WEBVTT 00:00:00.000 --> 00:00:01.964 align:middle line:90% [CREAKING] 00:00:01.964 --> 00:00:04.419 align:middle line:90% [RUSTLING] 00:00:04.419 --> 00:00:07.856 align:middle line:90% [CLICKING] 00:00:07.856 --> 00:00:10.330 align:middle line:90% 00:00:10.330 --> 00:00:14.140 align:middle line:84% SIGI ZHENG: There's a thinking approach that you need to grasp. 00:00:14.140 --> 00:00:22.420 align:middle line:84% So I know I'm not as funny and as smart as Juan and Azizi. 00:00:22.420 --> 00:00:25.430 align:middle line:84% So I need to make sure you like me. 00:00:25.430 --> 00:00:30.820 align:middle line:84% So I brought some food from Trader Joe's for today. 00:00:30.820 --> 00:00:37.060 align:middle line:84% And it's a way to make this very, very technical lecture 00:00:37.060 --> 00:00:38.410 align:middle line:90% a little bit funny. 00:00:38.410 --> 00:00:40.190 align:middle line:90% So that's about today. 00:00:40.190 --> 00:00:42.340 align:middle line:84% So today, we will talk about-- we already 00:00:42.340 --> 00:00:45.620 align:middle line:84% talked about green buildings and healthy buildings. 00:00:45.620 --> 00:00:50.800 align:middle line:84% Now we're going to talk about how to quantitatively measure 00:00:50.800 --> 00:00:52.660 align:middle line:84% the benefit of the green building 00:00:52.660 --> 00:00:56.530 align:middle line:84% or the or the green city and how to do a model. 00:00:56.530 --> 00:00:58.250 align:middle line:90% But don't be so worried. 00:00:58.250 --> 00:01:01.900 align:middle line:84% And this model is so simple, and I'm sure you can understand. 00:01:01.900 --> 00:01:04.000 align:middle line:90% So just to quickly recap-- 00:01:04.000 --> 00:01:06.350 align:middle line:84% the quickly recap is the following. 00:01:06.350 --> 00:01:12.940 align:middle line:84% So for the entire this course, we talk about the business case. 00:01:12.940 --> 00:01:14.510 align:middle line:90% We talk about business case. 00:01:14.510 --> 00:01:18.610 align:middle line:84% Our point is you cannot [INAUDIBLE] for people, 00:01:18.610 --> 00:01:21.130 align:middle line:84% for the planet with three bottom lines-- 00:01:21.130 --> 00:01:25.910 align:middle line:84% and very, very important bottom line, the circle, is profit. 00:01:25.910 --> 00:01:27.640 align:middle line:84% So we need to understand whether there 00:01:27.640 --> 00:01:30.880 align:middle line:84% is a profit in the market that can incentivize 00:01:30.880 --> 00:01:34.840 align:middle line:84% the different stakeholders to build and supply, 00:01:34.840 --> 00:01:38.960 align:middle line:84% to own and to occupy the green and healthy buildings. 00:01:38.960 --> 00:01:40.180 align:middle line:90% So that's the cash flow-- 00:01:40.180 --> 00:01:41.410 align:middle line:90% I don't need to go through. 00:01:41.410 --> 00:01:46.270 align:middle line:84% But I want to run through this-- in my first and second green 00:01:46.270 --> 00:01:49.150 align:middle line:84% building lecture, I just showed you some papers-- 00:01:49.150 --> 00:01:51.590 align:middle line:90% scholarly work, some literature. 00:01:51.590 --> 00:01:56.000 align:middle line:84% That's a literature to show there's a green premium. 00:01:56.000 --> 00:01:59.020 align:middle line:84% And here is the tenants-- they are 00:01:59.020 --> 00:02:02.660 align:middle line:84% willing to pay higher rent or the buyers, 00:02:02.660 --> 00:02:06.110 align:middle line:84% they are willing to pay higher price. 00:02:06.110 --> 00:02:08.000 align:middle line:90% And then I show you this. 00:02:08.000 --> 00:02:11.840 align:middle line:84% Remember the rental premium of the green buildings? 00:02:11.840 --> 00:02:14.180 align:middle line:84% So I say these are all the studies, 00:02:14.180 --> 00:02:16.260 align:middle line:90% and Siqi's study is here-- 00:02:16.260 --> 00:02:20.360 align:middle line:84% and from all the studies, I give you a takeaway message, 00:02:20.360 --> 00:02:24.800 align:middle line:84% which is a significant rent premium of the green buildings, 00:02:24.800 --> 00:02:26.160 align:middle line:90% about 6%. 00:02:26.160 --> 00:02:27.980 align:middle line:90% And there's some distribution. 00:02:27.980 --> 00:02:29.060 align:middle line:90% There's a range. 00:02:29.060 --> 00:02:32.300 align:middle line:84% So how did these scholars-- how did 00:02:32.300 --> 00:02:35.090 align:middle line:84% these researchers got to this conclusion 00:02:35.090 --> 00:02:37.860 align:middle line:90% of the 6% of the premium? 00:02:37.860 --> 00:02:40.490 align:middle line:90% That will be our today's focus. 00:02:40.490 --> 00:02:44.690 align:middle line:84% And we also saw some this price premium. 00:02:44.690 --> 00:02:48.290 align:middle line:84% This price premium say, oh, these-- all the papers. 00:02:48.290 --> 00:02:52.820 align:middle line:84% And the price premium overall is some negative, some positive, 00:02:52.820 --> 00:02:54.710 align:middle line:84% some significant, some not significant, 00:02:54.710 --> 00:02:59.480 align:middle line:84% but overall is 7.6% So that's basically 00:02:59.480 --> 00:03:03.110 align:middle line:84% today's topic is how to understand this premium 00:03:03.110 --> 00:03:05.920 align:middle line:90% and how to quantify. 00:03:05.920 --> 00:03:08.360 align:middle line:84% So our outlines are following today-- 00:03:08.360 --> 00:03:11.530 align:middle line:84% first, I will still talk about the green building as a starting 00:03:11.530 --> 00:03:14.350 align:middle line:84% point, but since this is the methodology, 00:03:14.350 --> 00:03:16.720 align:middle line:84% this can apply later to all kinds of things-- 00:03:16.720 --> 00:03:18.910 align:middle line:84% healthy building and all climate resilience, 00:03:18.910 --> 00:03:20.560 align:middle line:90% all kind of things you want-- 00:03:20.560 --> 00:03:22.390 align:middle line:90% how to quantify the premium. 00:03:22.390 --> 00:03:25.010 align:middle line:84% And then talk about some cities-- 00:03:25.010 --> 00:03:27.010 align:middle line:84% if you understand the green building part, 00:03:27.010 --> 00:03:28.690 align:middle line:90% the green city is very easy. 00:03:28.690 --> 00:03:30.670 align:middle line:84% And I will talk about some little bit 00:03:30.670 --> 00:03:34.280 align:middle line:84% how to apply that to business and policy decision-making 00:03:34.280 --> 00:03:37.210 align:middle line:90% if we can estimate that premium. 00:03:37.210 --> 00:03:43.990 align:middle line:84% First, has anyone heard of this hedonic pricing model? 00:03:43.990 --> 00:03:47.990 align:middle line:84% Raise your hand if you have heard of this terminology, 00:03:47.990 --> 00:03:48.670 align:middle line:90% "hedonic." 00:03:48.670 --> 00:03:52.240 align:middle line:84% Have you ever used that in your research-- 00:03:52.240 --> 00:03:55.900 align:middle line:84% or not research, like p-set exercise. 00:03:55.900 --> 00:04:02.014 align:middle line:84% OK, Wilson, what kind of p-sets you did in Wisconsin-Madison? 00:04:02.014 --> 00:04:04.050 align:middle line:90% AUDIENCE: No, I [INAUDIBLE]. 00:04:04.050 --> 00:04:07.880 align:middle line:90% 00:04:07.880 --> 00:04:10.900 align:middle line:84% SIGI ZHENG: From your research assistant work with me. 00:04:10.900 --> 00:04:12.800 align:middle line:90% OK, good to know. 00:04:12.800 --> 00:04:14.690 align:middle line:90% So this is my-- 00:04:14.690 --> 00:04:19.000 align:middle line:84% I use hedonic pricing as a way to make a living, so my papers-- 00:04:19.000 --> 00:04:20.390 align:middle line:90% so that's why. 00:04:20.390 --> 00:04:24.200 align:middle line:84% And so here, after today's lecture, 00:04:24.200 --> 00:04:28.040 align:middle line:84% you need to understand this, and you should be able to use that. 00:04:28.040 --> 00:04:30.410 align:middle line:90% So that's the point. 00:04:30.410 --> 00:04:33.680 align:middle line:84% But, first of all, I want to give you a big picture of this. 00:04:33.680 --> 00:04:36.860 align:middle line:84% So basically, when we want to say, OK, now 00:04:36.860 --> 00:04:40.340 align:middle line:84% we want to understand there's some benefit. 00:04:40.340 --> 00:04:43.740 align:middle line:90% This benefit is not market good. 00:04:43.740 --> 00:04:47.750 align:middle line:84% If this is a market good, like a coffee, like a computer, 00:04:47.750 --> 00:04:51.390 align:middle line:84% we can observe its price easily to form the market. 00:04:51.390 --> 00:04:55.670 align:middle line:84% But if this is nonmarket goods, like clean air-- 00:04:55.670 --> 00:04:57.170 align:middle line:90% nonmarket goods-- 00:04:57.170 --> 00:05:02.660 align:middle line:84% so this carbon, until it's priced, is nonmarket goods. 00:05:02.660 --> 00:05:04.220 align:middle line:90% Global warming-- nonmarket. 00:05:04.220 --> 00:05:09.670 align:middle line:84% You cannot really observe the price of those nonmarket goods. 00:05:09.670 --> 00:05:11.340 align:middle line:90% So there are two ways. 00:05:11.340 --> 00:05:15.390 align:middle line:84% There are two ways to derive, indirectly, 00:05:15.390 --> 00:05:17.890 align:middle line:90% the people's willingness to pay. 00:05:17.890 --> 00:05:22.350 align:middle line:84% So that's the first terminology I want you to understand. 00:05:22.350 --> 00:05:23.670 align:middle line:90% This is willing to pay-- 00:05:23.670 --> 00:05:24.400 align:middle line:90% WPP. 00:05:24.400 --> 00:05:26.622 align:middle line:84% We will see a lot of these words today. 00:05:26.622 --> 00:05:28.205 align:middle line:84% That's why I need the blackboard here. 00:05:28.205 --> 00:05:33.360 align:middle line:90% 00:05:33.360 --> 00:05:37.530 align:middle line:84% Consumers', no matter is buyers or tenants, 00:05:37.530 --> 00:05:40.110 align:middle line:84% willingness to pay for something. 00:05:40.110 --> 00:05:44.220 align:middle line:84% So there are two ways to derive this if you cannot observe from 00:05:44.220 --> 00:05:45.270 align:middle line:90% the market. 00:05:45.270 --> 00:05:47.730 align:middle line:90% One is called stated preference. 00:05:47.730 --> 00:05:49.620 align:middle line:84% The other is revealed preference. 00:05:49.620 --> 00:05:52.530 align:middle line:84% To help you memorize this-- stated preference 00:05:52.530 --> 00:05:54.460 align:middle line:90% is basically talking. 00:05:54.460 --> 00:05:55.300 align:middle line:90% You talk. 00:05:55.300 --> 00:05:56.950 align:middle line:90% You speak. 00:05:56.950 --> 00:05:58.900 align:middle line:90% And reveal is doing. 00:05:58.900 --> 00:06:00.300 align:middle line:90% One is talking. 00:06:00.300 --> 00:06:03.110 align:middle line:90% The other is doing. 00:06:03.110 --> 00:06:05.210 align:middle line:84% So today, we will focus on the doing part. 00:06:05.210 --> 00:06:07.760 align:middle line:84% But I want to give you a sense of the talking part. 00:06:07.760 --> 00:06:11.080 align:middle line:90% 00:06:11.080 --> 00:06:14.230 align:middle line:84% So basically, this short video gives you 00:06:14.230 --> 00:06:17.860 align:middle line:84% a sense of this willingness to pay by talking. 00:06:17.860 --> 00:06:21.080 align:middle line:84% So I have done this for many times for my own research. 00:06:21.080 --> 00:06:24.850 align:middle line:84% For example, I send students to survey these residents 00:06:24.850 --> 00:06:27.050 align:middle line:84% about their willingness to pay for a green building. 00:06:27.050 --> 00:06:29.230 align:middle line:90% So that's kind of-- 00:06:29.230 --> 00:06:31.360 align:middle line:84% I think, basically in your thesis later, 00:06:31.360 --> 00:06:34.000 align:middle line:84% maybe you can also apply this method 00:06:34.000 --> 00:06:36.380 align:middle line:84% if you want to do a survey and ask. 00:06:36.380 --> 00:06:39.610 align:middle line:84% But today, this is not our focus because I just 00:06:39.610 --> 00:06:43.390 align:middle line:84% mentioned in that short video, this talking 00:06:43.390 --> 00:06:46.750 align:middle line:84% is not reliable for many, many reasons. 00:06:46.750 --> 00:06:49.930 align:middle line:84% If the respondent says that, oh, you 00:06:49.930 --> 00:06:52.900 align:middle line:84% ask me-- willing to pay for any public policy 00:06:52.900 --> 00:06:54.850 align:middle line:84% to reduce air pollution or all the things, 00:06:54.850 --> 00:06:57.580 align:middle line:84% and later, you will use as evidence to increase tax, 00:06:57.580 --> 00:07:00.760 align:middle line:84% then they have a tendency to lower their willingness 00:07:00.760 --> 00:07:04.400 align:middle line:84% to pay if they think they can just be free riders. 00:07:04.400 --> 00:07:06.070 align:middle line:84% So they will increase their willingness 00:07:06.070 --> 00:07:09.970 align:middle line:84% to pay as a pressure to others to get this done. 00:07:09.970 --> 00:07:13.780 align:middle line:84% Now housing market or real estate market 00:07:13.780 --> 00:07:18.060 align:middle line:84% really provide us a much better opportunity for this 00:07:18.060 --> 00:07:19.000 align:middle line:90% just by doing. 00:07:19.000 --> 00:07:20.420 align:middle line:90% This is hedonic. 00:07:20.420 --> 00:07:23.450 align:middle line:84% So it's not about talking, but we observe. 00:07:23.450 --> 00:07:25.570 align:middle line:84% We observe from the real estate market 00:07:25.570 --> 00:07:28.630 align:middle line:84% of the real transactions-- price, rent. 00:07:28.630 --> 00:07:31.360 align:middle line:84% And you start to back out people's willingness 00:07:31.360 --> 00:07:34.970 align:middle line:84% to pay for market or nonmarket attributes. 00:07:34.970 --> 00:07:37.460 align:middle line:90% So that's a point. 00:07:37.460 --> 00:07:40.920 align:middle line:84% So why housing market provide us a better opportunity? 00:07:40.920 --> 00:07:43.160 align:middle line:84% That's because I think all of you 00:07:43.160 --> 00:07:47.900 align:middle line:84% understand that housing market, when you buy a house is not 00:07:47.900 --> 00:07:49.970 align:middle line:90% just the structure itself. 00:07:49.970 --> 00:07:51.300 align:middle line:90% It's not the concrete. 00:07:51.300 --> 00:07:54.030 align:middle line:84% It's not just the columns and the walls and the windows. 00:07:54.030 --> 00:07:57.680 align:middle line:84% No, you are buying a place to live. 00:07:57.680 --> 00:07:59.510 align:middle line:84% And you are buying that neighborhood 00:07:59.510 --> 00:08:02.840 align:middle line:84% that you want to access to many, many opportunities. 00:08:02.840 --> 00:08:04.650 align:middle line:90% So that's the point of housing. 00:08:04.650 --> 00:08:05.900 align:middle line:90% Housing is so important. 00:08:05.900 --> 00:08:07.470 align:middle line:90% That's location based. 00:08:07.470 --> 00:08:11.550 align:middle line:84% And you buy or rent a house for so many other things. 00:08:11.550 --> 00:08:14.040 align:middle line:84% It's not just the rooms or the kitchen 00:08:14.040 --> 00:08:17.780 align:middle line:90% or the walls or the space. 00:08:17.780 --> 00:08:21.350 align:middle line:84% So now give us a way to decompose-- 00:08:21.350 --> 00:08:23.570 align:middle line:84% using the housing transaction price 00:08:23.570 --> 00:08:26.900 align:middle line:84% to decompose why people are buying this house-- particularly 00:08:26.900 --> 00:08:27.810 align:middle line:90% location. 00:08:27.810 --> 00:08:31.070 align:middle line:84% So I give you a very simple example for you 00:08:31.070 --> 00:08:33.299 align:middle line:90% to understand this. 00:08:33.299 --> 00:08:38.820 align:middle line:84% For example, there is a single family house, $1 million. 00:08:38.820 --> 00:08:44.250 align:middle line:84% And actually, why this buyer is buying this house by $1 million? 00:08:44.250 --> 00:08:48.150 align:middle line:84% Because, of course, we have the household has two kids, 00:08:48.150 --> 00:08:50.080 align:middle line:90% and then they need three rooms. 00:08:50.080 --> 00:08:51.100 align:middle line:90% They need three rooms. 00:08:51.100 --> 00:08:52.930 align:middle line:84% They cannot live in a very small place. 00:08:52.930 --> 00:08:54.108 align:middle line:90% So that's three rooms. 00:08:54.108 --> 00:08:54.900 align:middle line:90% That's a component. 00:08:54.900 --> 00:08:56.340 align:middle line:90% That's very tangible. 00:08:56.340 --> 00:08:59.430 align:middle line:84% And then the household also wants 00:08:59.430 --> 00:09:02.110 align:middle line:84% to live in a place very convenient. 00:09:02.110 --> 00:09:05.340 align:middle line:84% So this household wants to pay for this location 00:09:05.340 --> 00:09:08.740 align:middle line:84% because there are two subway stops nearby, for example. 00:09:08.740 --> 00:09:10.410 align:middle line:90% That's a hypothetical example. 00:09:10.410 --> 00:09:12.750 align:middle line:84% And then this is so convenient, then 00:09:12.750 --> 00:09:18.480 align:middle line:84% they are willing to pay 0.1 million for this accessibility 00:09:18.480 --> 00:09:20.180 align:middle line:90% to subway stops. 00:09:20.180 --> 00:09:23.430 align:middle line:84% And then there's such a nice park nearby. 00:09:23.430 --> 00:09:24.980 align:middle line:84% This household really likes parks. 00:09:24.980 --> 00:09:27.260 align:middle line:84% So the [INAUDIBLE] want to buy this location 00:09:27.260 --> 00:09:32.850 align:middle line:84% because $1.1 million to pay for the park and plus all this. 00:09:32.850 --> 00:09:35.410 align:middle line:84% Another thing is this is a green building. 00:09:35.410 --> 00:09:36.980 align:middle line:84% Suppose this household really want 00:09:36.980 --> 00:09:38.820 align:middle line:90% to pay for the green building. 00:09:38.820 --> 00:09:41.330 align:middle line:84% So because of the green building, 00:09:41.330 --> 00:09:43.800 align:middle line:84% they also want to get $1.1 million. 00:09:43.800 --> 00:09:46.520 align:middle line:84% So, of course, this is a few examples-- 00:09:46.520 --> 00:09:50.990 align:middle line:84% four examples of the real space of the house, transportation 00:09:50.990 --> 00:09:54.710 align:middle line:84% accessibility, some environmental park, 00:09:54.710 --> 00:09:56.730 align:middle line:90% and the green building stuff. 00:09:56.730 --> 00:09:59.580 align:middle line:84% So that's the four things I give-- four examples. 00:09:59.580 --> 00:10:01.760 align:middle line:84% However, there are so many other things. 00:10:01.760 --> 00:10:04.500 align:middle line:84% There's a thousands of things related to this house. 00:10:04.500 --> 00:10:05.760 align:middle line:90% We cannot do all this. 00:10:05.760 --> 00:10:08.010 align:middle line:84% Then we always have this last point, 00:10:08.010 --> 00:10:09.840 align:middle line:90% last component called residual. 00:10:09.840 --> 00:10:10.770 align:middle line:90% It's a residual. 00:10:10.770 --> 00:10:11.880 align:middle line:90% It's like a trash can. 00:10:11.880 --> 00:10:15.740 align:middle line:84% It's all the things you cannot observe and you cannot measure, 00:10:15.740 --> 00:10:20.670 align:middle line:84% but you also must value those things go into this [INAUDIBLE]. 00:10:20.670 --> 00:10:24.430 align:middle line:84% So then you clearly can see this is hedonic model. 00:10:24.430 --> 00:10:27.010 align:middle line:90% Hedonic model is decomposition. 00:10:27.010 --> 00:10:32.680 align:middle line:84% And you decompose a very complex thing, which is a house-- 00:10:32.680 --> 00:10:34.530 align:middle line:90% it's a complex commodity-- 00:10:34.530 --> 00:10:40.980 align:middle line:84% into many pieces, a bundle of pieces that you are using money 00:10:40.980 --> 00:10:42.150 align:middle line:90% to buy. 00:10:42.150 --> 00:10:44.580 align:middle line:84% Now I just randomly give you some numbers, 00:10:44.580 --> 00:10:48.870 align:middle line:84% but there is a scientific way to estimate this. 00:10:48.870 --> 00:10:51.210 align:middle line:84% So I just throw all these numbers. 00:10:51.210 --> 00:10:53.580 align:middle line:84% Of course, cannot be doing this-- 00:10:53.580 --> 00:10:57.090 align:middle line:90% scientific way is to estimate. 00:10:57.090 --> 00:10:58.350 align:middle line:90% So how to estimate? 00:10:58.350 --> 00:10:59.910 align:middle line:90% That's the hedonic model. 00:10:59.910 --> 00:11:01.770 align:middle line:90% So that's the following. 00:11:01.770 --> 00:11:04.300 align:middle line:84% 1 million on the left-hand side-- 00:11:04.300 --> 00:11:08.040 align:middle line:84% that's the total price, total price, total willingness to pay. 00:11:08.040 --> 00:11:14.100 align:middle line:84% Then decompose to now four components plus a trashcan. 00:11:14.100 --> 00:11:17.070 align:middle line:84% These four components for each components, 00:11:17.070 --> 00:11:21.540 align:middle line:84% there is a unit price times quantity. 00:11:21.540 --> 00:11:24.850 align:middle line:84% When we buy apples you have unit price and you have quantity. 00:11:24.850 --> 00:11:25.680 align:middle line:90% How many apples? 00:11:25.680 --> 00:11:26.970 align:middle line:90% So that's the same thing. 00:11:26.970 --> 00:11:32.210 align:middle line:84% So basically, for three rooms-- each room, $0.2 million. 00:11:32.210 --> 00:11:35.990 align:middle line:84% Two subway stop-- each subway stop $0.05 million. 00:11:35.990 --> 00:11:41.150 align:middle line:84% Then the LEED is a kind of one thing, one tangible thing. 00:11:41.150 --> 00:11:43.310 align:middle line:84% And then one nice park-- if you have 00:11:43.310 --> 00:11:47.150 align:middle line:84% two nice parks, that will be two parks, but not this one park. 00:11:47.150 --> 00:11:50.120 align:middle line:84% So that's a way to decompose the price-- 00:11:50.120 --> 00:11:54.230 align:middle line:84% total value to 4 components plus a trash 00:11:54.230 --> 00:11:57.750 align:middle line:84% can and each component you have price and quantity. 00:11:57.750 --> 00:11:59.010 align:middle line:90% So that's clear. 00:11:59.010 --> 00:12:00.600 align:middle line:90% So that's the way. 00:12:00.600 --> 00:12:04.980 align:middle line:84% And how you get these numbers This is a model. 00:12:04.980 --> 00:12:07.530 align:middle line:90% This alpha is the unit price. 00:12:07.530 --> 00:12:10.800 align:middle line:90% X is the quantity [INAUDIBLE]. 00:12:10.800 --> 00:12:15.500 align:middle line:90% How to estimate-- you need data. 00:12:15.500 --> 00:12:18.960 align:middle line:84% You cannot just randomly come up with numbers without any data. 00:12:18.960 --> 00:12:20.770 align:middle line:90% You have input and output. 00:12:20.770 --> 00:12:24.250 align:middle line:84% So the way to do this is to do a regression. 00:12:24.250 --> 00:12:26.390 align:middle line:90% [INAUDIBLE] will tell you more. 00:12:26.390 --> 00:12:28.010 align:middle line:84% Yes, they give you cover regression? 00:12:28.010 --> 00:12:30.370 align:middle line:90% No-- next recitation. 00:12:30.370 --> 00:12:33.290 align:middle line:84% Regression-- for example, we cannot just get one house. 00:12:33.290 --> 00:12:36.010 align:middle line:84% If you can only get one house, you cannot do this. 00:12:36.010 --> 00:12:39.760 align:middle line:84% You need thousands, hundreds of thousands of observations. 00:12:39.760 --> 00:12:41.620 align:middle line:90% Now we have our big data. 00:12:41.620 --> 00:12:43.780 align:middle line:90% We have Zillow. 00:12:43.780 --> 00:12:45.950 align:middle line:84% After, we will share with you some Zillow data. 00:12:45.950 --> 00:12:49.750 align:middle line:84% So the Zillow data-- we will get the housing transactions 00:12:49.750 --> 00:12:54.800 align:middle line:84% in seaport area and the downtown Boston area. 00:12:54.800 --> 00:12:58.010 align:middle line:84% Suppose you get 1,000 observations 00:12:58.010 --> 00:12:59.780 align:middle line:90% of the transactions. 00:12:59.780 --> 00:13:03.650 align:middle line:84% Then the blue boxes are the input. 00:13:03.650 --> 00:13:07.760 align:middle line:84% For each transaction, you will have the total value transaction 00:13:07.760 --> 00:13:08.420 align:middle line:90% price. 00:13:08.420 --> 00:13:10.190 align:middle line:90% You will have those quantities. 00:13:10.190 --> 00:13:13.760 align:middle line:84% You'll know each house is three bedrooms, two bedrooms, or one 00:13:13.760 --> 00:13:17.180 align:middle line:84% subway stop, two subway stop, or zero subway stop, 00:13:17.180 --> 00:13:18.960 align:middle line:90% and whether it's LEED or not. 00:13:18.960 --> 00:13:23.850 align:middle line:84% So you have the input of the quantity and the total price. 00:13:23.850 --> 00:13:28.640 align:middle line:84% Then you throw the data into software, which is R, 00:13:28.640 --> 00:13:33.740 align:middle line:84% and then the software will report this coefficient. 00:13:33.740 --> 00:13:38.510 align:middle line:84% This red things are reported from the software where you fit 00:13:38.510 --> 00:13:40.250 align:middle line:90% all the transactions into that. 00:13:40.250 --> 00:13:43.230 align:middle line:90% So that's a regression. 00:13:43.230 --> 00:13:48.380 align:middle line:84% Then the point is from all those [INAUDIBLE] regressions, 00:13:48.380 --> 00:13:51.780 align:middle line:90% you get p, X1, X2, X3, X4. 00:13:51.780 --> 00:13:52.870 align:middle line:90% Then you're wrong. 00:13:52.870 --> 00:13:55.570 align:middle line:84% Then you get alpha 1, alpha 2, alpha 3, alpha 4. 00:13:55.570 --> 00:13:58.530 align:middle line:84% And it also reported a R-squared-- 00:13:58.530 --> 00:13:59.440 align:middle line:90% what's this? 00:13:59.440 --> 00:14:02.555 align:middle line:90% 00:14:02.555 --> 00:14:03.480 align:middle line:90% You know? 00:14:03.480 --> 00:14:05.860 align:middle line:90% OK, please. 00:14:05.860 --> 00:14:08.460 align:middle line:84% AUDIENCE: [INAUDIBLE] the higher the correlation 00:14:08.460 --> 00:14:12.370 align:middle line:84% of the variables [INAUDIBLE] dependent variables [INAUDIBLE] 00:14:12.370 --> 00:14:13.700 align:middle line:90% SIGI ZHENG: Good, yes. 00:14:13.700 --> 00:14:17.620 align:middle line:84% This is called this explanation power of this equation. 00:14:17.620 --> 00:14:21.860 align:middle line:84% So we want this to be as high as possible, but we are not so-- 00:14:21.860 --> 00:14:23.630 align:middle line:90% cannot go to 100%. 00:14:23.630 --> 00:14:28.060 align:middle line:84% So this 83% means these four variables 00:14:28.060 --> 00:14:33.820 align:middle line:84% can explain 83% of the variation among all those observations. 00:14:33.820 --> 00:14:38.350 align:middle line:90% 17% is in the trash can. 00:14:38.350 --> 00:14:42.280 align:middle line:84% So this, basically, 17% is here because other things you cannot 00:14:42.280 --> 00:14:43.000 align:middle line:90% observe. 00:14:43.000 --> 00:14:44.950 align:middle line:84% So that's the point of the hedonic-- 00:14:44.950 --> 00:14:46.000 align:middle line:90% easy. 00:14:46.000 --> 00:14:49.180 align:middle line:84% So that's basically today's point. 00:14:49.180 --> 00:14:52.240 align:middle line:84% And then, of course, there's many, many complexities 00:14:52.240 --> 00:14:52.940 align:middle line:90% behind this. 00:14:52.940 --> 00:14:54.850 align:middle line:90% But this case is here. 00:14:54.850 --> 00:14:58.420 align:middle line:84% And when you do-- the real regression won't be 4, maybe 5, 00:14:58.420 --> 00:14:59.660 align:middle line:90% 6, no matter what-- 00:14:59.660 --> 00:15:03.500 align:middle line:84% and always, when you do this, you are not without purpose. 00:15:03.500 --> 00:15:06.670 align:middle line:84% When you do this, you always have a purpose. 00:15:06.670 --> 00:15:09.430 align:middle line:84% Purpose can, in general, you want to do an exercise 00:15:09.430 --> 00:15:14.160 align:middle line:84% or you are mostly interested in one coefficient. 00:15:14.160 --> 00:15:17.570 align:middle line:84% For example, if I was a researcher to do this, 00:15:17.570 --> 00:15:20.120 align:middle line:90% I want to understand alpha 3. 00:15:20.120 --> 00:15:22.880 align:middle line:84% I want to understand alpha 3 is people's willingness 00:15:22.880 --> 00:15:25.460 align:middle line:90% to pay for a LEED building. 00:15:25.460 --> 00:15:27.800 align:middle line:90% So this willingness to pay-- 00:15:27.800 --> 00:15:31.010 align:middle line:84% and it can be many willing to pay. 00:15:31.010 --> 00:15:34.100 align:middle line:90% Now we have 4-- 00:15:34.100 --> 00:15:36.060 align:middle line:90% 1, 2, 3, 4. 00:15:36.060 --> 00:15:39.830 align:middle line:84% I'm most interested in 3, which is the LEED. 00:15:39.830 --> 00:15:43.140 align:middle line:84% Of course, if I want to get alpha 3, 00:15:43.140 --> 00:15:46.080 align:middle line:84% I must first run this and get all these others. 00:15:46.080 --> 00:15:47.900 align:middle line:90% Then I focus on alpha 3. 00:15:47.900 --> 00:15:50.510 align:middle line:90% Others are byproducts. 00:15:50.510 --> 00:15:53.070 align:middle line:84% For example, there is a transportation researcher-- 00:15:53.070 --> 00:15:54.870 align:middle line:84% Mackenzie is a transportation researcher. 00:15:54.870 --> 00:15:57.050 align:middle line:84% Then she is not interested in alpha 3. 00:15:57.050 --> 00:16:00.080 align:middle line:90% She is interested in alpha 2. 00:16:00.080 --> 00:16:04.130 align:middle line:84% The accessibility to subway stop means what? 00:16:04.130 --> 00:16:09.210 align:middle line:84% So then that later can relate to other policies. 00:16:09.210 --> 00:16:12.300 align:middle line:84% A little bit of algebra, but not so difficult-- 00:16:12.300 --> 00:16:16.620 align:middle line:84% so when just now I see a very intuitive way to decompose-- 00:16:16.620 --> 00:16:20.190 align:middle line:84% when you read papers or any research piece, 00:16:20.190 --> 00:16:23.880 align:middle line:84% you will see such a model in this way that vectors-- 00:16:23.880 --> 00:16:27.120 align:middle line:84% always they take log because it's nonlinear. 00:16:27.120 --> 00:16:34.450 align:middle line:84% Log price as a function of some components-- x1 now is a vector. 00:16:34.450 --> 00:16:38.520 align:middle line:84% This is the physical attributes as to maybe locational 00:16:38.520 --> 00:16:41.370 align:middle line:84% attributes-- maybe each vector has several variables-- 00:16:41.370 --> 00:16:45.510 align:middle line:84% number of bedrooms, number of living rooms, and whether these 00:16:45.510 --> 00:16:49.150 align:middle line:84% are detached or attached, lots of physical things. 00:16:49.150 --> 00:16:54.930 align:middle line:84% So you have a physical basket and some locational basket 00:16:54.930 --> 00:16:56.100 align:middle line:90% like that. 00:16:56.100 --> 00:17:02.070 align:middle line:84% Then [INAUDIBLE] this log, how to interpret the log? 00:17:02.070 --> 00:17:04.470 align:middle line:84% And just now, I show this level-- 00:17:04.470 --> 00:17:05.795 align:middle line:90% is no log thing. 00:17:05.795 --> 00:17:07.170 align:middle line:84% Then, you have the absolute value 00:17:07.170 --> 00:17:09.460 align:middle line:84% of this is-- that's easy to understand. 00:17:09.460 --> 00:17:11.829 align:middle line:84% When you take log on the left-hand side, 00:17:11.829 --> 00:17:14.910 align:middle line:84% how you interpret the coefficient-- 00:17:14.910 --> 00:17:17.514 align:middle line:90% as a small test of your math? 00:17:17.514 --> 00:17:18.990 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:17:18.990 --> 00:17:21.839 align:middle line:84% SIGI ZHENG: Yes, percentage change. 00:17:21.839 --> 00:17:25.290 align:middle line:84% So, for example, you can tell me-- 00:17:25.290 --> 00:17:25.917 align:middle line:90% look at this. 00:17:25.917 --> 00:17:26.875 align:middle line:90% This is a hypothetical. 00:17:26.875 --> 00:17:29.170 align:middle line:90% No price is like this. 00:17:29.170 --> 00:17:34.830 align:middle line:84% And I will ask if, let's say, living rooms, number of living 00:17:34.830 --> 00:17:37.230 align:middle line:84% rooms, if there is an additional one living 00:17:37.230 --> 00:17:41.510 align:middle line:84% room that will increase housing price by how much? 00:17:41.510 --> 00:17:44.098 align:middle line:90% 5%. 00:17:44.098 --> 00:17:45.140 align:middle line:90% AUDIENCE: [INAUDIBLE] 5%. 00:17:45.140 --> 00:17:45.807 align:middle line:90% SIGI ZHENG: Yes. 00:17:45.807 --> 00:17:47.720 align:middle line:84% So this one will become percentages 00:17:47.720 --> 00:17:55.940 align:middle line:84% because we all know that log y basically does this-- 00:17:55.940 --> 00:17:57.140 align:middle line:90% not so basic thing. 00:17:57.140 --> 00:17:58.730 align:middle line:90% Then you take the derivative. 00:17:58.730 --> 00:18:03.710 align:middle line:84% Then this part will become the percentage. 00:18:03.710 --> 00:18:07.320 align:middle line:84% So that's basically start of y divided by y. 00:18:07.320 --> 00:18:08.460 align:middle line:90% That's a percentage. 00:18:08.460 --> 00:18:13.010 align:middle line:84% So this thing will be become percentage, which means 00:18:13.010 --> 00:18:16.770 align:middle line:90% incremental one unit increase. 00:18:16.770 --> 00:18:21.240 align:middle line:84% One actual living room will lead to a 5% increase 00:18:21.240 --> 00:18:22.560 align:middle line:90% in housing price. 00:18:22.560 --> 00:18:26.850 align:middle line:90% And this one green park-- 00:18:26.850 --> 00:18:29.020 align:middle line:90% then it will increase by 5%. 00:18:29.020 --> 00:18:31.170 align:middle line:84% So that's an explanation of this when 00:18:31.170 --> 00:18:33.000 align:middle line:84% we take a log because when we look 00:18:33.000 --> 00:18:35.620 align:middle line:84% at all these empirical studies, that's all in log. 00:18:35.620 --> 00:18:38.250 align:middle line:90% It's not in level. 00:18:38.250 --> 00:18:40.960 align:middle line:90% Here, this is the real paper. 00:18:40.960 --> 00:18:43.290 align:middle line:90% So you can see the real paper-- 00:18:43.290 --> 00:18:46.410 align:middle line:90% that this paper is by this 2009. 00:18:46.410 --> 00:18:50.430 align:middle line:84% And then they say, OK, all other variables, all controlled. 00:18:50.430 --> 00:18:52.060 align:middle line:90% There's so many other things. 00:18:52.060 --> 00:18:54.420 align:middle line:84% And one-- they say, all other variables 00:18:54.420 --> 00:18:59.640 align:middle line:84% controlled-- then one variable called LEED and the other 00:18:59.640 --> 00:19:01.380 align:middle line:90% is called Energy Star. 00:19:01.380 --> 00:19:08.250 align:middle line:84% So remember, if a number of rooms, not a numerical variable, 00:19:08.250 --> 00:19:12.930 align:middle line:84% like one room, two rooms, three rooms-- sometimes, it's a dummy. 00:19:12.930 --> 00:19:16.235 align:middle line:90% It's a dummy variable is dummy-- 00:19:16.235 --> 00:19:19.550 align:middle line:90% 00:19:19.550 --> 00:19:22.880 align:middle line:84% it takes only two values, 0 or 1. 00:19:22.880 --> 00:19:28.330 align:middle line:84% And then, for example, in this example, subway is a dummy. 00:19:28.330 --> 00:19:32.510 align:middle line:84% I say dummy means within one mile of subway buffer. 00:19:32.510 --> 00:19:40.730 align:middle line:84% If this house-- this is a subway, and this is one mile. 00:19:40.730 --> 00:19:42.000 align:middle line:90% The house is here. 00:19:42.000 --> 00:19:44.300 align:middle line:90% Then it takes a value of 1. 00:19:44.300 --> 00:19:47.270 align:middle line:84% If the house is here, it takes a value of 0. 00:19:47.270 --> 00:19:48.730 align:middle line:90% So that's a dummy. 00:19:48.730 --> 00:19:50.780 align:middle line:84% So if the dummy, then the coefficient 00:19:50.780 --> 00:19:53.300 align:middle line:90% means from 0 to 1 change-- 00:19:53.300 --> 00:19:58.310 align:middle line:84% that will incur the housing price increase by 8%. 00:19:58.310 --> 00:19:59.590 align:middle line:90% So that's a dummy. 00:19:59.590 --> 00:20:05.330 align:middle line:84% So, based on this principle, how to Interpret 00:20:05.330 --> 00:20:11.960 align:middle line:90% the LEED coefficient of 0.06? 00:20:11.960 --> 00:20:14.210 align:middle line:90% LEED is the dummy, right? 00:20:14.210 --> 00:20:15.650 align:middle line:90% OK. 00:20:15.650 --> 00:20:18.190 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:20:18.190 --> 00:20:18.962 align:middle line:90% 00:20:18.962 --> 00:20:19.670 align:middle line:90% SIGI ZHENG: Good. 00:20:19.670 --> 00:20:21.950 align:middle line:84% So that's why, from all the empirical studies, 00:20:21.950 --> 00:20:23.990 align:middle line:90% we got all those coefficients. 00:20:23.990 --> 00:20:29.430 align:middle line:84% Then Energy Star similarly say also 6% price premium. 00:20:29.430 --> 00:20:33.260 align:middle line:84% So that's tell you how those papers estimate the price 00:20:33.260 --> 00:20:36.980 align:middle line:84% premium by this hedonic regressions. 00:20:36.980 --> 00:20:38.930 align:middle line:90% The point is decompose-- 00:20:38.930 --> 00:20:41.250 align:middle line:84% they decompose all the housing values. 00:20:41.250 --> 00:20:43.460 align:middle line:84% Then they have all other attributes 00:20:43.460 --> 00:20:48.440 align:middle line:84% and the coefficient of LEED is 6%. 00:20:48.440 --> 00:20:50.230 align:middle line:90% So that's the price premium. 00:20:50.230 --> 00:20:53.740 align:middle line:90% It's not difficult at all. 00:20:53.740 --> 00:21:00.430 align:middle line:84% Then a little bit go back to the Winthrop Center that-- 00:21:00.430 --> 00:21:02.320 align:middle line:90% this is a teaching case. 00:21:02.320 --> 00:21:03.370 align:middle line:90% Remember that? 00:21:03.370 --> 00:21:05.140 align:middle line:90% And, at that time-- 00:21:05.140 --> 00:21:06.880 align:middle line:90% let me see. 00:21:06.880 --> 00:21:08.860 align:middle line:84% You cannot see-- this is a building, right? 00:21:08.860 --> 00:21:11.450 align:middle line:90% The lower part is an office. 00:21:11.450 --> 00:21:12.800 align:middle line:90% The upper part is a condo. 00:21:12.800 --> 00:21:13.990 align:middle line:90% We talk about this a lot. 00:21:13.990 --> 00:21:18.070 align:middle line:84% And we have our guest speaker come in, 00:21:18.070 --> 00:21:22.990 align:middle line:84% and we understood that for the lower part of office, 00:21:22.990 --> 00:21:26.980 align:middle line:84% they chose very, very green, a much higher level of energy 00:21:26.980 --> 00:21:27.640 align:middle line:90% efficiency-- 00:21:27.640 --> 00:21:28.750 align:middle line:90% Passive House. 00:21:28.750 --> 00:21:31.450 align:middle line:84% For the upper level, they didn't because they 00:21:31.450 --> 00:21:34.690 align:middle line:84% need to keep the view, and they don't think that's important. 00:21:34.690 --> 00:21:37.630 align:middle line:84% Of course, I don't think they run the hedonic 00:21:37.630 --> 00:21:39.020 align:middle line:90% when they made this decision. 00:21:39.020 --> 00:21:41.680 align:middle line:84% But we can think through this in our way. 00:21:41.680 --> 00:21:45.790 align:middle line:84% For example, Winthrop Center's team brand team 00:21:45.790 --> 00:21:48.370 align:middle line:90% can collect the data of nearby-- 00:21:48.370 --> 00:21:49.490 align:middle line:90% that's Boston area. 00:21:49.490 --> 00:21:52.660 align:middle line:84% If they can collect data from nearby-- 00:21:52.660 --> 00:21:57.760 align:middle line:84% office market and condo market because Winthrop center 00:21:57.760 --> 00:21:58.970 align:middle line:90% is still under construction. 00:21:58.970 --> 00:22:00.850 align:middle line:84% They don't have their own transaction price, 00:22:00.850 --> 00:22:04.630 align:middle line:84% but they can collect nearby comparable-- these comparable 00:22:04.630 --> 00:22:05.860 align:middle line:90% properties. 00:22:05.860 --> 00:22:09.170 align:middle line:84% For example, they collect it and they run this regression 00:22:09.170 --> 00:22:10.740 align:middle line:90% like the following. 00:22:10.740 --> 00:22:11.240 align:middle line:90% I show this. 00:22:11.240 --> 00:22:14.330 align:middle line:84% Of course, there are so many other things. 00:22:14.330 --> 00:22:19.490 align:middle line:84% And for the upper part, the condo market nearby, 00:22:19.490 --> 00:22:21.760 align:middle line:90% they want this-- they got this. 00:22:21.760 --> 00:22:24.900 align:middle line:84% This I made up-- just want to show the logic-- 00:22:24.900 --> 00:22:26.690 align:middle line:84% and the [INAUDIBLE] is like this. 00:22:26.690 --> 00:22:32.110 align:middle line:84% If they say for the upper part, these condo buyers value 00:22:32.110 --> 00:22:35.140 align:middle line:84% view so much compared to the lower 00:22:35.140 --> 00:22:36.610 align:middle line:84% part because lower part no view-- 00:22:36.610 --> 00:22:38.900 align:middle line:90% remember, ugly buildings nearby. 00:22:38.900 --> 00:22:43.060 align:middle line:84% So this coefficient 0.3, this coefficient 0.01, 00:22:43.060 --> 00:22:46.195 align:middle line:84% which means the people's willing to pay for view 00:22:46.195 --> 00:22:48.840 align:middle line:84% is so high compared to the lower. 00:22:48.840 --> 00:22:53.350 align:middle line:84% However, the lower part of the green and the house, 00:22:53.350 --> 00:22:54.760 align:middle line:84% they have much higher willingness 00:22:54.760 --> 00:22:57.670 align:middle line:90% to pay compared to the condo. 00:22:57.670 --> 00:23:01.930 align:middle line:84% This is the relative magnitude of the coefficient can tell you 00:23:01.930 --> 00:23:05.410 align:middle line:84% what is the comparative advantage of that part 00:23:05.410 --> 00:23:06.820 align:middle line:90% of those attributes. 00:23:06.820 --> 00:23:10.540 align:middle line:84% Then this can justify why they chose 00:23:10.540 --> 00:23:13.480 align:middle line:84% the upper, the view as the comparative advantage 00:23:13.480 --> 00:23:15.580 align:middle line:84% because much higher willingness to pay, 00:23:15.580 --> 00:23:19.570 align:middle line:84% but for lower, they chose green and the house 00:23:19.570 --> 00:23:21.590 align:middle line:90% as the important attributes. 00:23:21.590 --> 00:23:24.190 align:middle line:84% So that's an example of how the market analysis can 00:23:24.190 --> 00:23:31.210 align:middle line:84% use hedonic model to assist their quantitative judgment. 00:23:31.210 --> 00:23:34.040 align:middle line:90% OK, not challenges-- 00:23:34.040 --> 00:23:36.260 align:middle line:90% Wilson, you have a question? 00:23:36.260 --> 00:23:39.576 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:23:39.576 --> 00:23:51.670 align:middle line:90% 00:23:51.670 --> 00:23:53.470 align:middle line:84% SIGI ZHENG: Yeah, the thing is that you 00:23:53.470 --> 00:23:57.035 align:middle line:84% don't have enough observations because, at that time, 00:23:57.035 --> 00:23:59.590 align:middle line:84% Passive House was still very new and maybe you couldn't 00:23:59.590 --> 00:24:01.340 align:middle line:90% get enough observation. 00:24:01.340 --> 00:24:02.590 align:middle line:90% That's a very real challenge. 00:24:02.590 --> 00:24:04.510 align:middle line:84% But besides this practical challenge, 00:24:04.510 --> 00:24:07.060 align:middle line:84% suppose we can get a lot of observations. 00:24:07.060 --> 00:24:09.520 align:middle line:84% And do you think this kind of thing 00:24:09.520 --> 00:24:13.860 align:middle line:84% to derive the coefficient has a challenge? 00:24:13.860 --> 00:24:15.263 align:middle line:90% Other challenges? 00:24:15.263 --> 00:24:16.930 align:middle line:84% AUDIENCE: I think it also depends on how 00:24:16.930 --> 00:24:19.510 align:middle line:90% you define your variables. 00:24:19.510 --> 00:24:23.170 align:middle line:84% For example, you're not controlling for the level 00:24:23.170 --> 00:24:24.840 align:middle line:90% of quality of the building. 00:24:24.840 --> 00:24:26.607 align:middle line:84% You're just saying it's LEED or not. 00:24:26.607 --> 00:24:28.690 align:middle line:84% Maybe you're measuring the quality of the building 00:24:28.690 --> 00:24:31.420 align:middle line:84% with the LEED certification-- it's getting confused. 00:24:31.420 --> 00:24:33.553 align:middle line:90% So it's how you define it. 00:24:33.553 --> 00:24:35.470 align:middle line:84% SIGI ZHENG: Yeah, you are saying although they 00:24:35.470 --> 00:24:37.053 align:middle line:84% call themselves LEED, maybe they don't 00:24:37.053 --> 00:24:39.335 align:middle line:84% have that LEED-specific [INAUDIBLE] features. 00:24:39.335 --> 00:24:41.710 align:middle line:84% AUDIENCE: No, maybe if you're controlling for, let's say, 00:24:41.710 --> 00:24:43.100 align:middle line:90% three variables-- 00:24:43.100 --> 00:24:44.900 align:middle line:90% LEED, view, and health-- 00:24:44.900 --> 00:24:48.020 align:middle line:84% maybe the LEED variable is not giving you 00:24:48.020 --> 00:24:49.730 align:middle line:84% the value of the LEED building-- it's 00:24:49.730 --> 00:24:52.950 align:middle line:84% only giving you giving you the value of a new prime building. 00:24:52.950 --> 00:24:56.060 align:middle line:84% SIGI ZHENG: Prime building, like class A, like that-- exactly. 00:24:56.060 --> 00:24:59.180 align:middle line:90% How to call this problem? 00:24:59.180 --> 00:25:02.060 align:middle line:84% How to call this problem, this just 00:25:02.060 --> 00:25:05.090 align:middle line:84% saying that it's a challenge that may LEED not just to LEED, 00:25:05.090 --> 00:25:07.520 align:middle line:84% but also other things bundled into LEED. 00:25:07.520 --> 00:25:08.850 align:middle line:90% The LEED building is newer. 00:25:08.850 --> 00:25:11.300 align:middle line:90% The LEED building is-- 00:25:11.300 --> 00:25:16.050 align:middle line:84% there's famous developers and those kind of things. 00:25:16.050 --> 00:25:17.100 align:middle line:90% How to call this? 00:25:17.100 --> 00:25:18.597 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:25:18.597 --> 00:25:19.305 align:middle line:90% SIGI ZHENG: Good. 00:25:19.305 --> 00:25:22.060 align:middle line:90% 00:25:22.060 --> 00:25:25.530 align:middle line:84% So I call it a very intuitive way. 00:25:25.530 --> 00:25:29.550 align:middle line:84% It's called this is not an apples-to-apples comparison. 00:25:29.550 --> 00:25:32.190 align:middle line:84% These are apples-to-oranges comparison. 00:25:32.190 --> 00:25:35.520 align:middle line:84% So that's why I got this Trader Joe thing 00:25:35.520 --> 00:25:41.500 align:middle line:84% because I want you to understand this apple-to-orange comparison. 00:25:41.500 --> 00:25:43.470 align:middle line:84% And if you answer my question correctly, 00:25:43.470 --> 00:25:44.730 align:middle line:90% I will give this to you. 00:25:44.730 --> 00:25:49.430 align:middle line:90% So now I put here-- 00:25:49.430 --> 00:25:51.708 align:middle line:84% apples-to-oranges comparison, I look at this paper. 00:25:51.708 --> 00:25:52.500 align:middle line:90% Look at this paper. 00:25:52.500 --> 00:25:58.100 align:middle line:84% This is a very famous paper of one of the very early paper 00:25:58.100 --> 00:25:59.630 align:middle line:90% of the green building thing. 00:25:59.630 --> 00:26:02.270 align:middle line:84% And they collect data of the United States. 00:26:02.270 --> 00:26:06.080 align:middle line:84% And then, of course, they collect some green building 00:26:06.080 --> 00:26:06.750 align:middle line:90% data. 00:26:06.750 --> 00:26:08.870 align:middle line:84% And then they collect nearby building data 00:26:08.870 --> 00:26:12.350 align:middle line:84% because they need to-- they need observations, they need data, 00:26:12.350 --> 00:26:14.310 align:middle line:90% and then they give this table. 00:26:14.310 --> 00:26:15.060 align:middle line:90% It's too small. 00:26:15.060 --> 00:26:18.710 align:middle line:84% But I want to show you that they first compare-- 00:26:18.710 --> 00:26:22.430 align:middle line:84% treated buildings means green buildings, LEED buildings. 00:26:22.430 --> 00:26:25.470 align:middle line:84% Control buildings means nearby non-LEED buildings. 00:26:25.470 --> 00:26:27.470 align:middle line:90% That's a terminology. 00:26:27.470 --> 00:26:29.220 align:middle line:84% I need to put the terminology here so 00:26:29.220 --> 00:26:34.720 align:middle line:90% that you can remember control. 00:26:34.720 --> 00:26:38.210 align:middle line:90% 00:26:38.210 --> 00:26:41.910 align:middle line:84% Treated means the thing that we are studying, which is green. 00:26:41.910 --> 00:26:44.660 align:middle line:84% Control means others that are not green. 00:26:44.660 --> 00:26:47.180 align:middle line:90% Then they compare. 00:26:47.180 --> 00:26:49.720 align:middle line:84% Of course, they say, oh, the rent of 00:26:49.720 --> 00:26:51.560 align:middle line:90% treated so high-- much higher. 00:26:51.560 --> 00:26:55.400 align:middle line:84% However, we also observe class A percentage. 00:26:55.400 --> 00:26:57.140 align:middle line:90% The treated green buildings-- 00:26:57.140 --> 00:27:03.095 align:middle line:84% 75% of them, they are class A. Only 25% of the control 00:27:03.095 --> 00:27:06.280 align:middle line:90% are class A. Age-- 00:27:06.280 --> 00:27:09.490 align:middle line:84% the treated buildings are average 24 years old 00:27:09.490 --> 00:27:10.330 align:middle line:90% and control-- 00:27:10.330 --> 00:27:11.860 align:middle line:90% 53 years old. 00:27:11.860 --> 00:27:13.900 align:middle line:90% So that gives you a sense. 00:27:13.900 --> 00:27:15.260 align:middle line:90% They are not the same thing. 00:27:15.260 --> 00:27:18.650 align:middle line:84% They are not the same animals, this apple and orange. 00:27:18.650 --> 00:27:25.030 align:middle line:84% So that's an apple and orange comparison. 00:27:25.030 --> 00:27:28.035 align:middle line:84% So, in our sense, think about this. 00:27:28.035 --> 00:27:33.160 align:middle line:90% 00:27:33.160 --> 00:27:35.720 align:middle line:84% We observe, oh, this is a green building. 00:27:35.720 --> 00:27:37.520 align:middle line:90% This is a nongreen building. 00:27:37.520 --> 00:27:40.980 align:middle line:84% The price difference is, like, $1 million. 00:27:40.980 --> 00:27:42.580 align:middle line:90% It's a lot. 00:27:42.580 --> 00:27:46.250 align:middle line:84% However, how can we so confident? 00:27:46.250 --> 00:27:49.210 align:middle line:84% How can we be confident that's because of the greenness. 00:27:49.210 --> 00:27:51.820 align:middle line:84% Because the green building also located in such 00:27:51.820 --> 00:27:53.470 align:middle line:90% a beautiful neighborhood-- 00:27:53.470 --> 00:27:55.990 align:middle line:84% trees, green park, and all the things. 00:27:55.990 --> 00:27:59.710 align:middle line:84% This building is such an ugly place and a factory 00:27:59.710 --> 00:28:01.600 align:middle line:84% and pollution, all the things nearby-- 00:28:01.600 --> 00:28:03.520 align:middle line:90% not compared. 00:28:03.520 --> 00:28:07.840 align:middle line:84% And so then that's basically the problem. 00:28:07.840 --> 00:28:10.750 align:middle line:84% How to better get from this apple 00:28:10.750 --> 00:28:13.725 align:middle line:90% to orange to get apple to apple? 00:28:13.725 --> 00:28:16.280 align:middle line:90% 00:28:16.280 --> 00:28:19.085 align:middle line:90% Do you have any answer for that? 00:28:19.085 --> 00:28:20.960 align:middle line:84% AUDIENCE: You would want to compare buildings 00:28:20.960 --> 00:28:23.280 align:middle line:84% that were built [INAUDIBLE] close by, 00:28:23.280 --> 00:28:27.280 align:middle line:84% and then, also, you would get the percentage of class A 00:28:27.280 --> 00:28:30.180 align:middle line:90% buildings [INAUDIBLE]. 00:28:30.180 --> 00:28:31.510 align:middle line:90% SIGI ZHENG: You said close by. 00:28:31.510 --> 00:28:33.293 align:middle line:84% That's basically location difference. 00:28:33.293 --> 00:28:35.710 align:middle line:84% You won't get this-- one is a nice neighborhood, the other 00:28:35.710 --> 00:28:37.620 align:middle line:90% in a factory neighborhood. 00:28:37.620 --> 00:28:40.800 align:middle line:84% But you say even nearby may be different. 00:28:40.800 --> 00:28:43.860 align:middle line:90% It's the old and the new one-- 00:28:43.860 --> 00:28:45.150 align:middle line:90% how to do this? 00:28:45.150 --> 00:28:47.940 align:middle line:84% Even nearby, two adjacent buildings-- one is new. 00:28:47.940 --> 00:28:49.720 align:middle line:90% The other is old. 00:28:49.720 --> 00:28:54.310 align:middle line:84% But this one's for you because I have several. 00:28:54.310 --> 00:28:55.905 align:middle line:90% This one.-- yeah, please. 00:28:55.905 --> 00:28:57.280 align:middle line:84% AUDIENCE: Wouldn't there be a way 00:28:57.280 --> 00:29:00.735 align:middle line:84% to normalize it so you can compare buildings that are old 00:29:00.735 --> 00:29:08.010 align:middle line:84% [INAUDIBLE] so that you're not comparing the same figures, 00:29:08.010 --> 00:29:11.280 align:middle line:84% but you're comparing [INAUDIBLE] and you can still 00:29:11.280 --> 00:29:13.340 align:middle line:90% use an older building? 00:29:13.340 --> 00:29:15.847 align:middle line:84% SIGI ZHENG: OK, your point-- could you repeat how-- 00:29:15.847 --> 00:29:17.180 align:middle line:90% AUDIENCE: I don't know exactly-- 00:29:17.180 --> 00:29:22.440 align:middle line:84% SIGI ZHENG: You just want to make this normalized way. 00:29:22.440 --> 00:29:23.950 align:middle line:90% Yeah, normalized is a point. 00:29:23.950 --> 00:29:25.860 align:middle line:90% And then this one is for you. 00:29:25.860 --> 00:29:27.958 align:middle line:90% And I ran out of-- please. 00:29:27.958 --> 00:29:29.750 align:middle line:84% AUDIENCE: You want to add as many variables 00:29:29.750 --> 00:29:32.210 align:middle line:84% as you can to make sure that you're comparing the product, 00:29:32.210 --> 00:29:34.793 align:middle line:84% so you're going to add in the room count like you were saying, 00:29:34.793 --> 00:29:38.570 align:middle line:84% the relative age, condition of the exterior [INAUDIBLE] 00:29:38.570 --> 00:29:41.240 align:middle line:84% or the interior, et cetera, so you have a bunch of data points, 00:29:41.240 --> 00:29:43.737 align:middle line:84% and then you can look at the relative value of each 00:29:43.737 --> 00:29:46.070 align:middle line:84% of those individual data points and fits into the larger 00:29:46.070 --> 00:29:46.470 align:middle line:90% formula. 00:29:46.470 --> 00:29:47.178 align:middle line:90% SIGI ZHENG: Good. 00:29:47.178 --> 00:29:48.900 align:middle line:84% So we should add as much as possible. 00:29:48.900 --> 00:29:50.870 align:middle line:84% The controls-- we call it controls. 00:29:50.870 --> 00:29:53.450 align:middle line:84% We can call this control variables, 00:29:53.450 --> 00:29:56.900 align:middle line:84% but as we know that always something in the trash 00:29:56.900 --> 00:29:59.930 align:middle line:84% can because there's so much difference and how can 00:29:59.930 --> 00:30:01.870 align:middle line:90% you control. 00:30:01.870 --> 00:30:05.320 align:middle line:84% So now I'm moving on a little bit, 00:30:05.320 --> 00:30:08.320 align:middle line:84% and I still need apple to orange. 00:30:08.320 --> 00:30:10.930 align:middle line:90% We have two. 00:30:10.930 --> 00:30:15.040 align:middle line:84% Now this is called causal inference. 00:30:15.040 --> 00:30:17.440 align:middle line:84% This I think you really need to understand, 00:30:17.440 --> 00:30:19.120 align:middle line:84% although maybe you don't want to get 00:30:19.120 --> 00:30:23.260 align:middle line:84% technical details, but the terminology and the thinking. 00:30:23.260 --> 00:30:27.010 align:middle line:84% Causal inference, whether it's just a correlation or it's 00:30:27.010 --> 00:30:28.630 align:middle line:90% causal-- we want causal. 00:30:28.630 --> 00:30:31.570 align:middle line:84% We don't want to just play with the correlations. 00:30:31.570 --> 00:30:37.450 align:middle line:84% And the causal inference-- the challenge is we cannot observe 00:30:37.450 --> 00:30:38.840 align:middle line:90% the counterfactual. 00:30:38.840 --> 00:30:41.080 align:middle line:84% This is a new terminology, "counterfactual." 00:30:41.080 --> 00:30:44.890 align:middle line:84% Counterfactual-- something in your dream you really want, 00:30:44.890 --> 00:30:45.920 align:middle line:90% but you cannot. 00:30:45.920 --> 00:30:47.620 align:middle line:90% That's a counterfactual. 00:30:47.620 --> 00:30:50.140 align:middle line:90% And I will try to show this. 00:30:50.140 --> 00:30:52.430 align:middle line:90% Let me see. 00:30:52.430 --> 00:30:55.490 align:middle line:84% Compare LEED building-- for example, 00:30:55.490 --> 00:30:58.460 align:middle line:84% this is a green building-- and another brown building. 00:30:58.460 --> 00:30:59.390 align:middle line:90% That's a treatment. 00:30:59.390 --> 00:31:00.910 align:middle line:90% This is control. 00:31:00.910 --> 00:31:03.620 align:middle line:84% Then we know the building's in a place 00:31:03.620 --> 00:31:06.740 align:middle line:84% so good that I put a small green tree 00:31:06.740 --> 00:31:09.110 align:middle line:90% to show this nice neighborhood. 00:31:09.110 --> 00:31:11.390 align:middle line:84% And the green building-- the building 00:31:11.390 --> 00:31:15.770 align:middle line:84% is in an ugly place-- rocks and random things. 00:31:15.770 --> 00:31:21.160 align:middle line:84% So now if we know-- if we compare these two, 00:31:21.160 --> 00:31:23.720 align:middle line:90% we know these are different. 00:31:23.720 --> 00:31:25.370 align:middle line:90% This is an orange. 00:31:25.370 --> 00:31:27.430 align:middle line:90% This is an apple-- cannot-- 00:31:27.430 --> 00:31:29.350 align:middle line:90% two different animals. 00:31:29.350 --> 00:31:33.070 align:middle line:90% And then how can we do? 00:31:33.070 --> 00:31:36.130 align:middle line:84% Ideally, we want a counterfactual. 00:31:36.130 --> 00:31:38.060 align:middle line:90% This is called counterfactual. 00:31:38.060 --> 00:31:40.030 align:middle line:90% This is a dream, not real. 00:31:40.030 --> 00:31:43.960 align:middle line:84% This counterfactual is all other things 00:31:43.960 --> 00:31:47.320 align:middle line:90% are the same as this real one. 00:31:47.320 --> 00:31:49.630 align:middle line:90% Let's say a tree is here. 00:31:49.630 --> 00:31:53.200 align:middle line:84% All the same as this one-- but the only difference 00:31:53.200 --> 00:31:55.070 align:middle line:90% is whether you are green or not. 00:31:55.070 --> 00:31:58.300 align:middle line:84% The nearby things, the size and all things-- 00:31:58.300 --> 00:32:01.660 align:middle line:84% not counterfactual, but of course-- let's see. 00:32:01.660 --> 00:32:03.610 align:middle line:90% I have this one. 00:32:03.610 --> 00:32:11.580 align:middle line:84% This is-- and then we know we want this one. 00:32:11.580 --> 00:32:14.250 align:middle line:84% All else equal, the effect of LEED-- 00:32:14.250 --> 00:32:17.950 align:middle line:84% is this real LEED minus counterfactual? 00:32:17.950 --> 00:32:20.610 align:middle line:84% So that's the real thing we want. 00:32:20.610 --> 00:32:23.890 align:middle line:90% However, we don't want this. 00:32:23.890 --> 00:32:25.660 align:middle line:90% This is called bias. 00:32:25.660 --> 00:32:31.260 align:middle line:84% Bias means this brown building and that brown building-- 00:32:31.260 --> 00:32:32.910 align:middle line:84% they have different other things. 00:32:32.910 --> 00:32:33.840 align:middle line:90% That's called bias. 00:32:33.840 --> 00:32:38.000 align:middle line:84% Bias is disturbing us from reaching 00:32:38.000 --> 00:32:44.150 align:middle line:90% the real result. Algebra-- 00:32:44.150 --> 00:32:46.580 align:middle line:84% difficult, and so small you cannot see. 00:32:46.580 --> 00:32:49.400 align:middle line:84% But just want to tell you, but then, to make 00:32:49.400 --> 00:32:53.150 align:middle line:84% sure you understand, I use my fruits 00:32:53.150 --> 00:32:55.820 align:middle line:84% from Trader Joe's for you to understand. 00:32:55.820 --> 00:32:57.830 align:middle line:84% And later, after many years, we will 00:32:57.830 --> 00:33:00.260 align:middle line:84% think about the causal inference-- think 00:33:00.260 --> 00:33:03.257 align:middle line:90% about Sigi and the fruits. 00:33:03.257 --> 00:33:04.090 align:middle line:90% Then you understand. 00:33:04.090 --> 00:33:06.020 align:middle line:90% You remember. 00:33:06.020 --> 00:33:09.950 align:middle line:90% So now the following is a lot. 00:33:09.950 --> 00:33:15.020 align:middle line:84% This green building-- say this is green apple-- 00:33:15.020 --> 00:33:15.560 align:middle line:90% LEED. 00:33:15.560 --> 00:33:17.720 align:middle line:90% Green apple-- green. 00:33:17.720 --> 00:33:19.560 align:middle line:90% This is an orange-- 00:33:19.560 --> 00:33:20.900 align:middle line:90% totally different thing. 00:33:20.900 --> 00:33:25.040 align:middle line:84% Now we need to create a counterfactual which 00:33:25.040 --> 00:33:28.940 align:middle line:84% is also apple-- same animal, same fruit-- apple. 00:33:28.940 --> 00:33:32.480 align:middle line:84% But this is a red apple or brown apple. 00:33:32.480 --> 00:33:36.330 align:middle line:84% So this is a counterfactual, So these are two real things. 00:33:36.330 --> 00:33:37.580 align:middle line:90% This is in dream-- 00:33:37.580 --> 00:33:39.500 align:middle line:90% cannot-- this dream. 00:33:39.500 --> 00:33:41.090 align:middle line:90% We observe these two. 00:33:41.090 --> 00:33:44.570 align:middle line:84% But we know these two are different. 00:33:44.570 --> 00:33:50.570 align:middle line:84% Now we create-- make up one thing is a counterfactual. 00:33:50.570 --> 00:33:55.430 align:middle line:84% The same animal, but the only difference is the color. 00:33:55.430 --> 00:33:56.880 align:middle line:90% So that's the point. 00:33:56.880 --> 00:34:00.860 align:middle line:84% The difference between these two are the real effect of LEED. 00:34:00.860 --> 00:34:02.330 align:middle line:90% Understand? 00:34:02.330 --> 00:34:04.343 align:middle line:90% And these two-- bias. 00:34:04.343 --> 00:34:08.560 align:middle line:90% 00:34:08.560 --> 00:34:11.699 align:middle line:84% And you need algebra to understand. 00:34:11.699 --> 00:34:13.710 align:middle line:90% If you understand this-- 00:34:13.710 --> 00:34:15.150 align:middle line:90% econometrics master. 00:34:15.150 --> 00:34:23.409 align:middle line:84% So what we observe is this green apple and an orange-- 00:34:23.409 --> 00:34:25.210 align:middle line:90% what we observe. 00:34:25.210 --> 00:34:27.969 align:middle line:84% And how to make sure you understand this? 00:34:27.969 --> 00:34:33.219 align:middle line:90% We make some real arrangement-- 00:34:33.219 --> 00:34:36.380 align:middle line:84% minus red apple, plus a red apple. 00:34:36.380 --> 00:34:37.840 align:middle line:90% The same-- cancel out. 00:34:37.840 --> 00:34:42.550 align:middle line:84% But we just plus a red apple minus a red apple-- 00:34:42.550 --> 00:34:43.580 align:middle line:90% minus and plus. 00:34:43.580 --> 00:34:48.250 align:middle line:84% Then this part, called treatment effect, that's 00:34:48.250 --> 00:34:53.800 align:middle line:90% a green apple minus red apple-- 00:34:53.800 --> 00:34:56.080 align:middle line:90% same apples. 00:34:56.080 --> 00:34:58.810 align:middle line:90% And this one bothers us-- 00:34:58.810 --> 00:35:00.280 align:middle line:90% bias. 00:35:00.280 --> 00:35:02.900 align:middle line:90% The red apple-- counterfactual. 00:35:02.900 --> 00:35:07.160 align:middle line:84% What's the counterfactual difference between this orange? 00:35:07.160 --> 00:35:07.710 align:middle line:90% Understand? 00:35:07.710 --> 00:35:11.270 align:middle line:84% So this is from the algebra, you artificially 00:35:11.270 --> 00:35:16.890 align:middle line:84% add minus that counterfactual will add the counterfactual. 00:35:16.890 --> 00:35:18.320 align:middle line:90% Then you have two terms. 00:35:18.320 --> 00:35:20.760 align:middle line:90% The upper is the real treatment. 00:35:20.760 --> 00:35:22.010 align:middle line:90% This is control. 00:35:22.010 --> 00:35:23.630 align:middle line:90% This is bias. 00:35:23.630 --> 00:35:30.440 align:middle line:84% So that's the algebra in this graphical way is suddenly 00:35:30.440 --> 00:35:33.260 align:middle line:90% your dream come to this. 00:35:33.260 --> 00:35:34.880 align:middle line:90% This is not real. 00:35:34.880 --> 00:35:36.620 align:middle line:90% And we need this. 00:35:36.620 --> 00:35:37.670 align:middle line:90% We don't want this. 00:35:37.670 --> 00:35:39.900 align:middle line:84% But in reality, we only have this. 00:35:39.900 --> 00:35:45.110 align:middle line:84% So of course, this is such a challenge. 00:35:45.110 --> 00:35:46.650 align:middle line:90% Of course, this is a challenge. 00:35:46.650 --> 00:35:48.350 align:middle line:90% Everyone knows it's a challenge. 00:35:48.350 --> 00:35:55.880 align:middle line:84% And then very smart people came up with techniques 00:35:55.880 --> 00:36:01.750 align:middle line:90% to make sure we can achieve. 00:36:01.750 --> 00:36:05.410 align:middle line:84% And so, basically, as I mentioned, 00:36:05.410 --> 00:36:07.690 align:middle line:90% this is the counterfactual. 00:36:07.690 --> 00:36:11.410 align:middle line:84% The point of this is to make apples-to-oranges comparison 00:36:11.410 --> 00:36:13.240 align:middle line:84% to an apples-to-apples comparison. 00:36:13.240 --> 00:36:14.450 align:middle line:90% So that's the key. 00:36:14.450 --> 00:36:18.490 align:middle line:84% Without doing so, cannot argue your causal inference. 00:36:18.490 --> 00:36:21.340 align:middle line:84% And then now because some of you, 00:36:21.340 --> 00:36:23.960 align:middle line:84% it seems that you understand this, 00:36:23.960 --> 00:36:26.300 align:middle line:90% and then what are the things-- 00:36:26.300 --> 00:36:29.570 align:middle line:84% what are the things why there is such a bother as the bias? 00:36:29.570 --> 00:36:33.320 align:middle line:90% What are the challenges to this? 00:36:33.320 --> 00:36:37.060 align:middle line:84% We call it a terminology called endogeneity. 00:36:37.060 --> 00:36:41.930 align:middle line:84% Endogeneity is a challenge to causal inference. 00:36:41.930 --> 00:36:46.960 align:middle line:84% If we can really observe two apples, no endogeneity, 00:36:46.960 --> 00:36:49.090 align:middle line:90% it means we estimate. 00:36:49.090 --> 00:36:50.440 align:middle line:90% In reality, we cannot. 00:36:50.440 --> 00:36:53.360 align:middle line:84% And then what are the real challenges? 00:36:53.360 --> 00:36:56.067 align:middle line:90% Anyone know some examples? 00:36:56.067 --> 00:36:57.400 align:middle line:90% Think about the green buildings. 00:36:57.400 --> 00:37:01.020 align:middle line:84% Let's use a real example to say why we cannot. 00:37:01.020 --> 00:37:04.760 align:middle line:84% Why there are some things that bother us-- 00:37:04.760 --> 00:37:07.910 align:middle line:90% not make this bias exist? 00:37:07.910 --> 00:37:09.560 align:middle line:90% This is a nongreen building. 00:37:09.560 --> 00:37:11.270 align:middle line:90% This is a green building. 00:37:11.270 --> 00:37:15.647 align:middle line:84% Both are nongreen building, but why there is a difference? 00:37:15.647 --> 00:37:16.355 align:middle line:90% Think about this. 00:37:16.355 --> 00:37:19.000 align:middle line:90% 00:37:19.000 --> 00:37:20.165 align:middle line:90% Go ahead. 00:37:20.165 --> 00:37:22.620 align:middle line:84% AUDIENCE: [INAUDIBLE] about relevant comparator data, 00:37:22.620 --> 00:37:32.850 align:middle line:84% so there would be other factors that [INAUDIBLE] not relevant 00:37:32.850 --> 00:37:33.968 align:middle line:90% to what we have-- 00:37:33.968 --> 00:37:35.760 align:middle line:84% SIGI ZHENG: Can you-- this is too abstract. 00:37:35.760 --> 00:37:38.700 align:middle line:84% Can you make this very intuitive? 00:37:38.700 --> 00:37:41.690 align:middle line:90% Green building-- LEED building. 00:37:41.690 --> 00:37:45.690 align:middle line:90% 00:37:45.690 --> 00:37:49.060 align:middle line:84% AUDIENCE: If we are observing a building in Boston 00:37:49.060 --> 00:37:51.070 align:middle line:84% and another building in New York, the market 00:37:51.070 --> 00:37:54.020 align:middle line:90% factors [INAUDIBLE] different. 00:37:54.020 --> 00:37:56.563 align:middle line:84% So it's not really an apples-to-apples comparison 00:37:56.563 --> 00:37:58.480 align:middle line:84% because the location is a different component. 00:37:58.480 --> 00:38:00.230 align:middle line:84% SIGI ZHENG: Yeah, but always, if we really 00:38:00.230 --> 00:38:02.000 align:middle line:84% do this green building premium estimation, 00:38:02.000 --> 00:38:03.500 align:middle line:90% we don't go to two markets-- 00:38:03.500 --> 00:38:05.450 align:middle line:84% gather some data from New York and Boston. 00:38:05.450 --> 00:38:07.190 align:middle line:84% For example, we're all from Boston. 00:38:07.190 --> 00:38:09.565 align:middle line:84% And then is there any problem for that? 00:38:09.565 --> 00:38:12.740 align:middle line:90% 00:38:12.740 --> 00:38:15.200 align:middle line:84% Still different neighborhoods-- still 00:38:15.200 --> 00:38:17.985 align:middle line:84% neighborhoods because maybe some rich people live in this area 00:38:17.985 --> 00:38:19.610 align:middle line:84% and some poor people live in that area. 00:38:19.610 --> 00:38:21.720 align:middle line:84% And for poor people, the green is nothing. 00:38:21.720 --> 00:38:23.420 align:middle line:90% They want space. 00:38:23.420 --> 00:38:26.240 align:middle line:84% They don't care about the environmental things first. 00:38:26.240 --> 00:38:28.340 align:middle line:84% OK, this give you this green building. 00:38:28.340 --> 00:38:30.410 align:middle line:90% And other things? 00:38:30.410 --> 00:38:34.760 align:middle line:84% Any possible other way because there's some difference you 00:38:34.760 --> 00:38:35.580 align:middle line:90% cannot observe. 00:38:35.580 --> 00:38:37.580 align:middle line:90% That's called omitted variables. 00:38:37.580 --> 00:38:41.530 align:middle line:90% And what other things? 00:38:41.530 --> 00:38:45.760 align:middle line:84% Besides [INAUDIBLE] variables, anything you can imagine? 00:38:45.760 --> 00:38:52.490 align:middle line:84% AUDIENCE: [INAUDIBLE] building dependent on the [INAUDIBLE] 00:38:52.490 --> 00:38:53.865 align:middle line:90% the brown building has-- 00:38:53.865 --> 00:38:54.740 align:middle line:90% SIGI ZHENG: Low rent. 00:38:54.740 --> 00:39:06.790 align:middle line:84% AUDIENCE: [INAUDIBLE] the office tenants [INAUDIBLE] they will 00:39:06.790 --> 00:39:09.240 align:middle line:90% prefer to-- 00:39:09.240 --> 00:39:15.740 align:middle line:84% if they are preferred to the high rent, maybe surrounded 00:39:15.740 --> 00:39:22.170 align:middle line:90% this building [INAUDIBLE] 00:39:22.170 --> 00:39:25.590 align:middle line:84% SIGI ZHENG: Yeah, but that's not the bias. 00:39:25.590 --> 00:39:28.090 align:middle line:90% That's what we want to see. 00:39:28.090 --> 00:39:40.673 align:middle line:84% AUDIENCE: Actually, if the [INAUDIBLE] bias [INAUDIBLE] 00:39:40.673 --> 00:39:42.840 align:middle line:84% SIGI ZHENG: OK, [INAUDIBLE] the company-- yes, good. 00:39:42.840 --> 00:39:45.770 align:middle line:84% So not just the location, but also the company. 00:39:45.770 --> 00:39:48.270 align:middle line:84% Some big companies-- they really value their image. 00:39:48.270 --> 00:39:50.060 align:middle line:90% So this is for you. 00:39:50.060 --> 00:39:53.780 align:middle line:84% Now I'm going to say this very complicated terminologies, 00:39:53.780 --> 00:39:56.220 align:middle line:90% but don't be scared. 00:39:56.220 --> 00:40:00.680 align:middle line:84% So in econometrics, we run our model y 00:40:00.680 --> 00:40:03.000 align:middle line:84% as a function of some other things. 00:40:03.000 --> 00:40:04.770 align:middle line:90% We call this treatment. 00:40:04.770 --> 00:40:10.430 align:middle line:84% The thing we are most interested in is D equals 1 is treated-- 00:40:10.430 --> 00:40:11.940 align:middle line:90% otherwise, D equals 0. 00:40:11.940 --> 00:40:13.740 align:middle line:90% So this is LEED or not. 00:40:13.740 --> 00:40:17.210 align:middle line:84% All these other variables are these controls [INAUDIBLE] 00:40:17.210 --> 00:40:19.010 align:middle line:90% X as a vector. 00:40:19.010 --> 00:40:23.110 align:middle line:84% So this endogeneity means there's 00:40:23.110 --> 00:40:24.700 align:middle line:90% something in the trash can. 00:40:24.700 --> 00:40:25.850 align:middle line:90% This is a trash can. 00:40:25.850 --> 00:40:27.910 align:middle line:84% Remember, always you have a trash can, 00:40:27.910 --> 00:40:31.420 align:middle line:84% and the trash can and the treated variable, 00:40:31.420 --> 00:40:33.250 align:middle line:90% they are correlated. 00:40:33.250 --> 00:40:34.140 align:middle line:90% They are correlated. 00:40:34.140 --> 00:40:35.300 align:middle line:90% They are not independent. 00:40:35.300 --> 00:40:38.080 align:middle line:84% If the trash can has so many things in the trash 00:40:38.080 --> 00:40:40.960 align:middle line:84% can, but the trash can has nothing to do with the treated, 00:40:40.960 --> 00:40:43.570 align:middle line:90% then you won't need to worry. 00:40:43.570 --> 00:40:46.850 align:middle line:84% But if they are correlated-- very. 00:40:46.850 --> 00:40:48.770 align:middle line:84% And three times-- just now we already 00:40:48.770 --> 00:40:52.640 align:middle line:84% talked about omitted variables, either from the location 00:40:52.640 --> 00:40:53.940 align:middle line:90% or from the company. 00:40:53.940 --> 00:40:56.600 align:middle line:84% So basically, there's so many things you cannot observe 00:40:56.600 --> 00:40:58.150 align:middle line:90% in a big company. 00:40:58.150 --> 00:41:03.410 align:middle line:84% You must build this-- a small company investing in this. 00:41:03.410 --> 00:41:06.830 align:middle line:84% And this is a nice [INAUDIBLE] neighborhood, all the things. 00:41:06.830 --> 00:41:09.950 align:middle line:84% And second one is called reverse causality. 00:41:09.950 --> 00:41:13.850 align:middle line:84% Reverse causality is the story we want to tell-- 00:41:13.850 --> 00:41:16.070 align:middle line:84% is from the right-hand side to the left-hand side. 00:41:16.070 --> 00:41:17.750 align:middle line:90% That's a regression point. 00:41:17.750 --> 00:41:20.730 align:middle line:84% If it's treated, if this green was a premium-- 00:41:20.730 --> 00:41:23.150 align:middle line:84% so that's from the right-hand side to left-hand side. 00:41:23.150 --> 00:41:26.090 align:middle line:84% Unfortunately, there is another channel from left-hand side 00:41:26.090 --> 00:41:28.200 align:middle line:90% to the right-hand side. 00:41:28.200 --> 00:41:35.400 align:middle line:84% For example, for those very expensive buildings because 00:41:35.400 --> 00:41:39.180 align:middle line:84% of the location or because of anything-- so expensive 00:41:39.180 --> 00:41:43.290 align:middle line:84% and charged so high rental price not because 00:41:43.290 --> 00:41:46.560 align:middle line:84% of the energy-related things, but they are 00:41:46.560 --> 00:41:48.910 align:middle line:84% so-- they collect so much revenue. 00:41:48.910 --> 00:41:51.780 align:middle line:84% And the board of the company say we have so much revenue-- 00:41:51.780 --> 00:41:53.460 align:middle line:90% let's renovate. 00:41:53.460 --> 00:41:56.050 align:middle line:84% If you cannot charge enough, then you have money-- 00:41:56.050 --> 00:41:57.810 align:middle line:90% you need to cover daily things. 00:41:57.810 --> 00:42:01.510 align:middle line:84% Now for the very rich, for the very expensive buildings, 00:42:01.510 --> 00:42:02.640 align:middle line:90% they have a lot of revenue. 00:42:02.640 --> 00:42:05.790 align:middle line:84% Then they decide to treat, not because they treated 00:42:05.790 --> 00:42:08.520 align:middle line:84% them become higher, but it's because they 00:42:08.520 --> 00:42:10.920 align:middle line:84% collect more revenue, then they decide to treat. 00:42:10.920 --> 00:42:13.440 align:middle line:84% That's opposite to what we want to say-- 00:42:13.440 --> 00:42:14.600 align:middle line:90% well, the estimate. 00:42:14.600 --> 00:42:17.160 align:middle line:90% So that costs reverse causality. 00:42:17.160 --> 00:42:20.260 align:middle line:84% The third thing is called selection bias. 00:42:20.260 --> 00:42:22.300 align:middle line:90% That's called selective sample. 00:42:22.300 --> 00:42:24.840 align:middle line:84% So the sample you selected, you collect, 00:42:24.840 --> 00:42:27.640 align:middle line:84% is not representative of the entire market. 00:42:27.640 --> 00:42:32.830 align:middle line:84% For example, if we say let's estimate the green premium, 00:42:32.830 --> 00:42:36.990 align:middle line:84% the green price premium for green buildings, 00:42:36.990 --> 00:42:40.190 align:middle line:84% then you go to Zillow and you collect. 00:42:40.190 --> 00:42:43.850 align:middle line:84% However, we know the transactions 00:42:43.850 --> 00:42:46.280 align:middle line:84% for the new houses-- you always transacted. 00:42:46.280 --> 00:42:48.200 align:middle line:84% For the old houses, it's not always-- 00:42:48.200 --> 00:42:49.340 align:middle line:90% low frequency. 00:42:49.340 --> 00:42:51.950 align:middle line:84% Then for the transaction sample you 00:42:51.950 --> 00:42:55.640 align:middle line:84% collect it's not the representative 00:42:55.640 --> 00:42:57.570 align:middle line:90% of the entire housing stock. 00:42:57.570 --> 00:43:00.080 align:middle line:84% You always have more new ones transacted, 00:43:00.080 --> 00:43:01.830 align:middle line:90% less old ones transacted. 00:43:01.830 --> 00:43:04.280 align:middle line:84% So then you have a selected buyers 00:43:04.280 --> 00:43:07.370 align:middle line:84% because you use a selected sample to estimate this. 00:43:07.370 --> 00:43:08.720 align:middle line:90% I'm not sure whether you-- 00:43:08.720 --> 00:43:12.000 align:middle line:84% so, basically, this is not underlying population. 00:43:12.000 --> 00:43:13.890 align:middle line:90% So that's called selection bias. 00:43:13.890 --> 00:43:17.870 align:middle line:84% All three reasons will always cause this problem 00:43:17.870 --> 00:43:20.360 align:middle line:90% of called endogeneity. 00:43:20.360 --> 00:43:26.810 align:middle line:84% So this is conceptual level, and there are so many technologies. 00:43:26.810 --> 00:43:32.210 align:middle line:84% And then we need to do apple-to-apple comparison. 00:43:32.210 --> 00:43:35.120 align:middle line:84% So the apple-to-apple comparison-- 00:43:35.120 --> 00:43:38.090 align:middle line:84% there are some techniques the economists 00:43:38.090 --> 00:43:40.930 align:middle line:84% and the econometrics scholars already developed. 00:43:40.930 --> 00:43:42.990 align:middle line:90% Anyone knows any names of them? 00:43:42.990 --> 00:43:47.630 align:middle line:84% This kind of how to get to apple-to-apple-- 00:43:47.630 --> 00:43:49.040 align:middle line:90% using any terminologies? 00:43:49.040 --> 00:43:51.070 align:middle line:90% I feel like you understand. 00:43:51.070 --> 00:43:51.920 align:middle line:90% Yeah, please? 00:43:51.920 --> 00:43:55.399 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:43:55.399 --> 00:43:59.665 align:middle line:90% 00:43:59.665 --> 00:44:00.540 align:middle line:90% SIGI ZHENG: OK, good. 00:44:00.540 --> 00:44:01.390 align:middle line:90% That's one example. 00:44:01.390 --> 00:44:03.470 align:middle line:84% Do you have one thing I gave you? 00:44:03.470 --> 00:44:07.550 align:middle line:90% No, OK, this one is for you. 00:44:07.550 --> 00:44:10.490 align:middle line:84% Mackenzie said regression discontinuity, yes, 00:44:10.490 --> 00:44:12.110 align:middle line:84% but maybe you don't know what's this. 00:44:12.110 --> 00:44:13.650 align:middle line:90% And other terminologies? 00:44:13.650 --> 00:44:16.850 align:middle line:90% 00:44:16.850 --> 00:44:21.597 align:middle line:84% Yes-- from your RA work for me, you know something-- please. 00:44:21.597 --> 00:44:22.930 align:middle line:90% AUDIENCE: Instrumental variable. 00:44:22.930 --> 00:44:25.640 align:middle line:90% 00:44:25.640 --> 00:44:27.500 align:middle line:84% SIGI ZHENG: Instrumental variable, yes. 00:44:27.500 --> 00:44:29.790 align:middle line:84% Since I already paid you, so no fruit-- 00:44:29.790 --> 00:44:35.720 align:middle line:84% OK, fine-- because you learn the knowledge from me. 00:44:35.720 --> 00:44:38.360 align:middle line:84% But it's so many things-- so many technologies. 00:44:38.360 --> 00:44:39.930 align:middle line:90% It's not for you to understand. 00:44:39.930 --> 00:44:43.670 align:middle line:84% If you understand this, graduate from the econ department, 00:44:43.670 --> 00:44:46.070 align:middle line:90% master's degree or PhD. 00:44:46.070 --> 00:44:49.420 align:middle line:84% So, yes, even after PhD, I still cannot do this. 00:44:49.420 --> 00:44:53.050 align:middle line:84% So the thing is, it's so small, but I give you a sense of this. 00:44:53.050 --> 00:44:54.910 align:middle line:90% This is fascinating. 00:44:54.910 --> 00:44:57.620 align:middle line:84% There's so many good technologies. 00:44:57.620 --> 00:45:01.220 align:middle line:84% One, the big category is the following. 00:45:01.220 --> 00:45:04.980 align:middle line:84% At first, we come from some naive ways. 00:45:04.980 --> 00:45:08.900 align:middle line:84% Naive ways will be accurate responsibility-- not naive. 00:45:08.900 --> 00:45:10.430 align:middle line:90% Basic-- not naive. 00:45:10.430 --> 00:45:12.290 align:middle line:84% Accuracy is very high level, so we 00:45:12.290 --> 00:45:15.680 align:middle line:84% will teach you OLS, which is a basic regression. 00:45:15.680 --> 00:45:18.440 align:middle line:84% Even for basic regression, we have ways to control, 00:45:18.440 --> 00:45:21.388 align:middle line:84% for example, so-called fixed effect. 00:45:21.388 --> 00:45:22.680 align:middle line:90% But I don't want to go to that. 00:45:22.680 --> 00:45:25.950 align:middle line:84% But anyway, if you do our basic-- 00:45:25.950 --> 00:45:28.590 align:middle line:84% we call the OLS, Ordinary Least Square. 00:45:28.590 --> 00:45:29.760 align:middle line:90% So that's the thing. 00:45:29.760 --> 00:45:31.770 align:middle line:84% Then at least you can try your best 00:45:31.770 --> 00:45:34.410 align:middle line:84% to put all enough controls-- all the data. 00:45:34.410 --> 00:45:38.850 align:middle line:84% If you have so many controls [INAUDIBLE] and have other ways. 00:45:38.850 --> 00:45:40.770 align:middle line:90% And then move on-- 00:45:40.770 --> 00:45:42.120 align:middle line:90% more advanced. 00:45:42.120 --> 00:45:46.160 align:middle line:84% More advanced to solve this apple to apple comparison thing 00:45:46.160 --> 00:45:48.160 align:middle line:90% to achieve this apple-to-apple-- 00:45:48.160 --> 00:45:49.590 align:middle line:90% I have some more-- 00:45:49.590 --> 00:45:52.260 align:middle line:90% the thing is the following. 00:45:52.260 --> 00:45:57.420 align:middle line:90% The golden way, so-called RCT-- 00:45:57.420 --> 00:46:01.170 align:middle line:84% that's called Randomized Controlled Trial, RCT. 00:46:01.170 --> 00:46:03.360 align:middle line:84% That's really you do an experiment. 00:46:03.360 --> 00:46:07.030 align:middle line:90% You do experiment in a lab. 00:46:07.030 --> 00:46:08.710 align:middle line:90% Then you really observe-- 00:46:08.710 --> 00:46:11.490 align:middle line:84% it's not like you just observe from the real world. 00:46:11.490 --> 00:46:14.830 align:middle line:90% You do a lab thing, RCT. 00:46:14.830 --> 00:46:17.170 align:middle line:84% And if you cannot do an experiment-- 00:46:17.170 --> 00:46:19.420 align:middle line:90% experiment is very hard to run-- 00:46:19.420 --> 00:46:23.230 align:middle line:84% and then you have quasi-experiment, like half, 00:46:23.230 --> 00:46:24.930 align:middle line:90% [INAUDIBLE]. 00:46:24.930 --> 00:46:27.040 align:middle line:84% Then you have some things-- today, 00:46:27.040 --> 00:46:29.430 align:middle line:84% I will talk a little bit about one advanced, 00:46:29.430 --> 00:46:30.780 align:middle line:90% which is called matching. 00:46:30.780 --> 00:46:32.250 align:middle line:90% You match. 00:46:32.250 --> 00:46:36.512 align:middle line:84% And then just now mechanism and this regression discontinuity-- 00:46:36.512 --> 00:46:38.970 align:middle line:84% that's later, maybe in our later sections of the climate we 00:46:38.970 --> 00:46:41.070 align:middle line:90% will have done. 00:46:41.070 --> 00:46:43.800 align:middle line:84% And Wilson said the instrumental variable, 00:46:43.800 --> 00:46:46.530 align:middle line:84% and another is called difference-in-differences. 00:46:46.530 --> 00:46:48.600 align:middle line:90% So this one, DID-- 00:46:48.600 --> 00:46:49.740 align:middle line:90% always use. 00:46:49.740 --> 00:46:51.150 align:middle line:90% I'm a DID master. 00:46:51.150 --> 00:46:53.700 align:middle line:84% I always use DID, and I will teach you 00:46:53.700 --> 00:46:57.300 align:middle line:90% how to do DID for climate-- 00:46:57.300 --> 00:47:00.480 align:middle line:84% the impact of climate risk on housing prices. 00:47:00.480 --> 00:47:01.720 align:middle line:90% That's for later session. 00:47:01.720 --> 00:47:07.440 align:middle line:84% But today, matching is very easy, and not that difficult 00:47:07.440 --> 00:47:08.910 align:middle line:90% one to understand. 00:47:08.910 --> 00:47:13.140 align:middle line:84% But the golden way is to do experiment. 00:47:13.140 --> 00:47:17.050 align:middle line:84% So experiment-- let me give you a simple idea of the experiment. 00:47:17.050 --> 00:47:20.670 align:middle line:84% Remember, once-- lecture on healthy buildings. 00:47:20.670 --> 00:47:22.980 align:middle line:84% I think I mentioned it will be the healthy buildings. 00:47:22.980 --> 00:47:25.760 align:middle line:84% So, for example, we have a very healthy building-- 00:47:25.760 --> 00:47:30.100 align:middle line:84% CO2 concentration's 800 PPM is rather healthy. 00:47:30.100 --> 00:47:36.940 align:middle line:84% This is unhealthy-- all the CO2 concentration. 00:47:36.940 --> 00:47:39.370 align:middle line:90% Then you have students. 00:47:39.370 --> 00:47:41.350 align:middle line:90% You have students-- you. 00:47:41.350 --> 00:47:45.130 align:middle line:84% And then we are not going to tell you which are which-- 00:47:45.130 --> 00:47:46.460 align:middle line:90% two classrooms. 00:47:46.460 --> 00:47:48.040 align:middle line:90% One is healthy. 00:47:48.040 --> 00:47:49.580 align:middle line:90% The other is unhealthy. 00:47:49.580 --> 00:47:54.610 align:middle line:84% We randomly allocate half of you to one classroom, the other half 00:47:54.610 --> 00:47:56.060 align:middle line:90% of you to the other classroom. 00:47:56.060 --> 00:47:58.510 align:middle line:84% You don't know, and you don't understand why. 00:47:58.510 --> 00:48:03.890 align:middle line:84% Then we do our final exam, [INAUDIBLE] exam or final exam. 00:48:03.890 --> 00:48:09.160 align:middle line:84% Then you do-- and then we compare. 00:48:09.160 --> 00:48:11.770 align:middle line:84% We come up-- because we randomized all of you 00:48:11.770 --> 00:48:13.390 align:middle line:84% and your other attributes should not 00:48:13.390 --> 00:48:15.170 align:middle line:90% have a systematic difference. 00:48:15.170 --> 00:48:19.620 align:middle line:84% Of course, each of you are different, but when randomized-- 00:48:19.620 --> 00:48:21.430 align:middle line:90% nondifference. 00:48:21.430 --> 00:48:24.650 align:middle line:84% And only difference will be this healthy and unhealthy-- 00:48:24.650 --> 00:48:28.570 align:middle line:84% then we measure these biological things and your heart beat 00:48:28.570 --> 00:48:32.620 align:middle line:84% and all the cognitive performance and your test score. 00:48:32.620 --> 00:48:39.570 align:middle line:84% So this kind of way was used very widely if the researchers 00:48:39.570 --> 00:48:40.620 align:middle line:90% have enough resources. 00:48:40.620 --> 00:48:41.580 align:middle line:90% It's very hard. 00:48:41.580 --> 00:48:46.410 align:middle line:84% How can you get permission from MIT institute 00:48:46.410 --> 00:48:50.280 align:middle line:84% to randomize you put half of you to a very unhealthy place-- 00:48:50.280 --> 00:48:51.620 align:middle line:90% ethical problem. 00:48:51.620 --> 00:48:56.340 align:middle line:84% But if these are manageable, this 00:48:56.340 --> 00:48:58.830 align:middle line:84% is a golden way to get apple-to-apple 00:48:58.830 --> 00:49:02.200 align:middle line:84% because it's totally randomized apple-to-apple. 00:49:02.200 --> 00:49:05.510 align:middle line:90% Then you compare. 00:49:05.510 --> 00:49:11.430 align:middle line:84% For example, primary elementary school-- 00:49:11.430 --> 00:49:13.370 align:middle line:90% then you put some air filters. 00:49:13.370 --> 00:49:15.620 align:middle line:84% For example, it's a developing country 00:49:15.620 --> 00:49:17.510 align:middle line:90% like China-- very polluted. 00:49:17.510 --> 00:49:20.770 align:middle line:84% And then for elementary school, it's many classrooms. 00:49:20.770 --> 00:49:25.520 align:middle line:84% You put some very good air filters in some classrooms, 00:49:25.520 --> 00:49:27.830 align:middle line:90% help clean the air inside. 00:49:27.830 --> 00:49:30.170 align:middle line:90% But others, we call it placebo-- 00:49:30.170 --> 00:49:33.080 align:middle line:84% also same machine, and the kids, they 00:49:33.080 --> 00:49:34.520 align:middle line:90% don't understand the difference. 00:49:34.520 --> 00:49:38.600 align:middle line:84% But it's not a real filter, but it's placebo. 00:49:38.600 --> 00:49:41.510 align:middle line:84% And then you run them, put the students in those classrooms. 00:49:41.510 --> 00:49:43.110 align:middle line:90% After some days, you observe. 00:49:43.110 --> 00:49:45.020 align:middle line:90% So that's called the experiment. 00:49:45.020 --> 00:49:46.380 align:middle line:90% That's so hard. 00:49:46.380 --> 00:49:50.270 align:middle line:84% You must have the collaboration with the elementary school, 00:49:50.270 --> 00:49:52.220 align:middle line:84% ministry of education, [INAUDIBLE] 00:49:52.220 --> 00:49:56.280 align:middle line:84% center and the parents and all the things. 00:49:56.280 --> 00:50:00.540 align:middle line:84% So that's why-- well, if you do experiment, 00:50:00.540 --> 00:50:02.400 align:middle line:90% you publish very well. 00:50:02.400 --> 00:50:05.000 align:middle line:84% That's all our scholars' dream to do 00:50:05.000 --> 00:50:08.900 align:middle line:84% is experiment and publish well, but it's so hard. 00:50:08.900 --> 00:50:13.130 align:middle line:84% For example, we did one in pandemic together-- 00:50:13.130 --> 00:50:15.920 align:middle line:84% my lab's team, including several researchers. 00:50:15.920 --> 00:50:21.110 align:middle line:84% And we did one publishing a very top journal, PNAS. 00:50:21.110 --> 00:50:23.180 align:middle line:84% That's basically you randomize people, give them 00:50:23.180 --> 00:50:25.453 align:middle line:84% different information, and observe their behaviors. 00:50:25.453 --> 00:50:26.870 align:middle line:84% I don't want to go to the details, 00:50:26.870 --> 00:50:28.700 align:middle line:90% but I just want to let you know. 00:50:28.700 --> 00:50:33.530 align:middle line:84% Do you know two very famous professors at MIT-- 00:50:33.530 --> 00:50:35.330 align:middle line:90% econ department? 00:50:35.330 --> 00:50:39.440 align:middle line:84% They got Nobel Prize because of this randomized controlled trial 00:50:39.440 --> 00:50:42.120 align:middle line:90% in developing countries. 00:50:42.120 --> 00:50:44.929 align:middle line:90% Who knows their name? 00:50:44.929 --> 00:50:47.890 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:50:47.890 --> 00:50:48.956 align:middle line:90% SIGI ZHENG: Yes. 00:50:48.956 --> 00:50:51.140 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:50:51.140 --> 00:50:52.490 align:middle line:90% SIGI ZHENG: And her husband. 00:50:52.490 --> 00:50:54.300 align:middle line:90% Yes, very good. 00:50:54.300 --> 00:50:56.058 align:middle line:90% How did you know them? 00:50:56.058 --> 00:50:57.810 align:middle line:90% AUDIENCE: [INAUDIBLE] 00:50:57.810 --> 00:51:00.270 align:middle line:84% SIGI ZHENG: Oh, yes, they are doing product in India. 00:51:00.270 --> 00:51:02.540 align:middle line:90% Yeah so there's a couple-- 00:51:02.540 --> 00:51:04.890 align:middle line:90% power couple of two professors. 00:51:04.890 --> 00:51:10.190 align:middle line:84% They are the husband and wife, Duflo and Banerjee. 00:51:10.190 --> 00:51:14.070 align:middle line:90% They won Nobel Prize for this. 00:51:14.070 --> 00:51:15.950 align:middle line:84% They did all kinds of things in India 00:51:15.950 --> 00:51:18.770 align:middle line:90% and other countries-- so golden. 00:51:18.770 --> 00:51:20.220 align:middle line:90% We are common people. 00:51:20.220 --> 00:51:21.740 align:middle line:90% We cannot do that every day. 00:51:21.740 --> 00:51:25.500 align:middle line:84% So we need to rely on other technologies. 00:51:25.500 --> 00:51:28.670 align:middle line:84% Now I'm going to tell you, as a common people, student 00:51:28.670 --> 00:51:34.700 align:middle line:84% or researcher, we don't need to do all those perfect things, 00:51:34.700 --> 00:51:37.750 align:middle line:84% but at least we can do something. 00:51:37.750 --> 00:51:40.130 align:middle line:84% The thing I want you to learn from today's class 00:51:40.130 --> 00:51:41.600 align:middle line:90% is we need to try. 00:51:41.600 --> 00:51:43.040 align:middle line:90% We cannot say I don't care. 00:51:43.040 --> 00:51:45.320 align:middle line:84% I just observed correlation and-- 00:51:45.320 --> 00:51:47.120 align:middle line:90% no, that's not rigorous. 00:51:47.120 --> 00:51:51.050 align:middle line:84% As an MIT student, rigorous-- and say we understand. 00:51:51.050 --> 00:51:53.600 align:middle line:84% If we just do a naive regression, 00:51:53.600 --> 00:51:56.060 align:middle line:84% get a coefficient of a variable green-- 00:51:56.060 --> 00:51:57.110 align:middle line:90% not real. 00:51:57.110 --> 00:51:59.270 align:middle line:84% There are some omitted variables. 00:51:59.270 --> 00:52:02.660 align:middle line:84% Endogeneity problem is not a causal inference. 00:52:02.660 --> 00:52:04.820 align:middle line:90% How to do? 00:52:04.820 --> 00:52:08.450 align:middle line:90% Then I'm going to talk matching. 00:52:08.450 --> 00:52:12.310 align:middle line:90% I have so many. 00:52:12.310 --> 00:52:18.140 align:middle line:84% And then apple-to-apple-- twins, that's perfect. 00:52:18.140 --> 00:52:20.400 align:middle line:90% Some scholars, they did do that. 00:52:20.400 --> 00:52:22.730 align:middle line:90% They tracked the twins. 00:52:22.730 --> 00:52:25.220 align:middle line:90% The twins-- they were born. 00:52:25.220 --> 00:52:27.800 align:middle line:84% Same parents, same environment like that, 00:52:27.800 --> 00:52:31.310 align:middle line:84% and later, different outcomes because of the environment-- 00:52:31.310 --> 00:52:35.720 align:middle line:84% when a boy went to this city, the girl to that city, 00:52:35.720 --> 00:52:39.540 align:middle line:84% and the environment-- then they are long-term, lifetime outcomes 00:52:39.540 --> 00:52:42.140 align:middle line:90% [INAUDIBLE] 00:52:42.140 --> 00:52:44.160 align:middle line:84% One thing is geographic matching. 00:52:44.160 --> 00:52:46.370 align:middle line:84% So matching-- twins are also matching, 00:52:46.370 --> 00:52:48.230 align:middle line:90% but the natural matching-- 00:52:48.230 --> 00:52:51.150 align:middle line:84% and a more advanced called propensity score matching. 00:52:51.150 --> 00:52:54.440 align:middle line:84% I'm going to skip this, but I will talk about a little bit. 00:52:54.440 --> 00:52:58.190 align:middle line:84% PSM is advanced matching, but the thing is geographic 00:52:58.190 --> 00:53:00.380 align:middle line:90% matching that's very well-- 00:53:00.380 --> 00:53:02.750 align:middle line:84% widely used in the real estate market 00:53:02.750 --> 00:53:04.590 align:middle line:90% because we are location things. 00:53:04.590 --> 00:53:07.790 align:middle line:84% So I think some of you already mentioned this. 00:53:07.790 --> 00:53:10.040 align:middle line:84% If there's a green building, you are not 00:53:10.040 --> 00:53:13.040 align:middle line:84% going to compare a green building in this location 00:53:13.040 --> 00:53:15.830 align:middle line:84% to a green building in totally a different location. 00:53:15.830 --> 00:53:18.150 align:middle line:90% Like this is Boston downtown. 00:53:18.150 --> 00:53:19.410 align:middle line:90% This is a green building. 00:53:19.410 --> 00:53:25.190 align:middle line:84% And then you find another place in a suburban someplace-- no. 00:53:25.190 --> 00:53:31.950 align:middle line:84% Compare this green building with nearby nongreen building-- 00:53:31.950 --> 00:53:33.210 align:middle line:90% very close nearby. 00:53:33.210 --> 00:53:34.940 align:middle line:84% That's called geographic matching. 00:53:34.940 --> 00:53:38.780 align:middle line:84% Say, 0.2 square miles-- this circle. 00:53:38.780 --> 00:53:41.860 align:middle line:90% So that was what they did. 00:53:41.860 --> 00:53:44.710 align:middle line:84% Now [INAUDIBLE] paper-- remember, the [INAUDIBLE] paper? 00:53:44.710 --> 00:53:47.360 align:middle line:90% They did this. 00:53:47.360 --> 00:53:51.850 align:middle line:84% After this match, they found-- ooh, become so similar. 00:53:51.850 --> 00:53:54.790 align:middle line:90% The class A both are 75%. 00:53:54.790 --> 00:53:55.990 align:middle line:90% One is 75%. 00:53:55.990 --> 00:53:57.370 align:middle line:90% The other is 71%. 00:53:57.370 --> 00:54:00.677 align:middle line:84% Of course, not perfect matching, but if you 00:54:00.677 --> 00:54:03.010 align:middle line:84% get the green building and there's two brown building so 00:54:03.010 --> 00:54:07.750 align:middle line:84% nearby, so same location, and will be similar, 00:54:07.750 --> 00:54:10.510 align:middle line:84% then they create [INAUDIBLE] not going 00:54:10.510 --> 00:54:14.980 align:middle line:84% to do this, not suburban area to compare-- no. 00:54:14.980 --> 00:54:19.540 align:middle line:84% And nearby buildings-- then this is treated this is control. 00:54:19.540 --> 00:54:20.720 align:middle line:90% Then they compare. 00:54:20.720 --> 00:54:22.780 align:middle line:90% So that's basically matching. 00:54:22.780 --> 00:54:25.030 align:middle line:90% So this is not difficult. 00:54:25.030 --> 00:54:29.050 align:middle line:84% After matching, they say, oh, I'm more confident. 00:54:29.050 --> 00:54:30.930 align:middle line:84% They also say we are more confident. 00:54:30.930 --> 00:54:34.840 align:middle line:84% We are estimating some apple-to-apple thing. 00:54:34.840 --> 00:54:40.760 align:middle line:84% This is green rating [INAUDIBLE] 2.6% premium 00:54:40.760 --> 00:54:44.360 align:middle line:90% and sale price 13.3% premium. 00:54:44.360 --> 00:54:47.300 align:middle line:90% In fact, [INAUDIBLE] is 77.6%. 00:54:47.300 --> 00:54:52.580 align:middle line:84% So after matching, they are more confident on this difference 00:54:52.580 --> 00:54:55.220 align:middle line:90% instead of a random other place. 00:54:55.220 --> 00:54:59.770 align:middle line:90% So that's the matching thing. 00:54:59.770 --> 00:55:04.210 align:middle line:84% Now I'm going to do more of this as-- 00:55:04.210 --> 00:55:05.860 align:middle line:90% think about this. 00:55:05.860 --> 00:55:09.700 align:middle line:84% So now we understand there's so much difficulties. 00:55:09.700 --> 00:55:11.380 align:middle line:90% There's causal inference. 00:55:11.380 --> 00:55:16.510 align:middle line:84% And we need some techniques to do apple-to-apple. 00:55:16.510 --> 00:55:23.260 align:middle line:84% I want to further understand the truth, the root of why 00:55:23.260 --> 00:55:24.890 align:middle line:90% we have problems. 00:55:24.890 --> 00:55:27.950 align:middle line:84% I will go back to this naive example. 00:55:27.950 --> 00:55:30.700 align:middle line:84% This [INAUDIBLE] come up when I introduce hedonic-- 00:55:30.700 --> 00:55:32.540 align:middle line:90% decompose. 00:55:32.540 --> 00:55:35.030 align:middle line:90% Remember, $1 million house-- 00:55:35.030 --> 00:55:39.670 align:middle line:84% 1, 2, 3, 4 components plus a trash can. 00:55:39.670 --> 00:55:42.280 align:middle line:84% Then this hedonic thing-- how to think 00:55:42.280 --> 00:55:45.880 align:middle line:84% about this biasing hedonic setting? 00:55:45.880 --> 00:55:49.240 align:middle line:90% That's the point. 00:55:49.240 --> 00:55:51.475 align:middle line:84% Let me see-- just now, I show this. 00:55:51.475 --> 00:55:52.350 align:middle line:90% This is all the same. 00:55:52.350 --> 00:55:55.950 align:middle line:90% 00:55:55.950 --> 00:56:01.900 align:middle line:84% Now suppose, in the trash can, although we cannot measure, 00:56:01.900 --> 00:56:04.240 align:middle line:84% but we understand from our intuition, 00:56:04.240 --> 00:56:07.325 align:middle line:84% we understand the trash can is something called fancy. 00:56:07.325 --> 00:56:09.797 align:middle line:90% 00:56:09.797 --> 00:56:12.380 align:middle line:84% Fancy, of course, is so hard to measure how fancy-- like this, 00:56:12.380 --> 00:56:12.880 align:middle line:90% like that. 00:56:12.880 --> 00:56:15.820 align:middle line:84% But we know some buildings are fancier. 00:56:15.820 --> 00:56:17.920 align:middle line:84% Some other buildings are less fancy. 00:56:17.920 --> 00:56:19.820 align:middle line:90% And then we cannot measure. 00:56:19.820 --> 00:56:21.640 align:middle line:84% But we know there's something here. 00:56:21.640 --> 00:56:25.450 align:middle line:84% And maybe the fancy thing is correlated with LEED, 00:56:25.450 --> 00:56:29.140 align:middle line:84% like the LEED building, the famous designer, 00:56:29.140 --> 00:56:30.080 align:middle line:90% and the fancier. 00:56:30.080 --> 00:56:33.610 align:middle line:84% Some of you went to New York City, and together with you, 00:56:33.610 --> 00:56:36.140 align:middle line:84% we saw some very fancy buildings. 00:56:36.140 --> 00:56:37.730 align:middle line:90% So these are correlated. 00:56:37.730 --> 00:56:41.890 align:middle line:84% If these are correlated and you cannot observe fancy in that 00:56:41.890 --> 00:56:43.035 align:middle line:90% trash can problem-- 00:56:43.035 --> 00:56:47.230 align:middle line:90% 00:56:47.230 --> 00:56:52.130 align:middle line:84% suppose if we could observe fancy. 00:56:52.130 --> 00:56:54.310 align:middle line:90% That fancy is $0.05 million. 00:56:54.310 --> 00:56:55.690 align:middle line:90% Suppose we could. 00:56:55.690 --> 00:56:56.930 align:middle line:90% That's a hypothetical thing. 00:56:56.930 --> 00:56:57.847 align:middle line:90% Actually, we couldn't. 00:56:57.847 --> 00:56:59.960 align:middle line:90% But suppose. 00:56:59.960 --> 00:57:08.410 align:middle line:90% And then, let me see if we can-- 00:57:08.410 --> 00:57:11.470 align:middle line:84% so fortunately, we can have a matter of fancy-- 00:57:11.470 --> 00:57:13.480 align:middle line:90% then no problem. 00:57:13.480 --> 00:57:17.080 align:middle line:84% We decompose the trash can to a real fancy 00:57:17.080 --> 00:57:20.380 align:middle line:90% and the leftover trash can. 00:57:20.380 --> 00:57:21.270 align:middle line:90% Everything is done. 00:57:21.270 --> 00:57:22.640 align:middle line:90% Good. 00:57:22.640 --> 00:57:25.180 align:middle line:90% Then this one is good. 00:57:25.180 --> 00:57:28.110 align:middle line:84% However, in reality, we cannot observe. 00:57:28.110 --> 00:57:32.520 align:middle line:84% And to the extreme case, suppose the fancy-- basically, 00:57:32.520 --> 00:57:34.470 align:middle line:90% only the buildings are fancier. 00:57:34.470 --> 00:57:37.020 align:middle line:84% Or not only the building-- are nonfancy. 00:57:37.020 --> 00:57:39.573 align:middle line:84% And we couldn't-- then we wrongly. 00:57:39.573 --> 00:57:42.550 align:middle line:90% 00:57:42.550 --> 00:57:45.680 align:middle line:84% overestimate because all the buildings are fancier. 00:57:45.680 --> 00:57:47.890 align:middle line:84% So you will see all the buildings 00:57:47.890 --> 00:57:50.680 align:middle line:90% are $0.15 million more. 00:57:50.680 --> 00:57:53.680 align:middle line:90% But the real LEED is $1.1. 00:57:53.680 --> 00:57:55.150 align:middle line:90% And because the fancier-- 00:57:55.150 --> 00:57:58.300 align:middle line:84% they are so correlated, 100% perfectly correlated, 00:57:58.300 --> 00:58:01.100 align:middle line:84% then all the LEED buildings are fancy buildings. 00:58:01.100 --> 00:58:02.540 align:middle line:90% So you bundle them together. 00:58:02.540 --> 00:58:05.020 align:middle line:84% You cannot separate and you cannot observe. 00:58:05.020 --> 00:58:09.970 align:middle line:84% Then you overestimate the LEED-- become $0.15 million. 00:58:09.970 --> 00:58:11.440 align:middle line:90% Is that clear? 00:58:11.440 --> 00:58:13.480 align:middle line:90% Lots of problem. 00:58:13.480 --> 00:58:18.720 align:middle line:84% So all the things you cannot observe if the residual-- 00:58:18.720 --> 00:58:20.820 align:middle line:84% they are correlated with your interest 00:58:20.820 --> 00:58:22.780 align:middle line:84% of the variable-- variable of interest. 00:58:22.780 --> 00:58:26.490 align:middle line:84% Then you overestimate or underestimate. 00:58:26.490 --> 00:58:28.770 align:middle line:90% This case is overestimate. 00:58:28.770 --> 00:58:30.730 align:middle line:90% So that's a problem. 00:58:30.730 --> 00:58:33.940 align:middle line:90% Any questions here? 00:58:33.940 --> 00:58:38.620 align:middle line:84% So I want to use this to show you because just now I know 00:58:38.620 --> 00:58:41.720 align:middle line:90% it's very abstract in this way. 00:58:41.720 --> 00:58:42.260 align:middle line:90% Let me see. 00:58:42.260 --> 00:58:44.690 align:middle line:90% Yeah, remember this way? 00:58:44.690 --> 00:58:47.570 align:middle line:90% Algebra so abstract. 00:58:47.570 --> 00:58:52.400 align:middle line:84% At that time, I said [INAUDIBLE] the trash can got correlated 00:58:52.400 --> 00:58:55.100 align:middle line:90% with D. The D is a green. 00:58:55.100 --> 00:58:58.080 align:middle line:84% But from algebra, maybe you feel so abstract, 00:58:58.080 --> 00:59:02.717 align:middle line:84% so I'm going to just not use this fancy way to show you. 00:59:02.717 --> 00:59:04.300 align:middle line:84% The trash can-- something in the trash 00:59:04.300 --> 00:59:06.250 align:middle line:84% can got correlated, then problem. 00:59:06.250 --> 00:59:07.240 align:middle line:90% You cannot observe. 00:59:07.240 --> 00:59:08.480 align:middle line:90% You bundle them together. 00:59:08.480 --> 00:59:10.950 align:middle line:84% You estimate a bigger thing, which is not real. 00:59:10.950 --> 00:59:13.760 align:middle line:90% 00:59:13.760 --> 00:59:21.600 align:middle line:84% So the thing-- again, think about this: apple to orange. 00:59:21.600 --> 00:59:26.020 align:middle line:84% Well, as we observe a LEED building and a brown building-- 00:59:26.020 --> 00:59:29.850 align:middle line:84% LEED fancy, brown building not fancy-- 00:59:29.850 --> 00:59:32.430 align:middle line:84% the difference is $0.15 million-- 00:59:32.430 --> 00:59:33.300 align:middle line:90% naive. 00:59:33.300 --> 00:59:36.090 align:middle line:90% This is very naive [INAUDIBLE]. 00:59:36.090 --> 00:59:39.240 align:middle line:84% We need to get this another apple, 00:59:39.240 --> 00:59:40.980 align:middle line:90% which is an apple-to-apple. 00:59:40.980 --> 00:59:45.570 align:middle line:84% And then at first, because this-- 00:59:45.570 --> 00:59:48.780 align:middle line:84% first, this [INAUDIBLE] is like this-- 00:59:48.780 --> 00:59:52.410 align:middle line:84% fancier green, nonfancy, nongreen. 00:59:52.410 --> 00:59:52.990 align:middle line:90% Not right. 00:59:52.990 --> 00:59:56.970 align:middle line:84% And we need to get the counterfactual, which 00:59:56.970 --> 00:59:59.660 align:middle line:90% is also fancy. 00:59:59.660 --> 01:00:02.960 align:middle line:84% And the numbering-- then we calculate real. 01:00:02.960 --> 01:00:05.690 align:middle line:90% And then that's an issue. 01:00:05.690 --> 01:00:06.590 align:middle line:90% It's not that. 01:00:06.590 --> 01:00:08.100 align:middle line:90% Then you match. 01:00:08.100 --> 01:00:10.760 align:middle line:84% Suppose that neighborhood are all fancy-- 01:00:10.760 --> 01:00:12.200 align:middle line:90% that's our assumption. 01:00:12.200 --> 01:00:15.770 align:middle line:84% Suppose geography matching-- we suppose this neighborhood 01:00:15.770 --> 01:00:17.370 align:middle line:90% are all fancy buildings. 01:00:17.370 --> 01:00:18.660 align:middle line:90% So we compare them. 01:00:18.660 --> 01:00:23.840 align:middle line:84% We rule out the fancy factor, and we get the green factor. 01:00:23.840 --> 01:00:27.740 align:middle line:90% I hope you understand. 01:00:27.740 --> 01:00:29.930 align:middle line:84% Now move on a little bit-- given all your-- 01:00:29.930 --> 01:00:34.580 align:middle line:84% if you understand all this, so easy to extend to say this. 01:00:34.580 --> 01:00:41.110 align:middle line:84% That's why I spend so much time [INAUDIBLE] 01:00:41.110 --> 01:00:43.300 align:middle line:90% the same thing-- hedonic. 01:00:43.300 --> 01:00:46.930 align:middle line:84% And then nothing-- remember, just now, 01:00:46.930 --> 01:00:49.570 align:middle line:84% we had all the attributes in X1, which 01:00:49.570 --> 01:00:53.230 align:middle line:84% is building attributes-- green or not, how many rooms, like 01:00:53.230 --> 01:00:54.100 align:middle line:90% that. 01:00:54.100 --> 01:00:57.020 align:middle line:84% There's another factor we call the locational things. 01:00:57.020 --> 01:00:58.270 align:middle line:90% That's also important. 01:00:58.270 --> 01:01:00.710 align:middle line:84% Then you decompose locational things. 01:01:00.710 --> 01:01:05.900 align:middle line:84% You buy a house not because you only need the space. 01:01:05.900 --> 01:01:08.390 align:middle line:84% You also want to be a good school district. 01:01:08.390 --> 01:01:12.040 align:middle line:84% That's why I bought my house in Lexington in the first place. 01:01:12.040 --> 01:01:14.840 align:middle line:84% At that time, I knew nothing about this city. 01:01:14.840 --> 01:01:16.520 align:middle line:90% I asked Azizi for advice. 01:01:16.520 --> 01:01:19.580 align:middle line:84% Then he told me several good towns with good schools-- 01:01:19.580 --> 01:01:20.770 align:middle line:90% one is Lexington. 01:01:20.770 --> 01:01:22.990 align:middle line:90% Then that's why I got Lexington. 01:01:22.990 --> 01:01:26.950 align:middle line:84% I said, my first priority is to send my son to a good school 01:01:26.950 --> 01:01:32.980 align:middle line:84% so that he later can get at least to a college. 01:01:32.980 --> 01:01:34.570 align:middle line:90% So that's school. 01:01:34.570 --> 01:01:37.060 align:middle line:84% Subway-- important transportation. 01:01:37.060 --> 01:01:41.020 align:middle line:90% Green park-- air pollution. 01:01:41.020 --> 01:01:44.390 align:middle line:84% No one wants to live close to a factory, I think, 01:01:44.390 --> 01:01:46.810 align:middle line:84% unless you don't know that's a problem. 01:01:46.810 --> 01:01:51.790 align:middle line:84% So if you can decompose a similar things as what we did-- 01:01:51.790 --> 01:01:55.720 align:middle line:84% green space, clean air, walkability, other things-- 01:01:55.720 --> 01:01:57.260 align:middle line:90% estimate. 01:01:57.260 --> 01:02:05.220 align:middle line:84% These slide can summarize many, many past faculty research-- 01:02:05.220 --> 01:02:09.070 align:middle line:90% our urban studies and planning. 01:02:09.070 --> 01:02:13.570 align:middle line:84% We have a transportation [INAUDIBLE] two professors, 01:02:13.570 --> 01:02:16.570 align:middle line:84% [INAUDIBLE] if they have the data, 01:02:16.570 --> 01:02:20.140 align:middle line:84% they estimate subway is important-- 01:02:20.140 --> 01:02:21.460 align:middle line:90% public transit. 01:02:21.460 --> 01:02:24.820 align:middle line:84% They need to come to Siqi, they don't know how to do-- 01:02:24.820 --> 01:02:27.700 align:middle line:84% they come to Siqi and say, can you give me 01:02:27.700 --> 01:02:29.260 align:middle line:90% some housing transaction data? 01:02:29.260 --> 01:02:30.160 align:middle line:90% No problem. 01:02:30.160 --> 01:02:31.370 align:middle line:90% I give them data. 01:02:31.370 --> 01:02:33.760 align:middle line:84% Then they use housing transactions to run the models, 01:02:33.760 --> 01:02:38.020 align:middle line:84% and they estimate the price premium of the subway station 01:02:38.020 --> 01:02:39.850 align:middle line:90% or bus stop, no matter what-- 01:02:39.850 --> 01:02:40.810 align:middle line:90% premium. 01:02:40.810 --> 01:02:41.440 align:middle line:90% Good. 01:02:41.440 --> 01:02:43.930 align:middle line:90% Then we publish a paper. 01:02:43.930 --> 01:02:47.950 align:middle line:84% Next day, a professor called [INAUDIBLE] came-- 01:02:47.950 --> 01:02:50.210 align:middle line:90% Siqi, I want to work with you. 01:02:50.210 --> 01:02:54.620 align:middle line:84% Housing transaction estimate-- the green parks. 01:02:54.620 --> 01:02:56.080 align:middle line:90% That's a housing effect. 01:02:56.080 --> 01:02:59.020 align:middle line:84% Why people want to live close to park 01:02:59.020 --> 01:03:01.480 align:middle line:84% or [INAUDIBLE] easier to exercise, all the air 01:03:01.480 --> 01:03:03.230 align:middle line:84% pollution-- good for their health. 01:03:03.230 --> 01:03:05.530 align:middle line:90% They have willingness to pay. 01:03:05.530 --> 01:03:09.200 align:middle line:84% Because this is bad thing-- if [INAUDIBLE] bad thing, negative. 01:03:09.200 --> 01:03:11.710 align:middle line:84% It's not like all the coefficient 01:03:11.710 --> 01:03:13.910 align:middle line:90% are positive that matters-- 01:03:13.910 --> 01:03:15.740 align:middle line:84% depends on how you measure that thing. 01:03:15.740 --> 01:03:18.730 align:middle line:84% The positive thing-- we call it positive amenities. 01:03:18.730 --> 01:03:20.950 align:middle line:84% Negative thing, we call it negative amenities. 01:03:20.950 --> 01:03:22.400 align:middle line:84% Air pollution is negative amenity. 01:03:22.400 --> 01:03:24.580 align:middle line:90% So that's why it's negative. 01:03:24.580 --> 01:03:27.580 align:middle line:90% Then we use this to do-- 01:03:27.580 --> 01:03:29.680 align:middle line:90% and have a publish a paper-- 01:03:29.680 --> 01:03:30.530 align:middle line:90% published house. 01:03:30.530 --> 01:03:32.320 align:middle line:90% So that's kind of-- 01:03:32.320 --> 01:03:35.210 align:middle line:84% myself are so interested in school, not just because 01:03:35.210 --> 01:03:39.100 align:middle line:84% of my son, but I just want to understand the school premium 01:03:39.100 --> 01:03:39.920 align:middle line:90% housing price. 01:03:39.920 --> 01:03:42.620 align:middle line:84% I did several papers using China's data. 01:03:42.620 --> 01:03:43.730 align:middle line:90% This is real data. 01:03:43.730 --> 01:03:45.490 align:middle line:84% I estimated a good school district-- 01:03:45.490 --> 01:03:47.030 align:middle line:90% 8% price premium. 01:03:47.030 --> 01:03:48.850 align:middle line:90% So that's my real paper. 01:03:48.850 --> 01:03:53.830 align:middle line:90% So that's all cities. 01:03:53.830 --> 01:03:55.900 align:middle line:84% I just want to showcase my paper. 01:03:55.900 --> 01:03:58.210 align:middle line:84% But you don't need to get into this. 01:03:58.210 --> 01:04:00.640 align:middle line:90% So I just want to tell a story. 01:04:00.640 --> 01:04:03.100 align:middle line:84% At first, I didn't know how to do this. 01:04:03.100 --> 01:04:07.030 align:middle line:84% When I came here, 2005, as a visiting scholar, 01:04:07.030 --> 01:04:09.620 align:middle line:90% I randomly run into a professor. 01:04:09.620 --> 01:04:11.620 align:middle line:90% Now professor is my coauthor. 01:04:11.620 --> 01:04:15.910 align:middle line:84% My coauthor told me you cannot just look at real estate 01:04:15.910 --> 01:04:17.750 align:middle line:90% as a-- inside the real estate. 01:04:17.750 --> 01:04:20.500 align:middle line:84% Real estate is very useful for other things. 01:04:20.500 --> 01:04:24.710 align:middle line:84% Study other things using real estate as a channel, as a tool. 01:04:24.710 --> 01:04:25.820 align:middle line:90% Real estate is a tool. 01:04:25.820 --> 01:04:26.890 align:middle line:90% Housing is a tool-- 01:04:26.890 --> 01:04:31.780 align:middle line:84% instead of your destination because people 01:04:31.780 --> 01:04:33.580 align:middle line:84% choose where to live, where to work, 01:04:33.580 --> 01:04:36.970 align:middle line:84% and where to play through real estate. 01:04:36.970 --> 01:04:39.170 align:middle line:84% And many things you cannot observe. 01:04:39.170 --> 01:04:41.680 align:middle line:84% You use real estate to break out out people's willingness 01:04:41.680 --> 01:04:43.280 align:middle line:90% to pay for so many things. 01:04:43.280 --> 01:04:45.070 align:middle line:84% I think, good, let's work together. 01:04:45.070 --> 01:04:46.810 align:middle line:90% Then we produce this paper. 01:04:46.810 --> 01:04:49.240 align:middle line:84% And that paper, using Beijing transaction data-- 01:04:49.240 --> 01:04:53.740 align:middle line:84% that's many years ago, when I was just graduate student. 01:04:53.740 --> 01:04:55.930 align:middle line:84% So I collect all the data [INAUDIBLE] 01:04:55.930 --> 01:04:58.120 align:middle line:90% with 920 transactions. 01:04:58.120 --> 01:05:01.360 align:middle line:84% At that time, I learned the GIS, and I didn't know. 01:05:01.360 --> 01:05:03.100 align:middle line:90% Now I learned GIS. 01:05:03.100 --> 01:05:06.220 align:middle line:84% These are things I saw at that time. 01:05:06.220 --> 01:05:10.000 align:middle line:84% And land and the subway and bus stop and [INAUDIBLE] subway 01:05:10.000 --> 01:05:12.280 align:middle line:90% stops. 01:05:12.280 --> 01:05:15.610 align:middle line:90% Crime-- high crime areas. 01:05:15.610 --> 01:05:17.150 align:middle line:90% Good schools. 01:05:17.150 --> 01:05:20.530 align:middle line:84% Good schools-- where is concentrated [INAUDIBLE] 01:05:20.530 --> 01:05:22.910 align:middle line:90% and major universities. 01:05:22.910 --> 01:05:24.160 align:middle line:90% Universities are also good. 01:05:24.160 --> 01:05:25.940 align:middle line:90% Everyone want to live close-- 01:05:25.940 --> 01:05:31.030 align:middle line:84% then you have all this human capital and your mental-- 01:05:31.030 --> 01:05:32.530 align:middle line:90% I mean, it's like here. 01:05:32.530 --> 01:05:36.610 align:middle line:84% This air pollution-- bad air pollution, good air pollution, 01:05:36.610 --> 01:05:38.420 align:middle line:90% and the park-- big parks. 01:05:38.420 --> 01:05:39.560 align:middle line:90% I collect all this. 01:05:39.560 --> 01:05:41.290 align:middle line:90% I work day and night. 01:05:41.290 --> 01:05:44.380 align:middle line:84% So that's something I want to tell another small story, when 01:05:44.380 --> 01:05:45.200 align:middle line:90% I was young. 01:05:45.200 --> 01:05:48.400 align:middle line:84% Now I'm not that old, but when I was relatively young, 01:05:48.400 --> 01:05:51.040 align:middle line:84% I worked so hard on these things, 01:05:51.040 --> 01:05:55.180 align:middle line:84% and I stayed up very late to collect data [INAUDIBLE] 01:05:55.180 --> 01:05:56.800 align:middle line:90% run regression. 01:05:56.800 --> 01:05:59.900 align:middle line:90% And I understand the hedonic. 01:05:59.900 --> 01:06:03.910 align:middle line:84% And later, when I become more senior as a professor, 01:06:03.910 --> 01:06:04.990 align:middle line:90% I have my students-- 01:06:04.990 --> 01:06:06.410 align:middle line:90% I ask my students to do. 01:06:06.410 --> 01:06:08.875 align:middle line:90% But I understand the rationale. 01:06:08.875 --> 01:06:11.470 align:middle line:84% And why I tell this story-- because many of you may say, 01:06:11.470 --> 01:06:14.222 align:middle line:90% oh, this is too technical. 01:06:14.222 --> 01:06:15.680 align:middle line:84% Oh, I don't need to run regression, 01:06:15.680 --> 01:06:17.130 align:middle line:90% I don't want to do the p-set. 01:06:17.130 --> 01:06:20.210 align:middle line:84% But if you do the pset right now, 01:06:20.210 --> 01:06:23.780 align:middle line:84% later, you become a senior manager or boss, 01:06:23.780 --> 01:06:27.410 align:middle line:84% and you know how your employees are doing. 01:06:27.410 --> 01:06:31.110 align:middle line:84% You can point out, you didn't consider this variable. 01:06:31.110 --> 01:06:32.730 align:middle line:90% Did you consider fancy? 01:06:32.730 --> 01:06:34.130 align:middle line:90% Did you consider this and that? 01:06:34.130 --> 01:06:35.910 align:middle line:90% This estimate is not right. 01:06:35.910 --> 01:06:40.018 align:middle line:84% I know because back to MIT, I learned that from Siqi's class, 01:06:40.018 --> 01:06:41.310 align:middle line:90% is apples-to-apples comparison. 01:06:41.310 --> 01:06:44.660 align:middle line:84% You are not doing apples-to-apples. 01:06:44.660 --> 01:06:46.430 align:middle line:84% Now don't think that later you will 01:06:46.430 --> 01:06:49.080 align:middle line:84% be so senior you don't need to understand this-- today, 01:06:49.080 --> 01:06:49.925 align:middle line:90% you understand. 01:06:49.925 --> 01:06:53.430 align:middle line:84% Later, you can advise your employees, high level. 01:06:53.430 --> 01:06:55.160 align:middle line:84% So that's why I'm-- now, I'm doing it. 01:06:55.160 --> 01:06:57.300 align:middle line:84% Now I don't want those kind of things. 01:06:57.300 --> 01:07:00.170 align:middle line:84% But because I run this, and I stayed up late 01:07:00.170 --> 01:07:03.230 align:middle line:84% running this 20 years ago, now I understand very well. 01:07:03.230 --> 01:07:05.960 align:middle line:84% Nobody, no my student, can cheat me. 01:07:05.960 --> 01:07:07.320 align:middle line:90% Everything point out. 01:07:07.320 --> 01:07:11.490 align:middle line:84% So that's why I want you to do your p sets. 01:07:11.490 --> 01:07:14.850 align:middle line:84% OK, at that time, I was so junior, the other professor, 01:07:14.850 --> 01:07:17.380 align:middle line:84% so senior, he said, you run this, you do this, you do this. 01:07:17.380 --> 01:07:18.150 align:middle line:90% I did. 01:07:18.150 --> 01:07:20.430 align:middle line:90% And then, I did this. 01:07:20.430 --> 01:07:21.540 align:middle line:90% That's right. 01:07:21.540 --> 01:07:24.280 align:middle line:84% You must go through that process. 01:07:24.280 --> 01:07:25.980 align:middle line:90% Then I collect all this. 01:07:25.980 --> 01:07:28.780 align:middle line:84% At that time, not that easy to get this data from China. 01:07:28.780 --> 01:07:30.560 align:middle line:90% So I got all the data. 01:07:30.560 --> 01:07:33.250 align:middle line:84% Then GISA, [INAUDIBLE] I run regression. 01:07:33.250 --> 01:07:37.810 align:middle line:84% I learned at that time, Stata, not R. So I did this. 01:07:37.810 --> 01:07:41.170 align:middle line:84% OK, finally, I'm not going to talk hard work. 01:07:41.170 --> 01:07:42.940 align:middle line:84% I'm going to talk some fun stuff. 01:07:42.940 --> 01:07:48.580 align:middle line:84% Many of us, we went to New York City with Carlos. 01:07:48.580 --> 01:07:50.780 align:middle line:90% And we did so many projects. 01:07:50.780 --> 01:07:54.010 align:middle line:84% And I went there one night before because I didn't want 01:07:54.010 --> 01:07:55.700 align:middle line:90% to take the bus early morning. 01:07:55.700 --> 01:07:57.280 align:middle line:90% So I went there early before-- 01:07:57.280 --> 01:07:58.220 align:middle line:90% the day before. 01:07:58.220 --> 01:08:01.100 align:middle line:84% I enjoyed some Broadway show and other things. 01:08:01.100 --> 01:08:03.280 align:middle line:84% And the other morning, I walked around. 01:08:03.280 --> 01:08:04.970 align:middle line:90% I walked through this park. 01:08:04.970 --> 01:08:07.850 align:middle line:84% I walked by this park because I knew I'm going to study this. 01:08:07.850 --> 01:08:09.580 align:middle line:90% And the other one is High Line. 01:08:09.580 --> 01:08:12.980 align:middle line:84% So some of you told me, where is the High Line Park? 01:08:12.980 --> 01:08:16.149 align:middle line:90% So I want to use this example. 01:08:16.149 --> 01:08:21.220 align:middle line:84% So this Bryant Park is very close by to our first visit, 01:08:21.220 --> 01:08:24.130 align:middle line:84% Rockefeller Center, I walked through. 01:08:24.130 --> 01:08:26.270 align:middle line:90% And it's not always fancy. 01:08:26.270 --> 01:08:28.272 align:middle line:90% Not always fancy. 01:08:28.272 --> 01:08:29.590 align:middle line:90% I looked through this. 01:08:29.590 --> 01:08:31.010 align:middle line:90% Now, it's fancy. 01:08:31.010 --> 01:08:40.300 align:middle line:84% But back to 1980s, 1970s, it was very poor, poorly managed. 01:08:40.300 --> 01:08:42.050 align:middle line:90% The city ran out of money. 01:08:42.050 --> 01:08:48.060 align:middle line:84% New York city ran out of money of maintaining all the parks 01:08:48.060 --> 01:08:49.229 align:middle line:90% in New York City. 01:08:49.229 --> 01:08:50.640 align:middle line:90% They almost gave up this park. 01:08:50.640 --> 01:08:54.090 align:middle line:90% At that time, crime and the-- 01:08:54.090 --> 01:08:58.664 align:middle line:84% drug, drug trade, and crime, and all this trash 01:08:58.664 --> 01:09:00.040 align:middle line:90% all over the place. 01:09:00.040 --> 01:09:02.450 align:middle line:90% So that suffered so much. 01:09:02.450 --> 01:09:06.220 align:middle line:84% And then, the city ran out of money. 01:09:06.220 --> 01:09:06.720 align:middle line:90% No money. 01:09:06.720 --> 01:09:07.840 align:middle line:90% No public money. 01:09:07.840 --> 01:09:09.229 align:middle line:90% They're going to give up. 01:09:09.229 --> 01:09:13.029 align:middle line:84% Then the people around communities, those buildings, 01:09:13.029 --> 01:09:16.939 align:middle line:84% those buildings, the companies, and some communities nearby, 01:09:16.939 --> 01:09:19.515 align:middle line:90% they self organized together. 01:09:19.515 --> 01:09:21.970 align:middle line:90% They went to the government. 01:09:21.970 --> 01:09:26.000 align:middle line:84% They negotiate with government saying, we can do this. 01:09:26.000 --> 01:09:27.970 align:middle line:90% OK, you don't have money? 01:09:27.970 --> 01:09:28.990 align:middle line:90% We are going to do. 01:09:28.990 --> 01:09:33.970 align:middle line:84% They organize a restoration association of this park 01:09:33.970 --> 01:09:36.490 align:middle line:84% and they collect money, each one contribute a little bit 01:09:36.490 --> 01:09:37.600 align:middle line:90% into a pool. 01:09:37.600 --> 01:09:43.189 align:middle line:84% And they get a 15-year long-term lease from the government-- 01:09:43.189 --> 01:09:44.300 align:middle line:90% from the government. 01:09:44.300 --> 01:09:46.819 align:middle line:84% And then later, they extend to another five years. 01:09:46.819 --> 01:09:49.819 align:middle line:84% And then they immediately shut down the park and they 01:09:49.819 --> 01:09:53.090 align:middle line:84% started the restoration, the redevelopment of that. 01:09:53.090 --> 01:09:57.440 align:middle line:84% And after five years, they opened and become very good-- 01:09:57.440 --> 01:09:58.220 align:middle line:90% very good place. 01:09:58.220 --> 01:10:01.340 align:middle line:84% And people stay here, winter shops, and the summer 01:10:01.340 --> 01:10:03.500 align:middle line:90% all this concert. 01:10:03.500 --> 01:10:08.270 align:middle line:84% And then the nearby housing got appreciation. 01:10:08.270 --> 01:10:09.740 align:middle line:90% So that's a way? 01:10:09.740 --> 01:10:10.945 align:middle line:90% Why, this nearby. 01:10:10.945 --> 01:10:12.320 align:middle line:84% There is a company called green-- 01:10:12.320 --> 01:10:14.990 align:middle line:84% Grace Building Company that's a office building nearby. 01:10:14.990 --> 01:10:17.670 align:middle line:84% At that time, they put in how many million dollars, 01:10:17.670 --> 01:10:22.130 align:middle line:84% I forgot the number, into that pool with the expectation later 01:10:22.130 --> 01:10:25.580 align:middle line:84% become from a form of negative amenity, ugly, 01:10:25.580 --> 01:10:30.512 align:middle line:84% to a positive amenity, increase the appreciation value 01:10:30.512 --> 01:10:31.470 align:middle line:90% of the office building. 01:10:31.470 --> 01:10:32.430 align:middle line:90% They get money back. 01:10:32.430 --> 01:10:35.190 align:middle line:84% So that's why they are willing to put money first in. 01:10:35.190 --> 01:10:37.460 align:middle line:90% So that's a rationale. 01:10:37.460 --> 01:10:40.550 align:middle line:84% Not that all our philosophy [INAUDIBLE] 01:10:40.550 --> 01:10:43.530 align:middle line:84% but also, this price appreciation. 01:10:43.530 --> 01:10:46.790 align:middle line:84% And then they-- in during the recession, 01:10:46.790 --> 01:10:50.840 align:middle line:84% during the recession, the financial crisis, all office 01:10:50.840 --> 01:10:52.610 align:middle line:90% buildings in Manhattan suffered. 01:10:52.610 --> 01:10:55.670 align:middle line:84% But they kept gradually higher occupancy rate, 01:10:55.670 --> 01:10:57.450 align:middle line:90% and higher rent at that time. 01:10:57.450 --> 01:11:00.147 align:middle line:84% So that's example of the private-public partnership. 01:11:00.147 --> 01:11:01.730 align:middle line:84% Then they went to the state, they say, 01:11:01.730 --> 01:11:04.040 align:middle line:90% we continue to run this park. 01:11:04.040 --> 01:11:07.830 align:middle line:84% And we have our civic group to run instead of using money. 01:11:07.830 --> 01:11:11.210 align:middle line:84% And New York City learned this lesson from this park. 01:11:11.210 --> 01:11:15.060 align:middle line:84% And then from that time, they didn't put all their own money. 01:11:15.060 --> 01:11:18.710 align:middle line:84% They always partner with a private sector group, 01:11:18.710 --> 01:11:22.460 align:middle line:84% or the community group to do this maintenance. 01:11:22.460 --> 01:11:25.511 align:middle line:84% I think many of us, we went to New York City 01:11:25.511 --> 01:11:30.530 align:middle line:84% and Kairos brought us to five mixed-use projects, remember. 01:11:30.530 --> 01:11:33.230 align:middle line:84% And we know it's mixed-use projects always 01:11:33.230 --> 01:11:37.040 align:middle line:84% is a collaboration between the government and the city, 01:11:37.040 --> 01:11:40.640 align:middle line:84% between the city and the private developer, and the owner. 01:11:40.640 --> 01:11:43.020 align:middle line:90% So that way. 01:11:43.020 --> 01:11:46.117 align:middle line:84% And then, of course, this is very clear. 01:11:46.117 --> 01:11:48.200 align:middle line:84% And I think it was also mentioned a lot about this 01:11:48.200 --> 01:11:50.390 align:middle line:90% in [INAUDIBLE] if you are there. 01:11:50.390 --> 01:11:55.030 align:middle line:84% But from my side, I want you to understand scientific. 01:11:55.030 --> 01:11:58.023 align:middle line:84% When you think about future, in your mind, 01:11:58.023 --> 01:12:00.190 align:middle line:84% that no matter whether quantitative or quantitative, 01:12:00.190 --> 01:12:02.110 align:middle line:84% in your mind, if you later become a decision 01:12:02.110 --> 01:12:05.770 align:middle line:84% maker, whether there's some group come to you, can 01:12:05.770 --> 01:12:08.800 align:middle line:84% you give us money to the pool, and we will do this 01:12:08.800 --> 01:12:11.860 align:middle line:84% and that, a nearby straight beauty, beauty, find 01:12:11.860 --> 01:12:13.660 align:middle line:90% a nearby neighborhood? 01:12:13.660 --> 01:12:16.210 align:middle line:84% Then how much money you are going to give? 01:12:16.210 --> 01:12:17.680 align:middle line:90% You estimate. 01:12:17.680 --> 01:12:21.640 align:middle line:84% In your mind, you run a quick conceptual-level hedonic model. 01:12:21.640 --> 01:12:22.330 align:middle line:90% Then, OK. 01:12:22.330 --> 01:12:26.930 align:middle line:84% The park here, then the park is green park, better park. 01:12:26.930 --> 01:12:30.800 align:middle line:84% That will increase my property's value by how much? 01:12:30.800 --> 01:12:32.350 align:middle line:90% Then, I compare. 01:12:32.350 --> 01:12:34.760 align:middle line:84% Even conceptual level, that's important. 01:12:34.760 --> 01:12:37.870 align:middle line:84% Instead of say, oh, even the park has nothing to do with me. 01:12:37.870 --> 01:12:43.030 align:middle line:84% Then, that's basically in the business decision making. 01:12:43.030 --> 01:12:45.020 align:middle line:90% I did this. 01:12:45.020 --> 01:12:49.700 align:middle line:84% This is a consultancy project when I was in China. 01:12:49.700 --> 01:12:52.560 align:middle line:84% A city is going to build a subway network. 01:12:52.560 --> 01:12:54.930 align:middle line:84% They are going to partner with the private sector. 01:12:54.930 --> 01:12:58.790 align:middle line:84% It's a Hong Kong subway company to build the subway. 01:12:58.790 --> 01:13:03.500 align:middle line:84% They want to understand how much money they are going to invest, 01:13:03.500 --> 01:13:05.540 align:middle line:84% and how much money they want that Hong Kong 01:13:05.540 --> 01:13:06.780 align:middle line:90% company to invest. 01:13:06.780 --> 01:13:08.750 align:middle line:84% They promise to the Hong Kong company, 01:13:08.750 --> 01:13:11.090 align:middle line:84% if they build the subway, invest, 01:13:11.090 --> 01:13:13.950 align:middle line:84% they will get some land around subway stations. 01:13:13.950 --> 01:13:15.920 align:middle line:90% That's called TOD. 01:13:15.920 --> 01:13:19.310 align:middle line:84% If you invest, the Hong Kong company invest money 01:13:19.310 --> 01:13:22.670 align:middle line:84% to build a subway, they get some land parcels around. 01:13:22.670 --> 01:13:24.810 align:middle line:84% And they can build the real estate later. 01:13:24.810 --> 01:13:27.810 align:middle line:84% So that's always the Hong Kong model of the subway building. 01:13:27.810 --> 01:13:29.390 align:middle line:90% Subway construction. 01:13:29.390 --> 01:13:31.560 align:middle line:90% Then I say, no problem. 01:13:31.560 --> 01:13:33.260 align:middle line:90% Siqi, hedonic master. 01:13:33.260 --> 01:13:34.640 align:middle line:90% I love hedonic. 01:13:34.640 --> 01:13:40.760 align:middle line:84% And I tell them, OK, after the subway, the premium 0.5 01:13:40.760 --> 01:13:43.620 align:middle line:90% kilometer circle, 14%. 01:13:43.620 --> 01:13:50.190 align:middle line:84% 0.5 to 1, 7.5%, we estimate benefit this. 01:13:50.190 --> 01:13:52.740 align:middle line:84% And I estimate if the Hong Kong company can 01:13:52.740 --> 01:13:56.020 align:middle line:84% get land parcels around, they can get a benefit of what? 01:13:56.020 --> 01:13:57.750 align:middle line:84% And then they say, oh, I understand. 01:13:57.750 --> 01:14:00.630 align:middle line:84% The Hong Kong company, you need to invest this money in this 01:14:00.630 --> 01:14:03.030 align:middle line:84% so that I can give you real estate. 01:14:03.030 --> 01:14:06.110 align:middle line:90% So that's a final way. 01:14:06.110 --> 01:14:08.190 align:middle line:90% OK, High Line Park. 01:14:08.190 --> 01:14:09.690 align:middle line:84% At first I learned this from Google. 01:14:09.690 --> 01:14:10.898 align:middle line:90% Then I went to New York city. 01:14:10.898 --> 01:14:12.530 align:middle line:90% You showed me High Line Park. 01:14:12.530 --> 01:14:14.360 align:middle line:90% So that's a thing. 01:14:14.360 --> 01:14:17.090 align:middle line:84% At first, it was relatively now [INAUDIBLE] 01:14:17.090 --> 01:14:22.130 align:middle line:84% area in the west of Chelsea, right? 01:14:22.130 --> 01:14:24.170 align:middle line:90% That's Chelsea, Manhattan. 01:14:24.170 --> 01:14:29.030 align:middle line:84% Some low-end shops around, but some relatively middle 01:14:29.030 --> 01:14:33.080 align:middle line:84% to low-income people, they have small houses nearby. 01:14:33.080 --> 01:14:35.930 align:middle line:84% The city upgrade to High Line Park, 01:14:35.930 --> 01:14:39.710 align:middle line:84% and then think about WPP, finally, WPP. 01:14:39.710 --> 01:14:43.160 align:middle line:90% We go back to WPP. 01:14:43.160 --> 01:14:45.880 align:middle line:90% At first, like this. 01:14:45.880 --> 01:14:46.710 align:middle line:90% This blue line. 01:14:46.710 --> 01:14:47.502 align:middle line:90% Right? 01:14:47.502 --> 01:14:50.600 align:middle line:84% This is High Line Park or any green space. 01:14:50.600 --> 01:14:52.610 align:middle line:90% The blue line is willing to pay. 01:14:52.610 --> 01:14:54.710 align:middle line:90% Closer, pay higher. 01:14:54.710 --> 01:14:56.190 align:middle line:90% The far one is lower. 01:14:56.190 --> 01:14:59.330 align:middle line:84% But because at that time, it's just an abandoned railway, 01:14:59.330 --> 01:15:00.320 align:middle line:90% not that good. 01:15:00.320 --> 01:15:01.910 align:middle line:90% So it's very flat. 01:15:01.910 --> 01:15:05.990 align:middle line:84% Then, the city invests money, build this High Line Park, 01:15:05.990 --> 01:15:08.570 align:middle line:84% and then become so fancy, and then 01:15:08.570 --> 01:15:10.370 align:middle line:90% become the richer household. 01:15:10.370 --> 01:15:14.150 align:middle line:84% They're willing to live nearby and some hotels and restaurants 01:15:14.150 --> 01:15:15.080 align:middle line:90% nearby. 01:15:15.080 --> 01:15:18.260 align:middle line:84% Then willing to [INAUDIBLE] pay this distance to the High Line 01:15:18.260 --> 01:15:20.300 align:middle line:90% Park become like this. 01:15:20.300 --> 01:15:21.790 align:middle line:90% So that's the rationale. 01:15:21.790 --> 01:15:29.460 align:middle line:84% But think about the final angle is environmental gentrification. 01:15:29.460 --> 01:15:33.030 align:middle line:84% So the good thing is the city become beautiful. 01:15:33.030 --> 01:15:34.470 align:middle line:90% High Line Park. 01:15:34.470 --> 01:15:40.020 align:middle line:84% However, inevitably, the red line, rich people's 01:15:40.020 --> 01:15:42.600 align:middle line:84% willingness to pay to be close, much higher 01:15:42.600 --> 01:15:44.310 align:middle line:90% than the original residents'. 01:15:44.310 --> 01:15:45.640 align:middle line:90% They bought in. 01:15:45.640 --> 01:15:47.080 align:middle line:90% They squeezed out. 01:15:47.080 --> 01:15:48.930 align:middle line:90% They pushed out others. 01:15:48.930 --> 01:15:51.450 align:middle line:90% So it's called gentrification. 01:15:51.450 --> 01:15:53.520 align:middle line:90% So it's always no free lunch. 01:15:53.520 --> 01:15:57.300 align:middle line:84% If there are green, amenities, gentrification. 01:15:57.300 --> 01:16:02.100 align:middle line:84% Rich people who have higher willingness to pay occupy. 01:16:02.100 --> 01:16:05.460 align:middle line:84% Poor people move out from small shops 01:16:05.460 --> 01:16:08.640 align:middle line:84% into fancy restaurant and hotel, five-star hotels. 01:16:08.640 --> 01:16:11.130 align:middle line:90% So that's a debate. 01:16:11.130 --> 01:16:12.990 align:middle line:90% Later, we can discuss this. 01:16:12.990 --> 01:16:15.990 align:middle line:84% But that's a debate, and not like, oh, everything is so good, 01:16:15.990 --> 01:16:17.950 align:middle line:84% we invest, then housing price go up. 01:16:17.950 --> 01:16:20.340 align:middle line:84% If you go to this urban planning department, 01:16:20.340 --> 01:16:21.690 align:middle line:90% you need to think about this. 01:16:21.690 --> 01:16:23.610 align:middle line:90% Gentrification. 01:16:23.610 --> 01:16:26.830 align:middle line:84% OK, so I think that's all I might finish. 01:16:26.830 --> 01:16:28.200 align:middle line:90% And I'm going to give this. 01:16:28.200 --> 01:16:30.070 align:middle line:90% Thank you so much. 01:16:30.070 --> 01:16:33.060 align:middle line:84% And I will give it to three people on the front line 01:16:33.060 --> 01:16:36.240 align:middle line:90% because I want you to sit front. 01:16:36.240 --> 01:16:38.480 align:middle line:90% OK, see you next time. 01:16:38.480 --> 01:16:54.000 align:middle line:90%