1
00:00:00,090 --> 00:00:02,490
The following content is
provided under a Creative

2
00:00:02,490 --> 00:00:04,059
Commons license.

3
00:00:04,059 --> 00:00:06,330
Your support will help
MIT OpenCourseWare

4
00:00:06,330 --> 00:00:10,720
continue to offer high quality
educational resources for free.

5
00:00:10,720 --> 00:00:13,350
To make a donation or
view additional materials

6
00:00:13,350 --> 00:00:17,280
from hundreds of MIT courses,
visit MIT OpenCourseWare

7
00:00:17,280 --> 00:00:18,480
at ocw.mit.edu.

8
00:00:28,024 --> 00:00:31,880
ROBERT TOWNSEND: We want to
talk about theoretical models.

9
00:00:31,880 --> 00:00:34,610
They tell us stories.

10
00:00:34,610 --> 00:00:37,310
The key variables are very
much of interest-- growth,

11
00:00:37,310 --> 00:00:42,130
inequality, poverty, but just
telling a story as elegant,

12
00:00:42,130 --> 00:00:45,070
as it may be, even
with surprises does not

13
00:00:45,070 --> 00:00:49,060
even necessarily tell us that
it matches up to reality,

14
00:00:49,060 --> 00:00:54,550
and therefore, one might be a
little bit cautious in terms

15
00:00:54,550 --> 00:00:57,220
of formulation of policy.

16
00:00:57,220 --> 00:01:02,890
Now it is true that having
sort of a good conceptual

17
00:01:02,890 --> 00:01:06,010
foundation, a good
well-articulated theoretical

18
00:01:06,010 --> 00:01:07,840
model is a very important step.

19
00:01:07,840 --> 00:01:10,740
I'm not demeaning
that step at all.

20
00:01:10,740 --> 00:01:15,365
What I'm trying to say is
it's a very useful first step,

21
00:01:15,365 --> 00:01:16,740
but in the literature
we're going

22
00:01:16,740 --> 00:01:19,930
to review today, we've
gone beyond that,

23
00:01:19,930 --> 00:01:25,840
and to try to take the next
step and see how well it fits.

24
00:01:25,840 --> 00:01:29,630
And how are we going
to, quote, fit it?

25
00:01:29,630 --> 00:01:36,160
We're going to take the assumed
micro-financial underpinnings

26
00:01:36,160 --> 00:01:42,100
and estimate those
underpinnings,

27
00:01:42,100 --> 00:01:45,580
and get the bulk of the key
parameters of the model pinned

28
00:01:45,580 --> 00:01:50,700
down in that way
from the micro data.

29
00:01:50,700 --> 00:01:53,320
There'll be a few
remaining parameters

30
00:01:53,320 --> 00:01:57,670
that we can't get from household
choice or business choice,

31
00:01:57,670 --> 00:01:59,410
and we'll calibrate those.

32
00:02:02,230 --> 00:02:06,320
And then we'll
simulate the models

33
00:02:06,320 --> 00:02:10,789
and get what the model would
predict for growth inequality--

34
00:02:10,789 --> 00:02:15,550
income, savings,
labor's share of GDP,

35
00:02:15,550 --> 00:02:18,820
and actually also some
of the key underpinnings

36
00:02:18,820 --> 00:02:20,770
of each model, and
the first one it's

37
00:02:20,770 --> 00:02:23,830
going to be the fraction of
households running businesses

38
00:02:23,830 --> 00:02:25,270
and the second
one is going to be

39
00:02:25,270 --> 00:02:28,390
the fraction of
households participating

40
00:02:28,390 --> 00:02:29,735
in the financial sector.

41
00:02:35,380 --> 00:02:38,425
So what do we do with these?

42
00:02:41,580 --> 00:02:44,430
Each one, then,
after we get done

43
00:02:44,430 --> 00:02:47,370
looking at the
goodness of fit, we

44
00:02:47,370 --> 00:02:53,400
can do welfare analysis
of policy intervention.

45
00:02:53,400 --> 00:02:56,910
The first model has,
to make it easier,

46
00:02:56,910 --> 00:03:00,210
an exogenously expanding
financial sector

47
00:03:00,210 --> 00:03:04,240
at exactly the rate
we see it in the data,

48
00:03:04,240 --> 00:03:08,460
and that's going to
create winners or losers.

49
00:03:08,460 --> 00:03:12,120
As the financial
system expands, we'll

50
00:03:12,120 --> 00:03:14,250
get changes on the
extensive margin

51
00:03:14,250 --> 00:03:19,200
because some households
will go into business.

52
00:03:19,200 --> 00:03:21,960
We'll get changes as
wealth accumulates

53
00:03:21,960 --> 00:03:25,410
in the amount to take
and invest in business,

54
00:03:25,410 --> 00:03:28,890
and also because it's going to
be a general equilibrium model,

55
00:03:28,890 --> 00:03:33,120
we're going to get changes in
equilibrium wages and interest

56
00:03:33,120 --> 00:03:38,490
rates which have effects even
beyond the direct effects,

57
00:03:38,490 --> 00:03:40,860
often quite big.

58
00:03:40,860 --> 00:03:47,460
In fact, there's a debate,
depending on which political

59
00:03:47,460 --> 00:03:54,370
position you take,
about trickle-down,

60
00:03:54,370 --> 00:03:57,210
which is sometimes
taken negatively,

61
00:03:57,210 --> 00:04:00,360
trickle-down meaning only
the rich benefit from various

62
00:04:00,360 --> 00:04:02,430
things and the poor--

63
00:04:02,430 --> 00:04:05,910
workers are lucky to get
anything, a better way

64
00:04:05,910 --> 00:04:11,190
to say it is a rising
tide raises all boats.

65
00:04:11,190 --> 00:04:13,980
The model kind of takes
a stand on these things,

66
00:04:13,980 --> 00:04:16,980
and in particular,
the wage, when

67
00:04:16,980 --> 00:04:20,700
you have financial
deepening, can go up,

68
00:04:20,700 --> 00:04:25,980
and that is an enormous benefit
to poor people, even unskilled

69
00:04:25,980 --> 00:04:29,490
poor people who have better
employment and better wage

70
00:04:29,490 --> 00:04:30,690
opportunities.

71
00:04:30,690 --> 00:04:35,570
The second welfare experiment
is when, in the second model,

72
00:04:35,570 --> 00:04:39,090
the government takes
over the banking system.

73
00:04:39,090 --> 00:04:42,540
I alluded to this at
the end of last time,

74
00:04:42,540 --> 00:04:45,395
and again, when they
liberalized this system

75
00:04:45,395 --> 00:04:48,750
and turned it back over
to the commercial sector,

76
00:04:48,750 --> 00:04:54,070
that also creates welfare
gains, which can be quite large,

77
00:04:54,070 --> 00:04:55,290
and we can--

78
00:04:55,290 --> 00:04:58,510
the point is, we can use
each of these models,

79
00:04:58,510 --> 00:05:00,840
if we believe the
model and we believe

80
00:05:00,840 --> 00:05:04,200
more or less the numbers we
put in for the parameters,

81
00:05:04,200 --> 00:05:06,570
we can quantify these
welfare gains and losses.

82
00:05:11,530 --> 00:05:15,165
There's some corollaries
that come out of this.

83
00:05:17,720 --> 00:05:19,850
First of all, growth
doesn't necessarily

84
00:05:19,850 --> 00:05:23,810
have to go up as the
financial system expands.

85
00:05:23,810 --> 00:05:25,940
We reviewed some
literature last time

86
00:05:25,940 --> 00:05:30,880
about finance causes growth that
seemed plausible and reasonable

87
00:05:30,880 --> 00:05:34,070
and at least at a
reduced form since it

88
00:05:34,070 --> 00:05:37,280
seemed to fit the data.

89
00:05:37,280 --> 00:05:41,720
But in the second model,
as opportunities expand,

90
00:05:41,720 --> 00:05:45,800
you can create endogenously
more financial infrastructure,

91
00:05:45,800 --> 00:05:47,800
but it comes at a cost.

92
00:05:47,800 --> 00:05:51,020
And just like anything
else, investment not

93
00:05:51,020 --> 00:05:53,700
may not pay off right away.

94
00:05:53,700 --> 00:05:55,550
So some resources,
quote-unquote,

95
00:05:55,550 --> 00:05:59,420
disappear in this creation
of financial infrastructure,

96
00:05:59,420 --> 00:06:03,140
and so in effect, the
growth rate levels go down

97
00:06:03,140 --> 00:06:06,110
and growth rates are
slower for a while

98
00:06:06,110 --> 00:06:10,520
than they would have been
without the liberalization.

99
00:06:10,520 --> 00:06:13,520
Another caveat
coming from the model

100
00:06:13,520 --> 00:06:17,630
is, both models today
are transition models--

101
00:06:17,630 --> 00:06:21,050
that is to say, they are
not yet in steady-state,

102
00:06:21,050 --> 00:06:24,500
although if I remember,
I'll try to say something

103
00:06:24,500 --> 00:06:27,990
about the steady states
of each of these.

104
00:06:27,990 --> 00:06:30,920
And so in some
respects, the models

105
00:06:30,920 --> 00:06:35,090
are not generating
stationary time series.

106
00:06:35,090 --> 00:06:40,580
Implicitly when you run a
regression, OLS and so on,

107
00:06:40,580 --> 00:06:43,610
you're kind of assuming
you're getting data drawn

108
00:06:43,610 --> 00:06:46,895
from as stationary time series.

109
00:06:49,620 --> 00:06:54,320
So we run regressions on some
of these model-generated data

110
00:06:54,320 --> 00:06:58,130
where we know exactly what's
going on, and sure enough,

111
00:06:58,130 --> 00:07:02,870
you get kind of spurious,
bogus, or erratic coefficients

112
00:07:02,870 --> 00:07:05,840
on the very same regressions
that the literature has

113
00:07:05,840 --> 00:07:09,480
been running with actual
cross-country data.

114
00:07:09,480 --> 00:07:12,972
So it's sort of a caution.

115
00:07:12,972 --> 00:07:20,330
Now toward the end, there's
a tendency to, in any class,

116
00:07:20,330 --> 00:07:24,650
any lecture for any course,
you see models and you'll see,

117
00:07:24,650 --> 00:07:28,550
oh, there's a model, let's
learn how to manipulate that one

118
00:07:28,550 --> 00:07:31,640
and maybe do a few
extensions of it,

119
00:07:31,640 --> 00:07:35,090
or a second class
with a second model.

120
00:07:35,090 --> 00:07:36,920
Toward the end, I
will deliberately

121
00:07:36,920 --> 00:07:41,000
try to engage you in a
comparison of these two models

122
00:07:41,000 --> 00:07:44,720
to make the point that the
real research that goes on

123
00:07:44,720 --> 00:07:48,770
is trying to decide not just
how to an extent a given model,

124
00:07:48,770 --> 00:07:51,560
but sort of what
we learn overall.

125
00:07:51,560 --> 00:07:55,040
It's not typically published,
this stuff, actually.

126
00:07:55,040 --> 00:07:58,640
Mostly these conversations take
place in offices and hallways

127
00:07:58,640 --> 00:08:05,870
or over beer and so on, but
it's actually a big part

128
00:08:05,870 --> 00:08:08,540
of the research process.

129
00:08:08,540 --> 00:08:16,710
So as an introduction overall,
a few words from a paper I

130
00:08:16,710 --> 00:08:28,490
wrote with Hyeok
Jeong, which repeats

131
00:08:28,490 --> 00:08:29,630
all what I've said already.

132
00:08:29,630 --> 00:08:32,600
Basically we're not grabbing
parameters out of thin air,

133
00:08:32,600 --> 00:08:36,200
we're going to estimate
them with the micro data.

134
00:08:36,200 --> 00:08:38,600
We compare the
predictions of the model

135
00:08:38,600 --> 00:08:39,739
with the actual dynamics.

136
00:08:44,179 --> 00:08:50,390
There is a well-established
tradition in other sciences

137
00:08:50,390 --> 00:08:54,200
that has to do with
model validation.

138
00:08:54,200 --> 00:08:58,310
So it's quite common
to basically calibrate

139
00:08:58,310 --> 00:09:03,620
the machine or the model
or the process, and then

140
00:09:03,620 --> 00:09:06,950
sort of simulate and
see how well it works

141
00:09:06,950 --> 00:09:10,455
without being recalibrated.

142
00:09:10,455 --> 00:09:12,830
Or put another way, if you
have enough degrees of freedom

143
00:09:12,830 --> 00:09:16,490
and enough parameters and
you're constantly tinkering,

144
00:09:16,490 --> 00:09:20,230
there's a tendency to
be able to fit stuff.

145
00:09:20,230 --> 00:09:25,150
You've seen this probably in
your first course in statistics

146
00:09:25,150 --> 00:09:26,830
and ordinary least squares.

147
00:09:26,830 --> 00:09:29,530
If you run out of degrees of--

148
00:09:29,530 --> 00:09:35,580
you can fit-- align two points,
you can perfectly well--

149
00:09:35,580 --> 00:09:42,580
or you can fit a polynomial
even with nonlinear data,

150
00:09:42,580 --> 00:09:46,090
but then out of sample, you
want to predict some more

151
00:09:46,090 --> 00:09:49,460
and it's just crazy,
it's just completely off.

152
00:09:49,460 --> 00:09:52,180
So here, we followed the
tradition of kind of not--

153
00:09:52,180 --> 00:09:56,530
deliberately not
using all of the data.

154
00:09:56,530 --> 00:10:00,040
It is a matter of choice,
though, what data to use

155
00:10:00,040 --> 00:10:04,570
and what to set aside, and
I could imagine more, quote,

156
00:10:04,570 --> 00:10:06,880
experiments than
we actually did.

157
00:10:06,880 --> 00:10:10,150
So there isn't a fixed
rule for how to do this,

158
00:10:10,150 --> 00:10:13,090
and it's not even as
commonly used in economics

159
00:10:13,090 --> 00:10:16,090
as I think it
probably ought to be.

160
00:10:22,770 --> 00:10:24,850
And I guess you
know the main theme

161
00:10:24,850 --> 00:10:32,460
today is this marriage
of theory and data.

162
00:10:32,460 --> 00:10:37,690
Frisch very early on in
the '30s was a big champion

163
00:10:37,690 --> 00:10:41,860
of the mutual penetration of
quantitative economic theory

164
00:10:41,860 --> 00:10:44,920
with statistical observations.

165
00:10:44,920 --> 00:10:52,000
I think that's actually on
the econometric cover, if not

166
00:10:52,000 --> 00:10:53,500
their website.

167
00:10:53,500 --> 00:11:00,210
And in particular, we're
somewhere between theory

168
00:11:00,210 --> 00:11:04,520
without measurement and
measurement without theory.

169
00:11:04,520 --> 00:11:08,920
A big controversy in
business cycle analysis

170
00:11:08,920 --> 00:11:13,870
along that line, a big debate
between [INAUDIBLE] and Burns

171
00:11:13,870 --> 00:11:16,240
and Mitchell.

172
00:11:16,240 --> 00:11:17,260
OK.

173
00:11:17,260 --> 00:11:22,390
So we're going to do
these two papers that

174
00:11:22,390 --> 00:11:25,630
came from the literature.

175
00:11:25,630 --> 00:11:28,990
It's kind of a plus and a minus.

176
00:11:28,990 --> 00:11:32,330
One is Lloyd-Ellis
and Dan Bernhardt,

177
00:11:32,330 --> 00:11:35,170
and the other is
Greenwood and Jovanovic.

178
00:11:35,170 --> 00:11:37,750
Each tell a plausible
story about what

179
00:11:37,750 --> 00:11:41,380
lies behind the growth
process and the relationship

180
00:11:41,380 --> 00:11:45,370
between growth and equality,
financial deepening,

181
00:11:45,370 --> 00:11:49,386
and those other
variables that I listed.

182
00:11:49,386 --> 00:11:51,720
And we're going to try
to estimate those models.

183
00:11:51,720 --> 00:11:53,370
Well, the negative
part of this thing

184
00:11:53,370 --> 00:11:57,570
is we learned how to do
statistics and econometrics

185
00:11:57,570 --> 00:11:59,760
in part through the
lens of these models

186
00:11:59,760 --> 00:12:05,160
by realizing how problematic
certain parts of those models

187
00:12:05,160 --> 00:12:06,390
are.

188
00:12:06,390 --> 00:12:11,940
And I would not start this
way if I were starting over.

189
00:12:14,740 --> 00:12:16,340
And as I go through
the lecture, I

190
00:12:16,340 --> 00:12:21,770
will try to point out which
particularly worrisome

191
00:12:21,770 --> 00:12:25,610
or problematic parts I
have in mind in terms

192
00:12:25,610 --> 00:12:27,530
of taking these
models to the data.

193
00:12:27,530 --> 00:12:30,330
But anyway, again, back
on the positive side,

194
00:12:30,330 --> 00:12:31,670
that's literally what we did.

195
00:12:31,670 --> 00:12:37,040
We adopted wholesale,
without any modification

196
00:12:37,040 --> 00:12:40,940
or qualification,
exactly the equations

197
00:12:40,940 --> 00:12:44,766
that these theoretical
authors used in their models.

198
00:12:44,766 --> 00:12:48,240
Uh oh.

199
00:12:48,240 --> 00:12:49,060
That's not good.

200
00:13:05,180 --> 00:13:12,240
So Lloyd-Ellis and
Bernhardt, LEB for short,

201
00:13:12,240 --> 00:13:14,630
there are three occupations--

202
00:13:14,630 --> 00:13:20,840
farmers, workers, and
entrepreneurs, and it's

203
00:13:20,840 --> 00:13:23,780
costly to set up a firm.

204
00:13:23,780 --> 00:13:27,320
But these costs are like talent,
they vary with the individual.

205
00:13:27,320 --> 00:13:31,670
The more talented you are,
the less costly it is.

206
00:13:31,670 --> 00:13:37,190
It's assumed that the cumulative
distribution of talent

207
00:13:37,190 --> 00:13:39,530
is distributed
according to capital H

208
00:13:39,530 --> 00:13:43,040
here where x is
the level of cost.

209
00:13:46,730 --> 00:13:51,680
And you have to be
careful with the notation

210
00:13:51,680 --> 00:13:53,330
where we're talking
about densities

211
00:13:53,330 --> 00:13:56,870
versus cumulative distributions,
so just kind of bear

212
00:13:56,870 --> 00:13:58,270
that in mind.

213
00:13:58,270 --> 00:14:01,370
There's also an initial
distribution of wealth

214
00:14:01,370 --> 00:14:07,910
that starts off the model, G of
b where b is sort of bequest.

215
00:14:07,910 --> 00:14:11,870
Just to alert you, this
model has a relatively naive

216
00:14:11,870 --> 00:14:14,060
dynamic part.

217
00:14:14,060 --> 00:14:17,270
People essentially live for
two periods-- they eat today

218
00:14:17,270 --> 00:14:20,060
and they leave
money for their kids

219
00:14:20,060 --> 00:14:27,260
or their tomorrow's next period
as if leaving a bequest, hence

220
00:14:27,260 --> 00:14:31,520
the notation b for a Bequest.

221
00:14:31,520 --> 00:14:35,780
So what earnings two people
have in a given period?

222
00:14:35,780 --> 00:14:41,640
Farmers earn kind of a
subsistence living in farming,

223
00:14:41,640 --> 00:14:44,270
and they carry their
wealth over and have it

224
00:14:44,270 --> 00:14:46,940
at the end of the period.

225
00:14:46,940 --> 00:14:54,890
Workers earn wages-- excuse me--
and entrepreneurs make profits.

226
00:14:54,890 --> 00:14:57,440
Everybody has this
initial wealth

227
00:14:57,440 --> 00:14:59,720
which can vary
across individuals

228
00:14:59,720 --> 00:15:06,860
carried over to the end of
the period, essentially.

229
00:15:06,860 --> 00:15:10,700
A little bit of an exception
for the entrepreneurs.

230
00:15:10,700 --> 00:15:14,810
When Dan wrote his
model, he kind of

231
00:15:14,810 --> 00:15:17,930
had in mind an earlier
development literature

232
00:15:17,930 --> 00:15:20,240
which makes
agriculture out there

233
00:15:20,240 --> 00:15:26,330
in the rural areas versus
factories and manufacturing

234
00:15:26,330 --> 00:15:26,930
in the cities.

235
00:15:26,930 --> 00:15:32,090
So they had in mind a kind
of cost of living adjustment,

236
00:15:32,090 --> 00:15:37,970
so basically workers
earn their wage,

237
00:15:37,970 --> 00:15:40,730
but they have to pay this
cost of living, which

238
00:15:40,730 --> 00:15:45,920
gives a little bit advantage
to being in the subsistence

239
00:15:45,920 --> 00:15:47,090
sector.

240
00:15:47,090 --> 00:15:53,480
Actually, you can see
it right away down here,

241
00:15:53,480 --> 00:16:00,260
if sort of in equilibrium to
have both farmers and workers

242
00:16:00,260 --> 00:16:06,030
coexisting, these two equations
have to be equal to each other.

243
00:16:06,030 --> 00:16:09,740
So as long as the
wage does not exceed--

244
00:16:09,740 --> 00:16:13,460
the wage less this cost of
living does not exceed gamma,

245
00:16:13,460 --> 00:16:16,040
then a household
would be indifferent

246
00:16:16,040 --> 00:16:19,430
about whether to migrate
to the city for a wage

247
00:16:19,430 --> 00:16:22,830
and pay that cost
versus stay in farming.

248
00:16:22,830 --> 00:16:23,330
Yes?

249
00:16:23,330 --> 00:16:24,525
AUDIENCE: So is this
going to be abstracting

250
00:16:24,525 --> 00:16:25,945
all the way from land markets?

251
00:16:25,945 --> 00:16:27,070
ROBERT TOWNSEND: Yes it is.

252
00:16:33,380 --> 00:16:37,170
And-- OK, let's get to the firm.

253
00:16:37,170 --> 00:16:39,940
Well, let me just
follow up on that.

254
00:16:39,940 --> 00:16:43,230
The model as estimated
and calibrated

255
00:16:43,230 --> 00:16:47,340
will actually deliver the fact
that for fairly long periods

256
00:16:47,340 --> 00:16:51,000
of time, the wage
stays relatively low

257
00:16:51,000 --> 00:16:54,600
and you have all three sectors
coexisting, although there will

258
00:16:54,600 --> 00:16:58,650
come this knife-edge time where
the wage starts to pop up,

259
00:16:58,650 --> 00:17:01,230
and then quite
unrealistically there

260
00:17:01,230 --> 00:17:04,140
should be no more farmers.

261
00:17:04,140 --> 00:17:05,700
You would think
this model is doomed

262
00:17:05,700 --> 00:17:08,849
to fail with that
kind of knife edge,

263
00:17:08,849 --> 00:17:15,240
but surprisingly in Thailand,
there are still farmers,

264
00:17:15,240 --> 00:17:23,920
but the wage did stay relatively
flat for quite a long time

265
00:17:23,920 --> 00:17:26,319
and then grew rather quickly.

266
00:17:26,319 --> 00:17:27,290
What about firms?

267
00:17:27,290 --> 00:17:31,270
So firms maximize
basically profit.

268
00:17:31,270 --> 00:17:33,960
The price of the output
or the consumption goods

269
00:17:33,960 --> 00:17:36,130
is normalized to 1.

270
00:17:36,130 --> 00:17:38,950
So firms have output,
which is produced

271
00:17:38,950 --> 00:17:43,060
using capital and labor.

272
00:17:43,060 --> 00:17:44,380
Labor can be hired.

273
00:17:44,380 --> 00:17:48,310
Those are the workers, you
have to pay them a wage.

274
00:17:48,310 --> 00:17:57,470
And capital, basically, is
coming out of initial wealth.

275
00:17:57,470 --> 00:18:00,680
So capital and consumption
are kind of 1 to 1.

276
00:18:03,230 --> 00:18:07,700
So you have your sort
of initial wealth,

277
00:18:07,700 --> 00:18:10,850
and the capital has to
be financed by that.

278
00:18:10,850 --> 00:18:15,410
Capital cannot
exceed that amount.

279
00:18:15,410 --> 00:18:17,150
And actually, there's
this fixed cost,

280
00:18:17,150 --> 00:18:20,300
and that also has to get paid.

281
00:18:20,300 --> 00:18:22,910
There's a lot going
on right here,

282
00:18:22,910 --> 00:18:26,330
because there's two constraints.

283
00:18:26,330 --> 00:18:28,950
One is you have to
pay x to get started,

284
00:18:28,950 --> 00:18:35,300
and the second is that
you can't borrow at all.

285
00:18:35,300 --> 00:18:39,050
All the investment is
basically self-financed.

286
00:18:39,050 --> 00:18:42,950
Now that said, you could imagine
kind of weakening this a bit

287
00:18:42,950 --> 00:18:45,500
and allowing the amount
you could be borrowing

288
00:18:45,500 --> 00:18:52,020
to be some multiple of
a b, and next class,

289
00:18:52,020 --> 00:18:54,410
we'll talk about
many models that

290
00:18:54,410 --> 00:18:57,200
do something exactly like that.

291
00:18:59,760 --> 00:19:01,850
But for now, no credit at all.

292
00:19:04,900 --> 00:19:07,410
So this just-- these costs
just subtract right off

293
00:19:07,410 --> 00:19:12,060
of output as if at
the end of the period,

294
00:19:12,060 --> 00:19:15,900
or b minus these
things is what you

295
00:19:15,900 --> 00:19:18,862
have at the end of the period.

296
00:19:18,862 --> 00:19:21,195
Then you get you get the
output and you pay the workers.

297
00:19:26,810 --> 00:19:29,570
So the key in this
paper and so many

298
00:19:29,570 --> 00:19:34,580
other papers is an occupation
choice map that really

299
00:19:34,580 --> 00:19:44,890
summarizes the possibilities.

300
00:19:44,890 --> 00:19:49,050
So the two key variables from
a household's point of view

301
00:19:49,050 --> 00:19:50,760
is the initial
wealth that they have

302
00:19:50,760 --> 00:19:52,960
at the beginning of the period--

303
00:19:52,960 --> 00:19:54,930
it could be high or low--

304
00:19:54,930 --> 00:20:00,830
and the setup cost x
to running a business.

305
00:20:00,830 --> 00:20:06,120
And naturally, if your wealth
is high and your costs are low,

306
00:20:06,120 --> 00:20:10,350
you're a talented person,
you're, say, not only

307
00:20:10,350 --> 00:20:15,000
likely to go into business, but
that finance constraint is not

308
00:20:15,000 --> 00:20:17,470
going to be binding either.

309
00:20:17,470 --> 00:20:22,450
You're going to run the marginal
product of capital down to 1,

310
00:20:22,450 --> 00:20:24,402
and you still have
more money left over,

311
00:20:24,402 --> 00:20:26,610
so you kind of carry it off
to the end of the period.

312
00:20:29,420 --> 00:20:34,320
There is a region
of lower wealth,

313
00:20:34,320 --> 00:20:36,420
but yet somewhat
talented people who

314
00:20:36,420 --> 00:20:39,150
are constrained
entrepreneurs, they

315
00:20:39,150 --> 00:20:44,330
would actually like to employ
higher levels of capital k

316
00:20:44,330 --> 00:20:48,450
than they are able to finance
out of their own wealth.

317
00:20:48,450 --> 00:20:52,110
They would borrow if they could,
but there's no credit yet.

318
00:20:52,110 --> 00:20:57,960
And actually, this straight
line is when they can't even

319
00:20:57,960 --> 00:21:00,460
cover their setup
costs, or on the margin,

320
00:21:00,460 --> 00:21:03,630
they're just able to
cover their setup cost

321
00:21:03,630 --> 00:21:07,240
x out of their wealth b, and
there's nothing left over,

322
00:21:07,240 --> 00:21:12,150
so they can't even run the
firm on the intensive margin.

323
00:21:12,150 --> 00:21:14,970
And then the rest
of the guys, they're

324
00:21:14,970 --> 00:21:20,360
kind of workers or
agriculturalists.

325
00:21:20,360 --> 00:21:26,590
So the shape of this is
intuitively reasonable, right?

326
00:21:26,590 --> 00:21:28,280
The higher is
wealth and the lower

327
00:21:28,280 --> 00:21:31,910
is talent, the more likely you'd
be a firm or an unconstrained

328
00:21:31,910 --> 00:21:32,870
for a more--

329
00:21:32,870 --> 00:21:38,000
the lower is your wealth
and the more likely

330
00:21:38,000 --> 00:21:39,710
you are to be a worker.

331
00:21:39,710 --> 00:21:44,720
And the thing is, even
when your talent is low,

332
00:21:44,720 --> 00:21:48,650
if your wealth is quite low,
you can't be in business.

333
00:21:48,650 --> 00:21:51,770
These are potentially
profitable entrepreneurs

334
00:21:51,770 --> 00:21:55,100
who are constrained on
the extensive margin

335
00:21:55,100 --> 00:21:57,260
at these parameter values.

336
00:21:57,260 --> 00:21:59,420
Now you say, what
parameter values?

337
00:21:59,420 --> 00:22:02,390
Well, I drew some lines.

338
00:22:02,390 --> 00:22:05,630
The point of the
micro estimation

339
00:22:05,630 --> 00:22:11,990
is to actually be able
to draw those lines based

340
00:22:11,990 --> 00:22:13,130
on the micro data.

341
00:22:16,820 --> 00:22:18,800
So-- oh my goodness.

342
00:22:25,408 --> 00:22:27,200
I don't know how often
this is going to be.

343
00:22:32,270 --> 00:22:40,760
So let me just grab
three equations

344
00:22:40,760 --> 00:22:44,570
and let you focus on this.

345
00:22:44,570 --> 00:22:47,960
Here are the functions,
the structural functions

346
00:22:47,960 --> 00:22:49,220
of the model.

347
00:22:49,220 --> 00:22:52,160
There is a trade-off between
consumption and bequests

348
00:22:52,160 --> 00:22:54,380
under this utility function.

349
00:22:54,380 --> 00:22:57,170
It's basically top Douglas
in today's consumption

350
00:22:57,170 --> 00:22:58,910
and tomorrow's wealth.

351
00:22:58,910 --> 00:23:01,580
This little parameter
here is omega.

352
00:23:01,580 --> 00:23:06,290
Omega is basically going to
determine the savings rate.

353
00:23:06,290 --> 00:23:08,810
And we're going to
have to calibrate it.

354
00:23:11,420 --> 00:23:15,800
This production
function is ugly.

355
00:23:15,800 --> 00:23:20,690
It was meant to be
general, which sounds good.

356
00:23:20,690 --> 00:23:22,910
So it has capital,
it has labor, it

357
00:23:22,910 --> 00:23:26,570
has the square of each of them,
and it has an interaction term.

358
00:23:26,570 --> 00:23:29,270
Unfortunate-- so
it's like translog.

359
00:23:29,270 --> 00:23:32,930
Approximately-- it can
approximate almost any function

360
00:23:32,930 --> 00:23:35,350
reasonably well,
but the problem is,

361
00:23:35,350 --> 00:23:38,450
it carries with it
five parameters.

362
00:23:38,450 --> 00:23:48,883
Alpha, beta, xi, rho, and an
unreadable script, I'm sorry.

363
00:23:48,883 --> 00:23:51,175
I thought it was sigma, but
it doesn't look like sigma.

364
00:23:54,640 --> 00:24:00,940
And this is one of
the regrets that I

365
00:24:00,940 --> 00:24:03,940
have about pushing so
hard in these models,

366
00:24:03,940 --> 00:24:08,540
because it turns
out in hindsight,

367
00:24:08,540 --> 00:24:14,590
we generated data from
that production function

368
00:24:14,590 --> 00:24:16,230
knowing the model,
and it's still

369
00:24:16,230 --> 00:24:17,910
hard to recover
these parameters.

370
00:24:17,910 --> 00:24:22,440
That's something you should
get into the habit of doing.

371
00:24:22,440 --> 00:24:26,550
It's hard enough with
real data, and you already

372
00:24:26,550 --> 00:24:30,168
know, the actual world
is not like your model,

373
00:24:30,168 --> 00:24:31,710
it's an abstraction,
but at least you

374
00:24:31,710 --> 00:24:33,200
ought to make sure that--

375
00:24:33,200 --> 00:24:35,910
and that you can't
really do too,

376
00:24:35,910 --> 00:24:38,640
too much about, because
you have to make decisions

377
00:24:38,640 --> 00:24:39,995
about your model.

378
00:24:39,995 --> 00:24:41,370
But at least you
should make sure

379
00:24:41,370 --> 00:24:44,910
your estimation routine
works well if the model--

380
00:24:44,910 --> 00:24:47,190
if the world were
exactly like the model

381
00:24:47,190 --> 00:24:51,550
and generated the data
that you're about to use.

382
00:24:51,550 --> 00:24:55,020
Anyway, we did the
best we could on that.

383
00:24:55,020 --> 00:25:00,030
And this is this
distribution of setup costs,

384
00:25:00,030 --> 00:25:02,220
the cumulative distribution.

385
00:25:02,220 --> 00:25:05,340
The easiest way to
disentangle this is to say,

386
00:25:05,340 --> 00:25:09,660
suppose m is equal to
0, then this disappears,

387
00:25:09,660 --> 00:25:15,300
and the cumulative distribution
is just rising linearly in x.

388
00:25:15,300 --> 00:25:18,510
In other words, the
density is flat.

389
00:25:18,510 --> 00:25:22,680
So this is just a standard
uniform distribution.

390
00:25:22,680 --> 00:25:29,860
That's also kind of
nice on the diagram,

391
00:25:29,860 --> 00:25:33,880
because although I don't want
to enlarge this, basically

392
00:25:33,880 --> 00:25:37,120
you want to know how many
people of a certain x

393
00:25:37,120 --> 00:25:42,010
there are given a certain b
how many people of a certain x

394
00:25:42,010 --> 00:25:46,260
there are before you go
from firms to workers.

395
00:25:46,260 --> 00:25:48,940
And that's just the
length of that line,

396
00:25:48,940 --> 00:25:59,100
and so a uniform density is
convenient for the sort of--

397
00:25:59,100 --> 00:26:02,800
the geometry of that figure.

398
00:26:02,800 --> 00:26:05,590
However, if m is you
know more than 0,

399
00:26:05,590 --> 00:26:07,360
then you get that squared term.

400
00:26:07,360 --> 00:26:12,880
It just means the distribution
of talent is skewed to the--

401
00:26:12,880 --> 00:26:15,620
really, the distribution of
cost is skewed to the right,

402
00:26:15,620 --> 00:26:19,870
so there are some very
untalented people,

403
00:26:19,870 --> 00:26:23,440
and conversely
when m is below 0.

404
00:26:23,440 --> 00:26:28,180
Now all this big to-do
about micro underpinnings,

405
00:26:28,180 --> 00:26:30,760
what do we see in the data?

406
00:26:30,760 --> 00:26:34,290
Well let's imagine
we see the wage,

407
00:26:34,290 --> 00:26:37,320
and we also see sort
of the net worth

408
00:26:37,320 --> 00:26:44,790
or wealth of the household,
and we know their occupation.

409
00:26:44,790 --> 00:26:49,160
So what does the model predict?

410
00:26:49,160 --> 00:26:55,030
The model predicts that the
occupation is, say, firm--

411
00:26:55,030 --> 00:26:59,170
or yi equals 1 in
a binary sense--

412
00:26:59,170 --> 00:27:01,730
when that distribution
of cost is--

413
00:27:01,730 --> 00:27:04,600
this is-- you should
read this as an x--

414
00:27:04,600 --> 00:27:11,520
is on or below that partition
between workers and firms.

415
00:27:11,520 --> 00:27:15,970
So you add up, as I was just
doing, all those households

416
00:27:15,970 --> 00:27:18,200
with talent less than that.

417
00:27:18,200 --> 00:27:21,790
Likewise, if you're not
going to be yi equal to 1,

418
00:27:21,790 --> 00:27:24,820
you're going to be yi equal
to 0 and a wage earner,

419
00:27:24,820 --> 00:27:27,190
and that's the rest
of the residual mass

420
00:27:27,190 --> 00:27:29,110
or 1 minus this probability.

421
00:27:32,210 --> 00:27:34,900
So what is the likelihood?

422
00:27:34,900 --> 00:27:40,780
So just quick review,
imagine you have a coin,

423
00:27:40,780 --> 00:27:44,980
and usually coins
are fair coins, so--

424
00:27:44,980 --> 00:27:48,310
I'm not sure about
the Super Bowl--

425
00:27:48,310 --> 00:27:51,520
equal probability of
being heads or tails,

426
00:27:51,520 --> 00:27:57,490
and you see a bunch of
data, 4 heads and 6 tails,

427
00:27:57,490 --> 00:27:59,600
and you say, well what
were the odds of that?

428
00:27:59,600 --> 00:28:01,540
Actually, it's a
bit more interesting

429
00:28:01,540 --> 00:28:06,370
if the coin is not
fair, because then it's

430
00:28:06,370 --> 00:28:10,210
not probability 1/2 for
everything that turns up.

431
00:28:10,210 --> 00:28:13,150
But you just basically
take the data,

432
00:28:13,150 --> 00:28:16,240
which is heads, tails, tails,
heads, and so on, or a worker,

433
00:28:16,240 --> 00:28:21,470
firm, firm, worker, and start
adding up the probabilities.

434
00:28:21,470 --> 00:28:22,595
They're independent events.

435
00:28:25,770 --> 00:28:29,360
So this says log here,
but if you undid the log,

436
00:28:29,360 --> 00:28:34,790
you would have the
probability of being a firm,

437
00:28:34,790 --> 00:28:38,960
and then this yi would
just be the 0 or 1,

438
00:28:38,960 --> 00:28:40,520
and when it's equal
to 1, you're just

439
00:28:40,520 --> 00:28:43,760
getting the number of, say,
heads or the number of firms

440
00:28:43,760 --> 00:28:48,675
in the sample, and likewise
for the number of workers.

441
00:28:48,675 --> 00:28:50,300
It's very convenient
to take the logs--

442
00:28:52,940 --> 00:28:57,403
and it takes this form because
logs bring the exponent down

443
00:28:57,403 --> 00:28:58,445
in front of the variable.

444
00:29:04,410 --> 00:29:06,090
It's a cute trick,
too, because you want

445
00:29:06,090 --> 00:29:08,460
to maximize the likelihood.

446
00:29:08,460 --> 00:29:11,828
Logs and levels or
monotone in each other,

447
00:29:11,828 --> 00:29:13,370
you find the max of
one, you're going

448
00:29:13,370 --> 00:29:14,790
to find the max of the other.

449
00:29:14,790 --> 00:29:19,380
So econometricians
usually maximize logs

450
00:29:19,380 --> 00:29:23,320
for reasons like this.

451
00:29:23,320 --> 00:29:27,030
And it turns out with
all that algebra,

452
00:29:27,030 --> 00:29:30,450
those particular
functional forms,

453
00:29:30,450 --> 00:29:33,630
for any set of parameters
in the production function--

454
00:29:33,630 --> 00:29:36,750
alpha, beta, rho, sigma,
xi, et cetera, right?

455
00:29:36,750 --> 00:29:41,620
You can draw all those
lines, and therefore,

456
00:29:41,620 --> 00:29:44,570
determine all these
probabilities.

457
00:29:44,570 --> 00:29:46,930
So for a given guess
about the parameters,

458
00:29:46,930 --> 00:29:50,330
you'll have a certain value
for the log likelihood,

459
00:29:50,330 --> 00:29:54,580
and then you just maximize
that log likelihood

460
00:29:54,580 --> 00:29:55,910
by choosing those parameters.

461
00:29:55,910 --> 00:29:59,170
Essentially the data come
first there on the diagram,

462
00:29:59,170 --> 00:30:01,210
you have sort of wealth--

463
00:30:01,210 --> 00:30:04,060
thousands of data points,
but you have wealth.

464
00:30:04,060 --> 00:30:10,090
And you don't see talent,
that's an unobservable,

465
00:30:10,090 --> 00:30:12,100
but you do see their occupation.

466
00:30:12,100 --> 00:30:13,600
So you're moving
those curves around

467
00:30:13,600 --> 00:30:16,520
to try to fit as many
data points as possible.

468
00:30:16,520 --> 00:30:19,190
Here are the
estimated parameters.

469
00:30:19,190 --> 00:30:22,300
Alpha, beta, xi, rho,
sigma, et cetera, and it's

470
00:30:22,300 --> 00:30:25,702
also estimated
naturally, because that's

471
00:30:25,702 --> 00:30:27,160
kind of telling
you whether there's

472
00:30:27,160 --> 00:30:32,356
a large mass of high or low
cost people in the population.

473
00:30:32,356 --> 00:30:34,921
AUDIENCE: So the line of
matter is just the boundary

474
00:30:34,921 --> 00:30:36,870
between [INAUDIBLE] right?

475
00:30:36,870 --> 00:30:40,660
Divide [INAUDIBLE]
doesn't matter for--

476
00:30:40,660 --> 00:30:42,660
ROBERT TOWNSEND: Not in
this case, no we didn't.

477
00:30:42,660 --> 00:30:46,360
Now if I had from the data
households answering-- or firms

478
00:30:46,360 --> 00:30:50,190
answering the question,
are we constrained?

479
00:30:50,190 --> 00:30:52,560
And you could even get
more sophisticated.

480
00:30:52,560 --> 00:30:54,150
Running businesses,
would we like

481
00:30:54,150 --> 00:30:58,110
to invest more given
this shadow price?

482
00:30:58,110 --> 00:31:00,040
I've asked questions like that.

483
00:31:00,040 --> 00:31:03,210
Alp in the macro
lecture half an hour

484
00:31:03,210 --> 00:31:09,460
ago was describing something
like that in the US for firms

485
00:31:09,460 --> 00:31:11,070
right before the--

486
00:31:11,070 --> 00:31:13,470
during the financial crisis.

487
00:31:13,470 --> 00:31:16,440
And you could even do something
on the extensive margin

488
00:31:16,440 --> 00:31:18,360
in principle.

489
00:31:18,360 --> 00:31:21,930
You're worried that those are
lousy questions, like are you

490
00:31:21,930 --> 00:31:22,710
constrained?

491
00:31:26,040 --> 00:31:27,060
But anyway.

492
00:31:27,060 --> 00:31:30,180
It seems to actually
work better in practice

493
00:31:30,180 --> 00:31:33,720
than you think it
might in reality--

494
00:31:33,720 --> 00:31:35,110
in the classroom.

495
00:31:35,110 --> 00:31:40,170
Calibration, not everything can
be gotten from those occupation

496
00:31:40,170 --> 00:31:42,490
choice diagrams.

497
00:31:42,490 --> 00:31:46,950
So-- and actually, we changed
the model just slightly.

498
00:31:46,950 --> 00:31:49,980
We're going to allow
a very slight growth

499
00:31:49,980 --> 00:31:52,110
rate in subsistence
income as if there

500
00:31:52,110 --> 00:31:55,500
were some technological
progress in agriculture.

501
00:32:00,660 --> 00:32:07,260
That savings rate is not
estimable from the micro data

502
00:32:07,260 --> 00:32:09,870
the way we were doing it, and
neither is the subsistence

503
00:32:09,870 --> 00:32:11,550
income in farming.

504
00:32:11,550 --> 00:32:13,470
The level of
subsistence income we

505
00:32:13,470 --> 00:32:18,870
get from a socioeconomic survey
done in Thailand in 1976,

506
00:32:18,870 --> 00:32:24,740
basically matching to
the reservation wage,

507
00:32:24,740 --> 00:32:26,150
there is a slight growth--

508
00:32:26,150 --> 00:32:30,480
very small trend over
time in that wage rate,

509
00:32:30,480 --> 00:32:35,450
so we let it grow exogenously
at 5.5% per year--

510
00:32:35,450 --> 00:32:36,020
less than 1%.

511
00:32:41,490 --> 00:32:44,280
And the Cobb-Douglas says
the savings rate-- and again,

512
00:32:44,280 --> 00:32:45,420
we're going to get--

513
00:32:45,420 --> 00:32:51,690
we're going to set it high,
basically, at basically 1/4,

514
00:32:51,690 --> 00:32:55,390
but that matches evidence--
and you'll see this in China

515
00:32:55,390 --> 00:32:57,390
and many of the countries
you're about to study,

516
00:32:57,390 --> 00:33:01,130
the savings rates are very
high in these countries.

517
00:33:01,130 --> 00:33:02,630
So that's actually realistic.

518
00:33:10,650 --> 00:33:11,610
Learning by doing.

519
00:33:15,750 --> 00:33:24,730
So from the original Bernhardt
and Lloyd-Ellis paper,

520
00:33:24,730 --> 00:33:26,800
you begin to see a little
bit of the dynamics

521
00:33:26,800 --> 00:33:29,710
and also their take on
the development literature

522
00:33:29,710 --> 00:33:32,050
at the time they
wrote the paper.

523
00:33:36,040 --> 00:33:41,320
First of all, why that
myopic savings rate?

524
00:33:41,320 --> 00:33:44,830
Why just care about
today and then tomorrow?

525
00:33:44,830 --> 00:33:48,430
Well, it just helps enormously
in computing the solutions

526
00:33:48,430 --> 00:33:49,970
to these models.

527
00:33:49,970 --> 00:33:52,510
It's like every
day is a new day.

528
00:33:52,510 --> 00:33:54,520
Here's your wealth.

529
00:33:54,520 --> 00:33:55,810
How much am I going to save?

530
00:33:55,810 --> 00:33:58,570
That's a number, that's the
fraction of wealth you're

531
00:33:58,570 --> 00:34:01,450
going to save, and the
rest of it you know you've

532
00:34:01,450 --> 00:34:03,730
got available for consumption.

533
00:34:03,730 --> 00:34:07,480
So after you make your
occupation decision.

534
00:34:07,480 --> 00:34:09,940
So every period,
you kind of have

535
00:34:09,940 --> 00:34:13,360
like a static model where you're
deciding on the occupation.

536
00:34:13,360 --> 00:34:15,580
How well you do, et
cetera, depending

537
00:34:15,580 --> 00:34:17,469
on your occupation
choice and your talent,

538
00:34:17,469 --> 00:34:20,889
determines your end-of-period
wealth, and that you carry over

539
00:34:20,889 --> 00:34:22,469
to tomorrow.

540
00:34:22,469 --> 00:34:25,630
And tomorrow's a new day, the
whole thing starts over again.

541
00:34:25,630 --> 00:34:29,230
No, I don't believe these
guys only live two periods.

542
00:34:29,230 --> 00:34:31,420
Although you will
see, this tension

543
00:34:31,420 --> 00:34:35,469
in the literature between
realism and tractability,

544
00:34:35,469 --> 00:34:39,429
people still use quite heavily
overlapping generations models,

545
00:34:39,429 --> 00:34:41,199
and some--

546
00:34:41,199 --> 00:34:42,850
and I mean like
with three periods.

547
00:34:45,730 --> 00:34:49,239
So it's kind of hard to match
year-to-year data that way,

548
00:34:49,239 --> 00:34:52,090
but you can tell
interesting stories

549
00:34:52,090 --> 00:34:54,760
about the distribution of
wealth and the impact of credit

550
00:34:54,760 --> 00:34:57,610
constraints.

551
00:34:57,610 --> 00:35:00,010
So this is an extreme
version of that.

552
00:35:00,010 --> 00:35:02,860
This is like Rostow's
take-off, you remember?

553
00:35:02,860 --> 00:35:05,170
He worked for Kennedy and--

554
00:35:05,170 --> 00:35:09,940
so this-- he says at first
nothing much is happening,

555
00:35:09,940 --> 00:35:12,670
and then you get
in this transition,

556
00:35:12,670 --> 00:35:16,750
wages begin to rise, but
income and wealth are still

557
00:35:16,750 --> 00:35:18,190
growing first
order, and then you

558
00:35:18,190 --> 00:35:21,820
get the advanced economic
development and wages take off,

559
00:35:21,820 --> 00:35:27,310
and incomes and wealth
basically really, really become

560
00:35:27,310 --> 00:35:29,080
quite high.

561
00:35:29,080 --> 00:35:31,540
And I'll forget to
say it, but if you

562
00:35:31,540 --> 00:35:33,420
let this model run
long enough, it

563
00:35:33,420 --> 00:35:37,540
will converge to a steady
state in which there's

564
00:35:37,540 --> 00:35:40,060
no more any growth in wealth.

565
00:35:40,060 --> 00:35:43,320
It's no longer a growth economy.

566
00:35:43,320 --> 00:35:47,170
And that steady state, because
of this weird assumption

567
00:35:47,170 --> 00:35:51,460
about two periods
and these costs,

568
00:35:51,460 --> 00:35:55,380
you will see people
making transitions.

569
00:35:55,380 --> 00:35:57,180
They will have a certain
wealth and they're

570
00:35:57,180 --> 00:36:00,150
going to be talented or not
talented in certain things,

571
00:36:00,150 --> 00:36:02,280
and they will decide
to, say, forget

572
00:36:02,280 --> 00:36:05,440
what they were doing last period
and do something new today.

573
00:36:05,440 --> 00:36:09,000
So you'll see a lot of churning
in the occupation distribution,

574
00:36:09,000 --> 00:36:11,010
but you will not see
any trends in anything

575
00:36:11,010 --> 00:36:13,010
in their steady state.

576
00:36:13,010 --> 00:36:14,760
This is going to be a
recurrent theme when

577
00:36:14,760 --> 00:36:17,760
we go through these papers as
to which ones are essentially

578
00:36:17,760 --> 00:36:23,490
in steady state, and maybe we're
looking across a cross-section

579
00:36:23,490 --> 00:36:28,110
of economies that vary in some
way, each of which country

580
00:36:28,110 --> 00:36:30,450
is in a steady state
versus transitions.

581
00:36:30,450 --> 00:36:33,060
Today is about
transitions, and I'm not

582
00:36:33,060 --> 00:36:35,730
going to spend a
lot of attention

583
00:36:35,730 --> 00:36:38,040
to this steady state of this
model, which is probably

584
00:36:38,040 --> 00:36:42,040
relatively uninteresting.

585
00:36:42,040 --> 00:36:47,640
OK so now we get to the
financial deepening part.

586
00:36:50,230 --> 00:36:56,380
So what we do is create an
exogenous financial sector

587
00:36:56,380 --> 00:36:58,300
under which everybody's
intermediated.

588
00:37:01,400 --> 00:37:04,880
It's like one of
these dual economies,

589
00:37:04,880 --> 00:37:09,120
except for us, it's like you
have free banks and access

590
00:37:09,120 --> 00:37:11,030
or you have zero access.

591
00:37:11,030 --> 00:37:13,210
It's very extreme.

592
00:37:13,210 --> 00:37:15,160
In the banking
sector, so to speak,

593
00:37:15,160 --> 00:37:18,560
it's not just about
credit, it's about lending.

594
00:37:18,560 --> 00:37:21,220
So there's an equilibrium
interest rate that's

595
00:37:21,220 --> 00:37:24,730
going to clear the market.

596
00:37:24,730 --> 00:37:28,330
More savings is
equal to borrowing.

597
00:37:28,330 --> 00:37:32,590
And now you look at
the return to farmers,

598
00:37:32,590 --> 00:37:34,918
farmers have their
subsistence income,

599
00:37:34,918 --> 00:37:36,460
but they put their
money in the bank,

600
00:37:36,460 --> 00:37:40,800
so they get 1 plus little r or
big R rate of return on that.

601
00:37:40,800 --> 00:37:45,610
Everybody-- when
you have complete--

602
00:37:45,610 --> 00:37:49,560
there's no uncertainty
here, essentially.

603
00:37:49,560 --> 00:37:51,850
So you have a
neoclassical separation

604
00:37:51,850 --> 00:37:57,760
between households and firms,
which essentially means

605
00:37:57,760 --> 00:37:59,990
the financing decisions
of firms have nothing

606
00:37:59,990 --> 00:38:01,350
to do with households.

607
00:38:01,350 --> 00:38:04,880
Another way to put this is, just
put all your money in the bank

608
00:38:04,880 --> 00:38:09,350
and then borrow what you need
if you need to borrow it all.

609
00:38:09,350 --> 00:38:11,300
So all these guys are
putting their money

610
00:38:11,300 --> 00:38:14,690
in the bank,
including the firms,

611
00:38:14,690 --> 00:38:18,060
and then they borrow
back what they need.

612
00:38:18,060 --> 00:38:20,790
Now it is true that the
borrowing and lending

613
00:38:20,790 --> 00:38:22,710
rate are the same
here, and that's how

614
00:38:22,710 --> 00:38:26,670
on getting away with this trick.

615
00:38:26,670 --> 00:38:31,140
Two lectures from now we'll have
a paper focusing quite a bit

616
00:38:31,140 --> 00:38:34,080
on the spread between
borrowing and lending

617
00:38:34,080 --> 00:38:37,650
as a measure of the
cost of intermediation.

618
00:38:37,650 --> 00:38:39,900
This paper doesn't have
that, this paper just

619
00:38:39,900 --> 00:38:43,800
assumes perfectly
costless intermediation,

620
00:38:43,800 --> 00:38:47,420
and these rates are the same.

621
00:38:47,420 --> 00:38:47,920
OK.

622
00:38:47,920 --> 00:38:58,550
So then alpha
fraction of households

623
00:38:58,550 --> 00:39:01,850
are in this intermediate
sector at this date.

624
00:39:01,850 --> 00:39:05,540
So how much is in
the savings accounts?

625
00:39:05,540 --> 00:39:09,080
So you take the
total population,

626
00:39:09,080 --> 00:39:11,990
alpha fraction of the
population intermediated,

627
00:39:11,990 --> 00:39:13,700
everybody's putting--

628
00:39:13,700 --> 00:39:15,530
in that sector
putting b in the bank.

629
00:39:15,530 --> 00:39:18,740
So alpha times b is the saving.

630
00:39:18,740 --> 00:39:25,070
And what's basically
the borrowing level?

631
00:39:25,070 --> 00:39:27,890
Well firms, if you
are firm, you're

632
00:39:27,890 --> 00:39:29,990
going to borrow to
finance both your setup

633
00:39:29,990 --> 00:39:34,370
costs and your capital.

634
00:39:34,370 --> 00:39:40,790
So basically we add
up or integrate up

635
00:39:40,790 --> 00:39:43,220
over the two characteristics--

636
00:39:43,220 --> 00:39:48,320
that is to say the setup costs
themselves as well as wealth.

637
00:39:51,830 --> 00:39:55,400
You will only
integrate up over firms

638
00:39:55,400 --> 00:39:58,280
when x is below some threshold.

639
00:39:58,280 --> 00:40:01,760
Now this threshold no long--

640
00:40:01,760 --> 00:40:04,550
I should actually have drawn
it on the other diagram.

641
00:40:04,550 --> 00:40:07,460
It's a completely flat line.

642
00:40:07,460 --> 00:40:09,530
Whether or not you're
a firm or a worker--

643
00:40:09,530 --> 00:40:12,260
I said this-- has nothing
to do with your wealth,

644
00:40:12,260 --> 00:40:14,940
it just has to do
with your cost.

645
00:40:14,940 --> 00:40:19,070
So if your costs are high,
above this threshold,

646
00:40:19,070 --> 00:40:20,870
you're not going to be a firm.

647
00:40:20,870 --> 00:40:22,693
And if it's below, you are.

648
00:40:22,693 --> 00:40:24,110
If you don't have
a lot of wealth,

649
00:40:24,110 --> 00:40:25,652
you're going to have
to borrow a lot.

650
00:40:25,652 --> 00:40:28,610
So this is sort of
the demand for funds.

651
00:40:28,610 --> 00:40:34,810
And then basically you're going
to set the demand for funds

652
00:40:34,810 --> 00:40:38,380
equal to the supply for funds
and find an interest rate that

653
00:40:38,380 --> 00:40:41,190
does that.

654
00:40:41,190 --> 00:40:44,020
There is an inequality
here only because, again,

655
00:40:44,020 --> 00:40:47,200
the 1 plus little r
could go to 1, basically,

656
00:40:47,200 --> 00:40:49,090
so the cost of
funds is basically

657
00:40:49,090 --> 00:40:52,510
like storing grain in the
backyard, in which case

658
00:40:52,510 --> 00:40:55,150
there's some indeterminacy you
might as well just carry it

659
00:40:55,150 --> 00:40:59,320
over rather than
put it in a bank.

660
00:40:59,320 --> 00:41:03,050
But interest rates can
be higher than that.

661
00:41:03,050 --> 00:41:07,660
And likewise,
employment, we've got--

662
00:41:07,660 --> 00:41:10,180
we just integrate up
the demand for work.

663
00:41:14,690 --> 00:41:17,000
Employment is tricky.

664
00:41:17,000 --> 00:41:19,640
People have jobs.

665
00:41:19,640 --> 00:41:21,080
You're either going
to be a farmer

666
00:41:21,080 --> 00:41:23,960
or a worker or an entrepreneur.

667
00:41:23,960 --> 00:41:28,460
There's only so many
people in the population.

668
00:41:28,460 --> 00:41:34,170
So if the size of the
population is normalized to 1,

669
00:41:34,170 --> 00:41:36,570
you've got basically the
number of entrepreneurs

670
00:41:36,570 --> 00:41:41,200
and the number of work workers.

671
00:41:41,200 --> 00:41:43,120
The number of
workers, it just comes

672
00:41:43,120 --> 00:41:45,880
from the employment
of the existing firms

673
00:41:45,880 --> 00:41:52,290
where l is the labor
hired of a firm, facing R

674
00:41:52,290 --> 00:41:54,390
is the interest rate,
and w is the wage.

675
00:41:57,140 --> 00:41:59,380
There's two sectors
here, intermediated and

676
00:41:59,380 --> 00:42:01,090
non-intermediated.

677
00:42:01,090 --> 00:42:03,040
They're different
from each other,

678
00:42:03,040 --> 00:42:06,350
but the labor market is perfect.

679
00:42:06,350 --> 00:42:11,610
Again, not so credible,
but makes it much easier.

680
00:42:11,610 --> 00:42:14,370
So you can work for
any firm, there's

681
00:42:14,370 --> 00:42:16,680
not even any real
migration cost-- maybe

682
00:42:16,680 --> 00:42:20,780
that cost of living
costs, and that's it.

683
00:42:20,780 --> 00:42:24,360
So you'll be indifferent at
a given wage where you work.

684
00:42:24,360 --> 00:42:27,350
So you would just add
up all the employment,

685
00:42:27,350 --> 00:42:30,230
add up the number
of firms, and again,

686
00:42:30,230 --> 00:42:32,412
lots of the time, this
thing will be equal to 1,

687
00:42:32,412 --> 00:42:35,120
you used up all the
people, and you'll

688
00:42:35,120 --> 00:42:38,900
have to find a wage that equates
the demand and the supply

689
00:42:38,900 --> 00:42:41,420
for workers.

690
00:42:41,420 --> 00:42:45,860
So on Friday, Yan will kind of
talk you through a bit more.

691
00:42:45,860 --> 00:42:49,070
Easy to say, maybe
not so easy to do.

692
00:42:49,070 --> 00:42:52,040
Even though it's a
static model, you

693
00:42:52,040 --> 00:42:55,610
have to find these equilibrium
interest rates and wages that

694
00:42:55,610 --> 00:42:56,510
clear the market.

695
00:42:56,510 --> 00:42:58,420
You need an algorithm
for doing that.

696
00:42:58,420 --> 00:43:01,540
You need to have a good
guess about how to search

697
00:43:01,540 --> 00:43:05,630
and how to iterate.

698
00:43:05,630 --> 00:43:06,570
AUDIENCE: Question?

699
00:43:06,570 --> 00:43:07,507
ROBERT TOWNSEND: Yes?

700
00:43:07,507 --> 00:43:09,090
AUDIENCE: This may
be in all of these.

701
00:43:09,090 --> 00:43:11,340
So I want to think about
investments in human capital.

702
00:43:11,340 --> 00:43:12,530
Should we think of that as--

703
00:43:12,530 --> 00:43:13,988
that goes into x
or should we think

704
00:43:13,988 --> 00:43:16,302
of that as a different
kind of investment which--

705
00:43:16,302 --> 00:43:17,760
ROBERT TOWNSEND:
There really isn't

706
00:43:17,760 --> 00:43:20,180
any realistic human
capital in this model.

707
00:43:22,920 --> 00:43:25,000
These guys are choosing
whether to be--

708
00:43:25,000 --> 00:43:27,200
set up a firm or have--

709
00:43:27,200 --> 00:43:28,200
or what occupation.

710
00:43:28,200 --> 00:43:31,730
You could imagine a
model where the choice is

711
00:43:31,730 --> 00:43:36,280
whether or not to go to
school, but it's just not here.

712
00:43:36,280 --> 00:43:38,240
And it's not even
realistic in the sense

713
00:43:38,240 --> 00:43:41,030
that those x's, you might
say, well I went to school,

714
00:43:41,030 --> 00:43:44,540
I'm pretty talented, and I
know how to run a business,

715
00:43:44,540 --> 00:43:46,010
might want to do that.

716
00:43:46,010 --> 00:43:48,930
But that talent can change
from one period to the next,

717
00:43:48,930 --> 00:43:52,085
so it's not great
for human capital.

718
00:43:56,310 --> 00:43:57,210
OK.

719
00:43:57,210 --> 00:44:01,650
So here are income,
savings, labor share,

720
00:44:01,650 --> 00:44:04,410
the fraction of entrepreneurs,
and a measure of inequality--

721
00:44:04,410 --> 00:44:05,910
the Gini--

722
00:44:05,910 --> 00:44:08,430
as both predicted by
the model and what

723
00:44:08,430 --> 00:44:12,330
is actually in the data.

724
00:44:12,330 --> 00:44:14,610
The data are up here, actually.

725
00:44:14,610 --> 00:44:17,130
So this didn't work too well.

726
00:44:17,130 --> 00:44:21,930
Income growth in the
model is always less

727
00:44:21,930 --> 00:44:25,320
than it is in the actual data.

728
00:44:25,320 --> 00:44:27,510
Remember my little
bit for science?

729
00:44:27,510 --> 00:44:31,890
Where I said let's see
how well a model does?

730
00:44:31,890 --> 00:44:34,110
Yeah, we can match
these things much better

731
00:44:34,110 --> 00:44:37,090
if we also choose the
parameters to try to match them,

732
00:44:37,090 --> 00:44:40,200
but we didn't do that,
we're completely under--

733
00:44:40,200 --> 00:44:42,060
it's amazing, it's
doing in some sense,

734
00:44:42,060 --> 00:44:44,400
as well as it does at all--

735
00:44:44,400 --> 00:44:46,230
like look at
inequality, for example.

736
00:44:46,230 --> 00:44:48,370
You can't probably
tell from the back row

737
00:44:48,370 --> 00:44:52,530
the difference between
the model and the data.

738
00:44:52,530 --> 00:44:54,570
There's some intervals here.

739
00:44:54,570 --> 00:44:57,360
Basically, when you
estimate things,

740
00:44:57,360 --> 00:44:59,910
you have standard errors
on the parameter estimate.

741
00:44:59,910 --> 00:45:02,310
So there's a whole
confidence interval

742
00:45:02,310 --> 00:45:05,460
around all those curves I'm
drawing on the occupation

743
00:45:05,460 --> 00:45:06,510
choice.

744
00:45:06,510 --> 00:45:10,500
So we generated lots of
different paths drawing

745
00:45:10,500 --> 00:45:16,480
parameters within standard
confidence intervals.

746
00:45:16,480 --> 00:45:19,350
But the actual mean
path is right here

747
00:45:19,350 --> 00:45:21,700
and the actual Thai
data is right here.

748
00:45:21,700 --> 00:45:31,680
So this inequality kind of is
basically very well-tracked,

749
00:45:31,680 --> 00:45:33,990
and eventually goes down.

750
00:45:33,990 --> 00:45:36,720
When does this start going down?

751
00:45:36,720 --> 00:45:39,570
That's when there
are no more farms.

752
00:45:39,570 --> 00:45:42,550
Well that's kind of the
model version of it.

753
00:45:42,550 --> 00:45:45,090
There's a sharp
increase in the wage

754
00:45:45,090 --> 00:45:49,970
at around 1992 in Thailand.

755
00:45:49,970 --> 00:45:54,090
I think-- I don't know when
the turning point was in China.

756
00:45:54,090 --> 00:45:55,370
But this is pretty typical.

757
00:46:00,500 --> 00:46:02,720
Fraction of entrepreneurs
we're underpredicting,

758
00:46:02,720 --> 00:46:04,310
although eventually it rises.

759
00:46:04,310 --> 00:46:10,430
Labor share would do
quite well, and so on.

760
00:46:17,020 --> 00:46:21,400
So now with the model with
its pluses and minuses,

761
00:46:21,400 --> 00:46:30,890
we can do some
welfare experiments.

762
00:46:30,890 --> 00:46:35,340
And what I want
to do essentially

763
00:46:35,340 --> 00:46:39,430
is a couple of experiments.

764
00:46:39,430 --> 00:46:39,930
RCTs.

765
00:46:42,630 --> 00:46:46,530
One partial equilibrium and
one general equilibrium.

766
00:46:46,530 --> 00:46:51,480
So here's my randomized control
trial for partial equilibrium.

767
00:46:51,480 --> 00:46:54,360
I'm going to take a
household and move it

768
00:46:54,360 --> 00:46:56,760
from the non-intermediated
it sector

769
00:46:56,760 --> 00:47:00,160
and put it in the
intermediated sector

770
00:47:00,160 --> 00:47:01,960
and then look at
the welfare gains.

771
00:47:05,930 --> 00:47:18,870
Now actually, before 1992, the
wage in Thailand is the same,

772
00:47:18,870 --> 00:47:21,500
it's still at the
subsistence level

773
00:47:21,500 --> 00:47:25,530
regardless of this
expanding financial sector.

774
00:47:25,530 --> 00:47:28,448
So it's both a partial and
a general equilibrium model,

775
00:47:28,448 --> 00:47:30,740
but I don't have to worry
about the general equilibrium

776
00:47:30,740 --> 00:47:32,700
effects of the wage.

777
00:47:32,700 --> 00:47:38,450
It's not like I'm giving the
economy intermediation or not.

778
00:47:38,450 --> 00:47:41,720
Or it's equivalent
with having a guy moved

779
00:47:41,720 --> 00:47:45,800
from autarky to
financial intermediation,

780
00:47:45,800 --> 00:47:48,030
because a wage is the same.

781
00:47:48,030 --> 00:47:52,090
So what the hell is this?

782
00:47:52,090 --> 00:47:56,620
This huge spike
here are basically

783
00:47:56,620 --> 00:48:05,310
the gains equivalent--

784
00:48:05,310 --> 00:48:08,580
consumption gains
or wealth gains

785
00:48:08,580 --> 00:48:13,920
for households of low
wealth but high talent.

786
00:48:13,920 --> 00:48:20,670
So I pinpointed them earlier in
the occupation choice diagram.

787
00:48:20,670 --> 00:48:22,410
So they are no
longer constrained.

788
00:48:22,410 --> 00:48:24,150
So there are some
very talented people

789
00:48:24,150 --> 00:48:29,640
who will now set up businesses
and even hire workers and make

790
00:48:29,640 --> 00:48:31,880
a lot of money.

791
00:48:31,880 --> 00:48:36,450
After all, the flip side
of the wage being low

792
00:48:36,450 --> 00:48:40,290
is that when you're a firm, you
don't have to pay high costs.

793
00:48:40,290 --> 00:48:41,460
You make a ton of money.

794
00:48:44,310 --> 00:48:50,790
So these gains are so big that
they make the other things

795
00:48:50,790 --> 00:48:52,320
small in comparison.

796
00:48:52,320 --> 00:48:57,270
Actually, these sort
of orangey people

797
00:48:57,270 --> 00:49:02,490
are switching from workers
to firms, and these blue--

798
00:49:02,490 --> 00:49:05,785
light-blue people, those were--

799
00:49:05,785 --> 00:49:07,410
I don't know if this
will surprise you,

800
00:49:07,410 --> 00:49:11,610
those are inefficient firms
going out of business.

801
00:49:11,610 --> 00:49:17,190
I mean, they're an autarky, not
much to do with their money,

802
00:49:17,190 --> 00:49:20,310
throw it back into the business,
drive the marginal return down

803
00:49:20,310 --> 00:49:25,890
to the carrying costs
of grain, and that's

804
00:49:25,890 --> 00:49:34,800
the best thing other than
being a farmer that you can do.

805
00:49:34,800 --> 00:49:37,970
But when the interest
rate goes up early on--

806
00:49:37,970 --> 00:49:39,920
and early on the interest
rate in this economy

807
00:49:39,920 --> 00:49:44,560
will be high, because there
isn't that much wealth,

808
00:49:44,560 --> 00:49:47,880
and there's relatively high
marginal product in production.

809
00:49:47,880 --> 00:49:52,430
So these lazy, untalented,
richer people--

810
00:49:52,430 --> 00:49:54,110
sorry, lazy was excessive.

811
00:49:56,923 --> 00:49:58,340
They don't have a
choice about it,

812
00:49:58,340 --> 00:50:01,730
they're just not
good at anything.

813
00:50:01,730 --> 00:50:06,192
They take their money and
they put it in the bank

814
00:50:06,192 --> 00:50:07,400
and they get a higher return.

815
00:50:11,420 --> 00:50:14,210
We'll come back to this later
in the class when we actually

816
00:50:14,210 --> 00:50:18,820
come back to RCTs and
IV and all of that,

817
00:50:18,820 --> 00:50:20,780
but there is already
a caution here--

818
00:50:20,780 --> 00:50:23,170
some things are not monotonic.

819
00:50:23,170 --> 00:50:26,900
The treatment effect is not
monotonic in this model.

820
00:50:26,900 --> 00:50:29,360
Intermediation it's a
good thing in the sense

821
00:50:29,360 --> 00:50:33,350
that everyone can earn higher
income at least when there are

822
00:50:33,350 --> 00:50:36,600
no general equilibrium effects.

823
00:50:36,600 --> 00:50:38,670
But the way in which
you earn higher income

824
00:50:38,670 --> 00:50:41,040
may involve occupation shifts--

825
00:50:41,040 --> 00:50:43,380
some people going into
business and some people

826
00:50:43,380 --> 00:50:45,700
going out of business.

827
00:50:45,700 --> 00:50:47,850
So in a very simple
way, you can see

828
00:50:47,850 --> 00:50:55,280
how you would lose
monotonicity of the treatment.

829
00:50:55,280 --> 00:50:57,080
We'll come back to
that in a later day.

830
00:51:00,950 --> 00:51:03,530
So this is a prediction--

831
00:51:03,530 --> 00:51:06,070
oh, now let's jump countries.

832
00:51:06,070 --> 00:51:07,200
I'll go to Mexico.

833
00:51:07,200 --> 00:51:10,160
Actually, I think
the next slide is--

834
00:51:10,160 --> 00:51:12,260
oh, how annoying.

835
00:51:12,260 --> 00:51:13,140
OK, I'm learning.

836
00:51:15,880 --> 00:51:17,260
We won't have this--

837
00:51:17,260 --> 00:51:19,240
I'll try to avoid this
problem in the future.

838
00:51:23,390 --> 00:51:29,300
In Mexico, the
fractions of basically

839
00:51:29,300 --> 00:51:34,640
wage earners, the fraction of
agriculturalists or subsistence

840
00:51:34,640 --> 00:51:37,970
people, and the fraction
of entrepreneurs.

841
00:51:37,970 --> 00:51:41,900
And the earlier one I didn't
show you on that scale

842
00:51:41,900 --> 00:51:44,030
actually shows these can
be slightly different,

843
00:51:44,030 --> 00:51:47,450
but on this scale with
all the occupations,

844
00:51:47,450 --> 00:51:49,670
it works quite well.

845
00:51:49,670 --> 00:51:51,010
So yes?

846
00:51:51,010 --> 00:51:54,600
AUDIENCE: [INAUDIBLE]
imagine the fractions

847
00:51:54,600 --> 00:51:56,720
in each occupation.

848
00:51:56,720 --> 00:51:58,500
You mentioned some of that--

849
00:51:58,500 --> 00:52:01,060
there's a lot of movement
in cross-occupation type.

850
00:52:01,060 --> 00:52:04,373
That probably is a
[INAUDIBLE] from the data.

851
00:52:04,373 --> 00:52:05,790
ROBERT TOWNSEND:
Yeah, we're not--

852
00:52:05,790 --> 00:52:06,872
AUDIENCE: Times the time.

853
00:52:06,872 --> 00:52:07,830
ROBERT TOWNSEND: Right.

854
00:52:07,830 --> 00:52:13,800
We're not actually looking at
the panel of a given household

855
00:52:13,800 --> 00:52:17,520
at their career choice
or occupation choice

856
00:52:17,520 --> 00:52:18,930
over many time periods.

857
00:52:18,930 --> 00:52:21,900
We're just only using
the cross-section

858
00:52:21,900 --> 00:52:23,190
to do the estimation.

859
00:52:28,390 --> 00:52:31,030
So even though-- I think your
point is-- even though this

860
00:52:31,030 --> 00:52:34,000
seems to fit well,
beneath this picture,

861
00:52:34,000 --> 00:52:39,430
there may be households
flipping from one to the other

862
00:52:39,430 --> 00:52:41,110
and the model doesn't
say they should.

863
00:52:44,750 --> 00:52:47,690
Actually, when we get to
Buera and Shin and so on,

864
00:52:47,690 --> 00:52:50,570
that will be at center stage.

865
00:52:50,570 --> 00:52:53,150
They're going to be more
realistic about the talent

866
00:52:53,150 --> 00:52:56,900
distribution, and they're going
to even have forward-looking

867
00:52:56,900 --> 00:53:00,050
dynamics so you'll-- they'll
anticipate the way in which

868
00:53:00,050 --> 00:53:06,410
their talent may evolve in the
future and it will remedy this

869
00:53:06,410 --> 00:53:08,130
shortcoming.

870
00:53:08,130 --> 00:53:11,400
Now it looks like I'm
claiming a big success here

871
00:53:11,400 --> 00:53:14,220
in the sense of jumping
New Mexico and all

872
00:53:14,220 --> 00:53:17,040
of a sudden showing you
the outcome after all

873
00:53:17,040 --> 00:53:20,220
the same steps, but let me
show you something else.

874
00:53:32,640 --> 00:53:39,540
So this line is the
model predicted number

875
00:53:39,540 --> 00:53:42,420
of entrepreneurs in the
sector without credit,

876
00:53:42,420 --> 00:53:45,970
and this is the one with credit.

877
00:53:45,970 --> 00:53:46,840
It's bad news.

878
00:53:46,840 --> 00:53:49,540
It's kind of the opposite
of what you think.

879
00:53:49,540 --> 00:53:57,660
Financial autarky turns out
to be better in this sense.

880
00:53:57,660 --> 00:54:02,220
Now normal people don't
show you this stuff.

881
00:54:05,060 --> 00:54:07,760
But I'm showing it
to you for a reason.

882
00:54:07,760 --> 00:54:10,340
And it's in that book
that I mentioned, the one

883
00:54:10,340 --> 00:54:12,680
we're writing on Mexico.

884
00:54:12,680 --> 00:54:14,840
So what went wrong?

885
00:54:14,840 --> 00:54:17,150
Well I went back and I've
been worried about this

886
00:54:17,150 --> 00:54:19,110
for quite a long time.

887
00:54:19,110 --> 00:54:22,480
I went back and looked
at the chapters again.

888
00:54:22,480 --> 00:54:26,800
Am I going to put this
book in the trash can?

889
00:54:26,800 --> 00:54:30,490
And what I realized was
that occupation choice

890
00:54:30,490 --> 00:54:34,400
where you do these probits of--

891
00:54:34,400 --> 00:54:37,280
the variables like credit
weren't even showing up there

892
00:54:37,280 --> 00:54:42,980
as being significant,
whereas they do in Thailand.

893
00:54:42,980 --> 00:54:46,820
So the data-- the micro data
were essentially speaking to me

894
00:54:46,820 --> 00:54:50,770
and I wasn't hearing
it for a while.

895
00:54:50,770 --> 00:54:52,380
It makes the point
of the lecture,

896
00:54:52,380 --> 00:54:57,300
which is you've got to take the
micro underpinning seriously.

897
00:54:57,300 --> 00:55:01,500
And I followed an algorithm as
if I'd figured something out

898
00:55:01,500 --> 00:55:03,840
in Thailand and I
could do it in Mexico.

899
00:55:03,840 --> 00:55:07,410
But now I know why it wasn't
working so well in Mexico.

900
00:55:07,410 --> 00:55:10,230
It doesn't tell you what the
next step is, by the way,

901
00:55:10,230 --> 00:55:14,526
but clearly LEB isn't going
to work there very well.

902
00:55:14,526 --> 00:55:19,810
AUDIENCE: But isn't the firm
bigger In the credit sector?

903
00:55:19,810 --> 00:55:21,060
ROBERT TOWNSEND: In the model?

904
00:55:21,060 --> 00:55:21,700
AUDIENCE: Yeah.

905
00:55:21,700 --> 00:55:22,358
The firm is--

906
00:55:22,358 --> 00:55:24,275
ROBERT TOWNSEND: Yeah,
the model doesn't work.

907
00:55:24,275 --> 00:55:26,250
AUDIENCE: No, but--
so if there is a lower

908
00:55:26,250 --> 00:55:28,260
percentage [INAUDIBLE]
but the firm's bigger,

909
00:55:28,260 --> 00:55:29,210
that's fine, right?

910
00:55:29,210 --> 00:55:30,680
ROBERT TOWNSEND: Oh.

911
00:55:30,680 --> 00:55:34,170
All right, I don't
have firm size.

912
00:55:34,170 --> 00:55:36,720
I'll go back and
look if we actually--

913
00:55:36,720 --> 00:55:40,440
oh, I should say also
something related.

914
00:55:40,440 --> 00:55:43,140
We have-- like I
have for Thailand,

915
00:55:43,140 --> 00:55:45,840
and it's available
to you on Dataverse,

916
00:55:45,840 --> 00:55:48,690
we have a massive
archive of Mexican data.

917
00:55:51,390 --> 00:55:56,510
It should be on Dataverse,
and I meant to include--

918
00:55:56,510 --> 00:56:00,530
I hope there is included
already Jorge Moreno's summary

919
00:56:00,530 --> 00:56:03,930
of the Mexican data.

920
00:56:03,930 --> 00:56:06,030
And this book will
kind of give you

921
00:56:06,030 --> 00:56:09,270
a sense of the great variety
of data we have and--

922
00:56:09,270 --> 00:56:12,360
I can go back and look and
see whether we actually

923
00:56:12,360 --> 00:56:13,710
plotted firm size.

924
00:56:21,310 --> 00:56:23,390
OK, another model.

925
00:56:23,390 --> 00:56:28,210
So this is the one with
endogenous financial deepening,

926
00:56:28,210 --> 00:56:33,430
and I'm leading off with a
picture that should remind you

927
00:56:33,430 --> 00:56:36,530
of the lecture of last time,
which is this trade-off

928
00:56:36,530 --> 00:56:37,030
between--

929
00:56:40,300 --> 00:56:45,280
potentially between
growth and levels and also

930
00:56:45,280 --> 00:56:47,560
the variability of growth.

931
00:56:50,380 --> 00:56:59,070
In this case, at low
levels of income,

932
00:56:59,070 --> 00:57:01,950
you have higher
variability, which

933
00:57:01,950 --> 00:57:07,060
is kind of a fact that seems
to be true in the data.

934
00:57:07,060 --> 00:57:09,900
However, this is a
model-generated path

935
00:57:09,900 --> 00:57:15,540
in which over time, per-capita
income is going to be changing,

936
00:57:15,540 --> 00:57:17,670
and from those
points on, is going

937
00:57:17,670 --> 00:57:20,460
to change the
statistics of the model

938
00:57:20,460 --> 00:57:24,930
in terms of expected future
growth and the variance

939
00:57:24,930 --> 00:57:27,040
of the growth.

940
00:57:27,040 --> 00:57:29,830
So how does this model work?

941
00:57:29,830 --> 00:57:34,240
Well it is-- this one
is forward-looking.

942
00:57:34,240 --> 00:57:37,440
So households don't just care
about today and tomorrow,

943
00:57:37,440 --> 00:57:43,150
they care about the
discounted expected future,

944
00:57:43,150 --> 00:57:46,860
and they care about, in this
case, the log of consumption

945
00:57:46,860 --> 00:57:49,120
at every date.

946
00:57:49,120 --> 00:57:51,550
In a minute we're going
to talk about some-- only

947
00:57:51,550 --> 00:57:53,730
slightly more general
power functions.

948
00:57:53,730 --> 00:58:01,150
Beta is the discount
rate, and we're

949
00:58:01,150 --> 00:58:05,230
going to simplify sectors
and occupation choices,

950
00:58:05,230 --> 00:58:12,790
almost like telling the usual
riskless and risky asset.

951
00:58:12,790 --> 00:58:15,490
Rather than thinking
of them as occupations,

952
00:58:15,490 --> 00:58:17,150
there's a riskless asset.

953
00:58:17,150 --> 00:58:21,250
So when you invest at t
minus 1 a certain amount,

954
00:58:21,250 --> 00:58:26,850
you get delta times that amount
back in the following period.

955
00:58:26,850 --> 00:58:32,470
A constant rate of return,
and the risky technology

956
00:58:32,470 --> 00:58:37,261
when you invest at the
amount i at t minus 1,

957
00:58:37,261 --> 00:58:43,290
you get it back plus
or minus 2 components.

958
00:58:43,290 --> 00:58:46,780
There's an aggregate
shock, theta t, no i on it.

959
00:58:46,780 --> 00:58:50,200
It's common across
every household.

960
00:58:50,200 --> 00:58:52,360
It's like aggregate risk.

961
00:58:52,360 --> 00:58:54,790
And there's an epsilon,
which does have--

962
00:58:58,100 --> 00:59:00,980
this is household
j, sorry, not i.

963
00:59:00,980 --> 00:59:03,310
i as investment, j is household.

964
00:59:03,310 --> 00:59:06,160
This epsilon j is the
idiosyncratic shock

965
00:59:06,160 --> 00:59:09,640
that's hitting household
j, and definitely

966
00:59:09,640 --> 00:59:11,980
different across
different households.

967
00:59:16,240 --> 00:59:21,810
So if you're lucky
enough or you've

968
00:59:21,810 --> 00:59:30,240
paid the money to be in the club
of the financial intermediary,

969
00:59:30,240 --> 00:59:33,130
there's an advantage--

970
00:59:33,130 --> 00:59:36,120
two advantages, actually.

971
00:59:36,120 --> 00:59:42,790
Information and pooling
of idiosyncratic risk.

972
00:59:42,790 --> 00:59:49,860
So the return you get at t
minus 1 for carrying stuff over

973
00:59:49,860 --> 00:59:55,230
is the maximum, largely,
of either the safe return

974
00:59:55,230 --> 00:59:56,880
or the risky return.

975
00:59:56,880 --> 01:00:00,120
But the risky return
isn't risky anymore.

976
01:00:00,120 --> 01:00:03,240
This model goes too far, you
see it perfectly in advance,

977
01:00:03,240 --> 01:00:06,410
so you can kind of
pick off what to do.

978
01:00:06,410 --> 01:00:10,610
It's an extreme version
of intermediaries

979
01:00:10,610 --> 01:00:13,610
have an information advantage
and they can advise clients

980
01:00:13,610 --> 01:00:16,430
because they have
that advantage.

981
01:00:19,790 --> 01:00:22,040
Well not quite, there's
kind of an intermediation

982
01:00:22,040 --> 01:00:24,900
cost and that's gamma.

983
01:00:24,900 --> 01:00:28,260
So this is a wedge, this
is in inefficiency--

984
01:00:28,260 --> 01:00:33,210
it's not modeled in a deep
way, but it uses up resources.

985
01:00:33,210 --> 01:00:36,010
And we'll come back
to that momentarily.

986
01:00:36,010 --> 01:00:40,390
And the other thing
is, it's costly to get

987
01:00:40,390 --> 01:00:46,600
into the investment club,
and it cost basically alpha.

988
01:00:46,600 --> 01:00:51,090
Not the alpha of the previous
slide, just some new number.

989
01:00:51,090 --> 01:00:53,950
There's perfect competition
among financial intermediaries,

990
01:00:53,950 --> 01:00:57,100
they might be tempted
to try to make profits,

991
01:00:57,100 --> 01:00:58,840
but they're going
to be driven down

992
01:00:58,840 --> 01:01:02,440
to basically making
people pay the fixed cost

993
01:01:02,440 --> 01:01:05,350
to enter the financial
system at the true cost

994
01:01:05,350 --> 01:01:11,520
and to give them a return
that exactly covers the wedge

995
01:01:11,520 --> 01:01:14,440
and is otherwise equal to
the real rate of return

996
01:01:14,440 --> 01:01:16,080
on the investment.

997
01:01:16,080 --> 01:01:17,860
So where did the epsilon go?

998
01:01:21,247 --> 01:01:22,330
I said they got the risk--

999
01:01:25,340 --> 01:01:27,740
I didn't quite say it right.

1000
01:01:27,740 --> 01:01:30,830
You get the risky
thing for sure,

1001
01:01:30,830 --> 01:01:34,910
not only because theta,
but because these epsilons

1002
01:01:34,910 --> 01:01:36,320
disappear.

1003
01:01:36,320 --> 01:01:38,900
And the epsilons
are pooling across

1004
01:01:38,900 --> 01:01:40,440
many, many, many households.

1005
01:01:40,440 --> 01:01:44,330
It's IID, essentially,
so it goes to 0

1006
01:01:44,330 --> 01:01:46,020
as you have more
and more households.

1007
01:01:46,020 --> 01:01:49,620
So it's like a mutual fund.

1008
01:01:49,620 --> 01:01:52,820
It's not just a bank
with fixed borrowing,

1009
01:01:52,820 --> 01:01:53,960
it's more like equity.

1010
01:01:57,390 --> 01:01:59,370
You put your money
in the bank, it

1011
01:01:59,370 --> 01:02:03,120
gets kind of the average
return on the bank's portfolio,

1012
01:02:03,120 --> 01:02:05,910
you might want to borrow
some to finance investment,

1013
01:02:05,910 --> 01:02:07,597
but that's kind of separate.

1014
01:02:11,580 --> 01:02:13,810
So this looks like a
discrete choice problem.

1015
01:02:18,860 --> 01:02:25,190
And if you bear with me, It's
not as bad as it may seem.

1016
01:02:25,190 --> 01:02:27,980
Suppose at day t, you've
already made the decision

1017
01:02:27,980 --> 01:02:31,620
to keep yourself out of
the financial system.

1018
01:02:31,620 --> 01:02:34,520
So then you have
a value function

1019
01:02:34,520 --> 01:02:38,670
that depends on your
wealth, which is not

1020
01:02:38,670 --> 01:02:42,720
b here, but k at day
t, and you're going

1021
01:02:42,720 --> 01:02:44,220
to decide what to do with that.

1022
01:02:44,220 --> 01:02:49,140
You can save some of it, and the
difference between what you had

1023
01:02:49,140 --> 01:02:52,160
and what you save, you eat.

1024
01:02:52,160 --> 01:02:58,945
And the amount that you
save, you get to put into--

1025
01:03:04,080 --> 01:03:07,200
some fraction into the risky
thing and some fraction

1026
01:03:07,200 --> 01:03:08,670
into the safe thing.

1027
01:03:08,670 --> 01:03:11,400
Now you're not in the
bank by the decision

1028
01:03:11,400 --> 01:03:13,890
here, so you're definitely
going to be experiencing

1029
01:03:13,890 --> 01:03:17,056
this idiosyncratic risk.

1030
01:03:17,056 --> 01:03:24,730
It may make the save return
more attractive, but we'll see.

1031
01:03:24,730 --> 01:03:32,060
OK, so then the next
period rolls around,

1032
01:03:32,060 --> 01:03:34,050
and then at the
very, very beginning,

1033
01:03:34,050 --> 01:03:36,030
you kind of get to
decide whether to join

1034
01:03:36,030 --> 01:03:39,410
the financial system
or not at that date.

1035
01:03:39,410 --> 01:03:43,510
You're forward-looking.

1036
01:03:43,510 --> 01:03:49,470
So the value to, say, not
joining the financial system

1037
01:03:49,470 --> 01:03:52,800
is this W that was over here.

1038
01:03:52,800 --> 01:03:57,300
W means withdrawn or
not participating.

1039
01:03:57,300 --> 01:04:00,300
But if you do decide
to join, you're

1040
01:04:00,300 --> 01:04:04,560
going to get the other branch,
the value of participating,

1041
01:04:04,560 --> 01:04:10,340
and that's basically going to
be something called V. However,

1042
01:04:10,340 --> 01:04:14,040
not quite V, because that
instant that you join,

1043
01:04:14,040 --> 01:04:16,700
you're going to have to subtract
off those fixed costs q.

1044
01:04:19,560 --> 01:04:23,190
And in this model, it's once
and never more, actually.

1045
01:04:23,190 --> 01:04:27,520
Once you pay to be in,
you stay in forever.

1046
01:04:27,520 --> 01:04:31,200
So this could have been a
simpler slide in the sense

1047
01:04:31,200 --> 01:04:34,170
that once you join and
you join the V branch,

1048
01:04:34,170 --> 01:04:38,700
you'll never again have
to choose to withdraw,

1049
01:04:38,700 --> 01:04:40,860
you'll just stay in the
financial system-- you reap

1050
01:04:40,860 --> 01:04:44,908
all the advantages of
information and risk pooling

1051
01:04:44,908 --> 01:04:46,200
and there's no reason to leave.

1052
01:04:46,200 --> 01:04:51,650
You'd have to change this
model to get disintermediation.

1053
01:04:51,650 --> 01:04:55,970
But the focus here was on
increasing intermediation.

1054
01:05:01,810 --> 01:05:07,220
So this is probably
something you've seen--

1055
01:05:07,220 --> 01:05:08,990
I'm not quite so
sure what you've

1056
01:05:08,990 --> 01:05:10,640
had in your other courses.

1057
01:05:10,640 --> 01:05:12,890
This is a dynamic
discrete choice model.

1058
01:05:17,400 --> 01:05:26,240
And basically, you have
this value function W

1059
01:05:26,240 --> 01:05:27,560
for low wealth.

1060
01:05:27,560 --> 01:05:30,078
It's the highest
curve, and then you

1061
01:05:30,078 --> 01:05:31,370
say, no, well this is above it.

1062
01:05:31,370 --> 01:05:35,030
No, not really, because
at the instant you join,

1063
01:05:35,030 --> 01:05:37,320
you have to subtract q.

1064
01:05:37,320 --> 01:05:40,740
So you take this V
and subtract q off

1065
01:05:40,740 --> 01:05:43,780
of the capital level that
brings the whole curve down.

1066
01:05:43,780 --> 01:05:46,490
So where these-- to this.

1067
01:05:46,490 --> 01:05:49,520
So where this thing
crosses, that's

1068
01:05:49,520 --> 01:05:51,820
kind of the key entry point.

1069
01:05:51,820 --> 01:05:56,090
At k at 15, given the way
I've drawn these curves,

1070
01:05:56,090 --> 01:05:58,790
to the left of 15, you
won't be participating,

1071
01:05:58,790 --> 01:06:00,590
and to the right
a 15, you will be.

1072
01:06:03,480 --> 01:06:06,040
Now remember, there's not a
whole lot of heterogeneity

1073
01:06:06,040 --> 01:06:09,370
either, sort of wealth
is the key unique state

1074
01:06:09,370 --> 01:06:11,390
variable at the
beginning of the period,

1075
01:06:11,390 --> 01:06:15,970
so wealth alone is going
to predict if the model is

1076
01:06:15,970 --> 01:06:21,340
accurate, whether or not
you're in the financial system,

1077
01:06:21,340 --> 01:06:23,520
and I'll come back to that.

1078
01:06:23,520 --> 01:06:29,860
Here, here are these savings
functions and portfolio

1079
01:06:29,860 --> 01:06:32,540
choices, and 15--

1080
01:06:32,540 --> 01:06:36,725
or I guess I said this
is kind of the key mark.

1081
01:06:39,250 --> 01:06:42,940
One thing that makes sense is
the dynamics, which is you're

1082
01:06:42,940 --> 01:06:48,070
saving up more and more in
advance for this eventual entry

1083
01:06:48,070 --> 01:06:50,530
into the financial system.

1084
01:06:50,530 --> 01:06:52,600
Exactly when you
will go in depends

1085
01:06:52,600 --> 01:06:54,640
on the sequence of shocks
that you experience,

1086
01:06:54,640 --> 01:06:58,120
and you may experience setbacks
that push your wealth down

1087
01:06:58,120 --> 01:07:00,790
even though ex ante your
investment decision made sense.

1088
01:07:00,790 --> 01:07:02,950
But kind of slowly
over time, there's

1089
01:07:02,950 --> 01:07:07,530
this pressure moving
off to the right.

1090
01:07:07,530 --> 01:07:11,272
So it makes-- but there's
discounting, right?

1091
01:07:11,272 --> 01:07:13,230
So you don't want to save
up too much early on,

1092
01:07:13,230 --> 01:07:15,000
you kind of want to
be close, and then you

1093
01:07:15,000 --> 01:07:16,830
don't want to take all
that fixed cost out

1094
01:07:16,830 --> 01:07:21,870
of your wealth at that date,
you save up to finance it.

1095
01:07:21,870 --> 01:07:26,580
This is unexpected ex ante.

1096
01:07:26,580 --> 01:07:31,500
So the closer you get,
the riskier you go.

1097
01:07:31,500 --> 01:07:34,260
So you would think
people who can't

1098
01:07:34,260 --> 01:07:36,810
diversify outside the
financial system would

1099
01:07:36,810 --> 01:07:38,820
be doing the safe thing.

1100
01:07:38,820 --> 01:07:43,050
But actually, this fixed cost
is creating a non-convexity.

1101
01:07:43,050 --> 01:07:44,670
If you remember
way back when you

1102
01:07:44,670 --> 01:07:49,060
see Friedman and Savage,
sort of that utility function

1103
01:07:49,060 --> 01:07:53,160
which is not concave everywhere,
well you've got two branches.

1104
01:07:53,160 --> 01:07:56,310
You've got this value
function for participating

1105
01:07:56,310 --> 01:07:59,010
and the value function
for not participating.

1106
01:07:59,010 --> 01:08:03,010
And in principle, they kind of
cross like a butterfly, right?

1107
01:08:03,010 --> 01:08:08,010
So then you're getting this
non-concave part in the middle,

1108
01:08:08,010 --> 01:08:12,480
and that's creating this kind of
risk behavior where they start

1109
01:08:12,480 --> 01:08:15,480
doing riskier or risky
things that kind of span--

1110
01:08:15,480 --> 01:08:18,090
take a chance of
grabbing the brass ring

1111
01:08:18,090 --> 01:08:22,620
and bootstrapping up
to a higher level.

1112
01:08:22,620 --> 01:08:24,120
We didn't see that coming.

1113
01:08:24,120 --> 01:08:28,590
Actually, those diagrams
look like they paced smoothly

1114
01:08:28,590 --> 01:08:31,950
into one another in
those value functions.

1115
01:08:31,950 --> 01:08:37,130
That's after this
randomization is going on.

1116
01:08:37,130 --> 01:08:38,899
Otherwise-- and so
the butterfly kind of

1117
01:08:38,899 --> 01:08:43,760
disappears, but only
after this randomization.

1118
01:08:55,420 --> 01:09:00,810
So let me just say a few more in
terms of the actual functional

1119
01:09:00,810 --> 01:09:02,310
forms.

1120
01:09:02,310 --> 01:09:07,250
Constant relative risk aversion
of essentially degree sigma.

1121
01:09:07,250 --> 01:09:10,319
Beta, I said, was
the discount rate.

1122
01:09:10,319 --> 01:09:15,090
Sorry, this is now
xi rather than theta.

1123
01:09:15,090 --> 01:09:18,569
But in any event,
we have to specify--

1124
01:09:18,569 --> 01:09:21,300
and they're uniformly
distributed-- again,

1125
01:09:21,300 --> 01:09:24,840
not my favorite choice anymore.

1126
01:09:24,840 --> 01:09:26,640
If something is
uniformly distributed,

1127
01:09:26,640 --> 01:09:29,380
you have to have an
upper and lower bound.

1128
01:09:29,380 --> 01:09:33,600
So we have picked up two more
parameters for the epsilon

1129
01:09:33,600 --> 01:09:43,890
distribution, and even something
for the risky distribution.

1130
01:09:57,440 --> 01:10:01,700
So discrete choice-- d for,
I guess, discrete choice,

1131
01:10:01,700 --> 01:10:03,380
basically you're
going to enter or not

1132
01:10:03,380 --> 01:10:06,470
enter depending on whether
your value of participating

1133
01:10:06,470 --> 01:10:10,160
is higher than not
participating and vice versa.

1134
01:10:10,160 --> 01:10:13,100
The model that
those parameters is

1135
01:10:13,100 --> 01:10:17,270
going to give you the k star,
which, again in the example,

1136
01:10:17,270 --> 01:10:19,010
was 15.

1137
01:10:19,010 --> 01:10:22,880
In general, k star is
going to depend on this--

1138
01:10:22,880 --> 01:10:27,440
all these list of parameters,
and this list of parameters--

1139
01:10:27,440 --> 01:10:33,080
the fixed cost, the discount
rate, aversion and utility,

1140
01:10:33,080 --> 01:10:35,840
the wedge in
intermediation, the upper

1141
01:10:35,840 --> 01:10:39,770
and lower bounds on
those distributions,

1142
01:10:39,770 --> 01:10:41,390
all of those things--

1143
01:10:41,390 --> 01:10:45,760
theta for GI-- basically
determine k star.

1144
01:10:45,760 --> 01:10:54,830
And k star is going to determine
whether you at a given wealth

1145
01:10:54,830 --> 01:10:56,600
choose to enter or not enter.

1146
01:10:56,600 --> 01:11:00,110
So I'll skip this,
but basically there's

1147
01:11:00,110 --> 01:11:03,500
a complicated way in
which the parameters

1148
01:11:03,500 --> 01:11:06,620
of the model through
the lens of the model

1149
01:11:06,620 --> 01:11:12,030
are generating this
sort of threshold value.

1150
01:11:12,030 --> 01:11:17,610
And likewise, therefore, you can
imagine, if at a given wealth,

1151
01:11:17,610 --> 01:11:20,190
you kind of see the fractions
participating and not

1152
01:11:20,190 --> 01:11:23,040
participating in
the cross-section,

1153
01:11:23,040 --> 01:11:28,380
that you would be able
to sort of invert things

1154
01:11:28,380 --> 01:11:31,180
and estimate these parameters.

1155
01:11:31,180 --> 01:11:34,650
It's a bit tricky because
that theta, the level

1156
01:11:34,650 --> 01:11:37,120
of the aggregate
shock kind of matters,

1157
01:11:37,120 --> 01:11:39,690
so we have to integrate
up over that, too.

1158
01:11:39,690 --> 01:11:44,660
But when we do the estimation
using a socioeconomic survey,

1159
01:11:44,660 --> 01:11:48,150
the papers online, we
ended up estimating--

1160
01:11:48,150 --> 01:11:52,080
and actually, comfortingly,
a lot of the parameters

1161
01:11:52,080 --> 01:11:53,040
make sense.

1162
01:11:53,040 --> 01:11:56,640
The discount rate's close
to 1, but not exactly.

1163
01:11:56,640 --> 01:12:01,890
A degree of risk aversion
is like almost the log case

1164
01:12:01,890 --> 01:12:06,600
and the depreciation rate was
4% and so on and so forth,

1165
01:12:06,600 --> 01:12:07,730
quite remarkable.

1166
01:12:07,730 --> 01:12:08,230
Yes?

1167
01:12:08,230 --> 01:12:10,640
AUDIENCE: And so is that--
am I reading this correctly?

1168
01:12:10,640 --> 01:12:12,140
Is that gamma was
estimated to be 1?

1169
01:12:15,132 --> 01:12:16,590
ROBERT TOWNSEND:
In this case, yes.

1170
01:12:16,590 --> 01:12:17,132
AUDIENCE: OK.

1171
01:12:17,132 --> 01:12:19,720
So with that imply efficient--
like there's no wedge?

1172
01:12:19,720 --> 01:12:22,250
ROBERT TOWNSEND: Yeah, no wedge.

1173
01:12:22,250 --> 01:12:25,871
Yeah, that-- I'll come
back to that momentarily.

1174
01:12:33,970 --> 01:12:36,100
So again, we can
simulate the model--

1175
01:12:36,100 --> 01:12:42,910
and under various estimated
or robustness checks,

1176
01:12:42,910 --> 01:12:44,560
parameter values--

1177
01:12:44,560 --> 01:12:47,050
to get the growth rate,
the participation rate,

1178
01:12:47,050 --> 01:12:52,000
and the tile index
of inequality.

1179
01:12:52,000 --> 01:12:55,220
Now one caution on
the growth rate,

1180
01:12:55,220 --> 01:12:57,580
and I always have to
check to make sure--

1181
01:12:57,580 --> 01:13:00,850
this model has aggregate shocks.

1182
01:13:00,850 --> 01:13:04,040
So if you integrate out
over the expected shocks,

1183
01:13:04,040 --> 01:13:05,790
you're going to get a
smoother growth rate

1184
01:13:05,790 --> 01:13:08,340
path than in the actual data.

1185
01:13:08,340 --> 01:13:09,855
And what is it that
I have to check?

1186
01:13:09,855 --> 01:13:11,730
I always have to go back
and see whether this

1187
01:13:11,730 --> 01:13:15,030
was one where we use
simulated values of theta

1188
01:13:15,030 --> 01:13:16,755
or we took the average
over those thetas.

1189
01:13:19,640 --> 01:13:23,000
We missed this upturn
here and the growth rate,

1190
01:13:23,000 --> 01:13:26,180
and that's related to the fact
that financial participation

1191
01:13:26,180 --> 01:13:28,530
goes flat right here.

1192
01:13:28,530 --> 01:13:31,180
This model's great
at the trends.

1193
01:13:31,180 --> 01:13:36,240
It gives you this very smooth,
ever-increasing participation

1194
01:13:36,240 --> 01:13:39,120
in the financial
system, and actually, it

1195
01:13:39,120 --> 01:13:42,060
gives you for a
while a good shot

1196
01:13:42,060 --> 01:13:46,140
at the increasing inequality
as those firms made

1197
01:13:46,140 --> 01:13:50,300
a lot more money, but
it misses the downturn.

1198
01:13:50,300 --> 01:13:52,070
In fact, there are
no real prices,

1199
01:13:52,070 --> 01:13:54,230
there's nothing like a
wage or an interest rate,

1200
01:13:54,230 --> 01:13:56,060
that's a defect.

1201
01:13:56,060 --> 01:14:01,640
And so it's almost like a
linear world, in some sense,

1202
01:14:01,640 --> 01:14:04,470
and we can't get inequality
to go down in this model,

1203
01:14:04,470 --> 01:14:07,850
but we can get the
trends right for a while.

1204
01:14:07,850 --> 01:14:10,150
This participation
thing is going

1205
01:14:10,150 --> 01:14:12,620
to bring me back to the wedge.

1206
01:14:20,570 --> 01:14:22,780
So what do we do?

1207
01:14:22,780 --> 01:14:28,650
We're going to do a policy
experiment, basically.

1208
01:14:35,830 --> 01:14:40,060
We look at the government's
control of the banking system

1209
01:14:40,060 --> 01:14:43,610
in terms of the ratio of credit.

1210
01:14:43,610 --> 01:14:45,950
It's moving around a bit,
but essentially it's flat,

1211
01:14:45,950 --> 01:14:48,380
it goes up, and then
it comes down again.

1212
01:14:48,380 --> 01:14:51,830
This is the liberalization
and this is the retrenchment

1213
01:14:51,830 --> 01:14:53,960
that happened before it.

1214
01:14:53,960 --> 01:15:00,100
So what we do is, we vary
that parameter gamma.

1215
01:15:00,100 --> 01:15:03,297
It's not going to be
costless intermediation.

1216
01:15:03,297 --> 01:15:04,880
In fact, we're going
to subtract off--

1217
01:15:04,880 --> 01:15:11,410
I think it was
something like 3% or 4%

1218
01:15:11,410 --> 01:15:14,470
when the intermediation is
being done by government banks.

1219
01:15:17,150 --> 01:15:20,970
And we don't change
the model otherwise.

1220
01:15:20,970 --> 01:15:27,270
And amazingly, we get this
almost perfectly right.

1221
01:15:27,270 --> 01:15:32,250
So we can actually get the
slowdown in the economy

1222
01:15:32,250 --> 01:15:37,740
as a function of this, quote,
bad policy experiment where

1223
01:15:37,740 --> 01:15:40,380
the government, for
one reason or another,

1224
01:15:40,380 --> 01:15:42,840
essentially took over
the banking system.

1225
01:15:42,840 --> 01:15:48,930
By the way, we don't do very
well with growth differences

1226
01:15:48,930 --> 01:15:53,930
even after liberalization,
and we can--

1227
01:15:53,930 --> 01:15:57,670
and the reason is partly
because, at least through

1228
01:15:57,670 --> 01:16:00,580
the lens of OLS
regressions, it's--

1229
01:16:00,580 --> 01:16:02,830
I've alluded to this before--

1230
01:16:02,830 --> 01:16:04,540
it's just basically junk.

1231
01:16:04,540 --> 01:16:07,600
I mean, it's not a
stationary model.

1232
01:16:07,600 --> 01:16:09,070
The decision rules
are stationary,

1233
01:16:09,070 --> 01:16:12,420
but aggregates like
growth rates are not.

1234
01:16:12,420 --> 01:16:15,970
And you can use
this value functions

1235
01:16:15,970 --> 01:16:19,330
to back out the welfare gains.

1236
01:16:19,330 --> 01:16:21,020
There are a number
of ways to do that.

1237
01:16:21,020 --> 01:16:23,110
One thing we could do
is just surprise people.

1238
01:16:23,110 --> 01:16:26,470
They think the banking system
is going to stay restricted,

1239
01:16:26,470 --> 01:16:28,900
and they've made
their dynamic plans,

1240
01:16:28,900 --> 01:16:33,757
and then we surprise
them and liberalize.

1241
01:16:33,757 --> 01:16:36,340
And they're better off, and we
can calculate through the value

1242
01:16:36,340 --> 01:16:38,500
functions.

1243
01:16:38,500 --> 01:16:40,150
It's a bit complicated
and I'm not

1244
01:16:40,150 --> 01:16:41,775
going to take you
through these slides,

1245
01:16:41,775 --> 01:16:45,170
but the welfare gains
can be quite large.

1246
01:16:45,170 --> 01:16:47,530
Now we can imagine
surprising them,

1247
01:16:47,530 --> 01:16:50,980
like varying that wedge or
varying the fixed cost and it

1248
01:16:50,980 --> 01:16:55,810
actually matters a bit, and we
can take this model to Mexico

1249
01:16:55,810 --> 01:17:02,640
and it does pretty well with
the trend of the growth rate,

1250
01:17:02,640 --> 01:17:10,480
but around 1992, Mexico
had yet another crisis.

1251
01:17:10,480 --> 01:17:14,280
And not surprising, this model
is just very ill-equipped

1252
01:17:14,280 --> 01:17:17,550
to handle that kind of stuff.

1253
01:17:17,550 --> 01:17:20,190
Actually, things
started going better

1254
01:17:20,190 --> 01:17:23,130
with Vicente Fox around 2002.

1255
01:17:23,130 --> 01:17:27,810
We could restart the model
over here and it will do well.

1256
01:17:27,810 --> 01:17:30,930
But it's not going to explain--

1257
01:17:30,930 --> 01:17:33,810
this is why Mexico
is still poor.

1258
01:17:33,810 --> 01:17:35,880
I mean, they go upstairs,
they go upstairs,

1259
01:17:35,880 --> 01:17:37,410
and then they fall down again.

1260
01:17:37,410 --> 01:17:40,925
And they've been doing this
since the early '80s, you know?

1261
01:17:40,925 --> 01:17:43,050
Now hopefully they're not
going to fall down again.

1262
01:17:43,050 --> 01:17:44,850
But the model--

1263
01:17:44,850 --> 01:17:46,920
I promised you
shortcomings, right?

1264
01:17:46,920 --> 01:17:52,620
This model is not designed to
get at sudden stops or things

1265
01:17:52,620 --> 01:17:53,750
along that line.

1266
01:18:01,490 --> 01:18:08,020
So the last thing I
want to show you quickly

1267
01:18:08,020 --> 01:18:13,450
is another kind of experiment.

1268
01:18:13,450 --> 01:18:15,640
Instead of going to a
new country like Mexico,

1269
01:18:15,640 --> 01:18:19,210
we started to look interior
to the country of Thailand.

1270
01:18:19,210 --> 01:18:24,510
This is a picture
of basically wealth

1271
01:18:24,510 --> 01:18:27,550
using a principal component.

1272
01:18:27,550 --> 01:18:29,380
You can see in
and around Bangkok

1273
01:18:29,380 --> 01:18:31,840
and going north to Chiang Mai--

1274
01:18:31,840 --> 01:18:35,380
the wealth corridor, if
you want, of the country,

1275
01:18:35,380 --> 01:18:40,180
these green things are the
provinces of my survey.

1276
01:18:40,180 --> 01:18:44,560
And looking at them
again, you can actually

1277
01:18:44,560 --> 01:18:47,125
see where the villages are,
you can see the road network,

1278
01:18:47,125 --> 01:18:52,340
at least on major highways, you
can see the district centers.

1279
01:18:52,340 --> 01:18:55,850
And I promised I
would ask you this.

1280
01:18:55,850 --> 01:18:57,560
I mentioned it before.

1281
01:18:57,560 --> 01:19:01,310
If you have an interest in
going through some of the GIS

1282
01:19:01,310 --> 01:19:05,360
capabilities, I'm more
than happy to schedule

1283
01:19:05,360 --> 01:19:07,490
an extra session.

1284
01:19:07,490 --> 01:19:09,400
I have to check my GIS person.

1285
01:19:09,400 --> 01:19:11,110
I see some of you
shaking your heads.

1286
01:19:11,110 --> 01:19:14,320
You don't all have to go,
so that's enough for me.

1287
01:19:14,320 --> 01:19:17,770
So let me check with her
and maybe we can possibly

1288
01:19:17,770 --> 01:19:21,970
do that Friday of
next week, but it

1289
01:19:21,970 --> 01:19:24,790
would be over and above the
regular recitation section.

1290
01:19:27,730 --> 01:19:30,760
And so here is a
picture of where

1291
01:19:30,760 --> 01:19:33,190
the businesses are in Thailand.

1292
01:19:35,800 --> 01:19:40,000
And through this GIS,
something called the Moran,

1293
01:19:40,000 --> 01:19:44,500
you can see whether areas, which
have high levels of enterprise,

1294
01:19:44,500 --> 01:19:49,840
are surrounded by other areas
with high levels of enterprise

1295
01:19:49,840 --> 01:19:53,230
or low levels surrounded by low.

1296
01:19:53,230 --> 01:19:59,695
And so you see this quite
salient geographic picture.

1297
01:20:02,290 --> 01:20:13,660
And here, we took levels
of enterprise in 1986

1298
01:20:13,660 --> 01:20:17,410
and looked at the growth
rate of enterprise

1299
01:20:17,410 --> 01:20:22,570
between 1986 and 1996.

1300
01:20:22,570 --> 01:20:26,550
This area in here, the
colors are a bit pale--

1301
01:20:26,550 --> 01:20:28,430
it doesn't come out too well--

1302
01:20:28,430 --> 01:20:30,700
are areas of
convergence in the sense

1303
01:20:30,700 --> 01:20:32,830
that they started at
high levels but then they

1304
01:20:32,830 --> 01:20:33,925
shrink to lower levels.

1305
01:20:37,520 --> 01:20:40,910
This area around it,
though, these reds, those

1306
01:20:40,910 --> 01:20:42,500
are areas of agglomeration.

1307
01:20:42,500 --> 01:20:45,550
They have high levels
and high growth rates.

1308
01:20:45,550 --> 01:20:51,140
There's an ever-increasing
concentration of enterprise.

1309
01:20:51,140 --> 01:20:56,660
These pink areas are catch-up
areas along the corridor which

1310
01:20:56,660 --> 01:21:00,980
had low levels, and now they're
gaining a lot of enterprise.

1311
01:21:00,980 --> 01:21:04,910
And these blue areas out here
are kind of like the boonies.

1312
01:21:04,910 --> 01:21:09,070
They have low levels
and low growth rates.

1313
01:21:09,070 --> 01:21:11,130
So it's not a uniform picture.

1314
01:21:11,130 --> 01:21:15,030
In fact, you can take
this down to the provinces

1315
01:21:15,030 --> 01:21:19,980
and look in the data at
the areas of agglomeration

1316
01:21:19,980 --> 01:21:26,040
and relative low
levels, and then that

1317
01:21:26,040 --> 01:21:27,960
looks like almost
the same picture.

1318
01:21:27,960 --> 01:21:30,900
That's a
model-generated picture.

1319
01:21:30,900 --> 01:21:35,610
So we actually projected--
took Lloyd-Ellis and Bernhardt

1320
01:21:35,610 --> 01:21:39,090
and projected it down
at the village level,

1321
01:21:39,090 --> 01:21:41,760
and here you can
see agglomeration.

1322
01:21:41,760 --> 01:21:45,040
It's not exactly identical.

1323
01:21:45,040 --> 01:21:47,790
There are certain
predictions in the model

1324
01:21:47,790 --> 01:21:51,990
that don't happen in
the data, but largely

1325
01:21:51,990 --> 01:21:54,620
these sort of favorable areas--

1326
01:21:54,620 --> 01:21:58,200
a subset of the candidate
of favorable areas

1327
01:21:58,200 --> 01:22:01,140
are the ones that actually start
growing like crazy in Thailand.

1328
01:22:04,140 --> 01:22:07,810
And here at that same level
is Greenwood in Jovanovic.

1329
01:22:11,130 --> 01:22:14,490
And here, these
red areas are areas

1330
01:22:14,490 --> 01:22:17,700
where the model
overpredicts the data,

1331
01:22:17,700 --> 01:22:19,980
and the green areas
are places where

1332
01:22:19,980 --> 01:22:21,480
the model is underpredicting.

1333
01:22:21,480 --> 01:22:24,150
In other words, these
red areas, which turn out

1334
01:22:24,150 --> 01:22:28,630
to be the same urban areas,
Greenwood and Jovanovic

1335
01:22:28,630 --> 01:22:32,460
is predicting they should have
more banks than they actually

1336
01:22:32,460 --> 01:22:34,680
do.

1337
01:22:34,680 --> 01:22:37,020
So in some sense, even
though they're higher wealth,

1338
01:22:37,020 --> 01:22:40,930
they're underserved relative to
what the model would predict.

1339
01:22:40,930 --> 01:22:44,280
And these green
areas, those are areas

1340
01:22:44,280 --> 01:22:48,660
where there's more access
to credit than Greenwood

1341
01:22:48,660 --> 01:22:50,360
and Jovanovic would predict.

1342
01:22:50,360 --> 01:22:53,340
Now it turns out
this story has to do

1343
01:22:53,340 --> 01:22:56,250
with the interplay
between commercial banks

1344
01:22:56,250 --> 01:23:00,210
that operate in urban areas and
the bank for agriculture which

1345
01:23:00,210 --> 01:23:02,220
is operating in rural
areas, and we'll

1346
01:23:02,220 --> 01:23:06,340
come back to that in the
last lecture of the class.

1347
01:23:06,340 --> 01:23:10,200
So I'm done, essentially.

1348
01:23:10,200 --> 01:23:13,380
I'll just point you to
the rest of the slides.

1349
01:23:13,380 --> 01:23:18,660
Basically we did kind
of a rigorous comparison

1350
01:23:18,660 --> 01:23:23,940
of the successes
and the failures.

1351
01:23:23,940 --> 01:23:26,100
So you can actually go
through each of the models

1352
01:23:26,100 --> 01:23:29,700
and see on which data
which model's doing well

1353
01:23:29,700 --> 01:23:34,950
and which data models
are doing poorly,

1354
01:23:34,950 --> 01:23:39,240
which feeds into the
issue of model selection

1355
01:23:39,240 --> 01:23:40,810
and next steps in the research.

1356
01:23:40,810 --> 01:23:45,680
So OK, with that, I'm
slightly two minutes beyond.