1 00:00:00,090 --> 00:00:02,490 The following content is provided under a Creative 2 00:00:02,490 --> 00:00:04,059 Commons license. 3 00:00:04,059 --> 00:00:06,330 Your support will help MIT OpenCourseWare 4 00:00:06,330 --> 00:00:10,720 continue to offer high quality educational resources for free. 5 00:00:10,720 --> 00:00:13,350 To make a donation or view additional materials 6 00:00:13,350 --> 00:00:17,290 from hundreds of MIT courses, visit MIT OpenCourseWare 7 00:00:17,290 --> 00:00:18,480 at ocw.mit.edu. 8 00:00:28,940 --> 00:00:36,860 ROBERT TOWNSEND: So this is about evaluating microcredit. 9 00:00:36,860 --> 00:00:38,220 And how to do it. 10 00:00:38,220 --> 00:00:42,230 It's also a methods lecture in some sense, 11 00:00:42,230 --> 00:00:47,120 because it's going to be an application where 12 00:00:47,120 --> 00:00:50,990 it was done both as reduced form, IV, as well 13 00:00:50,990 --> 00:00:52,640 as structural. 14 00:00:52,640 --> 00:00:54,890 So that gives us this great opportunity 15 00:00:54,890 --> 00:00:59,540 to think about pros and cons, pluses and minuses, kind 16 00:00:59,540 --> 00:01:02,360 of what you can get from one and not from the other. 17 00:01:05,090 --> 00:01:08,580 And this is about a financial intervention. 18 00:01:08,580 --> 00:01:12,790 So we have been consistent in every single lecture 19 00:01:12,790 --> 00:01:15,350 in talking about financial constraints 20 00:01:15,350 --> 00:01:21,500 and financial modeling and talking about policy and policy 21 00:01:21,500 --> 00:01:24,810 implications of changing the financial system. 22 00:01:24,810 --> 00:01:28,190 So this is an illustrative to application 23 00:01:28,190 --> 00:01:30,440 of that general theme. 24 00:01:43,348 --> 00:01:44,890 Very little review of the literature, 25 00:01:44,890 --> 00:01:48,890 but there are two, three applications-- actually 26 00:01:48,890 --> 00:01:56,360 the truth is there is very few RCT evaluations of microcredit. 27 00:01:56,360 --> 00:02:00,490 Abhijit, Esther, Cynthia, and Rachel 28 00:02:00,490 --> 00:02:03,880 did one in Hyderabad, Spandana. 29 00:02:07,150 --> 00:02:09,610 And in many ways, the results are 30 00:02:09,610 --> 00:02:13,290 strikingly similar to what we're going 31 00:02:13,290 --> 00:02:16,565 to find in the Thai villages and even some similarity 32 00:02:16,565 --> 00:02:17,065 to Morocco. 33 00:02:19,770 --> 00:02:22,500 What I want to emphasize, and they do too, 34 00:02:22,500 --> 00:02:24,600 is the heterogeneity. 35 00:02:24,600 --> 00:02:27,350 So, for example, in Hyderabad, there's 36 00:02:27,350 --> 00:02:30,100 group-based lending, joint liability. 37 00:02:30,100 --> 00:02:33,130 They're small loans, female borrowers, 38 00:02:33,130 --> 00:02:36,400 subsidized low rates. 39 00:02:36,400 --> 00:02:40,240 On average, to be put in italics, 40 00:02:40,240 --> 00:02:45,100 there's no impact on total expenditures. 41 00:02:45,100 --> 00:02:48,930 But on average an increase in durables in the short run. 42 00:02:48,930 --> 00:02:52,390 And you might stop there, and that would be a mistake, 43 00:02:52,390 --> 00:02:54,190 because the really interesting stuff 44 00:02:54,190 --> 00:02:57,520 gets going when you start looking at the people who 45 00:02:57,520 --> 00:03:03,460 are running or might run businesses, business creation, 46 00:03:03,460 --> 00:03:09,010 business assets, self-employed hours and profits increase 47 00:03:09,010 --> 00:03:12,580 for those who already had existing businesses. 48 00:03:15,260 --> 00:03:19,480 Their durable consumption goes up. 49 00:03:19,480 --> 00:03:21,430 Non-durable does not. 50 00:03:21,430 --> 00:03:26,790 Now, the consumption part is similar to the aggregate. 51 00:03:26,790 --> 00:03:34,590 But if you look, at say, the those 52 00:03:34,590 --> 00:03:37,770 who have as in a probit, a high propensity 53 00:03:37,770 --> 00:03:42,450 to start new businesses in the sample, 54 00:03:42,450 --> 00:03:44,460 they increase their durable goods spending. 55 00:03:44,460 --> 00:03:47,150 But they decrease their non-durable goods spending. 56 00:03:47,150 --> 00:03:50,580 So you are seeing something constant on the average 57 00:03:50,580 --> 00:03:52,180 and going down for one group. 58 00:03:52,180 --> 00:03:55,710 And in fact, if you look at the flip side, which is not 59 00:03:55,710 --> 00:03:58,530 surprising given the average is what it is, 60 00:03:58,530 --> 00:04:00,420 the low propensity business owners 61 00:04:00,420 --> 00:04:05,040 actually increase their consumption of non-durables. 62 00:04:05,040 --> 00:04:11,160 So various heterogeneous impact based, by the way, 63 00:04:11,160 --> 00:04:16,589 on observables, namely current business owners, 64 00:04:16,589 --> 00:04:18,510 and unobservables that are backed out 65 00:04:18,510 --> 00:04:23,090 through these propensity scores, as in probits. 66 00:04:23,090 --> 00:04:27,720 So it's unobservable as well as observable heterogeneity 67 00:04:27,720 --> 00:04:28,860 that seems to matter. 68 00:04:31,950 --> 00:04:37,320 Morocco similar, larger loans, lending to men. 69 00:04:41,210 --> 00:04:43,920 No effects on consumption. 70 00:04:43,920 --> 00:04:50,240 Maybe some small reduction in consumption for those doing 71 00:04:50,240 --> 00:04:52,340 agriculture and livestock. 72 00:04:52,340 --> 00:04:58,340 No effect on business creation, but within existing businesses, 73 00:04:58,340 --> 00:05:02,570 certain activities were discontinued less often, 74 00:05:02,570 --> 00:05:04,700 and the scale of other activities, 75 00:05:04,700 --> 00:05:09,860 as in agriculture and livestock, increase. 76 00:05:09,860 --> 00:05:13,400 So again, another aspect, one, the consumption aspect 77 00:05:13,400 --> 00:05:17,390 is kind of similar in the sense that maybe for these guys that 78 00:05:17,390 --> 00:05:20,450 seem to have some business impact, 79 00:05:20,450 --> 00:05:24,200 consumption might be going down instead 80 00:05:24,200 --> 00:05:26,120 of on average staying constant. 81 00:05:26,120 --> 00:05:29,900 And then you have this kind of nebulous effect on businesses. 82 00:05:29,900 --> 00:05:33,260 A lot of people think of microcredit 83 00:05:33,260 --> 00:05:36,590 traditionally as an impetus to business creation. 84 00:05:36,590 --> 00:05:39,920 It's to allow talented, if poor, people 85 00:05:39,920 --> 00:05:43,400 to enter and expand and also enter into business. 86 00:05:43,400 --> 00:05:50,090 And, in fact, some of the early micro founded macro models 87 00:05:50,090 --> 00:05:52,520 that we went through in class very much feature 88 00:05:52,520 --> 00:05:55,550 that vehicle-- 89 00:05:55,550 --> 00:05:59,480 people going into business who couldn't previously, 90 00:05:59,480 --> 00:06:04,652 as their wealth expanded, you know, driving aggregate TFP. 91 00:06:04,652 --> 00:06:09,770 Now, that's not to say those studies were wrong. 92 00:06:09,770 --> 00:06:11,750 These things can vary across countries. 93 00:06:11,750 --> 00:06:14,240 And they can vary within a given country, 94 00:06:14,240 --> 00:06:17,220 according to kind of the state of the financial system. 95 00:06:17,220 --> 00:06:21,830 And the overall development path. 96 00:06:21,830 --> 00:06:25,220 But we'll see, again, in the Thai study, this a bit 97 00:06:25,220 --> 00:06:26,930 of a struggle to figure out what's 98 00:06:26,930 --> 00:06:30,410 going on within existing businesses that's allowing them 99 00:06:30,410 --> 00:06:32,780 to be more profitable. 100 00:06:32,780 --> 00:06:37,850 And that durable goods things, you know, again, 101 00:06:37,850 --> 00:06:39,920 as we've studied in the financial accounts, 102 00:06:39,920 --> 00:06:42,030 we make decisions about classification. 103 00:06:42,030 --> 00:06:46,880 But some durable goods are used not just as a service flow 104 00:06:46,880 --> 00:06:50,180 for household consumption, but also potentially 105 00:06:50,180 --> 00:06:51,420 used in the business. 106 00:06:51,420 --> 00:06:54,320 And if there are sort of increase in durables 107 00:06:54,320 --> 00:06:56,960 might suggest, they're actually expanding 108 00:06:56,960 --> 00:07:00,980 the business in some way, not just taking it out 109 00:07:00,980 --> 00:07:02,905 in terms of increased consumption. 110 00:07:05,570 --> 00:07:10,990 So in Thailand, it's the Million Baht Fund. 111 00:07:10,990 --> 00:07:14,100 It was a government intervention. 112 00:07:14,100 --> 00:07:18,155 It was not a randomized control trial. 113 00:07:23,390 --> 00:07:26,435 And I guess I'll deal with that right away. 114 00:07:29,030 --> 00:07:33,050 How are we doing an evaluation as if it were 115 00:07:33,050 --> 00:07:34,490 a randomized control trial? 116 00:07:34,490 --> 00:07:39,200 Well, it was basically a million baht per village. 117 00:07:39,200 --> 00:07:43,280 Villages are well-defined geopolitical entities. 118 00:07:43,280 --> 00:07:47,670 But they vary in terms of total population size. 119 00:07:47,670 --> 00:07:50,000 So all villages got the same amount of credit. 120 00:07:50,000 --> 00:07:51,800 The villages with fewer households 121 00:07:51,800 --> 00:07:55,230 got a higher per capita treatment. 122 00:07:55,230 --> 00:08:01,830 And there were 72,000 villages, by the way, in Thailand. 123 00:08:01,830 --> 00:08:07,050 And they all got something in the order of $24,000, $25,000. 124 00:08:07,050 --> 00:08:12,070 So this was approximately 1.5% percent of GDP. 125 00:08:12,070 --> 00:08:17,190 So you could easily claim it's the world's largest microcredit 126 00:08:17,190 --> 00:08:20,040 sort of expansion and evaluation. 127 00:08:20,040 --> 00:08:23,310 Now, it's a quasi-experimental design. 128 00:08:23,310 --> 00:08:26,760 You'll see in the slides, but I'll say it now, 129 00:08:26,760 --> 00:08:30,720 you know, we carefully look at pre-intervention trends 130 00:08:30,720 --> 00:08:35,510 to see whether there's anything else going on systematically 131 00:08:35,510 --> 00:08:39,360 in a comparison between small and large villages. 132 00:08:39,360 --> 00:08:41,280 And we can't find much. 133 00:08:41,280 --> 00:08:44,400 We've even gotten the maps out and used this community 134 00:08:44,400 --> 00:08:48,570 development department data for all the villages in Thailand 135 00:08:48,570 --> 00:08:53,460 and looked at village size to see if, for example, larger 136 00:08:53,460 --> 00:08:58,440 villages are near urban areas or nearer to main roads and so on. 137 00:08:58,440 --> 00:08:59,820 They're not. 138 00:08:59,820 --> 00:09:05,160 We only find one or two variables that are correlated. 139 00:09:05,160 --> 00:09:08,340 And you kind of expect if you're searching 140 00:09:08,340 --> 00:09:11,400 over hundreds of variables to find a few things now and then. 141 00:09:14,440 --> 00:09:16,470 So we're going to look at the effect of credit 142 00:09:16,470 --> 00:09:20,340 on consumption, investment, income. 143 00:09:20,340 --> 00:09:23,200 And we're also going to look at some general equilibrium 144 00:09:23,200 --> 00:09:23,700 effects. 145 00:09:27,440 --> 00:09:36,430 And again, just for worry that I'll forget to say it later, 146 00:09:36,430 --> 00:09:39,760 if you view villages as small open economies-- 147 00:09:39,760 --> 00:09:43,270 and we have seen papers on that-- 148 00:09:43,270 --> 00:09:45,870 then by intervening village by village, 149 00:09:45,870 --> 00:09:48,250 there's some sense in which you're intervening country 150 00:09:48,250 --> 00:09:49,840 by country. 151 00:09:49,840 --> 00:09:53,250 That's pretty rare opportunity. 152 00:09:53,250 --> 00:09:56,280 These models we've been dealing with 153 00:09:56,280 --> 00:10:00,270 take a stand on what might happen with wages, for example, 154 00:10:00,270 --> 00:10:02,820 in general equilibrium. 155 00:10:02,820 --> 00:10:07,580 So we can actually see if there's a wage effect going on 156 00:10:07,580 --> 00:10:09,680 over and above the potential benefits 157 00:10:09,680 --> 00:10:14,240 or other impacts on households who got credit. 158 00:10:14,240 --> 00:10:16,110 Households who don't even get credit, 159 00:10:16,110 --> 00:10:21,140 but live in a village where business might be expanding 160 00:10:21,140 --> 00:10:26,120 would be villages given certain obstacles to total migration, 161 00:10:26,120 --> 00:10:28,510 would be villages where the wages are going up. 162 00:10:28,510 --> 00:10:33,440 And again, that's what we saw in those first few lectures 163 00:10:33,440 --> 00:10:36,080 to be the impact of the expansion 164 00:10:36,080 --> 00:10:37,730 of the financial system. 165 00:10:37,730 --> 00:10:38,590 Question? 166 00:10:38,590 --> 00:10:42,680 AUDIENCE: So I guess, ex ante, did it surprise you 167 00:10:42,680 --> 00:10:44,955 that there was no correlation between things 168 00:10:44,955 --> 00:10:51,570 like village size and proximity to resources or productivity 169 00:10:51,570 --> 00:10:53,530 of the average business in the village or-- 170 00:10:53,530 --> 00:10:56,960 ROBERT TOWNSEND: One thing, it's a political decision. 171 00:10:56,960 --> 00:11:00,650 And they divide and subdivide villages. 172 00:11:00,650 --> 00:11:04,770 But they don't do it on a regular, persistent basis. 173 00:11:04,770 --> 00:11:06,950 So if a village were getting bigger and bigger 174 00:11:06,950 --> 00:11:09,680 for economic reasons, then it could 175 00:11:09,680 --> 00:11:13,720 get divided by the government into two separate villages. 176 00:11:13,720 --> 00:11:17,960 AUDIENCE: So what is the measure that you looked down? 177 00:11:17,960 --> 00:11:20,363 The one decided by the government? 178 00:11:20,363 --> 00:11:21,280 ROBERT TOWNSEND: Yeah. 179 00:11:21,280 --> 00:11:24,970 At a moment in time, we have the identity 180 00:11:24,970 --> 00:11:27,190 of each of the villages. 181 00:11:27,190 --> 00:11:29,890 And we count the number of households in the village. 182 00:11:29,890 --> 00:11:31,410 AUDIENCE: So potentially something 183 00:11:31,410 --> 00:11:34,390 that we count as two villages is actually one big village? 184 00:11:34,390 --> 00:11:38,320 ROBERT TOWNSEND: Well, so the real question 185 00:11:38,320 --> 00:11:40,330 is whether the funds could have leaked over 186 00:11:40,330 --> 00:11:42,360 to neighboring villages. 187 00:11:42,360 --> 00:11:47,380 The grant that established this village-level institution 188 00:11:47,380 --> 00:11:50,200 was for a village. 189 00:11:50,200 --> 00:11:54,640 You know, like Downtown Crossing gets a village fund. 190 00:11:54,640 --> 00:11:56,320 Beacon Hill doesn't get one. 191 00:11:56,320 --> 00:11:58,262 Sorry, close to home. 192 00:11:58,262 --> 00:12:01,102 AUDIENCE: There's huge incentive not share it, right? 193 00:12:01,102 --> 00:12:02,560 I mean, it belongs to your village. 194 00:12:02,560 --> 00:12:03,935 Why would you want someone else-- 195 00:12:03,935 --> 00:12:06,265 AUDIENCE: No, no, what I mugging, 196 00:12:06,265 --> 00:12:09,310 is I think sort of agricultural or resource economics 197 00:12:09,310 --> 00:12:11,230 suggests that village size should 198 00:12:11,230 --> 00:12:15,560 be endogenous to resources in an area. 199 00:12:15,560 --> 00:12:16,990 Like there should be more villages 200 00:12:16,990 --> 00:12:19,140 near freshwater source, like in, you know, 201 00:12:19,140 --> 00:12:20,950 maybe a thousand years ago or something. 202 00:12:20,950 --> 00:12:23,200 ROBERT TOWNSEND: Well, but then adjust for population. 203 00:12:23,200 --> 00:12:26,200 Actually, I think your argument goes the other way. 204 00:12:26,200 --> 00:12:27,700 AUDIENCE: OK, sure, I mean-- 205 00:12:27,700 --> 00:12:29,470 ROBERT TOWNSEND: If villages are an arbitrary collection 206 00:12:29,470 --> 00:12:31,600 of people, then you might want to have the same number 207 00:12:31,600 --> 00:12:32,433 in each one of them. 208 00:12:32,433 --> 00:12:34,330 There could be more villages. 209 00:12:34,330 --> 00:12:37,960 But potentially there are same number of people per village 210 00:12:37,960 --> 00:12:41,200 if the government has some systematic rule. 211 00:12:41,200 --> 00:12:44,800 Now, the government is messing around, but not as 212 00:12:44,800 --> 00:12:46,540 systematically. 213 00:12:46,540 --> 00:12:47,710 That's one comment. 214 00:12:47,710 --> 00:12:52,870 The other thing is this leaking, because if I could get credit 215 00:12:52,870 --> 00:12:55,330 from some other village and/or, as we've 216 00:12:55,330 --> 00:12:58,270 seen with the informal kinship networks, 217 00:12:58,270 --> 00:13:01,030 get more money from someone else who is 218 00:13:01,030 --> 00:13:03,160 connected to another village, then 219 00:13:03,160 --> 00:13:05,470 we wouldn't have any power. 220 00:13:05,470 --> 00:13:07,880 Now, we have looked at that too. 221 00:13:07,880 --> 00:13:10,080 And again, we've done the GIS. 222 00:13:10,080 --> 00:13:12,550 We've gotten all the maps out and looked 223 00:13:12,550 --> 00:13:16,000 at the impact on villages nearby, 224 00:13:16,000 --> 00:13:19,930 creating these sort of smooth neighborhood averages. 225 00:13:19,930 --> 00:13:26,740 And at least early on, the impact doesn't spill over. 226 00:13:26,740 --> 00:13:29,600 But that doesn't say in the long run it might spill over, 227 00:13:29,600 --> 00:13:32,150 which is another interesting potential thing to do. 228 00:13:32,150 --> 00:13:32,650 Yeah? 229 00:13:32,650 --> 00:13:34,510 AUDIENCE: I have a question that's similar-- 230 00:13:34,510 --> 00:13:39,136 how essentially villages could be subdivided cities, 231 00:13:39,136 --> 00:13:42,876 how you can consider them small-- 232 00:13:42,876 --> 00:13:44,626 ROBERT TOWNSEND: Yeah, have I answered it? 233 00:13:44,626 --> 00:13:45,376 AUDIENCE: I think. 234 00:13:45,376 --> 00:13:46,209 ROBERT TOWNSEND: OK. 235 00:13:46,209 --> 00:13:48,180 AUDIENCE: Well, I still don't get why you can, 236 00:13:48,180 --> 00:13:51,710 but if empirically it looks like there aren't spillover. 237 00:13:51,710 --> 00:13:53,580 I guess that's somewhat of a justification. 238 00:13:53,580 --> 00:13:57,110 AUDIENCE: Yeah, I accept what you're saying is true. 239 00:13:57,110 --> 00:14:00,380 But it confuses me from like a priori point of view 240 00:14:00,380 --> 00:14:02,130 why is that just-- 241 00:14:02,130 --> 00:14:03,803 it seams-- 242 00:14:03,803 --> 00:14:05,970 ROBERT TOWNSEND: Well, we didn't take it for granted 243 00:14:05,970 --> 00:14:08,070 and get 3/4 of the way through the study 244 00:14:08,070 --> 00:14:09,070 and then think about it. 245 00:14:09,070 --> 00:14:09,780 AUDIENCE: No, no, I know-- 246 00:14:09,780 --> 00:14:11,400 ROBERT TOWNSEND: We looked right away, 247 00:14:11,400 --> 00:14:12,270 and I would have thought-- 248 00:14:12,270 --> 00:14:12,810 AUDIENCE: I can tell from-- 249 00:14:12,810 --> 00:14:14,790 ROBERT TOWNSEND: My hunch was larger villages 250 00:14:14,790 --> 00:14:18,507 would be near the cities and the main roads. 251 00:14:18,507 --> 00:14:20,590 AUDIENCE: Yeah, it might have a higher average TFP 252 00:14:20,590 --> 00:14:21,420 or something. 253 00:14:21,420 --> 00:14:22,800 ROBERT TOWNSEND: There's a map in the paper. 254 00:14:22,800 --> 00:14:24,780 And, you know, I'd created these slides kind 255 00:14:24,780 --> 00:14:26,140 of too late to include it. 256 00:14:26,140 --> 00:14:28,140 But you'll be reassured by the map. 257 00:14:28,140 --> 00:14:29,470 AUDIENCE: No, I believe you. 258 00:14:29,470 --> 00:14:30,428 I am just confused by-- 259 00:14:30,428 --> 00:14:33,210 ROBERT TOWNSEND: You can rub your hands on it and-- 260 00:14:33,210 --> 00:14:34,980 AUDIENCE: Some sort of that's true. 261 00:14:34,980 --> 00:14:37,220 But it's not true, like you have some sort of hubs 262 00:14:37,220 --> 00:14:44,350 all scattered around, so like the size, you know-- 263 00:14:44,350 --> 00:14:48,420 AUDIENCE: So the one thing is that if it's really urbanized-- 264 00:14:48,420 --> 00:14:49,660 so there's a limit. 265 00:14:49,660 --> 00:14:51,630 Once you reach a certain population, 266 00:14:51,630 --> 00:14:54,520 so this takes care of some of the heterogeneity. 267 00:14:54,520 --> 00:14:56,640 Once the population reaches a certain size, 268 00:14:56,640 --> 00:14:59,555 like the government will just designate it a town instead, 269 00:14:59,555 --> 00:15:00,755 so it won't be a village. 270 00:15:00,755 --> 00:15:01,860 AUDIENCE: Sure, that would take care of some of it. 271 00:15:01,860 --> 00:15:03,540 But it shouldn't take care of all of it. 272 00:15:03,540 --> 00:15:06,450 There should still be some variation in the boundary. 273 00:15:06,450 --> 00:15:07,435 Well, one would there would be variation. 274 00:15:07,435 --> 00:15:09,018 ROBERT TOWNSEND: It would be depend on 275 00:15:09,018 --> 00:15:12,050 how quickly they subdivide them and so on and so forth. 276 00:15:12,050 --> 00:15:13,050 AUDIENCE: That's true. 277 00:15:13,050 --> 00:15:16,216 ROBERT TOWNSEND: We just looked empirically and got reassured. 278 00:15:16,216 --> 00:15:17,590 Yes. 279 00:15:17,590 --> 00:15:20,750 AUDIENCE: A related question is sort of who has the incentives 280 00:15:20,750 --> 00:15:22,970 to subdivide up a village? 281 00:15:22,970 --> 00:15:25,460 Like have there been other sort of village-level programs 282 00:15:25,460 --> 00:15:28,278 in the past which-- 283 00:15:28,278 --> 00:15:29,820 I'm not sure would got the incentive. 284 00:15:29,820 --> 00:15:32,050 But if predicted something like Million Baht Village 285 00:15:32,050 --> 00:15:34,630 was coming, you'd want to subdivide your village. 286 00:15:34,630 --> 00:15:37,800 ROBERT TOWNSEND: Yeah, so on that, 287 00:15:37,800 --> 00:15:42,810 the program was announced pretty suddenly and implemented. 288 00:15:42,810 --> 00:15:46,620 Thaksin ran on this and implemented it very quickly 289 00:15:46,620 --> 00:15:47,790 after being elected. 290 00:15:47,790 --> 00:15:51,880 So arguably it couldn't have been anticipated. 291 00:15:51,880 --> 00:15:54,540 Now, you know, the larger question is, 292 00:15:54,540 --> 00:15:56,970 why was if he was trying to maximize 293 00:15:56,970 --> 00:16:01,650 votes you'd want it per capita? 294 00:16:01,650 --> 00:16:03,640 Why give a lot of money to a village where 295 00:16:03,640 --> 00:16:04,890 there aren't very many people? 296 00:16:04,890 --> 00:16:08,880 But, again, in the Thai system, villages 297 00:16:08,880 --> 00:16:13,290 are thought of as part of the political system it's village, 298 00:16:13,290 --> 00:16:15,160 tambon, amphoe, and so on. 299 00:16:15,160 --> 00:16:19,200 So you give it to a committee within the village 300 00:16:19,200 --> 00:16:21,500 to run a fund. 301 00:16:21,500 --> 00:16:26,130 And somehow it seemed, quote, "fair." 302 00:16:26,130 --> 00:16:27,450 It's great for us, right? 303 00:16:27,450 --> 00:16:32,380 I mean, there's wonderful mistakes in some respect. 304 00:16:32,380 --> 00:16:34,610 It's kind of a unique opportunity. 305 00:16:41,660 --> 00:16:42,530 So we did that. 306 00:16:42,530 --> 00:16:43,400 Oops, I see. 307 00:16:43,400 --> 00:16:43,900 OK. 308 00:16:47,090 --> 00:16:49,700 So as I said, it's part of the GDP. 309 00:16:56,946 --> 00:17:00,773 AUDIENCE: What is the 77,000? 310 00:17:00,773 --> 00:17:02,690 ROBERT TOWNSEND: Yeah, I think that's supposed 311 00:17:02,690 --> 00:17:06,200 to be villages, not households. 312 00:17:06,200 --> 00:17:08,770 That's too low. 313 00:17:08,770 --> 00:17:09,910 This I've said. 314 00:17:09,910 --> 00:17:12,460 So if you don't mine, I'll just skip it. 315 00:17:12,460 --> 00:17:18,099 Loans are about $500, short-term, 316 00:17:18,099 --> 00:17:22,359 as in one year, typically two guarantors. 317 00:17:22,359 --> 00:17:26,530 The interest rate was higher initially at about 7%, 318 00:17:26,530 --> 00:17:30,130 although the village committees vote on those things. 319 00:17:30,130 --> 00:17:33,550 And they allowed loans for both consumption and investment, 320 00:17:33,550 --> 00:17:36,190 unlike the stereotypical program where 321 00:17:36,190 --> 00:17:39,820 somehow consumption smoothing is supposed to be a bad thing. 322 00:17:44,730 --> 00:17:46,380 So it's my data. 323 00:17:46,380 --> 00:17:48,330 But in this context, probably bears 324 00:17:48,330 --> 00:17:51,990 repeating because it's unusual that we didn't 325 00:17:51,990 --> 00:17:53,530 see this program coming either. 326 00:17:53,530 --> 00:17:56,910 We're out there gathering data year after year. 327 00:17:56,910 --> 00:18:02,772 And we have five years of the pre-intervention data. 328 00:18:02,772 --> 00:18:04,980 We're going to exploit that like crazy, because we're 329 00:18:04,980 --> 00:18:07,380 going to write down a model, which I'll tell you about, 330 00:18:07,380 --> 00:18:10,980 and estimate it, all based on the pre-intervention data, 331 00:18:10,980 --> 00:18:13,750 and then simulate the model in compared 332 00:18:13,750 --> 00:18:15,600 to what actually happened. 333 00:18:15,600 --> 00:18:19,260 We actually have six years of post-program data too, 334 00:18:19,260 --> 00:18:25,680 which is, again, rare, and here, quite interesting. 335 00:18:25,680 --> 00:18:28,530 Again, if you're thinking about, which you may not, 336 00:18:28,530 --> 00:18:32,460 but you might be thinking about budget constraints to do RCTs 337 00:18:32,460 --> 00:18:33,090 and so on. 338 00:18:33,090 --> 00:18:35,310 It's quite expensive. 339 00:18:35,310 --> 00:18:39,910 And gathering the data is a big part of the expense. 340 00:18:39,910 --> 00:18:43,050 So running a survey for a long time before the intervention 341 00:18:43,050 --> 00:18:45,270 and running it for a long time afterwards 342 00:18:45,270 --> 00:18:48,240 is almost prohibitively expensive. 343 00:18:48,240 --> 00:18:54,090 Or at least you wouldn't be able to evaluate very many things. 344 00:18:54,090 --> 00:18:56,670 But we didn't design the data gathering 345 00:18:56,670 --> 00:18:57,720 with this thing in mind. 346 00:18:57,720 --> 00:19:00,810 It just happened. 347 00:19:00,810 --> 00:19:05,150 I guess in that sense, again, we just got lucky. 348 00:19:05,150 --> 00:19:07,070 We're going to look at four kinds of outcomes, 349 00:19:07,070 --> 00:19:11,900 short-term credit, overall, I might add, 350 00:19:11,900 --> 00:19:15,110 not just from the village funds, including, therefore, 351 00:19:15,110 --> 00:19:20,520 borrowing from the BAC, the Ag Bank, commercial banks, 352 00:19:20,520 --> 00:19:22,490 the reasons for borrowing-- 353 00:19:22,490 --> 00:19:24,650 although I don't have those slides for you 354 00:19:24,650 --> 00:19:29,430 here because money's potentially fungible. 355 00:19:29,430 --> 00:19:32,970 We do look at interest rates default and so on. 356 00:19:32,970 --> 00:19:36,900 We have consumption broken down in these annual data, 357 00:19:36,900 --> 00:19:43,180 by the way, into these categories. 358 00:19:43,180 --> 00:19:44,640 Why these categories? 359 00:19:44,640 --> 00:19:47,860 Well, we had the socioeconomic survey. 360 00:19:47,860 --> 00:19:52,060 And we did analysis when we designed my survey 361 00:19:52,060 --> 00:19:54,460 to see what subset of consumption items 362 00:19:54,460 --> 00:20:00,700 would best predict the total in the socioeconomic survey. 363 00:20:00,700 --> 00:20:05,470 So that's kind of in the spirit of Blundell and Pistaferri 364 00:20:05,470 --> 00:20:10,990 and so on in terms of taking advantage, projecting to get 365 00:20:10,990 --> 00:20:13,305 something bigger than food. 366 00:20:13,305 --> 00:20:15,680 But actually there's some interesting stuff with alcohol. 367 00:20:15,680 --> 00:20:17,920 I'll show you, if I remember, when 368 00:20:17,920 --> 00:20:19,420 we get through these things. 369 00:20:19,420 --> 00:20:22,960 Income and production decisions is another class of categories. 370 00:20:22,960 --> 00:20:25,090 We have assets and income growth, even 371 00:20:25,090 --> 00:20:30,940 in the annual data, divided into business, wage and salary, 372 00:20:30,940 --> 00:20:37,150 and agriculture and so on. 373 00:20:37,150 --> 00:20:40,970 And we see revenues and expenses. 374 00:20:40,970 --> 00:20:44,410 And again, we have some ability to look 375 00:20:44,410 --> 00:20:48,040 at female headed as opposed to male headed households. 376 00:20:52,240 --> 00:20:56,330 OK, what's an outcome? 377 00:20:56,330 --> 00:21:01,130 y for household n at date t. 378 00:21:01,130 --> 00:21:06,260 This is like imagine running, oh, my god, a simple-minded OLS 379 00:21:06,260 --> 00:21:12,140 regression, the way people sometimes used to do. 380 00:21:14,690 --> 00:21:19,510 So Village Fund credit would include the Million Baht Fund. 381 00:21:19,510 --> 00:21:23,960 You want to see the impact causally on an outcome. 382 00:21:23,960 --> 00:21:25,710 Of course, we're going to instrument this. 383 00:21:25,710 --> 00:21:33,740 But the other things are these household characteristics, 384 00:21:33,740 --> 00:21:38,650 of various kinds, mainly demographics, age, education, 385 00:21:38,650 --> 00:21:41,910 number of kids, blah, blah, blah, gender. 386 00:21:41,910 --> 00:21:45,600 And then we have a time fixed effects and a household fixed 387 00:21:45,600 --> 00:21:48,660 effects, which are estimated in the data. 388 00:21:54,690 --> 00:22:01,370 And here's the endogenous variable being instrumented. 389 00:22:01,370 --> 00:22:03,490 And this is pretty simple. 390 00:22:03,490 --> 00:22:08,550 And we also have those x covariates here and time 391 00:22:08,550 --> 00:22:14,550 and household fixed effects for household short-term credit, 392 00:22:14,550 --> 00:22:16,450 household and at date t. 393 00:22:16,450 --> 00:22:18,360 But we've also got some dummies that 394 00:22:18,360 --> 00:22:23,447 kick in the first year and second year 395 00:22:23,447 --> 00:22:24,280 of the intervention. 396 00:22:24,280 --> 00:22:26,370 Those are chis not x's. 397 00:22:26,370 --> 00:22:28,410 Those are 0, 1 variables. 398 00:22:28,410 --> 00:22:32,940 And they're interacted with the inverse number of households. 399 00:22:32,940 --> 00:22:36,270 So inverse, that's just to get the sign right. 400 00:22:36,270 --> 00:22:45,960 Although it's a parabola, so it's not exactly not mattering. 401 00:22:45,960 --> 00:22:48,660 But if you wanted this thing to be positive, 402 00:22:48,660 --> 00:22:52,500 the larger the number of households, 403 00:22:52,500 --> 00:22:55,230 the smaller this number, and kind of the less 404 00:22:55,230 --> 00:22:57,900 you would expect the impact to be. 405 00:23:00,700 --> 00:23:05,880 And then we have the usual all the errors are uncorrelated 406 00:23:05,880 --> 00:23:07,680 with each other, et cetera, et cetera. 407 00:23:09,870 --> 00:23:10,370 Yep. 408 00:23:10,370 --> 00:23:13,040 AUDIENCE: How many villages was this? 409 00:23:13,040 --> 00:23:16,610 ROBERT TOWNSEND: This is 64, because it's my annual data. 410 00:23:16,610 --> 00:23:22,232 AUDIENCE: And can you also look at outcomes in the CDD data? 411 00:23:22,232 --> 00:23:27,160 You know, 450,000? 412 00:23:27,160 --> 00:23:34,173 ROBERT TOWNSEND: We have not amazingly done that. 413 00:23:34,173 --> 00:23:36,662 AUDIENCE: CDD is at the village level. 414 00:23:36,662 --> 00:23:38,922 This a regression at the household level. 415 00:23:38,922 --> 00:23:40,280 AUDIENCE: That's true. 416 00:23:40,280 --> 00:23:42,200 ROBERT TOWNSEND: But you could potentially 417 00:23:42,200 --> 00:23:46,190 do things across villages to see where the villages got 418 00:23:46,190 --> 00:23:47,430 more or less credit. 419 00:23:47,430 --> 00:23:50,450 So the spirit of it is possible. 420 00:23:50,450 --> 00:23:54,920 Someone actually looked at the socioeconomic survey 421 00:23:54,920 --> 00:23:58,040 and had enough connections to the Nash NSO 422 00:23:58,040 --> 00:24:01,940 to get it by village. 423 00:24:01,940 --> 00:24:05,780 But their timing was off, because their first year is 424 00:24:05,780 --> 00:24:07,370 the year of the intervention. 425 00:24:07,370 --> 00:24:10,670 And we have the pre-intervention data. 426 00:24:10,670 --> 00:24:12,080 But anyway, it's possible. 427 00:24:12,080 --> 00:24:13,310 No one has looked at that. 428 00:24:13,310 --> 00:24:16,327 AUDIENCE: What was year of the intervention? 429 00:24:16,327 --> 00:24:17,660 ROBERT TOWNSEND: The first year? 430 00:24:17,660 --> 00:24:18,680 AUDIENCE: Yeah. 431 00:24:18,680 --> 00:24:23,013 ROBERT TOWNSEND: 2002. 432 00:24:23,013 --> 00:24:25,090 AUDIENCE: Why don't they have the ISS data? 433 00:24:25,090 --> 00:24:26,382 They just don't have the years? 434 00:24:26,382 --> 00:24:28,590 I thought the ISS go back 30 years. 435 00:24:28,590 --> 00:24:30,310 ROBERT TOWNSEND: They don't have a panel. 436 00:24:30,310 --> 00:24:32,600 AUDIENCE: Oh, OK. 437 00:24:32,600 --> 00:24:34,340 But they don't have panel afterward? 438 00:24:34,340 --> 00:24:35,920 ROBERT TOWNSEND: They do. 439 00:24:35,920 --> 00:24:36,820 They did. 440 00:24:36,820 --> 00:24:37,320 Yeah. 441 00:24:37,320 --> 00:24:38,612 AUDIENCE: But we don't have it. 442 00:24:38,612 --> 00:24:42,050 ROBERT TOWNSEND: We discovered it after. 443 00:24:42,050 --> 00:24:42,580 We have it. 444 00:24:47,720 --> 00:24:52,100 So we're going to look at new short-term credit, consumption, 445 00:24:52,100 --> 00:24:53,270 asset growth, and income. 446 00:24:56,317 --> 00:24:57,650 Tried to do it pretty carefully. 447 00:24:57,650 --> 00:25:00,470 We actually did it about four more ways than this. 448 00:25:00,470 --> 00:25:03,130 But reduced the number of that we show. 449 00:25:03,130 --> 00:25:06,410 OLS with all the problems that you know about. 450 00:25:06,410 --> 00:25:12,790 Here's an IV regression with all villages as opposed to fifth-- 451 00:25:12,790 --> 00:25:14,610 this is kind of outliers in village. 452 00:25:14,610 --> 00:25:17,030 If you think about that parabola, 453 00:25:17,030 --> 00:25:20,300 you know, it's kind of getting really steep when household 454 00:25:20,300 --> 00:25:21,560 size is really small-- 455 00:25:21,560 --> 00:25:24,500 number of households is small and not moving much. 456 00:25:24,500 --> 00:25:30,890 So we clipped off, took the big chunk, 457 00:25:30,890 --> 00:25:33,770 but we clipped off the tails of small and large villages 458 00:25:33,770 --> 00:25:38,030 and dropped some outliers from the household data too. 459 00:25:38,030 --> 00:25:46,010 But every single way we do it, short-term village fund credit 460 00:25:46,010 --> 00:25:50,010 goes new, new, short-term village credit goes up. 461 00:25:50,010 --> 00:25:57,440 These are roughly the order of magnitude of $25,000 divided 462 00:25:57,440 --> 00:26:00,310 by the number of households. 463 00:26:00,310 --> 00:26:01,990 So, you know, credit goes up. 464 00:26:01,990 --> 00:26:03,060 You might be worried. 465 00:26:03,060 --> 00:26:06,270 And like Esther and Abhijit have papers where there's 466 00:26:06,270 --> 00:26:07,790 this substitution going on. 467 00:26:07,790 --> 00:26:09,540 You get cheap credit from one source. 468 00:26:09,540 --> 00:26:11,130 You borrow less from other sources. 469 00:26:11,130 --> 00:26:14,170 That doesn't happen. 470 00:26:14,170 --> 00:26:20,110 In fact, arguably there seems to be almost an amplifier effect, 471 00:26:20,110 --> 00:26:22,570 but more weakly. 472 00:26:22,570 --> 00:26:25,480 Short-term credit goes up, for sure, one to one. 473 00:26:25,480 --> 00:26:26,830 And you can see it in the data. 474 00:26:26,830 --> 00:26:31,750 I mean, this is this like a huge spike in the financial system. 475 00:26:31,750 --> 00:26:34,420 And it didn't go down. 476 00:26:34,420 --> 00:26:36,610 That part doesn't go down subsequently. 477 00:26:36,610 --> 00:26:37,133 Yep. 478 00:26:37,133 --> 00:26:39,050 AUDIENCE: If we thought that going up variable 479 00:26:39,050 --> 00:26:42,380 was-- took that seriously, could that be something like people 480 00:26:42,380 --> 00:26:44,450 getting extra funds and re-lending 481 00:26:44,450 --> 00:26:47,438 and so you have to sort of multiplier? 482 00:26:47,438 --> 00:26:48,730 ROBERT TOWNSEND: It's true one. 483 00:26:48,730 --> 00:26:51,830 Would have to-- that's a good point. 484 00:26:51,830 --> 00:26:54,570 Once could be just-- 485 00:26:54,570 --> 00:26:57,164 not to say that it's a bad thing. 486 00:26:57,164 --> 00:26:59,600 AUDIENCE: Sure. 487 00:26:59,600 --> 00:27:03,140 ROBERT TOWNSEND: But, you know, you 488 00:27:03,140 --> 00:27:06,350 have these sort of combinations of characteristics of loans. 489 00:27:06,350 --> 00:27:08,480 And if the village fund committee 490 00:27:08,480 --> 00:27:10,040 has this cap, which they're supposed 491 00:27:10,040 --> 00:27:12,680 to have on the loan size, then you 492 00:27:12,680 --> 00:27:15,350 get one leg up and think about doing something. 493 00:27:15,350 --> 00:27:18,500 Maybe you want to go the money lender or other sources 494 00:27:18,500 --> 00:27:23,065 and lever up even more when you have chunkiness. 495 00:27:23,065 --> 00:27:24,690 AUDIENCE: So this is because it's going 496 00:27:24,690 --> 00:27:27,050 more one to one, the credit? 497 00:27:27,050 --> 00:27:28,990 ROBERT TOWNSEND: Yeah. 498 00:27:28,990 --> 00:27:33,580 But the more conservative way to read it is just one for one 499 00:27:33,580 --> 00:27:34,090 anyway. 500 00:27:34,090 --> 00:27:37,450 So it's not like a substitute for other sources of money. 501 00:27:37,450 --> 00:27:39,130 AUDIENCE: So when I initially looked 502 00:27:39,130 --> 00:27:42,680 at it I thought this support for Esther and Abhijit's idea 503 00:27:42,680 --> 00:27:46,510 that if you can get over a hump, then 504 00:27:46,510 --> 00:27:48,260 you'll really want to expand or something. 505 00:27:48,260 --> 00:27:50,177 But I think what you were saying is could they 506 00:27:50,177 --> 00:27:52,190 be re-lending to each other? 507 00:27:52,190 --> 00:27:53,110 So I could I be-- 508 00:27:53,110 --> 00:27:53,980 ROBERT TOWNSEND: Yeah, he did say that. 509 00:27:53,980 --> 00:27:55,598 And, yeah, that could be going on. 510 00:27:55,598 --> 00:27:56,140 AUDIENCE: OK. 511 00:27:56,140 --> 00:27:57,843 That's a convenient explanation. 512 00:27:57,843 --> 00:27:59,260 ROBERT TOWNSEND: And I don't think 513 00:27:59,260 --> 00:28:00,520 we tried to take that out. 514 00:28:00,520 --> 00:28:04,780 We're just looking at the household 515 00:28:04,780 --> 00:28:08,860 and at date t in some village j and adding up 516 00:28:08,860 --> 00:28:11,690 all the short-term loans that they have. 517 00:28:11,690 --> 00:28:17,130 So if you only counted the loan and not the lending, 518 00:28:17,130 --> 00:28:19,800 if it wasn't net, and I think it wasn't. 519 00:28:19,800 --> 00:28:21,333 I think it's just gross. 520 00:28:21,333 --> 00:28:23,250 AUDIENCE: Not all loans are one year. 521 00:28:23,250 --> 00:28:24,750 They're are loans that are one week. 522 00:28:24,750 --> 00:28:25,792 AUDIENCE: No, sure, yeah. 523 00:28:28,200 --> 00:28:30,440 AUDIENCE: They almost have different lending, really, 524 00:28:30,440 --> 00:28:33,297 really short term. 525 00:28:33,297 --> 00:28:35,630 ROBERT TOWNSEND: So that's why we did short term instead 526 00:28:35,630 --> 00:28:36,800 of long term. 527 00:28:36,800 --> 00:28:40,250 Consumption levels, other than the OLS-- 528 00:28:40,250 --> 00:28:43,150 I'll come back to that-- go up. 529 00:28:43,150 --> 00:28:48,455 Significantly, asset growth in one specification goes down. 530 00:28:51,840 --> 00:28:58,940 Now, I can't help it but say that we're going 531 00:28:58,940 --> 00:29:00,023 to have to choose a model. 532 00:29:00,023 --> 00:29:01,107 What I'm going to do now-- 533 00:29:01,107 --> 00:29:02,710 I'm giving you these stylized facts-- 534 00:29:02,710 --> 00:29:09,730 but what model comes to your head on why things like savings 535 00:29:09,730 --> 00:29:10,540 might go down? 536 00:29:16,400 --> 00:29:18,260 Like a buffer stock model. 537 00:29:18,260 --> 00:29:20,840 If you're on your own, you're saving 538 00:29:20,840 --> 00:29:23,450 for these future disasters. 539 00:29:23,450 --> 00:29:25,970 Now, in the future, you could borrow 540 00:29:25,970 --> 00:29:28,120 if you believe the village fund is persistent. 541 00:29:28,120 --> 00:29:31,130 So you don't need to have so many buffer stocks. 542 00:29:31,130 --> 00:29:33,675 So net financial savings goes down actually. 543 00:29:37,080 --> 00:29:40,370 And we have trouble finding stuff with the capital stock. 544 00:29:40,370 --> 00:29:41,460 I'll mention that. 545 00:29:41,460 --> 00:29:44,430 By the way, these consumption magnitudes 546 00:29:44,430 --> 00:29:47,310 are very similar to the credit magnitudes. 547 00:29:47,310 --> 00:29:51,210 So a bit naively, you might say they 548 00:29:51,210 --> 00:29:54,490 used it all for consumption. 549 00:29:54,490 --> 00:29:57,540 Now, again, what model do we have 550 00:29:57,540 --> 00:29:59,460 that would predict that consumption 551 00:29:59,460 --> 00:30:02,220 ought to go up one for one with an increase in credit? 552 00:30:07,180 --> 00:30:09,660 AUDIENCE: Life time permanent income model. 553 00:30:09,660 --> 00:30:10,860 ROBERT TOWNSEND: What would the permanent income 554 00:30:10,860 --> 00:30:11,520 model give you? 555 00:30:16,557 --> 00:30:18,140 AUDIENCE: It would go up, wouldn't it? 556 00:30:18,140 --> 00:30:19,710 ROBERT TOWNSEND: How much? 557 00:30:19,710 --> 00:30:23,379 AUDIENCE: It has hazard shock is temporary or persistent. 558 00:30:23,379 --> 00:30:25,421 AUDIENCE: You would think this persistent, right? 559 00:30:25,421 --> 00:30:27,320 So then it should jump to the new-- 560 00:30:27,320 --> 00:30:28,572 but I don't know how much. 561 00:30:28,572 --> 00:30:30,280 ROBERT TOWNSEND: You'd put it in the bank 562 00:30:30,280 --> 00:30:33,658 basically and draw the interest off of it. 563 00:30:33,658 --> 00:30:35,450 AUDIENCE: This is a temporary shock, right? 564 00:30:35,450 --> 00:30:38,132 It's just once. 565 00:30:38,132 --> 00:30:39,840 AUDIENCE: Then you can get the funds back 566 00:30:39,840 --> 00:30:42,030 and you lend it again. 567 00:30:42,030 --> 00:30:44,725 ROBERT TOWNSEND: Either way. 568 00:30:44,725 --> 00:30:45,850 AUDIENCE: We're missing it. 569 00:30:45,850 --> 00:30:46,493 What's the-- 570 00:30:46,493 --> 00:30:48,410 ROBERT TOWNSEND: No, no, no, I'm just saying-- 571 00:30:48,410 --> 00:30:49,950 no, no, you got close. 572 00:30:49,950 --> 00:30:52,560 I mean, the point is the permanent income model 573 00:30:52,560 --> 00:30:56,110 isn't going to get you this. 574 00:30:56,110 --> 00:30:58,320 And you've got to have these credit constraints. 575 00:30:58,320 --> 00:31:01,600 You've got to be something that's generating hand 576 00:31:01,600 --> 00:31:04,060 to mouth like behavior. 577 00:31:04,060 --> 00:31:06,780 AUDIENCE: So this is a shock to the credit constraint not 578 00:31:06,780 --> 00:31:08,410 the shock to income. 579 00:31:08,410 --> 00:31:11,720 So you should be facing this credit constraint-- 580 00:31:11,720 --> 00:31:14,230 you should be constrained. 581 00:31:14,230 --> 00:31:15,770 ROBERT TOWNSEND: Yeah, that's right. 582 00:31:15,770 --> 00:31:20,450 And therefore, you're going to see that this overall average 583 00:31:20,450 --> 00:31:24,180 is masking a very big differential 584 00:31:24,180 --> 00:31:27,060 impact in the population depending on whether someone's 585 00:31:27,060 --> 00:31:28,485 at that constraint or not. 586 00:31:28,485 --> 00:31:29,860 AUDIENCE: But even if you are not 587 00:31:29,860 --> 00:31:31,902 constrained, if you anticipate that in the future 588 00:31:31,902 --> 00:31:35,630 I'm constrained, endogenously you-- 589 00:31:35,630 --> 00:31:38,040 ROBERT TOWNSEND: It depends on the interpretation. 590 00:31:38,040 --> 00:31:41,310 Suppose you thought, you know, the government 591 00:31:41,310 --> 00:31:45,030 gave the money lump sum to the households 592 00:31:45,030 --> 00:31:47,310 indirectly through the village fund. 593 00:31:47,310 --> 00:31:49,220 And they just put that in the bank. 594 00:31:49,220 --> 00:31:51,490 Then their net worth has gone up. 595 00:31:51,490 --> 00:31:53,610 They don't have an obligation. 596 00:31:53,610 --> 00:31:56,790 Within the village, they are borrowing and lending 597 00:31:56,790 --> 00:31:58,830 against this fund. 598 00:31:58,830 --> 00:32:02,990 But the village, as a whole, just got a whole lot richer. 599 00:32:02,990 --> 00:32:04,820 And they would potentially, if you 600 00:32:04,820 --> 00:32:06,600 believed in the permanent income, 601 00:32:06,600 --> 00:32:09,290 raise their asset level, right? 602 00:32:09,290 --> 00:32:13,050 And then get the flow off of it. 603 00:32:13,050 --> 00:32:13,550 Yes? 604 00:32:16,850 --> 00:32:19,040 AUDIENCE: How is it financed? 605 00:32:19,040 --> 00:32:20,150 Is it something like-- 606 00:32:20,150 --> 00:32:21,358 ROBERT TOWNSEND: Taxes. 607 00:32:21,358 --> 00:32:23,150 AUDIENCE: But like the taxes mainly hitting 608 00:32:23,150 --> 00:32:23,840 cities instead of villages? 609 00:32:23,840 --> 00:32:25,190 ROBERT TOWNSEND: Yeah. 610 00:32:25,190 --> 00:32:27,362 AUDIENCE: I see. 611 00:32:27,362 --> 00:32:28,295 ROBERT TOWNSEND: Matt. 612 00:32:28,295 --> 00:32:29,670 AUDIENCE: So I was a bit confused 613 00:32:29,670 --> 00:32:33,980 between the OLS and the IV, because what 614 00:32:33,980 --> 00:32:37,400 is the expansion variable in OLS I 615 00:32:37,400 --> 00:32:41,740 was thinking it must be credit per person. 616 00:32:41,740 --> 00:32:43,580 But then I don't understand really 617 00:32:43,580 --> 00:32:47,180 what the instrument is, right? 618 00:32:47,180 --> 00:32:49,730 Because if its credit per person it's just a million 619 00:32:49,730 --> 00:32:51,370 divided by number of households. 620 00:32:51,370 --> 00:32:54,155 And that instrument itself is per household. 621 00:32:54,155 --> 00:32:55,530 ROBERT TOWNSEND: So there's a lot 622 00:32:55,530 --> 00:32:59,670 of variation in short-term village credit, 623 00:32:59,670 --> 00:33:02,910 both among households within a village and across villages. 624 00:33:02,910 --> 00:33:04,830 So we're identifying this through 625 00:33:04,830 --> 00:33:07,020 the cross-sectional variation. 626 00:33:07,020 --> 00:33:11,370 We're saying what part of the increase in observed increase 627 00:33:11,370 --> 00:33:14,880 in short-term village credit can be explained simply 628 00:33:14,880 --> 00:33:19,620 by inverse population size? 629 00:33:19,620 --> 00:33:22,920 And the OLS doesn't do that? 630 00:33:22,920 --> 00:33:26,490 It's just a flat out regression credit on the right-hand side 631 00:33:26,490 --> 00:33:28,460 and impacts on the left. 632 00:33:28,460 --> 00:33:31,070 All credit. 633 00:33:31,070 --> 00:33:34,070 AUDIENCE: OK. 634 00:33:34,070 --> 00:33:38,560 ROBERT TOWNSEND: And, you know, net income growth is positive 635 00:33:38,560 --> 00:33:42,035 in 3 to the 4, including the OLS, 636 00:33:42,035 --> 00:33:45,600 or 2 of the 3 if you don't. 637 00:33:45,600 --> 00:33:49,130 And so there is some impact on overall growth, 638 00:33:49,130 --> 00:33:50,045 not levels, growth. 639 00:33:56,870 --> 00:33:58,635 And this is by source of income. 640 00:34:02,660 --> 00:34:06,490 You know, we see a hint of profits, 641 00:34:06,490 --> 00:34:10,929 apropos the first two three slides of the lecture today. 642 00:34:10,929 --> 00:34:12,639 Profits may be going up. 643 00:34:12,639 --> 00:34:16,300 And wages and salary payments are going up 644 00:34:16,300 --> 00:34:18,530 pretty consistently. 645 00:34:18,530 --> 00:34:22,010 So again, you know, what that's all about? 646 00:34:22,010 --> 00:34:24,060 But I've already given some of it away. 647 00:34:24,060 --> 00:34:27,790 The wage rate that's going up for villages 648 00:34:27,790 --> 00:34:32,920 that got a higher per capita treatment. 649 00:34:32,920 --> 00:34:35,580 And potentially employment goes up. 650 00:34:35,580 --> 00:34:36,219 Yep? 651 00:34:36,219 --> 00:34:38,742 AUDIENCE: So this slide and the previous slide as well. 652 00:34:38,742 --> 00:34:40,659 I mean I understand we're talking about bahts, 653 00:34:40,659 --> 00:34:43,760 so the quantities are big. 654 00:34:43,760 --> 00:34:45,742 But like some of yours are huge. 655 00:34:48,018 --> 00:34:49,810 ROBERT TOWNSEND: But you can see the stars. 656 00:34:49,810 --> 00:34:51,810 AUDIENCE: No, no, no, I understand that. 657 00:34:51,810 --> 00:34:54,610 But even in the previous slides, on stuff 658 00:34:54,610 --> 00:34:58,963 that's like the growth rates, the standard errors are big. 659 00:34:58,963 --> 00:35:00,130 ROBERT TOWNSEND: Yeah, Yeah. 660 00:35:00,130 --> 00:35:03,370 AUDIENCE: You don't get clear zeros and stuff like that. 661 00:35:03,370 --> 00:35:04,008 So-- 662 00:35:04,008 --> 00:35:06,300 ROBERT TOWNSEND: Wait, I'm not sure what you're saying. 663 00:35:06,300 --> 00:35:08,550 Yes, the standard errors can be big. 664 00:35:08,550 --> 00:35:11,820 That would make it tough to identify significant effects. 665 00:35:11,820 --> 00:35:13,847 But we on occasion do find significant effects. 666 00:35:13,847 --> 00:35:15,180 AUDIENCE: No, I understand that. 667 00:35:15,180 --> 00:35:18,546 I'm just wondering-- my question was more of as technical one. 668 00:35:18,546 --> 00:35:26,130 Like how-- is there a way to sort of make them smaller. 669 00:35:26,130 --> 00:35:29,670 I mean I guess it's an issue of ID and stuff like that. 670 00:35:29,670 --> 00:35:31,170 But-- 671 00:35:31,170 --> 00:35:33,330 ROBERT TOWNSEND: Well, you know, this 672 00:35:33,330 --> 00:35:37,680 is with clustered standard errors and the whole thing. 673 00:35:37,680 --> 00:35:43,500 And we try to eliminate the outliers to see if that's kind 674 00:35:43,500 --> 00:35:45,570 of causing a lot of the noise. 675 00:35:45,570 --> 00:35:48,330 Sometimes it does. 676 00:35:48,330 --> 00:35:52,070 It doesn't do that consistently. 677 00:35:52,070 --> 00:35:53,280 OK. 678 00:35:53,280 --> 00:35:58,610 We don't find much with ag or other sources 679 00:35:58,610 --> 00:36:00,490 of income or livestock. 680 00:36:00,490 --> 00:36:05,120 Now, as I said, we also have a lot 681 00:36:05,120 --> 00:36:10,340 of post data, post-intervention. 682 00:36:10,340 --> 00:36:15,470 Even when we wrote this thing, we're up to six years of data. 683 00:36:15,470 --> 00:36:16,910 But this is OLS. 684 00:36:16,910 --> 00:36:21,800 So you have to take it with a bit more of the grain of salt. 685 00:36:21,800 --> 00:36:25,220 For example, consumption isn't showing up, 686 00:36:25,220 --> 00:36:30,700 even initially because we're not instrumenting anymore. 687 00:36:30,700 --> 00:36:33,080 The probability of default's kind of moving around. 688 00:36:33,080 --> 00:36:34,100 And it doesn't go away. 689 00:36:34,100 --> 00:36:35,280 And that's a real issue. 690 00:36:35,280 --> 00:36:38,330 And let me show you the default rates. 691 00:36:38,330 --> 00:36:43,920 And, you know, potentially, by the way, there should be a lag. 692 00:36:43,920 --> 00:36:46,430 Depending on exactly where the 12 month falls, 693 00:36:46,430 --> 00:36:49,130 you borrow one year and you can't repay the next. 694 00:36:56,070 --> 00:36:57,270 Net income goes up. 695 00:36:57,270 --> 00:37:01,300 But it's not significant after that. 696 00:37:01,300 --> 00:37:08,580 And short-term village credit remains high. 697 00:37:08,580 --> 00:37:12,330 Well, what do I want to say? 698 00:37:12,330 --> 00:37:15,510 Back to models, you'll get one, don't worry. 699 00:37:18,670 --> 00:37:21,280 You know, the world isn't static. 700 00:37:21,280 --> 00:37:23,320 And we just shook things a little bit, 701 00:37:23,320 --> 00:37:27,160 because these villages arguably are wealthier 702 00:37:27,160 --> 00:37:30,430 than they were and they've also somehow used this 703 00:37:30,430 --> 00:37:37,125 to capitalize basically a savings and loan association 704 00:37:37,125 --> 00:37:38,250 within each of the village. 705 00:37:38,250 --> 00:37:41,400 So if you were in a steady state, 706 00:37:41,400 --> 00:37:44,430 and someone smacked you in the head with something good, 707 00:37:44,430 --> 00:37:48,150 you might expect the adjustment isn't necessarily insta-- 708 00:37:48,150 --> 00:37:52,710 all the dynamic models we have with frictions, Paco's, 709 00:37:52,710 --> 00:37:57,360 for example, suggests there are slow transitions even 710 00:37:57,360 --> 00:37:59,760 to some ultimate new steady state. 711 00:37:59,760 --> 00:38:04,650 So this is rare, this look at these-- now, granted, 712 00:38:04,650 --> 00:38:10,800 the model's going to be partial equilibrium, even though we 713 00:38:10,800 --> 00:38:12,315 have some data on local wages. 714 00:38:15,500 --> 00:38:17,100 And there's a guy named-- 715 00:38:17,100 --> 00:38:22,790 is it Furfine who has a study? 716 00:38:22,790 --> 00:38:25,270 Pandi and Burgess had this study in India 717 00:38:25,270 --> 00:38:30,280 where the government changed the branching rules 718 00:38:30,280 --> 00:38:33,620 and showed impacts on poverty and all of that stuff. 719 00:38:33,620 --> 00:38:37,350 But he went back after they published their paper 720 00:38:37,350 --> 00:38:40,090 and looked at the long-term impact. 721 00:38:40,090 --> 00:38:44,170 And he thinks something like this was going on in India too. 722 00:38:44,170 --> 00:38:48,490 So again, apropos, you know, RCTs, well, no one 723 00:38:48,490 --> 00:38:50,200 claimed the contrary, but if you don't 724 00:38:50,200 --> 00:38:52,870 know what happens afterwards, you're 725 00:38:52,870 --> 00:38:55,120 kind of left with the impression and hope 726 00:38:55,120 --> 00:39:00,640 that the program effects, if they were positive, persisted. 727 00:39:00,640 --> 00:39:03,980 Here, they're definitely mitigating over time. 728 00:39:03,980 --> 00:39:05,210 Yes. 729 00:39:05,210 --> 00:39:10,710 AUDIENCE: Do you have the long-term impact on wages? 730 00:39:10,710 --> 00:39:13,674 ROBERT TOWNSEND: I don't think we looked at that. 731 00:39:13,674 --> 00:39:16,120 That would be good to do. 732 00:39:16,120 --> 00:39:20,920 We struggle with the wage, because the annual data 733 00:39:20,920 --> 00:39:22,060 measure it very well. 734 00:39:22,060 --> 00:39:26,770 So this is the one thing that we use the monthly data for. 735 00:39:26,770 --> 00:39:29,410 And we only have 16 villages. 736 00:39:29,410 --> 00:39:31,810 So it's quite problematic. 737 00:39:31,810 --> 00:39:35,200 Most things we try don't show up because of the sample 738 00:39:35,200 --> 00:39:38,720 size is too low and the standard errors just kill us. 739 00:39:38,720 --> 00:39:42,170 But I don't think we actually looked at long-term wages. 740 00:39:42,170 --> 00:39:43,490 So that would be good to do. 741 00:39:43,490 --> 00:39:44,196 Yes. 742 00:39:44,196 --> 00:39:49,198 AUDIENCE: Do you think the wage data in CDD is any good? 743 00:39:49,198 --> 00:39:51,242 Daily wage for laborers. 744 00:39:51,242 --> 00:39:52,950 ROBERT TOWNSEND: I think when you sort of 745 00:39:52,950 --> 00:39:56,880 look at geographic averages of things, yeah, that it has-- 746 00:39:56,880 --> 00:40:00,360 there are beautiful maps showing how wages, not size, 747 00:40:00,360 --> 00:40:02,910 but wages move as you get near urban areas 748 00:40:02,910 --> 00:40:04,950 and get near Bangkok. 749 00:40:04,950 --> 00:40:08,980 And so arguably, something could potentially-- 750 00:40:08,980 --> 00:40:12,920 I wouldn't trust any one head man's response, 751 00:40:12,920 --> 00:40:20,370 but with 72,000 data points, you can do a lot. 752 00:40:20,370 --> 00:40:22,518 AUDIENCE: So that paper, his name is Fulford. 753 00:40:22,518 --> 00:40:23,560 ROBERT TOWNSEND: Fulford. 754 00:40:26,780 --> 00:40:32,230 OK, so this is a cautionary tale. 755 00:40:32,230 --> 00:40:33,250 I'm going to skip it. 756 00:40:33,250 --> 00:40:36,760 If we have time, I'll do it next time. 757 00:40:36,760 --> 00:40:38,874 That would be a better context for it. 758 00:40:41,540 --> 00:40:48,230 So I've been trying to clue you into the puzzles all along 759 00:40:48,230 --> 00:40:49,730 in terms of-- 760 00:40:49,730 --> 00:40:51,920 the one thing I didn't say was about investment. 761 00:40:55,210 --> 00:41:00,040 And I had mentioned in other lectures 762 00:41:00,040 --> 00:41:03,590 that investment doesn't happen very often. 763 00:41:03,590 --> 00:41:06,270 And when it happens, it's big. 764 00:41:06,270 --> 00:41:08,320 So it's quite lumpy. 765 00:41:08,320 --> 00:41:10,910 And we're going to want to incorporate that. 766 00:41:10,910 --> 00:41:12,990 But the flip side of that is we're 767 00:41:12,990 --> 00:41:15,500 going to have trouble finding effects of investment 768 00:41:15,500 --> 00:41:19,700 both in the IV and in the structural model, 769 00:41:19,700 --> 00:41:22,250 and the data generated from the structural model, 770 00:41:22,250 --> 00:41:25,730 if we limit that data to the same sample 771 00:41:25,730 --> 00:41:28,980 size that we have in reality. 772 00:41:28,980 --> 00:41:32,070 But when you see the structural model momentarily, 773 00:41:32,070 --> 00:41:35,390 you can imagine a 10-fold increase in the sample size 774 00:41:35,390 --> 00:41:37,820 because we can generate as much data as we want. 775 00:41:37,820 --> 00:41:40,940 And then, we definitely get these investment effects. 776 00:41:40,940 --> 00:41:45,460 So, you know, arguably one reason 777 00:41:45,460 --> 00:41:49,480 still on the table that we don't see microcredit directly 778 00:41:49,480 --> 00:41:51,130 in terms of investment activities 779 00:41:51,130 --> 00:41:53,950 is simply a sample size issue. 780 00:41:53,950 --> 00:41:55,900 I mentioned that because, you know, 781 00:41:55,900 --> 00:41:57,640 people write reviews of what do we 782 00:41:57,640 --> 00:42:00,760 know about microcredit and whether or not it 783 00:42:00,760 --> 00:42:02,710 having an impact. 784 00:42:02,710 --> 00:42:04,930 And there seems to be a bit of a consensus 785 00:42:04,930 --> 00:42:08,410 that at best, impact is quite mixed. 786 00:42:08,410 --> 00:42:11,110 And maybe, you know, non-existent in many studies. 787 00:42:11,110 --> 00:42:14,860 And people come away with this impression that, at most, 788 00:42:14,860 --> 00:42:16,150 it's all about consumption. 789 00:42:16,150 --> 00:42:21,910 But I think that's a little bit of a rush to judgment. 790 00:42:21,910 --> 00:42:23,460 I'm not saying it has to be this too. 791 00:42:23,460 --> 00:42:25,530 I don't have more data in reality than we have. 792 00:42:28,620 --> 00:42:32,260 So we're going to have this precautionary savings. 793 00:42:32,260 --> 00:42:36,130 We're going to have some limits because of those consumption 794 00:42:36,130 --> 00:42:37,100 numbers I showed you. 795 00:42:37,100 --> 00:42:41,588 We're going to allow default. It does happen in reality. 796 00:42:41,588 --> 00:42:43,630 And we're going to try to match the default rate. 797 00:42:46,390 --> 00:42:47,540 And we have income growth. 798 00:42:50,880 --> 00:42:57,860 That's actually arguably the hardest thing that we did. 799 00:42:57,860 --> 00:43:00,640 And I'll try to say something about why it was hard. 800 00:43:03,820 --> 00:43:11,990 But we actually allow persistent growth in the model. 801 00:43:11,990 --> 00:43:14,440 OK, so what is the model? 802 00:43:14,440 --> 00:43:22,080 Liquid stuff is just your current income plus your return 803 00:43:22,080 --> 00:43:24,030 on previous year savings. 804 00:43:24,030 --> 00:43:27,780 Now, this reminds me to say, you know, 805 00:43:27,780 --> 00:43:31,440 so why aren't you doing moral hazard or costly state 806 00:43:31,440 --> 00:43:34,890 verification or full information? 807 00:43:34,890 --> 00:43:37,970 Hey, we just had a lecture about that. 808 00:43:37,970 --> 00:43:40,670 Same kind of data. 809 00:43:40,670 --> 00:43:44,390 Alex and I showed that in the rural data, 810 00:43:44,390 --> 00:43:47,660 the best approximation is something simple like savings 811 00:43:47,660 --> 00:43:50,970 only or limited borrowing. 812 00:43:50,970 --> 00:43:53,030 Now, we got lucky, because Alex and I 813 00:43:53,030 --> 00:43:56,280 did that after Joe and I wrote this paper. 814 00:43:56,280 --> 00:43:59,270 But it is at least very comforting 815 00:43:59,270 --> 00:44:04,780 that the best micro underpinning that we now think 816 00:44:04,780 --> 00:44:08,560 exists out there was the micro underpinning that 817 00:44:08,560 --> 00:44:11,620 was being used in this paper. 818 00:44:11,620 --> 00:44:14,290 Likewise, I would not expect to get the same thing 819 00:44:14,290 --> 00:44:18,100 in the urban areas, because there, Alex and I 820 00:44:18,100 --> 00:44:23,030 found moral hazard was a better approximation. 821 00:44:23,030 --> 00:44:26,410 So here, it's sort of this incomplete market's macro type 822 00:44:26,410 --> 00:44:33,220 literature or development literature, for that matter. 823 00:44:33,220 --> 00:44:37,620 Rotsio swears by this stuff. 824 00:44:37,620 --> 00:44:38,920 And it's in logs. 825 00:44:38,920 --> 00:44:44,140 So income looks like it's just a multiplicative 826 00:44:44,140 --> 00:44:45,820 transitory and permanent shock. 827 00:44:45,820 --> 00:44:49,300 That makes the log of income equal to the log 828 00:44:49,300 --> 00:44:53,200 of some permanent thing and the log is some transitory thing. 829 00:44:53,200 --> 00:44:56,350 The log of the permanent thing almost 830 00:44:56,350 --> 00:45:02,250 looks like a random walk, where log Pt plus 1 831 00:45:02,250 --> 00:45:05,820 equals log Pt plus log N. So this is 832 00:45:05,820 --> 00:45:09,030 the shock to permanent income. 833 00:45:09,030 --> 00:45:12,510 This thing is the shock to transitory income, where 834 00:45:12,510 --> 00:45:13,990 income is measured in logs. 835 00:45:13,990 --> 00:45:16,170 Now, it's not quite that because G 836 00:45:16,170 --> 00:45:19,790 is this drift, this damn drift. 837 00:45:19,790 --> 00:45:21,920 It's 4%, by the way. 838 00:45:21,920 --> 00:45:25,135 So you know, there's kind of sustained growth, not modeled. 839 00:45:28,670 --> 00:45:32,700 You get this boost to your permanent income. 840 00:45:32,700 --> 00:45:34,290 It's like TFP type stuff. 841 00:45:37,060 --> 00:45:38,490 And then what's this? 842 00:45:38,490 --> 00:45:40,497 So investment, you know, why invest 843 00:45:40,497 --> 00:45:41,580 if it doesn't do anything? 844 00:45:41,580 --> 00:45:42,955 Well, on the contrary, investment 845 00:45:42,955 --> 00:45:48,670 gives you a boost of your permanent income, right? 846 00:45:48,670 --> 00:45:52,920 So this is this decision whether or not to invest, a dummy. 847 00:45:52,920 --> 00:45:55,990 And this is the amount of the investment. 848 00:45:55,990 --> 00:45:57,760 As I said, investment opportunities 849 00:45:57,760 --> 00:46:01,960 arrive stochastically from a distribution 850 00:46:01,960 --> 00:46:03,280 with a non-trivial mean. 851 00:46:03,280 --> 00:46:05,470 So on average, it's chunky, which 852 00:46:05,470 --> 00:46:07,840 means you may choose not to do it because it's 853 00:46:07,840 --> 00:46:10,380 too big to swallow. 854 00:46:10,380 --> 00:46:12,840 You don't get a choice about the size though. 855 00:46:12,840 --> 00:46:17,430 If you do it, you've got to do the size that arrived 856 00:46:17,430 --> 00:46:20,370 to you as an approximation. 857 00:46:20,370 --> 00:46:23,970 Although again, I've said this, we see this in the data. 858 00:46:23,970 --> 00:46:28,965 People don't do sort of half chicken coops. 859 00:46:28,965 --> 00:46:32,440 A chicken coop without a roof is not a coop. 860 00:46:32,440 --> 00:46:33,263 OK. 861 00:46:33,263 --> 00:46:36,401 AUDIENCE: I have a question about the drift term. 862 00:46:36,401 --> 00:46:39,930 I think we have in papers we have-- 863 00:46:39,930 --> 00:46:43,900 we basically assume some exogenous technical progress 864 00:46:43,900 --> 00:46:47,800 pushing wages up and then like papers 865 00:46:47,800 --> 00:46:49,780 where the focus on the structural model 866 00:46:49,780 --> 00:46:52,805 is to model that, because obviously 867 00:46:52,805 --> 00:46:56,720 like that's really important if Thailand over this period 868 00:46:56,720 --> 00:46:59,835 there was 4% drift where everyone or maybe-- 869 00:46:59,835 --> 00:47:01,210 ROBERT TOWNSEND: It is important. 870 00:47:01,210 --> 00:47:02,080 And we're not. 871 00:47:02,080 --> 00:47:04,420 A lot of the literature we've covered doesn't. 872 00:47:04,420 --> 00:47:06,190 Daron features that a lot. 873 00:47:06,190 --> 00:47:06,710 What is it? 874 00:47:06,710 --> 00:47:08,060 Is it innovation? 875 00:47:08,060 --> 00:47:10,911 What is this technological progress? 876 00:47:10,911 --> 00:47:12,715 AUDIENCE: So would it be-- 877 00:47:12,715 --> 00:47:14,950 Are there any inherent reasons why 878 00:47:14,950 --> 00:47:17,715 it would be difficult to translate Daron's models 879 00:47:17,715 --> 00:47:21,920 to the Thai data or the-- 880 00:47:21,920 --> 00:47:23,420 ROBERT TOWNSEND: It should be done-- 881 00:47:23,420 --> 00:47:25,572 AUDIENCE: With regional variation? 882 00:47:25,572 --> 00:47:27,030 ROBERT TOWNSEND: It should be done. 883 00:47:27,030 --> 00:47:28,760 It hasn't been done yet. 884 00:47:28,760 --> 00:47:31,760 No, I don't see any intrinsic difficulty. 885 00:47:31,760 --> 00:47:34,050 Actually, it's more like a real opportunity, 886 00:47:34,050 --> 00:47:36,290 because when you're on the ground like this, 887 00:47:36,290 --> 00:47:39,160 you should be able to see-- 888 00:47:39,160 --> 00:47:42,740 and in my data, you measure the capital stock and labor input 889 00:47:42,740 --> 00:47:43,240 and all of-- 890 00:47:47,380 --> 00:47:49,910 you know, if it's real technological improvement it's 891 00:47:49,910 --> 00:47:53,810 like, mm, getting rid of the water buffalo, 892 00:47:53,810 --> 00:47:55,880 and bringing in the walking tractor. 893 00:47:55,880 --> 00:47:58,310 I showed you a picture of the walking tractor. 894 00:47:58,310 --> 00:48:00,405 I said it was land, labor, and capital. 895 00:48:00,405 --> 00:48:01,850 Remember that one? 896 00:48:01,850 --> 00:48:03,380 Well, it used to be water buffalo 897 00:48:03,380 --> 00:48:06,740 and the people in Bangkok still think 898 00:48:06,740 --> 00:48:09,200 that you know the farmers are out there with their water 899 00:48:09,200 --> 00:48:09,730 buffalo. 900 00:48:09,730 --> 00:48:12,430 Well, I mean, it's just not true. 901 00:48:12,430 --> 00:48:14,550 You know, they have pickup trucks and all of that. 902 00:48:14,550 --> 00:48:22,970 So my point is that that's real technological progress. 903 00:48:22,970 --> 00:48:25,370 And so data like this potentially 904 00:48:25,370 --> 00:48:31,280 could tell you whether it's really that or just 905 00:48:31,280 --> 00:48:34,040 potentially something else. 906 00:48:34,040 --> 00:48:41,550 It could be better roads, you know, higher, easier access 907 00:48:41,550 --> 00:48:45,390 to markets that somehow showing up, 908 00:48:45,390 --> 00:48:51,080 everything else equal, but not modeled, and not 909 00:48:51,080 --> 00:48:51,890 at a deep level. 910 00:48:51,890 --> 00:48:52,390 OK. 911 00:48:55,840 --> 00:48:58,640 So here's the max problem. 912 00:48:58,640 --> 00:49:02,560 Maximize discounted expected utility of this household. 913 00:49:06,100 --> 00:49:10,080 Rho is sort of the related to the coefficient 914 00:49:10,080 --> 00:49:11,580 of relative risk aversion. 915 00:49:11,580 --> 00:49:14,030 Beta is the discount rate. 916 00:49:14,030 --> 00:49:16,450 This is as of, quote, "initial period." 917 00:49:19,900 --> 00:49:21,910 Here's the liquidity. 918 00:49:21,910 --> 00:49:24,700 That, by the way, was the same as interest 919 00:49:24,700 --> 00:49:29,350 on savings, short-term liquid savings plus current income. 920 00:49:29,350 --> 00:49:32,320 And it can be spent on consumption or saving 921 00:49:32,320 --> 00:49:34,550 or investment. 922 00:49:34,550 --> 00:49:40,990 If you decide, D, to do the size investment that has arrived. 923 00:49:40,990 --> 00:49:43,360 What are the state variables? 924 00:49:43,360 --> 00:49:47,600 In any given day-- this is true for every day, t. 925 00:49:47,600 --> 00:49:49,570 It would have been better, and I don't have it 926 00:49:49,570 --> 00:49:52,270 on the slides, to show you a traditional value function, 927 00:49:52,270 --> 00:49:56,530 where you'd have the utility today plus the value tomorrow. 928 00:49:56,530 --> 00:50:01,270 But this does tell you what the state variables are namely, 929 00:50:01,270 --> 00:50:05,140 permanent income today, liquidity today, 930 00:50:05,140 --> 00:50:08,170 and the size of the investment draw. 931 00:50:08,170 --> 00:50:11,100 So those are the key economic states. 932 00:50:11,100 --> 00:50:12,780 That's households are going to face. 933 00:50:12,780 --> 00:50:15,180 This is just a parameter. 934 00:50:15,180 --> 00:50:16,950 What is s bar? 935 00:50:16,950 --> 00:50:18,360 Here it is. 936 00:50:18,360 --> 00:50:21,480 Now, it looks like savings is bounded from below, 937 00:50:21,480 --> 00:50:24,370 weird, weird. 938 00:50:24,370 --> 00:50:29,290 Well, actually savings can go negative. 939 00:50:29,290 --> 00:50:29,980 That's fine. 940 00:50:29,980 --> 00:50:32,020 That's just borrowing. 941 00:50:32,020 --> 00:50:33,760 So this is a credit limit. 942 00:50:33,760 --> 00:50:37,120 It says savings can't go too negative, 943 00:50:37,120 --> 00:50:39,370 credit can't be too big. 944 00:50:39,370 --> 00:50:43,060 And it's scale by permanent income. 945 00:50:43,060 --> 00:50:44,880 In fact, almost everything in the model 946 00:50:44,880 --> 00:50:47,200 is going to get scaled by permanent income. 947 00:50:53,460 --> 00:50:57,013 Part of that is the permanent income, we wanted growth. 948 00:50:57,013 --> 00:50:58,680 And the way to get the growth is to have 949 00:50:58,680 --> 00:51:01,977 this expansion exogenously in permanent income. 950 00:51:01,977 --> 00:51:03,810 But the other thing is you look at the data, 951 00:51:03,810 --> 00:51:06,720 and you see, you know, like investments 952 00:51:06,720 --> 00:51:12,180 are large for small households relative to their assets. 953 00:51:12,180 --> 00:51:17,220 And you go to larger households, and their projects scale up. 954 00:51:17,220 --> 00:51:22,190 So it's not like you can sort of accumulate wealth 955 00:51:22,190 --> 00:51:26,020 and save your way out of these non-convexities. 956 00:51:26,020 --> 00:51:28,750 So we scaled everything-- go to the other extreme-- 957 00:51:28,750 --> 00:51:31,140 we scale everything by permanent income. 958 00:51:31,140 --> 00:51:35,160 The arrival of project sizes and so on and so forth 959 00:51:35,160 --> 00:51:38,010 and the shocks are all scale. 960 00:51:38,010 --> 00:51:43,160 And what I'm not writing down, but we do in the paper, 961 00:51:43,160 --> 00:51:49,260 is to actually turn this into a balanced growth path. 962 00:51:49,260 --> 00:51:51,900 And the intuition is pretty much what I said. 963 00:51:51,900 --> 00:51:55,250 The math is trickier, where you just 964 00:51:55,250 --> 00:51:58,960 divide through by p everywhere. 965 00:51:58,960 --> 00:52:01,540 So the control variables are things 966 00:52:01,540 --> 00:52:08,950 like consumption per permanent income, 967 00:52:08,950 --> 00:52:10,480 per unit permanent income. 968 00:52:13,590 --> 00:52:16,170 And then we have to solve the value functions. 969 00:52:19,250 --> 00:52:23,450 Now, these are the stochastic processes 970 00:52:23,450 --> 00:52:27,260 for the transitory shock. 971 00:52:27,260 --> 00:52:29,620 Again, everything is in logs. 972 00:52:29,620 --> 00:52:34,320 The permanent shock, so these are log normal, 973 00:52:34,320 --> 00:52:37,530 centered around zero with these standard deviations. 974 00:52:37,530 --> 00:52:39,720 Project side is not centered around zero. 975 00:52:39,720 --> 00:52:45,150 It has a non-trivial mean with the standard deviation. 976 00:52:45,150 --> 00:52:50,490 And you're going to see a table with parameters-- 977 00:52:50,490 --> 00:52:53,340 not to say you're really memorizing these. 978 00:52:53,340 --> 00:52:56,970 I am, because I want to go through the list 979 00:52:56,970 --> 00:52:58,150 when we get there-- 980 00:52:58,150 --> 00:53:01,110 mu sigma, U, N. 981 00:53:01,110 --> 00:53:02,850 Here's a problem. 982 00:53:07,780 --> 00:53:09,640 We do have a borrowing limit. 983 00:53:09,640 --> 00:53:12,500 And we do not put in sort of a natural borrowing. 984 00:53:12,500 --> 00:53:20,260 We let households default. They default in the data, about 18%. 985 00:53:20,260 --> 00:53:22,250 It's not trivial. 986 00:53:22,250 --> 00:53:26,180 It's also hard to nail down exactly when a loan is not 987 00:53:26,180 --> 00:53:29,090 performing, because they can stretch out the payments. 988 00:53:29,090 --> 00:53:31,970 So it's not nailed at exactly 18%. 989 00:53:31,970 --> 00:53:37,370 And, oh, yeah, so here's another story with a bonus. 990 00:53:40,100 --> 00:53:45,900 Econometrica requires the codes for any published paper. 991 00:53:45,900 --> 00:53:48,960 So you've got to give them the codes and the data, 992 00:53:48,960 --> 00:53:49,820 which is good. 993 00:53:49,820 --> 00:53:51,090 It's a good thing. 994 00:53:51,090 --> 00:53:53,220 Of course, when you're doing the work, 995 00:53:53,220 --> 00:53:59,018 you don't really annotate as thoroughly as you might do. 996 00:53:59,018 --> 00:54:01,060 And then when you're ready to publish this stuff, 997 00:54:01,060 --> 00:54:02,602 you're going to like go back over it, 998 00:54:02,602 --> 00:54:05,600 and we found a mistake. 999 00:54:05,600 --> 00:54:08,740 And the mistake was around this default rate. 1000 00:54:08,740 --> 00:54:11,050 So we actually know, for bad reasons, 1001 00:54:11,050 --> 00:54:12,355 that the default rate matters. 1002 00:54:14,950 --> 00:54:15,777 Yes. 1003 00:54:15,777 --> 00:54:18,710 AUDIENCE: So I saw two versions of a paper 1004 00:54:18,710 --> 00:54:22,077 and they had quite different welfare coverages. 1005 00:54:22,077 --> 00:54:23,067 Was that-- 1006 00:54:23,067 --> 00:54:25,150 ROBERT TOWNSEND: That's probably part of it, yeah. 1007 00:54:27,720 --> 00:54:30,600 Well, look, everyone makes mistakes in research. 1008 00:54:30,600 --> 00:54:33,660 And it's in some respect doing it over and over 1009 00:54:33,660 --> 00:54:36,270 is actually a good thing, because you 1010 00:54:36,270 --> 00:54:38,160 get this robustness check. 1011 00:54:38,160 --> 00:54:40,582 But, yeah-- 1012 00:54:40,582 --> 00:54:43,610 AUDIENCE: So in the data, when we say a household defaults, 1013 00:54:43,610 --> 00:54:47,052 what do we mean exactly by that? 1014 00:54:47,052 --> 00:54:48,760 ROBERT TOWNSEND: Never paid off the loan. 1015 00:54:48,760 --> 00:54:49,302 AUDIENCE: OK. 1016 00:54:49,302 --> 00:54:51,020 Because in most microcredit literature 1017 00:54:51,020 --> 00:54:54,360 a default is what we would call a delinquency 1018 00:54:54,360 --> 00:54:57,283 in the sort of developed country finance literatures. 1019 00:54:57,283 --> 00:54:58,200 ROBERT TOWNSEND: Yeah. 1020 00:54:58,200 --> 00:55:00,575 AUDIENCE: So you're not counting deliquincies as default. 1021 00:55:00,575 --> 00:55:02,610 You're counting actual defaults, like we just 1022 00:55:02,610 --> 00:55:04,222 never pay the money back. 1023 00:55:04,222 --> 00:55:06,680 ROBERT TOWNSEND: You know, I have to go back and make sure. 1024 00:55:06,680 --> 00:55:10,700 We could have put something like a two-year threshold on this. 1025 00:55:13,610 --> 00:55:16,600 But, yeah, I mean, I always actually go the other way. 1026 00:55:16,600 --> 00:55:19,360 I tend to think of defaults as a contingency in the loan 1027 00:55:19,360 --> 00:55:20,410 contract. 1028 00:55:20,410 --> 00:55:22,090 The Bank for Agriculture for sure 1029 00:55:22,090 --> 00:55:26,690 runs an insurance company this way, where they extend loans 1030 00:55:26,690 --> 00:55:29,780 and actually even more than that don't charge you 1031 00:55:29,780 --> 00:55:33,970 interest on the renewed principle and so on. 1032 00:55:33,970 --> 00:55:38,970 So, you know, people are risk averse, 1033 00:55:38,970 --> 00:55:41,520 having these contingencies are a good thing. 1034 00:55:41,520 --> 00:55:45,870 There is a literature on this, other literatures. 1035 00:55:50,060 --> 00:55:53,240 So anyway, there may be other ways to do it. 1036 00:55:53,240 --> 00:55:54,290 Measurement is tricky. 1037 00:55:54,290 --> 00:55:57,290 But we decided to put default in. 1038 00:55:57,290 --> 00:55:58,717 And what is this thing saying? 1039 00:55:58,717 --> 00:56:00,050 Well, it's a bit tricky to read. 1040 00:56:00,050 --> 00:56:03,170 But we don't drive people down to zero consumption. 1041 00:56:03,170 --> 00:56:05,780 We put in some minimum consumption. 1042 00:56:05,780 --> 00:56:10,100 No matter what, you're not going to go below c lower bar 1043 00:56:10,100 --> 00:56:15,530 and, well, scaled by your permanent income 1044 00:56:15,530 --> 00:56:17,960 as everything else is. 1045 00:56:17,960 --> 00:56:19,310 Now, what is this other thing? 1046 00:56:19,310 --> 00:56:19,940 Savings. 1047 00:56:19,940 --> 00:56:23,110 Well, put savings times p on the right-hand side. 1048 00:56:23,110 --> 00:56:25,550 It'll pick up a negative sign. 1049 00:56:25,550 --> 00:56:26,690 That's credit. 1050 00:56:26,690 --> 00:56:28,710 Minus savings is credit. 1051 00:56:28,710 --> 00:56:32,450 So it's liquidity plus the maximum loan 1052 00:56:32,450 --> 00:56:33,440 you could take out. 1053 00:56:36,360 --> 00:56:46,610 And still, you can't cover that minimum consumption. 1054 00:56:46,610 --> 00:56:50,230 So you're not respecting the budget constraint anymore. 1055 00:56:50,230 --> 00:56:51,610 You get this, quote, "gift." 1056 00:56:55,740 --> 00:56:59,460 And then here are the rules that happen and when you default. 1057 00:56:59,460 --> 00:57:01,800 You're not investing. 1058 00:57:01,800 --> 00:57:03,660 You're borrowing up to the limit. 1059 00:57:03,660 --> 00:57:06,375 And your consumption is the scaled version 1060 00:57:06,375 --> 00:57:07,550 of permanent income. 1061 00:57:10,730 --> 00:57:14,770 So let's see if I've done my homework, or you have. 1062 00:57:14,770 --> 00:57:16,550 R is the interest rate. 1063 00:57:16,550 --> 00:57:21,250 This is the standard deviation of the permanent shock, 1064 00:57:21,250 --> 00:57:23,830 standard deviation of the transitory shock. 1065 00:57:23,830 --> 00:57:24,850 This is new. 1066 00:57:24,850 --> 00:57:26,650 It's measurement error. 1067 00:57:26,650 --> 00:57:29,470 We imagine that we see the data contaminated. 1068 00:57:29,470 --> 00:57:33,710 It's a bit dirty centered around the mean. 1069 00:57:33,710 --> 00:57:36,600 But this is the drift term. 1070 00:57:36,600 --> 00:57:39,810 This is the lower bound on consumption under default. 1071 00:57:39,810 --> 00:57:40,980 Beta is the discount rate. 1072 00:57:40,980 --> 00:57:44,390 Rho is related to constant relative risk aversion. 1073 00:57:44,390 --> 00:57:47,160 Mu is the project size. 1074 00:57:47,160 --> 00:57:50,630 On average, sigma i is the standard deviation 1075 00:57:50,630 --> 00:57:51,510 of project size. 1076 00:57:51,510 --> 00:57:54,660 And this is the credit limit. 1077 00:57:54,660 --> 00:57:58,500 And R, big R, is the return on investment 1078 00:57:58,500 --> 00:58:00,750 that augments permanent income. 1079 00:58:00,750 --> 00:58:04,700 That's also quite problematic. 1080 00:58:04,700 --> 00:58:09,050 What we did was use these return on assets numbers. 1081 00:58:09,050 --> 00:58:12,770 But, Emily and Abhijit and I have been working subsequently. 1082 00:58:12,770 --> 00:58:17,300 And we think not only does R vary 1083 00:58:17,300 --> 00:58:20,180 in the population with some heterogeneity 1084 00:58:20,180 --> 00:58:23,000 that we don't have here, but it actually 1085 00:58:23,000 --> 00:58:25,430 is predictive of who's getting the money 1086 00:58:25,430 --> 00:58:30,920 and what they're doing with it, which may be another reason 1087 00:58:30,920 --> 00:58:35,000 that we didn't get as far as we thought we might 1088 00:58:35,000 --> 00:58:36,475 with the investment thing. 1089 00:58:36,475 --> 00:58:40,490 But anyway hopefully, we'll write that up real soon. 1090 00:58:40,490 --> 00:58:45,250 Here, it's just a number, calibrated, you might say. 1091 00:58:45,250 --> 00:58:48,090 And we're going to use method of moments. 1092 00:58:48,090 --> 00:58:52,080 Oh, now, a couple of questions for you guys, or one comment. 1093 00:58:52,080 --> 00:58:55,200 Back to the codes, you know, Emily 1094 00:58:55,200 --> 00:59:00,930 was about your cohort in a class like this. 1095 00:59:00,930 --> 00:59:03,300 And very shortly thereafter, she got interested 1096 00:59:03,300 --> 00:59:05,070 in the structural stuff. 1097 00:59:05,070 --> 00:59:06,180 The codes are all there. 1098 00:59:06,180 --> 00:59:09,840 And, you know, not only the published version, 1099 00:59:09,840 --> 00:59:12,090 but all the other things, you're welcome to have them. 1100 00:59:14,983 --> 00:59:16,900 Probably a little late to do it for the class, 1101 00:59:16,900 --> 00:59:21,460 but it's a resource that's available to you. 1102 00:59:21,460 --> 00:59:25,300 And, of course, in the end it wasn't the paper with Abhijit 1103 00:59:25,300 --> 00:59:26,860 that's still not quite done. 1104 00:59:26,860 --> 00:59:30,580 But she used structural modeling in her job market paper. 1105 00:59:30,580 --> 00:59:33,750 And she was learning it by looking at those codes. 1106 00:59:33,750 --> 00:59:38,550 So the other thing I want to ask you, 1107 00:59:38,550 --> 00:59:43,880 I hope you've seen at some point methods of simulated moments. 1108 00:59:43,880 --> 00:59:48,940 One yes, three, four yes, OK. 1109 00:59:48,940 --> 00:59:50,230 So we're going to-- 1110 00:59:50,230 --> 00:59:53,720 AUDIENCE: For those of us who took Devop 771 in the Fall, 1111 00:59:53,720 --> 00:59:55,910 Esther had a problem set on your papers. 1112 00:59:58,900 --> 01:00:00,900 ROBERT TOWNSEND: Have you learned have something 1113 01:00:00,900 --> 01:00:02,064 new today? 1114 01:00:02,064 --> 01:00:05,382 AUDIENCE: Yeah, yeah. 1115 01:00:05,382 --> 01:00:07,567 Like we did the methods of moments already. 1116 01:00:07,567 --> 01:00:08,400 ROBERT TOWNSEND: OK. 1117 01:00:11,440 --> 01:00:13,040 All right, that's excellent. 1118 01:00:13,040 --> 01:00:16,560 Yeah, that's excellent. 1119 01:00:16,560 --> 01:00:17,400 I guess I knew that. 1120 01:00:17,400 --> 01:00:19,640 I'd forgotten. 1121 01:00:19,640 --> 01:00:21,630 But anyone, I'm glad you know it. 1122 01:00:21,630 --> 01:00:24,130 So I won't try to reteach. 1123 01:00:24,130 --> 01:00:28,120 There's not time today. 1124 01:00:28,120 --> 01:00:32,160 OK, another controversy, what to do 1125 01:00:32,160 --> 01:00:34,230 with all the heterogeneity in the data 1126 01:00:34,230 --> 01:00:37,110 that we don't have in the model? 1127 01:00:37,110 --> 01:00:40,110 I'm emphasizing certain aspects of observed and unobserved 1128 01:00:40,110 --> 01:00:42,330 heterogeneity in the data. 1129 01:00:42,330 --> 01:00:45,159 That is in the model. 1130 01:00:45,159 --> 01:00:49,630 But we've also got all these demographic stuff. 1131 01:00:49,630 --> 01:00:55,730 Now, granted not everyone would do it this way. 1132 01:00:55,730 --> 01:00:57,490 We take it out. 1133 01:00:57,490 --> 01:00:59,420 So we filter the data basically. 1134 01:00:59,420 --> 01:01:04,800 We regress the household data on these observables 1135 01:01:04,800 --> 01:01:12,960 and as well as time trends, potential initial business 1136 01:01:12,960 --> 01:01:15,250 cycle, i.e. 1137 01:01:15,250 --> 01:01:18,000 the initial data is one year after the financial crisis, 1138 01:01:18,000 --> 01:01:19,890 the whole country is kind of-- 1139 01:01:19,890 --> 01:01:25,340 so the controversy is, well, why don't you just put it in-- 1140 01:01:25,340 --> 01:01:27,150 and I'm not going to have time this year. 1141 01:01:27,150 --> 01:01:29,370 But I have lectured in previous years 1142 01:01:29,370 --> 01:01:32,070 on Keane and Wolpin and stuff. 1143 01:01:32,070 --> 01:01:35,490 So if you want to see a really, really long specification 1144 01:01:35,490 --> 01:01:39,690 of what you can put into the utility function, 1145 01:01:39,690 --> 01:01:40,973 I'm not debunking it. 1146 01:01:40,973 --> 01:01:41,640 It's a perfect-- 1147 01:01:41,640 --> 01:01:46,086 [AUDIO OUT] 1148 01:02:18,393 --> 01:02:20,060 ROBERT TOWNSEND: We kind of estimate it. 1149 01:02:20,060 --> 01:02:21,920 But this is almost like calibration, 1150 01:02:21,920 --> 01:02:27,430 because we have earned interest, which is interest time savings. 1151 01:02:27,430 --> 01:02:30,200 And we have savings and interest in the data. 1152 01:02:30,200 --> 01:02:35,100 So we are trying to minimize to get the mean right. 1153 01:02:38,100 --> 01:02:39,780 The other things that are more fun-- 1154 01:02:39,780 --> 01:02:48,590 consumption, decision to invest, and so on is just basically 1155 01:02:48,590 --> 01:02:53,260 looking at consumption in the data. 1156 01:02:53,260 --> 01:02:57,550 And this is expected consumption through the lens of the model. 1157 01:02:57,550 --> 01:03:00,060 It's conditioned on observables. 1158 01:03:00,060 --> 01:03:01,810 Liquidity is observed. 1159 01:03:01,810 --> 01:03:03,990 And income is observed. 1160 01:03:03,990 --> 01:03:08,190 Now, one caution, these are not one to one with the key state 1161 01:03:08,190 --> 01:03:10,920 variables of the model itself. 1162 01:03:10,920 --> 01:03:13,950 There are something like permanent income 1163 01:03:13,950 --> 01:03:18,790 and project size relative to permanent income 1164 01:03:18,790 --> 01:03:20,050 are the key state variables. 1165 01:03:20,050 --> 01:03:21,600 And permanent income is unobserved. 1166 01:03:21,600 --> 01:03:28,840 So there's some work to back out these expectations. 1167 01:03:28,840 --> 01:03:30,940 But anyway, you get an error, which 1168 01:03:30,940 --> 01:03:32,530 is the difference between the observed 1169 01:03:32,530 --> 01:03:33,670 and the predicted values. 1170 01:03:33,670 --> 01:03:36,520 And you kind of like want to minimize the error term 1171 01:03:36,520 --> 01:03:42,310 by choosing the parameter values here. 1172 01:03:42,310 --> 01:03:50,410 Here's a la Blundell, Pistaferri and Preston. 1173 01:03:50,410 --> 01:03:57,270 We put in a log moving average process for income. 1174 01:03:57,270 --> 01:04:02,480 So if you're judicious about choosing log differences far 1175 01:04:02,480 --> 01:04:05,780 enough apart, you can pick up the drift. 1176 01:04:05,780 --> 01:04:10,250 It's really very similar to what BPP we're doing. 1177 01:04:13,230 --> 01:04:17,640 And that's how we get that G basically. 1178 01:04:17,640 --> 01:04:20,190 So these are time differences in growth rates. 1179 01:04:23,770 --> 01:04:26,310 Maybe I don't need to-- and then, of course, 1180 01:04:26,310 --> 01:04:31,570 you can pick up other moments by the orthogonality conditions 1181 01:04:31,570 --> 01:04:39,333 on error terms, assuming that they don't see something else. 1182 01:04:39,333 --> 01:04:40,750 If they were to see something else 1183 01:04:40,750 --> 01:04:43,003 we're not seeing in the model, then 1184 01:04:43,003 --> 01:04:44,920 we're going to make a mistake with these guys, 1185 01:04:44,920 --> 01:04:48,490 because there's going to be information contained 1186 01:04:48,490 --> 01:04:50,490 in the error. 1187 01:04:50,490 --> 01:04:52,360 But we assume not. 1188 01:04:52,360 --> 01:04:55,420 Simulating, what are we going to do? 1189 01:04:55,420 --> 01:05:00,200 Well, first of all, how do we relax the borrowing limit? 1190 01:05:00,200 --> 01:05:03,890 We allow that borrowing limit, that s bar, 1191 01:05:03,890 --> 01:05:09,850 to move from one village to the next in such a way 1192 01:05:09,850 --> 01:05:12,910 as to predict the increase in short-term village credit 1193 01:05:12,910 --> 01:05:15,020 that we see in the data. 1194 01:05:15,020 --> 01:05:19,170 So that's how we sort of calibrate the magnitude 1195 01:05:19,170 --> 01:05:20,507 of the intervention. 1196 01:05:24,890 --> 01:05:27,410 And the rest is simulation. 1197 01:05:27,410 --> 01:05:31,130 We're going to draw these shocks over and over again and get 1198 01:05:31,130 --> 01:05:32,210 repeated samples. 1199 01:05:32,210 --> 01:05:36,490 We're going to have basically 500 artificial data sets. 1200 01:05:36,490 --> 01:05:39,170 Any one data set is a series of draws 1201 01:05:39,170 --> 01:05:46,130 of the permanent transitory project size and so on. 1202 01:05:46,130 --> 01:05:50,180 And we have the pre-intervention years. 1203 01:05:50,180 --> 01:05:52,940 You can stop there and use the data 1204 01:05:52,940 --> 01:05:54,560 and estimate the parameters. 1205 01:05:54,560 --> 01:05:59,120 You can keep going and see what the model predicts 1206 01:05:59,120 --> 01:06:01,670 over and over again, not just one path, 1207 01:06:01,670 --> 01:06:04,280 but averaging over these 500 different paths. 1208 01:06:10,070 --> 01:06:10,850 Why do that? 1209 01:06:10,850 --> 01:06:14,160 Well, basically, you want to get average tendencies. 1210 01:06:14,160 --> 01:06:19,790 You don't want to be so sensitive to the luck 1211 01:06:19,790 --> 01:06:22,820 of the draw in the first year after the intervention 1212 01:06:22,820 --> 01:06:25,770 and the second year after the intervention, 1213 01:06:25,770 --> 01:06:29,370 because we don't know what-- 1214 01:06:29,370 --> 01:06:33,130 the model says-- we know something about averages. 1215 01:06:33,130 --> 01:06:34,255 In fact, then we run-- 1216 01:06:48,030 --> 01:06:50,520 this slide really deteriorated through the projector. 1217 01:06:50,520 --> 01:06:58,360 But these are the economic variables in the model itself. 1218 01:06:58,360 --> 01:07:01,000 Liquidity divided by permanent income, 1219 01:07:01,000 --> 01:07:04,920 projects size divided by permanent income. 1220 01:07:04,920 --> 01:07:11,350 And what's going on here, as you move in this direction, 1221 01:07:11,350 --> 01:07:16,120 there's three black dots here. 1222 01:07:16,120 --> 01:07:18,940 And they're kind of all fixing project 1223 01:07:18,940 --> 01:07:22,300 size and varying liquidity. 1224 01:07:22,300 --> 01:07:25,630 So these guys down here are bankrupt. 1225 01:07:25,630 --> 01:07:28,430 They're defaulting. 1226 01:07:28,430 --> 01:07:31,410 They have almost no liquidity. 1227 01:07:31,410 --> 01:07:36,480 They can't even really pay back their previous loans. 1228 01:07:36,480 --> 01:07:39,462 These guys, these are these hand-to-mouth guys. 1229 01:07:39,462 --> 01:07:40,545 This thing's really curvy. 1230 01:07:43,860 --> 01:07:45,390 They're constrained. 1231 01:07:45,390 --> 01:07:47,940 They're liquidity constrained. 1232 01:07:47,940 --> 01:07:49,980 That borrowing constraint is binding. 1233 01:07:49,980 --> 01:07:53,310 These guys have drifted into higher territory. 1234 01:07:53,310 --> 01:07:55,380 They're not actually constrained anymore. 1235 01:07:55,380 --> 01:07:58,820 But in the future, they know there are credit limits and so 1236 01:07:58,820 --> 01:08:00,480 on. 1237 01:08:00,480 --> 01:08:03,500 If you go this way, you're basically 1238 01:08:03,500 --> 01:08:08,340 fixing liquidity and varying project size. 1239 01:08:08,340 --> 01:08:10,980 Now, the most interesting part of the diagram 1240 01:08:10,980 --> 01:08:13,470 is the Grand Canyon here. 1241 01:08:16,279 --> 01:08:21,560 And this sort of pre-intervention, 1242 01:08:21,560 --> 01:08:28,229 this guy sitting close to the cliff, he's tempted to invest. 1243 01:08:28,229 --> 01:08:30,450 But given limited resources, investment 1244 01:08:30,450 --> 01:08:33,270 would drive his consumption really low. 1245 01:08:33,270 --> 01:08:36,010 And he's not willing to do it. 1246 01:08:36,010 --> 01:08:38,770 But then you have this sort of increment in the borrowing 1247 01:08:38,770 --> 01:08:40,210 resources. 1248 01:08:40,210 --> 01:09:01,189 It's like another [AUDIO OUT] that effect on consumption, 1249 01:09:01,189 --> 01:09:03,640 we've got these four. 1250 01:09:03,640 --> 01:09:06,859 These guys don't move consumption basically. 1251 01:09:06,859 --> 01:09:09,939 Oh, well, let me get at the welfare stuff right now. 1252 01:09:09,939 --> 01:09:18,330 So for them, this credit program is terrible. 1253 01:09:18,330 --> 01:09:19,710 Why? 1254 01:09:19,710 --> 01:09:22,170 Because we're not going to force them into default anymore. 1255 01:09:22,170 --> 01:09:23,545 We're going to make them take out 1256 01:09:23,545 --> 01:09:27,580 a loan at interest, whereas before they got exempt 1257 01:09:27,580 --> 01:09:29,359 from repayment. 1258 01:09:29,359 --> 01:09:33,350 So welfare goes down for these guys down here 1259 01:09:33,350 --> 01:09:35,750 as a result of the village fund program. 1260 01:09:35,750 --> 01:09:38,875 Welfare goes up for these credit constrained guys, 1261 01:09:38,875 --> 01:09:40,500 because they're at the borrowing limit, 1262 01:09:40,500 --> 01:09:42,330 and that's precisely what this new program 1263 01:09:42,330 --> 01:09:44,372 is doing, at least through the lens of the model. 1264 01:09:46,771 --> 01:09:48,979 These guys, you'd say, well, they're not constrained. 1265 01:09:48,979 --> 01:09:52,490 No, but they're these buffer stock guys. 1266 01:09:52,490 --> 01:09:56,850 And they now have excess liquidity. 1267 01:09:56,850 --> 01:09:59,650 Well, might as well spend some of it. 1268 01:09:59,650 --> 01:10:02,890 So their consumption also goes up. 1269 01:10:02,890 --> 01:10:04,540 And as I just said, these guys who 1270 01:10:04,540 --> 01:10:10,440 drop off the cliff, so to speak, they actually 1271 01:10:10,440 --> 01:10:11,940 have consumption drops. 1272 01:10:11,940 --> 01:10:16,440 So this consumption average that we see, 1 to 1, 1273 01:10:16,440 --> 01:10:23,340 is this mongrel weighted average of overall this heterogeneity, 1274 01:10:23,340 --> 01:10:25,980 of course, that we only create sort of 1275 01:10:25,980 --> 01:10:28,140 through the eye of the model. 1276 01:10:28,140 --> 01:10:28,840 Yes. 1277 01:10:28,840 --> 01:10:30,470 AUDIENCE: Do you understand the guy 1278 01:10:30,470 --> 01:10:33,580 who's like constantly bankrupt is worse off? 1279 01:10:33,580 --> 01:10:36,120 ROBERT TOWNSEND: Because before we would have forgiven 1280 01:10:36,120 --> 01:10:40,920 the loans, sent him to a minimum consumption c bar and let 1281 01:10:40,920 --> 01:10:42,660 him start over. 1282 01:10:42,660 --> 01:10:46,020 And now, he's saddled with all that previous debt. 1283 01:10:46,020 --> 01:10:51,120 And he can basically borrow more against it at interest rates 1284 01:10:51,120 --> 01:10:53,575 and then have to pay off in the future. 1285 01:10:53,575 --> 01:10:55,800 AUDIENCE: Oh, I see. 1286 01:10:55,800 --> 01:10:58,490 Because we expanded his ability to refinance, 1287 01:10:58,490 --> 01:11:00,720 now he's worse off. 1288 01:11:00,720 --> 01:11:02,410 ROBERT TOWNSEND: Yeah, exactly. 1289 01:11:02,410 --> 01:11:03,150 AUDIENCE: That's different-- 1290 01:11:03,150 --> 01:11:04,800 ROBERT TOWNSEND: If the government exempts your loan 1291 01:11:04,800 --> 01:11:05,883 and you stay in the house. 1292 01:11:05,883 --> 01:11:06,523 But-- 1293 01:11:06,523 --> 01:11:07,940 AUDIENCE: So if would be different 1294 01:11:07,940 --> 01:11:09,857 if you were borrowing from a money lending who 1295 01:11:09,857 --> 01:11:11,290 is going to break his knee. 1296 01:11:11,290 --> 01:11:14,570 ROBERT TOWNSEND: Yeah, so you could imagine default 1297 01:11:14,570 --> 01:11:15,920 doesn't work like this. 1298 01:11:15,920 --> 01:11:20,060 But that's the way it works in the model. 1299 01:11:20,060 --> 01:11:23,120 Anyway, and then you know after they climb out of the cliff, 1300 01:11:23,120 --> 01:11:26,300 you get a repeat in terms of-- 1301 01:11:26,300 --> 01:11:32,140 so if you looked at those welfare numbers, 1302 01:11:32,140 --> 01:11:34,450 it's almost misleading. 1303 01:11:34,450 --> 01:11:38,060 In fact, I thought it was wrong for a minute. 1304 01:11:38,060 --> 01:11:40,070 This looks like it's just-- 1305 01:11:40,070 --> 01:11:43,660 the bankruptcy region, there are these huge gains. 1306 01:11:43,660 --> 01:11:46,640 No, the bankruptcy, these guys are losing. 1307 01:11:46,640 --> 01:11:48,710 They're negative numbers down here. 1308 01:11:48,710 --> 01:11:52,740 What happens is very close, but not very 1309 01:11:52,740 --> 01:11:55,410 adjacent to that bankruptcy reason, 1310 01:11:55,410 --> 01:11:59,220 the welfare gains just you know rise like a tent. 1311 01:11:59,220 --> 01:12:00,870 It's actually not vertical. 1312 01:12:00,870 --> 01:12:04,200 It's just coming up real fast and then coming down. 1313 01:12:04,200 --> 01:12:05,410 It's a very steep tent. 1314 01:12:08,370 --> 01:12:11,880 So these guys are really in the program. 1315 01:12:11,880 --> 01:12:13,860 And then these other constrained people, 1316 01:12:13,860 --> 01:12:15,810 they're kind of liking the program. 1317 01:12:15,810 --> 01:12:22,540 All these people are actually relative to what? 1318 01:12:22,540 --> 01:12:27,150 Well, basically, we use the model 1319 01:12:27,150 --> 01:12:29,580 to look at an alternative program, 1320 01:12:29,580 --> 01:12:33,170 where they just get a lump sum transfer. 1321 01:12:33,170 --> 01:12:36,350 And we actually do this consumption equivalent 1322 01:12:36,350 --> 01:12:39,470 calculation, which is how much of a transfer 1323 01:12:39,470 --> 01:12:43,460 would you have to get to have a welfare 1324 01:12:43,460 --> 01:12:46,160 gain equivalent to the one you get under a Million Baht 1325 01:12:46,160 --> 01:12:47,800 Program. 1326 01:12:47,800 --> 01:12:53,710 And another way to say it, in this region, the government 1327 01:12:53,710 --> 01:12:55,600 could have saved money. 1328 01:12:55,600 --> 01:12:58,630 It could have gotten households the equivalent welfare gain 1329 01:12:58,630 --> 01:13:02,360 without actually putting so much money in the village. 1330 01:13:02,360 --> 01:13:04,730 But it's very heterogeneous. 1331 01:13:04,730 --> 01:13:07,460 Some people are tempted, and the abstract on the paper 1332 01:13:07,460 --> 01:13:11,000 actually talks about it not being a great program. 1333 01:13:11,000 --> 01:13:14,240 But I don't know, strictly speaking, Pareto criterion, 1334 01:13:14,240 --> 01:13:15,080 we can't say that. 1335 01:13:15,080 --> 01:13:18,890 24% of the population really love this program. 1336 01:13:18,890 --> 01:13:21,330 For them, it's wonderful. 1337 01:13:21,330 --> 01:13:24,390 But more than the majority would have 1338 01:13:24,390 --> 01:13:27,034 preferred a different program. 1339 01:13:27,034 --> 01:13:27,678 Yep. 1340 01:13:27,678 --> 01:13:29,220 AUDIENCE: So that type of comparisons 1341 01:13:29,220 --> 01:13:31,125 is just looking in people's welfare 1342 01:13:31,125 --> 01:13:34,820 in the model, not thinking about cushion being administer, 1343 01:13:34,820 --> 01:13:36,500 for instance. 1344 01:13:36,500 --> 01:13:38,500 ROBERT TOWNSEND: Oh, yeah, we're extracting away 1345 01:13:38,500 --> 01:13:41,670 from the people who are financing the programs. 1346 01:13:41,670 --> 01:13:43,600 So there's all those tax distortions 1347 01:13:43,600 --> 01:13:45,430 being created to generate the revenue 1348 01:13:45,430 --> 01:13:48,675 to fund this program that aren't involved in this calculation. 1349 01:13:48,675 --> 01:13:50,050 AUDIENCE: Oh, I was just thinking 1350 01:13:50,050 --> 01:13:51,050 organizational capacity. 1351 01:13:51,050 --> 01:13:54,260 So maybe one reason for loaning to each village 1352 01:13:54,260 --> 01:13:57,210 is that by head count, there's be a lot more variation-- 1353 01:13:57,210 --> 01:13:58,210 ROBERT TOWNSEND: Maybe-- 1354 01:13:58,210 --> 01:13:59,914 AUDIENCE: Than a simple version. 1355 01:14:02,670 --> 01:14:04,440 ROBERT TOWNSEND: And then I just want 1356 01:14:04,440 --> 01:14:10,140 to say that we come back to those IV regressions, two 1357 01:14:10,140 --> 01:14:10,710 comments. 1358 01:14:10,710 --> 01:14:15,630 First of all, what do these coefficients mean? 1359 01:14:15,630 --> 01:14:23,190 Well, you can think about them as impact of the program. 1360 01:14:23,190 --> 01:14:29,780 But that's kind of like a funny number, 1361 01:14:29,780 --> 01:14:33,730 because it's an average over all these different people who 1362 01:14:33,730 --> 01:14:36,430 got treated differently and had different views 1363 01:14:36,430 --> 01:14:38,380 about the program. 1364 01:14:38,380 --> 01:14:43,240 So we're lucky in a way that we saw this big salient number. 1365 01:14:43,240 --> 01:14:47,260 And it helped us figure out a model where something like that 1366 01:14:47,260 --> 01:14:47,890 could happen. 1367 01:14:47,890 --> 01:14:49,630 But then with that model, we realize 1368 01:14:49,630 --> 01:14:56,190 this number is not a uniform benefit for all the households. 1369 01:14:56,190 --> 01:14:58,580 So in other words, a big advantage 1370 01:14:58,580 --> 01:15:00,718 of a structural model conditioned 1371 01:15:00,718 --> 01:15:03,260 on believing that you're getting the structure somewhat right 1372 01:15:03,260 --> 01:15:05,750 is that it allows you to really look 1373 01:15:05,750 --> 01:15:10,000 at the distribution of welfare gains and losses. 1374 01:15:10,000 --> 01:15:13,030 The other comment is more favorable. 1375 01:15:13,030 --> 01:15:17,570 Since we did those IV regressions 1376 01:15:17,570 --> 01:15:20,510 and I reported them, we can run-- 1377 01:15:20,510 --> 01:15:22,370 we can generate-- we did generate-- 1378 01:15:22,370 --> 01:15:26,090 the data from the model over and over again, as I said, 1379 01:15:26,090 --> 01:15:30,230 and then do exactly the same IV type regressions on the data 1380 01:15:30,230 --> 01:15:34,500 from the model and take-- 1381 01:15:34,500 --> 01:15:36,195 so we're comparing apples to apples. 1382 01:15:39,800 --> 01:15:43,290 It's kind of a funny intermediate criterion, 1383 01:15:43,290 --> 01:15:45,630 but it is at least a consistent criterion 1384 01:15:45,630 --> 01:15:47,970 that's used on the model generated data as well 1385 01:15:47,970 --> 01:15:48,930 as the actual data. 1386 01:15:48,930 --> 01:15:55,095 And the fit is actually pretty good. 1387 01:15:55,095 --> 01:16:04,040 So I think that's it. 1388 01:16:04,040 --> 01:16:05,890 Thank you.