE. 00:46:25.970 --> 00:46:28.560 align:middle line:84% So the LATE may not-- going back to your earlier point, 00:46:28.560 --> 00:46:30.500 align:middle line:84% the LATE may not be the LATE of what 00:46:30.500 --> 00:46:34.550 align:middle line:84% would the LATE be if I was able to persuade everyone 00:46:34.550 --> 00:46:35.360 align:middle line:90% to do this program. 00:46:35.360 --> 00:46:37.490 align:middle line:84% But LATE in the left-hand design and LATE in the right-hand 00:46:37.490 --> 00:46:38.060 align:middle line:90% are the same. 00:46:38.060 --> 00:46:39.602 align:middle line:84% AUDIENCE: I think Paolo's saying what 00:46:39.602 --> 00:46:42.350 align:middle line:84% if just very few people in the country are interested. 00:46:42.350 --> 00:46:44.722 align:middle line:84% And then, at least you see that not much is happening 00:46:44.722 --> 00:46:45.680 align:middle line:90% in the right-hand side. 00:46:45.680 --> 00:46:48.260 align:middle line:84% And then on the left-hand side, oh, wow, in these four villages 00:46:48.260 --> 00:46:49.927 align:middle line:84% that were interested, something happens. 00:46:49.927 --> 00:46:50.427 align:middle line:90% But then-- 00:46:50.427 --> 00:46:52.177 align:middle line:84% BEN OLKEN: No, no, but you know that, too. 00:46:52.177 --> 00:46:53.690 align:middle line:84% You have this information over here. 00:46:53.690 --> 00:46:55.448 align:middle line:84% You know what facts people are interested. 00:46:55.448 --> 00:46:56.990 align:middle line:84% AUDIENCE: I guess I'm saying that you 00:46:56.990 --> 00:46:58.470 align:middle line:90% have to take that into account. 00:46:58.470 --> 00:47:00.607 align:middle line:90% And then just like [INAUDIBLE]. 00:47:00.607 --> 00:47:02.940 align:middle line:84% BEN OLKEN: Yes, you certainly have take it into account. 00:47:02.940 --> 00:47:03.710 align:middle line:90% Absolutely. 00:47:03.710 --> 00:47:07.460 align:middle line:84% But that piece of data you get in the same, both designs. 00:47:07.460 --> 00:47:08.240 align:middle line:90% Yeah. 00:47:08.240 --> 00:47:09.950 align:middle line:84% AUDIENCE: So on the right-hand side, 00:47:09.950 --> 00:47:13.310 align:middle line:84% if you were to then go to all the control villages 00:47:13.310 --> 00:47:16.175 align:middle line:84% and inform them, and then be, like, are you interested, 00:47:16.175 --> 00:47:22.100 align:middle line:84% and then just do nothing, do these become the same again? 00:47:22.100 --> 00:47:24.830 align:middle line:84% BEN OLKEN: Well, the problem is in reality, you can't. 00:47:24.830 --> 00:47:26.540 align:middle line:90% It's not quite credible. 00:47:26.540 --> 00:47:29.090 align:middle line:84% If you're going to inform them and then be, like, oh, sorry, 00:47:29.090 --> 00:47:31.020 align:middle line:84% you're not going to get it, the question is-- 00:47:31.020 --> 00:47:32.270 align:middle line:90% you have to be able to-- 00:47:32.270 --> 00:47:34.280 align:middle line:84% you want to basically inform them 00:47:34.280 --> 00:47:36.492 align:middle line:90% in exactly an identical way. 00:47:36.492 --> 00:47:38.950 align:middle line:84% So it's not the same to say, are you interested, and if so, 00:47:38.950 --> 00:47:40.080 align:middle line:84% we're going to do the program today, 00:47:40.080 --> 00:47:42.000 align:middle line:84% versus hypothetically speaking, if we did this program, 00:47:42.000 --> 00:47:43.250 align:middle line:90% would you be interested in it. 00:47:43.250 --> 00:47:44.800 align:middle line:90% Those are different questions. 00:47:44.800 --> 00:47:47.740 align:middle line:84% AUDIENCE: Well, couldn't you ask the one-- 00:47:47.740 --> 00:47:48.550 align:middle line:90% I see. 00:47:48.550 --> 00:47:50.810 align:middle line:84% And then you have to ask the same question to the-- 00:47:50.810 --> 00:47:54.960 align:middle line:84% BEN OLKEN: And that gives you the left-hand side. 00:47:54.960 --> 00:47:58.410 align:middle line:84% AUDIENCE: Wait, if you were asking people 00:47:58.410 --> 00:48:01.560 align:middle line:84% on the right-hand side of the left wing, 00:48:01.560 --> 00:48:04.260 align:middle line:84% hypothetically, if we had this program, would you 00:48:04.260 --> 00:48:09.660 align:middle line:84% be interested, and then also asking the right-hand side. 00:48:09.660 --> 00:48:13.540 align:middle line:90% But it is the same thing. 00:48:13.540 --> 00:48:15.540 align:middle line:84% BEN OLKEN: Yeah, but that is the left-hand side. 00:48:15.540 --> 00:48:16.310 align:middle line:90% AUDIENCE: Yes. 00:48:16.310 --> 00:48:18.522 align:middle line:84% It's the same thing, right, the [INAUDIBLE] 00:48:18.522 --> 00:48:21.090 align:middle line:84% measuring the right wing of the alternative. 00:48:21.090 --> 00:48:23.370 align:middle line:84% BEN OLKEN: If you're doing it exactly the same way, 00:48:23.370 --> 00:48:26.320 align:middle line:90% then that's this design. 00:48:26.320 --> 00:48:27.870 align:middle line:90% OK, so does everyone agree? 00:48:27.870 --> 00:48:30.360 align:middle line:84% So they're both valid internally. 00:48:30.360 --> 00:48:33.780 align:middle line:84% They actually get you the same LATE. 00:48:33.780 --> 00:48:37.670 align:middle line:90% So which is better? 00:48:37.670 --> 00:48:41.482 align:middle line:84% Are they equal or is there a strict one 00:48:41.482 --> 00:48:42.440 align:middle line:90% that's better or worse? 00:48:42.440 --> 00:48:44.470 align:middle line:90% Someone else? 00:48:44.470 --> 00:48:45.070 align:middle line:90% Somebody else? 00:48:45.070 --> 00:48:46.480 align:middle line:84% Nobody else I haven't heard from? 00:48:46.480 --> 00:48:47.230 align:middle line:90% What do you think? 00:48:47.230 --> 00:48:51.520 align:middle line:90% 00:48:51.520 --> 00:48:52.020 align:middle line:90% Yeah. 00:48:52.020 --> 00:48:53.145 align:middle line:84% AUDIENCE: I mean, my intuition is 00:48:53.145 --> 00:48:54.895 align:middle line:84% if we are estimating the exact same thing, 00:48:54.895 --> 00:48:57.000 align:middle line:84% we would rather have less than IV. 00:48:57.000 --> 00:48:58.315 align:middle line:90% BEN OLKEN: Why? 00:48:58.315 --> 00:48:59.940 align:middle line:84% AUDIENCE: IV standard errors are always 00:48:59.940 --> 00:49:02.600 align:middle line:84% much larger because they're fighting 00:49:02.600 --> 00:49:03.980 align:middle line:90% two estimates by each other. 00:49:03.980 --> 00:49:05.900 align:middle line:90% So like [INAUDIBLE]. 00:49:05.900 --> 00:49:10.133 align:middle line:90% So I don't have [INAUDIBLE]. 00:49:10.133 --> 00:49:10.800 align:middle line:90% BEN OLKEN: Yeah. 00:49:10.800 --> 00:49:12.830 align:middle line:90% [INAUDIBLE] 00:49:12.830 --> 00:49:14.925 align:middle line:84% AUDIENCE: I was in [INAUDIBLE] power. 00:49:14.925 --> 00:49:15.550 align:middle line:90% BEN OLKEN: Yes. 00:49:15.550 --> 00:49:16.690 align:middle line:90% So yes, you're right. 00:49:16.690 --> 00:49:19.810 align:middle line:84% So yes, your power is going to be greater 00:49:19.810 --> 00:49:21.460 align:middle line:90% in this one and this one. 00:49:21.460 --> 00:49:28.340 align:middle line:84% And one reasonable way to think about your power 00:49:28.340 --> 00:49:31.850 align:middle line:90% is that your power-- 00:49:31.850 --> 00:49:33.770 align:middle line:84% you can think about your statistical power 00:49:33.770 --> 00:49:37.850 align:middle line:84% as like the t-test on the reduced form. 00:49:37.850 --> 00:49:40.010 align:middle line:90% And the t-test on the reduced-- 00:49:40.010 --> 00:49:41.600 align:middle line:84% because in some sense, both cases, 00:49:41.600 --> 00:49:43.100 align:middle line:84% you're estimating your reduced form. 00:49:43.100 --> 00:49:44.390 align:middle line:84% And in the IV, you take the reduced form 00:49:44.390 --> 00:49:45.320 align:middle line:84% and divide it by the first [INAUDIBLE].. 00:49:45.320 --> 00:49:46.580 align:middle line:84% But in some sense, your power is never 00:49:46.580 --> 00:49:48.377 align:middle line:84% going to be better than the reduced form. 00:49:48.377 --> 00:49:49.835 align:middle line:84% Your statistical precision is never 00:49:49.835 --> 00:49:51.508 align:middle line:90% going to be better than your-- 00:49:51.508 --> 00:49:54.050 align:middle line:84% your t-stat is never going to be better than the reduced form 00:49:54.050 --> 00:49:55.070 align:middle line:84% t-stat because you're taking that and then 00:49:55.070 --> 00:49:57.403 align:middle line:84% dividing it by something that's not perfectly estimated. 00:49:57.403 --> 00:50:05.590 align:middle line:84% So which of these has a better t-stat on the reduced form? 00:50:05.590 --> 00:50:12.410 align:middle line:84% Well, here-- sorry, I probably have to write this down. 00:50:12.410 --> 00:50:22.230 align:middle line:90% 00:50:22.230 --> 00:50:28.720 align:middle line:84% OK, so on the right-hand side, let's suppose 00:50:28.720 --> 00:50:31.570 align:middle line:84% that fraction alpha are interested. 00:50:31.570 --> 00:50:40.210 align:middle line:90% 00:50:40.210 --> 00:50:43.570 align:middle line:84% And beta is the treatment effect. 00:50:43.570 --> 00:50:48.860 align:middle line:90% 00:50:48.860 --> 00:50:50.260 align:middle line:90% So what's my t-stat? 00:50:50.260 --> 00:50:51.985 align:middle line:90% My t-stat is going to be-- 00:50:51.985 --> 00:50:54.700 align:middle line:90% 00:50:54.700 --> 00:50:56.860 align:middle line:84% for the right-hand side, I don't care about this 00:50:56.860 --> 00:50:58.445 align:middle line:84% because I'm only restricting my sample 00:50:58.445 --> 00:51:00.047 align:middle line:90% to people are interested. 00:51:00.047 --> 00:51:01.630 align:middle line:84% Sorry, my t-state is going to be rough 00:51:01.630 --> 00:51:06.400 align:middle line:84% and N is the total sample, total number of villages. 00:51:06.400 --> 00:51:09.440 align:middle line:90% 00:51:09.440 --> 00:51:10.480 align:middle line:90% So what's my t-stat? 00:51:10.480 --> 00:51:16.870 align:middle line:84% It's roughly going to be proportional to beta divided 00:51:16.870 --> 00:51:22.090 align:middle line:84% by alpha, the square root of alpha times N. 00:51:22.090 --> 00:51:25.857 align:middle line:84% So your standard error is proportional to the-- so t-stat 00:51:25.857 --> 00:51:27.440 align:middle line:84% is beta divided by the standard error. 00:51:27.440 --> 00:51:29.500 align:middle line:84% Standard error is proportional to the square root 00:51:29.500 --> 00:51:31.640 align:middle line:90% of the sample size. 00:51:31.640 --> 00:51:37.777 align:middle line:84% So my sample size, I've reduced by alpha because I had to-- 00:51:37.777 --> 00:51:40.360 align:middle line:84% [INAUDIBLE] my alpha because all people who aren't interested, 00:51:40.360 --> 00:51:41.410 align:middle line:90% I chucked them out. 00:51:41.410 --> 00:51:45.540 align:middle line:84% So my reduced form t-stat looks something like this. 00:51:45.540 --> 00:51:49.010 align:middle line:84% Now, I'm sorry, this is the left-hand side. 00:51:49.010 --> 00:51:52.390 align:middle line:84% Now, on the right-hand side version, 00:51:52.390 --> 00:51:56.170 align:middle line:84% my t-stat is going to be roughly equal to-- 00:51:56.170 --> 00:51:59.260 align:middle line:84% well, I got everybody in there, but only a fraction alpha 00:51:59.260 --> 00:52:01.780 align:middle line:90% of them have a treatment. 00:52:01.780 --> 00:52:05.200 align:middle line:84% So it's roughly equal to alpha times beta, 00:52:05.200 --> 00:52:08.065 align:middle line:90% and I have the full sample. 00:52:08.065 --> 00:52:11.450 align:middle line:90% 00:52:11.450 --> 00:52:13.040 align:middle line:90% So you can see that this one-- 00:52:13.040 --> 00:52:18.275 align:middle line:90% 00:52:18.275 --> 00:52:19.490 align:middle line:90% did I do this right? 00:52:19.490 --> 00:52:24.133 align:middle line:90% 00:52:24.133 --> 00:52:25.050 align:middle line:90% I did something wrong. 00:52:25.050 --> 00:52:25.585 align:middle line:90% Hold on. 00:52:25.585 --> 00:52:26.085 align:middle line:90% No. 00:52:26.085 --> 00:52:28.690 align:middle line:90% 00:52:28.690 --> 00:52:30.340 align:middle line:84% My standard error is proportional to 1 00:52:30.340 --> 00:52:32.950 align:middle line:84% over the square root of the sample size, sorry. 00:52:32.950 --> 00:52:36.500 align:middle line:84% Bigger sample size means a smaller standard error. 00:52:36.500 --> 00:52:39.210 align:middle line:90% Yeah, there we go. 00:52:39.210 --> 00:52:47.110 align:middle line:90% 00:52:47.110 --> 00:52:49.590 align:middle line:90% Sorry, that's correct. 00:52:49.590 --> 00:52:51.210 align:middle line:90% So what is this? 00:52:51.210 --> 00:52:55.650 align:middle line:84% This is proportional to root alpha 00:52:55.650 --> 00:53:06.690 align:middle line:84% beta times root N. And this one is proportional to alpha 00:53:06.690 --> 00:53:13.530 align:middle line:84% beta times the square root of N. So the t-stat is going 00:53:13.530 --> 00:53:17.550 align:middle line:84% to be much bigger on the left-hand version rather 00:53:17.550 --> 00:53:19.620 align:middle line:90% than the right-hand version. 00:53:19.620 --> 00:53:20.910 align:middle line:90% And why? 00:53:20.910 --> 00:53:25.170 align:middle line:84% It's because of the question of, do I want to reduce my power? 00:53:25.170 --> 00:53:26.880 align:middle line:84% I have the same kind of alpha problem, 00:53:26.880 --> 00:53:29.547 align:middle line:84% and do I want to deal with it by just having a larger standard-- 00:53:29.547 --> 00:53:32.460 align:middle line:84% by reducing the first stage or by reducing with or dealing 00:53:32.460 --> 00:53:33.750 align:middle line:90% with more sample size? 00:53:33.750 --> 00:53:38.220 align:middle line:84% And the answer is, because of this square root of N thing, 00:53:38.220 --> 00:53:40.050 align:middle line:84% your power is going to be left better 00:53:40.050 --> 00:53:43.490 align:middle line:84% if you can deal with it upfront, rather than just 00:53:43.490 --> 00:53:44.490 align:middle line:90% you have to estimate it. 00:53:44.490 --> 00:53:47.040 align:middle line:90% 00:53:47.040 --> 00:53:50.830 align:middle line:90% Is that clear? 00:53:50.830 --> 00:53:51.330 align:middle line:90% No? 00:53:51.330 --> 00:53:52.140 align:middle line:90% [INAUDIBLE], you look confused. 00:53:52.140 --> 00:53:52.770 align:middle line:90% No? 00:53:52.770 --> 00:53:54.460 align:middle line:84% AUDIENCE: Yeah, sorry, why is that? 00:53:54.460 --> 00:53:57.840 align:middle line:84% Why are we comparing alpha N to N? 00:53:57.840 --> 00:54:00.757 align:middle line:84% Wouldn't we have the same sample size? 00:54:00.757 --> 00:54:01.340 align:middle line:90% BEN OLKEN: No. 00:54:01.340 --> 00:54:04.790 align:middle line:84% So in this example, remember I said, 00:54:04.790 --> 00:54:07.170 align:middle line:84% suppose we had 1,000 villages over here. 00:54:07.170 --> 00:54:08.920 align:middle line:84% AUDIENCE: OK, it's not like 1,000 villages 00:54:08.920 --> 00:54:10.570 align:middle line:90% to work with in the experiment. 00:54:10.570 --> 00:54:13.480 align:middle line:84% BEN OLKEN: I have 1,000 villages total that I can work with. 00:54:13.480 --> 00:54:17.155 align:middle line:84% And my question is, do I want to screen them out ex ante 00:54:17.155 --> 00:54:18.280 align:middle line:90% or screen them out ex post? 00:54:18.280 --> 00:54:20.830 align:middle line:84% Same 1,000 villages, and my point 00:54:20.830 --> 00:54:23.980 align:middle line:84% is that it's even better, by the way, if I 00:54:23.980 --> 00:54:26.420 align:middle line:90% have 1,000 villages to survey. 00:54:26.420 --> 00:54:29.372 align:middle line:84% That's way, way better over here. 00:54:29.372 --> 00:54:30.830 align:middle line:84% Because I don't even have to-- then 00:54:30.830 --> 00:54:36.000 align:middle line:84% it's just like beta times root 1,000. 00:54:36.000 --> 00:54:39.380 align:middle line:84% Over here, it's beta times whatever, 00:54:39.380 --> 00:54:44.650 align:middle line:84% alpha root 1,000 or whatever, or alpha beta times root 1,000. 00:54:44.650 --> 00:54:45.510 align:middle line:90% So it's way better. 00:54:45.510 --> 00:54:47.135 align:middle line:84% If I can [INAUDIBLE],, it's way better. 00:54:47.135 --> 00:54:49.140 align:middle line:84% I was making a harder case for myself, 00:54:49.140 --> 00:54:51.390 align:middle line:84% where I only have a fixed number of villages up front. 00:54:51.390 --> 00:54:52.927 align:middle line:84% And my question is-- and so where 00:54:52.927 --> 00:54:55.260 align:middle line:84% there's a cost on the left-hand side of having a smaller 00:54:55.260 --> 00:54:55.968 align:middle line:90% sample over here. 00:54:55.968 --> 00:55:00.000 align:middle line:84% And I'm saying, even then, even with a smaller 00:55:00.000 --> 00:55:02.010 align:middle line:84% sample, a smaller amount of data, 00:55:02.010 --> 00:55:05.200 align:middle line:84% I still want the left-hand design. 00:55:05.200 --> 00:55:05.700 align:middle line:90% Yeah. 00:55:05.700 --> 00:55:07.400 align:middle line:84% AUDIENCE: Can this intuitively be 00:55:07.400 --> 00:55:09.420 align:middle line:84% thought of as on the right-hand side, 00:55:09.420 --> 00:55:14.233 align:middle line:84% you are meaning to estimate alpha, 00:55:14.233 --> 00:55:15.650 align:middle line:84% whereas on the left-hand side, you 00:55:15.650 --> 00:55:16.900 align:middle line:90% don't need to estimate alpha? 00:55:16.900 --> 00:55:18.650 align:middle line:84% BEN OLKEN: No, I think it's not just that. 00:55:18.650 --> 00:55:20.645 align:middle line:90% No, because I know alpha. 00:55:20.645 --> 00:55:23.950 align:middle line:90% 00:55:23.950 --> 00:55:24.610 align:middle line:90% It's not that. 00:55:24.610 --> 00:55:25.163 align:middle line:90% The problem-- 00:55:25.163 --> 00:55:26.080 align:middle line:90% AUDIENCE: You do know? 00:55:26.080 --> 00:55:28.340 align:middle line:84% BEN OLKEN: Yeah, because [INAUDIBLE] treatment. 00:55:28.340 --> 00:55:29.445 align:middle line:90% So alpha just-- 00:55:29.445 --> 00:55:30.820 align:middle line:84% AUDIENCE: On the right-hand side? 00:55:30.820 --> 00:55:32.450 align:middle line:84% BEN OLKEN: No, well, it's statistically the same. 00:55:32.450 --> 00:55:32.950 align:middle line:90% So-- 00:55:32.950 --> 00:55:34.960 align:middle line:90% AUDIENCE: I see, I see. 00:55:34.960 --> 00:55:35.890 align:middle line:90% [INAUDIBLE], right? 00:55:35.890 --> 00:55:38.057 align:middle line:84% BEN OLKEN: Yeah, but I don't think that's the issue. 00:55:38.057 --> 00:55:39.220 align:middle line:90% I think the issue is-- 00:55:39.220 --> 00:55:42.020 align:middle line:84% I think really where this comes from is on the right-hand side, 00:55:42.020 --> 00:55:43.673 align:middle line:84% I don't know who the compliers are 00:55:43.673 --> 00:55:45.340 align:middle line:84% and who the never-takers are, and I have 00:55:45.340 --> 00:55:46.660 align:middle line:90% to lump them all in together. 00:55:46.660 --> 00:55:48.220 align:middle line:84% Whereas on the left-hand side, I know 00:55:48.220 --> 00:55:49.150 align:middle line:84% who the never-takers are for sure, 00:55:49.150 --> 00:55:51.127 align:middle line:84% and I just get rid of them for my sample. 00:55:51.127 --> 00:55:52.960 align:middle line:84% And that's where I think the real efficiency 00:55:52.960 --> 00:55:53.877 align:middle line:90% gains are coming from. 00:55:53.877 --> 00:55:56.740 align:middle line:90% 00:55:56.740 --> 00:55:59.330 align:middle line:90% OK, Clear? 00:55:59.330 --> 00:56:05.550 align:middle line:84% So that's why it is always a good idea, if you're ever 00:56:05.550 --> 00:56:08.220 align:middle line:84% doing this, to do as much screening 00:56:08.220 --> 00:56:11.610 align:middle line:84% as you can before the randomization 00:56:11.610 --> 00:56:17.520 align:middle line:84% so that your first stage conditional on randomizing 00:56:17.520 --> 00:56:19.140 align:middle line:84% people to treatment and control-- 00:56:19.140 --> 00:56:21.750 align:middle line:84% your first stage is as close to 1 as possible. 00:56:21.750 --> 00:56:26.440 align:middle line:84% It's always better to do that screening ex ante, 00:56:26.440 --> 00:56:28.063 align:middle line:90% or in general, it is better. 00:56:28.063 --> 00:56:29.980 align:middle line:84% Clearly, I'm talking a lot because my computer 00:56:29.980 --> 00:56:31.540 align:middle line:90% keeps turning off. 00:56:31.540 --> 00:56:35.900 align:middle line:84% So my final question that I wanted to ask, 00:56:35.900 --> 00:56:42.232 align:middle line:84% and as I said, writing exam questions that people seem 00:56:42.232 --> 00:56:43.690 align:middle line:84% confused about is a good diagnostic 00:56:43.690 --> 00:56:46.120 align:middle line:84% for things that should be taught in lecture. 00:56:46.120 --> 00:56:47.230 align:middle line:90% I try to avoid doing that. 00:56:47.230 --> 00:56:49.750 align:middle line:84% But every once in a while it happens. 00:56:49.750 --> 00:56:52.027 align:middle line:84% My final question is, is there a reason 00:56:52.027 --> 00:56:53.860 align:middle line:84% that you would prefer the right-hand design? 00:56:53.860 --> 00:56:55.443 align:middle line:84% And I will try to get my computer back 00:56:55.443 --> 00:56:58.192 align:middle line:84% so it can show you the right-hand design in a second. 00:56:58.192 --> 00:57:00.400 align:middle line:84% So I just told you all the reasons the left-hand side 00:57:00.400 --> 00:57:02.170 align:middle line:84% design is better because it gets you 00:57:02.170 --> 00:57:03.970 align:middle line:84% the same internal validity, the same external validity, 00:57:03.970 --> 00:57:04.960 align:middle line:90% but much better power. 00:57:04.960 --> 00:57:07.090 align:middle line:84% Is there a reason you prefer the left-hand design? 00:57:07.090 --> 00:57:09.275 align:middle line:90% Yeah, Paolo, what do you think? 00:57:09.275 --> 00:57:10.150 align:middle line:90% No, prefer the right. 00:57:10.150 --> 00:57:11.380 align:middle line:84% Is there a reason you prefer the right? 00:57:11.380 --> 00:57:12.130 align:middle line:90% Sorry. 00:57:12.130 --> 00:57:14.170 align:middle line:84% AUDIENCE: You do half as much screening 00:57:14.170 --> 00:57:16.780 align:middle line:84% and so your cost will basically be half. 00:57:16.780 --> 00:57:17.950 align:middle line:90% BEN OLKEN: Yeah, exactly. 00:57:17.950 --> 00:57:20.410 align:middle line:84% So if it turns out that the screening 00:57:20.410 --> 00:57:25.020 align:middle line:84% step is super duper expensive, then you 00:57:25.020 --> 00:57:28.070 align:middle line:90% might want the left-hand side. 00:57:28.070 --> 00:57:32.510 align:middle line:84% Because here on the left, I've got to screen everybody. 00:57:32.510 --> 00:57:35.640 align:middle line:84% And on the right, I only screen the treatment group. 00:57:35.640 --> 00:57:38.578 align:middle line:84% So if it turns out that that step is-- so 00:57:38.578 --> 00:57:41.120 align:middle line:84% the other thing you have to do in practicing different design 00:57:41.120 --> 00:57:43.120 align:middle line:84% is you have different costs at different levels. 00:57:43.120 --> 00:57:48.170 align:middle line:84% So there's a screening cost and there's a data collection cost. 00:57:48.170 --> 00:57:52.250 align:middle line:84% And sometimes, your data collection 00:57:52.250 --> 00:57:54.250 align:middle line:84% is really expensive, like I'm going door to door 00:57:54.250 --> 00:57:55.125 align:middle line:90% and running a survey. 00:57:55.125 --> 00:57:56.160 align:middle line:90% That's really expensive. 00:57:56.160 --> 00:57:59.340 align:middle line:84% Sometimes, your data collection is free, 00:57:59.340 --> 00:58:02.130 align:middle line:84% like if you're getting admin data and you just get all of it 00:58:02.130 --> 00:58:02.940 align:middle line:90% for free. 00:58:02.940 --> 00:58:03.900 align:middle line:90% There's no cost there. 00:58:03.900 --> 00:58:05.858 align:middle line:84% And actually, maybe screening people is costly. 00:58:05.858 --> 00:58:12.202 align:middle line:84% So in practice, if the total number N here is maybe large, 00:58:12.202 --> 00:58:14.160 align:middle line:84% but these different steps have different costs, 00:58:14.160 --> 00:58:16.202 align:middle line:84% that you have to think through the costs in this. 00:58:16.202 --> 00:58:19.580 align:middle line:84% And that's, I think, the one caveat I would make to this. 00:58:19.580 --> 00:58:20.240 align:middle line:90% Clear? 00:58:20.240 --> 00:58:21.930 align:middle line:90% Questions? 00:58:21.930 --> 00:58:23.620 align:middle line:84% OK, all right, so that's mostly what 00:58:23.620 --> 00:58:25.370 align:middle line:84% I want to say about the experiment design. 00:58:25.370 --> 00:58:29.660 align:middle line:84% I thought this was a nice example 00:58:29.660 --> 00:58:34.670 align:middle line:84% OK, back to land reform, back to the land titling, substantive. 00:58:34.670 --> 00:58:35.705 align:middle line:90% So what do they show? 00:58:35.705 --> 00:58:38.580 align:middle line:90% 00:58:38.580 --> 00:58:40.580 align:middle line:84% So the first stage-- so even here, by the way, 00:58:40.580 --> 00:58:43.880 align:middle line:90% their first stage is not-- 00:58:43.880 --> 00:58:51.452 align:middle line:84% or the impact on clear borders or whatever is not 1. 00:58:51.452 --> 00:58:52.910 align:middle line:84% So going through this whole process 00:58:52.910 --> 00:58:55.430 align:middle line:84% only increases it by 1/3 or whatever, or by 1/4. 00:58:55.430 --> 00:58:55.960 align:middle line:90% That's fine. 00:58:55.960 --> 00:58:57.380 align:middle line:90% It's whatever. 00:58:57.380 --> 00:59:00.890 align:middle line:84% But then, they actually do look like they 00:59:00.890 --> 00:59:03.680 align:middle line:84% find some impacts on things on tree 00:59:03.680 --> 00:59:06.050 align:middle line:90% planting and perennial crops. 00:59:06.050 --> 00:59:10.100 align:middle line:84% And the idea is these are things with very long time horizons. 00:59:10.100 --> 00:59:12.560 align:middle line:90% So I plant a tree. 00:59:12.560 --> 00:59:15.350 align:middle line:84% I need to know that A, I'm going to still own this land when 00:59:15.350 --> 00:59:18.597 align:middle line:84% it's time to collect the crops from the tree, and B, no one 00:59:18.597 --> 00:59:20.180 align:middle line:84% is going to come and cut down my tree. 00:59:20.180 --> 00:59:23.150 align:middle line:84% And so that requires pretty good property rights 00:59:23.150 --> 00:59:23.900 align:middle line:90% and clear borders. 00:59:23.900 --> 00:59:26.480 align:middle line:90% And so same thing with-- 00:59:26.480 --> 00:59:28.320 align:middle line:84% perennial crops come back year after year. 00:59:28.320 --> 00:59:34.240 align:middle line:84% So these things suggest that the clear property rights actually 00:59:34.240 --> 00:59:37.960 align:middle line:84% do change my time horizon, and that they lead people to take 00:59:37.960 --> 00:59:39.280 align:middle line:90% these longer term investments. 00:59:39.280 --> 00:59:43.030 align:middle line:84% And so that shows one type of inefficiency 00:59:43.030 --> 00:59:46.740 align:middle line:90% from having unclear borders. 00:59:46.740 --> 00:59:48.920 align:middle line:90% OK, clear? 00:59:48.920 --> 00:59:51.500 align:middle line:90% Questions? 00:59:51.500 --> 00:59:54.220 align:middle line:90% All right. 00:59:54.220 --> 00:59:57.930 align:middle line:84% So that was number one, substantively [INAUDIBLE].. 00:59:57.930 --> 01:00:00.522 align:middle line:84% This was investment increases with good property rights 01:00:00.522 --> 01:00:01.980 align:middle line:84% because of this time horizon point. 01:00:01.980 --> 01:00:04.903 align:middle line:84% Investments are I pay a fixed cost now, hopefully 01:00:04.903 --> 01:00:05.820 align:middle line:90% returns in the future. 01:00:05.820 --> 01:00:08.130 align:middle line:84% If I own the thing, I'm going to capture those returns. 01:00:08.130 --> 01:00:09.900 align:middle line:84% But even if I sell it, I still capture the returns 01:00:09.900 --> 01:00:10.650 align:middle line:90% in the sale price. 01:00:10.650 --> 01:00:12.317 align:middle line:84% So I make these longer term investments. 01:00:12.317 --> 01:00:13.610 align:middle line:90% Great. 01:00:13.610 --> 01:00:20.270 align:middle line:84% Second question is, do I have to do inefficient actions in order 01:00:20.270 --> 01:00:24.470 align:middle line:84% to protect my property rights if they are less clear? 01:00:24.470 --> 01:00:26.350 align:middle line:84% So if my property rights are bad, 01:00:26.350 --> 01:00:28.000 align:middle line:84% maybe I have to do things inefficiently 01:00:28.000 --> 01:00:30.520 align:middle line:84% to protect the property rights or to give me better property 01:00:30.520 --> 01:00:33.650 align:middle line:84% rights, I can not take those inefficient actions. 01:00:33.650 --> 01:00:35.930 align:middle line:90% So here are some examples. 01:00:35.930 --> 01:00:39.840 align:middle line:84% So this is one from Ghana, where people have weak property 01:00:39.840 --> 01:00:41.250 align:middle line:90% rights on plots. 01:00:41.250 --> 01:00:44.683 align:middle line:84% There are competing claims to the land, whatever, 01:00:44.683 --> 01:00:46.350 align:middle line:84% there's a lineage thing and the village. 01:00:46.350 --> 01:00:48.923 align:middle line:84% And it's never quite clear, as they argue in the paper, 01:00:48.923 --> 01:00:51.090 align:middle line:84% whether you have the right to farm this land or not. 01:00:51.090 --> 01:00:52.290 align:middle line:90% It's unclear. 01:00:52.290 --> 01:00:57.548 align:middle line:84% And in particular-- so as you may know, 01:00:57.548 --> 01:00:59.180 align:middle line:84% you want to leave your land fallow 01:00:59.180 --> 01:01:02.660 align:middle line:84% or not farm it every so many years to allow the land 01:01:02.660 --> 01:01:04.460 align:middle line:90% to regenerate itself. 01:01:04.460 --> 01:01:08.157 align:middle line:84% But the problem is, if I'm not actively farming the land, 01:01:08.157 --> 01:01:09.740 align:middle line:84% someone else may say, well, that land, 01:01:09.740 --> 01:01:10.580 align:middle line:90% you're not using that land. 01:01:10.580 --> 01:01:11.747 align:middle line:90% I'm going to take that land. 01:01:11.747 --> 01:01:15.650 align:middle line:84% And so in this context, the idea is that fallowing your land 01:01:15.650 --> 01:01:19.940 align:middle line:84% is a risky thing to do if you have insecure property rights. 01:01:19.940 --> 01:01:25.460 align:middle line:84% And so the idea here is, what they want to say, 01:01:25.460 --> 01:01:30.420 align:middle line:84% is people who have power in this relationship, or in this area. 01:01:30.420 --> 01:01:32.880 align:middle line:84% They are the ones who are less worried about their property 01:01:32.880 --> 01:01:33.060 align:middle line:90% rights. 01:01:33.060 --> 01:01:34.310 align:middle line:90% It's not a land titling paper. 01:01:34.310 --> 01:01:37.560 align:middle line:84% It just says in this event of who-- 01:01:37.560 --> 01:01:39.612 align:middle line:84% if I'm really powerful in the village, 01:01:39.612 --> 01:01:41.070 align:middle line:84% I'm not really worried you're going 01:01:41.070 --> 01:01:43.528 align:middle line:84% to take my land because I can boot you out if you did that. 01:01:43.528 --> 01:01:46.590 align:middle line:84% But if I'm not very powerful, then that's a real risk for me. 01:01:46.590 --> 01:01:51.490 align:middle line:84% And what they argue is that the people who are not powerful 01:01:51.490 --> 01:01:55.735 align:middle line:84% don't fallow their land as much for this reason. 01:01:55.735 --> 01:01:57.235 align:middle line:84% So that's the argument of the paper. 01:01:57.235 --> 01:01:59.920 align:middle line:90% 01:01:59.920 --> 01:02:08.170 align:middle line:84% So they show, for example, that office holders 01:02:08.170 --> 01:02:09.040 align:middle line:90% fallow their land. 01:02:09.040 --> 01:02:14.460 align:middle line:84% Inherited people fallow their land. 01:02:14.460 --> 01:02:17.320 align:middle line:84% And so I'm not going to go through that 01:02:17.320 --> 01:02:18.130 align:middle line:90% in a lot of detail. 01:02:18.130 --> 01:02:19.330 align:middle line:84% I think the question you want to think about 01:02:19.330 --> 01:02:20.140 align:middle line:84% if you're thinking about this paper 01:02:20.140 --> 01:02:21.370 align:middle line:84% is, what is this correlated with? 01:02:21.370 --> 01:02:22.390 align:middle line:84% And what are the [INAUDIBLE] variables? 01:02:22.390 --> 01:02:24.010 align:middle line:90% And let me not go into that. 01:02:24.010 --> 01:02:25.885 align:middle line:84% If you're interested, you can read the paper. 01:02:25.885 --> 01:02:28.180 align:middle line:84% But the point is, this is one potential example 01:02:28.180 --> 01:02:32.430 align:middle line:84% of a kind of inefficient action people could take. 01:02:32.430 --> 01:02:34.320 align:middle line:84% Another one that I want to talk about 01:02:34.320 --> 01:02:40.090 align:middle line:84% in a little bit more detail is this paper by Field in Peru. 01:02:40.090 --> 01:02:42.500 align:middle line:90% This is a land titling paper. 01:02:42.500 --> 01:02:45.820 align:middle line:84% So in this period, basically, they rapidly 01:02:45.820 --> 01:02:47.810 align:middle line:90% start issuing titles. 01:02:47.810 --> 01:02:52.370 align:middle line:84% And these are actually urban, going to be urban households. 01:02:52.370 --> 01:02:55.060 align:middle line:84% So you can think of people in kind of less formal areas, 01:02:55.060 --> 01:02:56.810 align:middle line:84% maybe in slum areas, who don't really have 01:02:56.810 --> 01:02:57.770 align:middle line:90% clear title over their lands. 01:02:57.770 --> 01:03:00.062 align:middle line:84% They're going to go through and title all these places. 01:03:00.062 --> 01:03:03.800 align:middle line:84% And they want to say, does just the act of getting that title 01:03:03.800 --> 01:03:05.480 align:middle line:84% and having that clear property rights 01:03:05.480 --> 01:03:08.840 align:middle line:84% mean that you no longer have to stick around the house 01:03:08.840 --> 01:03:10.405 align:middle line:90% to protect your house? 01:03:10.405 --> 01:03:11.405 align:middle line:90% That's kind of the idea. 01:03:11.405 --> 01:03:14.510 align:middle line:90% 01:03:14.510 --> 01:03:18.100 align:middle line:84% And it's going to be a different diff, basically. 01:03:18.100 --> 01:03:19.630 align:middle line:90% So in this paper-- 01:03:19.630 --> 01:03:23.280 align:middle line:90% 01:03:23.280 --> 01:03:31.550 align:middle line:84% it's basically like areas that get the title-- diff in diff. 01:03:31.550 --> 01:03:33.975 align:middle line:84% Areas get the titling program versus areas 01:03:33.975 --> 01:03:36.350 align:middle line:84% that don't get the titling program or haven't had it yet, 01:03:36.350 --> 01:03:37.970 align:middle line:84% and people who had titles to begin 01:03:37.970 --> 01:03:40.190 align:middle line:84% with who shouldn't be affected versus the ones who 01:03:40.190 --> 01:03:42.120 align:middle line:90% were untitled to begin with. 01:03:42.120 --> 01:03:43.730 align:middle line:84% So that's kind of the diff in diff. 01:03:43.730 --> 01:03:47.480 align:middle line:84% So lo and behold, the titling program 01:03:47.480 --> 01:03:51.980 align:middle line:84% gives you a title, probability 0.78. 01:03:51.980 --> 01:03:55.710 align:middle line:84% That says the program kind of worked. 01:03:55.710 --> 01:04:02.677 align:middle line:84% And it reduces your risk of eviction or property invasion 01:04:02.677 --> 01:04:04.510 align:middle line:84% substantially, at least your perceived risk. 01:04:04.510 --> 01:04:07.260 align:middle line:84% So it looks like people are updating about their risks 01:04:07.260 --> 01:04:10.740 align:middle line:90% based on titles, and so forth. 01:04:10.740 --> 01:04:11.880 align:middle line:90% Or do I consider it secure? 01:04:11.880 --> 01:04:12.490 align:middle line:90% That goes up. 01:04:12.490 --> 01:04:13.865 align:middle line:84% So in general, it seems like this 01:04:13.865 --> 01:04:15.990 align:middle line:84% is changing people's security about their houses. 01:04:15.990 --> 01:04:19.050 align:middle line:84% And the question is, what does this do to their labor market 01:04:19.050 --> 01:04:20.040 align:middle line:90% outcomes? 01:04:20.040 --> 01:04:22.470 align:middle line:84% And in particular, what you can see is they-- 01:04:22.470 --> 01:04:24.060 align:middle line:84% so what happens is they basically 01:04:24.060 --> 01:04:27.780 align:middle line:84% they end up more in the labor force. 01:04:27.780 --> 01:04:29.940 align:middle line:84% But I think what's more interesting is why. 01:04:29.940 --> 01:04:32.988 align:middle line:84% And in particular, these kind of things 01:04:32.988 --> 01:04:35.280 align:middle line:84% are, I think, actually, to me, the two most interesting 01:04:35.280 --> 01:04:35.780 align:middle line:90% results. 01:04:35.780 --> 01:04:40.050 align:middle line:84% They're am I using my residence for my economic activity? 01:04:40.050 --> 01:04:42.450 align:middle line:84% Do I kind of have to work from home 01:04:42.450 --> 01:04:45.210 align:middle line:84% because I need to be home to protect the house? 01:04:45.210 --> 01:04:47.280 align:middle line:90% That goes down with titling. 01:04:47.280 --> 01:04:50.040 align:middle line:84% And do people have really long commutes or are 01:04:50.040 --> 01:04:52.860 align:middle line:84% they kind of going really far away for work? 01:04:52.860 --> 01:04:54.480 align:middle line:90% That actually goes up. 01:04:54.480 --> 01:04:56.730 align:middle line:84% These are in urban states where people aren't actually 01:04:56.730 --> 01:04:57.630 align:middle line:90% selling their land. 01:04:57.630 --> 01:05:00.480 align:middle line:84% And this is not a case where the title let me sell the house 01:05:00.480 --> 01:05:01.650 align:middle line:90% and move somewhere else. 01:05:01.650 --> 01:05:03.897 align:middle line:84% What this case is doing is saying, 01:05:03.897 --> 01:05:05.730 align:middle line:84% there are good opportunities really far away 01:05:05.730 --> 01:05:10.590 align:middle line:84% in this big metro area, but I have to leave my house alone 01:05:10.590 --> 01:05:12.270 align:middle line:84% for the day and be confident it's going 01:05:12.270 --> 01:05:13.560 align:middle line:90% to be there when I come back. 01:05:13.560 --> 01:05:17.070 align:middle line:84% And they show that that goes up with better titles. 01:05:17.070 --> 01:05:18.628 align:middle line:84% And so the implication of that, also, 01:05:18.628 --> 01:05:20.670 align:middle line:84% is that the titling was causing an inefficiency-- 01:05:20.670 --> 01:05:23.070 align:middle line:84% the lack of titling was causing inefficiency 01:05:23.070 --> 01:05:24.240 align:middle line:90% because people weren't able. 01:05:24.240 --> 01:05:26.282 align:middle line:84% These were jobs they clearly prefer, might reveal 01:05:26.282 --> 01:05:29.810 align:middle line:84% a preference, but weren't able to do because they were worried 01:05:29.810 --> 01:05:31.060 align:middle line:90% about protecting their houses. 01:05:31.060 --> 01:05:31.560 align:middle line:90% Yeah. 01:05:31.560 --> 01:05:36.850 align:middle line:84% AUDIENCE: [INAUDIBLE] we studied land [INAUDIBLE].. 01:05:36.850 --> 01:05:39.612 align:middle line:84% Is [INAUDIBLE] global data very critical to this kind 01:05:39.612 --> 01:05:41.338 align:middle line:90% of thing [INAUDIBLE]? 01:05:41.338 --> 01:05:45.050 align:middle line:90% 01:05:45.050 --> 01:05:47.240 align:middle line:84% BEN OLKEN: Yeah, so it's a power question. 01:05:47.240 --> 01:05:52.357 align:middle line:84% So could you run-- suppose this thing was randomized 01:05:52.357 --> 01:05:54.440 align:middle line:84% so we didn't have to worry about pre-trends or any 01:05:54.440 --> 01:05:58.430 align:middle line:84% of that stuff, so we just had titled areas, non-titled areas. 01:05:58.430 --> 01:06:00.783 align:middle line:84% If you had enough of them, you could just 01:06:00.783 --> 01:06:03.200 align:middle line:84% compare in general, what happens in the titled areas, what 01:06:03.200 --> 01:06:04.533 align:middle line:90% happens in the non-titled areas. 01:06:04.533 --> 01:06:09.228 align:middle line:90% 01:06:09.228 --> 01:06:09.770 align:middle line:90% I'm thinking. 01:06:09.770 --> 01:06:12.920 align:middle line:90% 01:06:12.920 --> 01:06:14.660 align:middle line:84% Actually, the ideal thing, I think, 01:06:14.660 --> 01:06:18.770 align:middle line:84% would be to say, either at the house level or the person 01:06:18.770 --> 01:06:22.550 align:middle line:84% level, and the place where it's a little complicated 01:06:22.550 --> 01:06:23.803 align:middle line:90% is if you have-- 01:06:23.803 --> 01:06:25.220 align:middle line:84% so you either want to track what's 01:06:25.220 --> 01:06:27.230 align:middle line:84% happening to houses as kind of a unit, 01:06:27.230 --> 01:06:29.510 align:middle line:84% or what's happening to people kind of as a unit. 01:06:29.510 --> 01:06:33.450 align:middle line:84% What you might not want to do is track people in location. 01:06:33.450 --> 01:06:36.500 align:middle line:84% So you might not want to kind of do a cross-sectional ex post 01:06:36.500 --> 01:06:38.960 align:middle line:84% interview of who happens to be in that house at the time 01:06:38.960 --> 01:06:41.210 align:middle line:84% because that could be-- because people could swap back 01:06:41.210 --> 01:06:41.990 align:middle line:90% and forth between the places. 01:06:41.990 --> 01:06:43.430 align:middle line:90% So you see what I'm saying? 01:06:43.430 --> 01:06:45.270 align:middle line:84% So I think what you really want to do is-- 01:06:45.270 --> 01:06:47.570 align:middle line:90% AUDIENCE: [INAUDIBLE]. 01:06:47.570 --> 01:06:50.390 align:middle line:84% BEN OLKEN: Yeah, I mean, it's a thing, like what happens-- 01:06:50.390 --> 01:06:53.180 align:middle line:84% you're asking questions about the land. 01:06:53.180 --> 01:06:56.675 align:middle line:84% And that's-- yes, I mean, with the caveat the whole city is 01:06:56.675 --> 01:06:58.050 align:middle line:84% operating in general equilibrium. 01:06:58.050 --> 01:07:00.920 align:middle line:84% So an even better case is if you have isolated cities that 01:07:00.920 --> 01:07:02.310 align:middle line:90% are separate from each other. 01:07:02.310 --> 01:07:05.345 align:middle line:84% So if you're in a world where the city is in-- 01:07:05.345 --> 01:07:07.850 align:middle line:90% 01:07:07.850 --> 01:07:11.490 align:middle line:84% so for people, I think you're in good shape, ex ante people 01:07:11.490 --> 01:07:15.050 align:middle line:84% tracking them over time, that's fine. 01:07:15.050 --> 01:07:19.455 align:middle line:84% If the-- no, I guess if the city is in equilibrium, 01:07:19.455 --> 01:07:20.580 align:middle line:90% that's fine, too, actually. 01:07:20.580 --> 01:07:23.132 align:middle line:84% You're going to get what is the-- 01:07:23.132 --> 01:07:25.090 align:middle line:84% I think, the city actually, that's probably OK. 01:07:25.090 --> 01:07:26.770 align:middle line:84% Because in the pre-period where they're not titled, 01:07:26.770 --> 01:07:28.687 align:middle line:84% you're going to get-- you're going to read off 01:07:28.687 --> 01:07:31.760 align:middle line:84% what is the difference of the marginal return 01:07:31.760 --> 01:07:33.010 align:middle line:90% to this property being titled. 01:07:33.010 --> 01:07:33.970 align:middle line:90% So that's still valid. 01:07:33.970 --> 01:07:37.895 align:middle line:84% But yes, you do, I think, want to think through. 01:07:37.895 --> 01:07:39.520 align:middle line:84% If you're asking equilibrium questions, 01:07:39.520 --> 01:07:42.310 align:middle line:84% what are the equilibrium [INAUDIBLE]?? 01:07:42.310 --> 01:07:45.093 align:middle line:90% 01:07:45.093 --> 01:07:45.760 align:middle line:90% Other questions? 01:07:45.760 --> 01:07:46.305 align:middle line:90% Yeah. 01:07:46.305 --> 01:07:47.680 align:middle line:84% AUDIENCE: In this kind of things, 01:07:47.680 --> 01:07:49.810 align:middle line:84% why don't you see more [INAUDIBLE] 01:07:49.810 --> 01:07:51.045 align:middle line:90% insurance structures? 01:07:51.045 --> 01:07:51.920 align:middle line:90% BEN OLKEN: More what? 01:07:51.920 --> 01:07:53.295 align:middle line:84% AUDIENCE: More insurance company, 01:07:53.295 --> 01:07:56.180 align:middle line:84% maybe pricing, maybe provided by the powerful individuals 01:07:56.180 --> 01:07:56.680 align:middle line:90% to provide-- 01:07:56.680 --> 01:07:57.555 align:middle line:90% BEN OLKEN: Insurance. 01:07:57.555 --> 01:07:59.940 align:middle line:90% AUDIENCE: Yeah. 01:07:59.940 --> 01:08:01.447 align:middle line:84% [INAUDIBLE] if you'll go away, I'll 01:08:01.447 --> 01:08:03.530 align:middle line:84% guarantee you that no one's going to take over it. 01:08:03.530 --> 01:08:06.337 align:middle line:84% And if they do, I'll give you some money for it. 01:08:06.337 --> 01:08:07.670 align:middle line:90% BEN OLKEN: It's a good question. 01:08:07.670 --> 01:08:09.260 align:middle line:84% So that's actually really interesting. 01:08:09.260 --> 01:08:10.750 align:middle line:90% So I don't know. 01:08:10.750 --> 01:08:13.540 align:middle line:84% I mean, and you do hear about protection payments, 01:08:13.540 --> 01:08:17.262 align:middle line:90% for example, by informal-- 01:08:17.262 --> 01:08:20.800 align:middle line:84% often, they end up being by gangs and things, which 01:08:20.800 --> 01:08:23.515 align:middle line:84% are maybe better able to be in the business. 01:08:23.515 --> 01:08:25.390 align:middle line:84% The problem is the formal sector can't really 01:08:25.390 --> 01:08:27.430 align:middle line:84% do it because you don't have the formal title. 01:08:27.430 --> 01:08:31.312 align:middle line:84% So it becomes really tricky to be, like, what are you-- 01:08:31.312 --> 01:08:32.770 align:middle line:84% doing this thing through the form-- 01:08:32.770 --> 01:08:34.720 align:middle line:84% I think providing this kind of insurance 01:08:34.720 --> 01:08:36.399 align:middle line:84% through the formal sector is difficult 01:08:36.399 --> 01:08:40.950 align:middle line:84% if you don't have title because the formal sector needs 01:08:40.950 --> 01:08:42.880 align:middle line:90% paperwork, basically. 01:08:42.880 --> 01:08:44.340 align:middle line:90% So I don't know of any. 01:08:44.340 --> 01:08:45.399 align:middle line:90% I could be wrong. 01:08:45.399 --> 01:08:48.090 align:middle line:84% I don't know of any examples of the formal sector providing 01:08:48.090 --> 01:08:50.125 align:middle line:84% insurance against evictions for squatter, 01:08:50.125 --> 01:08:51.750 align:middle line:84% for people who don't have formal title. 01:08:51.750 --> 01:08:54.375 align:middle line:84% I could be wrong and that would be super interesting, actually, 01:08:54.375 --> 01:08:55.740 align:middle line:90% if you found examples of that. 01:08:55.740 --> 01:08:59.910 align:middle line:84% I do think there are all these informal institutions, 01:08:59.910 --> 01:09:03.630 align:middle line:84% like gangs, for example, which on the one hand 01:09:03.630 --> 01:09:04.620 align:middle line:90% are just taxing people. 01:09:04.620 --> 01:09:06.180 align:middle line:90% So they're providing that. 01:09:06.180 --> 01:09:08.430 align:middle line:84% So a lot of those payments are just pure straight rent 01:09:08.430 --> 01:09:09.540 align:middle line:90% extraction. 01:09:09.540 --> 01:09:12.359 align:middle line:84% But they sometimes are offering services in exchange for that. 01:09:12.359 --> 01:09:15.210 align:middle line:84% And because they kind of operate more informally, 01:09:15.210 --> 01:09:17.910 align:middle line:84% they can maybe enforce these things kind of more informally. 01:09:17.910 --> 01:09:21.819 align:middle line:84% I don't know of any examples where 01:09:21.819 --> 01:09:24.729 align:middle line:84% informal institutions like that are providing 01:09:24.729 --> 01:09:28.120 align:middle line:84% either good property protection against expropriation 01:09:28.120 --> 01:09:30.375 align:middle line:90% or insurance. 01:09:30.375 --> 01:09:31.750 align:middle line:84% But I think that would be, if you 01:09:31.750 --> 01:09:34.910 align:middle line:84% found one, that would be super interesting to study. 01:09:34.910 --> 01:09:36.439 align:middle line:90% I think it's super cool. 01:09:36.439 --> 01:09:39.660 align:middle line:90% No, you're right, absolutely. 01:09:39.660 --> 01:09:42.525 align:middle line:90% Other questions or comments? 01:09:42.525 --> 01:09:47.000 align:middle line:90% 01:09:47.000 --> 01:09:49.109 align:middle line:90% OK. 01:09:49.109 --> 01:09:51.598 align:middle line:84% All right, so number three in our list of what 01:09:51.598 --> 01:09:53.390 align:middle line:84% are the problems with poor property rights, 01:09:53.390 --> 01:09:54.229 align:middle line:90% is misallocation. 01:09:54.229 --> 01:09:56.050 align:middle line:84% I think somebody asked about that. 01:09:56.050 --> 01:10:06.900 align:middle line:84% So in principle, tradable property rights 01:10:06.900 --> 01:10:08.580 align:middle line:84% should lead to efficiency improvements. 01:10:08.580 --> 01:10:10.520 align:middle line:84% That's kind of the Coase theorem. 01:10:10.520 --> 01:10:12.892 align:middle line:84% As an aside, we have a current grad student 01:10:12.892 --> 01:10:15.350 align:middle line:84% who's actually working on some of this stuff in the context 01:10:15.350 --> 01:10:17.480 align:middle line:84% of fisheries and tradable fishing permits, 01:10:17.480 --> 01:10:19.563 align:middle line:84% exactly kind of asking these kind of Coase theorem 01:10:19.563 --> 01:10:20.960 align:middle line:90% questions. 01:10:20.960 --> 01:10:22.320 align:middle line:90% What about land? 01:10:22.320 --> 01:10:30.170 align:middle line:84% So this is another paper where I'm looking for a great land 01:10:30.170 --> 01:10:31.267 align:middle line:90% Coase theorem paper. 01:10:31.267 --> 01:10:33.350 align:middle line:84% I'm going to tell you about two that are relevant, 01:10:33.350 --> 01:10:37.310 align:middle line:84% but I don't think quite have exactly what we want here. 01:10:37.310 --> 01:10:39.650 align:middle line:84% But let me tell you the two papers 01:10:39.650 --> 01:10:41.252 align:middle line:84% that I know of that think about this. 01:10:41.252 --> 01:10:42.710 align:middle line:84% And also, I should say, by the way, 01:10:42.710 --> 01:10:45.710 align:middle line:84% I'm always open to, oh, Ben, you forgot about this paper. 01:10:45.710 --> 01:10:47.630 align:middle line:84% So if there are papers I should be aware of, 01:10:47.630 --> 01:10:48.688 align:middle line:90% always let me know. 01:10:48.688 --> 01:10:49.980 align:middle line:90% There's a lot to keep track of. 01:10:49.980 --> 01:10:53.673 align:middle line:84% So one paper that is not particularly well known 01:10:53.673 --> 01:10:55.090 align:middle line:84% but I happen to it because I think 01:10:55.090 --> 01:10:58.270 align:middle line:84% it has a very clear question is this paper by Ravallion and Van 01:10:58.270 --> 01:11:00.170 align:middle line:90% de Walle about Vietnam. 01:11:00.170 --> 01:11:03.860 align:middle line:84% So Vietnam was a strict communist country at one point, 01:11:03.860 --> 01:11:08.072 align:middle line:84% and then had a bunch of reforms and relaxed some of that, 01:11:08.072 --> 01:11:10.280 align:middle line:84% and allowed more tradable, more things that look more 01:11:10.280 --> 01:11:11.660 align:middle line:90% like ownership properties. 01:11:11.660 --> 01:11:12.770 align:middle line:84% So in particular, they're going to study 01:11:12.770 --> 01:11:14.630 align:middle line:84% the introduction of private tradable land in Vietnam. 01:11:14.630 --> 01:11:16.520 align:middle line:84% So it used to be the land was allocated by the government, 01:11:16.520 --> 01:11:17.312 align:middle line:90% couldn't be traded. 01:11:17.312 --> 01:11:19.560 align:middle line:90% And then it becomes tradable. 01:11:19.560 --> 01:11:21.270 align:middle line:84% And they want to argue that basically 01:11:21.270 --> 01:11:23.880 align:middle line:84% after privatization, what happens 01:11:23.880 --> 01:11:28.950 align:middle line:84% is the high-productivity farmers end up with more land, which 01:11:28.950 --> 01:11:34.180 align:middle line:84% you would expect in a standard kind of Coase theorem logic. 01:11:34.180 --> 01:11:34.930 align:middle line:90% I'm a good farmer. 01:11:34.930 --> 01:11:35.770 align:middle line:90% You're a bad farmer. 01:11:35.770 --> 01:11:37.510 align:middle line:90% I should buy your land-- 01:11:37.510 --> 01:11:39.040 align:middle line:90% to some point. 01:11:39.040 --> 01:11:41.530 align:middle line:90% If there's kind of concave-- 01:11:41.530 --> 01:11:43.900 align:middle line:84% if land productivity is concave in size, 01:11:43.900 --> 01:11:46.030 align:middle line:84% then I shouldn't necessarily buy all the land. 01:11:46.030 --> 01:11:49.905 align:middle line:84% But on the margin, we should see an allocation of the-- 01:11:49.905 --> 01:11:51.530 align:middle line:84% if we all have the same amount of land, 01:11:51.530 --> 01:11:53.320 align:middle line:84% the marginal productivity of the good farmers 01:11:53.320 --> 01:11:54.540 align:middle line:84% would be lower than the low farmers. 01:11:54.540 --> 01:11:55.300 align:middle line:90% So that's not an equilibrium. 01:11:55.300 --> 01:11:56.950 align:middle line:84% We should have more land to the-- in equilibrium, 01:11:56.950 --> 01:11:59.350 align:middle line:84% where we allocate, we equalize marginal products of land, 01:11:59.350 --> 01:12:01.450 align:middle line:84% should have more land for the higher farmer's 01:12:01.450 --> 01:12:03.580 align:middle line:90% standard [INAUDIBLE]. 01:12:03.580 --> 01:12:04.610 align:middle line:90% So how do they do this? 01:12:04.610 --> 01:12:06.670 align:middle line:90% Well, they do this structurally. 01:12:06.670 --> 01:12:09.765 align:middle line:84% And they say, in the pre-period, they 01:12:09.765 --> 01:12:11.390 align:middle line:84% estimate some characteristics that make 01:12:11.390 --> 01:12:13.670 align:middle line:90% you a more productive farmer. 01:12:13.670 --> 01:12:16.250 align:middle line:84% Then they write down a functional form 01:12:16.250 --> 01:12:20.330 align:middle line:84% where basically, they have this kind of C function, which 01:12:20.330 --> 01:12:26.358 align:middle line:84% is saying that the C function is the productivity 01:12:26.358 --> 01:12:28.400 align:middle line:84% function as a function of how much land you have, 01:12:28.400 --> 01:12:33.710 align:middle line:84% and how much, and the X characteristics. 01:12:33.710 --> 01:12:37.080 align:middle line:84% And they write it down parametric like this in logs, 01:12:37.080 --> 01:12:40.650 align:middle line:84% which means that in levels, these things are complements. 01:12:40.650 --> 01:12:42.030 align:middle line:84% So people with more X should have 01:12:42.030 --> 01:12:44.740 align:middle line:90% more l in that functional form. 01:12:44.740 --> 01:12:48.090 align:middle line:90% 01:12:48.090 --> 01:12:51.860 align:middle line:84% And so if this b coefficient is less than 1, 01:12:51.860 --> 01:12:53.617 align:middle line:90% this is concave in land. 01:12:53.617 --> 01:12:55.700 align:middle line:84% And we get the interior solution, what I basically 01:12:55.700 --> 01:12:58.520 align:middle line:84% just described, that basically, you have the people with better 01:12:58.520 --> 01:13:00.140 align:middle line:84% X's are the ones getting more land, 01:13:00.140 --> 01:13:01.473 align:middle line:90% but they don't get all the land. 01:13:01.473 --> 01:13:04.640 align:middle line:90% 01:13:04.640 --> 01:13:10.150 align:middle line:84% And so they then test whether land allocations increase with 01:13:10.150 --> 01:13:12.830 align:middle line:84% this kind of predicted optimal amount of land, 01:13:12.830 --> 01:13:15.750 align:middle line:84% which is all function of their X's. 01:13:15.750 --> 01:13:18.100 align:middle line:84% I think this is a very structural kind of approach 01:13:18.100 --> 01:13:18.600 align:middle line:90% to this. 01:13:18.600 --> 01:13:22.650 align:middle line:84% It requires getting these X's kind of right. 01:13:22.650 --> 01:13:29.430 align:middle line:84% And it also requires other admitted variables that are 01:13:29.430 --> 01:13:30.760 align:middle line:90% correlated with these X's. 01:13:30.760 --> 01:13:32.295 align:middle line:84% So I don't think this is perfect. 01:13:32.295 --> 01:13:35.230 align:middle line:90% 01:13:35.230 --> 01:13:38.920 align:middle line:84% For example, there could be other things, like for example, 01:13:38.920 --> 01:13:41.100 align:middle line:84% the X's might influence the production function. 01:13:41.100 --> 01:13:43.518 align:middle line:84% If I'm politically connected, maybe I 01:13:43.518 --> 01:13:45.060 align:middle line:84% look like I'm a good farmer because I 01:13:45.060 --> 01:13:50.328 align:middle line:84% can sell at high prices and maybe also get some output. 01:13:50.328 --> 01:13:52.620 align:middle line:84% Other things could be changing besides property rights. 01:13:52.620 --> 01:13:55.427 align:middle line:84% It's kind of just like a diff, essentially. 01:13:55.427 --> 01:13:57.510 align:middle line:84% I mean, it's a different diff by the l star thing, 01:13:57.510 --> 01:13:59.760 align:middle line:84% but there could be other things that are changing that 01:13:59.760 --> 01:14:01.530 align:middle line:90% change the allocation of land. 01:14:01.530 --> 01:14:03.120 align:middle line:90% A big one is actually they're-- 01:14:03.120 --> 01:14:05.910 align:middle line:84% the reason this is the C, this is actually 01:14:05.910 --> 01:14:07.330 align:middle line:90% consumption, not production. 01:14:07.330 --> 01:14:09.040 align:middle line:90% So that's a little weird. 01:14:09.040 --> 01:14:11.170 align:middle line:84% It's not necessarily what you want. 01:14:11.170 --> 01:14:13.150 align:middle line:84% And it's a very functional form of production, 01:14:13.150 --> 01:14:17.435 align:middle line:84% so very functional form dependent. 01:14:17.435 --> 01:14:19.060 align:middle line:84% It's very, very simple functional form. 01:14:19.060 --> 01:14:21.810 align:middle line:84% So I'm not going to go through this in a lot of detail. 01:14:21.810 --> 01:14:23.750 align:middle line:84% But I think that in principle, though, this 01:14:23.750 --> 01:14:26.243 align:middle line:84% is a cool idea to say, look, here's 01:14:26.243 --> 01:14:28.160 align:middle line:84% what the efficient allocation would look like. 01:14:28.160 --> 01:14:31.190 align:middle line:84% We relax property rates over here, not over here. 01:14:31.190 --> 01:14:33.710 align:middle line:84% If we had that extra diff, does the allocation 01:14:33.710 --> 01:14:37.910 align:middle line:84% move more in the direction of more to the high TFP focus? 01:14:37.910 --> 01:14:38.870 align:middle line:90% Yes, [INAUDIBLE]. 01:14:38.870 --> 01:14:41.310 align:middle line:84% AUDIENCE: Would it be a less structural approach 01:14:41.310 --> 01:14:44.120 align:middle line:90% to [INAUDIBLE]? 01:14:44.120 --> 01:14:46.877 align:middle line:84% BEN OLKEN: Well, it's a good question. 01:14:46.877 --> 01:14:48.710 align:middle line:84% Maybe it's not so much that it's structural. 01:14:48.710 --> 01:14:51.293 align:middle line:84% Maybe it's all these data issues and not having a second diff. 01:14:51.293 --> 01:14:52.370 align:middle line:90% Maybe that's the problem. 01:14:52.370 --> 01:14:53.090 align:middle line:90% Maybe you're right. 01:14:53.090 --> 01:14:54.260 align:middle line:84% I mean, at some level, you're right-- you 01:14:54.260 --> 01:14:55.400 align:middle line:90% do have to estimate. 01:14:55.400 --> 01:15:02.310 align:middle line:90% 01:15:02.310 --> 01:15:04.020 align:middle line:84% Well, one thing I could do, for example, 01:15:04.020 --> 01:15:05.395 align:middle line:84% is I could not protect it on X's. 01:15:05.395 --> 01:15:08.173 align:middle line:90% 01:15:08.173 --> 01:15:09.840 align:middle line:84% AUDIENCE: I'm sorry, how does that work? 01:15:09.840 --> 01:15:13.190 align:middle line:90% 01:15:13.190 --> 01:15:15.830 align:middle line:84% BEN OLKEN: OK, fine, it's a very fair question. 01:15:15.830 --> 01:15:20.840 align:middle line:84% So I could estimate individual farmer fixed characteristics 01:15:20.840 --> 01:15:24.530 align:middle line:84% and say, I could just take the production function-- 01:15:24.530 --> 01:15:26.660 align:middle line:84% I could estimate that the farm production function, 01:15:26.660 --> 01:15:29.360 align:middle line:84% and then a farmer fixed effects, and the farmer 01:15:29.360 --> 01:15:33.050 align:middle line:84% fixed effects is kind of your residual TFP. 01:15:33.050 --> 01:15:36.747 align:middle line:84% And then, I could see whether individual farmers who 01:15:36.747 --> 01:15:39.080 align:middle line:84% seem more productive get more land when they free it up. 01:15:39.080 --> 01:15:40.490 align:middle line:90% That would be a less-- 01:15:40.490 --> 01:15:44.390 align:middle line:84% it's still parametric in that I have to partial out other stuff 01:15:44.390 --> 01:15:50.310 align:middle line:84% to get your piece of it, but that would be one way to do it. 01:15:50.310 --> 01:15:53.300 align:middle line:84% So another way to do it would be to-- 01:15:53.300 --> 01:15:55.970 align:middle line:84% it's not perfect, it's got other downsides, 01:15:55.970 --> 01:16:04.450 align:middle line:84% but to say, look, I'm going to estimate 01:16:04.450 --> 01:16:09.420 align:middle line:90% the average productivity. 01:16:09.420 --> 01:16:10.670 align:middle line:90% I could take something simple. 01:16:10.670 --> 01:16:12.260 align:middle line:84% I could do average productivity of land. 01:16:12.260 --> 01:16:12.830 align:middle line:90% Maybe that's the same. 01:16:12.830 --> 01:16:13.730 align:middle line:90% It's the same as the X's. 01:16:13.730 --> 01:16:14.930 align:middle line:90% So maybe that's the answer. 01:16:14.930 --> 01:16:20.300 align:middle line:84% Is it A, better data, B, an extra diff, and C, 01:16:20.300 --> 01:16:22.040 align:middle line:84% I think maybe a richer structure where 01:16:22.040 --> 01:16:25.235 align:middle line:84% I don't load it all into a couple of X's maybe 01:16:25.235 --> 01:16:27.340 align:middle line:90% would be better. 01:16:27.340 --> 01:16:28.960 align:middle line:84% Fair question, though, fair question. 01:16:28.960 --> 01:16:34.500 align:middle line:90% 01:16:34.500 --> 01:16:35.430 align:middle line:90% But what do you see? 01:16:35.430 --> 01:16:44.480 align:middle line:84% The basic point is the land deficit is l star over-- 01:16:44.480 --> 01:16:46.480 align:middle line:84% these are the people that should have more land. 01:16:46.480 --> 01:16:49.680 align:middle line:84% This is like l star over l, optimal land compared 01:16:49.680 --> 01:16:50.460 align:middle line:90% to actual land. 01:16:50.460 --> 01:16:53.290 align:middle line:84% And this is the change in land during the reform period. 01:16:53.290 --> 01:16:56.760 align:middle line:84% So in general, just says during this five-year period, 01:16:56.760 --> 01:16:58.650 align:middle line:84% the people who we predict to get more land 01:16:58.650 --> 01:16:59.800 align:middle line:90% do actually get more land. 01:16:59.800 --> 01:17:05.100 align:middle line:84% So it suggests it's reallocation in the productivity dimension, 01:17:05.100 --> 01:17:07.350 align:middle line:90% with the caveats I mentioned. 01:17:07.350 --> 01:17:11.670 align:middle line:90% 01:17:11.670 --> 01:17:13.800 align:middle line:84% The last thing I want to say for today 01:17:13.800 --> 01:17:17.370 align:middle line:84% is that this Coase process, this reallocation process, 01:17:17.370 --> 01:17:19.343 align:middle line:90% can take a really long time. 01:17:19.343 --> 01:17:20.760 align:middle line:84% And here's a nice example of this. 01:17:20.760 --> 01:17:24.455 align:middle line:84% This is a paper by Bleakley and Ferrie. 01:17:24.455 --> 01:17:26.580 align:middle line:84% It's called, "Land Openings on the Georgia Frontier 01:17:26.580 --> 01:17:28.788 align:middle line:84% and the Coase Theorem in the Short and the Long Run." 01:17:28.788 --> 01:17:32.040 align:middle line:84% So this paper studied land allocations 01:17:32.040 --> 01:17:33.930 align:middle line:90% in the US State of Georgia. 01:17:33.930 --> 01:17:37.140 align:middle line:90% And the frontier in Georgia-- 01:17:37.140 --> 01:17:39.600 align:middle line:84% this was a long time ago if Georgia was the frontier. 01:17:39.600 --> 01:17:41.430 align:middle line:84% For those of you who know, the US 01:17:41.430 --> 01:17:44.010 align:middle line:84% was settled, approximately settled, 01:17:44.010 --> 01:17:47.910 align:middle line:84% certainly from coast to coast, mostly from East to West, 01:17:47.910 --> 01:17:54.870 align:middle line:84% with some caveats, by the colonists and so on. 01:17:54.870 --> 01:17:57.380 align:middle line:84% So the frontier was opened by the US government 01:17:57.380 --> 01:18:01.070 align:middle line:84% to settlers in waves heading west. 01:18:01.070 --> 01:18:04.560 align:middle line:84% And different waves used different parcel sizes. 01:18:04.560 --> 01:18:07.850 align:middle line:84% So at some area, I think they get 40-acre parcels, 01:18:07.850 --> 01:18:10.500 align:middle line:84% and they get 160-acre parcels, or something like that. 01:18:10.500 --> 01:18:14.135 align:middle line:84% And there's a discontinuity of the border between which 01:18:14.135 --> 01:18:15.510 align:middle line:84% of the areas where it was divided 01:18:15.510 --> 01:18:17.550 align:middle line:84% into big plots and the area where it 01:18:17.550 --> 01:18:19.080 align:middle line:90% was divided into small plots. 01:18:19.080 --> 01:18:24.240 align:middle line:84% And they argue that in the long run, if the Coase theorem is 01:18:24.240 --> 01:18:26.880 align:middle line:84% true, the optimal plot sizes should 01:18:26.880 --> 01:18:30.030 align:middle line:84% converge to the same thing on both sides of this border. 01:18:30.030 --> 01:18:32.220 align:middle line:84% There's an optimal farm size in Georgia, 01:18:32.220 --> 01:18:36.790 align:middle line:84% and we should buy and sell farms to get to the optimal farm 01:18:36.790 --> 01:18:37.290 align:middle line:90% size. 01:18:37.290 --> 01:18:40.260 align:middle line:84% So the initial allocation of big farms versus small farms 01:18:40.260 --> 01:18:42.450 align:middle line:84% shouldn't matter in the long run. 01:18:42.450 --> 01:18:45.960 align:middle line:84% And the question is, how long is the long run? 01:18:45.960 --> 01:18:49.930 align:middle line:84% That is the question of this paper. 01:18:49.930 --> 01:18:52.220 align:middle line:84% And the answer is, the long run's about 100 years. 01:18:52.220 --> 01:18:54.840 align:middle line:90% 01:18:54.840 --> 01:19:06.630 align:middle line:84% So they trace the parcel size across this border over time, 01:19:06.630 --> 01:19:11.160 align:middle line:84% and it doesn't really converge till almost 100 years later. 01:19:11.160 --> 01:19:13.335 align:middle line:90% 01:19:13.335 --> 01:19:15.460 align:middle line:84% So I just want to say that the Coase theorem works, 01:19:15.460 --> 01:19:17.320 align:middle line:84% but the Coase theorem is slow, was 01:19:17.320 --> 01:19:19.240 align:middle line:84% the point I wanted to make-- so just 01:19:19.240 --> 01:19:20.847 align:middle line:90% worth keeping in mind, as well. 01:19:20.847 --> 01:19:21.680 align:middle line:90% OK, I'm out of time. 01:19:21.680 --> 01:19:22.460 align:middle line:90% I'll stop here. 01:19:22.460 --> 01:19:26.110 align:middle line:84% I'll just finish up the last few things about land. 01:19:26.110 --> 01:19:29.762 align:middle line:84% The paper I'm going to start with very briefly next time 01:19:29.762 --> 01:19:30.262 align:middle line:90% is-- 01:19:30.262 --> 01:19:35.590 align:middle line:90% 01:19:35.590 --> 01:19:38.087 align:middle line:84% I'll talk a little bit about this Foster and Rosenzweig 01:19:38.087 --> 01:19:39.670 align:middle line:84% paper, if you have a quick look at it. 01:19:39.670 --> 01:19:41.080 align:middle line:90% And then we'll switch to the-- 01:19:41.080 --> 01:19:43.990 align:middle line:84% I think the PF lectures are next, on redistribution, 01:19:43.990 --> 01:19:46.450 align:middle line:90% are the next ones. 01:19:46.450 --> 01:19:48.470 align:middle line:90% So that's what's coming next. 01:19:48.470 --> 01:19:53.010 align:middle line:90% I'll see you guys on Wednesday. 01:19:53.010 --> 01:20:05.000 align:middle line:90%