1
00:00:00,000 --> 00:00:01,952
[SQUEAKING]

2
00:00:01,952 --> 00:00:04,392
[RUSTLING]

3
00:00:04,392 --> 00:00:07,320
[CLICKING]

4
00:00:10,113 --> 00:00:12,030
FRANK SCHILBACH: So I'm
going to briefly recap

5
00:00:12,030 --> 00:00:13,740
what we discussed,
and then sort of talk

6
00:00:13,740 --> 00:00:17,580
about a number of different
applications, ranging from,

7
00:00:17,580 --> 00:00:20,280
like, work, exercising, credit
cards, savings behavior,

8
00:00:20,280 --> 00:00:23,040
drinking, smoking,
fertilizer use, and so on.

9
00:00:23,040 --> 00:00:26,670
And then, that'll spill
into the next lecture,

10
00:00:26,670 --> 00:00:28,290
or go into next lecture as well.

11
00:00:28,290 --> 00:00:29,790
I'm going to sort of
summarize where we are.

12
00:00:29,790 --> 00:00:31,170
What do we know about
time preferences?

13
00:00:31,170 --> 00:00:32,250
What have you learned?

14
00:00:32,250 --> 00:00:34,542
What's useful from what have
you learned for the world,

15
00:00:34,542 --> 00:00:38,800
and what things might be
still up for investigation?

16
00:00:38,800 --> 00:00:40,770
OK, so what happened
so far is, like,

17
00:00:40,770 --> 00:00:45,030
I showed you a simple model
of exponential discounting.

18
00:00:45,030 --> 00:00:48,840
Again, that's the workhorse
model of discounting.

19
00:00:48,840 --> 00:00:51,090
In economics, it's one of
the most successful and most

20
00:00:51,090 --> 00:00:53,520
important models that
people have written down.

21
00:00:53,520 --> 00:00:56,580
In economics, the Solow model,
et cetera, like, many sort

22
00:00:56,580 --> 00:00:59,170
of different important models,
thinking about long-run growth

23
00:00:59,170 --> 00:01:02,680
and so on, use this
particular model.

24
00:01:02,680 --> 00:01:05,580
So it's been tremendously
important and successful.

25
00:01:05,580 --> 00:01:07,950
It has different implications
that we discussed

26
00:01:07,950 --> 00:01:09,930
at length, both in
class, and lecture,

27
00:01:09,930 --> 00:01:12,870
and in recitation, which
is constant discounting,

28
00:01:12,870 --> 00:01:15,330
dynamic consistency, and
no demand for commitment.

29
00:01:15,330 --> 00:01:17,880
We discussed some
evidence that shows

30
00:01:17,880 --> 00:01:20,520
that these assumptions--
these implications--

31
00:01:20,520 --> 00:01:21,760
are not warranted.

32
00:01:21,760 --> 00:01:23,698
And then we talked about
a different version

33
00:01:23,698 --> 00:01:25,740
of this model, or like an
extension, if you want,

34
00:01:25,740 --> 00:01:28,073
of that model, which is the
quasi-hyperbolic discounting

35
00:01:28,073 --> 00:01:31,380
model, which adds an additional
parameter that measures

36
00:01:31,380 --> 00:01:33,720
people's present bias
or present focus,

37
00:01:33,720 --> 00:01:36,810
as people call it sometimes
these days, which allows us

38
00:01:36,810 --> 00:01:39,120
to be more flexible
and be able to look

39
00:01:39,120 --> 00:01:43,590
at short-run and long-run
discounting in the same model,

40
00:01:43,590 --> 00:01:45,990
as in, like, there is one
parameter, beta, that measures

41
00:01:45,990 --> 00:01:50,070
people's short-run discount
factor, and another parameter,

42
00:01:50,070 --> 00:01:51,982
delta, which is
close to one off,

43
00:01:51,982 --> 00:01:53,940
and that measures the
long-run discount factor.

44
00:01:53,940 --> 00:01:55,650
That makes the
model more flexible,

45
00:01:55,650 --> 00:01:59,220
and we're able to explain
some phenomena that might

46
00:01:59,220 --> 00:02:02,270
be hard to explain otherwise.

47
00:02:02,270 --> 00:02:05,880
Any questions on that so far,
or last time, or the like?

48
00:02:11,230 --> 00:02:14,770
OK, so then next, we talked
about sophistication versus

49
00:02:14,770 --> 00:02:15,910
naiveté.

50
00:02:15,910 --> 00:02:18,230
So this is the issue
that we discussed before,

51
00:02:18,230 --> 00:02:22,030
which is present bias creates
time inconsistency, right?

52
00:02:22,030 --> 00:02:24,370
When thinking about the
future, we want to be patient.

53
00:02:24,370 --> 00:02:26,620
When the time actually comes,
when the future actually

54
00:02:26,620 --> 00:02:29,230
arrived, we are impatient.

55
00:02:29,230 --> 00:02:31,690
So then the key question you
might ask is, like, well,

56
00:02:31,690 --> 00:02:34,450
do people understand
the time consistency?

57
00:02:34,450 --> 00:02:37,000
We talked about two different
extreme assumptions or versions

58
00:02:37,000 --> 00:02:37,500
if this.

59
00:02:37,500 --> 00:02:40,720
One is full naiveté, which is
the idea that the person does

60
00:02:40,720 --> 00:02:43,330
not realize that she
will change her mind.

61
00:02:43,330 --> 00:02:45,670
When thinking about
the future, she

62
00:02:45,670 --> 00:02:48,310
thinks she's going to follow
through on her favorite plan.

63
00:02:48,310 --> 00:02:50,530
When the future comes,
she will be patient.

64
00:02:50,530 --> 00:02:54,010
But then of course, the future
arrives and surprises happen.

65
00:02:54,010 --> 00:02:57,230
And the person is surprised
by their own present bias.

66
00:02:57,230 --> 00:03:00,370
And sort of there is false
optimism about future patience,

67
00:03:00,370 --> 00:03:01,920
and sort of the--

68
00:03:01,920 --> 00:03:03,670
over and over again,
the person might say,

69
00:03:03,670 --> 00:03:06,880
this time is different.

70
00:03:06,880 --> 00:03:10,480
Second extreme assumption
is full sophistication.

71
00:03:10,480 --> 00:03:11,740
That is perfect foresight.

72
00:03:11,740 --> 00:03:15,130
The person actually understands
their beta perfectly well,

73
00:03:15,130 --> 00:03:17,710
and understands that
she, in the future,

74
00:03:17,710 --> 00:03:19,510
might not stick to the
plans that she has,

75
00:03:19,510 --> 00:03:21,340
and might change her mind.

76
00:03:21,340 --> 00:03:24,610
So she does sort of her
best, given the future self's

77
00:03:24,610 --> 00:03:27,800
anticipated changes in behavior.

78
00:03:27,800 --> 00:03:29,560
So taken into account
as a constraint what

79
00:03:29,560 --> 00:03:31,850
the person will
do in the future,

80
00:03:31,850 --> 00:03:34,900
the person optimizes that way.

81
00:03:34,900 --> 00:03:37,000
There's no surprises
about future present bias.

82
00:03:37,000 --> 00:03:39,820
The person has
rational expectations.

83
00:03:39,820 --> 00:03:44,260
Those are the two broad extreme
assumptions that we discussed.

84
00:03:44,260 --> 00:03:46,330
Now, how can we tell--
how can we actually tell--

85
00:03:46,330 --> 00:03:50,890
so if you wanted to
know about your friend,

86
00:03:50,890 --> 00:03:54,100
and try to understand, is this
person naive or sophisticated,

87
00:03:54,100 --> 00:03:56,907
how can we actually tell
whether that's the case?

88
00:03:56,907 --> 00:03:57,490
What do we do?

89
00:04:07,460 --> 00:04:09,968
What data could you collect
from your friend or from people

90
00:04:09,968 --> 00:04:10,510
that you see?

91
00:04:14,320 --> 00:04:15,220
Yes?

92
00:04:15,220 --> 00:04:16,887
AUDIENCE: Just giving
them [INAUDIBLE]..

93
00:04:16,887 --> 00:04:19,466
Trying to see, like, what they
procrastinate [INAUDIBLE]..

94
00:04:22,223 --> 00:04:24,390
FRANK SCHILBACH: And what
exactly would you collect?

95
00:04:28,640 --> 00:04:31,792
AUDIENCE: I guess if
you could, asking them

96
00:04:31,792 --> 00:04:33,500
what they think that
they're going to do,

97
00:04:33,500 --> 00:04:35,642
and then seeing what
they actually do.

98
00:04:35,642 --> 00:04:37,600
FRANK SCHILBACH: Right,
so one thing you can do

99
00:04:37,600 --> 00:04:39,190
is collect the person's beliefs.

100
00:04:39,190 --> 00:04:41,690
So you can ask them, what are
you going to do in the future?

101
00:04:41,690 --> 00:04:43,660
And if the person mispredicts
what they're going to be

102
00:04:43,660 --> 00:04:45,580
in the future-- in particular,
if the person thinks

103
00:04:45,580 --> 00:04:47,913
they're going to be more
patient in the future than they

104
00:04:47,913 --> 00:04:50,500
actually are, so if they think
their beta in the future is

105
00:04:50,500 --> 00:04:55,780
higher than it actually is,
that would suggest at least some

106
00:04:55,780 --> 00:04:58,870
naiveté, right?

107
00:04:58,870 --> 00:05:03,770
What else could we elicit?

108
00:05:03,770 --> 00:05:04,937
Yes?

109
00:05:04,937 --> 00:05:06,770
AUDIENCE: There are
some choices that people

110
00:05:06,770 --> 00:05:08,600
make that wouldn't make
sense if they're not

111
00:05:08,600 --> 00:05:09,433
being sophisticated.

112
00:05:09,433 --> 00:05:12,439
So for example, is we see
them restricting their choices

113
00:05:12,439 --> 00:05:14,897
in some way that we would
expect a sophisticated person--

114
00:05:14,897 --> 00:05:17,230
like that would make sense
if they're [INAUDIBLE] naive.

115
00:05:17,230 --> 00:05:18,490
FRANK SCHILBACH: Exactly,
so if you offered them

116
00:05:18,490 --> 00:05:21,340
commitment devices, and we said,
here's a commitment device.

117
00:05:21,340 --> 00:05:22,840
You can change your
future behavior

118
00:05:22,840 --> 00:05:25,600
in certain ways that make
certain behaviors in the future

119
00:05:25,600 --> 00:05:26,990
more expensive.

120
00:05:26,990 --> 00:05:29,950
So for example, you
might tell your friend,

121
00:05:29,950 --> 00:05:31,540
or your friend
might offer to you,

122
00:05:31,540 --> 00:05:34,972
if I don't do the problem
set until Friday, 5:00 PM,

123
00:05:34,972 --> 00:05:36,430
because you want
to have fun Friday

124
00:05:36,430 --> 00:05:39,880
night, I'm going
to pay you $100.

125
00:05:39,880 --> 00:05:44,080
Now, if you make that choice,
or if somebody offers you

126
00:05:44,080 --> 00:05:46,480
that option to pay
them $100 in case

127
00:05:46,480 --> 00:05:50,350
you haven't done the problem set
by Friday 5:00 PM, that choice

128
00:05:50,350 --> 00:05:53,410
does make doesn't make
any sense if the person is

129
00:05:53,410 --> 00:05:54,880
an exponential discounter.

130
00:05:54,880 --> 00:05:57,520
That person only makes sense
if you are a present bias

131
00:05:57,520 --> 00:05:59,950
or if you have self-control
problems in some way.

132
00:05:59,950 --> 00:06:02,240
And you must be
sophisticated in some sense.

133
00:06:02,240 --> 00:06:05,200
So it must indicate some
form of sophistication.

134
00:06:05,200 --> 00:06:08,740
To be clear, it doesn't
indicate perfect sophistication.

135
00:06:08,740 --> 00:06:11,110
You might be only
partially sophisticated.

136
00:06:11,110 --> 00:06:14,410
But at least it indicates
some form of sophistication.

137
00:06:14,410 --> 00:06:18,580
I'm going to talk about partial
sophistication in a bit.

138
00:06:18,580 --> 00:06:20,980
The same awareness
issue does not

139
00:06:20,980 --> 00:06:23,530
arise with exponential
discounting.

140
00:06:23,530 --> 00:06:24,140
Why is that?

141
00:06:24,140 --> 00:06:25,540
So in some sense, if you
thought about it, like,

142
00:06:25,540 --> 00:06:28,300
if you have looked at like the
exponential discounting model,

143
00:06:28,300 --> 00:06:30,370
there's nothing about
sophistication and naiveté.

144
00:06:30,370 --> 00:06:32,037
There's no parameter
that measures that.

145
00:06:32,037 --> 00:06:33,870
And why is that?

146
00:06:33,870 --> 00:06:36,270
Why do we need like a
delta hat, or whatever?

147
00:06:39,540 --> 00:06:40,040
Yes?

148
00:06:40,040 --> 00:06:44,347
AUDIENCE: [INAUDIBLE]

149
00:06:44,347 --> 00:06:45,680
FRANK SCHILBACH: Right, exactly.

150
00:06:45,680 --> 00:06:47,925
So the whole issue only
arises because there's

151
00:06:47,925 --> 00:06:48,800
time and consistency.

152
00:06:48,800 --> 00:06:51,170
The future self wants
different things

153
00:06:51,170 --> 00:06:52,520
that the current self does.

154
00:06:52,520 --> 00:06:55,760
In the exponential discounting
model, there's no such issue.

155
00:06:55,760 --> 00:06:57,390
There's no time and consistency.

156
00:06:57,390 --> 00:06:59,030
So the future self
will always do

157
00:06:59,030 --> 00:07:00,860
what the current self
actually wants to do

158
00:07:00,860 --> 00:07:02,372
unless circumstances change.

159
00:07:02,372 --> 00:07:04,580
So we don't need any parameter
that sort of looks at,

160
00:07:04,580 --> 00:07:06,200
like, how much is
the future self

161
00:07:06,200 --> 00:07:08,720
deviating, because
that's not even

162
00:07:08,720 --> 00:07:11,120
an issue in the first place.

163
00:07:11,120 --> 00:07:15,380
OK, so then we talked about
sort of extreme assumptions.

164
00:07:15,380 --> 00:07:17,930
So these are extreme
assumptions on the two ends.

165
00:07:17,930 --> 00:07:19,340
One is, like, full naiveté.

166
00:07:19,340 --> 00:07:20,330
You're entirely naive.

167
00:07:20,330 --> 00:07:24,308
You just cannot imagine that
your beta in the future will be

168
00:07:24,308 --> 00:07:25,100
different from one.

169
00:07:25,100 --> 00:07:26,840
That's full naiveté.

170
00:07:26,840 --> 00:07:29,310
And then there is the
other extreme assumption,

171
00:07:29,310 --> 00:07:31,880
which is full sophistication,
which is beta hat

172
00:07:31,880 --> 00:07:32,912
equals beta, right?

173
00:07:32,912 --> 00:07:35,120
Essentially, it's like you
understand entirely what's

174
00:07:35,120 --> 00:07:36,950
going on with your future beta.

175
00:07:36,950 --> 00:07:41,750
But of course, there's a whole
range in between from beta to 1

176
00:07:41,750 --> 00:07:44,930
that beta hat could take.

177
00:07:44,930 --> 00:07:48,110
And that's what we refer
to as partial naiveté.

178
00:07:48,110 --> 00:07:52,280
That is to say beta hat measures
the beliefs about future beta.

179
00:07:52,280 --> 00:07:55,670
The extreme cases are
useful to think about.

180
00:07:55,670 --> 00:07:58,225
But presumably, the truth
is somewhere in between.

181
00:07:58,225 --> 00:07:59,600
So the intermediate
case might be

182
00:07:59,600 --> 00:08:01,370
sort of the most
relevant one, which

183
00:08:01,370 --> 00:08:05,515
is beta hat is in
between beta and 1.

184
00:08:05,515 --> 00:08:06,890
So that is to say,
the individual

185
00:08:06,890 --> 00:08:09,230
understands that they will
experience present bias

186
00:08:09,230 --> 00:08:09,840
in the future.

187
00:08:09,840 --> 00:08:12,530
So I know that I have some
present bias in the future.

188
00:08:12,530 --> 00:08:16,400
As an example, say,
my beta might be 0.6.

189
00:08:16,400 --> 00:08:18,500
I might think-- I
understand to some degree

190
00:08:18,500 --> 00:08:20,210
that my future beta is not 1.

191
00:08:20,210 --> 00:08:22,620
But I might think it's
like 0.8 or the like.

192
00:08:22,620 --> 00:08:25,880
So I understand that I will
be present bias in the future,

193
00:08:25,880 --> 00:08:30,990
but I underestimate the degree
of present bias in the future.

194
00:08:30,990 --> 00:08:35,240
And so if I'm partially naive--
and this is what Maya was

195
00:08:35,240 --> 00:08:36,580
saying earlier--

196
00:08:36,580 --> 00:08:39,620
I might demand commitment
devices anyway.

197
00:08:39,620 --> 00:08:43,230
I might understand I have
a problem in the future.

198
00:08:43,230 --> 00:08:45,180
So I want some
commitment devices.

199
00:08:45,180 --> 00:08:46,370
But I could also overcommit.

200
00:08:46,370 --> 00:08:48,140
I could demand
commitment devices that

201
00:08:48,140 --> 00:08:50,270
are actually not useful for me.

202
00:08:50,270 --> 00:08:50,775
Why is that?

203
00:08:50,775 --> 00:08:52,400
Because I sort of
underestimate how bad

204
00:08:52,400 --> 00:08:53,540
my self-control problem is.

205
00:08:53,540 --> 00:08:55,373
I understand there's a
self-control problem.

206
00:08:55,373 --> 00:08:57,860
Somebody offers me a commitment
device that's fairly weak.

207
00:08:57,860 --> 00:08:59,660
I say great, that's
going to help me.

208
00:08:59,660 --> 00:09:01,550
That's going to help
me follow through.

209
00:09:01,550 --> 00:09:03,500
But then, surprise, my
self-control problem

210
00:09:03,500 --> 00:09:05,762
is actually worse
than I anticipated.

211
00:09:05,762 --> 00:09:07,220
And then I have a
commitment device

212
00:09:07,220 --> 00:09:09,170
and I'm actually failing
with that commitment device,

213
00:09:09,170 --> 00:09:10,837
because the self-control
problem happens

214
00:09:10,837 --> 00:09:13,940
to be worse than anticipated.

215
00:09:13,940 --> 00:09:15,020
Any questions on this?

216
00:09:21,900 --> 00:09:25,460
OK, we're going to talk
tomorrow about, like--

217
00:09:25,460 --> 00:09:28,282
a little bit about solving
problems with partial naiveté.

218
00:09:28,282 --> 00:09:30,740
We talked about solving problems
with full naiveté and full

219
00:09:30,740 --> 00:09:31,702
sophistication.

220
00:09:31,702 --> 00:09:33,410
Partial naiveté is a
little bit trickier,

221
00:09:33,410 --> 00:09:35,327
because it sort of
requires iterating forwards

222
00:09:35,327 --> 00:09:36,450
and backwards.

223
00:09:36,450 --> 00:09:39,340
We'll talk about this
briefly tomorrow.

224
00:09:39,340 --> 00:09:41,650
OK, so now, demand for
commitment, we already

225
00:09:41,650 --> 00:09:42,850
talked about this before.

226
00:09:42,850 --> 00:09:44,750
Here is sort of a
formal definition.

227
00:09:44,750 --> 00:09:47,710
It's defined as an arrangement
entered into by an agent who

228
00:09:47,710 --> 00:09:49,450
restricts his and
her future choice set

229
00:09:49,450 --> 00:09:52,300
by making certain choices more
expensive, perhaps infinitely

230
00:09:52,300 --> 00:09:53,162
expensive.

231
00:09:53,162 --> 00:09:55,120
That's to say, at the
margin, you might sort of

232
00:09:55,120 --> 00:09:59,290
pay for something that might
make restrict your choices

233
00:09:59,290 --> 00:10:01,690
or make your choices in
the future more expensive.

234
00:10:01,690 --> 00:10:03,500
If it's not
available at all, you

235
00:10:03,500 --> 00:10:06,050
can think of this as, like,
the price is infinite, right?

236
00:10:06,050 --> 00:10:08,530
So I don't want to
eat donuts tomorrow.

237
00:10:08,530 --> 00:10:10,080
I can make donuts
more expensive--

238
00:10:10,080 --> 00:10:11,080
more and more expensive.

239
00:10:11,080 --> 00:10:12,580
If it's infinitely
expensive, donuts

240
00:10:12,580 --> 00:10:16,280
are just not available to me.

241
00:10:16,280 --> 00:10:18,550
So now of course,
as we said before,

242
00:10:18,550 --> 00:10:21,250
time-inconsistent preferences,
the selves differ.

243
00:10:21,250 --> 00:10:23,140
And this is, again, repetition.

244
00:10:23,140 --> 00:10:24,880
Selfs differ between
what you want today

245
00:10:24,880 --> 00:10:25,960
versus in the future.

246
00:10:25,960 --> 00:10:28,600
You might sort of worry about
misbehaving in the future.

247
00:10:28,600 --> 00:10:30,940
If you understand
that, you might

248
00:10:30,940 --> 00:10:33,820
want to discipline your future
self by demanding a commitment

249
00:10:33,820 --> 00:10:34,960
device.

250
00:10:34,960 --> 00:10:38,785
OK, so who of you is using
commitment devices, or can

251
00:10:38,785 --> 00:10:40,660
you give me an example
of a commitment device

252
00:10:40,660 --> 00:10:43,700
that you have used in the
past, successful or not?

253
00:10:43,700 --> 00:10:44,576
Yes?

254
00:10:44,576 --> 00:10:47,034
AUDIENCE: I have an app on my
phone where I can set a timer

255
00:10:47,034 --> 00:10:49,570
[INAUDIBLE].

256
00:10:49,570 --> 00:10:50,820
FRANK SCHILBACH: Does it work?

257
00:10:50,820 --> 00:10:53,020
AUDIENCE: Yes.

258
00:10:53,020 --> 00:10:54,950
FRANK SCHILBACH:
I should try that.

259
00:10:54,950 --> 00:10:56,060
Yes?

260
00:10:56,060 --> 00:10:59,020
AUDIENCE: I have a
browser extension

261
00:10:59,020 --> 00:11:03,022
that launch certain sites
at certain times of the day.

262
00:11:03,022 --> 00:11:04,814
FRANK SCHILBACH: Does
it work for you, too?

263
00:11:04,814 --> 00:11:06,087
AUDIENCE: Somewhat.

264
00:11:06,087 --> 00:11:07,670
FRANK SCHILBACH: Why
does it not work?

265
00:11:07,670 --> 00:11:11,870
AUDIENCE: Because it's way
too easy to just turn it off.

266
00:11:11,870 --> 00:11:13,858
FRANK SCHILBACH: Right, exactly.

267
00:11:13,858 --> 00:11:15,650
So that's an example
of a commitment device

268
00:11:15,650 --> 00:11:17,150
that's sort of
partial in some ways.

269
00:11:17,150 --> 00:11:18,980
It's not sort of strong enough.

270
00:11:18,980 --> 00:11:20,610
So either you can
sort of substitute

271
00:11:20,610 --> 00:11:23,480
to, like, Firefox or whatever,
and another browser, you

272
00:11:23,480 --> 00:11:25,940
can substitute your phone, you
can substitute your friend's

273
00:11:25,940 --> 00:11:28,100
phone even, and so on.

274
00:11:28,100 --> 00:11:31,370
Or you might just actually be
able to turn it off yourself.

275
00:11:31,370 --> 00:11:32,360
But you can circumvent.

276
00:11:32,360 --> 00:11:36,440
I think there's some apps that
have sort of options that don't

277
00:11:36,440 --> 00:11:37,860
allow you to do that at all.

278
00:11:37,860 --> 00:11:40,380
But you guys are, like,
a lot of CS majors.

279
00:11:40,380 --> 00:11:43,170
So maybe you can get
around that as well.

280
00:11:43,170 --> 00:11:44,180
Any other examples?

281
00:11:44,180 --> 00:11:45,290
Yes?

282
00:11:45,290 --> 00:11:48,290
AUDIENCE: When you go shopping,
buying more vegetables

283
00:11:48,290 --> 00:11:50,340
or something so that
you feel like, oh, I

284
00:11:50,340 --> 00:11:54,710
have to eat them, otherwise I
just wasted my time and money.

285
00:11:54,710 --> 00:11:56,960
FRANK SCHILBACH: So what
do you do actually then?

286
00:11:56,960 --> 00:11:59,030
So what's your
commitment device?

287
00:11:59,030 --> 00:12:00,950
AUDIENCE: I guess it's
like making the choice

288
00:12:00,950 --> 00:12:03,240
to eat something else,
like, more expensive.

289
00:12:03,240 --> 00:12:05,532
Because I'd have to, like,
go back to the grocery store

290
00:12:05,532 --> 00:12:07,130
and I'd have to pay for it.

291
00:12:07,130 --> 00:12:09,530
FRANK SCHILBACH: But
how do you commit then?

292
00:12:09,530 --> 00:12:11,300
Or, like, what's
restricting your options?

293
00:12:11,300 --> 00:12:13,615
AUDIENCE: The fact
that I'm lazy and I

294
00:12:13,615 --> 00:12:16,860
don't want to go somewhere
to get other food.

295
00:12:16,860 --> 00:12:19,210
FRANK SCHILBACH:
But, like, so you

296
00:12:19,210 --> 00:12:21,710
know that you're going to do
that in the future potentially.

297
00:12:21,710 --> 00:12:24,810
So now, how do you avoid or
change your future behavior?

298
00:12:24,810 --> 00:12:27,080
Can you sort of
incentivize yourself?

299
00:12:27,080 --> 00:12:30,080
Or can you make sure that--
another version of that

300
00:12:30,080 --> 00:12:31,940
would be, like, so
what people often

301
00:12:31,940 --> 00:12:35,780
do is when they buy,
for example, potato

302
00:12:35,780 --> 00:12:38,990
chips or the like, they buy
really small bags, in part

303
00:12:38,990 --> 00:12:41,900
sort of knowing that if they buy
a big bag, which actually would

304
00:12:41,900 --> 00:12:43,837
be cheaper, they
would just eat it all.

305
00:12:43,837 --> 00:12:45,920
So then, you have, like,
only very small portions.

306
00:12:45,920 --> 00:12:47,660
And then if you
want another one,

307
00:12:47,660 --> 00:12:50,060
you have to sort of go to
the store and buy more.

308
00:12:50,060 --> 00:12:51,560
So that's like a
version-- it's sort

309
00:12:51,560 --> 00:12:53,420
of a version of a commitment
device, where essentially, you

310
00:12:53,420 --> 00:12:55,820
commit yourself to, if you
want to eat more potato

311
00:12:55,820 --> 00:12:57,560
chips in the future,
you have to go back

312
00:12:57,560 --> 00:12:59,227
to the store as opposed
to having, like,

313
00:12:59,227 --> 00:13:02,430
a big bag that you can just
consume in one session.

314
00:13:02,430 --> 00:13:03,350
Yeah.

315
00:13:03,350 --> 00:13:05,450
Any other examples?

316
00:13:05,450 --> 00:13:06,045
Yes?

317
00:13:06,045 --> 00:13:09,180
AUDIENCE: Carrying
cash instead of card.

318
00:13:09,180 --> 00:13:12,742
Like, if you [INAUDIBLE] cash
on you, you won't overspend.

319
00:13:12,742 --> 00:13:13,700
FRANK SCHILBACH: I see.

320
00:13:13,700 --> 00:13:15,680
And so that's interesting.

321
00:13:15,680 --> 00:13:16,340
Yeah, exactly.

322
00:13:16,340 --> 00:13:18,023
So, like, credit,
it's interesting.

323
00:13:18,023 --> 00:13:19,940
Because in some sense,
in some other settings,

324
00:13:19,940 --> 00:13:21,677
carrying cash is
sort of not helping.

325
00:13:21,677 --> 00:13:24,260
But you're saying, like, instead
of having a credit card where

326
00:13:24,260 --> 00:13:27,290
you can spend, usually, as much
as you like, depending on what

327
00:13:27,290 --> 00:13:29,090
your credit limit
is, you might say,

328
00:13:29,090 --> 00:13:32,180
I'm going to go out with
$100, and, like, once I've

329
00:13:32,180 --> 00:13:34,160
spent the $100, I'm not
going to spend more.

330
00:13:34,160 --> 00:13:36,620
I have to sort of go
back home or the like.

331
00:13:36,620 --> 00:13:39,390
And then, I might sort of
not give in to temptations.

332
00:13:39,390 --> 00:13:41,780
The reason I was
hesitating a little bit

333
00:13:41,780 --> 00:13:43,580
is, like, in
developing countries,

334
00:13:43,580 --> 00:13:47,120
in part, in some settings
where I work, lots of people

335
00:13:47,120 --> 00:13:49,310
have lots of cash on
hand from their work.

336
00:13:49,310 --> 00:13:51,500
For example, cycle
rickshaw drivers,

337
00:13:51,500 --> 00:13:54,048
they would have a
lot of cash on hand

338
00:13:54,048 --> 00:13:55,340
because they get, like, trips--

339
00:13:55,340 --> 00:13:57,043
they get paid for
every single trip.

340
00:13:57,043 --> 00:13:58,710
And then they have
lots of cash on hand.

341
00:13:58,710 --> 00:14:01,002
It's actually quite bad,
because then they can spend it

342
00:14:01,002 --> 00:14:03,230
on lots of things any day.

343
00:14:03,230 --> 00:14:07,070
And having it in a liquid,
or their savings account

344
00:14:07,070 --> 00:14:10,160
or the like, would be a
different form of commitment

345
00:14:10,160 --> 00:14:12,320
device that might help them.

346
00:14:12,320 --> 00:14:14,780
There was more-- yes?

347
00:14:14,780 --> 00:14:16,670
AUDIENCE: There's a
show, Nathan For You,

348
00:14:16,670 --> 00:14:18,972
where he wants to help
people lose weight.

349
00:14:18,972 --> 00:14:21,810
So he takes a picture of
them-- like a very embarrassing

350
00:14:21,810 --> 00:14:22,760
picture--

351
00:14:22,760 --> 00:14:25,376
and gets a notarized letter
that gets sent out in two weeks

352
00:14:25,376 --> 00:14:27,200
if they don't lose five pounds.

353
00:14:27,200 --> 00:14:28,855
FRANK SCHILBACH:
And does it work?

354
00:14:28,855 --> 00:14:29,480
AUDIENCE: Yeah.

355
00:14:29,480 --> 00:14:30,673
[LAUGHTER]

356
00:14:30,673 --> 00:14:32,090
FRANK SCHILBACH:
Wow, interesting.

357
00:14:32,090 --> 00:14:32,870
I've not heard of this.

358
00:14:32,870 --> 00:14:33,440
What is it called?

359
00:14:33,440 --> 00:14:34,470
AUDIENCE: That's how
embarrassing the picture is.

360
00:14:34,470 --> 00:14:35,990
FRANK SCHILBACH: I see.

361
00:14:35,990 --> 00:14:37,520
Yeah, you can see, like--

362
00:14:37,520 --> 00:14:40,970
yeah, it depends a
little bit, I guess--

363
00:14:40,970 --> 00:14:45,140
yeah, so when asked does it
work, does it work on average,

364
00:14:45,140 --> 00:14:47,282
or what's being
shown, or how costly

365
00:14:47,282 --> 00:14:48,740
is it to fail--
but it sounds like,

366
00:14:48,740 --> 00:14:52,044
at least for some people,
that's, in fact, effective.

367
00:14:52,044 --> 00:14:52,545
Yeah?

368
00:14:52,545 --> 00:14:54,170
AUDIENCE: There are
some accounts where

369
00:14:54,170 --> 00:14:56,240
you can put money,
and it's a bit harder

370
00:14:56,240 --> 00:14:58,490
to take that money
out [INAUDIBLE]..

371
00:14:58,490 --> 00:15:03,110
So that's kind of [INAUDIBLE]
and you can take it out

372
00:15:03,110 --> 00:15:03,610
[INAUDIBLE].

373
00:15:03,610 --> 00:15:03,860
FRANK SCHILBACH: Right.

374
00:15:03,860 --> 00:15:05,735
AUDIENCE: If you really
need it [INAUDIBLE]..

375
00:15:05,735 --> 00:15:08,420
FRANK SCHILBACH: Right, in fact,
a lot of retirement accounts

376
00:15:08,420 --> 00:15:12,480
in the US, in many places, have
penalties for early withdrawal.

377
00:15:12,480 --> 00:15:16,940
It's a lot of, like, many
employers that offer 401(k)

378
00:15:16,940 --> 00:15:19,310
and other savings vehicles--

379
00:15:19,310 --> 00:15:22,730
essentially, that's like
a retirement sort of--

380
00:15:22,730 --> 00:15:25,070
tax-deferred retirement savings
often sort of subsidized

381
00:15:25,070 --> 00:15:26,090
by the employer.

382
00:15:26,090 --> 00:15:29,720
But often, one condition is
that you have to pay a 10%,

383
00:15:29,720 --> 00:15:31,385
or something, penalty
to withdraw it.

384
00:15:31,385 --> 00:15:33,260
Similarly, I think for
some savings accounts,

385
00:15:33,260 --> 00:15:34,370
that's sort of similar.

386
00:15:34,370 --> 00:15:39,320
And the idea is very much, like,
helping you resist temptations

387
00:15:39,320 --> 00:15:41,240
to change your plan.

388
00:15:41,240 --> 00:15:43,940
So you're going to plan
for saving for retirement

389
00:15:43,940 --> 00:15:44,990
or something else.

390
00:15:44,990 --> 00:15:47,690
But in fact, you might
sort of withdraw early

391
00:15:47,690 --> 00:15:48,860
if you're tempted.

392
00:15:48,860 --> 00:15:50,900
And then usually,
the penalty is only

393
00:15:50,900 --> 00:15:52,693
something like 10% or the like.

394
00:15:52,693 --> 00:15:55,110
Because when people have actual
shocks, like health shocks

395
00:15:55,110 --> 00:15:57,480
or other issues, they want
to be able to take out

396
00:15:57,480 --> 00:15:59,892
the money at some penalty.

397
00:15:59,892 --> 00:16:01,350
So they don't want
to have it to be

398
00:16:01,350 --> 00:16:05,790
entirely liquid, because
that would be bad for them.

399
00:16:05,790 --> 00:16:07,230
Any other example?

400
00:16:07,230 --> 00:16:08,850
Yes?

401
00:16:08,850 --> 00:16:11,540
AUDIENCE: I have an
alarm clock on my phone

402
00:16:11,540 --> 00:16:13,705
where I have to take a
picture of something.

403
00:16:13,705 --> 00:16:15,330
So I'll get up, and
go to the bathroom,

404
00:16:15,330 --> 00:16:17,245
and take a picture of the thing.

405
00:16:17,245 --> 00:16:18,495
FRANK SCHILBACH: Does it work?

406
00:16:18,495 --> 00:16:20,745
AUDIENCE: No, I take the
picture and I go back to bed.

407
00:16:20,745 --> 00:16:23,460
FRANK SCHILBACH: I
see, fair enough.

408
00:16:23,460 --> 00:16:24,920
Does it work sometimes?

409
00:16:24,920 --> 00:16:25,920
AUDIENCE: Yeah, kind of.

410
00:16:25,920 --> 00:16:27,463
FRANK SCHILBACH:
Sometimes, I see.

411
00:16:27,463 --> 00:16:29,130
Yeah, so there the
issue, in some sense,

412
00:16:29,130 --> 00:16:31,110
is, like, if it
doesn't work, then

413
00:16:31,110 --> 00:16:32,700
you're kind of worse
off than before,

414
00:16:32,700 --> 00:16:36,730
because you slept worse than
you would have otherwise.

415
00:16:36,730 --> 00:16:40,240
And you know, there's no benefit
of you getting actually up.

416
00:16:40,240 --> 00:16:41,100
Yeah.

417
00:16:41,100 --> 00:16:42,300
One more?

418
00:16:42,300 --> 00:16:42,878
Yes?

419
00:16:42,878 --> 00:16:45,045
AUDIENCE: Would, like,
societally implied commitment

420
00:16:45,045 --> 00:16:46,587
about-- it's like,
No Shave November,

421
00:16:46,587 --> 00:16:48,788
would some sort of
variance on that work?

422
00:16:48,788 --> 00:16:50,205
FRANK SCHILBACH:
Of what, exactly?

423
00:16:50,205 --> 00:16:52,710
AUDIENCE: Like No Shave
November or variants of this,

424
00:16:52,710 --> 00:16:55,230
would that imply a
commitment device?

425
00:16:55,230 --> 00:16:58,210
FRANK SCHILBACH: Yeah, I
mean, I think in some sense,

426
00:16:58,210 --> 00:16:59,280
there's often--

427
00:16:59,280 --> 00:17:01,530
some commitment devices are
essentially sort of public

428
00:17:01,530 --> 00:17:04,388
attention or the like, where
people are announcing one way

429
00:17:04,388 --> 00:17:06,930
or the other publicly-- which
is kind of what you were saying

430
00:17:06,930 --> 00:17:08,310
earlier--

431
00:17:08,310 --> 00:17:12,766
where society, in some sense,
has some influence where people

432
00:17:12,766 --> 00:17:14,849
would say they would
publicly declare that they're

433
00:17:14,849 --> 00:17:15,724
going to lose weight.

434
00:17:15,724 --> 00:17:18,460
They would publicly declare that
they're going to save money,

435
00:17:18,460 --> 00:17:18,970
and so on.

436
00:17:18,970 --> 00:17:20,595
And then, there would
be social shaming

437
00:17:20,595 --> 00:17:22,700
in some way or the
other to do that.

438
00:17:22,700 --> 00:17:24,670
Now, what you want--
so in some sense,

439
00:17:24,670 --> 00:17:26,910
if it's just society
imposing things on people,

440
00:17:26,910 --> 00:17:27,690
that's not enough.

441
00:17:27,690 --> 00:17:29,130
What you want is
a person actively

442
00:17:29,130 --> 00:17:31,770
sort of making some
announcement or some statements

443
00:17:31,770 --> 00:17:35,970
with a goal of making it more
costly not to follow through.

444
00:17:35,970 --> 00:17:39,390
OK, so as I said before,
demand for commitment

445
00:17:39,390 --> 00:17:42,120
requires at least some
partial sophistication.

446
00:17:42,120 --> 00:17:43,615
I have some examples
for you here.

447
00:17:43,615 --> 00:17:44,490
One is, like, StickK.

448
00:17:44,490 --> 00:17:45,600
I don't know if you
have any of-- has

449
00:17:45,600 --> 00:17:46,558
any of you used StickK?

450
00:17:50,490 --> 00:17:51,090
No?

451
00:17:51,090 --> 00:17:55,200
OK, so StickK is a website
founded by an academic, Dean

452
00:17:55,200 --> 00:17:56,730
Karlan and co-authors.

453
00:17:56,730 --> 00:17:59,610
What StickK does is,
it's a commitment device

454
00:17:59,610 --> 00:18:02,700
that works, in some sense, in
the way I was saying before.

455
00:18:02,700 --> 00:18:06,825
For example, if you want to of
commit to certain behaviors,

456
00:18:06,825 --> 00:18:09,030
for example, if I
wanted to finish a paper

457
00:18:09,030 --> 00:18:12,180
draft by Friday night, I
would have to find a referee.

458
00:18:12,180 --> 00:18:14,580
I would, for example, ask
Aaron to be my referee.

459
00:18:14,580 --> 00:18:16,890
And I would say, Aaron,
I'm going to give you

460
00:18:16,890 --> 00:18:18,840
my credit card information.

461
00:18:18,840 --> 00:18:23,850
And if I don't finish the
paper by Friday night,

462
00:18:23,850 --> 00:18:27,570
I'm going to pay $100 either
to Aaron or to some charity--

463
00:18:27,570 --> 00:18:31,540
could be anti charity, could be
the pro-smoking, pro-whatever

464
00:18:31,540 --> 00:18:33,150
society.

465
00:18:33,150 --> 00:18:35,410
The money is going to be gone.

466
00:18:35,410 --> 00:18:36,870
And so then, Friday night comes.

467
00:18:36,870 --> 00:18:41,208
Aaron is the referee who
can then decide, or will

468
00:18:41,208 --> 00:18:43,500
decide, then, whether I have
actually followed through.

469
00:18:43,500 --> 00:18:44,730
Did I actually write a draft?

470
00:18:44,730 --> 00:18:47,220
Is there an actual
paper that's done?

471
00:18:47,220 --> 00:18:49,750
And if it's done,
then nothing happens.

472
00:18:49,750 --> 00:18:51,030
I don't have to pay any money.

473
00:18:51,030 --> 00:18:53,520
If it's not done,
the money is gone.

474
00:18:53,520 --> 00:18:56,940
And since the credit card
information is on the website,

475
00:18:56,940 --> 00:18:59,430
they can actually just
withdraw that money.

476
00:18:59,430 --> 00:19:02,617
What's the problem with this
commitment device, potentially?

477
00:19:05,250 --> 00:19:05,750
Yes?

478
00:19:05,750 --> 00:19:07,950
AUDIENCE: The referee not
wanting to enforce it?

479
00:19:07,950 --> 00:19:09,200
FRANK SCHILBACH: Yes, exactly.

480
00:19:09,200 --> 00:19:12,110
So, like, I tried to do
this a few times in school.

481
00:19:12,110 --> 00:19:13,423
And my friend was my referee.

482
00:19:13,423 --> 00:19:14,840
And then, Friday
night would come.

483
00:19:14,840 --> 00:19:17,298
And then I would sort of just
convince my friend to, like--

484
00:19:17,298 --> 00:19:21,823
I'll pay you dinner with
if you let me get through.

485
00:19:21,823 --> 00:19:23,240
So you need a very
good friend who

486
00:19:23,240 --> 00:19:24,698
is actually sort
of then insisting,

487
00:19:24,698 --> 00:19:26,600
and sort of credible,
potentially.

488
00:19:26,600 --> 00:19:29,150
Or your enemy could be your
referee, if you wanted,

489
00:19:29,150 --> 00:19:31,260
who is happy sort of
then to take your money.

490
00:19:31,260 --> 00:19:32,510
But I think it's worth trying.

491
00:19:32,510 --> 00:19:34,343
You can try it out and
see whether it works.

492
00:19:34,343 --> 00:19:37,820
I'd love to hear some thoughts
on whether it works or not.

493
00:19:37,820 --> 00:19:42,020
There's Clocky which I
think is coming out of MIT.

494
00:19:42,020 --> 00:19:43,700
It's an alarm clock
that runs away,

495
00:19:43,700 --> 00:19:48,365
which is kind of similar to
what you mentioned earlier.

496
00:19:48,365 --> 00:19:50,990
There's also Tocky, which is the
alarm clock that jumps around,

497
00:19:50,990 --> 00:19:52,682
and then you have
to sort of find it.

498
00:19:52,682 --> 00:19:54,068
[LAUGHTER]

499
00:19:54,068 --> 00:19:56,690
There are some forms of--

500
00:19:56,690 --> 00:19:58,532
this is a fake alarm clock--

501
00:19:58,532 --> 00:20:00,740
I don't think it actually
exists-- where essentially,

502
00:20:00,740 --> 00:20:03,860
if you don't get up and
press a button on your alarm,

503
00:20:03,860 --> 00:20:05,630
you essentially give
money to charity

504
00:20:05,630 --> 00:20:08,720
or anti-charities if you want.

505
00:20:08,720 --> 00:20:11,340
There's a thing called Antabuse,
which is quite interesting,

506
00:20:11,340 --> 00:20:14,720
which is a drug that
essentially interferes

507
00:20:14,720 --> 00:20:17,250
with the metabolism of alcohol.

508
00:20:17,250 --> 00:20:19,250
So this is some people
have trouble metabolizing

509
00:20:19,250 --> 00:20:20,178
alcohol anyway.

510
00:20:20,178 --> 00:20:21,470
So when they drink, they flush.

511
00:20:21,470 --> 00:20:25,080
They turn-- they get headaches.

512
00:20:25,080 --> 00:20:27,390
They have to throw up, and
feel unwell, and so on.

513
00:20:27,390 --> 00:20:29,510
So for them, drinking is
very unpleasant as it is.

514
00:20:29,510 --> 00:20:31,010
It turns out,
there's a drug that

515
00:20:31,010 --> 00:20:36,590
can actually sort of reproduce
those kinds of behaviors

516
00:20:36,590 --> 00:20:38,150
or feelings.

517
00:20:38,150 --> 00:20:40,480
And it's essentially--
you can take it.

518
00:20:40,480 --> 00:20:42,650
And then for the
next 24 to 48 hours,

519
00:20:42,650 --> 00:20:44,548
you will not be able
to metabolize alcohol.

520
00:20:44,548 --> 00:20:45,965
There's also
versions of that that

521
00:20:45,965 --> 00:20:48,113
are sort of implants
for a month or the like.

522
00:20:48,113 --> 00:20:50,030
But essentially, they
make alcohol consumption

523
00:20:50,030 --> 00:20:50,930
very unpleasant.

524
00:20:50,930 --> 00:20:53,680
And this is sort of
meant to help people

525
00:20:53,680 --> 00:20:55,880
who have problems with
alcohol consumption,

526
00:20:55,880 --> 00:20:58,370
that if you don't want
to drink tomorrow,

527
00:20:58,370 --> 00:21:02,510
or in two days from now, or
the like, or the next month,

528
00:21:02,510 --> 00:21:06,240
you can take it and then
commit to not doing that.

529
00:21:06,240 --> 00:21:09,710
And then alcohol consumption
becomes very unpleasant

530
00:21:09,710 --> 00:21:11,780
when consuming alcohol anyway.

531
00:21:11,780 --> 00:21:13,310
It's a little bit dangerous.

532
00:21:13,310 --> 00:21:15,950
Or it can be quite dangerous
if people then drink anyway,

533
00:21:15,950 --> 00:21:17,840
because there's sort
of an alcohol antabuse

534
00:21:17,840 --> 00:21:21,210
sort of reaction
that's quite dangerous.

535
00:21:21,210 --> 00:21:23,300
So it hasn't been
particularly successful,

536
00:21:23,300 --> 00:21:26,450
in part also because people
sort of fall off the bandwagon.

537
00:21:26,450 --> 00:21:29,060
So if you essentially
take that daily, or, like,

538
00:21:29,060 --> 00:21:31,100
every second day,
what happens then

539
00:21:31,100 --> 00:21:33,317
is that people start
wanting to drink.

540
00:21:33,317 --> 00:21:35,150
And then they just stop
taking the Antabuse.

541
00:21:35,150 --> 00:21:35,960
And then they have
to sort of just

542
00:21:35,960 --> 00:21:37,910
wait it out for 24
hours or the like

543
00:21:37,910 --> 00:21:39,930
until the drug is
out of their system.

544
00:21:39,930 --> 00:21:42,177
And then they drink anyway.

545
00:21:42,177 --> 00:21:44,510
But in principle, it's like
the ideal commitment device.

546
00:21:44,510 --> 00:21:46,343
Because essentially,
the only reason for you

547
00:21:46,343 --> 00:21:51,740
to take this drug would be to
reduce your future drinking.

548
00:21:51,740 --> 00:21:54,560
There's also sort of various
versions of self-control

549
00:21:54,560 --> 00:21:58,615
devices which essentially either
restrict certain websites that

550
00:21:58,615 --> 00:21:59,240
you can go to--

551
00:21:59,240 --> 00:22:01,580
you can restrict yourself--
or it restricts them

552
00:22:01,580 --> 00:22:03,730
to certain hours, or you
can sort of give yourself

553
00:22:03,730 --> 00:22:06,680
a budget for a number
of hours that you

554
00:22:06,680 --> 00:22:09,260
can use certain websites for.

555
00:22:09,260 --> 00:22:10,715
Any questions?

556
00:22:16,760 --> 00:22:19,700
So I also have a video
of commitment devices,

557
00:22:19,700 --> 00:22:22,280
which I'm not sure whether
the audio actually works.

558
00:22:22,280 --> 00:22:23,460
So we have to try this out.

559
00:22:23,460 --> 00:22:26,510
If not, I'll show
it to you tomorrow.

560
00:22:26,510 --> 00:22:29,750
OK, so as you can see,
I think we can learn

561
00:22:29,750 --> 00:22:31,580
a lot from watching movies.

562
00:22:31,580 --> 00:22:32,240
And I love TV.

563
00:22:32,240 --> 00:22:34,690
So what can we learn from this?

564
00:22:34,690 --> 00:22:37,190
I think that it can actually--
we can learn a lot about sort

565
00:22:37,190 --> 00:22:39,260
of commitment devices,
and whether they work,

566
00:22:39,260 --> 00:22:41,290
and why they work, or
why they might not work.

567
00:22:41,290 --> 00:22:42,950
So what did what did we learn?

568
00:22:45,720 --> 00:22:47,002
Yes?

569
00:22:47,002 --> 00:22:50,404
AUDIENCE: [INAUDIBLE] reversible
commitment devices are not

570
00:22:50,404 --> 00:22:52,445
as effective.

571
00:22:52,445 --> 00:22:54,070
FRANK SCHILBACH:
Right, so just making,

572
00:22:54,070 --> 00:22:56,910
like, future choices more
expensive might not be enough,

573
00:22:56,910 --> 00:22:57,410
right?

574
00:22:57,410 --> 00:22:59,860
I might sort of say I might
have self-control problems.

575
00:22:59,860 --> 00:23:02,545
Now, I'm going to increase the
price of my future behavior.

576
00:23:02,545 --> 00:23:04,420
But that doesn't mean
that it actually works.

577
00:23:04,420 --> 00:23:05,500
It might just be more costly.

578
00:23:05,500 --> 00:23:07,458
Now, I have to climb up
the ladder or whatever.

579
00:23:07,458 --> 00:23:11,702
But then I might still eat the
cookie, and that's not helping.

580
00:23:11,702 --> 00:23:12,660
What else did we learn?

581
00:23:17,620 --> 00:23:18,358
Yeah?

582
00:23:18,358 --> 00:23:20,400
AUDIENCE: But I think you
have to find a balance.

583
00:23:20,400 --> 00:23:22,240
Because if you make
the penalty too high,

584
00:23:22,240 --> 00:23:24,540
you're not going to use
the commitment device.

585
00:23:24,540 --> 00:23:25,300
FRANK SCHILBACH: Right, exactly.

586
00:23:25,300 --> 00:23:27,370
So what they did, essentially,
they made the penalty--

587
00:23:27,370 --> 00:23:28,787
or they essentially
made the price

588
00:23:28,787 --> 00:23:29,950
of cookies infinitely high.

589
00:23:29,950 --> 00:23:31,750
They just sort of gave
them away to the birds.

590
00:23:31,750 --> 00:23:34,083
Now they don't have any cookies
at all, and sort of now,

591
00:23:34,083 --> 00:23:36,080
they're going to eat
no cookies whatsoever,

592
00:23:36,080 --> 00:23:40,150
which is kind of not what
they wanted either, right?

593
00:23:40,150 --> 00:23:43,060
AUDIENCE: But I think
another lesson is that once

594
00:23:43,060 --> 00:23:46,625
you've made one of
these commitment devices

595
00:23:46,625 --> 00:23:50,680
so costly that you
actually just transfer it

596
00:23:50,680 --> 00:23:52,250
to another temptation
[INAUDIBLE]..

597
00:23:52,250 --> 00:23:53,500
FRANK SCHILBACH: Yes, exactly.

598
00:23:53,500 --> 00:23:54,730
That's what I was
saying earlier.

599
00:23:54,730 --> 00:23:56,260
You might sort of
restrict your browser.

600
00:23:56,260 --> 00:23:58,343
You might sort of, like,
shut down Chrome, or even

601
00:23:58,343 --> 00:23:59,605
your entire laptop.

602
00:23:59,605 --> 00:24:01,480
But then, if you start
surfing on your phone,

603
00:24:01,480 --> 00:24:02,737
that's not really helping.

604
00:24:02,737 --> 00:24:04,570
So in some sense there's,
like, substitution

605
00:24:04,570 --> 00:24:08,110
to different potential either
sort of goods, or devices,

606
00:24:08,110 --> 00:24:10,840
and so on, or technologies
that might sort of undo

607
00:24:10,840 --> 00:24:12,310
your commitment device.

608
00:24:12,310 --> 00:24:15,040
So they need to be foolproof
in some sense, in the sense of,

609
00:24:15,040 --> 00:24:17,920
like, being able to not
substitute to other things.

610
00:24:24,140 --> 00:24:27,530
OK, so first, substitution
across temptation goods

611
00:24:27,530 --> 00:24:29,360
can mitigate the
usefulness of commitment.

612
00:24:29,360 --> 00:24:31,882
If, essentially, you have
other vices or other things

613
00:24:31,882 --> 00:24:33,590
you're going to do
instead, then it's not

614
00:24:33,590 --> 00:24:36,560
really helpful to shut
down one thing only.

615
00:24:36,560 --> 00:24:40,160
So you have to think about, if
I don't do the behavior that I'm

616
00:24:40,160 --> 00:24:42,620
trying to prevent, is there
sort of other behaviors

617
00:24:42,620 --> 00:24:46,020
that am I going to
substitute towards,

618
00:24:46,020 --> 00:24:49,740
and then it's not going
to be helpful either?

619
00:24:49,740 --> 00:24:52,880
Now, it's also helpful to think
about what is it actually--

620
00:24:52,880 --> 00:24:55,670
what is it actually required
for a commitment device

621
00:24:55,670 --> 00:24:57,253
to be helpful to a person?

622
00:24:57,253 --> 00:24:59,420
So first, kind of obviously
in some ways, the person

623
00:24:59,420 --> 00:25:01,800
needs to have some self-control
problem in the first place.

624
00:25:01,800 --> 00:25:04,342
This needs to be something that,
in the future, some behavior

625
00:25:04,342 --> 00:25:05,362
you want to change.

626
00:25:05,362 --> 00:25:07,570
Second, there needs to be
some sophistication, right?

627
00:25:07,570 --> 00:25:09,020
Like Frog and Toad,
they kind of know

628
00:25:09,020 --> 00:25:10,812
that they're going to
eat too many cookies.

629
00:25:10,812 --> 00:25:12,380
They need to sort
of understand that.

630
00:25:12,380 --> 00:25:15,240
Third, and this is
what was said earlier,

631
00:25:15,240 --> 00:25:17,240
the commitment device
needs to be effective.

632
00:25:17,240 --> 00:25:19,490
We need to have something
that's sort of strong enough

633
00:25:19,490 --> 00:25:24,860
to be able to help us overcome
our self-control problems.

634
00:25:24,860 --> 00:25:26,563
And then fourth,
it needs to be--

635
00:25:26,563 --> 00:25:27,980
the person needs
to actually think

636
00:25:27,980 --> 00:25:29,733
that the device is effective.

637
00:25:29,733 --> 00:25:31,400
If I sort of don't
have faith in myself,

638
00:25:31,400 --> 00:25:35,240
I think this is not useful, then
I'm never going to take it up.

639
00:25:35,240 --> 00:25:39,170
Or conversely, I guess, there's
issues potentially with, like,

640
00:25:39,170 --> 00:25:40,160
naiveté.

641
00:25:40,160 --> 00:25:42,050
There's two things
that naiveté could do.

642
00:25:42,050 --> 00:25:43,980
So one part in naiveté--

643
00:25:43,980 --> 00:25:45,990
what I was already
talking about before--

644
00:25:45,990 --> 00:25:48,317
which is to say, I might
be naive in a sense

645
00:25:48,317 --> 00:25:49,400
I'm saying, like, I know--

646
00:25:49,400 --> 00:25:51,442
I don't have-- if I'm
fully naive, I might think,

647
00:25:51,442 --> 00:25:52,942
I don't have a
self-control problem.

648
00:25:52,942 --> 00:25:53,790
It's not really bad.

649
00:25:53,790 --> 00:25:55,070
So I don't really need
a commitment device

650
00:25:55,070 --> 00:25:56,300
in the first place.

651
00:25:56,300 --> 00:25:59,160
That's one version where I
sort of under demand commitment

652
00:25:59,160 --> 00:26:00,890
by saying I don't
actually need it.

653
00:26:00,890 --> 00:26:03,472
It's not helpful for me anyway.

654
00:26:03,472 --> 00:26:04,680
In the future, I will behave.

655
00:26:04,680 --> 00:26:06,290
So I might not
demand commitment.

656
00:26:06,290 --> 00:26:08,450
Another version is to
say, I might actually

657
00:26:08,450 --> 00:26:11,720
demand commitment devices where
I overcommit myself in a sense,

658
00:26:11,720 --> 00:26:14,270
like, I think the commitment
device is actually helpful

659
00:26:14,270 --> 00:26:15,920
when in fact it is not.

660
00:26:15,920 --> 00:26:17,630
So Frog and Toad
were not doing that.

661
00:26:17,630 --> 00:26:19,463
In some sense, they
sort of like understood,

662
00:26:19,463 --> 00:26:22,250
to some degree at least,
pretty well what they're

663
00:26:22,250 --> 00:26:23,430
going to do in the future.

664
00:26:23,430 --> 00:26:25,730
I think they sort of
underestimated the substitution

665
00:26:25,730 --> 00:26:27,438
in this case, in the
sense of, like, they

666
00:26:27,438 --> 00:26:31,680
thought giving the cookies to
the birds would be helpful.

667
00:26:31,680 --> 00:26:34,240
But in fact, then they
sort of underestimated

668
00:26:34,240 --> 00:26:35,360
the substitution.

669
00:26:35,360 --> 00:26:37,040
And in the end, you know,
they're going to be worse off.

670
00:26:37,040 --> 00:26:39,415
Because now, he's going to
make the cake anyway, and then

671
00:26:39,415 --> 00:26:40,190
eat a lot of cake.

672
00:26:40,190 --> 00:26:43,850
He could have as well just eaten
the cookies in the first place.

673
00:26:43,850 --> 00:26:44,870
Any questions on this?

674
00:26:50,430 --> 00:26:54,740
OK, so now, let me sort of turn
toward some academic papers,

675
00:26:54,740 --> 00:26:57,990
and sort of actual
empirical setups.

676
00:26:57,990 --> 00:27:00,800
So this is the paper that
you, I hope, all read,

677
00:27:00,800 --> 00:27:02,780
which is Ariely and Wertenbroch.

678
00:27:02,780 --> 00:27:05,540
Dan Ariely was actually a
professor here at Sloan,

679
00:27:05,540 --> 00:27:09,350
and did a bunch of experiments,
mostly with MIT students

680
00:27:09,350 --> 00:27:10,990
at Sloan--

681
00:27:10,990 --> 00:27:12,380
MBAs.

682
00:27:12,380 --> 00:27:16,310
And so these are 51
executives at Sloan.

683
00:27:16,310 --> 00:27:18,410
These are sort of highly
incentivized individuals,

684
00:27:18,410 --> 00:27:20,180
in the sense that
they take classes,

685
00:27:20,180 --> 00:27:21,920
and if they don't--
if they fail a class,

686
00:27:21,920 --> 00:27:24,170
if they do badly in a class,
they have to actually pay

687
00:27:24,170 --> 00:27:24,960
for it themselves.

688
00:27:24,960 --> 00:27:27,773
So they really are
incentivized to do well.

689
00:27:27,773 --> 00:27:29,690
They had to submit three
papers in their class

690
00:27:29,690 --> 00:27:31,970
to get a 1% grade penalty
for late submissions, which

691
00:27:31,970 --> 00:27:34,200
is quite a bit.

692
00:27:34,200 --> 00:27:35,330
There were two groups.

693
00:27:35,330 --> 00:27:37,220
Group A had
evenly-spaced deadlines.

694
00:27:37,220 --> 00:27:40,520
Group B had the option to
set their own deadlines.

695
00:27:40,520 --> 00:27:46,440
And now, the first
sort of result

696
00:27:46,440 --> 00:27:48,860
here is that there was demand
for commitment in the sense

697
00:27:48,860 --> 00:27:52,970
that, like, 68% of
people chose deadlines

698
00:27:52,970 --> 00:27:54,230
prior to the last week.

699
00:27:54,230 --> 00:27:55,730
So when given the
option when do you

700
00:27:55,730 --> 00:27:58,188
want to set your deadlines over
the course of the semester,

701
00:27:58,188 --> 00:28:02,540
68% of people chose deadlines
that were not the last week.

702
00:28:02,540 --> 00:28:04,316
Why is that demand
for commitment?

703
00:28:07,910 --> 00:28:08,410
Yes?

704
00:28:08,410 --> 00:28:09,548
AUDIENCE: More
choices as to when

705
00:28:09,548 --> 00:28:11,830
to do the paper, just to
make a deadline at the end.

706
00:28:11,830 --> 00:28:13,270
FRANK SCHILBACH: Exactly,
like, you-- there's

707
00:28:13,270 --> 00:28:15,728
no reason, unless you have sort
of issues with self-control

708
00:28:15,728 --> 00:28:19,360
and the like, you just
make it essentially costly

709
00:28:19,360 --> 00:28:20,710
for you to submit late.

710
00:28:20,710 --> 00:28:23,110
If you are worried about
submitting everything late,

711
00:28:23,110 --> 00:28:26,050
that's a good thing for you
to do, and it might help you.

712
00:28:26,050 --> 00:28:29,380
Again, notice that it
requires some sophistication.

713
00:28:29,380 --> 00:28:32,200
You need to sort of understand
your future behavior

714
00:28:32,200 --> 00:28:32,883
to do that.

715
00:28:32,883 --> 00:28:34,300
Now, what they
find is, like, they

716
00:28:34,300 --> 00:28:36,940
find no late submissions,
which is interesting by itself.

717
00:28:36,940 --> 00:28:40,090
But in particular, they find
that group A has higher grades

718
00:28:40,090 --> 00:28:41,680
than group B, OK?

719
00:28:41,680 --> 00:28:45,820
Group A is the group that
has evenly-spaced deadlines,

720
00:28:45,820 --> 00:28:48,280
compared to group B
that has the option

721
00:28:48,280 --> 00:28:50,497
to set their own deadlines.

722
00:28:50,497 --> 00:28:52,330
It is consistent with
self-control problems,

723
00:28:52,330 --> 00:28:55,420
in the sense that people
set their own deadlines that

724
00:28:55,420 --> 00:28:57,130
are early.

725
00:28:57,130 --> 00:28:58,570
What else is it consistent with?

726
00:28:58,570 --> 00:29:00,678
What did we learn from that?

727
00:29:00,678 --> 00:29:02,970
What do we learn from the
fact that group A does better

728
00:29:02,970 --> 00:29:03,570
than group B?

729
00:29:07,460 --> 00:29:08,223
Yes?

730
00:29:08,223 --> 00:29:11,630
AUDIENCE: That the commitment
device wasn't strong enough?

731
00:29:11,630 --> 00:29:14,420
FRANK SCHILBACH: Yeah, so in
some sense, the commitment

732
00:29:14,420 --> 00:29:17,150
devices that people chose
for themselves was not

733
00:29:17,150 --> 00:29:19,340
strong enough, or,
like, weaker than

734
00:29:19,340 --> 00:29:21,080
the evenly-spaced deadlines.

735
00:29:21,080 --> 00:29:23,813
Now, notice that
evenly-spaced deadlines are

736
00:29:23,813 --> 00:29:24,980
in the choice set of people.

737
00:29:24,980 --> 00:29:27,355
Like, if you choose your own
deadlines, you can just say,

738
00:29:27,355 --> 00:29:29,327
I'm choosing
evenly-spaced deadlines.

739
00:29:29,327 --> 00:29:31,910
And you should do as well-- if
you're perfectly sophisticated,

740
00:29:31,910 --> 00:29:34,760
you would do as
well as the group A.

741
00:29:34,760 --> 00:29:36,140
But the fact that the groups--

742
00:29:36,140 --> 00:29:39,050
group B is doing worse sort of
tells us that people are not

743
00:29:39,050 --> 00:29:40,453
optimally setting deadlines.

744
00:29:40,453 --> 00:29:42,620
Either they don't-- there's
a fraction of people who

745
00:29:42,620 --> 00:29:43,910
don't choose any deadlines.

746
00:29:43,910 --> 00:29:46,910
There's at least 32% of
people who sort of say,

747
00:29:46,910 --> 00:29:48,330
my deadlines are all at the end.

748
00:29:48,330 --> 00:29:49,910
So maybe these are naive people.

749
00:29:49,910 --> 00:29:51,710
Or even among the people
who set deadlines,

750
00:29:51,710 --> 00:29:54,470
they might choose deadlines set
further back in the semester,

751
00:29:54,470 --> 00:29:57,200
and they do worse than in
the evenly-spaced deadlines.

752
00:29:57,200 --> 00:29:58,310
So what did we learn?

753
00:29:58,310 --> 00:29:59,940
So people do set
early deadlines--

754
00:29:59,940 --> 00:30:01,080
quite a few of them.

755
00:30:01,080 --> 00:30:03,500
So we learn about people
are at least partially--

756
00:30:03,500 --> 00:30:05,062
or some people
are sophisticated.

757
00:30:05,062 --> 00:30:06,770
But they're only
partially sophisticated.

758
00:30:06,770 --> 00:30:08,990
They don't set
deadlines optimally,

759
00:30:08,990 --> 00:30:11,240
which is sort of consistent
with people being naive,

760
00:30:11,240 --> 00:30:12,200
at least in some way.

761
00:30:12,200 --> 00:30:12,700
Yeah?

762
00:30:12,700 --> 00:30:14,310
AUDIENCE: I was
wondering, does it

763
00:30:14,310 --> 00:30:18,780
make sense to try to find
the difference between not

764
00:30:18,780 --> 00:30:24,500
adequately measuring your
commitment [INAUDIBLE]

765
00:30:24,500 --> 00:30:26,860
estimating how long it would
take you to do the paper?

766
00:30:26,860 --> 00:30:28,860
FRANK SCHILBACH: Right,
that's a great question.

767
00:30:28,860 --> 00:30:31,100
So the question is, so
there's two things that

768
00:30:31,100 --> 00:30:33,350
are perhaps uncertain when
you think about the future.

769
00:30:33,350 --> 00:30:36,130
One is your present bias--
your preference parameter

770
00:30:36,130 --> 00:30:39,370
about how much work
you're going to put

771
00:30:39,370 --> 00:30:41,050
in today versus
tomorrow, and how much

772
00:30:41,050 --> 00:30:42,440
you're going to procrastinate.

773
00:30:42,440 --> 00:30:45,148
A second question is about
your beliefs about your effort

774
00:30:45,148 --> 00:30:46,690
costs in some way
or the other, which

775
00:30:46,690 --> 00:30:50,410
is how costly is it for you to
actually do the problem set,

776
00:30:50,410 --> 00:30:51,790
or write the paper, and so on.

777
00:30:51,790 --> 00:30:54,187
And you might underestimate
how tedious it is to do.

778
00:30:54,187 --> 00:30:56,020
There's sort of a large
literature on what's

779
00:30:56,020 --> 00:30:57,270
called the "planning fallacy."

780
00:30:57,270 --> 00:30:59,110
People always plan
too many things,

781
00:30:59,110 --> 00:31:03,560
and think whatever they plan
takes shorter than it actually

782
00:31:03,560 --> 00:31:04,060
does.

783
00:31:04,060 --> 00:31:05,830
Or put differently,
things always

784
00:31:05,830 --> 00:31:08,822
take longer than people
think when they make plans.

785
00:31:08,822 --> 00:31:10,780
And the odd thing is that
also is the case even

786
00:31:10,780 --> 00:31:11,990
if you take that into account.

787
00:31:11,990 --> 00:31:13,930
So in some sense, even if you're
aware of the planning fallacy,

788
00:31:13,930 --> 00:31:15,820
and you plan, and you know that
people are going to-- things

789
00:31:15,820 --> 00:31:17,950
are going to take longer,
somewhat oddly enough,

790
00:31:17,950 --> 00:31:19,810
people are even
taking longer even

791
00:31:19,810 --> 00:31:21,410
if you take that into account.

792
00:31:21,410 --> 00:31:23,440
But anyway, so the
question is kind of

793
00:31:23,440 --> 00:31:27,950
like how can we separate that.

794
00:31:27,950 --> 00:31:30,872
I think, from this evidence,
we cannot separate this.

795
00:31:30,872 --> 00:31:32,830
There are sort of like
cleaner experiments that

796
00:31:32,830 --> 00:31:34,600
are more careful about this.

797
00:31:34,600 --> 00:31:36,970
In some sense, if
it were just about

798
00:31:36,970 --> 00:31:39,010
like underestimating
effort costs,

799
00:31:39,010 --> 00:31:41,800
then you would not set your
own deadlines necessarily.

800
00:31:41,800 --> 00:31:43,780
Like, the setting
early deadlines

801
00:31:43,780 --> 00:31:46,537
is consistent with
self-control issues.

802
00:31:46,537 --> 00:31:48,370
But just from the
performance, you're right.

803
00:31:48,370 --> 00:31:51,280
That could just be people
underestimate effort costs as

804
00:31:51,280 --> 00:31:52,340
well.

805
00:31:52,340 --> 00:31:55,810
In general, I think this is hard
to separate in many situations.

806
00:31:55,810 --> 00:31:57,670
The key part about the
self-control problems

807
00:31:57,670 --> 00:32:00,670
about the time-inconsistent
is a demand for commitment

808
00:32:00,670 --> 00:32:02,140
which sort of
reveals that people

809
00:32:02,140 --> 00:32:05,530
have some form of
self-control problems related

810
00:32:05,530 --> 00:32:07,690
to time preferences.

811
00:32:07,690 --> 00:32:10,570
Any other question?

812
00:32:10,570 --> 00:32:13,600
So now, there's one concern
is the two sessions are not

813
00:32:13,600 --> 00:32:14,440
randomly assigned.

814
00:32:14,440 --> 00:32:15,898
Why is that a
problem, potentially?

815
00:32:15,898 --> 00:32:19,060
So we have, like, 51 executives.

816
00:32:19,060 --> 00:32:22,930
25 or something-- 26 are
in one and 25 are another,

817
00:32:22,930 --> 00:32:25,180
but they're not
randomly assigned

818
00:32:25,180 --> 00:32:27,310
across the two sessions.

819
00:32:27,310 --> 00:32:27,810
Yes?

820
00:32:27,810 --> 00:32:29,762
AUDIENCE: Well, if
you're just [INAUDIBLE]

821
00:32:29,762 --> 00:32:34,043
you might want to pick the
class with the [INAUDIBLE]..

822
00:32:34,043 --> 00:32:36,460
FRANK SCHILBACH: Right, so
there could be selection people

823
00:32:36,460 --> 00:32:38,110
sort of switch sessions.

824
00:32:38,110 --> 00:32:39,250
I am not sure recitations--

825
00:32:39,250 --> 00:32:42,760
I'm not sure that's
actually possible.

826
00:32:42,760 --> 00:32:45,130
But setting that-- so
that's surely one problem,

827
00:32:45,130 --> 00:32:47,800
like, selection into
the different groups.

828
00:32:47,800 --> 00:32:51,190
Setting that aside, what
other problem do we have?

829
00:32:55,180 --> 00:32:56,450
Yes?

830
00:32:56,450 --> 00:32:57,950
AUDIENCE: [INAUDIBLE]
something like

831
00:32:57,950 --> 00:32:59,556
if the two sessions
are at different times,

832
00:32:59,556 --> 00:33:01,202
and there's some
advanced class that we

833
00:33:01,202 --> 00:33:03,252
can see as one session
but not the other,

834
00:33:03,252 --> 00:33:05,435
you could end up with
more advanced students

835
00:33:05,435 --> 00:33:08,820
in [INAUDIBLE] section, which
could affect how good they

836
00:33:08,820 --> 00:33:09,960
are at writing [INAUDIBLE].

837
00:33:09,960 --> 00:33:12,300
FRANK SCHILBACH: Exactly,
there could just be other--

838
00:33:12,300 --> 00:33:14,050
sorry, this is a version of
what you were saying earlier.

839
00:33:14,050 --> 00:33:15,383
So I misunderstood a little bit.

840
00:33:15,383 --> 00:33:17,910
Exactly, there could
be, like, people are--

841
00:33:17,910 --> 00:33:20,430
for other reasons than
the actual treatment,

842
00:33:20,430 --> 00:33:21,715
people might--

843
00:33:21,715 --> 00:33:22,590
one session is early.

844
00:33:22,590 --> 00:33:23,208
One is late.

845
00:33:23,208 --> 00:33:24,000
One is on Thursday.

846
00:33:24,000 --> 00:33:25,380
One is on Friday.

847
00:33:25,380 --> 00:33:26,387
One has a better TA.

848
00:33:26,387 --> 00:33:27,720
One has a worse TA, or whatever.

849
00:33:27,720 --> 00:33:29,845
People might sort of select
into different-- so the

850
00:33:29,845 --> 00:33:34,050
characteristics that people
have when they select into

851
00:33:34,050 --> 00:33:36,120
or when they end up in
session A versus session B

852
00:33:36,120 --> 00:33:37,510
might be different.

853
00:33:37,510 --> 00:33:39,630
So what you compare is
essentially different types

854
00:33:39,630 --> 00:33:41,340
of people.

855
00:33:41,340 --> 00:33:43,440
Second, there's only two
sessions or sections.

856
00:33:43,440 --> 00:33:45,480
That's essentially
an issue of, like,

857
00:33:45,480 --> 00:33:46,852
sample size and statistics.

858
00:33:46,852 --> 00:33:48,810
They're going to talk to
one of the recitations

859
00:33:48,810 --> 00:33:50,320
about that in more detail.

860
00:33:50,320 --> 00:33:54,270
But essentially, the bottom line
is that if you randomize, add--

861
00:33:54,270 --> 00:33:56,700
or if you were to randomize,
which they haven't even

862
00:33:56,700 --> 00:33:58,830
done here, but if
you were randomize

863
00:33:58,830 --> 00:34:01,620
at the level of a section,
then if you only have,

864
00:34:01,620 --> 00:34:04,530
like, two sections, that's
essentially not enough.

865
00:34:04,530 --> 00:34:07,110
Because you might face
correlated shocks.

866
00:34:07,110 --> 00:34:07,652
What is that?

867
00:34:07,652 --> 00:34:09,152
For example, it
might be like one TA

868
00:34:09,152 --> 00:34:10,239
is better than the other.

869
00:34:10,239 --> 00:34:12,840
And one class, A or
B, the group might

870
00:34:12,840 --> 00:34:15,449
do much better than the other,
not because of the deadline

871
00:34:15,449 --> 00:34:17,620
policy, but just
because of the TA.

872
00:34:17,620 --> 00:34:19,770
Or again, it could be that,
like, the recitation is

873
00:34:19,770 --> 00:34:21,978
early in the morning, and
people don't pay attention,

874
00:34:21,978 --> 00:34:22,889
or whatever.

875
00:34:22,889 --> 00:34:25,250
There might be many other
things happening and so on.

876
00:34:25,250 --> 00:34:27,090
So in some sense,
the sample size

877
00:34:27,090 --> 00:34:28,980
is A, too small to start with.

878
00:34:28,980 --> 00:34:31,409
50 people is very small
for an experiment.

879
00:34:31,409 --> 00:34:34,650
Two, if you randomize or if you
have only, like, two treatment

880
00:34:34,650 --> 00:34:39,570
groups, you need more
clusters, as people call it--

881
00:34:39,570 --> 00:34:45,389
more sort of items of
randomization overall.

882
00:34:45,389 --> 00:34:47,435
So now, there's a
second experiment

883
00:34:47,435 --> 00:34:49,560
that they did sort of to
deal with the first issue.

884
00:34:49,560 --> 00:34:53,100
This was sort of like,
again, not a huge sample,

885
00:34:53,100 --> 00:34:54,780
but somewhat better.

886
00:34:54,780 --> 00:34:55,889
It's 60 people.

887
00:34:55,889 --> 00:35:00,190
And now, it's randomized
into three treatment groups.

888
00:35:00,190 --> 00:35:01,920
One is evenly-spaced deadlines.

889
00:35:01,920 --> 00:35:04,260
B is, like, no deadlines.

890
00:35:04,260 --> 00:35:06,400
Three is, like,
self-imposed deadlines.

891
00:35:06,400 --> 00:35:09,540
This is now a proofreading
exercise over 21 days.

892
00:35:09,540 --> 00:35:11,560
This is now randomized
at the individual level,

893
00:35:11,560 --> 00:35:12,700
so it's not in groups anymore.

894
00:35:12,700 --> 00:35:14,850
So we get a little bit away
from this cluster issue

895
00:35:14,850 --> 00:35:16,450
that we had before.

896
00:35:16,450 --> 00:35:18,180
Now, what does an
exponential discounter

897
00:35:18,180 --> 00:35:20,940
do in terms of performance?

898
00:35:20,940 --> 00:35:23,660
What do we predict
what's going to happen?

899
00:35:23,660 --> 00:35:25,680
So like, a person
with beta equals 1,

900
00:35:25,680 --> 00:35:27,342
what do we say is
going to happen

901
00:35:27,342 --> 00:35:28,800
to the performance
of these groups?

902
00:35:33,580 --> 00:35:35,905
AUDIENCE: It should be the same?

903
00:35:35,905 --> 00:35:36,780
FRANK SCHILBACH: Yes.

904
00:35:36,780 --> 00:35:37,150
Which one?

905
00:35:37,150 --> 00:35:37,650
All of them?

906
00:35:45,766 --> 00:35:49,250
AUDIENCE: I'm not sure.

907
00:35:49,250 --> 00:35:51,950
FRANK SCHILBACH: So let's start
with the self-imposed deadlines

908
00:35:51,950 --> 00:35:54,442
and the no deadlines.

909
00:35:54,442 --> 00:35:56,650
What kinds of deadlines is
the exponential discounter

910
00:35:56,650 --> 00:35:57,994
going to set?

911
00:35:57,994 --> 00:35:58,619
AUDIENCE: None?

912
00:35:58,619 --> 00:36:00,161
FRANK SCHILBACH:
None, somebody said?

913
00:36:00,161 --> 00:36:01,290
AUDIENCE: [INAUDIBLE]

914
00:36:01,290 --> 00:36:01,980
FRANK SCHILBACH: Yes, and--

915
00:36:01,980 --> 00:36:03,300
AUDIENCE: I don't think
they need the deadline.

916
00:36:03,300 --> 00:36:04,900
FRANK SCHILBACH: Yeah, like,
why would you set a deadline?

917
00:36:04,900 --> 00:36:05,490
You didn't need a deadline.

918
00:36:05,490 --> 00:36:07,800
You're going to just do it
whenever it's best to do.

919
00:36:07,800 --> 00:36:09,630
There might be some--
you might get sick

920
00:36:09,630 --> 00:36:11,670
or, like, some other
shocks might appear

921
00:36:11,670 --> 00:36:13,060
in the course of the semester.

922
00:36:13,060 --> 00:36:15,090
So deadlines will just
restrict yourself.

923
00:36:15,090 --> 00:36:18,210
So you will not
self-impose any deadlines.

924
00:36:18,210 --> 00:36:20,700
So early deadlines essentially
just limit flexibility.

925
00:36:20,700 --> 00:36:22,140
That's not helpful to do.

926
00:36:22,140 --> 00:36:25,320
You don't need to sort of
self-impose any deadlines.

927
00:36:25,320 --> 00:36:28,920
And then, you might sort
of do weekly worse--

928
00:36:28,920 --> 00:36:31,050
in group A, you might do
it a little bit worse,

929
00:36:31,050 --> 00:36:34,163
because essentially, shocks
might happen or the like.

930
00:36:34,163 --> 00:36:35,580
There's no reason
to believe that,

931
00:36:35,580 --> 00:36:38,100
like, if you restrict your
option, you might do better.

932
00:36:38,100 --> 00:36:41,070
Group A might happen to do as
well as C and B. But at least

933
00:36:41,070 --> 00:36:42,930
they are going to
be weekly worse,

934
00:36:42,930 --> 00:36:47,670
in part because people get sick
around deadlines or the like.

935
00:36:47,670 --> 00:36:48,450
OK?

936
00:36:48,450 --> 00:36:53,340
So what about a sophisticated
person with beta hat

937
00:36:53,340 --> 00:36:54,600
equals beta smaller than 1?

938
00:37:07,590 --> 00:37:08,420
Yes?

939
00:37:08,420 --> 00:37:10,054
AUDIENCE: Self-imposed
deadlines should

940
00:37:10,054 --> 00:37:12,290
equal evenly-spaced deadlines?

941
00:37:12,290 --> 00:37:15,148
FRANK SCHILBACH: Can
you explain more?

942
00:37:15,148 --> 00:37:18,040
AUDIENCE: [INAUDIBLE]

943
00:37:21,753 --> 00:37:23,670
FRANK SCHILBACH: So the
evenly-spaced deadline

944
00:37:23,670 --> 00:37:26,720
will do better or worse than
no deadlines, to start with?

945
00:37:26,720 --> 00:37:28,130
AUDIENCE: They would do better?

946
00:37:28,130 --> 00:37:30,222
FRANK SCHILBACH:
And why is that?

947
00:37:30,222 --> 00:37:35,389
AUDIENCE: Because they--
if they have present bias,

948
00:37:35,389 --> 00:37:36,514
then they'll procrastinate.

949
00:37:36,514 --> 00:37:39,420
So then they would have three
things due at the same time.

950
00:37:39,420 --> 00:37:40,590
FRANK SCHILBACH: Exactly.

951
00:37:40,590 --> 00:37:42,000
Exactly, so that should help.

952
00:37:42,000 --> 00:37:44,250
And then the self-imposed
deadlines, there,

953
00:37:44,250 --> 00:37:45,830
you can optimally
set the deadlines,

954
00:37:45,830 --> 00:37:48,150
so you could probably do
at least weekly better than

955
00:37:48,150 --> 00:37:50,640
the evenly-spaced ones,
because you might also take

956
00:37:50,640 --> 00:37:53,280
into account some constraints
that you have in a semester,

957
00:37:53,280 --> 00:37:56,100
like your family is
visiting or the like,

958
00:37:56,100 --> 00:37:57,630
that might be sort of--

959
00:37:57,630 --> 00:37:59,640
so evenly-spaced
deadlines, some might not

960
00:37:59,640 --> 00:38:01,050
be optimal because of that.

961
00:38:01,050 --> 00:38:03,508
If you self-impose them and
you're perfectly sophisticated,

962
00:38:03,508 --> 00:38:05,490
you can take that into account.

963
00:38:05,490 --> 00:38:07,380
So deadlines can help.

964
00:38:07,380 --> 00:38:10,590
And flexible deadlines
tend to be preferable,

965
00:38:10,590 --> 00:38:11,970
because you can
take into account

966
00:38:11,970 --> 00:38:16,320
your individual specific
costs that you might face.

967
00:38:16,320 --> 00:38:18,120
OK, what about the
fully naive person?

968
00:38:25,800 --> 00:38:26,580
Yes?

969
00:38:26,580 --> 00:38:28,878
AUDIENCE: So no deadlines
and self-imposed deadlines

970
00:38:28,878 --> 00:38:29,670
should be the same.

971
00:38:29,670 --> 00:38:31,362
Because if you're
fully naive, you

972
00:38:31,362 --> 00:38:33,570
don't think you need that
much [INAUDIBLE] commitment

973
00:38:33,570 --> 00:38:35,237
devices, so you just
set them to the end

974
00:38:35,237 --> 00:38:37,060
as if you're expontential.

975
00:38:37,060 --> 00:38:39,630
And evenly-spaced
should be weekly better,

976
00:38:39,630 --> 00:38:43,163
because even if it's
not fully optimal,

977
00:38:43,163 --> 00:38:44,830
it should stop some
of the [INAUDIBLE]..

978
00:38:44,830 --> 00:38:45,872
FRANK SCHILBACH: Correct.

979
00:38:45,872 --> 00:38:48,170
Correct, so as before, the
evenly-spaced deadlines,

980
00:38:48,170 --> 00:38:50,290
since the person has beta
smaller-- or some people

981
00:38:50,290 --> 00:38:53,530
have beta smaller than 1, the
evenly-spaced deadlines likely

982
00:38:53,530 --> 00:38:55,313
do better, or help
people do better,

983
00:38:55,313 --> 00:38:57,730
unless they sort of miss a lot
of deadlines anyway because

984
00:38:57,730 --> 00:38:58,522
of procrastination.

985
00:38:58,522 --> 00:39:01,150
But assuming the deadlines are
sort of helping people space

986
00:39:01,150 --> 00:39:04,180
things out better,
they're going to do better

987
00:39:04,180 --> 00:39:06,160
than the no deadlines case.

988
00:39:06,160 --> 00:39:07,950
And then, the fully
naive person says,

989
00:39:07,950 --> 00:39:09,200
why would I set any deadlines?

990
00:39:09,200 --> 00:39:10,060
I don't need any deadline.

991
00:39:10,060 --> 00:39:11,007
Flexibility is good.

992
00:39:11,007 --> 00:39:12,340
I'm going to do it early anyway.

993
00:39:12,340 --> 00:39:14,450
Of course, that's
not going to happen.

994
00:39:14,450 --> 00:39:16,840
But then essentially, B
is doing the same as C,

995
00:39:16,840 --> 00:39:21,070
because the person does not
set any deadlines at all, OK?

996
00:39:21,070 --> 00:39:22,990
So now, what about the
partially naive person?

997
00:39:37,730 --> 00:39:38,840
Yes?

998
00:39:38,840 --> 00:39:41,390
AUDIENCE: A is better than
C which is better than B?

999
00:39:41,390 --> 00:39:43,160
FRANK SCHILBACH:
Can you say more?

1000
00:39:43,160 --> 00:39:47,300
AUDIENCE: Sure,
so I mean, I guess

1001
00:39:47,300 --> 00:39:49,460
if they're sufficiently
sophisticated, then

1002
00:39:49,460 --> 00:39:53,330
it's is possible that Cis better
than A which is better than B.

1003
00:39:53,330 --> 00:39:56,173
But if they're mostly
naive, then they

1004
00:39:56,173 --> 00:39:59,820
will do some kind of
deadline imposition

1005
00:39:59,820 --> 00:40:04,150
which is not the absence of
deadlines in its entirety,

1006
00:40:04,150 --> 00:40:07,080
but it might not be as
good as [INAUDIBLE]..

1007
00:40:07,080 --> 00:40:09,240
FRANK SCHILBACH: So I
guess the answer is, like,

1008
00:40:09,240 --> 00:40:10,560
to some degree, it depends.

1009
00:40:10,560 --> 00:40:13,060
So we still have, like,
A is better than B.

1010
00:40:13,060 --> 00:40:15,240
So if people have
beta smaller than 1,

1011
00:40:15,240 --> 00:40:16,860
the evenly-spaced
deadlines still,

1012
00:40:16,860 --> 00:40:20,070
as before, are going
to do somewhat better

1013
00:40:20,070 --> 00:40:21,570
than the no deadlines.

1014
00:40:21,570 --> 00:40:23,580
Now, for the
self-imposed deadlines,

1015
00:40:23,580 --> 00:40:25,375
essentially two
things can happen.

1016
00:40:25,375 --> 00:40:27,000
One is, like, the
person might actually

1017
00:40:27,000 --> 00:40:29,260
set some deadlines that
happen to be actually helpful.

1018
00:40:29,260 --> 00:40:31,260
If they're sufficiently
sophisticated,

1019
00:40:31,260 --> 00:40:34,170
the deadlines might actually
help at least some people.

1020
00:40:34,170 --> 00:40:36,983
But for some people, they might
set overambitious deadlines

1021
00:40:36,983 --> 00:40:39,150
and say, oh, I'm going to
get it done all next week.

1022
00:40:39,150 --> 00:40:41,100
Well, it turns out
that's actually not true.

1023
00:40:41,100 --> 00:40:42,810
And then they sort of
said too-early deadlines

1024
00:40:42,810 --> 00:40:44,880
and then have to rush, or they
miss the deadline, or the like.

1025
00:40:44,880 --> 00:40:46,390
They might actually do worse.

1026
00:40:46,390 --> 00:40:49,680
So there, we have essentially
no clear predictions.

1027
00:40:49,680 --> 00:40:50,970
Some commitment should help.

1028
00:40:50,970 --> 00:40:53,850
So C could be better than B.
But individuals might also

1029
00:40:53,850 --> 00:40:54,450
overcommit.

1030
00:40:54,450 --> 00:40:56,530
So C might actually
be worse than B.

1031
00:40:56,530 --> 00:40:59,280
So that's-- essentially,
there's no empirical--

1032
00:40:59,280 --> 00:41:01,320
clear empirical prediction.

1033
00:41:01,320 --> 00:41:04,500
Any questions on what
I showed you here?

1034
00:41:04,500 --> 00:41:05,764
Yes?

1035
00:41:05,764 --> 00:41:14,373
AUDIENCE: [INAUDIBLE]

1036
00:41:14,373 --> 00:41:15,290
FRANK SCHILBACH: Yeah.

1037
00:41:15,290 --> 00:41:23,412
AUDIENCE: [INAUDIBLE]

1038
00:41:23,412 --> 00:41:24,870
FRANK SCHILBACH:
So you mean, like,

1039
00:41:24,870 --> 00:41:26,760
a subsample of people
who happen to choose

1040
00:41:26,760 --> 00:41:29,100
evenly-spaced deadlines?

1041
00:41:29,100 --> 00:41:29,917
Yeah.

1042
00:41:29,917 --> 00:41:33,326
AUDIENCE: So the
concerns [INAUDIBLE]----

1043
00:41:33,326 --> 00:41:39,657
so I guess I'm just not sure
[INAUDIBLE] could that possibly

1044
00:41:39,657 --> 00:41:43,513
[INAUDIBLE] leave the
concerns that you [INAUDIBLE]..

1045
00:41:43,513 --> 00:41:45,930
FRANK SCHILBACH: Yeah, so the
issue there is essentially--

1046
00:41:45,930 --> 00:41:48,097
so one thing they were doing
in the paper is to say,

1047
00:41:48,097 --> 00:41:50,700
let's look at the people who
set evenly-spaced deadlines,

1048
00:41:50,700 --> 00:41:52,290
and say, when they--

1049
00:41:52,290 --> 00:41:55,320
how are they doing compared to
people in the control group,

1050
00:41:55,320 --> 00:41:56,220
or the like?

1051
00:41:56,220 --> 00:41:58,200
The problem with that
is, like, these people

1052
00:41:58,200 --> 00:41:59,670
are selected in certain ways.

1053
00:41:59,670 --> 00:42:02,430
Like, if I sort of just--

1054
00:42:02,430 --> 00:42:05,320
this is a subset of people
that's non-randomly selected.

1055
00:42:05,320 --> 00:42:07,070
We know that they chose
certain deadlines.

1056
00:42:07,070 --> 00:42:09,112
So in some sense, we know
that they're different,

1057
00:42:09,112 --> 00:42:11,620
in some ways, compared to the
average person in the group.

1058
00:42:11,620 --> 00:42:12,780
And these could be
people who are, like,

1059
00:42:12,780 --> 00:42:14,580
more productive, less
productive, et cetera, and so

1060
00:42:14,580 --> 00:42:15,203
on.

1061
00:42:15,203 --> 00:42:16,620
So the clean design,
how you would

1062
00:42:16,620 --> 00:42:18,495
do that is essentially,
what you would do is,

1063
00:42:18,495 --> 00:42:23,400
you would ask people to--

1064
00:42:23,400 --> 00:42:25,170
ask for their deadline
choices, and then

1065
00:42:25,170 --> 00:42:27,960
randomize them into
different deadline regimens.

1066
00:42:27,960 --> 00:42:29,638
So either they get
their choice or not.

1067
00:42:29,638 --> 00:42:31,680
So you can actually estimate
the treatment effect

1068
00:42:31,680 --> 00:42:33,097
on those kinds of
people, and then

1069
00:42:33,097 --> 00:42:35,730
look at are
evenly-spaced deadlines--

1070
00:42:35,730 --> 00:42:38,490
are those treatment effects
on those people worse or not?

1071
00:42:38,490 --> 00:42:41,280
But I worry that in
that kind of analysis,

1072
00:42:41,280 --> 00:42:43,500
what's non-random here--
or what they're choosing

1073
00:42:43,500 --> 00:42:46,290
as a non-random subset and
compare that with a control

1074
00:42:46,290 --> 00:42:46,830
group--

1075
00:42:46,830 --> 00:42:49,038
they're sort of-- that's
correlated with other stuff.

1076
00:42:49,038 --> 00:42:55,080
And that's not necessarily
getting your causal effect.

1077
00:42:55,080 --> 00:43:01,650
Having said that, when you look
at the evenly-spaced deadlines,

1078
00:43:01,650 --> 00:43:03,660
they could sort make
things worse in some ways,

1079
00:43:03,660 --> 00:43:05,573
in the sense of,
like, even for people

1080
00:43:05,573 --> 00:43:06,990
with beta smaller
than 1, it could

1081
00:43:06,990 --> 00:43:08,280
be that people
missed the deadlines

1082
00:43:08,280 --> 00:43:10,655
because they're procrastinating,
and so on, and so forth.

1083
00:43:10,655 --> 00:43:12,533
Or they had some
shocks and the like,

1084
00:43:12,533 --> 00:43:14,700
so it's not necessarily
obvious that they do better.

1085
00:43:14,700 --> 00:43:17,400
That's sort of all
under some assumptions.

1086
00:43:21,000 --> 00:43:22,424
Any other questions?

1087
00:43:27,640 --> 00:43:29,670
OK, so now, what
we find is, like,

1088
00:43:29,670 --> 00:43:32,730
A is doing better than C
is doing better than B.

1089
00:43:32,730 --> 00:43:35,970
So this essentially true
for errors detected.

1090
00:43:35,970 --> 00:43:37,650
It's true for delays
in submission.

1091
00:43:37,650 --> 00:43:39,480
It's also true for
people's earnings

1092
00:43:39,480 --> 00:43:41,558
in the entire exercise.

1093
00:43:41,558 --> 00:43:43,350
So now, what you have
is, like, A is better

1094
00:43:43,350 --> 00:43:46,530
than C is better than B. So
what is that consistent with?

1095
00:43:46,530 --> 00:43:49,470
Well, it seems like it's
consistent with some partial

1096
00:43:49,470 --> 00:43:50,430
naiveté.

1097
00:43:50,430 --> 00:43:55,947
In some sense, admittedly, the
prediction of partial naiveté

1098
00:43:55,947 --> 00:43:57,780
is a little bit loose
in a sense that, like,

1099
00:43:57,780 --> 00:44:02,190
all we're saying is A is
better than B. And you know,

1100
00:44:02,190 --> 00:44:04,840
then the comparison with
C could go either way.

1101
00:44:04,840 --> 00:44:07,680
So sure, it's consistent
with partial naiveté,

1102
00:44:07,680 --> 00:44:10,050
but it's also only
suggestive in some sense.

1103
00:44:10,050 --> 00:44:13,210
Because there's no clear
prediction in that case.

1104
00:44:13,210 --> 00:44:15,770
However, we can reject--
sort of like the other cases,

1105
00:44:15,770 --> 00:44:17,520
we can reject the
exponential discountings

1106
00:44:17,520 --> 00:44:20,940
of sophisticated present bias
to fully sophisticated one,

1107
00:44:20,940 --> 00:44:23,580
or the fully naive
present biased person,

1108
00:44:23,580 --> 00:44:26,200
because those predictions
are not borne out.

1109
00:44:26,200 --> 00:44:32,370
So the only case that really
is left is the partial naiveté

1110
00:44:32,370 --> 00:44:32,880
one.

1111
00:44:32,880 --> 00:44:35,230
So let me-- sort of to
summarize what we saw,

1112
00:44:35,230 --> 00:44:37,830
so the result is that deadline
setting improves performance,

1113
00:44:37,830 --> 00:44:40,740
that is evidence of some
present bias-- being essentially

1114
00:44:40,740 --> 00:44:42,960
beta being smaller than 1.

1115
00:44:42,960 --> 00:44:46,720
It's also consistent
with partial naiveté.

1116
00:44:46,720 --> 00:44:48,840
People set deadlines,
which is consistent.

1117
00:44:48,840 --> 00:44:50,580
They set early
deadlines, which shows

1118
00:44:50,580 --> 00:44:54,360
some demand for commitment,
some form of sophistication.

1119
00:44:54,360 --> 00:44:57,360
Then, result 2 says deadline
setting is suboptimal, which

1120
00:44:57,360 --> 00:45:00,250
sort of suggests that beta
is smaller than beta hat,

1121
00:45:00,250 --> 00:45:03,000
or beta hat is bigger than beta,
which means that essentially,

1122
00:45:03,000 --> 00:45:07,470
people seem to underestimate the
self-control problems that they

1123
00:45:07,470 --> 00:45:08,610
have in the future.

1124
00:45:08,610 --> 00:45:10,110
So that's, sort of
broadly speaking,

1125
00:45:10,110 --> 00:45:16,650
some support of the beta delta
model with partial naiveté.

1126
00:45:16,650 --> 00:45:17,900
Any questions on this paper?

1127
00:45:26,700 --> 00:45:27,450
OK.

1128
00:45:27,450 --> 00:45:29,670
Yes, sorry, go ahead.

1129
00:45:29,670 --> 00:45:32,043
AUDIENCE: I was confused by
what you meant by partial

1130
00:45:32,043 --> 00:45:33,460
sophistication or
partial naiveté.

1131
00:45:33,460 --> 00:45:35,876
FRANK SCHILBACH: So this is
what I was discussing earlier,

1132
00:45:35,876 --> 00:45:37,735
partial naivete
being-- meaning that--

1133
00:45:37,735 --> 00:45:39,110
let me just go
back for a second.

1134
00:45:45,844 --> 00:45:47,710
This is [INAUDIBLE]
discussing here earlier,

1135
00:45:47,710 --> 00:45:49,600
partial naiveté meaning that--

1136
00:45:49,600 --> 00:45:51,380
so we discussed
two cases before,

1137
00:45:51,380 --> 00:45:54,700
which was full naiveté
or full sophistication--

1138
00:45:54,700 --> 00:45:57,670
full naiveté meaning
that beta hat equals 1,

1139
00:45:57,670 --> 00:46:01,900
full sophistication meaning
that beta hat equals beta.

1140
00:46:01,900 --> 00:46:06,610
And now, if you sort of take
the case in between, which is--

1141
00:46:06,610 --> 00:46:09,040
where you see-- which I have
here at the bottom, which

1142
00:46:09,040 --> 00:46:13,180
is beta hat is in the
middle of beta and 1,

1143
00:46:13,180 --> 00:46:15,160
or in between beta and 1.

1144
00:46:15,160 --> 00:46:18,220
So essentially, you understand
that your beta in the future

1145
00:46:18,220 --> 00:46:19,660
is not 1.

1146
00:46:19,660 --> 00:46:24,920
But you sort of underestimate
how small it is.

1147
00:46:24,920 --> 00:46:28,150
So as I said before,
my beta might be 0.6.

1148
00:46:28,150 --> 00:46:31,060
I might understand my
future beta is not 1.

1149
00:46:31,060 --> 00:46:32,920
So I understand
it's smaller than 1.

1150
00:46:32,920 --> 00:46:35,530
But I might think it's,
like, 0.8 or 0.9, which

1151
00:46:35,530 --> 00:46:39,430
is to say I overestimate my
future beta to some degree.

1152
00:46:39,430 --> 00:46:41,142
While I get it right
that it's not 1,

1153
00:46:41,142 --> 00:46:42,850
I might sort of get
it wrong in the sense

1154
00:46:42,850 --> 00:46:44,200
of saying it's not 0.6.

1155
00:46:44,200 --> 00:46:47,360
In fact, I might think
it's higher than that.

1156
00:46:47,360 --> 00:46:47,860
OK.

1157
00:46:54,450 --> 00:46:55,710
OK, let me go back to--

1158
00:46:55,710 --> 00:47:00,517
OK, so in some sense,
these are nice experiments.

1159
00:47:00,517 --> 00:47:02,850
They're useful experiments
to help us sort of understand

1160
00:47:02,850 --> 00:47:04,260
is there evidence
for this model,

1161
00:47:04,260 --> 00:47:07,920
is there evidence for some of
the predictions that we have?

1162
00:47:07,920 --> 00:47:10,020
But it's still fairly
contrived in the sense of,

1163
00:47:10,020 --> 00:47:12,227
like, it's kind of
like-- especially

1164
00:47:12,227 --> 00:47:14,310
in the second experiment--
it's kind of like a lab

1165
00:47:14,310 --> 00:47:15,540
experiment and the like.

1166
00:47:15,540 --> 00:47:18,270
Really, what we want to
know is, in the real world,

1167
00:47:18,270 --> 00:47:20,700
are commitment devices
A, helpful in some sense?

1168
00:47:20,700 --> 00:47:21,930
Do people want them?

1169
00:47:21,930 --> 00:47:23,460
And B, are they
actually improving

1170
00:47:23,460 --> 00:47:24,750
behaviors in certain ways?

1171
00:47:24,750 --> 00:47:26,340
Should a firm use
commitment devices?

1172
00:47:26,340 --> 00:47:30,182
This is actually helpful for
improving firm performance.

1173
00:47:30,182 --> 00:47:32,390
So there's a very nice paper
by Supreet Kaur, Sendhil

1174
00:47:32,390 --> 00:47:33,765
Mullainathan, and
Michael Kremer,

1175
00:47:33,765 --> 00:47:36,570
who sort of investigate
this question using

1176
00:47:36,570 --> 00:47:38,650
data entry workers in India.

1177
00:47:38,650 --> 00:47:41,970
So this is a full-time
job for people.

1178
00:47:41,970 --> 00:47:45,190
This is the primary source of
earnings for these workers.

1179
00:47:45,190 --> 00:47:48,030
These are people who are sort of
recruited as data entry workers

1180
00:47:48,030 --> 00:47:51,930
for the course of 13
months-- so over a year--

1181
00:47:51,930 --> 00:47:53,812
as a typical data entry worker.

1182
00:47:53,812 --> 00:47:55,770
So in some sense, there
was a job ad that said,

1183
00:47:55,770 --> 00:47:56,940
we're looking for a
data entry worker.

1184
00:47:56,940 --> 00:47:58,350
We have data to be entered.

1185
00:47:58,350 --> 00:48:00,060
People would answer
the job ad and then

1186
00:48:00,060 --> 00:48:04,530
sort of start working
to do data entry.

1187
00:48:04,530 --> 00:48:08,610
Output is measured, as it is
usual done in many data entry

1188
00:48:08,610 --> 00:48:11,910
companies, by measuring
the number of accurate

1189
00:48:11,910 --> 00:48:13,920
fields entered in a day.

1190
00:48:13,920 --> 00:48:17,250
This is measured by doing dual
entry, as in to say the data

1191
00:48:17,250 --> 00:48:17,970
is entered twice.

1192
00:48:17,970 --> 00:48:19,860
It's then matched
with the other person.

1193
00:48:19,860 --> 00:48:23,140
And the number of
errors or discrepancies

1194
00:48:23,140 --> 00:48:25,390
are then defined
as an error rate.

1195
00:48:25,390 --> 00:48:27,870
And if the error rate--

1196
00:48:27,870 --> 00:48:29,950
you're matched to
different people over time,

1197
00:48:29,950 --> 00:48:31,755
so I can sort of back
out your error rate--

1198
00:48:31,755 --> 00:48:33,397
how good or bad you are.

1199
00:48:33,397 --> 00:48:34,980
So people are
essentially incentivized

1200
00:48:34,980 --> 00:48:37,440
to do to work fast,
and incentivized

1201
00:48:37,440 --> 00:48:40,560
to enter stuff correctly.

1202
00:48:40,560 --> 00:48:44,280
Workers are paid a piece rate
using the weekly paycheck that

1203
00:48:44,280 --> 00:48:45,900
paid once a week.

1204
00:48:45,900 --> 00:48:47,280
There's no
restrictions on hours.

1205
00:48:47,280 --> 00:48:48,863
People can show up
whenever they like.

1206
00:48:48,863 --> 00:48:52,240
They can stay for as many hours
as they like on any given day.

1207
00:48:52,240 --> 00:48:53,820
There are no penalties
for absences.

1208
00:48:53,820 --> 00:48:56,070
So you can just-- if you
don't show up on a given day,

1209
00:48:56,070 --> 00:48:57,207
nothing happens.

1210
00:48:57,207 --> 00:48:59,040
This is what the data
entry task looks like.

1211
00:48:59,040 --> 00:49:00,210
There's essentially some data.

1212
00:49:00,210 --> 00:49:01,377
This is kind of interesting.

1213
00:49:01,377 --> 00:49:02,730
This is data coming from Kenya.

1214
00:49:02,730 --> 00:49:04,480
So Michael Kremer and
Sendhil Mullainathan

1215
00:49:04,480 --> 00:49:06,480
were working on some
Kenyan data where

1216
00:49:06,480 --> 00:49:10,410
they had a huge amount of data
from Kenyan sugarcane farmers

1217
00:49:10,410 --> 00:49:14,070
over the course of 30 years--
some historical records

1218
00:49:14,070 --> 00:49:15,030
that they found.

1219
00:49:15,030 --> 00:49:16,390
They wanted them to get entered.

1220
00:49:16,390 --> 00:49:17,940
So like, how do we
get them entered?

1221
00:49:17,940 --> 00:49:19,568
Well, India is the
place to do that.

1222
00:49:19,568 --> 00:49:20,610
So they got them entered.

1223
00:49:20,610 --> 00:49:23,520
And then, when they hired lots
of people to enter the data,

1224
00:49:23,520 --> 00:49:26,250
they thought, well, why
not run an experiment

1225
00:49:26,250 --> 00:49:29,768
to learn about self-control
among workers while doing that?

1226
00:49:29,768 --> 00:49:32,310
So what they would do is they
would show these scanned images

1227
00:49:32,310 --> 00:49:33,520
of these pages.

1228
00:49:33,520 --> 00:49:36,990
And then there's like a computer
software, pretty rudimentary,

1229
00:49:36,990 --> 00:49:41,860
where you can sort of enter
the fields in your data set.

1230
00:49:41,860 --> 00:49:46,200
Now, the commitment device
is a dominated contract

1231
00:49:46,200 --> 00:49:47,350
that looks as follows.

1232
00:49:47,350 --> 00:49:49,230
So the-- as I told
you before, workers

1233
00:49:49,230 --> 00:49:51,030
are paid a piece rate w.

1234
00:49:51,030 --> 00:49:52,560
It's a linear piece rate.

1235
00:49:52,560 --> 00:49:55,600
That's the control contract
for their production.

1236
00:49:55,600 --> 00:49:58,020
So forever production, think
of production as how much

1237
00:49:58,020 --> 00:49:59,850
the correct entries
that you produce.

1238
00:49:59,850 --> 00:50:01,550
For each correct
entry that you make,

1239
00:50:01,550 --> 00:50:04,050
you're going to get paid w.

1240
00:50:04,050 --> 00:50:06,960
Now, the dominated
contract looks as this--

1241
00:50:06,960 --> 00:50:08,860
this is the red one
that you see here--

1242
00:50:08,860 --> 00:50:12,180
which is until a
target T. We'll talk

1243
00:50:12,180 --> 00:50:13,530
about the target in a second.

1244
00:50:13,530 --> 00:50:17,010
Until the target T, you
paid w over 2, which

1245
00:50:17,010 --> 00:50:18,870
is like half as much as before.

1246
00:50:18,870 --> 00:50:21,630
And as soon as you
reach T, the target,

1247
00:50:21,630 --> 00:50:24,450
you're going to be paid w
for the entire production

1248
00:50:24,450 --> 00:50:25,440
that you make.

1249
00:50:25,440 --> 00:50:26,895
Why is this a
commitment contract?

1250
00:50:29,940 --> 00:50:32,630
So if I ask you to
choose your target T,

1251
00:50:32,630 --> 00:50:34,130
if you choose a
positive target, why

1252
00:50:34,130 --> 00:50:35,450
is this a commitment contract?

1253
00:50:35,450 --> 00:50:35,690
Yes?

1254
00:50:35,690 --> 00:50:37,315
AUDIENCE: So if you
go less then you're

1255
00:50:37,315 --> 00:50:40,460
penalized for it by making
less money [INAUDIBLE]..

1256
00:50:40,460 --> 00:50:41,508
FRANK SCHILBACH: Right.

1257
00:50:41,508 --> 00:50:42,050
That's right.

1258
00:50:42,050 --> 00:50:46,230
So there's no value of
choosing this contract.

1259
00:50:46,230 --> 00:50:48,860
So if you are an
exponential discounter,

1260
00:50:48,860 --> 00:50:51,140
you might just
choose T equals 0,

1261
00:50:51,140 --> 00:50:53,570
the reason being that
there's no point in choosing

1262
00:50:53,570 --> 00:50:54,515
a high target.

1263
00:50:54,515 --> 00:50:55,890
Who knows what's
going to happen?

1264
00:50:55,890 --> 00:50:56,810
You might get a headache.

1265
00:50:56,810 --> 00:50:57,860
Your kid might get sick.

1266
00:50:57,860 --> 00:50:59,450
The computer might
not work, or whatever.

1267
00:50:59,450 --> 00:51:00,908
For whatever reason,
there might be

1268
00:51:00,908 --> 00:51:04,070
shocks, uncertainty that might
make you not reach your target.

1269
00:51:04,070 --> 00:51:06,530
And so if you choose a positive
target, the only thing you

1270
00:51:06,530 --> 00:51:08,405
can do from this positive
target is, you see,

1271
00:51:08,405 --> 00:51:09,740
you can do only worse, right?

1272
00:51:09,740 --> 00:51:13,890
It's a weakly dominated contract
in terms of your payments,

1273
00:51:13,890 --> 00:51:16,290
so there's no reason
to actually do that.

1274
00:51:16,290 --> 00:51:18,430
But why might the worker
actually choose it anyway?

1275
00:51:22,960 --> 00:51:24,122
Yes?

1276
00:51:24,122 --> 00:51:27,400
AUDIENCE: Because
earlier, you specified

1277
00:51:27,400 --> 00:51:29,332
that the workers can
come and go as they

1278
00:51:29,332 --> 00:51:30,540
please once they get to work.

1279
00:51:30,540 --> 00:51:32,520
And so there's no
real set schedule.

1280
00:51:32,520 --> 00:51:36,040
So [INAUDIBLE] or think that
they know themselves, and think

1281
00:51:36,040 --> 00:51:38,470
that they'll push off all
the work until the very

1282
00:51:38,470 --> 00:51:39,970
last minute, et
cetera, et cetera.

1283
00:51:39,970 --> 00:51:42,290
And so they might want
the commitment device

1284
00:51:42,290 --> 00:51:44,883
to incentivize themselves to
finish their work [INAUDIBLE]..

1285
00:51:44,883 --> 00:51:46,800
FRANK SCHILBACH: Exactly,
I might sort of just

1286
00:51:46,800 --> 00:51:49,575
be slightly below T or the like.

1287
00:51:49,575 --> 00:51:52,320
I might sort of be
tempted to go home.

1288
00:51:52,320 --> 00:51:54,210
You know, my friends
call or whatever,

1289
00:51:54,210 --> 00:51:55,690
and I might just
not want to do it.

1290
00:51:55,690 --> 00:51:56,940
It might be just a tedious day.

1291
00:51:56,940 --> 00:51:57,898
It might be really hot.

1292
00:51:57,898 --> 00:51:59,730
I had really goals
to work a lot,

1293
00:51:59,730 --> 00:52:02,970
and I want to incentivize
my future self

1294
00:52:02,970 --> 00:52:04,980
to reach certain targets.

1295
00:52:04,980 --> 00:52:06,030
That's exactly right.

1296
00:52:06,030 --> 00:52:08,640
So I think this is--

1297
00:52:08,640 --> 00:52:10,230
we've already said all of that.

1298
00:52:10,230 --> 00:52:11,825
Workers can choose
the T in advance.

1299
00:52:11,825 --> 00:52:13,950
They can choose-- essentially,
on the previous day,

1300
00:52:13,950 --> 00:52:16,800
they can choose T equals 0.

1301
00:52:16,800 --> 00:52:19,590
They also have
randomization of paydays

1302
00:52:19,590 --> 00:52:21,750
to be able to look
at payday effects.

1303
00:52:21,750 --> 00:52:23,140
What do I mean by that?

1304
00:52:23,140 --> 00:52:26,200
I told you, like,
workers are paid weekly.

1305
00:52:26,200 --> 00:52:27,660
And so they randomize
the paydays.

1306
00:52:27,660 --> 00:52:29,993
What they do is they randomize
the pages to be Tuesdays,

1307
00:52:29,993 --> 00:52:32,543
Thursdays, or Saturdays
so that you can allow

1308
00:52:32,543 --> 00:52:33,960
for day-of-the-week
fixed effects.

1309
00:52:33,960 --> 00:52:35,760
That's kind of
[INAUDIBLE] interesting.

1310
00:52:35,760 --> 00:52:37,830
But essentially,
you can look at,

1311
00:52:37,830 --> 00:52:39,882
when a worker is
paid in a given week,

1312
00:52:39,882 --> 00:52:42,090
are they more or less
productive on that day compared

1313
00:52:42,090 --> 00:52:44,560
to the previous days,
and so on and so forth.

1314
00:52:44,560 --> 00:52:46,480
What does the exponential
discounting model

1315
00:52:46,480 --> 00:52:50,100
predict about paydays
or payday effects?

1316
00:52:50,100 --> 00:52:51,900
Should you work
harder on a payday?

1317
00:53:05,590 --> 00:53:06,560
Yes?

1318
00:53:06,560 --> 00:53:09,177
AUDIENCE: No, because
you're time-consistent.

1319
00:53:09,177 --> 00:53:10,510
FRANK SCHILBACH: Right, exactly.

1320
00:53:10,510 --> 00:53:13,210
Like, your delta is
0.95 or the like.

1321
00:53:13,210 --> 00:53:16,780
It doesn't really matter whether
the daily discounting should

1322
00:53:16,780 --> 00:53:18,460
be essentially
close-- there should

1323
00:53:18,460 --> 00:53:20,080
be essentially close
to no discounting

1324
00:53:20,080 --> 00:53:21,310
between different days.

1325
00:53:21,310 --> 00:53:24,250
You might have a yearly
delta of 0.95 or the like.

1326
00:53:24,250 --> 00:53:27,255
But then, between today
and tomorrow, or two days

1327
00:53:27,255 --> 00:53:29,800
and three days from now and so
on, it doesn't really matter.

1328
00:53:29,800 --> 00:53:32,770
So if I'm working today
and I'm paid today,

1329
00:53:32,770 --> 00:53:34,450
it's the same as if
I'm working today

1330
00:53:34,450 --> 00:53:36,800
and I'm paid in like three days.

1331
00:53:36,800 --> 00:53:39,310
However, if I'm a
quasi-hyperbolic discounter,

1332
00:53:39,310 --> 00:53:41,800
well, what's going to
be the case is, like,

1333
00:53:41,800 --> 00:53:44,200
if I'm working on
Sundays, I'm working today

1334
00:53:44,200 --> 00:53:45,965
and I'm going to be paid today.

1335
00:53:45,965 --> 00:53:47,590
Well, then I'm going
to go work harder.

1336
00:53:47,590 --> 00:53:49,090
Maybe I get the
money right away.

1337
00:53:49,090 --> 00:53:51,280
I can buy a nice
meal, my family is

1338
00:53:51,280 --> 00:53:53,450
going to be happier,
and so on and so forth.

1339
00:53:53,450 --> 00:53:56,120
Now notice, that requires some
form of liquidity constraints,

1340
00:53:56,120 --> 00:53:56,620
right?

1341
00:53:56,620 --> 00:53:58,450
In the sense that if I have
a bunch of cash anyway,

1342
00:53:58,450 --> 00:53:59,950
it doesn't really
matter whether I'm

1343
00:53:59,950 --> 00:54:01,720
going to be paid
today versus tomorrow.

1344
00:54:01,720 --> 00:54:04,840
But assuming that some workers
are liquidity constrained,

1345
00:54:04,840 --> 00:54:06,580
essentially whether
they're paid today

1346
00:54:06,580 --> 00:54:09,850
versus tomorrow matters a
lot when the reward comes.

1347
00:54:09,850 --> 00:54:12,140
If you think about today,
it's like 10:00 AM.

1348
00:54:12,140 --> 00:54:13,630
I'm typing and typing.

1349
00:54:13,630 --> 00:54:15,820
Now, I think about, like,
when do I get the reward.

1350
00:54:15,820 --> 00:54:17,290
Either it's on
the payday, you're

1351
00:54:17,290 --> 00:54:19,030
going to be paid
at, like, 5:00 PM,

1352
00:54:19,030 --> 00:54:20,947
so you're going to get
essentially a nice meal

1353
00:54:20,947 --> 00:54:21,822
on the same day.

1354
00:54:21,822 --> 00:54:24,280
Versus if you're paid in five
days, that's really far away.

1355
00:54:24,280 --> 00:54:27,910
In the beta delta world,
your reward for working hard

1356
00:54:27,910 --> 00:54:31,382
is coming much later, OK?

1357
00:54:31,382 --> 00:54:33,340
So you would think
quasi-hyperbolic discounters

1358
00:54:33,340 --> 00:54:35,950
would put in higher
efforts on paydays.

1359
00:54:35,950 --> 00:54:39,153
Assuming there is some form
of liquidity constraints,

1360
00:54:39,153 --> 00:54:41,570
there should be close to no
difference between other days.

1361
00:54:41,570 --> 00:54:44,195
That depends a little bit, like,
what's the horizon of the beta

1362
00:54:44,195 --> 00:54:46,690
that we talked about before.

1363
00:54:46,690 --> 00:54:49,090
OK, so now what do they
find in this paper?

1364
00:54:49,090 --> 00:54:51,160
They find essentially
three results.

1365
00:54:51,160 --> 00:54:53,410
The first result is there's
demand for commitment.

1366
00:54:53,410 --> 00:54:56,310
There's sort of-- people
choose dominated contracts.

1367
00:54:56,310 --> 00:54:59,170
That is to say, workers
select dominated contracts

1368
00:54:59,170 --> 00:55:01,210
about 36% of the time.

1369
00:55:01,210 --> 00:55:04,660
This is a lower bound
for the extent of time

1370
00:55:04,660 --> 00:55:06,710
inconsistency for three reasons.

1371
00:55:06,710 --> 00:55:08,632
Reason one is some
workers might be naive.

1372
00:55:08,632 --> 00:55:10,840
They might sort of think
they don't need any targets.

1373
00:55:10,840 --> 00:55:12,950
Well, in fact, they would
benefit from targets.

1374
00:55:12,950 --> 00:55:16,510
Or they might underinvest
in this technology.

1375
00:55:16,510 --> 00:55:19,060
Second, people might think this
commitment device is actually

1376
00:55:19,060 --> 00:55:20,710
ineffective, in a
sense saying, look,

1377
00:55:20,710 --> 00:55:22,960
this is sort of what we're
talking about with the Frog

1378
00:55:22,960 --> 00:55:23,890
and the Toad.

1379
00:55:23,890 --> 00:55:26,630
There might be-- they might
have time-inconsistency issues.

1380
00:55:26,630 --> 00:55:29,080
It might just not be a strong
enough commitment device.

1381
00:55:29,080 --> 00:55:31,210
And so then, if I have
a day where I just

1382
00:55:31,210 --> 00:55:33,700
don't want to work, then if
that's not going to get you

1383
00:55:33,700 --> 00:55:36,550
over the target,
well then I might not

1384
00:55:36,550 --> 00:55:39,580
choose this commitment
device, not because I'm not--

1385
00:55:39,580 --> 00:55:42,550
I don't have present bias,
but because it's just

1386
00:55:42,550 --> 00:55:46,530
not effective enough of a
punishment for me to do.

1387
00:55:46,530 --> 00:55:49,390
Number three is people
might prefer flexibility

1388
00:55:49,390 --> 00:55:50,650
and they're risk-averse.

1389
00:55:50,650 --> 00:55:53,518
These are issues like they might
have children that get sick,

1390
00:55:53,518 --> 00:55:55,060
they might have
headaches, and so on.

1391
00:55:55,060 --> 00:55:57,770
They might just be risks that's
unrelated to their present bias

1392
00:55:57,770 --> 00:55:58,510
and the like.

1393
00:55:58,510 --> 00:55:59,710
The computers might be bad.

1394
00:55:59,710 --> 00:56:01,840
So they had different
computers at different speed.

1395
00:56:01,840 --> 00:56:03,730
You might end up with
a really bad computer,

1396
00:56:03,730 --> 00:56:06,358
and then you might
not reach the target.

1397
00:56:06,358 --> 00:56:08,650
That's nothing to do with
self-control or present bias,

1398
00:56:08,650 --> 00:56:10,510
but rather with sort
of external risk.

1399
00:56:10,510 --> 00:56:12,010
If you're really
worried about risk,

1400
00:56:12,010 --> 00:56:16,390
then you might not choose
a positive target either.

1401
00:56:16,390 --> 00:56:20,060
Second, they find offering the
dominated contracts increases

1402
00:56:20,060 --> 00:56:20,560
output.

1403
00:56:20,560 --> 00:56:23,200
That is to say, if you compare
groups that were, on some days,

1404
00:56:23,200 --> 00:56:25,720
offered those contracts,
compared to other groups that

1405
00:56:25,720 --> 00:56:28,270
were randomly selected
to be not offered

1406
00:56:28,270 --> 00:56:31,060
those contracts on
those same days,

1407
00:56:31,060 --> 00:56:34,240
they find that being offered
this commitment contract

1408
00:56:34,240 --> 00:56:36,580
increases production by 2.3%.

1409
00:56:36,580 --> 00:56:39,243
Now you might say 2%
is actually pretty low.

1410
00:56:39,243 --> 00:56:40,660
And that's true
in absolute terms.

1411
00:56:40,660 --> 00:56:42,050
That's not a lot of money.

1412
00:56:42,050 --> 00:56:44,950
But if you think about
what other options does

1413
00:56:44,950 --> 00:56:46,840
the employer have,
well, one thing

1414
00:56:46,840 --> 00:56:48,850
you might want to do to
increase worker's output

1415
00:56:48,850 --> 00:56:52,450
is sort of a double their wages
or just increase wages overall.

1416
00:56:52,450 --> 00:56:57,160
Now it turns out that they also
have some piece rate variations

1417
00:56:57,160 --> 00:57:00,010
where they can actually estimate
how much of a piece rate

1418
00:57:00,010 --> 00:57:03,070
increase do we need to achieve
these effects on productivity.

1419
00:57:03,070 --> 00:57:05,200
And it turns out, the
impact is actually

1420
00:57:05,200 --> 00:57:07,810
corresponding to, like, an
18% increase in the piece rate

1421
00:57:07,810 --> 00:57:08,310
wage.

1422
00:57:08,310 --> 00:57:10,810
Now, if you're an
employer, you don't

1423
00:57:10,810 --> 00:57:12,970
want to pay people,
like, an 80% higher piece

1424
00:57:12,970 --> 00:57:16,843
rate for, like, a 2% or 3%
increase in productivity.

1425
00:57:16,843 --> 00:57:18,010
That's just not worth doing.

1426
00:57:18,010 --> 00:57:19,093
It's just like a bad deal.

1427
00:57:19,093 --> 00:57:21,385
I mean, depends a little bit
on your costs and benefits

1428
00:57:21,385 --> 00:57:22,090
of production.

1429
00:57:22,090 --> 00:57:23,620
But that's a very
ineffective way

1430
00:57:23,620 --> 00:57:27,070
of getting workers to
be more productive,

1431
00:57:27,070 --> 00:57:28,810
and it's a costly thing to do.

1432
00:57:28,810 --> 00:57:31,420
You're going to incentivize a
bunch of intramarginal people

1433
00:57:31,420 --> 00:57:32,260
to do that.

1434
00:57:32,260 --> 00:57:34,660
Instead, offering commitment
devices in this setting

1435
00:57:34,660 --> 00:57:35,860
is actually free.

1436
00:57:35,860 --> 00:57:37,370
And in some sense,
for the employer,

1437
00:57:37,370 --> 00:57:38,870
you might actually
pay people less.

1438
00:57:38,870 --> 00:57:40,245
Some people don't
reach a target,

1439
00:57:40,245 --> 00:57:42,053
you can actually pay w over 2.

1440
00:57:42,053 --> 00:57:43,720
Now, you actually
don't want to do that,

1441
00:57:43,720 --> 00:57:46,600
because workers will be
not happy and, you know,

1442
00:57:46,600 --> 00:57:49,300
probably get annoyed with
you if you do that too much.

1443
00:57:49,300 --> 00:57:51,147
But the technology
is actually free.

1444
00:57:51,147 --> 00:57:53,230
Here's a thing that you
can offer to your workers.

1445
00:57:53,230 --> 00:57:56,300
They're going to be 2%
to 3% more productive.

1446
00:57:56,300 --> 00:57:59,260
That's actually a big deal
for a lot of companies,

1447
00:57:59,260 --> 00:58:01,750
in terms of the margins
are often small,

1448
00:58:01,750 --> 00:58:03,790
and a much better instrument
and getting workers

1449
00:58:03,790 --> 00:58:06,520
to be more productive compared
to increasing people's wages.

1450
00:58:06,520 --> 00:58:09,720
Because wages, obviously,
you need to pay.

1451
00:58:09,720 --> 00:58:12,430
Any questions on that?

1452
00:58:12,430 --> 00:58:14,160
And then a third--
and this is sort of

1453
00:58:14,160 --> 00:58:16,710
consistent with a model
of self-control control

1454
00:58:16,710 --> 00:58:18,390
being really the
driver of this--

1455
00:58:18,390 --> 00:58:21,670
is payday effects predict
people's demand for commitment.

1456
00:58:21,670 --> 00:58:23,512
I'm going to show
you this in a second.

1457
00:58:23,512 --> 00:58:25,470
But essentially, what
they find is that they're

1458
00:58:25,470 --> 00:58:26,820
A-- they're payday effects.

1459
00:58:26,820 --> 00:58:28,570
What do I mean by that?

1460
00:58:28,570 --> 00:58:31,470
If you plot, essentially,
production over the course

1461
00:58:31,470 --> 00:58:32,890
of the day cycle--

1462
00:58:32,890 --> 00:58:35,280
so here's-- on the very right
hand side of this graph,

1463
00:58:35,280 --> 00:58:37,590
you see people's payday
productivity compared

1464
00:58:37,590 --> 00:58:39,630
to the productivity on
the day after the payday,

1465
00:58:39,630 --> 00:58:41,360
or seven days before the payday.

1466
00:58:41,360 --> 00:58:43,860
You'll see people essentially
are more productive on payday.

1467
00:58:43,860 --> 00:58:45,600
They produce more
on paydays compared

1468
00:58:45,600 --> 00:58:46,687
to, like, previous days.

1469
00:58:46,687 --> 00:58:48,270
And so that what you
see, essentially,

1470
00:58:48,270 --> 00:58:51,300
is a constant increase
towards the day of the payday.

1471
00:58:51,300 --> 00:58:52,920
Now, somebody was
asking me previously

1472
00:58:52,920 --> 00:58:54,510
about quasi-hyperbolic
discounting,

1473
00:58:54,510 --> 00:58:58,340
or hyperbolic discounting,
which model is right.

1474
00:58:58,340 --> 00:59:00,090
When you look at this
graph, this actually

1475
00:59:00,090 --> 00:59:02,790
doesn't look very much like
quasi-hyperbolic discounting.

1476
00:59:02,790 --> 00:59:05,243
It looks a lot more like
hyperbolic discounting,

1477
00:59:05,243 --> 00:59:07,410
as in, like, there's sort
of like a smooth increase.

1478
00:59:07,410 --> 00:59:09,510
There's not a jump up.

1479
00:59:09,510 --> 00:59:12,325
On the day of the payday
people are more productive,

1480
00:59:12,325 --> 00:59:13,950
but it's rather sort
of more consistent

1481
00:59:13,950 --> 00:59:16,222
with true hyperbolic
discounting.

1482
00:59:16,222 --> 00:59:17,430
That's more of a detail here.

1483
00:59:17,430 --> 00:59:18,847
That's not that
important for you.

1484
00:59:18,847 --> 00:59:20,310
But I wanted to point that out.

1485
00:59:20,310 --> 00:59:20,810
Yes?

1486
00:59:20,810 --> 00:59:23,220
AUDIENCE: What's the scale
and unit on the y-axis?

1487
00:59:23,220 --> 00:59:24,440
FRANK SCHILBACH: That
is a great question.

1488
00:59:24,440 --> 00:59:25,480
I think its production.

1489
00:59:25,480 --> 00:59:27,540
So I think this is not a--

1490
00:59:27,540 --> 00:59:28,447
this is not a great--

1491
00:59:28,447 --> 00:59:30,030
I think these are
units of production.

1492
00:59:30,030 --> 00:59:32,820
I don't know exactly
what the fraction

1493
00:59:32,820 --> 00:59:34,480
is in terms of production.

1494
00:59:34,480 --> 00:59:36,780
But it must be, like, a
few percent of production.

1495
00:59:36,780 --> 00:59:38,043
I can look this up.

1496
00:59:38,043 --> 00:59:39,460
But it's like units
of production,

1497
00:59:39,460 --> 00:59:42,570
which is surely not what
you want to use here.

1498
00:59:42,570 --> 00:59:44,160
Yeah, and so then--

1499
00:59:44,160 --> 00:59:46,980
so what we find here is
that, essentially, people

1500
00:59:46,980 --> 00:59:48,722
are more productive
on paydays compared

1501
00:59:48,722 --> 00:59:51,180
to the day after the payday,
or compared to, like, six days

1502
00:59:51,180 --> 00:59:53,430
before the payday.

1503
00:59:53,430 --> 00:59:55,470
And then we find
another graph, which

1504
00:59:55,470 --> 00:59:59,490
is a quite nice one, which is
to say high payday workers,

1505
00:59:59,490 --> 01:00:02,160
or high payday effect workers,
are more likely to select

1506
01:00:02,160 --> 01:00:03,660
positive targets.

1507
01:00:03,660 --> 01:00:06,990
What I mean by that is
they split the sample

1508
01:00:06,990 --> 01:00:11,250
into workers who have high
payday impacts and low payday

1509
01:00:11,250 --> 01:00:11,790
impacts.

1510
01:00:11,790 --> 01:00:14,490
That is to say, when you look
at the workers and look at,

1511
01:00:14,490 --> 01:00:17,430
like, which workers are more
productive on paydays compared

1512
01:00:17,430 --> 01:00:20,820
to on other days, you can sort
of split the sample into two.

1513
01:00:20,820 --> 01:00:22,410
Some workers are
much more productive

1514
01:00:22,410 --> 01:00:24,060
on paydays compared
to other days.

1515
01:00:24,060 --> 01:00:25,695
Some workers are not.

1516
01:00:25,695 --> 01:00:27,570
So what you see here in
the graph is you see,

1517
01:00:27,570 --> 01:00:29,400
essentially, the blue dots.

1518
01:00:29,400 --> 01:00:31,945
These are workers who
have high payday impacts.

1519
01:00:31,945 --> 01:00:34,320
These are workers who are much
more productive on paydays

1520
01:00:34,320 --> 01:00:35,880
compared to other days.

1521
01:00:35,880 --> 01:00:38,100
And you see the red
dots, which are workers

1522
01:00:38,100 --> 01:00:40,410
who are, essentially,
pretty much

1523
01:00:40,410 --> 01:00:42,390
they have low payday
impact, meaning

1524
01:00:42,390 --> 01:00:45,780
they kind of work the same on
paydays versus on other days.

1525
01:00:45,780 --> 01:00:48,630
And then what they show on
the-- so then, on the x-axis,

1526
01:00:48,630 --> 01:00:49,590
they show experience.

1527
01:00:49,590 --> 01:00:51,510
This is essentially--
think of this

1528
01:00:51,510 --> 01:00:52,750
as the course of the study.

1529
01:00:52,750 --> 01:00:56,040
These are something
like 150 days.

1530
01:00:56,040 --> 01:00:59,760
They have different workdays
in the sample, where

1531
01:00:59,760 --> 01:01:02,760
essentially, over time,
people get to choose

1532
01:01:02,760 --> 01:01:04,530
over and over their targets.

1533
01:01:04,530 --> 01:01:08,310
And what you see, essentially,
is that the blue graph--

1534
01:01:08,310 --> 01:01:11,490
the blue dots tend to be A,
higher than the red graphs,

1535
01:01:11,490 --> 01:01:13,502
meaning that people have
high payday impacts.

1536
01:01:13,502 --> 01:01:15,210
These are workers of
high payday effects.

1537
01:01:15,210 --> 01:01:17,190
They're like more
productive on paydays.

1538
01:01:17,190 --> 01:01:21,192
They're more likely to
choose positive targets.

1539
01:01:21,192 --> 01:01:22,900
Sorry, I should have
said, on the y-axis,

1540
01:01:22,900 --> 01:01:26,880
you see the fraction of people
choosing positive targets, OK?

1541
01:01:26,880 --> 01:01:28,600
So what we see
essentially is two things.

1542
01:01:28,600 --> 01:01:31,020
One is the blue dots are
higher than the red dots,

1543
01:01:31,020 --> 01:01:33,600
meaning that the people
have high payday impacts.

1544
01:01:33,600 --> 01:01:37,140
They are more likely to
choose positive targets, sort

1545
01:01:37,140 --> 01:01:39,630
of suggesting that
the underlying reason

1546
01:01:39,630 --> 01:01:42,270
why people have high payday
impact is self-control.

1547
01:01:42,270 --> 01:01:45,510
You're more, essentially,
productive on paydays

1548
01:01:45,510 --> 01:01:48,070
because you're present
biased one way or the other.

1549
01:01:48,070 --> 01:01:51,557
And that also then predicts
whether you choose the target.

1550
01:01:51,557 --> 01:01:53,640
So people who have higher
self-control problems as

1551
01:01:53,640 --> 01:01:55,587
revealed by the
payday effects, those

1552
01:01:55,587 --> 01:01:57,420
are the people who are
more likely to choose

1553
01:01:57,420 --> 01:01:59,130
positive targets.

1554
01:01:59,130 --> 01:02:02,310
Second, we also see
that the blue dots

1555
01:02:02,310 --> 01:02:05,202
tend to be trending
upwards in the graph.

1556
01:02:05,202 --> 01:02:06,660
You see on the
right, the blue dots

1557
01:02:06,660 --> 01:02:08,970
are higher than on
the left in the graph,

1558
01:02:08,970 --> 01:02:11,850
meaning that as people have more
experience in the study, when

1559
01:02:11,850 --> 01:02:14,380
they have like 100,
150 days in the study,

1560
01:02:14,380 --> 01:02:17,100
they're more likely to
choose positive targets,

1561
01:02:17,100 --> 01:02:20,920
consistent with some learning
about self-control over time.

1562
01:02:20,920 --> 01:02:23,340
So they learn over time that
choosing positive targets

1563
01:02:23,340 --> 01:02:25,132
is a good thing for
them, potentially, that

1564
01:02:25,132 --> 01:02:26,850
makes them more productive.

1565
01:02:26,850 --> 01:02:29,340
You don't see such a
thing for the red dots

1566
01:02:29,340 --> 01:02:33,198
that seem to be essentially
flat or constant over time.

1567
01:02:33,198 --> 01:02:34,740
Sorry, that was a
lot of information.

1568
01:02:34,740 --> 01:02:36,270
Are there any questions on that?

1569
01:02:40,950 --> 01:02:43,520
Yes?

1570
01:02:43,520 --> 01:02:45,880
AUDIENCE: Arguably,
there's lots to do

1571
01:02:45,880 --> 01:02:47,468
with the savings of
the payday, right?

1572
01:02:47,468 --> 01:02:49,510
So on the payday, I have
to come in to the office

1573
01:02:49,510 --> 01:02:50,674
and do the work.

1574
01:02:50,674 --> 01:02:53,880
So I might as well type a little
longer to earn a little more.

1575
01:02:53,880 --> 01:02:56,138
Another is you don't
really have an incentive.

1576
01:02:56,138 --> 01:02:57,180
They're not getting paid.

1577
01:02:57,180 --> 01:02:58,780
So there are other
jobs where you

1578
01:02:58,780 --> 01:03:00,990
get paid at the end
of the day, every day.

1579
01:03:00,990 --> 01:03:03,790
And then you don't really
find the payday effects

1580
01:03:03,790 --> 01:03:05,760
in those kinds of jobs, right?

1581
01:03:05,760 --> 01:03:07,220
FRANK SCHILBACH:
Right, so exactly.

1582
01:03:07,220 --> 01:03:08,970
So there's an issue
on, when you think

1583
01:03:08,970 --> 01:03:12,602
about these payday effects, what
exactly do we learn from that?

1584
01:03:12,602 --> 01:03:14,310
One way to think about
the payday effects

1585
01:03:14,310 --> 01:03:16,143
is kind of like the
story I was telling you,

1586
01:03:16,143 --> 01:03:18,870
which is to say, on days when
you come in, when you type,

1587
01:03:18,870 --> 01:03:21,420
when you work, you're going
to be paid in the evening.

1588
01:03:21,420 --> 01:03:24,180
And you're going to work
harder, because your reward is

1589
01:03:24,180 --> 01:03:25,440
closer to you--

1590
01:03:25,440 --> 01:03:26,773
to your effort.

1591
01:03:26,773 --> 01:03:28,440
Another explanation
that you were saying

1592
01:03:28,440 --> 01:03:29,850
is like, well, what
if workers are just,

1593
01:03:29,850 --> 01:03:31,060
like, liquidity constrained.

1594
01:03:31,060 --> 01:03:33,910
What if their children are
hungry, et cetera, and so on.

1595
01:03:33,910 --> 01:03:37,920
So you come in on that day
because you want a paycheck.

1596
01:03:37,920 --> 01:03:40,380
Once you show up anyway, you
might as well do some work.

1597
01:03:40,380 --> 01:03:45,540
And you're going to end up
being more productive that way.

1598
01:03:45,540 --> 01:03:48,090
That's surely, in
part, going on.

1599
01:03:48,090 --> 01:03:49,990
I have sort of two
responses to that.

1600
01:03:49,990 --> 01:03:53,250
So one response is, like, well,
we find that the workers with

1601
01:03:53,250 --> 01:03:56,580
high payday impacts are more
likely to choose positive

1602
01:03:56,580 --> 01:03:59,550
targets, which sort of suggests
there is an underlying issue

1603
01:03:59,550 --> 01:04:03,660
which is self-control problems
that drives both the payday

1604
01:04:03,660 --> 01:04:06,770
effects and the high--

1605
01:04:06,770 --> 01:04:08,790
the positive targets
that people choose

1606
01:04:08,790 --> 01:04:10,530
to demand for commitment.

1607
01:04:10,530 --> 01:04:13,470
Second, and more subtle in
some ways, is, like, people

1608
01:04:13,470 --> 01:04:14,670
are liquidity constrained.

1609
01:04:14,670 --> 01:04:16,313
People who don't
have cash, there's

1610
01:04:16,313 --> 01:04:17,730
often a reason why
they don't have

1611
01:04:17,730 --> 01:04:19,560
cash, which is often
because they haven't

1612
01:04:19,560 --> 01:04:20,832
saved in the first place.

1613
01:04:20,832 --> 01:04:22,290
Again, that's a
little complicated.

1614
01:04:22,290 --> 01:04:25,290
But essentially, if somebody
is very liquidity constrained,

1615
01:04:25,290 --> 01:04:27,510
never has cash,
often that's coming--

1616
01:04:27,510 --> 01:04:28,980
or the underlying
reason might be

1617
01:04:28,980 --> 01:04:31,260
present bias, or some form
of self-control problems

1618
01:04:31,260 --> 01:04:32,490
to start with.

1619
01:04:32,490 --> 01:04:34,770
The reason being that,
like, if you really

1620
01:04:34,770 --> 01:04:36,838
had such a value-- high
value of having cash,

1621
01:04:36,838 --> 01:04:38,880
you should just save it
and have it in your home,

1622
01:04:38,880 --> 01:04:42,070
or try to save money in
a bank account and so on.

1623
01:04:42,070 --> 01:04:44,070
So usually, we think that
when people are really

1624
01:04:44,070 --> 01:04:45,810
liquidity constrained,
often that's

1625
01:04:45,810 --> 01:04:48,960
an underlying reason
of present bias

1626
01:04:48,960 --> 01:04:50,998
or some form of
self-control problems.

1627
01:04:50,998 --> 01:04:53,040
Having said that, there's
also some other reasons

1628
01:04:53,040 --> 01:04:54,030
why people can't save.

1629
01:04:54,030 --> 01:04:56,640
Because it might be they don't
have access to bank accounts,

1630
01:04:56,640 --> 01:04:58,170
and so on, and so forth.

1631
01:05:03,400 --> 01:05:05,440
OK.

1632
01:05:05,440 --> 01:05:09,230
OK, so let me tell you
one more application.

1633
01:05:09,230 --> 01:05:13,700
And let me sort of continue
with the rest of those tomorrow.

1634
01:05:13,700 --> 01:05:17,620
So this is sort of like
a classic study, in fact,

1635
01:05:17,620 --> 01:05:20,308
done in Boston in Boston
area health clubs.

1636
01:05:20,308 --> 01:05:21,850
So this is Dellavigna
and Malmendier.

1637
01:05:21,850 --> 01:05:24,495
And what they did is they
looked at health clubs

1638
01:05:24,495 --> 01:05:25,870
and different
options that people

1639
01:05:25,870 --> 01:05:29,320
did in gyms when they had
choices between the following,

1640
01:05:29,320 --> 01:05:33,070
or the following kind, which
is monthly fees of over $70

1641
01:05:33,070 --> 01:05:38,080
for unlimited use of the gym,
or a pay-per-visit fee of $10.

1642
01:05:38,080 --> 01:05:39,938
If you had that option
or those two options,

1643
01:05:39,938 --> 01:05:41,230
which options would you choose?

1644
01:05:41,230 --> 01:05:42,938
Or why would you choose
one or the other?

1645
01:05:47,050 --> 01:05:47,550
Yes?

1646
01:05:47,550 --> 01:05:49,008
AUDIENCE: If you
think you're going

1647
01:05:49,008 --> 01:05:52,330
to use the gym more than seven
times in a month, [INAUDIBLE]..

1648
01:05:55,547 --> 01:05:56,880
FRANK SCHILBACH: Right, exactly.

1649
01:05:56,880 --> 01:06:00,600
So why pay more than-- why pick
option number one if you only

1650
01:06:00,600 --> 01:06:03,102
go three times?

1651
01:06:03,102 --> 01:06:04,560
When I was teaching
the first time,

1652
01:06:04,560 --> 01:06:08,090
I was doing exactly
that, in fact.

1653
01:06:08,090 --> 01:06:10,380
But exactly as you
say, if you make--

1654
01:06:10,380 --> 01:06:13,200
if you understand how
often you go to the gym,

1655
01:06:13,200 --> 01:06:16,710
why choose option one if you
don't go at least seven times?

1656
01:06:16,710 --> 01:06:18,930
There could be some
transaction costs and the like.

1657
01:06:18,930 --> 01:06:21,040
But surely, if you go
only once or twice,

1658
01:06:21,040 --> 01:06:23,580
you should not
choose option one.

1659
01:06:23,580 --> 01:06:26,160
Now of course, some people
might choose option one anyway.

1660
01:06:26,160 --> 01:06:27,820
Why might you do that?

1661
01:06:27,820 --> 01:06:28,320
Yes?

1662
01:06:28,320 --> 01:06:30,237
AUDIENCE: Would it be
for a commitment device?

1663
01:06:30,237 --> 01:06:32,070
Like, I would pay for
this month [INAUDIBLE]

1664
01:06:32,070 --> 01:06:33,430
I want to get my money's worth.

1665
01:06:33,430 --> 01:06:36,540
FRANK SCHILBACH:
Right, So one option

1666
01:06:36,540 --> 01:06:39,580
is, like, essentially
change their future prices.

1667
01:06:39,580 --> 01:06:42,330
So notice that the marginal cost
of going to the gym in option

1668
01:06:42,330 --> 01:06:44,070
two is $10, right?

1669
01:06:44,070 --> 01:06:46,710
Every time you go, your
marginal cost is $10.

1670
01:06:46,710 --> 01:06:48,953
In option one, the
marginal cost is 0.

1671
01:06:48,953 --> 01:06:51,120
So what I'm trying to do
is change the marginal cost

1672
01:06:51,120 --> 01:06:56,910
in the future to make it
more favorable for me to go.

1673
01:06:56,910 --> 01:06:59,622
What else could be going on?

1674
01:06:59,622 --> 01:07:01,830
So one is they could use
this as a commitment device.

1675
01:07:01,830 --> 01:07:05,160
What's another explanation?

1676
01:07:05,160 --> 01:07:05,660
Yeah?

1677
01:07:05,660 --> 01:07:08,005
AUDIENCE: Overestimation of
how many times you would go?

1678
01:07:08,005 --> 01:07:10,380
FRANK SCHILBACH: Exactly, it
could be like, essentially--

1679
01:07:10,380 --> 01:07:11,755
and there's two
versions of that.

1680
01:07:11,755 --> 01:07:15,313
One version has to do with beta
delta or present bias, which

1681
01:07:15,313 --> 01:07:16,730
is to say I'm just
underestimating

1682
01:07:16,730 --> 01:07:18,350
my future present bias.

1683
01:07:18,350 --> 01:07:21,620
I think my beta is, like,
1, but in fact, it's 0.5.

1684
01:07:21,620 --> 01:07:23,520
I think I'm going
to go 17 times,

1685
01:07:23,520 --> 01:07:26,670
but in fact, I'm
going only twice.

1686
01:07:26,670 --> 01:07:27,630
That's one option.

1687
01:07:27,630 --> 01:07:29,720
Another option that was
also mentioned earlier

1688
01:07:29,720 --> 01:07:32,990
was about, like, underestimating
the costs of future exercising.

1689
01:07:32,990 --> 01:07:34,490
That is to say,
what often happens

1690
01:07:34,490 --> 01:07:36,503
when people sign up
for the gym, often

1691
01:07:36,503 --> 01:07:37,670
they are already at the gym.

1692
01:07:37,670 --> 01:07:40,100
They are really excited
to exercise, and so on.

1693
01:07:40,100 --> 01:07:42,915
And they might underestimate
how costly it actually

1694
01:07:42,915 --> 01:07:44,540
is once they sit at
home in the evening

1695
01:07:44,540 --> 01:07:46,250
coming back home
from work and so on.

1696
01:07:46,250 --> 01:07:49,040
They might underestimate how
it feels, how costly it is

1697
01:07:49,040 --> 01:07:50,645
for them to actually exercise.

1698
01:07:50,645 --> 01:07:52,520
Nothing to do with
self-control specifically,

1699
01:07:52,520 --> 01:07:54,180
but it's just it's a
tedious thing to do.

1700
01:07:54,180 --> 01:07:55,460
They might
underestimate how tired

1701
01:07:55,460 --> 01:07:57,043
they are, and so on,
and so forth-- so

1702
01:07:57,043 --> 01:07:59,790
some form of like
underestimation of the cost.

1703
01:07:59,790 --> 01:08:02,480
That's, again, hard to
separate in many cases.

1704
01:08:02,480 --> 01:08:04,650
But that could also be going on.

1705
01:08:04,650 --> 01:08:09,140
So what they find then is people
exercise, on average, 4.3 times

1706
01:08:09,140 --> 01:08:10,440
a month in the first year.

1707
01:08:10,440 --> 01:08:12,470
That's about $17 per visit.

1708
01:08:12,470 --> 01:08:15,650
And of course, you should
be choosing option two.

1709
01:08:15,650 --> 01:08:19,130
Before canceling, consumers
go 2.3 months, on average,

1710
01:08:19,130 --> 01:08:20,840
without using the
gym at all, which

1711
01:08:20,840 --> 01:08:24,380
is kind of like it's even
tedious to cancel the gym

1712
01:08:24,380 --> 01:08:25,520
and actually go there.

1713
01:08:25,520 --> 01:08:27,760
And I did, actually,
exactly that as well.

1714
01:08:27,760 --> 01:08:31,109
And so how do we think about
quasi-hyperbolic discounting

1715
01:08:31,109 --> 01:08:31,609
here?

1716
01:08:31,609 --> 01:08:33,800
Well, the gym-goers
would like to exercise

1717
01:08:33,800 --> 01:08:34,771
a lot in the future.

1718
01:08:34,771 --> 01:08:36,229
Being naive, that's
what they think

1719
01:08:36,229 --> 01:08:38,750
they will do, so, you know,
overestimate what they will do.

1720
01:08:38,750 --> 01:08:42,979
To save on gym costs, they
buy the monthly membership.

1721
01:08:42,979 --> 01:08:44,990
So that's kind of what
the naive person does.

1722
01:08:44,990 --> 01:08:46,460
When it comes down
to exercising,

1723
01:08:46,460 --> 01:08:48,050
their short-run
inpatient kicks in,

1724
01:08:48,050 --> 01:08:50,255
and they do end up ending
the membership much.

1725
01:08:50,255 --> 01:08:52,130
That's very much like
sort of underestimating

1726
01:08:52,130 --> 01:08:54,590
or overestimating their
beta or underestimating

1727
01:08:54,590 --> 01:08:55,910
their self-control problem.

1728
01:08:55,910 --> 01:08:57,729
Now, what does the
sophisticated person do?

1729
01:08:57,729 --> 01:09:00,620
We already discussed that
as well, which is they also

1730
01:09:00,620 --> 01:09:02,450
prefer to exercise
a lot in the future,

1731
01:09:02,450 --> 01:09:04,740
but they realize they
don't want to do so later.

1732
01:09:04,740 --> 01:09:08,840
So what they want to do is they
choose the monthly contract

1733
01:09:08,840 --> 01:09:11,090
as a form of commitment
device, in the sense of,

1734
01:09:11,090 --> 01:09:14,600
like, making future actions
changing prices in the future,

1735
01:09:14,600 --> 01:09:16,622
they make it now cheaper
to exercise compared

1736
01:09:16,622 --> 01:09:18,830
to sit at home by sort of
changing the marginal costs

1737
01:09:18,830 --> 01:09:19,640
in the future.

1738
01:09:19,640 --> 01:09:21,870
And they might even be
willing to pay for that.

1739
01:09:21,870 --> 01:09:23,510
So they might
actually say, I know

1740
01:09:23,510 --> 01:09:25,970
that I'm only going to go
like three or four times.

1741
01:09:25,970 --> 01:09:27,470
But if I don't
choose that contract,

1742
01:09:27,470 --> 01:09:31,220
I'm going to only go 0 or 1
times, and it's worth for me

1743
01:09:31,220 --> 01:09:33,212
to do that.

1744
01:09:33,212 --> 01:09:34,670
So now, there's a
different version

1745
01:09:34,670 --> 01:09:36,295
of-- in which one
could do a commitment

1746
01:09:36,295 --> 01:09:39,020
devices about dealing
with temptation, which

1747
01:09:39,020 --> 01:09:42,590
is a very clever bundling of
temptations paper by Milkman

1748
01:09:42,590 --> 01:09:43,399
and others.

1749
01:09:43,399 --> 01:09:45,859
What they did-- instead of
offering commitment devices

1750
01:09:45,859 --> 01:09:50,000
directly, what they did
is they bundle two things.

1751
01:09:50,000 --> 01:09:52,399
I told you previously about
investment goods and leisure

1752
01:09:52,399 --> 01:09:52,939
goods.

1753
01:09:52,939 --> 01:09:55,340
And so investment goods are
things like going to the gym

1754
01:09:55,340 --> 01:09:56,990
that you do not enough of.

1755
01:09:56,990 --> 01:09:59,480
Leisure goods are things
that you enjoy in the present

1756
01:09:59,480 --> 01:10:01,480
and you might do
too much of that.

1757
01:10:01,480 --> 01:10:03,230
So now, what you can
do is you essentially

1758
01:10:03,230 --> 01:10:04,170
bundle those things.

1759
01:10:04,170 --> 01:10:07,220
So what they did is they offered
people the option to only

1760
01:10:07,220 --> 01:10:10,070
listen to audiobooks--
addictive audiobooks--

1761
01:10:10,070 --> 01:10:11,240
at the gym.

1762
01:10:11,240 --> 01:10:12,637
And so then, you
can only sort of

1763
01:10:12,637 --> 01:10:13,970
do that while you're at the gym.

1764
01:10:13,970 --> 01:10:15,170
And you might come
back to the gym,

1765
01:10:15,170 --> 01:10:17,045
not necessarily because
you want to exercise,

1766
01:10:17,045 --> 01:10:19,850
but because you want to
listen to your audiobook.

1767
01:10:19,850 --> 01:10:20,960
This can also backfire.

1768
01:10:20,960 --> 01:10:23,750
I have a friend of mine,
also in grad school, who

1769
01:10:23,750 --> 01:10:25,430
was watching lots of TV shows.

1770
01:10:25,430 --> 01:10:28,850
And he convinced himself that
he could only watch TV shows

1771
01:10:28,850 --> 01:10:30,793
while being at the gym.

1772
01:10:30,793 --> 01:10:32,210
So then, I would
be in the office,

1773
01:10:32,210 --> 01:10:34,835
and he would sometimes come back
from the gym totally exhausted

1774
01:10:34,835 --> 01:10:37,730
because he had watched like
three episodes of certain TV

1775
01:10:37,730 --> 01:10:40,550
shows while being
on the treadmill.

1776
01:10:40,550 --> 01:10:43,760
And so he was kind of
like overexercising

1777
01:10:43,760 --> 01:10:47,840
because of his consumption--
his bundling of temptations.

1778
01:10:47,840 --> 01:10:49,190
But in this case, for--

1779
01:10:49,190 --> 01:10:50,810
so you have to sort of figure
out how to calibrate this,

1780
01:10:50,810 --> 01:10:51,350
right?

1781
01:10:51,350 --> 01:10:53,433
But it could be a way of
getting you, essentially,

1782
01:10:53,433 --> 01:10:56,060
to-- by bundling, essentially,
pleasant and unpleasant things

1783
01:10:56,060 --> 01:10:58,925
at different points in time, by
bundling them in certain ways,

1784
01:10:58,925 --> 01:11:00,800
it might make you-- sort
of help you overcome

1785
01:11:00,800 --> 01:11:03,380
two issues at the time, which
is one, you go to the gym

1786
01:11:03,380 --> 01:11:04,250
more often.

1787
01:11:04,250 --> 01:11:06,710
B, you might not sort
of watch too much TV

1788
01:11:06,710 --> 01:11:09,620
if you can convince yourself
to actually follow through.

1789
01:11:09,620 --> 01:11:11,690
They do find, in fact,
bundling of temptation

1790
01:11:11,690 --> 01:11:14,060
is effective until
the Thanksgiving

1791
01:11:14,060 --> 01:11:17,810
break, when sort of everything
goes downhill in their study.

1792
01:11:17,810 --> 01:11:19,190
OK, that's all I have for now.

1793
01:11:19,190 --> 01:11:22,250
I'm going to continue on
tomorrow with the remaining

1794
01:11:22,250 --> 01:11:24,300
parts of time preferences.

1795
01:11:24,300 --> 01:11:25,990
Thank you.