1
00:00:00,135 --> 00:00:01,770
The following
content is provided

2
00:00:01,770 --> 00:00:04,059
under a creative
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 --> 00:00:16,560
from hundreds of MIT courses,
visit MIT OpenCourseWare

7
00:00:16,560 --> 00:00:17,785
at ocw.mit.edu.

8
00:00:27,266 --> 00:00:30,850
ROBERT TOWNSEND: That's
the starting point,

9
00:00:30,850 --> 00:00:35,500
but we're going to then move,
in the latter half of the class,

10
00:00:35,500 --> 00:00:38,290
to what to make of it, what
happens if you find that it's

11
00:00:38,290 --> 00:00:41,110
complete or incomplete,
how to think

12
00:00:41,110 --> 00:00:46,980
about policy through
the lens of this model,

13
00:00:46,980 --> 00:00:51,310
how to design insurance
instruments, what damage they

14
00:00:51,310 --> 00:00:54,790
actually might do, what the
distribution of benefits

15
00:00:54,790 --> 00:00:56,320
might be.

16
00:00:56,320 --> 00:00:58,420
There is a recurring
theme today,

17
00:00:58,420 --> 00:01:01,750
which wouldn't surprise
you, knowing me, which

18
00:01:01,750 --> 00:01:09,050
is theory is guiding
the whole discussion,

19
00:01:09,050 --> 00:01:13,730
rather than starting with
some randomized control trial

20
00:01:13,730 --> 00:01:16,850
and targeting, but we
will get to those control

21
00:01:16,850 --> 00:01:18,410
trials in the discussion.

22
00:01:20,960 --> 00:01:22,150
OK.

23
00:01:22,150 --> 00:01:23,260
So here's the outline.

24
00:01:27,550 --> 00:01:29,500
We'll look at risk and
insurance and village

25
00:01:29,500 --> 00:01:32,650
India, move to a
benchmark way of measuring

26
00:01:32,650 --> 00:01:36,700
the effectiveness of
financial institutions

27
00:01:36,700 --> 00:01:38,070
on the formal side.

28
00:01:38,070 --> 00:01:41,380
We'll then couple that with
work on the informal side

29
00:01:41,380 --> 00:01:46,060
and networks and so on.

30
00:01:46,060 --> 00:01:47,080
We'll get to this--

31
00:01:47,080 --> 00:01:49,540
can well-intended
interventions actually

32
00:01:49,540 --> 00:01:53,740
result in well welfare losses?

33
00:01:53,740 --> 00:02:01,300
And then issues of why insurance
take up in many circumstances

34
00:02:01,300 --> 00:02:02,710
is actually quite low.

35
00:02:16,320 --> 00:02:18,750
The idea here is that
people in village India,

36
00:02:18,750 --> 00:02:20,700
and developing economies
more generally,

37
00:02:20,700 --> 00:02:24,900
live in very high-risk
environments.

38
00:02:24,900 --> 00:02:27,510
There are ways to
cope with this risk--

39
00:02:27,510 --> 00:02:29,990
ex-ante and ex-post.

40
00:02:29,990 --> 00:02:35,010
Ex-ante, you can diversify
your crops, fields, activities,

41
00:02:35,010 --> 00:02:37,950
and occupations.

42
00:02:37,950 --> 00:02:43,905
Ex-post, you can have financial
transactions, gifts, loans.

43
00:02:43,905 --> 00:02:46,680
I actually mentioned
this the first lecture.

44
00:02:50,670 --> 00:02:53,610
How well are households doing?

45
00:02:53,610 --> 00:02:59,130
So the idea in this
paper is forget

46
00:02:59,130 --> 00:03:00,900
enumerating all the mechanisms.

47
00:03:00,900 --> 00:03:03,750
You're likely to miss one
anyway and the measurement

48
00:03:03,750 --> 00:03:04,920
might be really bad.

49
00:03:04,920 --> 00:03:07,080
Let's just look at the outcomes.

50
00:03:07,080 --> 00:03:11,700
The null hypothesis-- If
risk sharing were perfect,

51
00:03:11,700 --> 00:03:18,210
what would you expect to see
in consumption and income data?

52
00:03:18,210 --> 00:03:23,740
I think this style of work,
although nowadays it's

53
00:03:23,740 --> 00:03:27,370
not only accepted, it's highly
utilized all over the place,

54
00:03:27,370 --> 00:03:32,710
but initially there was a
confusion that perfect means

55
00:03:32,710 --> 00:03:35,590
complete as incomplete
markets, and somehow you're

56
00:03:35,590 --> 00:03:39,070
saying you're going to have all
these Arrow-Debreu securities

57
00:03:39,070 --> 00:03:41,350
being traded in a
village economy,

58
00:03:41,350 --> 00:03:44,560
and no, that's not the idea.

59
00:03:44,560 --> 00:03:49,270
Informal institutions can be
at work, as well as some spot

60
00:03:49,270 --> 00:03:50,920
market exchanges.

61
00:03:50,920 --> 00:03:52,930
You don't need to take
a stand on any of that

62
00:03:52,930 --> 00:03:54,805
if you're just going to
look at the outcomes.

63
00:03:58,440 --> 00:04:02,170
So the idea is
individual consumption

64
00:04:02,170 --> 00:04:07,210
should actually not depend on
individual income, stunningly.

65
00:04:10,030 --> 00:04:14,120
Once you control for
aggregate consumption,

66
00:04:14,120 --> 00:04:18,560
it's as if all these households
lived in a big risk syndicate,

67
00:04:18,560 --> 00:04:20,839
and they were just pooling
all their grain and so

68
00:04:20,839 --> 00:04:24,530
on in one big pile, and
then depending on shares,

69
00:04:24,530 --> 00:04:27,380
and some people are
more equal than others,

70
00:04:27,380 --> 00:04:31,540
they get a fraction
of the grain.

71
00:04:31,540 --> 00:04:35,840
A pile of the grain can go up
and down with village aggregate

72
00:04:35,840 --> 00:04:40,910
shocks, but controlling
for those shocks,

73
00:04:40,910 --> 00:04:44,390
your contribution has nothing
to do with what you take out.

74
00:04:51,610 --> 00:04:54,475
So where to test?

75
00:04:56,917 --> 00:04:58,750
We're going to do these
tests at the village

76
00:04:58,750 --> 00:05:04,500
level in this starting point of
this lecture, but we'll move.

77
00:05:04,500 --> 00:05:08,790
It could be kinship
networks within a village,

78
00:05:08,790 --> 00:05:10,200
family related people.

79
00:05:10,200 --> 00:05:12,450
It could be bigger than that.

80
00:05:12,450 --> 00:05:14,680
It could be a
cross-village risk sharing.

81
00:05:14,680 --> 00:05:18,480
It could be that being a member
of a financial institution,

82
00:05:18,480 --> 00:05:23,580
for example, is allowing
you to be part of a larger

83
00:05:23,580 --> 00:05:27,580
risk pooling population.

84
00:05:27,580 --> 00:05:30,330
So it's a judgment call
about where to apply it.

85
00:05:30,330 --> 00:05:33,930
Fortunately in the work
today, I can show you it

86
00:05:33,930 --> 00:05:37,470
being applied at various
distinct levels, which

87
00:05:37,470 --> 00:05:42,210
is a theme we'll come back
to next Thursday as well.

88
00:05:42,210 --> 00:05:43,700
Why did I choose villages?

89
00:05:43,700 --> 00:05:47,130
I had some idea that
everybody knows each other,

90
00:05:47,130 --> 00:05:50,580
and the environment
is relatively simple,

91
00:05:50,580 --> 00:05:54,620
and it just seemed like
the idealized setting.

92
00:05:54,620 --> 00:05:58,190
Of course, in practice not
all villages are alike,

93
00:05:58,190 --> 00:06:00,320
and they're pretty
complicated, and you've

94
00:06:00,320 --> 00:06:02,240
got people interacting
with one another,

95
00:06:02,240 --> 00:06:04,280
and they don't all
love each other.

96
00:06:07,090 --> 00:06:09,520
But that was the
motivation to begin with.

97
00:06:12,970 --> 00:06:17,180
By the way, if there's
no idiosyncratic risk

98
00:06:17,180 --> 00:06:23,900
there's nothing to test,
because the benchmark

99
00:06:23,900 --> 00:06:26,090
says controlling
for the aggregate is

100
00:06:26,090 --> 00:06:30,170
the idiosyncratic influencing
consumption or not.

101
00:06:30,170 --> 00:06:33,620
And if there's no idiosyncratic,
it's all aggregate,

102
00:06:33,620 --> 00:06:37,080
you can't do that test.

103
00:06:37,080 --> 00:06:39,880
This extends, of course,
beyond households

104
00:06:39,880 --> 00:06:43,900
within a village to tests
across villages and so on.

105
00:06:48,060 --> 00:06:51,500
Now, many people have the
idea that villages which

106
00:06:51,500 --> 00:06:55,910
they may never go
into are all alike,

107
00:06:55,910 --> 00:06:57,860
and it's all a function
of the rainfall,

108
00:06:57,860 --> 00:07:01,460
and it's all aggregate, and
that's just not true at all.

109
00:07:01,460 --> 00:07:07,880
And I'll show you the front end
of this paper, some statistics.

110
00:07:07,880 --> 00:07:10,130
Although it is true there
may be a lot of diversity

111
00:07:10,130 --> 00:07:13,580
in a village, if they all
completely diversified

112
00:07:13,580 --> 00:07:17,060
and did not specialize and they
all held a common portfolio,

113
00:07:17,060 --> 00:07:19,250
then, again, there
would be nothing

114
00:07:19,250 --> 00:07:24,130
to test as if they're doing it
all ex-ante and not ex-post,

115
00:07:24,130 --> 00:07:27,700
because the starting point would
be the income coming from all

116
00:07:27,700 --> 00:07:29,410
these diversified activities.

117
00:07:33,120 --> 00:07:34,530
But again, they don't do that.

118
00:07:43,250 --> 00:07:46,360
So let's take a look, these
so-called ICRISAT villages--

119
00:07:46,360 --> 00:07:49,090
Institute for Crops Research
in the Semi-Arid Tropic--

120
00:07:49,090 --> 00:07:52,990
at one point that was the source
of virtually the only panel

121
00:07:52,990 --> 00:08:00,030
data on developing
countries in the world.

122
00:08:00,030 --> 00:08:02,830
And it was gathered monthly,
and essentially for 10 years.

123
00:08:02,830 --> 00:08:05,860
They've started it up
again, interestingly.

124
00:08:05,860 --> 00:08:07,840
And there are other data
sets like that, where

125
00:08:07,840 --> 00:08:11,930
there are gaps of decades.

126
00:08:11,930 --> 00:08:15,670
So it's not true that
ICRISAT was the only source

127
00:08:15,670 --> 00:08:17,470
of panel variation.

128
00:08:17,470 --> 00:08:22,210
In Aurepalle, they're
growing castor and sorghum.

129
00:08:22,210 --> 00:08:25,120
Each crop is quite risky if
you look at the coefficient

130
00:08:25,120 --> 00:08:28,300
of variation.

131
00:08:28,300 --> 00:08:30,100
Those numbers are
comparable to something

132
00:08:30,100 --> 00:08:36,159
I did in medieval England,
where roughly every 12,

133
00:08:36,159 --> 00:08:40,080
13 years, they would
experience a famine.

134
00:08:40,080 --> 00:08:45,630
So these are high-risk
environments.

135
00:08:45,630 --> 00:08:50,320
Of course you could
diversify across these crops.

136
00:08:50,320 --> 00:08:51,730
That helps.

137
00:08:51,730 --> 00:08:54,520
Depending on the pair
of crops that you pick,

138
00:08:54,520 --> 00:08:57,910
the cross-crop correlations
could be as low as 0.09

139
00:08:57,910 --> 00:09:00,100
or something much higher.

140
00:09:00,100 --> 00:09:02,110
A low number is a good
thing, because you

141
00:09:02,110 --> 00:09:03,700
can spread your risk.

142
00:09:06,510 --> 00:09:09,420
Soil isn't uniform either.

143
00:09:09,420 --> 00:09:11,340
You can look, for a
given crop, at how

144
00:09:11,340 --> 00:09:16,990
the coefficient of variation
varies from one type of soil

145
00:09:16,990 --> 00:09:18,350
to another.

146
00:09:18,350 --> 00:09:21,580
And again, the cross-soil
correlation for a given crop

147
00:09:21,580 --> 00:09:25,780
is 0.37, and relatively low.

148
00:09:25,780 --> 00:09:27,880
Nevertheless, as I
said, it's not true

149
00:09:27,880 --> 00:09:31,120
that all households hold
a diversified basket

150
00:09:31,120 --> 00:09:32,540
of all these things.

151
00:09:32,540 --> 00:09:33,040
Yes.

152
00:09:33,040 --> 00:09:34,498
AUDIENCE: Are there any
crops that are negatively

153
00:09:34,498 --> 00:09:35,956
correlated with each other?

154
00:09:39,507 --> 00:09:41,090
ROBERT TOWNSEND: Not
in these numbers,

155
00:09:41,090 --> 00:09:43,190
although I certainly
wouldn't rule it out.

156
00:09:48,440 --> 00:09:50,750
Well, and some of those
numbers are point estimates,

157
00:09:50,750 --> 00:09:51,590
relatively low.

158
00:09:51,590 --> 00:09:55,935
And you put the standard
error bands, and yeah.

159
00:09:55,935 --> 00:09:56,920
OK.

160
00:09:56,920 --> 00:10:01,480
This is by cruder categories,
like labor income, income

161
00:10:01,480 --> 00:10:04,690
from trade and handicrafts,
animal husbandry.

162
00:10:04,690 --> 00:10:09,280
And it's stratified, as ICRISAT
did, by the amount of land

163
00:10:09,280 --> 00:10:10,240
that they hold--

164
00:10:10,240 --> 00:10:13,810
none, small, medium,
and large landholders.

165
00:10:13,810 --> 00:10:19,120
And these are sort of
basically fractions of income

166
00:10:19,120 --> 00:10:21,130
that they get from such things.

167
00:10:21,130 --> 00:10:26,710
Large landholders have the bulk
of their income from crops,

168
00:10:26,710 --> 00:10:29,860
and a substantial amount
from animal husbandry.

169
00:10:29,860 --> 00:10:33,910
People without land virtually
have no crop income.

170
00:10:33,910 --> 00:10:35,830
And the dominant
source is from labor.

171
00:10:35,830 --> 00:10:38,930
And there's everything
else in between,

172
00:10:38,930 --> 00:10:41,080
as you might anticipate.

173
00:10:41,080 --> 00:10:43,030
There were three
villages involved.

174
00:10:43,030 --> 00:10:46,480
That pattern is pretty common.

175
00:10:46,480 --> 00:10:49,270
And here, again, by
those income sources--

176
00:10:49,270 --> 00:10:50,860
(EXCITEDLY) hey, a
negative number--

177
00:10:54,790 --> 00:10:58,070
again, in the lower bound of
the stair of the confidence

178
00:10:58,070 --> 00:10:58,570
interval.

179
00:11:01,270 --> 00:11:04,180
Well, actually this is--

180
00:11:04,180 --> 00:11:08,980
even the point
estimates went negative.

181
00:11:08,980 --> 00:11:10,990
Sorry, no no.

182
00:11:10,990 --> 00:11:12,040
Yes, yes.

183
00:11:12,040 --> 00:11:15,880
This is hard to read because
these items should also

184
00:11:15,880 --> 00:11:18,020
be listed on the rows.

185
00:11:18,020 --> 00:11:21,730
So if you're going
down the diagonal,

186
00:11:21,730 --> 00:11:24,820
you're reading the
coefficient of variation,

187
00:11:24,820 --> 00:11:27,040
and then above and
below the diagonal,

188
00:11:27,040 --> 00:11:33,190
you're reading the
correlation of, say,

189
00:11:33,190 --> 00:11:37,975
in this case, livestock
income with profit income.

190
00:11:37,975 --> 00:11:40,600
It just throws you off, because
it should be written down here.

191
00:11:40,600 --> 00:11:42,763
And then to avoid
double counting,

192
00:11:42,763 --> 00:11:44,555
you don't fill in the
bottom of the matrix.

193
00:11:50,470 --> 00:11:53,470
So I've shown you
this picture before,

194
00:11:53,470 --> 00:12:01,458
the Rocky Mountain picture,
arguably the most famous graph

195
00:12:01,458 --> 00:12:02,250
I've ever produced.

196
00:12:04,980 --> 00:12:07,920
This is 10 years of data.

197
00:12:07,920 --> 00:12:10,260
And then the households
are lined up--

198
00:12:10,260 --> 00:12:13,500
which is why I keep showing it--

199
00:12:13,500 --> 00:12:15,270
the households are
lined up roughly

200
00:12:15,270 --> 00:12:17,800
by the size of landholdings.

201
00:12:17,800 --> 00:12:22,800
And you can see the levels
kind of going up as they

202
00:12:22,800 --> 00:12:26,250
have more and more land.

203
00:12:26,250 --> 00:12:30,270
And there's a 0 here.

204
00:12:30,270 --> 00:12:32,220
And so everything
on this diagram

205
00:12:32,220 --> 00:12:37,170
is deviation from the village
average at a point in time.

206
00:12:37,170 --> 00:12:39,900
Some people are below and
some people are above.

207
00:12:42,750 --> 00:12:45,240
You're seeing that
in the cross section,

208
00:12:45,240 --> 00:12:48,270
and you're seeing
that over time.

209
00:12:48,270 --> 00:12:50,100
The point of the diagram,
I've said before,

210
00:12:50,100 --> 00:12:52,800
is that some of the valleys
lie behind the peaks.

211
00:12:52,800 --> 00:12:56,130
And that means not everyone
is experiencing good years

212
00:12:56,130 --> 00:12:59,970
and bad years at the same time.

213
00:12:59,970 --> 00:13:06,420
It's not like undulating
waves of grain, as in Kansas.

214
00:13:06,420 --> 00:13:08,760
This is the consumption
picture that

215
00:13:08,760 --> 00:13:11,940
is on the same scale
as the income picture.

216
00:13:11,940 --> 00:13:14,130
If I wanted to get
bigger differences,

217
00:13:14,130 --> 00:13:17,690
I could rescale this graph.

218
00:13:17,690 --> 00:13:19,710
You know, it's all just
suppressed down there.

219
00:13:22,740 --> 00:13:26,160
But the statistical
work is really

220
00:13:26,160 --> 00:13:29,760
testing to see if there
are common fixed effects,

221
00:13:29,760 --> 00:13:34,090
and also whether this diagram
somehow, through the lens

222
00:13:34,090 --> 00:13:37,480
of the model, is related to
the income diagram previously.

223
00:13:38,835 --> 00:13:40,543
AUDIENCE: Have you
observed this in terms

224
00:13:40,543 --> 00:13:42,983
of like greater
weights on people?

225
00:13:42,983 --> 00:13:45,150
Because it looks like there's
not only risk-sharing,

226
00:13:45,150 --> 00:13:48,000
but also the rich guys are--

227
00:13:48,000 --> 00:13:49,350
if it's the same [INAUDIBLE].

228
00:13:49,350 --> 00:13:50,175
ROBERT TOWNSEND: Yeah, exactly.

229
00:13:50,175 --> 00:13:51,830
AUDIENCE: They're not
just getting the average.

230
00:13:51,830 --> 00:13:53,080
ROBERT TOWNSEND: That's right.

231
00:13:53,080 --> 00:13:55,530
So risk-sharing is
sometimes misinterpreted

232
00:13:55,530 --> 00:13:59,170
to mean complete pooling, and
everyone has an equal share.

233
00:13:59,170 --> 00:14:00,660
And that's not necessary.

234
00:14:00,660 --> 00:14:03,360
And when we look at the
benchmark regression,

235
00:14:03,360 --> 00:14:05,050
essentially it
has fixed effects.

236
00:14:05,050 --> 00:14:07,860
And those fixed effects
are almost exactly

237
00:14:07,860 --> 00:14:11,070
the Pareto weights.

238
00:14:11,070 --> 00:14:12,810
And then we can
test and see what

239
00:14:12,810 --> 00:14:14,340
those weights are related to.

240
00:14:20,230 --> 00:14:25,510
So here's the
optimization problem.

241
00:14:25,510 --> 00:14:30,010
You just basically-- there
is a one-to-one relationship

242
00:14:30,010 --> 00:14:35,230
between all the Pareto-optimal
allocations in this economy

243
00:14:35,230 --> 00:14:38,350
and solutions to this
programming problem.

244
00:14:42,236 --> 00:14:44,570
And that's asking about
the Pareto weights.

245
00:14:44,570 --> 00:14:47,720
Here they are-- lambda
sub-k, different

246
00:14:47,720 --> 00:14:50,000
for different households, k.

247
00:14:50,000 --> 00:14:53,120
The household is the
basic unit, although we

248
00:14:53,120 --> 00:14:54,860
will control for
the demographics

249
00:14:54,860 --> 00:14:56,700
within the household.

250
00:14:56,700 --> 00:15:00,720
So just to read
across this, it's

251
00:15:00,720 --> 00:15:04,470
a weighted average, lambda,
Pareto-weighted average

252
00:15:04,470 --> 00:15:09,060
of the discounted
expected utility of, say,

253
00:15:09,060 --> 00:15:11,610
in this case, household k.

254
00:15:11,610 --> 00:15:15,410
OK, so it's over time, which
could be finite or infinite.

255
00:15:15,410 --> 00:15:17,850
It's discounted
at a common rate.

256
00:15:17,850 --> 00:15:19,510
I'll come back to that.

257
00:15:19,510 --> 00:15:21,720
It's presumed everyone
has equal probability

258
00:15:21,720 --> 00:15:23,310
over these states of the world.

259
00:15:26,520 --> 00:15:32,070
And these states of
the world are not just

260
00:15:32,070 --> 00:15:36,200
the contemporaneous
realization of all the shocks.

261
00:15:36,200 --> 00:15:40,580
It's the entire history, from
the initial date out to that,

262
00:15:40,580 --> 00:15:42,350
like the branches of a tree.

263
00:15:45,420 --> 00:15:47,610
And then utility
contemporaneously

264
00:15:47,610 --> 00:15:54,550
depends on consumption and
these gender/age weights.

265
00:15:54,550 --> 00:16:00,480
So and you just want to--
whatever sums of consumption

266
00:16:00,480 --> 00:16:03,090
is, it's this average
and the aggregate.

267
00:16:03,090 --> 00:16:05,460
And then the idea is how to
distribute that aggregate

268
00:16:05,460 --> 00:16:07,365
to maximize this
subjective function.

269
00:16:09,970 --> 00:16:12,520
You can put labor in here.

270
00:16:12,520 --> 00:16:14,020
We'll have a whole
lecture on that.

271
00:16:16,470 --> 00:16:17,470
So what is the solution?

272
00:16:20,140 --> 00:16:24,280
Let's just pick consumption
of household k at date t

273
00:16:24,280 --> 00:16:26,800
is entering in here,
and it's entering

274
00:16:26,800 --> 00:16:28,030
in this resource constraint.

275
00:16:28,030 --> 00:16:30,850
This supplies, you know, for
the whole history of states

276
00:16:30,850 --> 00:16:34,300
up to date, t,
for every date, t.

277
00:16:34,300 --> 00:16:37,030
So we're going to
get derivative here,

278
00:16:37,030 --> 00:16:38,650
and a Lagrange multiplier here.

279
00:16:38,650 --> 00:16:40,700
This is common
resource constraints.

280
00:16:40,700 --> 00:16:43,660
So you pick up a common
Lagrange multiplier.

281
00:16:43,660 --> 00:16:44,946
Yes.

282
00:16:44,946 --> 00:16:48,390
AUDIENCE: Does the [? ht ?]
include the use of [INAUDIBLE]

283
00:16:48,390 --> 00:16:49,520
ROBERT TOWNSEND: Yes.

284
00:16:49,520 --> 00:16:50,020
Yep.

285
00:16:52,790 --> 00:16:57,040
A complete enumeration of
everything, and not just crops,

286
00:16:57,040 --> 00:17:00,700
but babies, and deaths, and the
whole demographic structure.

287
00:17:08,230 --> 00:17:10,869
So again, taking
that derivative,

288
00:17:10,869 --> 00:17:12,400
putting this thing
over here, you'd

289
00:17:12,400 --> 00:17:17,260
get a beta-discounted
probability, lambda k weighted

290
00:17:17,260 --> 00:17:18,280
marginal utility.

291
00:17:18,280 --> 00:17:20,410
And everyone's margin
utility is being

292
00:17:20,410 --> 00:17:23,680
equated to this common
Lagrange multiplier.

293
00:17:29,480 --> 00:17:36,020
You know, so if you assumed a
particular exponential utility

294
00:17:36,020 --> 00:17:40,190
function, for example,
then you can actually

295
00:17:40,190 --> 00:17:42,680
get sort of closed-form
analytic solutions

296
00:17:42,680 --> 00:17:44,820
for these first-order
conditions.

297
00:17:44,820 --> 00:17:47,240
So this just says, in
levels, that the consumption

298
00:17:47,240 --> 00:17:50,180
of household k depends
on the Pareto--

299
00:17:50,180 --> 00:17:53,510
the log Pareto weight
of household k.

300
00:17:53,510 --> 00:17:57,080
It has something to do with
the demographics and something

301
00:17:57,080 --> 00:18:00,260
to do with this common
shadow price of consumption,

302
00:18:00,260 --> 00:18:01,610
if I dare call it that.

303
00:18:06,350 --> 00:18:08,980
So here are the levels that
we were just talking about.

304
00:18:14,290 --> 00:18:17,260
This is too small to
read, but basically--

305
00:18:25,090 --> 00:18:27,850
another nice thing about
that exponential utility,

306
00:18:27,850 --> 00:18:29,740
if you're willing to
make strong assumptions

307
00:18:29,740 --> 00:18:33,190
about the composition
of the household,

308
00:18:33,190 --> 00:18:37,520
is it basically would
dictate how to weight people.

309
00:18:37,520 --> 00:18:43,890
And little babies don't eat
as much as 18-year-old males.

310
00:18:43,890 --> 00:18:50,110
And in turn, consumption is
dropping for the elderly.

311
00:18:50,110 --> 00:18:52,260
So we actually had
a dietary survey.

312
00:18:52,260 --> 00:18:55,240
And we assumed that
strong functional form,

313
00:18:55,240 --> 00:18:56,900
and we made all
those corrections.

314
00:18:56,900 --> 00:19:01,560
So you don't actually
see people eating.

315
00:19:01,560 --> 00:19:04,320
So you don't see their
individual consumptions.

316
00:19:04,320 --> 00:19:09,900
You see the consumption
for the whole household, k,

317
00:19:09,900 --> 00:19:12,840
but you put it in per
capita terms by dividing

318
00:19:12,840 --> 00:19:17,460
by the age/gender-weighted
number of people.

319
00:19:17,460 --> 00:19:20,520
Likewise, this
thing over here is

320
00:19:20,520 --> 00:19:26,000
the sum in the whole population
of per-capita consumption.

321
00:19:26,000 --> 00:19:30,570
And sort of reading
out loud, this term

322
00:19:30,570 --> 00:19:35,850
is actually the risk tolerance,
the inverse risk aversion,

323
00:19:35,850 --> 00:19:36,825
of household k.

324
00:19:41,580 --> 00:19:42,890
It's actually a j here.

325
00:19:42,890 --> 00:19:44,300
Sorry.

326
00:19:44,300 --> 00:19:46,730
Household j relative
to the sum of the risk

327
00:19:46,730 --> 00:19:51,200
tolerances of all
the other households.

328
00:19:51,200 --> 00:19:54,880
And I'll come back
to this often.

329
00:19:54,880 --> 00:19:59,140
If you assumed homogeneous risk
aversion, which is standard,

330
00:19:59,140 --> 00:20:02,660
then this ought to have
a coefficient of 1.

331
00:20:02,660 --> 00:20:05,740
So everybody's consumption
is co-moving, one to one,

332
00:20:05,740 --> 00:20:09,560
with the per capita average.

333
00:20:09,560 --> 00:20:11,600
If you have different
risk aversions,

334
00:20:11,600 --> 00:20:14,060
there is a very nice
interpretation here, which

335
00:20:14,060 --> 00:20:16,010
is this is aggregate risk.

336
00:20:16,010 --> 00:20:18,800
This is what's left over
after all the other smoothing

337
00:20:18,800 --> 00:20:20,900
that they're doing
intertemporally, blah, blah,

338
00:20:20,900 --> 00:20:24,470
blah, than what
they actually eat.

339
00:20:24,470 --> 00:20:27,320
So this is like the macro risk.

340
00:20:27,320 --> 00:20:32,000
And then the higher this
risk tolerance, the more they

341
00:20:32,000 --> 00:20:36,740
should be willing to bear
that, quote, "helping"

342
00:20:36,740 --> 00:20:39,610
out their village neighbors
who might be quite risk-averse.

343
00:20:42,920 --> 00:20:44,480
And we'll come back
to that in terms

344
00:20:44,480 --> 00:20:47,290
of measuring risk aversion.

345
00:20:47,290 --> 00:20:50,020
OK, so here's the standard
regression equation.

346
00:20:50,020 --> 00:20:54,500
If the level of consumption
depends on basically

347
00:20:54,500 --> 00:20:59,350
what is now a constant
household-specific term, which

348
00:20:59,350 --> 00:21:02,960
are those Pareto weights,
sum-- coefficients, time,

349
00:21:02,960 --> 00:21:06,380
village average
consumption-- at date t,

350
00:21:06,380 --> 00:21:08,840
adjustment for demographics,
and something else that

351
00:21:08,840 --> 00:21:12,420
ought to be zero, like income.

352
00:21:20,250 --> 00:21:23,490
You're going to have a devil
of a time running this.

353
00:21:23,490 --> 00:21:28,160
Even when you run one
household at a time ignoring

354
00:21:28,160 --> 00:21:32,360
all the others, even when
you use the leave out mean--

355
00:21:32,360 --> 00:21:35,060
so the household
is not included--

356
00:21:35,060 --> 00:21:36,950
it will turn out,
amazingly enough,

357
00:21:36,950 --> 00:21:42,260
that these coefficients,
one for each household, when

358
00:21:42,260 --> 00:21:47,860
you average them up, have
to take on the value of 1.

359
00:21:47,860 --> 00:21:53,230
So there are sort of
econometric issues

360
00:21:53,230 --> 00:21:57,160
here that you could
mistakenly think

361
00:21:57,160 --> 00:21:58,882
that you've got a
lot of co-movement,

362
00:21:58,882 --> 00:22:01,090
and it's just going to
happen-- you know, why is this

363
00:22:01,090 --> 00:22:02,620
happening, essentially?

364
00:22:02,620 --> 00:22:06,610
Well, it's more obvious if you
had the panel in front of us

365
00:22:06,610 --> 00:22:08,440
rather than one
household at a time.

366
00:22:08,440 --> 00:22:11,920
You're looking at household
consumption on the left,

367
00:22:11,920 --> 00:22:14,980
and the average of household
consumption on the right.

368
00:22:14,980 --> 00:22:18,020
Now take the average
of the whole thing,

369
00:22:18,020 --> 00:22:20,350
and the averages appearing
on the left and the right,

370
00:22:20,350 --> 00:22:22,210
and it ought to take
on a value of 1.

371
00:22:22,210 --> 00:22:25,630
So the average tendency
of the dependent variable

372
00:22:25,630 --> 00:22:28,570
is kind of like an intercept.

373
00:22:28,570 --> 00:22:30,430
So that's kind of the intuition.

374
00:22:30,430 --> 00:22:34,780
This also comes up in
many situations where you

375
00:22:34,780 --> 00:22:36,770
have this confounding problem.

376
00:22:36,770 --> 00:22:37,270
Yes.

377
00:22:37,270 --> 00:22:39,645
AUDIENCE: [INAUDIBLE] talked
about [? instead of their ?]

378
00:22:39,645 --> 00:22:40,690
[? compatibility? ?]

379
00:22:40,690 --> 00:22:42,357
ROBERT TOWNSEND: No,
because we're going

380
00:22:42,357 --> 00:22:44,110
to assume full commitment--

381
00:22:44,110 --> 00:22:45,408
instead of compatibility?

382
00:22:45,408 --> 00:22:46,450
Did I hear you correctly?

383
00:22:46,450 --> 00:22:47,033
AUDIENCE: Yes.

384
00:22:47,033 --> 00:22:48,550
ROBERT TOWNSEND: Yeah.

385
00:22:48,550 --> 00:22:52,630
We can put in
incentive constraints.

386
00:22:52,630 --> 00:22:54,880
And we will later.

387
00:22:54,880 --> 00:22:58,660
So we're going to build
on this basic framework,

388
00:22:58,660 --> 00:23:00,700
and get to all the
obstacles to trade,

389
00:23:00,700 --> 00:23:03,240
and how to estimate them.

390
00:23:03,240 --> 00:23:07,250
But you may think this thing
would be rejected out of hand.

391
00:23:07,250 --> 00:23:10,100
And it is statistically
rejected most of the time.

392
00:23:10,100 --> 00:23:13,520
But it turns out to be a
very, very useful benchmark.

393
00:23:18,830 --> 00:23:23,300
So this is the time series,
one household at a time.

394
00:23:23,300 --> 00:23:25,030
I won't say much
about this other

395
00:23:25,030 --> 00:23:27,260
than, does the data
have any power.

396
00:23:27,260 --> 00:23:31,070
The problem is, you test one
null, like full risk-sharing.

397
00:23:31,070 --> 00:23:33,500
And you can't reject it because
the confidence intervals

398
00:23:33,500 --> 00:23:34,310
are so large.

399
00:23:34,310 --> 00:23:37,700
So you want to test other
things like its autarky.

400
00:23:37,700 --> 00:23:40,940
And the weight of the
evidence is it's not autarky.

401
00:23:40,940 --> 00:23:43,265
We can pretty much
soundly reject--

402
00:23:48,858 --> 00:23:50,400
And risk-sharing is
doing quite well.

403
00:23:50,400 --> 00:23:52,300
And this is with a panel.

404
00:23:52,300 --> 00:23:54,920
And again, I'm going to
spare you all the details.

405
00:23:54,920 --> 00:23:56,940
But we did it three
different ways--

406
00:23:56,940 --> 00:24:02,790
standard, first
difference, Jerry Hausman--

407
00:24:02,790 --> 00:24:04,050
and estimated.

408
00:24:04,050 --> 00:24:06,510
Now, here we're
picking up coefficients

409
00:24:06,510 --> 00:24:09,960
on various sources of
income, none of which

410
00:24:09,960 --> 00:24:11,940
should be positive
and significant.

411
00:24:11,940 --> 00:24:14,670
But some of them kind of are.

412
00:24:14,670 --> 00:24:17,430
For example, the
highest number here

413
00:24:17,430 --> 00:24:21,100
is profits from trade and
handicrafts in Aurepalle.

414
00:24:21,100 --> 00:24:24,810
They're climbing palm
trees and tapping--

415
00:24:24,810 --> 00:24:27,270
and then making liquor, toddy.

416
00:24:30,540 --> 00:24:32,280
And it's not insured, evidently.

417
00:24:34,980 --> 00:24:40,860
Labor in general has a
high number pretty much

418
00:24:40,860 --> 00:24:42,160
in all the villages.

419
00:24:42,160 --> 00:24:46,680
So now we're kind of easing
into the policy thing, which

420
00:24:46,680 --> 00:24:49,080
is, if you wanted a
target, wouldn't you

421
00:24:49,080 --> 00:24:51,150
want to run this thing
first and see who's

422
00:24:51,150 --> 00:24:54,720
the most vulnerable,
and then sort of guide

423
00:24:54,720 --> 00:24:56,460
your interventions.

424
00:24:56,460 --> 00:24:58,990
Based on that, I wasn't
thinking along that line,

425
00:24:58,990 --> 00:25:01,530
I was just being
honest about where

426
00:25:01,530 --> 00:25:03,300
we're rejecting and so on.

427
00:25:07,790 --> 00:25:13,790
And here, finally, Matt, is
where we took this coefficient

428
00:25:13,790 --> 00:25:19,680
and regressed it
against other things

429
00:25:19,680 --> 00:25:24,840
like the area operated by
the household, the value

430
00:25:24,840 --> 00:25:27,990
of their plow animals,
their inheritance,

431
00:25:27,990 --> 00:25:29,250
and so on and so forth.

432
00:25:32,760 --> 00:25:34,530
And I'm not sure what
you would expect.

433
00:25:34,530 --> 00:25:38,220
It looks good at first
blush, in the sense

434
00:25:38,220 --> 00:25:41,910
that the amount of land and
the value of the animals

435
00:25:41,910 --> 00:25:44,172
is correlated with
those intercepts.

436
00:25:47,130 --> 00:25:51,240
The problem is things are
more turbulent than that

437
00:25:51,240 --> 00:25:52,290
at the village level.

438
00:25:52,290 --> 00:25:55,140
And people do sell, or
buy, or acquire land.

439
00:25:55,140 --> 00:25:57,990
And certainly those
plow animals aren't just

440
00:25:57,990 --> 00:26:01,650
locked in place year
after year after year.

441
00:26:01,650 --> 00:26:05,070
So that's kind of some
evidence of some kind

442
00:26:05,070 --> 00:26:10,740
of long-term impact that shows
up in terms of the assets

443
00:26:10,740 --> 00:26:12,570
they end up holding.

444
00:26:12,570 --> 00:26:14,460
And that's very much
along the lines of,

445
00:26:14,460 --> 00:26:17,820
is there a moral hazard
problem or some kind

446
00:26:17,820 --> 00:26:19,140
of incentive constraint?

447
00:26:19,140 --> 00:26:24,300
And how should wealth levels
behave if that were the case?

448
00:26:28,170 --> 00:26:31,970
The biggest puzzle in
these data, in some sense,

449
00:26:31,970 --> 00:26:38,670
is almost the opposite,
which is although we see them

450
00:26:38,670 --> 00:26:48,150
over 10 years, we see the
upper-caste, white-robed

451
00:26:48,150 --> 00:26:51,540
brahman-type guys in the
village just consistently eating

452
00:26:51,540 --> 00:26:55,270
far less than their
income, constantly.

453
00:26:55,270 --> 00:26:58,470
And so is this just a
social transfer mechanism,

454
00:26:58,470 --> 00:27:01,680
or are they actually
insuring themselves

455
00:27:01,680 --> 00:27:05,760
against future relabeling of--

456
00:27:05,760 --> 00:27:07,590
castes, by the way
do get relabeled,

457
00:27:07,590 --> 00:27:09,070
and histories get recreated.

458
00:27:09,070 --> 00:27:13,730
It it's a much more complicated
history than you might--

459
00:27:13,730 --> 00:27:17,130
at least that's what the
anthropologists tell me.

460
00:27:17,130 --> 00:27:25,970
OK, so that's risk and
insurance in village India.

461
00:27:30,020 --> 00:27:32,840
So let's take that framework
and start knocking off

462
00:27:32,840 --> 00:27:34,180
some policy questions.

463
00:27:34,180 --> 00:27:38,930
So first is about
the effectiveness

464
00:27:38,930 --> 00:27:40,310
of financial institutions.

465
00:27:40,310 --> 00:27:45,410
Now, let me just say, the
immediate policy implication

466
00:27:45,410 --> 00:27:50,870
coming out of village
India was basically,

467
00:27:50,870 --> 00:27:56,200
miraculously, somehow,
they're pooling all that risk.

468
00:27:56,200 --> 00:28:00,480
It's not a bunch of fragmented
households on their own,

469
00:28:00,480 --> 00:28:01,830
in autarky.

470
00:28:01,830 --> 00:28:04,380
It seems to be
more the opposite.

471
00:28:04,380 --> 00:28:08,010
So there was a huge
firestorm people

472
00:28:08,010 --> 00:28:11,790
had a hard time accepting
that something like that

473
00:28:11,790 --> 00:28:12,420
could happen.

474
00:28:15,450 --> 00:28:18,900
But villages, even if they
were amazingly good not,

475
00:28:18,900 --> 00:28:22,440
are not necessarily
interacting with each other.

476
00:28:22,440 --> 00:28:24,150
I'll come back to that.

477
00:28:24,150 --> 00:28:27,420
And there may be, as there
is in Thailand, transactions

478
00:28:27,420 --> 00:28:29,230
with outside financial
institutions.

479
00:28:29,230 --> 00:28:33,990
So the question is, do outside
formal financial institutions

480
00:28:33,990 --> 00:28:39,070
help, if not within the
village, then across village,

481
00:28:39,070 --> 00:28:41,350
in the mitigation of risk?

482
00:28:48,380 --> 00:28:51,610
So roughly speaking,
the treatment group

483
00:28:51,610 --> 00:28:54,640
is a set of households
who are customers

484
00:28:54,640 --> 00:28:56,710
of financial institutions.

485
00:28:56,710 --> 00:28:57,760
A little typo there.

486
00:29:00,790 --> 00:29:03,940
And that's measured-- their
use of the institution

487
00:29:03,940 --> 00:29:07,450
is actually measured in the
data, which is reassuring.

488
00:29:07,450 --> 00:29:13,060
And I'll tell you about the
instruments we use momentarily.

489
00:29:13,060 --> 00:29:15,580
So we're going to go back to
that consumption equation,

490
00:29:15,580 --> 00:29:21,100
and also derive a corresponding
investment equation,

491
00:29:21,100 --> 00:29:27,250
and then see whether being
a customer or a member

492
00:29:27,250 --> 00:29:29,260
lowers the vulnerability
in the sense

493
00:29:29,260 --> 00:29:31,940
of that coefficient
on idiosyncratic risk.

494
00:29:31,940 --> 00:29:32,440
Yes.

495
00:29:32,440 --> 00:29:36,050
AUDIENCE: So when we say
those with financial access

496
00:29:36,050 --> 00:29:38,290
are compared to those
without, so does

497
00:29:38,290 --> 00:29:40,420
that really mean
those with access

498
00:29:40,420 --> 00:29:43,187
to the formal financial
sector are compared without--

499
00:29:43,187 --> 00:29:44,770
ROBERT TOWNSEND:
That's what it means.

500
00:29:44,770 --> 00:29:46,600
AUDIENCE: Yeah, because
there's money-lending sectors

501
00:29:46,600 --> 00:29:46,820
et cetera.

502
00:29:46,820 --> 00:29:47,500
ROBERT TOWNSEND: Exactly.

503
00:29:47,500 --> 00:29:48,430
AUDIENCE: Yeah.

504
00:29:48,430 --> 00:29:52,350
ROBERT TOWNSEND: So yeah,
that's like one topic away.

505
00:29:52,350 --> 00:29:55,270
I'll come right back to that.

506
00:29:55,270 --> 00:29:58,240
But to answer your
question, it's

507
00:29:58,240 --> 00:30:01,270
going to turn out that
directly connected,

508
00:30:01,270 --> 00:30:05,020
at least in consumption,
is helpful, but indirectly

509
00:30:05,020 --> 00:30:07,810
connected, through
gifts and loans

510
00:30:07,810 --> 00:30:10,660
to someone in the village,
even though you're not,

511
00:30:10,660 --> 00:30:14,230
is as effective in
consumption-smoothing.

512
00:30:14,230 --> 00:30:20,350
So that's then consistent
with the previous paper,

513
00:30:20,350 --> 00:30:22,875
although it's not quite
so clean with investment.

514
00:30:30,620 --> 00:30:32,080
So what are these institutions?

515
00:30:32,080 --> 00:30:35,470
Well, by now you probably
know the names of them.

516
00:30:35,470 --> 00:30:39,640
They've been coming up
in previous lectures.

517
00:30:39,640 --> 00:30:42,660
We find that this
bank for agriculture

518
00:30:42,660 --> 00:30:45,460
and agricultural cooperatives
is helpful in smoothing

519
00:30:45,460 --> 00:30:46,840
consumption and investment.

520
00:30:46,840 --> 00:30:49,300
It looks, from the
transactions data,

521
00:30:49,300 --> 00:30:52,390
is if that has to do with
its credit operation.

522
00:30:52,390 --> 00:30:55,180
They're basically embedding
credit and insurance

523
00:30:55,180 --> 00:30:57,640
in their loan contracts.

524
00:30:57,640 --> 00:31:01,000
It's not, "you pay back
the loan or default."

525
00:31:01,000 --> 00:31:04,900
There's a whole
schedule for repayment.

526
00:31:04,900 --> 00:31:09,470
And it's adjusted by aggregate
local/regional shocks,

527
00:31:09,470 --> 00:31:10,990
and so on.

528
00:31:10,990 --> 00:31:15,260
It's quite sophisticated.

529
00:31:15,260 --> 00:31:18,260
In the financial
crisis, the World Bank

530
00:31:18,260 --> 00:31:22,190
went in there determined to
see this stereotype everywhere,

531
00:31:22,190 --> 00:31:26,330
since remember, finance
companies and commercial bank

532
00:31:26,330 --> 00:31:29,270
failures triggered the
Asia financial crisis.

533
00:31:32,070 --> 00:31:35,870
So they looked at the BAAC
history of over arrears,

534
00:31:35,870 --> 00:31:38,300
and concluded that it
was another disaster.

535
00:31:38,300 --> 00:31:41,870
But actually they were running
an insurance operation.

536
00:31:41,870 --> 00:31:44,140
And that's what shows up
in the consumption data.

537
00:31:46,780 --> 00:31:49,240
Commercial banks are
helping to smooth somewhat.

538
00:31:49,240 --> 00:31:53,770
It's largely, though, through
formal savings accounts.

539
00:31:53,770 --> 00:31:55,720
Loans are pretty
thin in the data.

540
00:32:00,380 --> 00:32:02,920
So what are we going to do?

541
00:32:02,920 --> 00:32:05,680
You've seen this model before.

542
00:32:05,680 --> 00:32:08,720
This is Greenwood and Javanovic.

543
00:32:08,720 --> 00:32:14,570
By the way, we're sort of in the
data micro part of the class.

544
00:32:14,570 --> 00:32:17,470
Don't forget about
all that macro.

545
00:32:17,470 --> 00:32:20,770
So the Greenwood-Jovanovic
model assumes

546
00:32:20,770 --> 00:32:23,620
that the benefit of being
in a financial institution

547
00:32:23,620 --> 00:32:28,390
is better information
and risk-sharing.

548
00:32:28,390 --> 00:32:31,120
So here we are
looking at the data

549
00:32:31,120 --> 00:32:33,730
to see if that's a
good assumption or not,

550
00:32:33,730 --> 00:32:36,130
actually institution
by institution.

551
00:32:36,130 --> 00:32:41,350
So there is a direct,
immediate link

552
00:32:41,350 --> 00:32:45,130
of this micro work with
those macro models.

553
00:32:45,130 --> 00:32:47,110
Or even if we reject,
then it should

554
00:32:47,110 --> 00:32:51,390
inform what we assume
within those macro models.

555
00:32:51,390 --> 00:32:54,100
So remember, in
Greenwood and Javanovic,

556
00:32:54,100 --> 00:32:58,390
there was a cost to joining
the financial system.

557
00:32:58,390 --> 00:33:02,770
We're going to let this be
a household-specific cost.

558
00:33:02,770 --> 00:33:04,720
And basically we're
going to want something

559
00:33:04,720 --> 00:33:10,420
like instruments, like
distance from the branch, which

560
00:33:10,420 --> 00:33:11,860
varies across households.

561
00:33:17,830 --> 00:33:20,920
Just a little bit on
the underlying model,

562
00:33:20,920 --> 00:33:23,620
just to remind you, because
you've actually seen it before.

563
00:33:23,620 --> 00:33:28,570
You've got output, which is a
function of the current capital

564
00:33:28,570 --> 00:33:36,400
stock, and then the sum of an
aggregate idiosyncratic shock,

565
00:33:36,400 --> 00:33:39,970
aggregate at the whatever
level, regional/national level,

566
00:33:39,970 --> 00:33:43,620
idiosyncratic at the
level of the households.

567
00:33:43,620 --> 00:33:48,150
As I said, we're going to
embed investment in this,

568
00:33:48,150 --> 00:33:51,240
adding that to the risk and
insurance in village India.

569
00:33:51,240 --> 00:33:54,000
So there's going to be some
adjustment costs to adjusting

570
00:33:54,000 --> 00:33:55,620
your capital stock.

571
00:33:55,620 --> 00:33:59,895
This is also buffeted
around by IID shocks.

572
00:34:03,400 --> 00:34:05,350
This is new.

573
00:34:05,350 --> 00:34:06,760
This is a preference shock.

574
00:34:06,760 --> 00:34:09,550
We're going to put something
in the utility function which

575
00:34:09,550 --> 00:34:10,960
is autocorrelated.

576
00:34:13,770 --> 00:34:16,360
I'm actually bending
over backwards

577
00:34:16,360 --> 00:34:19,650
to create a bias here.

578
00:34:19,650 --> 00:34:24,280
What we're going to want to
be able to do an instrument is

579
00:34:24,280 --> 00:34:27,639
to say that the
instrument has to do with,

580
00:34:27,639 --> 00:34:30,159
does influence the
selection of being

581
00:34:30,159 --> 00:34:32,889
a member of a
financial institution,

582
00:34:32,889 --> 00:34:38,500
and conditioned on that, does
not influence the outcome,

583
00:34:38,500 --> 00:34:39,699
in the future, at least.

584
00:34:39,699 --> 00:34:42,219
But you can imagine, with
autocorrelated preference

585
00:34:42,219 --> 00:34:44,500
shocks, that part
of the decision

586
00:34:44,500 --> 00:34:46,659
to join the financial
institution today

587
00:34:46,659 --> 00:34:50,170
has to do with that
urgency or patience.

588
00:34:50,170 --> 00:34:51,940
And if that carries
on into the future,

589
00:34:51,940 --> 00:34:54,695
it will also affect the
endogenous variables.

590
00:34:54,695 --> 00:34:56,320
So we want to make
it hard, and then we

591
00:34:56,320 --> 00:35:02,380
want to get rid of
the problem, to show

592
00:35:02,380 --> 00:35:05,140
that we have the ability
to get rid of the problem.

593
00:35:05,140 --> 00:35:08,200
You may remember how
this works, though.

594
00:35:08,200 --> 00:35:12,400
You create a value function
for being a participant

595
00:35:12,400 --> 00:35:14,560
in the financial system.

596
00:35:14,560 --> 00:35:18,650
If you join, you subtract
off that fixed cost.

597
00:35:18,650 --> 00:35:24,570
And this is just the array
of the other variables.

598
00:35:24,570 --> 00:35:27,150
You're joining after the
realization of the adjustment

599
00:35:27,150 --> 00:35:30,570
cost shock, the idiosyncratic
and aggregate shock,

600
00:35:30,570 --> 00:35:32,070
and the preference shock.

601
00:35:32,070 --> 00:35:34,260
You could choose not to join.

602
00:35:34,260 --> 00:35:36,330
You'd have another
value function.

603
00:35:36,330 --> 00:35:41,300
I think we even used V and
W in the earlier lecture.

604
00:35:41,300 --> 00:35:44,490
So but you don't
subtract off the cost

605
00:35:44,490 --> 00:35:46,380
from the capital stock.

606
00:35:46,380 --> 00:35:50,370
If you saw the autarky problem,
then you've kind of already

607
00:35:50,370 --> 00:35:52,830
seen the solution to
the consumption part

608
00:35:52,830 --> 00:35:56,460
of the intermediated problem.

609
00:35:56,460 --> 00:36:02,670
You can just sort of list
the equations for p equal 1,

610
00:36:02,670 --> 00:36:04,320
participant.

611
00:36:04,320 --> 00:36:07,170
They've got the
V value function.

612
00:36:07,170 --> 00:36:12,390
You've got consumption as a
function of your preference

613
00:36:12,390 --> 00:36:14,820
shock, and aggregate
consumption, and your Pareto

614
00:36:14,820 --> 00:36:16,710
weight.

615
00:36:16,710 --> 00:36:19,450
By the way, you haven't
seen this today,

616
00:36:19,450 --> 00:36:23,160
but investment is also a
function of the adjustment

617
00:36:23,160 --> 00:36:29,520
cost, the capital stock, and
the same aggregate consumption.

618
00:36:29,520 --> 00:36:32,920
Because it's the same
resource constraint.

619
00:36:32,920 --> 00:36:35,340
So the price of consumption
today, actually,

620
00:36:35,340 --> 00:36:38,550
versus tomorrow, is
not only influencing

621
00:36:38,550 --> 00:36:44,040
consumption, it's influencing
investment intertemporally.

622
00:36:47,040 --> 00:36:53,730
And we'll come back
to that and repeat it

623
00:36:53,730 --> 00:36:55,750
in the lecture on Thursday.

624
00:36:55,750 --> 00:36:59,280
And this is just the
autarky equation.

625
00:36:59,280 --> 00:37:01,410
You don't see aggregate
consumption anywhere.

626
00:37:01,410 --> 00:37:03,930
People are on their
own, and so on.

627
00:37:03,930 --> 00:37:05,910
There's no way you
could see and totally

628
00:37:05,910 --> 00:37:10,410
understand all the individual
variables at this level.

629
00:37:10,410 --> 00:37:12,060
I'm going too fast for that.

630
00:37:12,060 --> 00:37:16,940
But I just want you to think
about what you see in the data.

631
00:37:16,940 --> 00:37:21,460
You're going to see kind
of a weighted average

632
00:37:21,460 --> 00:37:26,350
of these things, weighted
by participation.

633
00:37:26,350 --> 00:37:28,990
So actually you're
either in this branch

634
00:37:28,990 --> 00:37:31,610
or you're in this one.

635
00:37:31,610 --> 00:37:33,485
And being an instrument
for these Ps.

636
00:37:36,220 --> 00:37:39,700
But anyway, it's convenient
to write it down this way,

637
00:37:39,700 --> 00:37:44,200
and derive the sort of
econometric equations

638
00:37:44,200 --> 00:37:45,850
you're going to
take to the data.

639
00:37:51,530 --> 00:37:53,050
I actually take
first differences,

640
00:37:53,050 --> 00:37:55,600
but the same principle applies.

641
00:37:55,600 --> 00:38:00,400
These instruments are the
history of the person's

642
00:38:00,400 --> 00:38:01,660
institutional use.

643
00:38:01,660 --> 00:38:04,300
Like predetermined
variables is an indicator

644
00:38:04,300 --> 00:38:11,500
for their current participation
at the village level,

645
00:38:11,500 --> 00:38:14,020
actually, not at
the household level.

646
00:38:14,020 --> 00:38:16,822
The time to the district
center is the most compelling.

647
00:38:16,822 --> 00:38:19,030
That's like how long does
it take to get to the bank.

648
00:38:22,220 --> 00:38:26,080
And then we have some
GIS measure of access

649
00:38:26,080 --> 00:38:29,260
in the local area, which is
supposed to pick up, like,

650
00:38:29,260 --> 00:38:33,340
their active credit
officers in one spot

651
00:38:33,340 --> 00:38:38,005
and some not-so-active credit
officers in other areas.

652
00:38:42,720 --> 00:38:53,580
And then the data you know,
this was from May of 1997,

653
00:38:53,580 --> 00:39:02,860
in this case, when we wrote
this paper, through 2001.

654
00:39:02,860 --> 00:39:06,370
And I already gave you
the conclusions basically.

655
00:39:13,062 --> 00:39:17,060
And in fact you're going
to see the equations again

656
00:39:17,060 --> 00:39:19,790
when we get to this
informal, because it's

657
00:39:19,790 --> 00:39:24,320
very similar-looking
no point in repeating.

658
00:39:24,320 --> 00:39:33,650
So this is what I've been doing
with Cynthia and [? Whit. ?]

659
00:39:33,650 --> 00:39:37,580
So we have the measured chains
of financial transactions,

660
00:39:37,580 --> 00:39:40,565
gifts and loans, from
one family to another.

661
00:39:48,380 --> 00:39:50,810
So we've already
talked about the fact

662
00:39:50,810 --> 00:39:52,970
that people with
indirect connections who

663
00:39:52,970 --> 00:39:55,490
are not directly
customers but are

664
00:39:55,490 --> 00:39:57,890
connected through
other households

665
00:39:57,890 --> 00:40:00,050
are going to do quite well.

666
00:40:00,050 --> 00:40:02,660
There's a flip side
to that, is we've been

667
00:40:02,660 --> 00:40:07,060
measuring the control group.

668
00:40:07,060 --> 00:40:11,100
The control group was
an average of people

669
00:40:11,100 --> 00:40:15,030
who were indirect previously,
indirectly connected or not

670
00:40:15,030 --> 00:40:16,030
connected at all.

671
00:40:16,030 --> 00:40:18,150
And when we separate
that out, we

672
00:40:18,150 --> 00:40:20,340
have the "not connected
at all" people.

673
00:40:20,340 --> 00:40:24,430
And it turns out they're
actually the most vulnerable.

674
00:40:24,430 --> 00:40:30,300
So their vulnerability
was being concealed.

675
00:40:30,300 --> 00:40:35,010
And there's some work in Mexico
that is very, very similar

676
00:40:35,010 --> 00:40:35,850
in spirit.

677
00:40:35,850 --> 00:40:37,890
The data is a bit
different, so the details

678
00:40:37,890 --> 00:40:43,200
are a bit different, Andgelucci,
and Rangel, Rasul, and so on.

679
00:40:48,420 --> 00:40:51,090
Investment's going to be a
bit of a different story.

680
00:40:51,090 --> 00:40:53,130
Kin are going to
matter, but it's almost

681
00:40:53,130 --> 00:40:54,735
like an
out-of-equilibrium story.

682
00:40:59,175 --> 00:41:01,650
Let me come back to that.

683
00:41:01,650 --> 00:41:03,540
I won't dwell on this slide.

684
00:41:03,540 --> 00:41:06,570
It's just showing you that
there is a lot of borrowing,

685
00:41:06,570 --> 00:41:09,720
and lending, and gift-giving.

686
00:41:09,720 --> 00:41:15,580
And when it happens,
it happens typically,

687
00:41:15,580 --> 00:41:18,440
you know, depending
on which line

688
00:41:18,440 --> 00:41:25,550
you're looking at, three to
five times during a sample.

689
00:41:25,550 --> 00:41:26,750
And the amounts are large.

690
00:41:26,750 --> 00:41:31,320
The borrowing transaction is
like 60% your monthly income.

691
00:41:31,320 --> 00:41:36,340
So you know, take your
student stipend and--

692
00:41:36,340 --> 00:41:38,190
it's a big number.

693
00:41:38,190 --> 00:41:39,610
No, I'm not saying your stipend.

694
00:41:39,610 --> 00:41:40,600
It's a big percent.

695
00:41:40,600 --> 00:41:44,450
AUDIENCE: [CHUCKLES]
We would never

696
00:41:44,450 --> 00:41:46,508
make that misunderstanding
[INAUDIBLE]..

697
00:41:46,508 --> 00:41:53,320
ROBERT TOWNSEND:
[CHUCKLES] So this

698
00:41:53,320 --> 00:41:55,930
is the equation I was alluding
to that I didn't show you

699
00:41:55,930 --> 00:41:59,280
on the previous paper with Alem.

700
00:41:59,280 --> 00:42:01,450
But basically,
first-differenced, we

701
00:42:01,450 --> 00:42:04,270
have the change in
household consumption

702
00:42:04,270 --> 00:42:06,520
in village v at date t.

703
00:42:06,520 --> 00:42:08,990
This is the change in
the income of household

704
00:42:08,990 --> 00:42:10,670
i in village v at date t.

705
00:42:10,670 --> 00:42:13,390
This is the
vulnerability coefficient

706
00:42:13,390 --> 00:42:18,010
that we're interested in
seeing how big or small it is.

707
00:42:18,010 --> 00:42:25,720
And then we start interacting
that change in income with

708
00:42:25,720 --> 00:42:32,200
basically a direct member or
"indirectly-connected," quote,

709
00:42:32,200 --> 00:42:35,060
member of a financial
institution.

710
00:42:38,410 --> 00:42:43,720
And one hopes that this
might be not trivial,

711
00:42:43,720 --> 00:42:49,550
but this would be
negative and subtract off

712
00:42:49,550 --> 00:42:52,730
for those who are members,
and customers, and so on.

713
00:42:52,730 --> 00:42:54,800
K is kinship.

714
00:42:54,800 --> 00:42:56,510
That's also entering--
we're trying

715
00:42:56,510 --> 00:42:58,790
to distinguish sort
of having family

716
00:42:58,790 --> 00:43:01,880
in the village as opposed to
having measured transactions.

717
00:43:01,880 --> 00:43:02,870
It's a bit heroic.

718
00:43:02,870 --> 00:43:06,470
We don't have instruments
for whether you're

719
00:43:06,470 --> 00:43:07,610
in a network or not.

720
00:43:07,610 --> 00:43:13,850
It's a problem for us and
for a lot of this literature.

721
00:43:13,850 --> 00:43:17,330
And then this is
controlling for net worth.

722
00:43:17,330 --> 00:43:21,950
I mean, as you've seen from the
previous slides, the relatively

723
00:43:21,950 --> 00:43:26,180
poor people, at least in
India, were more vulnerable.

724
00:43:26,180 --> 00:43:28,010
So knowing that
a priori, we want

725
00:43:28,010 --> 00:43:32,750
to take that effect out and
control for it, at least

726
00:43:32,750 --> 00:43:36,070
in this additive way.

727
00:43:36,070 --> 00:43:38,215
This is the investment equation.

728
00:43:41,420 --> 00:43:43,090
It's basically very similar.

729
00:43:43,090 --> 00:43:44,690
I is just investment.

730
00:43:44,690 --> 00:43:47,140
It's actually normalized
by the capital stock.

731
00:43:47,140 --> 00:43:50,020
So it's investment
per unit capital.

732
00:43:50,020 --> 00:43:52,300
And again, it allows
for the direct effect

733
00:43:52,300 --> 00:43:55,150
of the income per
unit of capital,

734
00:43:55,150 --> 00:43:59,860
as well as the direct
and indirect connections.

735
00:44:03,010 --> 00:44:07,420
I should have highlighted that,
in both of these equations,

736
00:44:07,420 --> 00:44:14,180
you've got these time-varying
calm and fixed effects.

737
00:44:14,180 --> 00:44:16,700
This is what the theory
is telling us to do.

738
00:44:16,700 --> 00:44:19,990
It's not as though consumption
doesn't move with income.

739
00:44:19,990 --> 00:44:23,620
Consumption can move
with this fixed effect.

740
00:44:23,620 --> 00:44:25,570
I should have also
said, previously,

741
00:44:25,570 --> 00:44:28,720
with my mention of the
econometric problems

742
00:44:28,720 --> 00:44:30,760
of measuring average
consumption and putting it

743
00:44:30,760 --> 00:44:32,740
on the right-hand
side, that you can just

744
00:44:32,740 --> 00:44:36,010
replace that with an
unobserved fixed effect.

745
00:44:36,010 --> 00:44:39,310
And the theory gives you the
interpretation of what it is.

746
00:44:39,310 --> 00:44:44,360
Or you could subtract it all
off from the left-hand side.

747
00:44:44,360 --> 00:44:46,930
And when the
coefficient is 1, that's

748
00:44:46,930 --> 00:44:48,310
also a reasonable thing to do.

749
00:44:48,310 --> 00:44:51,355
Here we're just putting in these
time-varying fixed effects.

750
00:44:59,260 --> 00:45:02,380
So this is the coefficient
for those people

751
00:45:02,380 --> 00:45:05,140
without any access,
direct or indirect.

752
00:45:07,720 --> 00:45:11,220
This is how much you'd subtract
if you look at direct bank--

753
00:45:11,220 --> 00:45:14,880
yes, these are very
similar numbers.

754
00:45:14,880 --> 00:45:18,610
I always get accused
of biasing this stuff.

755
00:45:18,610 --> 00:45:21,700
Anyway, so the sum
of these things

756
00:45:21,700 --> 00:45:23,920
is significantly
different from zero.

757
00:45:23,920 --> 00:45:30,190
And the point is you've
achieved the benchmark standard.

758
00:45:30,190 --> 00:45:33,280
But equally interesting
is, when you

759
00:45:33,280 --> 00:45:35,680
have the indirect connections,
you're subtracting off

760
00:45:35,680 --> 00:45:38,430
virtually the same number.

761
00:45:38,430 --> 00:45:40,050
Kin didn't help in this case.

762
00:45:43,954 --> 00:45:45,906
AUDIENCE: Sorry, is
there a way to augment--

763
00:45:45,906 --> 00:45:47,281
because we have
direct connection

764
00:45:47,281 --> 00:45:48,590
and indirect connection.

765
00:45:48,590 --> 00:45:52,027
Would it help me
more to have both?

766
00:45:52,027 --> 00:45:53,860
ROBERT TOWNSEND: That's
what indirect means.

767
00:45:53,860 --> 00:45:54,240
Sorry.

768
00:45:54,240 --> 00:45:54,580
AUDIENCE: Oh, OK.

769
00:45:54,580 --> 00:45:54,920
All right.

770
00:45:54,920 --> 00:45:55,740
ROBERT TOWNSEND: Yeah, yeah.

771
00:45:55,740 --> 00:45:56,580
It's mislabeled.

772
00:45:56,580 --> 00:45:57,210
It means both.

773
00:45:57,210 --> 00:45:58,210
AUDIENCE: Oh, all right.

774
00:45:58,210 --> 00:46:00,430
ROBERT TOWNSEND:
Yeah, good point.

775
00:46:00,430 --> 00:46:03,960
And then this is
with investment.

776
00:46:03,960 --> 00:46:05,250
Now, what are the differences?

777
00:46:05,250 --> 00:46:06,630
Well, two, basically.

778
00:46:06,630 --> 00:46:11,190
First of all, whatever
we're subtracting off,

779
00:46:11,190 --> 00:46:14,310
it doesn't lower
the total to zero.

780
00:46:14,310 --> 00:46:17,520
And if you tested, there's
still residual vulnerability.

781
00:46:17,520 --> 00:46:21,900
So with investment, smoothing
investment against cash flow,

782
00:46:21,900 --> 00:46:25,350
you are left with some
idiosyncratic risk.

783
00:46:25,350 --> 00:46:32,180
Although kin are quite
helpful, but measured

784
00:46:32,180 --> 00:46:34,526
transactions are not.

785
00:46:34,526 --> 00:46:36,880
It's a little too early
to get into this story.

786
00:46:39,520 --> 00:46:41,710
Cynthia and I are
still working on it.

787
00:46:41,710 --> 00:46:45,580
But the idea is
something like, just

788
00:46:45,580 --> 00:46:50,080
to anticipate when we get
to obstacles to trade,

789
00:46:50,080 --> 00:46:52,900
you can't force someone to
be in a risk-sharing group,

790
00:46:52,900 --> 00:46:54,340
but they kind of
like being there

791
00:46:54,340 --> 00:46:57,500
because they can smooth
their future income.

792
00:46:57,500 --> 00:47:00,310
So you can't force them,
but they voluntarily

793
00:47:00,310 --> 00:47:02,440
choose to stay even
though, in any period,

794
00:47:02,440 --> 00:47:04,180
they have the option to leave.

795
00:47:04,180 --> 00:47:07,800
It's, like, self-sustaining.

796
00:47:07,800 --> 00:47:14,060
With investment-- and this
is above median investment

797
00:47:14,060 --> 00:47:17,680
relative to the capital stock--

798
00:47:17,680 --> 00:47:19,660
investment is big.

799
00:47:19,660 --> 00:47:24,190
The typical number is like
20% of your physical capital

800
00:47:24,190 --> 00:47:26,140
when it happens.

801
00:47:26,140 --> 00:47:29,340
So it's not little,
incremental, teeny-weeny stuff.

802
00:47:29,340 --> 00:47:34,240
OK, now imagine, say,
someone comes to me

803
00:47:34,240 --> 00:47:35,920
and asks me for a loan--

804
00:47:35,920 --> 00:47:40,020
I'm the community head, in
case you didn't notice--

805
00:47:40,020 --> 00:47:45,850
and you're going to
put it in your project,

806
00:47:45,850 --> 00:47:47,770
and it's quite a good project.

807
00:47:47,770 --> 00:47:50,500
And then, next period,
you got a lot more income

808
00:47:50,500 --> 00:47:54,760
than you used to, now it's
much more tempting to renege.

809
00:47:54,760 --> 00:47:57,910
Because you're much better
off in autarky than you

810
00:47:57,910 --> 00:48:02,180
would have been if you hadn't
gotten the investment done.

811
00:48:02,180 --> 00:48:09,310
And so the theory is, A, that
either doesn't happen, or B,

812
00:48:09,310 --> 00:48:13,180
all your relatives come
and do their thing.

813
00:48:16,580 --> 00:48:18,580
And that's consistent
with these patterns

814
00:48:18,580 --> 00:48:22,000
when we're working on
numerical estimates.

815
00:48:22,000 --> 00:48:23,042
Yes.

816
00:48:23,042 --> 00:48:25,600
AUDIENCE: I'm not sure if you
can see this from this data,

817
00:48:25,600 --> 00:48:29,910
but if I'm talking indirectly,
accessing the financial system,

818
00:48:29,910 --> 00:48:33,120
is that link between us,
does that person typically

819
00:48:33,120 --> 00:48:35,410
have assets on stock,
perhaps, because they

820
00:48:35,410 --> 00:48:38,478
know people will borrow from
them, and they want to do it?

821
00:48:38,478 --> 00:48:40,520
Or are they actually just
turning around and then

822
00:48:40,520 --> 00:48:42,680
borrowing the
pharmacist, and then

823
00:48:42,680 --> 00:48:44,915
being more like an
intermediary for a loan.

824
00:48:44,915 --> 00:48:47,290
ROBERT TOWNSEND: I'm not sure
exactly what you're asking.

825
00:48:47,290 --> 00:48:50,770
[? Whit ?] and I have this--

826
00:48:50,770 --> 00:48:53,830
somebody will take a
formal-sector loan.

827
00:48:53,830 --> 00:48:54,850
Then it's due.

828
00:48:54,850 --> 00:48:57,290
And they're not doing
too well in that year.

829
00:48:57,290 --> 00:49:00,340
So they will turn around and
get an informal short-term loan

830
00:49:00,340 --> 00:49:03,100
from a friend, or
relative, or money lender,

831
00:49:03,100 --> 00:49:04,780
pay off the long-term
loan, and then

832
00:49:04,780 --> 00:49:06,730
with the proceeds of
the long-term loan, they

833
00:49:06,730 --> 00:49:08,240
pay off the short-term loan.

834
00:49:08,240 --> 00:49:10,350
AUDIENCE: And that friend
or relative, should I--

835
00:49:10,350 --> 00:49:12,892
maybe the answer is it's both--
should I think of that person

836
00:49:12,892 --> 00:49:16,521
as having money lying
around to be ready to lend

837
00:49:16,521 --> 00:49:21,070
to their friend who needs
it, or did they intend to--

838
00:49:21,070 --> 00:49:22,650
AUDIENCE: They
can also take it--

839
00:49:22,650 --> 00:49:23,975
there could be a lot of shame.

840
00:49:23,975 --> 00:49:26,350
And they could also get it
from the financial institution

841
00:49:26,350 --> 00:49:28,160
themself if they're
a normal guy.

842
00:49:28,160 --> 00:49:30,660
But there is some people who
are acting as money lender,

843
00:49:30,660 --> 00:49:32,860
and they have money
ready [INAUDIBLE]..

844
00:49:32,860 --> 00:49:34,298
So it depends on the person.

845
00:49:34,298 --> 00:49:36,340
If it's a relative, mostly
they're not that rich.

846
00:49:36,340 --> 00:49:38,362
So they'd probably borrow.

847
00:49:38,362 --> 00:49:39,820
AUDIENCE: I guess
I'm just speaking

848
00:49:39,820 --> 00:49:42,540
of one of the things the
portfolios and the [INAUDIBLE]

849
00:49:42,540 --> 00:49:45,160
book, is that people will
simultaneously borrow and save

850
00:49:45,160 --> 00:49:46,380
at the same time.

851
00:49:46,380 --> 00:49:49,785
And maybe this fits in,
if part of the reason

852
00:49:49,785 --> 00:49:52,970
I have that savings is because
I might need to lend it out

853
00:49:52,970 --> 00:49:55,050
at short notice to somebody.

854
00:49:55,050 --> 00:49:56,200
ROBERT TOWNSEND: Yeah.

855
00:49:56,200 --> 00:49:58,330
So there are all kinds
of financial strategies.

856
00:49:58,330 --> 00:50:01,420
You know, savings and borrowing
gets pretty complicated.

857
00:50:01,420 --> 00:50:03,370
Because even if you get
a loan, then the way

858
00:50:03,370 --> 00:50:05,110
you get the loan
proceeds is someone

859
00:50:05,110 --> 00:50:06,460
opens up a savings account.

860
00:50:06,460 --> 00:50:09,460
You don't necessarily
withdraw it right away.

861
00:50:09,460 --> 00:50:12,400
So that's one mechanical
reason to get that.

862
00:50:12,400 --> 00:50:15,520
But loans can be long term.

863
00:50:15,520 --> 00:50:17,800
And then you have a
liquidity problem.

864
00:50:17,800 --> 00:50:19,330
And you don't call in the loan.

865
00:50:19,330 --> 00:50:25,510
So you might want to keep
savings to balance that risk.

866
00:50:25,510 --> 00:50:27,250
I mean, this is like
the bread and butter

867
00:50:27,250 --> 00:50:28,953
of a lot of things
people do in macro

868
00:50:28,953 --> 00:50:30,370
when they're looking
at investment

869
00:50:30,370 --> 00:50:33,140
and trying to sort all that out.

870
00:50:37,280 --> 00:50:37,780
Yep.

871
00:50:37,780 --> 00:50:41,011
AUDIENCE: Why [INAUDIBLE]
smallest investment?

872
00:50:44,840 --> 00:50:47,090
ROBERT TOWNSEND: They want
to undertake the investment

873
00:50:47,090 --> 00:50:49,340
when they have an opportunity.

874
00:50:49,340 --> 00:50:52,010
And they may not
be able to do it,

875
00:50:52,010 --> 00:50:53,720
or they may have
to wait to do it

876
00:50:53,720 --> 00:50:56,720
until they have the liquidity.

877
00:50:56,720 --> 00:50:58,000
And that's costly.

878
00:50:58,000 --> 00:51:00,510
AUDIENCE: But [INAUDIBLE]
differentiate [INAUDIBLE]

879
00:51:00,510 --> 00:51:02,350
tp avoid two cases.

880
00:51:02,350 --> 00:51:05,930
The first one is that there
is insufficient insurance,

881
00:51:05,930 --> 00:51:07,760
so I cannot borrow
money from the bank.

882
00:51:07,760 --> 00:51:11,090
The second one is that I
[? just ?] [? want ?] to invest

883
00:51:11,090 --> 00:51:15,100
because productivity
[INAUDIBLE]..

884
00:51:15,100 --> 00:51:19,460
ROBERT TOWNSEND: Yeah, so we'll
get to this structural model

885
00:51:19,460 --> 00:51:25,430
of village funds, which takes
a stand on the process that

886
00:51:25,430 --> 00:51:27,110
generate shocks.

887
00:51:27,110 --> 00:51:29,000
And then we have to
take the data, and kind

888
00:51:29,000 --> 00:51:31,430
of estimate that process.

889
00:51:31,430 --> 00:51:37,590
And then that model allows the--

890
00:51:37,590 --> 00:51:39,890
and then you get these
investment opportunities

891
00:51:39,890 --> 00:51:44,430
which are like random variables,
large, chunky opportunities.

892
00:51:44,430 --> 00:51:49,610
AUDIENCE: Here we assume that
we don't have to differentiate

893
00:51:49,610 --> 00:51:51,570
[INAUDIBLE]

894
00:51:51,570 --> 00:51:53,400
ROBERT TOWNSEND: This
is reduced form--

895
00:51:53,400 --> 00:51:56,743
I mean, here, I was talking
about a model just then.

896
00:51:56,743 --> 00:51:59,160
But we're not quite ready in
the class, because we haven't

897
00:51:59,160 --> 00:52:02,820
really explicitly written
down these obstacles,

898
00:52:02,820 --> 00:52:06,200
neither the moral hazard
or the limited commitment.

899
00:52:06,200 --> 00:52:09,530
So I was talking you
through it like what next.

900
00:52:09,530 --> 00:52:11,870
But actually this is
just reduced-form stuff.

901
00:52:11,870 --> 00:52:13,595
It doesn't really tell you.

902
00:52:13,595 --> 00:52:18,380
But we're telling a story which
I think is reasonably accurate.

903
00:52:18,380 --> 00:52:20,270
But Cynthia and I are
writing down a model,

904
00:52:20,270 --> 00:52:22,280
and trying to
numerically compute

905
00:52:22,280 --> 00:52:26,630
it to see if it's
a reasonable story.

906
00:52:31,600 --> 00:52:34,690
This is what I said in words.

907
00:52:34,690 --> 00:52:36,460
And then this is
the Mexican paper.

908
00:52:39,430 --> 00:52:42,580
So it has to do with Progresa.

909
00:52:42,580 --> 00:52:46,010
And they have
treatment and control.

910
00:52:46,010 --> 00:52:49,550
And the details are
a bit different.

911
00:52:49,550 --> 00:52:53,170
But I think you
have these slides.

912
00:52:53,170 --> 00:52:56,450
And there's only like
four or five of them.

913
00:52:56,450 --> 00:53:02,670
But I will just say, then,
it's not just about Thailand,

914
00:53:02,670 --> 00:53:04,890
and we're not the
only one doing this.

915
00:53:04,890 --> 00:53:06,770
Networks is a really hot topic.

916
00:53:06,770 --> 00:53:10,090
Everybody wants to play
around with networks.

917
00:53:10,090 --> 00:53:14,760
Esther and [? Arun ?] have
pretty good network data

918
00:53:14,760 --> 00:53:15,600
in parts of India.

919
00:53:21,870 --> 00:53:25,950
So one example
probably is enough.

920
00:53:25,950 --> 00:53:30,390
Now, I said something about
well-intended interventions

921
00:53:30,390 --> 00:53:32,580
causing losses.

922
00:53:32,580 --> 00:53:43,160
This is some joint
work We first of all do

923
00:53:43,160 --> 00:53:46,020
estimate heterogeneity
and risk preferences.

924
00:53:49,990 --> 00:53:53,240
And then the idea, splicing
together these papers,

925
00:53:53,240 --> 00:53:55,340
it's like village India.

926
00:53:55,340 --> 00:53:58,040
So there's aggregate
village level risk that's

927
00:53:58,040 --> 00:54:00,920
not necessarily shared very--

928
00:54:00,920 --> 00:54:04,160
which you have to control for,
and it's not "shared," quote,

929
00:54:04,160 --> 00:54:06,650
unquote--

930
00:54:06,650 --> 00:54:08,810
let me say it again.

931
00:54:08,810 --> 00:54:10,790
You've got all the
idiosyncratic stuff

932
00:54:10,790 --> 00:54:12,620
going on within the village.

933
00:54:12,620 --> 00:54:16,580
So it's as if almost
complete markets there.

934
00:54:16,580 --> 00:54:22,420
Then you village-level risk,
which is aggregate risk.

935
00:54:22,420 --> 00:54:25,660
And then someone comes
in ostensibly trying

936
00:54:25,660 --> 00:54:28,390
to insure that aggregate risk.

937
00:54:28,390 --> 00:54:30,730
So you would think, oh great,
they're already covered

938
00:54:30,730 --> 00:54:33,070
on the idiosyncratic side.

939
00:54:33,070 --> 00:54:34,960
Now someone else can
come in and cover them

940
00:54:34,960 --> 00:54:37,160
on the aggregate side.

941
00:54:37,160 --> 00:54:39,900
Must be a good thing.

942
00:54:39,900 --> 00:54:42,810
And the answer is, well,
no, not necessarily.

943
00:54:42,810 --> 00:54:45,720
I've already basically
given this away.

944
00:54:45,720 --> 00:54:48,390
Because you have
risk-tolerant people

945
00:54:48,390 --> 00:54:51,120
in the village who were
willing to underwrite

946
00:54:51,120 --> 00:54:53,410
that aggregate risk.

947
00:54:53,410 --> 00:54:57,280
And it shows up in higher
levels of consumption for them,

948
00:54:57,280 --> 00:54:58,900
on average.

949
00:54:58,900 --> 00:55:01,480
It's not as though
people wrote a check,

950
00:55:01,480 --> 00:55:02,750
and you got the premium.

951
00:55:02,750 --> 00:55:05,960
But it's showing up,
in whatever mechanism

952
00:55:05,960 --> 00:55:11,010
it is, in the give-and-take
with informal transactions

953
00:55:11,010 --> 00:55:12,960
and so on.

954
00:55:21,840 --> 00:55:26,040
So to see how
we're estimating it

955
00:55:26,040 --> 00:55:32,760
and get some intuition for how
we estimate the risk aversion,

956
00:55:32,760 --> 00:55:36,600
if two households had more
or less the same coefficient

957
00:55:36,600 --> 00:55:38,610
of risk tolerance,
then their consumption

958
00:55:38,610 --> 00:55:41,760
should more or
less move together.

959
00:55:41,760 --> 00:55:45,300
But the opposite, if one were
very almost risk-neutral,

960
00:55:45,300 --> 00:55:50,580
and the other very risk-averse,
then, practically, consumption

961
00:55:50,580 --> 00:55:54,600
is almost smooth
for the risk-averse

962
00:55:54,600 --> 00:55:58,290
guy by virtue of
the risk-neutral guy

963
00:55:58,290 --> 00:56:00,480
underwriting that risk.

964
00:56:00,480 --> 00:56:05,250
So you can imagine-- and
people have done this--

965
00:56:05,250 --> 00:56:07,860
looking at households
pairwise, and just

966
00:56:07,860 --> 00:56:10,470
trying to see how many
consumption reversals there

967
00:56:10,470 --> 00:56:11,730
are, and things.

968
00:56:11,730 --> 00:56:12,990
We don't quite do that.

969
00:56:12,990 --> 00:56:15,930
We don't have enough data.

970
00:56:15,930 --> 00:56:18,750
And I'm not sure it's very
effective econometrically.

971
00:56:18,750 --> 00:56:24,510
We do take a household, j, take
their measure of co-movement,

972
00:56:24,510 --> 00:56:27,480
essentially, compared to
the co-movements of all

973
00:56:27,480 --> 00:56:29,200
the other households.

974
00:56:29,200 --> 00:56:32,600
So it's always j,
less time, compared

975
00:56:32,600 --> 00:56:33,600
with all the other ones.

976
00:56:33,600 --> 00:56:36,670
Well, it's probably easier
to do it with the notation.

977
00:56:36,670 --> 00:56:41,190
So here is that basic
first-order condition.

978
00:56:41,190 --> 00:56:44,010
With constant relative
risk-aversion, by the way,

979
00:56:44,010 --> 00:56:47,610
it's in logs, not levels.

980
00:56:47,610 --> 00:56:48,930
But it's the same idea.

981
00:56:48,930 --> 00:56:51,240
You've got the Pareto weight.

982
00:56:51,240 --> 00:56:56,340
We're actually going to allow
differences in discount rates

983
00:56:56,340 --> 00:56:58,480
across households.

984
00:56:58,480 --> 00:57:01,830
And so that sends
sets up a trend,

985
00:57:01,830 --> 00:57:06,740
where some people are going up
and some people are going down

986
00:57:06,740 --> 00:57:07,800
long term.

987
00:57:07,800 --> 00:57:09,720
We allow preference shocks--

988
00:57:09,720 --> 00:57:14,550
actually monthly shocks, because
there are holidays and things.

989
00:57:14,550 --> 00:57:17,790
So the idea is to put as many
controls in, in some sense,

990
00:57:17,790 --> 00:57:19,680
that the theory would allow.

991
00:57:19,680 --> 00:57:24,660
And then we have this sort
of common fixed effect.

992
00:57:24,660 --> 00:57:30,300
And note in particular
that actually everything,

993
00:57:30,300 --> 00:57:33,210
but in particular this
thing, is normalized

994
00:57:33,210 --> 00:57:41,110
by basically the risk aversion.

995
00:57:41,110 --> 00:57:45,760
If risk aversion is really high,
this coefficient is really low,

996
00:57:45,760 --> 00:57:48,850
and so that household's
not moving around

997
00:57:48,850 --> 00:57:50,117
with the aggregate.

998
00:57:54,990 --> 00:57:59,730
So basically-- sorry--
almost what we're doing

999
00:57:59,730 --> 00:58:02,430
is taking consumption,
regressing it

1000
00:58:02,430 --> 00:58:05,850
against trends, and
monthly effects,

1001
00:58:05,850 --> 00:58:10,800
and household fixed effects,
and getting a residual.

1002
00:58:10,800 --> 00:58:13,260
The residual is
capturing all of this--

1003
00:58:13,260 --> 00:58:17,660
call it v or eta, and
then just sort of see.

1004
00:58:17,660 --> 00:58:21,900
What the correlation
is between a household

1005
00:58:21,900 --> 00:58:27,510
i's residual, which is a
stand-in for that aggregate

1006
00:58:27,510 --> 00:58:31,560
that's moving consumption, and
any other household's i-prime.

1007
00:58:34,130 --> 00:58:36,990
So this is like a
moment, basically.

1008
00:58:36,990 --> 00:58:41,150
And we can use
method of moments,

1009
00:58:41,150 --> 00:58:44,670
in principle, for all
the households, pairwise.

1010
00:58:44,670 --> 00:58:47,850
But we didn't have
enough data for that,

1011
00:58:47,850 --> 00:58:50,870
even though we have a lot of
households and a lot of months.

1012
00:58:50,870 --> 00:58:53,960
So we just summed that we have
household i's risk tolerance--

1013
00:58:53,960 --> 00:58:57,020
essentially what this
thing turns out to be

1014
00:58:57,020 --> 00:59:00,680
is household i's risk
tolerance times the sum

1015
00:59:00,680 --> 00:59:06,200
of the risk toleri of the other
households in the village.

1016
00:59:06,200 --> 00:59:08,090
But hopefully you're
getting some intuition

1017
00:59:08,090 --> 00:59:12,515
for where we're getting
the risk aversion measures.

1018
00:59:17,850 --> 00:59:21,590
This, I think, you
can ignore today.

1019
00:59:21,590 --> 00:59:25,460
It has to do with whether or
not the conventional regressions

1020
00:59:25,460 --> 00:59:31,460
that ignored heterogeneity
are biased in some way.

1021
00:59:31,460 --> 00:59:33,590
And early work--
actually Sam's work--

1022
00:59:33,590 --> 00:59:41,260
suggested that it's even harder
to reject full risk-sharing

1023
00:59:41,260 --> 00:59:44,770
if you allow for
the heterogeneity.

1024
00:59:44,770 --> 00:59:48,460
Or it could seem that you're
rejecting full risk-sharing

1025
00:59:48,460 --> 00:59:51,850
when you assume homogeneity,
because you haven't allowed

1026
00:59:51,850 --> 00:59:54,220
this asymmetric
adjustment that's

1027
00:59:54,220 --> 00:59:56,530
possible on the
part of households,

1028
00:59:56,530 --> 00:59:58,970
depending on their
risk tolerance.

1029
00:59:58,970 --> 01:00:03,730
And that's all I want to say,
probably even false intuition.

1030
01:00:03,730 --> 01:00:06,250
Because it may not
be true in general.

1031
01:00:13,330 --> 01:00:15,910
But then you kind of want
to measure welfare gains.

1032
01:00:15,910 --> 01:00:17,830
OK, so how much
would a household

1033
01:00:17,830 --> 01:00:22,960
be willing to pay living
in a risky situation

1034
01:00:22,960 --> 01:00:25,640
to remove that risk.

1035
01:00:25,640 --> 01:00:29,830
In other words, what fraction
of average consumption would

1036
01:00:29,830 --> 01:00:33,220
they be willing to take a
hit on and live in a world

1037
01:00:33,220 --> 01:00:36,700
without the risk, as
opposed to the world

1038
01:00:36,700 --> 01:00:39,190
that they're currently in.

1039
01:00:39,190 --> 01:00:41,740
Or how much would they be
willing to pay for someone

1040
01:00:41,740 --> 01:00:45,310
to come in as an outsider
targeting insurance

1041
01:00:45,310 --> 01:00:47,155
against, say, rainfall?

1042
01:00:50,840 --> 01:00:52,340
Lucas did this with
business cycles.

1043
01:00:54,980 --> 01:01:00,610
So this number, k, which
is the fraction you're

1044
01:01:00,610 --> 01:01:03,460
willing to give up, if
positive means that you

1045
01:01:03,460 --> 01:01:07,810
gain from eliminating
aggregate risk, but k

1046
01:01:07,810 --> 01:01:09,670
can actually turn
out to be negative.

1047
01:01:09,670 --> 01:01:12,020
Here is an expression for it.

1048
01:01:12,020 --> 01:01:16,380
Suffice it to say that it has
to do with the risk tolerance.

1049
01:01:16,380 --> 01:01:31,790
And although this slide is a bit
tiny, village by village, this

1050
01:01:31,790 --> 01:01:33,690
is 0.

1051
01:01:33,690 --> 01:01:37,520
So basically we're
lining up the households

1052
01:01:37,520 --> 01:01:42,510
by their degree of tolerance,
from very risk-averse to not

1053
01:01:42,510 --> 01:01:43,860
very.

1054
01:01:43,860 --> 01:01:46,170
And these are
downward-sloping lines.

1055
01:01:46,170 --> 01:01:57,040
So basically the higher this
risk tolerance, the more likely

1056
01:01:57,040 --> 01:02:02,230
it is that this k or this
welfare gain is negative.

1057
01:02:02,230 --> 01:02:03,930
So it's not trivial.

1058
01:02:03,930 --> 01:02:06,730
And by the way, it's not
correlated with characteristics

1059
01:02:06,730 --> 01:02:07,590
either.

1060
01:02:07,590 --> 01:02:09,410
You say, oh, it's
the rich people.

1061
01:02:09,410 --> 01:02:11,430
No, that's wrong.

1062
01:02:11,430 --> 01:02:16,210
Risk aversion is not
correlated with wealth.

1063
01:02:16,210 --> 01:02:21,042
It's as if complete
markets, remember?

1064
01:02:21,042 --> 01:02:23,500
AUDIENCE: Could it be due to
[INAUDIBLE] within the village

1065
01:02:23,500 --> 01:02:26,900
[INAUDIBLE] institutions.

1066
01:02:26,900 --> 01:02:30,180
So two equally risk-averse
people, but one [INAUDIBLE]

1067
01:02:30,180 --> 01:02:32,280
so they [INAUDIBLE].

1068
01:02:32,280 --> 01:02:35,250
ROBERT TOWNSEND: We actually
initially did this just

1069
01:02:35,250 --> 01:02:38,190
on the network
households, because I

1070
01:02:38,190 --> 01:02:40,530
was worried about having
the non-network households

1071
01:02:40,530 --> 01:02:42,930
in there.

1072
01:02:42,930 --> 01:02:46,020
Although when we actually
run the full test now,

1073
01:02:46,020 --> 01:02:48,780
for risk-sharing, taking into
account the heterogeneity,

1074
01:02:48,780 --> 01:02:52,410
we don't reject either.

1075
01:02:52,410 --> 01:02:53,700
I don't totally believe it.

1076
01:02:53,700 --> 01:02:56,220
I still think there are
relatively poor people who tend

1077
01:02:56,220 --> 01:02:58,200
to be a little more vulnerable.

1078
01:02:58,200 --> 01:03:00,330
But somehow, with all
this heterogeneity,

1079
01:03:00,330 --> 01:03:04,560
it's not showing up the
way I would have thought.

1080
01:03:04,560 --> 01:03:07,484
It's not soundly
rejecting full insurance.

1081
01:03:07,484 --> 01:03:09,276
AUDIENCE: From the
data, can you see, then,

1082
01:03:09,276 --> 01:03:13,608
if the risk tolerance
is greater amongst--

1083
01:03:13,608 --> 01:03:16,150
is there a distinction between
risk tolerance and [INAUDIBLE]

1084
01:03:16,150 --> 01:03:19,500
network [INAUDIBLE]?

1085
01:03:19,500 --> 01:03:22,940
ROBERT TOWNSEND:
Oh, I'm not sure we

1086
01:03:22,940 --> 01:03:26,390
have enough non-network
households to do it.

1087
01:03:26,390 --> 01:03:28,640
But the counter to that
is that we certainly

1088
01:03:28,640 --> 01:03:31,860
had enough in network with
Cynthia to find something.

1089
01:03:31,860 --> 01:03:33,700
So it's probably worth trying.

1090
01:03:33,700 --> 01:03:35,700
I haven't done it yet.

1091
01:03:35,700 --> 01:03:38,990
Yeah, the worry here is that
we've left something out.

1092
01:03:38,990 --> 01:03:42,260
The model is just making a
lot of strong assumptions

1093
01:03:42,260 --> 01:03:45,140
that there's no alternatives.

1094
01:03:45,140 --> 01:03:46,640
To try to say it
in words, we're not

1095
01:03:46,640 --> 01:03:48,680
looking at some indirect
utility function.

1096
01:03:48,680 --> 01:03:51,490
We're looking at the
direct utility function.

1097
01:03:51,490 --> 01:03:54,815
And indirect might be there if
they have other ways of coping.

1098
01:03:57,710 --> 01:04:00,410
When we get to labor
supply, we will

1099
01:04:00,410 --> 01:04:03,650
be quite explicit
about supplying labor

1100
01:04:03,650 --> 01:04:06,880
as a way to cope with risk, and
then revisit these equations.

1101
01:04:15,990 --> 01:04:21,310
So people have--

1102
01:04:21,310 --> 01:04:26,200
I have, too-- offered insurance
products like rainfall

1103
01:04:26,200 --> 01:04:28,840
insurance in India.

1104
01:04:28,840 --> 01:04:32,220
We haven't quite done
it yet in Thailand.

1105
01:04:32,220 --> 01:04:36,550
And I'll tell you why we
have misgivings about it,

1106
01:04:36,550 --> 01:04:37,690
and some of the puzzles.

1107
01:04:37,690 --> 01:04:42,480
But basically with less of an
exception in Ghana and Chris

1108
01:04:42,480 --> 01:04:45,940
Udry, with Dean,
most experiments

1109
01:04:45,940 --> 01:04:48,370
in the world, the take-up
of these insurance products

1110
01:04:48,370 --> 01:04:49,450
is pretty low.

1111
01:04:49,450 --> 01:04:53,800
Or in our case, in India,
people will insure one plot,

1112
01:04:53,800 --> 01:04:56,890
and not insure the other
plots that they own.

1113
01:04:56,890 --> 01:05:01,270
So you know, it's
not totally terrible.

1114
01:05:01,270 --> 01:05:03,160
There is some price elasticity.

1115
01:05:03,160 --> 01:05:05,680
But there are puzzles.

1116
01:05:05,680 --> 01:05:08,710
I mean, if people are so
poor and so vulnerable,

1117
01:05:08,710 --> 01:05:10,330
then why aren't
taking it up more?

1118
01:05:10,330 --> 01:05:15,150
Now, one answer you
may have already seen.

1119
01:05:15,150 --> 01:05:18,850
They don't really need it.

1120
01:05:18,850 --> 01:05:20,858
Because even if they're
somewhat vulnerable,

1121
01:05:20,858 --> 01:05:22,900
there are people in the
village willing to assume

1122
01:05:22,900 --> 01:05:23,930
some of that risk.

1123
01:05:23,930 --> 01:05:27,250
And even if it isn't perfect,
at the end of the day, given

1124
01:05:27,250 --> 01:05:30,220
the way the policies
are priced, maybe you're

1125
01:05:30,220 --> 01:05:32,470
just not going to
see that take-up.

1126
01:05:32,470 --> 01:05:35,320
Unfortunately-- and people
aren't either taking

1127
01:05:35,320 --> 01:05:40,480
the time to get baseline data
or don't bother with it at all.

1128
01:05:40,480 --> 01:05:44,470
They just go in
there with an RCT.

1129
01:05:44,470 --> 01:05:46,490
Well, we did.

1130
01:05:46,490 --> 01:05:47,060
Yeah.

1131
01:05:47,060 --> 01:05:50,246
AUDIENCE: So is that
just rainfall insurance?

1132
01:05:50,246 --> 01:05:51,670
Or any kinds of insurance?

1133
01:05:51,670 --> 01:05:53,860
ROBERT TOWNSEND:
Could be anything.

1134
01:05:53,860 --> 01:05:55,930
I'm going to focus on
rainfall as just a way

1135
01:05:55,930 --> 01:05:57,280
to give it a label.

1136
01:05:57,280 --> 01:06:00,380
Could be any kind
of weather risk.

1137
01:06:00,380 --> 01:06:03,880
It could actually be price risk.

1138
01:06:03,880 --> 01:06:07,510
Case-Shiller index in the
US is about price risk.

1139
01:06:07,510 --> 01:06:09,640
It wasn't just to
measure housing prices

1140
01:06:09,640 --> 01:06:12,460
to get his name in The Wall
Street Journal every month.

1141
01:06:12,460 --> 01:06:17,230
It was originally designed
because Bob Shiller had spent

1142
01:06:17,230 --> 01:06:21,880
years creating a risk
instrument which he was trying

1143
01:06:21,880 --> 01:06:24,790
to promote on Wall Street.

1144
01:06:24,790 --> 01:06:26,510
So yeah, so it could be--

1145
01:06:26,510 --> 01:06:28,450
and in urban areas,
maybe it is more

1146
01:06:28,450 --> 01:06:34,090
likely to be vulnerability to
price rather than rainfall.

1147
01:06:34,090 --> 01:06:37,420
So rainfall here is kind
of like, partly, a metaphor

1148
01:06:37,420 --> 01:06:38,770
for the problem in general.

1149
01:06:38,770 --> 01:06:41,680
But I'll come back
with a vengeance

1150
01:06:41,680 --> 01:06:45,196
to look at rainfall in
these Thai villages.

1151
01:06:45,196 --> 01:06:48,650
AUDIENCE: [INAUDIBLE]
even summer and winter,

1152
01:06:48,650 --> 01:06:52,783
they say that rain
affects [INAUDIBLE]..

1153
01:06:52,783 --> 01:06:53,700
ROBERT TOWNSEND: Yeah.

1154
01:06:53,700 --> 01:06:57,660
People don't come and buy
when it's raining, and so on.

1155
01:06:57,660 --> 01:07:01,650
But you came up with a situation
where it was actually good.

1156
01:07:01,650 --> 01:07:03,150
Just trying to
remember what it was.

1157
01:07:03,150 --> 01:07:04,530
AUDIENCE: I think
we said, umbrellas.

1158
01:07:04,530 --> 01:07:06,322
ROBERT TOWNSEND: Yeah,
something like that.

1159
01:07:10,580 --> 01:07:17,690
OK, so here's the vengeance
paper with Kamilya.

1160
01:07:17,690 --> 01:07:19,850
So this is all about rain.

1161
01:07:19,850 --> 01:07:29,270
So we want to see the effect
of rain on rice production.

1162
01:07:29,270 --> 01:07:31,940
Now again, the typical
thing that you'll

1163
01:07:31,940 --> 01:07:33,440
see in the
development literature

1164
01:07:33,440 --> 01:07:36,470
is rainfall is the risk.

1165
01:07:36,470 --> 01:07:39,560
Let's just sort of instrument
the relevant variable

1166
01:07:39,560 --> 01:07:43,340
by rainfall, and then
start looking at migration

1167
01:07:43,340 --> 01:07:44,918
and all kinds of stuff.

1168
01:07:44,918 --> 01:07:46,460
So here, the idea
is, let's just look

1169
01:07:46,460 --> 01:07:50,110
at rainfall in the village,
start instrumenting income

1170
01:07:50,110 --> 01:07:53,900
or rice production by
rainfall, and then look

1171
01:07:53,900 --> 01:07:58,790
at the vulnerability
of consumption

1172
01:07:58,790 --> 01:08:04,470
to instrumented income,
which people do, a lot.

1173
01:08:04,470 --> 01:08:08,800
Now, it's going to
turn out that things

1174
01:08:08,800 --> 01:08:11,500
are more complicated than
that within the village.

1175
01:08:11,500 --> 01:08:12,670
The soil is different.

1176
01:08:12,670 --> 01:08:15,220
The hydrology and water
flow are different.

1177
01:08:15,220 --> 01:08:20,069
They actually plant
at different times.

1178
01:08:20,069 --> 01:08:25,050
So a common rainfall shock,
even if it is actually common,

1179
01:08:25,050 --> 01:08:28,770
and shows up and all the
rain gauges which we have,

1180
01:08:28,770 --> 01:08:31,740
it's still the case that
it has differential impact

1181
01:08:31,740 --> 01:08:36,990
on the households, especially
depending on when they plant,

1182
01:08:36,990 --> 01:08:42,350
which is in turn a
function of soil and so on.

1183
01:08:42,350 --> 01:08:44,859
So what are Kamilya and I doing?

1184
01:08:44,859 --> 01:08:48,069
We're going to estimate
a three-stage production

1185
01:08:48,069 --> 01:08:54,550
function which takes into
account the sequential nature.

1186
01:08:54,550 --> 01:08:55,990
There's planting, for sure.

1187
01:08:55,990 --> 01:08:59,680
And then you have sort of
the germination period,

1188
01:08:59,680 --> 01:09:01,810
and the planting of
the rice, and so on,

1189
01:09:01,810 --> 01:09:03,729
with maybe some
fertilizing and so on.

1190
01:09:03,729 --> 01:09:05,925
Then you've got
sort of midseason,

1191
01:09:05,925 --> 01:09:07,300
and all the stuff
you need to do.

1192
01:09:07,300 --> 01:09:10,615
And then you finally
approach and include harvest.

1193
01:09:14,040 --> 01:09:20,960
You know, I'm going to jump,
and just show you the notation.

1194
01:09:20,960 --> 01:09:29,920
So here basically is stage
I, with predecessor stages I

1195
01:09:29,920 --> 01:09:30,609
minus 1.

1196
01:09:33,550 --> 01:09:36,250
It's going to be looking
forward to future stages

1197
01:09:36,250 --> 01:09:38,870
if you're not at the last one.

1198
01:09:38,870 --> 01:09:43,160
So if you're at the last one,
this would be the harvest.

1199
01:09:43,160 --> 01:09:45,560
That's the easiest
thing to think about.

1200
01:09:45,560 --> 01:09:47,569
If you're in the
next-to-last one,

1201
01:09:47,569 --> 01:09:51,020
the output is
basically the condition

1202
01:09:51,020 --> 01:09:54,410
of the crop prior to harvest.

1203
01:09:54,410 --> 01:09:58,040
And what influences that?

1204
01:09:58,040 --> 01:10:00,120
Well, if you've got
land, labor, capital,

1205
01:10:00,120 --> 01:10:04,850
and so on, as Cobb-Douglas
type inputs in the production

1206
01:10:04,850 --> 01:10:08,300
function--

1207
01:10:08,300 --> 01:10:14,370
but you've also got the
state of the crop plot

1208
01:10:14,370 --> 01:10:18,790
that you inherited from
the previous stage.

1209
01:10:18,790 --> 01:10:21,130
And then all of this gets
hit, at the end of the period,

1210
01:10:21,130 --> 01:10:24,565
with more shocks,
including rain, and so on,

1211
01:10:24,565 --> 01:10:27,210
and temperature.

1212
01:10:27,210 --> 01:10:30,870
So this is not a Cobb-Douglas.

1213
01:10:30,870 --> 01:10:36,510
It's a CES that allows
some substitutability

1214
01:10:36,510 --> 01:10:40,080
to be estimated between
the current state

1215
01:10:40,080 --> 01:10:43,930
and contemporaneous inputs.

1216
01:10:47,800 --> 01:10:50,890
It may not be as
extreme as Leontief.

1217
01:10:50,890 --> 01:10:54,070
So you're kind of doomed by
the state of the crop plot

1218
01:10:54,070 --> 01:10:56,620
at this time,
based on everything

1219
01:10:56,620 --> 01:10:58,990
that happened in the past.

1220
01:10:58,990 --> 01:11:01,660
And then you just have to
adjust inputs in proportion.

1221
01:11:01,660 --> 01:11:04,240
Doesn't have to be that
bad, but it turns out

1222
01:11:04,240 --> 01:11:06,263
not to be like Cobb-Douglas.

1223
01:11:06,263 --> 01:11:07,930
If you want to think
about Cobb-Douglas,

1224
01:11:07,930 --> 01:11:10,410
there's a whole lot
of substitutability--

1225
01:11:10,410 --> 01:11:12,055
labor today, labor tomorrow.

1226
01:11:12,055 --> 01:11:13,750
Oh, it's not that
much of a difference.

1227
01:11:13,750 --> 01:11:17,230
Why don't we wait and figure
out what the rainfall is first,

1228
01:11:17,230 --> 01:11:20,820
and then adjust
the inputs later?

1229
01:11:20,820 --> 01:11:24,465
So this allows us to estimate
the ability to adjust.

1230
01:11:27,350 --> 01:11:29,660
And essentially the
reason you can't

1231
01:11:29,660 --> 01:11:33,200
use rain, aggregated up
over the whole season,

1232
01:11:33,200 --> 01:11:35,840
is because of this
imperfect substitutability

1233
01:11:35,840 --> 01:11:36,710
across the stages.

1234
01:11:39,410 --> 01:11:40,078
Yep.

1235
01:11:40,078 --> 01:11:43,420
AUDIENCE: It says
that it introduces--

1236
01:11:43,420 --> 01:11:46,000
[INAUDIBLE] that
[INAUDIBLE] induces

1237
01:11:46,000 --> 01:11:48,462
people to plant at different
times, for example.

1238
01:11:48,462 --> 01:11:50,170
ROBERT TOWNSEND: No,
that's not in there.

1239
01:11:50,170 --> 01:11:53,650
There's a separate equation
that has to do-- like a probit,

1240
01:11:53,650 --> 01:11:57,310
basically, which determines
planting times as we see it

1241
01:11:57,310 --> 01:11:58,000
in the data.

1242
01:12:05,970 --> 01:12:16,870
So we can look at the effect
of rainfall on rice production.

1243
01:12:16,870 --> 01:12:20,520
And we can actually do
something with climate change.

1244
01:12:20,520 --> 01:12:23,430
Let me show you this thing.

1245
01:12:23,430 --> 01:12:32,020
So this black thing
here is the distribution

1246
01:12:32,020 --> 01:12:35,290
of actual yields
across households--

1247
01:12:40,680 --> 01:12:42,710
and actually it's
households and time.

1248
01:12:45,540 --> 01:12:55,340
This peak thing is just
using a weather index.

1249
01:12:55,340 --> 01:12:57,860
So you regress yields--

1250
01:12:57,860 --> 01:13:02,750
same data, our data, onto
temperature and rainfall,

1251
01:13:02,750 --> 01:13:04,760
and a whole suite of
them, a big, long vector.

1252
01:13:04,760 --> 01:13:07,420
It doesn't have to be,
narrowly, rainfall.

1253
01:13:07,420 --> 01:13:10,510
Try to be as
realistic as possible.

1254
01:13:10,510 --> 01:13:12,250
And then you run a regression.

1255
01:13:12,250 --> 01:13:15,370
And then you get basically
the predicted through the lens

1256
01:13:15,370 --> 01:13:20,570
of that weather index model.

1257
01:13:20,570 --> 01:13:22,040
And it's close.

1258
01:13:22,040 --> 01:13:23,370
It's very, very peaked.

1259
01:13:26,450 --> 01:13:28,250
Well, sorry, that's
the first thing.

1260
01:13:28,250 --> 01:13:31,350
Then this thing is
using the full model--

1261
01:13:31,350 --> 01:13:34,610
sorry, using the model, but only
the observed rainfall variable,

1262
01:13:34,610 --> 01:13:37,760
and setting everything else at
their sample average-- common

1263
01:13:37,760 --> 01:13:42,110
soil, common fertilizer.

1264
01:13:42,110 --> 01:13:47,120
And then the big jump
down here toward the data

1265
01:13:47,120 --> 01:13:50,600
is when you start to allow the
variation that the model has

1266
01:13:50,600 --> 01:13:56,850
in it, certainly soil variation,
and eventually full variation.

1267
01:13:56,850 --> 01:14:01,580
So we're not one to one with
the actual distribution,

1268
01:14:01,580 --> 01:14:03,980
but we've made a
substantial improvement

1269
01:14:03,980 --> 01:14:06,440
from simple weather indices.

1270
01:14:12,310 --> 01:14:15,910
So people talk about basis risk.

1271
01:14:15,910 --> 01:14:18,400
And what they mean
typically is the reason

1272
01:14:18,400 --> 01:14:20,500
there's not take-up of
a rainfall insurance

1273
01:14:20,500 --> 01:14:24,340
product is because the rain
gauge is way out there,

1274
01:14:24,340 --> 01:14:29,620
so distant that the correlation
of the rainfall at the gauge

1275
01:14:29,620 --> 01:14:33,880
with the correlation of the
rainfall in the village is low.

1276
01:14:33,880 --> 01:14:37,450
In fact, the Bank for
Agriculture introduced--

1277
01:14:37,450 --> 01:14:40,150
I guess it's now
three years ago--

1278
01:14:40,150 --> 01:14:43,460
an index product.

1279
01:14:43,460 --> 01:14:46,270
And they already restricted
it to be something

1280
01:14:46,270 --> 01:14:48,580
like 20, 25 kilometers.

1281
01:14:48,580 --> 01:14:51,010
And the take-up was terrible.

1282
01:14:51,010 --> 01:14:54,340
So then they said, oh,
we know, it's basis risk.

1283
01:14:54,340 --> 01:14:57,460
So then they went down
to something like 15,

1284
01:14:57,460 --> 01:14:59,860
lowered the radius.

1285
01:14:59,860 --> 01:15:01,760
And take-up was
still pretty bad.

1286
01:15:01,760 --> 01:15:04,177
And I think they're going to
have to withdraw the product.

1287
01:15:06,580 --> 01:15:09,780
So this is a kind of basis risk.

1288
01:15:09,780 --> 01:15:12,270
What we're saying
is, even though we're

1289
01:15:12,270 --> 01:15:14,580
sitting on top of
the rainfall gauge,

1290
01:15:14,580 --> 01:15:17,100
within the village, the
way that that rainfall

1291
01:15:17,100 --> 01:15:22,710
is impacting these farmers is
different across the farmers.

1292
01:15:22,710 --> 01:15:27,630
So we're busy sort of thinking
about designing, still,

1293
01:15:27,630 --> 01:15:31,230
a rainfall product, but
trying to take into account

1294
01:15:31,230 --> 01:15:32,320
this heterogeneity.

1295
01:15:32,320 --> 01:15:32,820
Yes.

1296
01:15:32,820 --> 01:15:35,112
AUDIENCE: Maybe this is kind
of what you're getting at,

1297
01:15:35,112 --> 01:15:38,540
[INAUDIBLE] how flexible is the
sort of payout structure given?

1298
01:15:38,540 --> 01:15:40,623
I'm just trying to-- because
it seems like there's

1299
01:15:40,623 --> 01:15:42,340
kind of like, a [INAUDIBLE]

1300
01:15:42,340 --> 01:15:46,210
ROBERT TOWNSEND: Yeah, so
let me show you Kapphan's--

1301
01:15:48,850 --> 01:15:50,500
her dissertation.

1302
01:15:58,710 --> 01:16:01,467
So here's her idea,
which is basically--

1303
01:16:01,467 --> 01:16:03,050
and this is typical
of the literature.

1304
01:16:03,050 --> 01:16:05,090
She did better than most.

1305
01:16:05,090 --> 01:16:09,530
You have a utility
function over consumption.

1306
01:16:09,530 --> 01:16:12,830
Consumption would
be autarky income.

1307
01:16:12,830 --> 01:16:14,210
But you have a payout--

1308
01:16:14,210 --> 01:16:16,640
could be positive or negative.

1309
01:16:16,640 --> 01:16:20,720
And that's a function of
this weather index, z.

1310
01:16:20,720 --> 01:16:23,210
So this is supposed
to be smoothing.

1311
01:16:23,210 --> 01:16:26,600
And this is an optimal
design problem.

1312
01:16:26,600 --> 01:16:29,472
It's output conditioned
on the weather index,

1313
01:16:29,472 --> 01:16:31,430
and then the distribution
of the weather index.

1314
01:16:31,430 --> 01:16:33,305
By the way, where is
she getting all of this?

1315
01:16:33,305 --> 01:16:35,480
She's doing it in the lab.

1316
01:16:35,480 --> 01:16:38,690
She's basically simulated
millet production

1317
01:16:38,690 --> 01:16:40,460
in Switzerland or something.

1318
01:16:40,460 --> 01:16:44,540
But she's not out looking
at the actual crops.

1319
01:16:44,540 --> 01:16:46,605
AUDIENCE: Is that
[INAUDIBLE] soil?

1320
01:16:46,605 --> 01:16:47,480
ROBERT TOWNSEND: Hmm?

1321
01:16:47,480 --> 01:16:50,040
AUDIENCE: Is it
[INAUDIBLE] soil?

1322
01:16:50,040 --> 01:16:51,800
ROBERT TOWNSEND: I
don't remember exactly.

1323
01:16:51,800 --> 01:16:53,960
But that is a big
problem in practice

1324
01:16:53,960 --> 01:16:57,140
when you take these sort
of lab-type experiments.

1325
01:16:57,140 --> 01:17:00,670
I forgot to say, we actually
use a crop grow model as well.

1326
01:17:00,670 --> 01:17:02,660
It's called DSSAT.

1327
01:17:02,660 --> 01:17:08,060
And we carefully feed in soil
depth and soil into that model

1328
01:17:08,060 --> 01:17:09,380
I just showed you.

1329
01:17:09,380 --> 01:17:12,590
But to address
your question, sort

1330
01:17:12,590 --> 01:17:15,960
of optimized over this thing,
this is the policy design.

1331
01:17:15,960 --> 01:17:18,380
This is the optimal
contract design problem.

1332
01:17:18,380 --> 01:17:19,660
How cool.

1333
01:17:19,660 --> 01:17:21,970
I mean, it really is cool.

1334
01:17:21,970 --> 01:17:28,110
And you know, you can see,
in her simulated data,

1335
01:17:28,110 --> 01:17:30,360
the distribution of
yields for any index.

1336
01:17:33,790 --> 01:17:36,960
And then here's the
optimized payout,

1337
01:17:36,960 --> 01:17:42,270
which is basically positive,
definitely non-linear.

1338
01:17:42,270 --> 01:17:45,220
And basically it's
like [WHOOSHING NOISE]..

1339
01:17:45,220 --> 01:17:48,630
And many of the products that
people already came up with

1340
01:17:48,630 --> 01:17:54,630
are like these kind of
linear, flat schedules.

1341
01:17:54,630 --> 01:17:57,900
This has more wiggles, and
it's more non-linear than that.

1342
01:17:57,900 --> 01:17:59,650
But this is all
negative territory.

1343
01:17:59,650 --> 01:18:02,470
This is basically
ex-post premia down here.

1344
01:18:02,470 --> 01:18:03,270
Yes.

1345
01:18:03,270 --> 01:18:05,590
AUDIENCE: So these
[INAUDIBLE] models,

1346
01:18:05,590 --> 01:18:08,390
do they try to account for
price changes that also result

1347
01:18:08,390 --> 01:18:10,360
from aggregate weather shocks?

1348
01:18:10,360 --> 01:18:10,860
[INAUDIBLE]

1349
01:18:10,860 --> 01:18:10,980
ROBERT TOWNSEND: No.

1350
01:18:10,980 --> 01:18:13,322
AUDIENCE: [INAUDIBLE]
everybody [INAUDIBLE]..

1351
01:18:13,322 --> 01:18:15,530
Because that could push it
in the opposite direction.

1352
01:18:15,530 --> 01:18:18,590
AUDIENCE: Yes, absolutely.

1353
01:18:18,590 --> 01:18:21,160
And I couldn't agree more.

1354
01:18:21,160 --> 01:18:23,660
Now, at some level, you've
got to either use data

1355
01:18:23,660 --> 01:18:26,840
to decide how price is moving--

1356
01:18:26,840 --> 01:18:28,690
that's a good starting point--

1357
01:18:28,690 --> 01:18:33,230
and/or incorporate in your
model like we saw with the macro

1358
01:18:33,230 --> 01:18:34,880
development models,
the movements

1359
01:18:34,880 --> 01:18:39,270
of wages and interest rates
and so on with interventions.

1360
01:18:39,270 --> 01:18:44,280
So you don't have to
do just one thing,

1361
01:18:44,280 --> 01:18:47,400
but you're right-- you need
to take it into account.

1362
01:18:47,400 --> 01:18:50,592
In Thailand, rice is
the dominant crop.

1363
01:18:50,592 --> 01:18:52,800
It's pretty hard to argue
that like the price of rice

1364
01:18:52,800 --> 01:18:54,960
in the village is a
function of village or even

1365
01:18:54,960 --> 01:18:56,475
necessarily regional rainfall.

1366
01:19:00,595 --> 01:19:02,762
AUDIENCE: It would have to
be a pretty big aggregate

1367
01:19:02,762 --> 01:19:03,873
[INAUDIBLE] price.

1368
01:19:03,873 --> 01:19:04,790
ROBERT TOWNSEND: Yeah.

1369
01:19:09,530 --> 01:19:16,730
So we want to do something
like that, that optimal design,

1370
01:19:16,730 --> 01:19:18,500
to take into account
what we've learned

1371
01:19:18,500 --> 01:19:22,190
about within-village insurance,
and their heterogeneity,

1372
01:19:22,190 --> 01:19:24,230
and so on.

1373
01:19:24,230 --> 01:19:27,770
Just two comments--
first of all,

1374
01:19:27,770 --> 01:19:30,310
is it true that
within-village stuff is great

1375
01:19:30,310 --> 01:19:33,830
and across-village stuff is not?

1376
01:19:33,830 --> 01:19:36,470
Well, in Thailand it's
not obvious at all.

1377
01:19:36,470 --> 01:19:38,180
And in Hong's helped with this.

1378
01:19:38,180 --> 01:19:40,250
We updated some earlier results.

1379
01:19:40,250 --> 01:19:43,190
But basically this
section of the slides,

1380
01:19:43,190 --> 01:19:45,770
which I won't
attempt to go over,

1381
01:19:45,770 --> 01:19:49,920
is how to test not only within
villages but across villages.

1382
01:19:49,920 --> 01:19:55,040
And then you can aggregate
it up, villages, and test

1383
01:19:55,040 --> 01:19:57,480
one county versus another.

1384
01:19:57,480 --> 01:20:01,940
So you have you have layered
sort of degrees of aggregation.

1385
01:20:01,940 --> 01:20:07,610
And you can use [INAUDIBLE]
method-- she did it in Kenya--

1386
01:20:07,610 --> 01:20:11,120
to see how good the insurance
is across the villages.

1387
01:20:11,120 --> 01:20:15,260
And in Thailand it's hard to
say that it's worse than what's

1388
01:20:15,260 --> 01:20:17,660
going on within the village.

1389
01:20:17,660 --> 01:20:21,260
So it's just
massive remittances.

1390
01:20:21,260 --> 01:20:25,160
There's just huge
financial flows coming in,

1391
01:20:25,160 --> 01:20:28,400
which we pick up in
the transactions data.

1392
01:20:28,400 --> 01:20:37,520
So I still have misgivings
about pushing too hard on this.

1393
01:20:37,520 --> 01:20:39,690
I'm sort of torn.

1394
01:20:39,690 --> 01:20:43,350
I'm torn by the attractiveness
of designing the intervention

1395
01:20:43,350 --> 01:20:46,680
based on the data, and yet
worried that they really don't

1396
01:20:46,680 --> 01:20:48,900
need this rainfall product.

1397
01:20:48,900 --> 01:20:55,910
And the other thing I want
to just point to so you can--

1398
01:20:55,910 --> 01:20:58,790
sorry, Hong, all your
great stuff here--

1399
01:21:02,720 --> 01:21:06,440
is to say that Chris Udry,
with Dean and co-authors,

1400
01:21:06,440 --> 01:21:12,800
have been doing stuff
in Ghana, and they

1401
01:21:12,800 --> 01:21:14,180
have various treatments.

1402
01:21:14,180 --> 01:21:16,940
They have rainfall insurance.

1403
01:21:16,940 --> 01:21:19,210
They have credit.

1404
01:21:19,210 --> 01:21:20,270
They have both.

1405
01:21:20,270 --> 01:21:21,610
They have one and the other.

1406
01:21:21,610 --> 01:21:23,530
They have controls.

1407
01:21:23,530 --> 01:21:26,110
And then you see stuff.

1408
01:21:26,110 --> 01:21:28,300
Like you see
sometimes that credit

1409
01:21:28,300 --> 01:21:30,760
seems to increase investment.

1410
01:21:30,760 --> 01:21:36,610
But you also see that the
combination of insurance

1411
01:21:36,610 --> 01:21:42,300
and credit is actually not
helping, although insurance

1412
01:21:42,300 --> 01:21:44,640
alone actually can help.

1413
01:21:44,640 --> 01:21:49,380
So they wrote down a model,
a simple two-period model,

1414
01:21:49,380 --> 01:21:52,380
which is clear about the
optimization problem.

1415
01:21:52,380 --> 01:21:54,120
And it's similar
in spirit to what

1416
01:21:54,120 --> 01:21:56,100
we're talking about
today, which is what's

1417
01:21:56,100 --> 01:21:59,160
the shadow price of current
consumption, what's the shadow

1418
01:21:59,160 --> 01:22:01,170
price of future consumption.

1419
01:22:01,170 --> 01:22:03,540
If there is full insurance--
which they assume

1420
01:22:03,540 --> 01:22:04,722
in the village--

1421
01:22:04,722 --> 01:22:06,180
and there are no
aggregate shocks--

1422
01:22:06,180 --> 01:22:08,360
which they assume in the model--

1423
01:22:08,360 --> 01:22:13,590
then basically you can get
counterintuitive effects going

1424
01:22:13,590 --> 01:22:17,670
on when you offer insurance.

1425
01:22:17,670 --> 01:22:19,530
Because insurance
moves tomorrow, it

1426
01:22:19,530 --> 01:22:22,080
makes tomorrow kind of less
valuable in some sense.

1427
01:22:22,080 --> 01:22:25,410
It actually seems to
undercut investment today.

1428
01:22:25,410 --> 01:22:31,020
And these other papers are again
treatments, RCTs and so on.

1429
01:22:33,840 --> 01:22:38,570
It's easy to get lost in
the details of these things,

1430
01:22:38,570 --> 01:22:41,510
and get caught up with
the language of take-up,

1431
01:22:41,510 --> 01:22:46,010
and behavioral responses, and
some very real things that

1432
01:22:46,010 --> 01:22:47,480
are out there.

1433
01:22:47,480 --> 01:22:51,290
But it's important
not to get lost,

1434
01:22:51,290 --> 01:22:54,290
and to remember the
basic sort of framework

1435
01:22:54,290 --> 01:22:57,530
that you're taking
in if you choose

1436
01:22:57,530 --> 01:23:03,010
to design one of these
randomized control trials.

1437
01:23:03,010 --> 01:23:05,030
Or let me put it another way.

1438
01:23:05,030 --> 01:23:09,380
If Udry had not been thinking
about those kinds of models,

1439
01:23:09,380 --> 01:23:12,440
he would be left with some
very strange puzzles that

1440
01:23:12,440 --> 01:23:14,670
does beg for an interpretation.

1441
01:23:14,670 --> 01:23:16,280
But even better
yet, when you have

1442
01:23:16,280 --> 01:23:19,610
the model and the structure,
ideally, with enough data,

1443
01:23:19,610 --> 01:23:21,740
you can actually
design interventions

1444
01:23:21,740 --> 01:23:25,730
which are informative about
key parameters in the model

1445
01:23:25,730 --> 01:23:30,520
and your guesses
about basic obstacles.