Popular in Social Sciences

Banks and Policy Organisations

Macroeconomic modelling of CBDC: a critical review

European Central Bank Ulrich Bindseil

Social Identities and Divisions

A new explanation of antisemitism: Jew hatred as a civilisational phenomenon

Boston University Liah Greenfeld

Banks and Policy Organisations

Financing Private Credit

Federal Reserve Bank of New York Nina Boyarchenko

Economics

Failing Banks

Massachusetts Institute of Technology Emil Verner

Culture and Society

The City as a Growth Machine: Toward a Political Economy of Place

New York University Harvey Moltoch

American Politics

Keeping the Republic

Boston College Marc Landy

British Politics

Entrepreneurial thinking: the politics and practice of policy impact

University of Sheffield Matthew Flinders

Consumption dynamics and welfare under non-Gaussian earnings risk

View more insights in Social Sciences - Economics - Labor Economics

View more figures from University of Minnesota

View the page of Fatih Guvenen

Read the Study

Next in Economics

xtevent: Estimation and Visualization in the Linear Panel Event-Study Design

Simon Freyaldenhoven | Federal Reserve Bank of Philadelphia

Time-varying Persistence of House Price Growth

C.Y. Choi | Federal Reserve Bank of Dallas

How does non-Gaussian earnings risk impact individual welfare and consumption patterns in different economic settings, especially considering the key role of earnings dynamics in shaping economic behavior?

Image courtesy of interviewee. November 19, 2024

Log-in or Sign-up to Faculti
Currently viewing this subject insight as a guest. You have insight(s) remaining for this month. Login to view 8000+ figures on the platform.

Generating Workspace...

Export as PDF

This video discusses a seminal study by macroeconomist Bob Lucas from 1987, which posed a hypothetical scenario in which all recessions could be eliminated, leading to consistent economic growth. Lucas found that the cost of eliminating business cycle fluctuations would be minimal—0.1% of consumption per year. This video contradicted the widespread perception of the significance of recessions.

Over the years, numerous economists challenged this conclusion, and Lucas eventually updated his findings in a 2003 address, revising the cost down to an even smaller figure, 0.01%. This video sparked further debate about the true costs of income fluctuations individuals experience, which were found to be drastically larger than macroeconomic indicators like GDP fluctuations suggest.

The author highlights that while overall GDP fluctuations appear minor (1% to 2% standard deviation), individual income volatility is around 50%. This video discrepancy implies that people are more affected by income risks than macroeconomic cycles. The study's purpose is to explore the costs individuals would be willing to pay to eliminate income fluctuations. Results suggest that individuals might be willing to pay up to 40% of their income to avoid such risks.

The analysis benefited from improved access to comprehensive earnings data post-Great Recession, allowing for more accurate modeling of individual risks without reliance on outdated Gaussian assumptions. The findings indicate significant differences in income management and the long-term effects of both positive and negative shocks. Notably, negative shocks are often persistent for low-income individuals, while positive shocks for higher earners tend to be more durable.

The implications stress the need for better economic models that accurately reflect the realities of income risk. The Gaussian model traditionally underestimates these risks, leading to inappropriate policies. Therefore, understanding the severe income fluctuations and their true societal costs is vital for crafting effective economic policies. Overall, the paper argues for a shift in economic modeling frameworks to better account for the asymmetric nature of income risks and their broader implications on policy-making.

Language:

Export as PDF

Export as PDF
Here are five key points from the video:

1. **Bob Lucas's Hypothetical Question**: In 1987, economist Bob Lucas posed a question about the "cost of business cycles," hypothesizing whether eliminating recessions and maintaining a steady economic growth rate would significantly affect consumption. His original calculation suggested that the public would only be willing to pay a minuscule 0.1% of their consumption per year to eliminate these fluctuations. Subsequent research indicated an even smaller number of 0.01%.

2. **Income Volatility and Individual Risk**: The findings suggest that while the volatility of GDP growth is relatively low (around 1-2% per year), individual income fluctuations are much more significant, with estimates of volatility around 50%. This discrepancy highlights that discussions about economic risk often overlook the larger, more persistent risks individuals face, which are not solely tied to recessions.

3. **Impact of Big Data on Economic Research**: The availability of large datasets, particularly in the wake of the Great Recession, has transformed economic research. Researchers now have access to comprehensive earnings histories, allowing for a clearer understanding of income risks and the ability to discard outdated assumptions about income shocks, such as the Gaussian distribution.

4. **Non-Gaussian Income Shocks**: The recent research demonstrates that income shocks do not follow a Gaussian distribution. Instead, they exhibit features like long tails and high peaks, indicating that extreme income changes (both positive and negative) are more common than previously thought. This understanding alters how income risk is assessed, suggesting that traditional models significantly underestimate the probability and impact of severe income shocks.

5. **Policy Implications**: The flawed assumptions surrounding income risk modeling can lead to inadequate policy responses, particularly regarding unemployment insurance, retirement savings, and taxation. The research urges a reevaluation of how income risks are modeled to better reflect reality, emphasizing that policies should consider the high likelihood of extreme income drops, which has significant implications for social safety nets and economic resilience.

Export as PDF
Here are some key quotes and ideas extracted from the video:

1. **Bob Lucas's Hypothetical Question**:
- "He said, suppose that I can eliminate all recessions... how much would you be willing to pay me?"

2. **Cost of Business Cycles**:
- "The number was not 0.1% of consumption per year... the updated number was 0.01%, it was even smaller."

3. **Individual versus Aggregate Income Fluctuations**:
- "If you look at the individuals... the volatility is close to 50%, not one or 2%."

4. **Big Data and Income Risk**:
- "After the Great Recession... we went from data sets on earnings from surveys of a thousand people to data sets on entire populations."

5. **Non-Gaussian Income Shocks**:
- "What we found... looked nothing like Gaussian. We had huge peaks in the middle. We had very long tails."

6. **Positive and Negative Shocks**:
- "Positive shocks usually are associated with a good outcome... negative shocks are associated with loss or adverse events."

7. **Risk Aversion and Income Stability**:
- "...people would be willing to give up 40% of their income to eliminate these fluctuations."

8. **Implications of Gaussian Assumptions**:
- "If we have the wrong model... we may just be happy driving our car without worrying about..."

9. **Policy Implications**:
- "If you actually take this income process, it's more than 40%... these are two very different pictures."

10. **Minimum Income Floor**:
- "We said that even if the model predicts that your income goes down to zero, your consumption will not be close to zero."

These quotes capture the essence of the discussion regarding business cycles, income risk, the impact of big data on economic modeling, and the implications for economic policy.

X-TIMESTAMP-MAP=LOCAL:00:00:00.000,MPEGTS:90000

1
00:00:00.035 --> 00:00:04.565
Need to go back a little bit to 1987 when the very famous,

2
00:00:04.785 --> 00:00:09.525
macro economist, Bob Lucas, asked a question,

3
00:00:09.935 --> 00:00:13.085
about what we now call the cost of business cycles, right?

4
00:00:13.105 --> 00:00:16.205
We talked so much about recessions financial crisis,

5
00:00:17.145 --> 00:00:20.005
and, he had done 20 years of work on that,

6
00:00:20.975 --> 00:00:23.475
but then he got convinced that economic growth,

7
00:00:24.235 --> 00:00:25.355
studying growth as opposed

8
00:00:25.355 --> 00:00:27.595
to fluctuations could be more important.

9
00:00:28.095 --> 00:00:29.835
So he asked this hypothetical question.

10
00:00:29.935 --> 00:00:33.475
He said, suppose that this, hypothetical,

11
00:00:33.475 --> 00:00:36.835
he called like an alien central banker, came to earth,

12
00:00:37.095 --> 00:00:40.395
and he said that I can eliminate all recessions.

13
00:00:41.175 --> 00:00:44.715
You know, the economy will grow at a constant 2% rate

14
00:00:44.765 --> 00:00:45.835
along the trend line.

15
00:00:46.415 --> 00:00:49.075
Um, how much would you be willing to pay me

16
00:00:49.455 --> 00:00:50.755
to the, to the US public?

17
00:00:51.645 --> 00:00:54.345
And this is a simple calculation as it turns out, you know,

18
00:00:54.345 --> 00:00:57.385
when, when he calculated under different assumptions,

19
00:00:58.165 --> 00:01:02.185
he found the number to be like literally ridiculously small.

20
00:01:02.185 --> 00:01:06.265
It was like 0.1% of consumption per year.

21
00:01:07.085 --> 00:01:08.945
So to translate it is, it was

22
00:01:08.945 --> 00:01:11.825
as if like people didn't care about this fluctuations.

23
00:01:11.825 --> 00:01:15.065
Now, when we go through them, we find them to be, you know,

24
00:01:15.365 --> 00:01:17.495
really big, deal.

25
00:01:18.075 --> 00:01:22.495
But when you look in the context of the entire history,

26
00:01:22.605 --> 00:01:26.295
they were really overwhelmed by the trend growth rate.

27
00:01:27.155 --> 00:01:29.535
So a, a large literature emerged

28
00:01:29.535 --> 00:01:31.535
after that trying to overturn this

29
00:01:31.535 --> 00:01:33.495
because he himself included,

30
00:01:33.705 --> 00:01:36.095
wrote papers on business cycles for 20 years.

31
00:01:36.395 --> 00:01:38.415
And people said, no, no, no, you just, you know,

32
00:01:38.485 --> 00:01:39.575
ignored this fact.

33
00:01:39.855 --> 00:01:42.295
I will edit to my model, it'll be a bigger number.

34
00:01:42.685 --> 00:01:44.215
That fact, it'll be a bigger number.

35
00:01:44.875 --> 00:01:48.295
So to make a long story short, over 15 years

36
00:01:48.305 --> 00:01:51.545
after his paper, maybe hundreds

37
00:01:51.545 --> 00:01:52.865
of papers were written on this,

38
00:01:53.485 --> 00:01:56.345
and, Bob Lucas gave this presidential address

39
00:01:56.345 --> 00:01:58.785
to the American Economic Association in 2003.

40
00:01:59.485 --> 00:02:02.505
And he said, well, you know, we have all this work

41
00:02:02.505 --> 00:02:03.705
that challenged my paper.

42
00:02:04.045 --> 00:02:06.905
And he kind of wished that some of them could be true,

43
00:02:06.905 --> 00:02:08.625
because, you know, he invested so much in that.

44
00:02:09.005 --> 00:02:10.985
He said, here is my updated number.

45
00:02:11.035 --> 00:02:15.265
Based on all the new research, the number was not 0.1%

46
00:02:15.285 --> 00:02:17.305
of consumption per year.

47
00:02:17.605 --> 00:02:19.705
So if you think about, you know, like,

48
00:02:19.965 --> 00:02:24.225
income per capita being $60,000, that is something like,

49
00:02:24.405 --> 00:02:26.625
you know, $600 a year, right?

50
00:02:26.625 --> 00:02:27.705
So it's something so small.

51
00:02:28.285 --> 00:02:32.065
The updated number was 0.01%,

52
00:02:32.925 --> 00:02:34.065
it was even smaller.

53
00:02:34.205 --> 00:02:35.825
So again, that caused an uproar,

54
00:02:36.795 --> 00:02:40.215
but there's so much work on that that

55
00:02:41.435 --> 00:02:42.735
we thought about something

56
00:02:42.805 --> 00:02:45.215
that is much more closer to us, right?

57
00:02:45.235 --> 00:02:49.455
You know, if you look at overall wage fluctuations

58
00:02:49.485 --> 00:02:52.375
that a worker experiences during their lifetime,

59
00:02:53.545 --> 00:02:55.975
these aggregate fluctuations, the reason

60
00:02:56.005 --> 00:02:59.285
that calculation turned out to be so small is

61
00:02:59.285 --> 00:03:02.845
because the volatility, which we measure by, say,

62
00:03:02.845 --> 00:03:06.685
the standard deviation of GDP growth, okay,

63
00:03:06.945 --> 00:03:11.845
is something like 1% per year, or at max it is 2% per year.

64
00:03:12.305 --> 00:03:13.805
And that's minuscule in a way.

65
00:03:14.265 --> 00:03:18.245
But if you look at the individuals like you and me,

66
00:03:18.465 --> 00:03:19.885
and you know, many others out there,

67
00:03:20.545 --> 00:03:24.125
if you calculate the same statistic of volatility

68
00:03:25.015 --> 00:03:27.235
or individual income,

69
00:03:27.455 --> 00:03:30.035
in the US in many countries actually that we have data on,

70
00:03:30.295 --> 00:03:33.955
it is close to 50%, not one or 2%.

71
00:03:34.185 --> 00:03:36.675
It's not like our income is not fluctuating,

72
00:03:36.935 --> 00:03:39.875
but it is fluctuating for many reasons

73
00:03:40.185 --> 00:03:42.395
that are not the recessions.

74
00:03:42.395 --> 00:03:44.235
They're not the financial, you know, crisis.

75
00:03:45.335 --> 00:03:48.435
And we realize we have never asked the same question

76
00:03:49.185 --> 00:03:53.955
that Bob asked, Bob Lucas asked for this much bigger risk.

77
00:03:54.295 --> 00:03:56.795
So that is motivation for the paper to ask.

78
00:03:57.245 --> 00:04:00.595
Again, suppose an alien kind of scientist came from,

79
00:04:00.695 --> 00:04:03.035
you know, Mars, maybe we'll meet them if we go there

80
00:04:03.375 --> 00:04:05.275
and said, how much are you willing to pay me?

81
00:04:05.815 --> 00:04:10.195
And our answer turns out to be substantial, that actually

82
00:04:10.985 --> 00:04:14.075
even for a moderate risk conversion,

83
00:04:14.135 --> 00:04:17.555
people will be willing to pay, up to 40%

84
00:04:17.575 --> 00:04:21.995
of their annual income to eliminate these fluctuations.

85
00:04:22.425 --> 00:04:24.875
This paper builds, it's like a continuation

86
00:04:24.895 --> 00:04:27.075
or outgrowth of my,

87
00:04:27.235 --> 00:04:29.355
research agenda in the last maybe 15 years.

88
00:04:30.055 --> 00:04:33.115
Um, actually I worked on income risk throughout my career,

89
00:04:33.615 --> 00:04:36.555
but in the last 15 years, what has changed,

90
00:04:37.255 --> 00:04:39.795
has been the availability of big data.

91
00:04:40.095 --> 00:04:42.595
And this will surprise no one, right? We hear it everywhere.

92
00:04:43.175 --> 00:04:46.315
And in economics, it's, this started about maybe 15

93
00:04:46.315 --> 00:04:49.435
or so years ago, I would say, after the great recession.

94
00:04:50.255 --> 00:04:53.955
And, I view it as a silver lining, so to speak,

95
00:04:53.955 --> 00:04:56.795
of the great recession that a lot of government officials,

96
00:04:57.835 --> 00:04:59.955
statistics, statistical agencies,

97
00:05:00.795 --> 00:05:03.725
what they realized is we are sitting on this mountain

98
00:05:04.385 --> 00:05:08.845
of really good quality, earnings, histories

99
00:05:08.945 --> 00:05:12.165
of individuals, that's the Social Security Administration

100
00:05:12.465 --> 00:05:15.005
or IRS tax agencies and so on.

101
00:05:15.705 --> 00:05:18.845
And pretty much every developed country has these records,

102
00:05:19.465 --> 00:05:22.965
but they were not being, made available to researchers.

103
00:05:23.745 --> 00:05:26.565
So after the great recession when, you know, it was

104
00:05:26.585 --> 00:05:27.885
so painful and so long,

105
00:05:28.465 --> 00:05:30.965
and there was a pushback against economists about, you know,

106
00:05:31.355 --> 00:05:33.565
kind of, why didn't you tell us, you know,

107
00:05:33.565 --> 00:05:36.605
this was gonna happen or tell us what to do now?

108
00:05:37.025 --> 00:05:40.005
And the economists went back and said, well, we don't know.

109
00:05:40.005 --> 00:05:42.805
Like there are so many questions whose answers we didn't

110
00:05:42.805 --> 00:05:46.165
know because we simply couldn't access the data

111
00:05:46.165 --> 00:05:47.645
that we needed to do research.

112
00:05:48.265 --> 00:05:51.445
So that led to an effort throughout the world actually,

113
00:05:51.535 --> 00:05:53.605
which made these data more available.

114
00:05:54.545 --> 00:05:56.845
So when you have, to give you an

115
00:05:56.845 --> 00:05:58.845
Example, we went from,

116
00:05:58.915 --> 00:06:03.565
data sets on earnings from surveys of a thousand people

117
00:06:03.665 --> 00:06:08.485
or 2000 people to data sets on entire populations,

118
00:06:08.555 --> 00:06:11.365
like 150 million people in the labor force in the US

119
00:06:11.795 --> 00:06:12.805
over 40 years.

120
00:06:13.385 --> 00:06:16.085
So this was not an order of magnitude.

121
00:06:16.155 --> 00:06:18.045
This was like, you know, several orders

122
00:06:18.105 --> 00:06:19.565
of magnitude jump instantly.

123
00:06:20.395 --> 00:06:23.885
What that allowed was, in a way it is like,

124
00:06:23.915 --> 00:06:26.245
suppose you are really nearsighted, right?

125
00:06:26.625 --> 00:06:30.405
So the word looks like, a lot of,

126
00:06:31.065 --> 00:06:34.245
you see the word behind this really dirty glass,

127
00:06:34.625 --> 00:06:36.565
and somebody comes and just clears your glasses

128
00:06:36.745 --> 00:06:39.245
or they give you kind of, that was my,

129
00:06:39.815 --> 00:06:42.965
experience about 10 years ago when I realized that I,

130
00:06:43.045 --> 00:06:44.445
I couldn't read as well.

131
00:06:44.465 --> 00:06:46.565
And you put them on and you are like, my God, like,

132
00:06:46.565 --> 00:06:47.925
the world is so much sharper.

133
00:06:48.265 --> 00:06:51.605
And what you thought was like a horse actually turns out

134
00:06:51.605 --> 00:06:55.045
to be, you know, like, I don't know, like,

135
00:06:55.425 --> 00:06:57.285
the face of someone or vice versa.

136
00:06:58.065 --> 00:07:01.085
So we got this much clearer idea about

137
00:07:01.085 --> 00:07:02.725
how income risks looked like.

138
00:07:03.345 --> 00:07:05.125
And one outgrowth of that was

139
00:07:05.635 --> 00:07:07.485
when we couldn't see things very clearly.

140
00:07:07.745 --> 00:07:10.125
The data was small, there was measurement error,

141
00:07:10.265 --> 00:07:11.765
and a lot of other issues.

142
00:07:12.785 --> 00:07:17.605
We had to substitute data with our assumptions.

143
00:07:17.665 --> 00:07:21.205
We had to impose assumptions on top of the data to be able

144
00:07:21.205 --> 00:07:22.565
to learn anything from it.

145
00:07:22.865 --> 00:07:26.245
And a very common assumption, almost universal assumption

146
00:07:26.275 --> 00:07:29.125
that I made and everybody else in the profession made 20

147
00:07:29.405 --> 00:07:34.045
years ago, was that the, the income shocks were Gaussian,

148
00:07:34.155 --> 00:07:35.765
they were bell shaped basically.

149
00:07:36.785 --> 00:07:41.445
And there were a, a few kind of,

150
00:07:42.025 --> 00:07:44.005
clues that they may,

151
00:07:44.005 --> 00:07:45.325
that may not be the best assumption,

152
00:07:46.065 --> 00:07:48.845
but we did not have the quality of data to,

153
00:07:48.985 --> 00:07:50.085
to, to push that line.

154
00:07:50.665 --> 00:07:54.335
So with these big data sets, I said, let's go crazy.

155
00:07:54.595 --> 00:07:57.655
You know, like, why do we, let's remove all the assumptions

156
00:07:57.795 --> 00:08:00.255
and do what we call non-parametric work.

157
00:08:00.835 --> 00:08:03.655
And the non-parametric work, which is a paper published,

158
00:08:03.825 --> 00:08:08.135
three years ago in Econometrica with, who is also part

159
00:08:08.135 --> 00:08:10.215
of this project and, and some other economist,

160
00:08:11.035 --> 00:08:15.895
we found like, some, some features of income shocks

161
00:08:15.965 --> 00:08:18.335
that looked nothing like Gaussian.

162
00:08:18.765 --> 00:08:21.775
Like you couldn't be as far from Gaussian almost as you,

163
00:08:21.955 --> 00:08:25.455
you know, you want, we had like huge peaks in the middle.

164
00:08:25.675 --> 00:08:30.375
We had very long tails, meaning the probability of income,

165
00:08:30.675 --> 00:08:32.415
going up by tenfold

166
00:08:33.475 --> 00:08:36.215
was something like 80 times larger in the

167
00:08:36.215 --> 00:08:37.655
data than what you get.

168
00:08:37.715 --> 00:08:38.855
Uh, under Gaussian.

169
00:08:39.315 --> 00:08:43.415
The probability of your income going down by 90% was, again,

170
00:08:43.415 --> 00:08:47.895
like 150 times higher than what a gian assumption implied.

171
00:08:48.515 --> 00:08:52.055
And the asymmetric came part of that, that one thing we saw

172
00:08:52.755 --> 00:08:56.245
was that, you know, income shocks were the, the,

173
00:08:56.245 --> 00:08:57.605
the common assumption was

174
00:08:57.605 --> 00:09:00.965
what we call like an auto aggressive process, meaning

175
00:09:00.995 --> 00:09:03.245
that income shocks mean revert.

176
00:09:03.785 --> 00:09:07.845
Um, so if your income goes up today, it tends to revert back

177
00:09:07.905 --> 00:09:09.205
to, you know, the mean,

178
00:09:09.865 --> 00:09:13.445
but this was symmetric, for positive shocks

179
00:09:13.485 --> 00:09:14.525
and negative shocks.

180
00:09:15.105 --> 00:09:16.685
And what we found out was that

181
00:09:16.875 --> 00:09:18.565
that was not really a good description,

182
00:09:19.215 --> 00:09:22.965
especially when you look at like workers at the low end,

183
00:09:22.965 --> 00:09:27.005
with low incomes, when they got a negative shock, you know,

184
00:09:27.005 --> 00:09:30.125
those shocks were actually quite, transitory,

185
00:09:30.125 --> 00:09:33.165
meaning they were not lasting very long.

186
00:09:33.385 --> 00:09:36.725
And, whereas if they got positive shocks,

187
00:09:36.725 --> 00:09:38.365
they were really pers persistent,

188
00:09:39.225 --> 00:09:40.965
and the opposite was true at the top.

189
00:09:41.345 --> 00:09:44.325
So we found a number of differences like this.

190
00:09:44.385 --> 00:09:47.935
And when you think about the real life positive

191
00:09:47.935 --> 00:09:50.695
and negative shocks, we call both of them shocks,

192
00:09:50.835 --> 00:09:54.255
but you know, a positive shocks usually is associated

193
00:09:54.255 --> 00:09:55.935
with a good outcome, right?

194
00:09:55.955 --> 00:09:59.455
You get a promotion, you know, you find a much better job,

195
00:09:59.915 --> 00:10:02.815
you finish college and you find your first job.

196
00:10:03.475 --> 00:10:06.215
Um, whereas negative shocks are usually associated

197
00:10:06.215 --> 00:10:07.375
with negative outcomes.

198
00:10:08.075 --> 00:10:09.335
You know, you lose your job,

199
00:10:09.685 --> 00:10:11.575
your industry gets a very big shock,

200
00:10:11.995 --> 00:10:13.575
you get sick, you cannot work.

201
00:10:14.315 --> 00:10:15.655
So fundamentally,

202
00:10:15.865 --> 00:10:18.615
these are different animals, so to speak, right?

203
00:10:19.075 --> 00:10:21.655
And we want to kind of treat them as such.

204
00:10:22.355 --> 00:10:25.055
So what we did in this paper, we took,

205
00:10:26.475 --> 00:10:30.655
we took a fairly, general, a very rich,

206
00:10:31.275 --> 00:10:33.735
what we call a lifecycle model,

207
00:10:33.835 --> 00:10:36.575
in which basically we model individuals

208
00:10:36.795 --> 00:10:38.495
as realistically as possible.

209
00:10:38.565 --> 00:10:41.095
There's another class of models that, you know, if

210
00:10:41.615 --> 00:10:43.895
somebody's not doing this kind of work when they first hear,

211
00:10:43.895 --> 00:10:45.055
they're surprised about it.

212
00:10:45.435 --> 00:10:47.175
Uh, they are infinite horizon models.

213
00:10:47.315 --> 00:10:49.215
So it assumes that people live forever.

214
00:10:49.695 --> 00:10:51.255
I wish that was true. Uh,

215
00:10:52.595 --> 00:10:54.815
but in reality, we live a certain number of years,

216
00:10:54.845 --> 00:10:58.215
then we retire, and then sadly, at some point we die

217
00:10:58.795 --> 00:11:00.775
and we leave some bequest to our children,

218
00:11:01.035 --> 00:11:05.055
and children get some bequest we save for retirement.

219
00:11:05.595 --> 00:11:07.335
Um, we have borrowing constraints.

220
00:11:07.475 --> 00:11:10.895
We cannot borrow, you know, too smooth a a,

221
00:11:11.015 --> 00:11:12.895
a big income fall all the time, right?

222
00:11:12.895 --> 00:11:15.215
There's a limit to that. But at the same time,

223
00:11:15.215 --> 00:11:17.055
there's government insurance, right?

224
00:11:17.115 --> 00:11:18.535
You know, the, the, the society,

225
00:11:18.595 --> 00:11:21.215
the social safety net does not allow people

226
00:11:21.215 --> 00:11:24.215
to fall too far down when they get an income risk, right?

227
00:11:24.595 --> 00:11:27.375
You can think about unemployment, insurance, welfare system,

228
00:11:27.835 --> 00:11:30.775
but you can also think about like, you know, charities and,

229
00:11:30.795 --> 00:11:32.455
and churches and families

230
00:11:32.515 --> 00:11:34.815
and many other informal sources of,

231
00:11:34.835 --> 00:11:36.335
of this, insurance.

232
00:11:37.035 --> 00:11:38.615
So we allow for all of that

233
00:11:38.615 --> 00:11:43.015
because the reason this matters is if you don't allow for

234
00:11:43.675 --> 00:11:47.135
the way people can mitigate these shocks, right?

235
00:11:47.235 --> 00:11:50.495
So if my income goes down by 50% this year,

236
00:11:52.185 --> 00:11:56.405
if I am, I am forced to reduce my spending by 50%,

237
00:11:56.995 --> 00:11:59.565
that will hurt a lot, right?

238
00:11:59.675 --> 00:12:02.445
Because for most people, a lot

239
00:12:02.445 --> 00:12:04.045
of our incomes are spoken for.

240
00:12:04.945 --> 00:12:07.045
We have what are called like commitment goods, right?

241
00:12:07.125 --> 00:12:09.365
I already have a mortgage, I already have a car payment,

242
00:12:09.605 --> 00:12:12.085
I already have, you know, health insurance and so on.

243
00:12:12.625 --> 00:12:16.845
So I, what do I do? Do I sell my house? Do I sell my car?

244
00:12:17.505 --> 00:12:20.285
But if I can smooth that by borrowing

245
00:12:20.385 --> 00:12:21.805
or if the government helps me,

246
00:12:21.985 --> 00:12:23.285
that's a complete different game.

247
00:12:23.745 --> 00:12:26.605
So to measure. So that's why this question is

248
00:12:27.385 --> 00:12:32.085
far more harder to answer then what Bob Lucas asked.

249
00:12:32.085 --> 00:12:34.685
Now, in 1987, of course we are, you know,

250
00:12:34.685 --> 00:12:36.925
like 30 plus years later, we have better tools,

251
00:12:36.925 --> 00:12:38.445
better computers and so on.

252
00:12:39.755 --> 00:12:42.695
So we put all of these to get an accurate picture.

253
00:12:43.575 --> 00:12:47.555
And one thing we did was to ask the same questions.

254
00:12:47.705 --> 00:12:50.875
Suppose I can eliminate all of these fluctuations,

255
00:12:51.615 --> 00:12:55.275
and your income just grows at the average level

256
00:12:55.375 --> 00:12:58.555
for the entire population, which follows this nice,

257
00:12:58.775 --> 00:13:00.595
you know, hump shaped kind of curve.

258
00:13:00.865 --> 00:13:03.355
Your income grows up to age 45, 50

259
00:13:03.735 --> 00:13:06.555
and then slightly declines, you know, before retirement.

260
00:13:07.835 --> 00:13:10.295
And it turned out like the number I mentioned that you know,

261
00:13:10.435 --> 00:13:11.775
you, with a,

262
00:13:11.775 --> 00:13:13.975
with a very moderate risk aversion, by the way.

263
00:13:14.315 --> 00:13:16.415
So there's this concept of risk aversion, right?

264
00:13:16.415 --> 00:13:18.375
That that determines how much people

265
00:13:18.645 --> 00:13:19.855
dislike, you know, risk.

266
00:13:20.795 --> 00:13:23.335
And you can get inflated numbers if you assume

267
00:13:23.335 --> 00:13:25.095
that people are super risk averse.

268
00:13:25.115 --> 00:13:29.615
And that's kinda an easy way to get a sensational number.

269
00:13:30.275 --> 00:13:32.615
But at the same time, if you assume that people are

270
00:13:32.615 --> 00:13:35.295
so risk averse, then people should not be

271
00:13:35.535 --> 00:13:36.655
crossing the street, okay?

272
00:13:36.915 --> 00:13:38.255
So we know their behavior.

273
00:13:38.595 --> 00:13:40.615
And so we took a number pretty on,

274
00:13:40.615 --> 00:13:42.855
a pretty conservative lower end,

275
00:13:43.115 --> 00:13:45.935
and we still found that people would be willing to,

276
00:13:46.155 --> 00:13:49.535
to give up 40% of their income.

277
00:13:50.405 --> 00:13:52.265
And the second question we asked was,

278
00:13:52.615 --> 00:13:56.225
what if we did not have this new evidence on income risk?

279
00:13:56.645 --> 00:13:58.305
That's why it's in the title, right?

280
00:13:58.305 --> 00:14:00.905
The title says, you know, non GA earnings risk

281
00:14:00.905 --> 00:14:03.585
because until a few years ago,

282
00:14:04.615 --> 00:14:07.225
most work empirical work in this area

283
00:14:07.255 --> 00:14:09.945
that modeled income risk assumed Gaussian shocks,

284
00:14:10.445 --> 00:14:14.535
and we found that you would be understating even.

285
00:14:14.645 --> 00:14:17.735
It's, it's kind of an apples to apples comparison

286
00:14:17.735 --> 00:14:20.415
because we are taking Gaussian shocks now,

287
00:14:20.915 --> 00:14:24.175
but we are matching the same target in the data.

288
00:14:24.395 --> 00:14:27.815
So they are in a way inferred from the same data.

289
00:14:28.635 --> 00:14:32.015
But in one case, we, use the older approach

290
00:14:32.075 --> 00:14:35.295
of imposing gian distribution in the new one.

291
00:14:35.395 --> 00:14:37.975
We don't, and we found that you'll be understating,

292
00:14:38.085 --> 00:14:39.655
like the, the income risk

293
00:14:39.655 --> 00:14:43.655
that you get is about 40% can be 40% or less.

294
00:14:44.635 --> 00:14:46.495
So it's 40% or 40%.

295
00:14:46.595 --> 00:14:51.295
So basically, lemme put it the other way, accounting

296
00:14:51.345 --> 00:14:56.205
for non Gaussian income risk, gives you an estimate

297
00:14:56.305 --> 00:14:59.685
of the cost that is about two to four times higher

298
00:15:00.315 --> 00:15:01.805
than under the Gaussian risk.

299
00:15:02.005 --> 00:15:04.965
I don't know if like the listeners remember,

300
00:15:06.775 --> 00:15:10.105
this was before even like when the great recession happened.

301
00:15:10.605 --> 00:15:14.945
Uh, it started with the failures of a few banks, right?

302
00:15:15.325 --> 00:15:17.265
The first one I think was in the spring,

303
00:15:17.445 --> 00:15:20.025
if I remember correctly, of 2007,

304
00:15:20.285 --> 00:15:22.865
it was bear terms, or 2008.

305
00:15:23.845 --> 00:15:27.145
And then a while later came Lehman Brothers, right?

306
00:15:27.145 --> 00:15:31.345
That also went down. And, but there's a story before that.

307
00:15:31.345 --> 00:15:32.345
In 1997,

308
00:15:32.435 --> 00:15:35.585
there was another hedge fund called the Long-term Capital

309
00:15:35.585 --> 00:15:39.385
Management, LTCM, that went bankrupt.

310
00:15:40.265 --> 00:15:43.525
And, the interesting thing about that hedge fund was

311
00:15:43.525 --> 00:15:44.885
that it was founded, one

312
00:15:44.885 --> 00:15:49.405
of the founders was a Nobel Prize winning economist from the

313
00:15:49.405 --> 00:15:52.565
auction pricing formula, the, the, the,

314
00:15:52.565 --> 00:15:53.565
black Shaw formula.

315
00:15:55.775 --> 00:15:57.995
And so people were like, wow, you know,

316
00:15:57.995 --> 00:15:59.955
like the smartest economist,

317
00:15:59.955 --> 00:16:01.515
have this hedge fund and it goes bankrupt.

318
00:16:02.215 --> 00:16:06.245
And, in 1997, the answer they gave was,

319
00:16:06.755 --> 00:16:08.245
they said, what happened?

320
00:16:08.825 --> 00:16:11.245
Um, and the story is long, I won't go into details,

321
00:16:11.665 --> 00:16:14.005
but there was like, I think Russia in,

322
00:16:14.385 --> 00:16:16.885
had put like a moratorium on their debt payments.

323
00:16:17.545 --> 00:16:20.045
And so it was something really remote kind of

324
00:16:20.045 --> 00:16:21.565
that happened in another part of the world.

325
00:16:21.785 --> 00:16:24.805
And I might be mixing this, whether it was the Thailand,

326
00:16:25.025 --> 00:16:27.725
you know, Christ in 97, it's one of the two,

327
00:16:28.385 --> 00:16:30.645
but the response that was given was,

328
00:16:31.395 --> 00:16:34.805
what happened was a 40 standard deviation event.

329
00:16:35.505 --> 00:16:36.565
So what does that mean?

330
00:16:36.985 --> 00:16:39.925
So a Gaussian distribution is something that is very,

331
00:16:40.035 --> 00:16:41.765
very common in nature.

332
00:16:41.945 --> 00:16:44.565
It is the most common distribution that we see,

333
00:16:44.905 --> 00:16:47.685
and everybody knows it as like the bell curve, right?

334
00:16:47.685 --> 00:16:49.725
That's a smooth, like bell shaped.

335
00:16:50.765 --> 00:16:52.975
That distribution has a measure a,

336
00:16:53.135 --> 00:16:56.175
a feature called standard deviation, which is

337
00:16:56.275 --> 00:16:59.015
how wide the tails of the distribution is.

338
00:16:59.515 --> 00:17:02.575
In other words, it tells you compared to the mean,

339
00:17:03.205 --> 00:17:07.175
what is the likelihood of getting an outcome that is say,

340
00:17:07.435 --> 00:17:09.335
20% higher than the mean?

341
00:17:09.685 --> 00:17:12.215
Suppose that the distribution standard deviation

342
00:17:12.435 --> 00:17:15.975
or sigma in this language is 20%, okay,

343
00:17:16.165 --> 00:17:19.135
hundred G in distribution events

344
00:17:19.205 --> 00:17:23.135
that are plus one standard deviation, that is 20%

345
00:17:23.265 --> 00:17:25.935
above the mean or 20% below the mean.

346
00:17:26.195 --> 00:17:29.855
And in between they have about a two thirds probability.

347
00:17:30.275 --> 00:17:35.015
So about 65% of the time, you'll not observe an outcome

348
00:17:35.015 --> 00:17:39.375
that is 20% above the mean or 20% below the,

349
00:17:39.715 --> 00:17:41.295
or larger than that, okay?

350
00:17:42.235 --> 00:17:44.575
You can do two sigma, two standard deviation.

351
00:17:45.155 --> 00:17:47.335
Ask the same question. What is the likelihood

352
00:17:47.335 --> 00:17:49.810
that you'll observe an outcome 40%

353
00:17:49.945 --> 00:17:54.045
or higher than the mean if the standard deviation is 20%?

354
00:17:54.235 --> 00:17:56.165
That probability is something like 2%.

355
00:17:57.185 --> 00:18:00.525
Now, if you go to 40 standard deviations, okay, that's

356
00:18:00.525 --> 00:18:02.605
what they meant, they said, so 40 Sigma event,

357
00:18:02.995 --> 00:18:06.765
what they meant was this was an event that had almost like

358
00:18:07.555 --> 00:18:11.445
zero point, you know, put 20 zeros and then a one, okay?

359
00:18:11.445 --> 00:18:12.965
Even that is an overstatement, actually.

360
00:18:13.265 --> 00:18:14.685
So they said this was something

361
00:18:14.685 --> 00:18:16.245
that should have never happened.

362
00:18:17.165 --> 00:18:20.585
Now, when the great recession came, and you know, a IG

363
00:18:20.585 --> 00:18:23.225
and others, basically these, these insurance companies

364
00:18:23.775 --> 00:18:27.665
were struggling and they were selling before the crisis.

365
00:18:27.835 --> 00:18:31.785
These securities, they, the, the, the critique was

366
00:18:31.785 --> 00:18:34.225
that they were basing some of their pricing

367
00:18:34.405 --> 00:18:37.145
or most of their pricing on the Gaussian

368
00:18:37.145 --> 00:18:38.385
assumption again, okay?

369
00:18:38.805 --> 00:18:40.825
So what that means is that they had these numbers

370
00:18:41.095 --> 00:18:45.105
that I have in mind that, plus two standard deviation

371
00:18:45.285 --> 00:18:48.945
or larger event has only a two, 3% probability,

372
00:18:49.395 --> 00:18:51.985
three standard deviation has this much probability.

373
00:18:53.555 --> 00:18:54.935
The problem is the world,

374
00:18:55.885 --> 00:18:58.775
some key variables in the world are not Gaussian.

375
00:18:59.155 --> 00:19:02.775
And we knew that in some financial sector variables,

376
00:19:02.805 --> 00:19:04.615
usually like in, you know,

377
00:19:04.615 --> 00:19:08.215
complicated derivative securities, auction price and so on.

378
00:19:09.075 --> 00:19:12.815
But what we did in our work is like what we kind of

379
00:19:13.415 --> 00:19:14.615
discovered in a way, it,

380
00:19:14.755 --> 00:19:15.935
by the way, we are not the first one.

381
00:19:15.935 --> 00:19:18.655
There's actually a very, very nice paper in two year 2000 by

382
00:19:19.325 --> 00:19:21.775
Akia and Keen who emphasized this,

383
00:19:21.915 --> 00:19:24.815
but they, you know, it wasn't followed up, too much.

384
00:19:25.355 --> 00:19:27.975
So let's move now to non Gaussian. What does that mean?

385
00:19:28.085 --> 00:19:31.255
Okay, I gave you a number at the beginning.

386
00:19:31.415 --> 00:19:33.215
I said that the standard deviation

387
00:19:33.355 --> 00:19:36.135
of individual income growth is about 50%.

388
00:19:36.995 --> 00:19:38.455
Now, under Gaussian,

389
00:19:38.635 --> 00:19:41.255
the standard deviation has a second interpretation,

390
00:19:41.345 --> 00:19:45.955
which is, it is the size of the average shock, okay?

391
00:19:46.285 --> 00:19:50.115
Which means in a typical year, a typical person can expect

392
00:19:50.635 --> 00:19:53.035
a plus one or minus one standard deviation shock.

393
00:19:53.735 --> 00:19:56.235
But I don't know you, but if I, I look at myself

394
00:19:56.255 --> 00:19:59.075
and you know, my sister, my, you know, my friends,

395
00:19:59.355 --> 00:20:02.555
I don't see that many people whose income fluctuate

396
00:20:02.555 --> 00:20:03.915
by 50% every year.

397
00:20:04.805 --> 00:20:06.895
That is because the data is not Gaussian.

398
00:20:06.955 --> 00:20:08.135
So what does it look like?

399
00:20:08.595 --> 00:20:12.255
It looks like this almost like a cone, okay?

400
00:20:12.725 --> 00:20:16.255
That is, basically very,

401
00:20:16.525 --> 00:20:18.455
very concentrated in the middle,

402
00:20:18.715 --> 00:20:20.055
the probability distribution.

403
00:20:20.475 --> 00:20:23.535
By that we mean most of the time most people

404
00:20:24.245 --> 00:20:26.895
only experience like a five, 10,

405
00:20:26.895 --> 00:20:29.375
15% income change year to year.

406
00:20:30.555 --> 00:20:33.095
But again, compared to the Gaussian, what is different?

407
00:20:33.115 --> 00:20:35.495
So why do we get that use standard deviation, right?

408
00:20:36.005 --> 00:20:38.895
It's because the standard deviation takes into account the

409
00:20:38.895 --> 00:20:39.895
entire distribution.

410
00:20:40.355 --> 00:20:42.375
And that is what I said a few minutes ago,

411
00:20:42.375 --> 00:20:45.335
that the tail events, okay,

412
00:20:45.925 --> 00:20:48.525
that LTCM and Bear Stearns

413
00:20:48.525 --> 00:20:49.725
and everybody else said,

414
00:20:49.725 --> 00:20:51.485
this is a 40 standard deviation event.

415
00:20:52.385 --> 00:20:54.085
No, like it wasn't,

416
00:20:54.225 --> 00:20:57.565
it was actually those tail events are much more likely,

417
00:20:58.025 --> 00:20:59.965
as I mentioned before, the vol,

418
00:20:59.965 --> 00:21:03.125
that your income will go down by 90, 95%

419
00:21:03.665 --> 00:21:08.525
is about 150 times higher in the actual data than

420
00:21:08.525 --> 00:21:10.485
what the GA in distribution tells you.

421
00:21:10.785 --> 00:21:12.445
So if you work with that distribution,

422
00:21:12.905 --> 00:21:16.685
you are vastly over underestimating the risk

423
00:21:16.685 --> 00:21:18.965
that's out there, the tail risk that's out there.

424
00:21:19.505 --> 00:21:21.165
And, you know, we are human beings.

425
00:21:21.165 --> 00:21:24.725
It, it doesn't take too much convincing to, realize

426
00:21:24.795 --> 00:21:27.445
that that's exactly the risk that we hate.

427
00:21:27.875 --> 00:21:32.085
Because if I, my income goes down by 10%, I can manage,

428
00:21:32.685 --> 00:21:34.565
I can reduce my consumption by 10%,

429
00:21:34.925 --> 00:21:36.325
I can borrow temporarily.

430
00:21:36.985 --> 00:21:40.365
But if it goes down by 90%, 95%,

431
00:21:40.545 --> 00:21:45.005
and if that shock is persistent, then it's very hard for

432
00:21:45.585 --> 00:21:49.325
myself to insure myself, for the society to insure me to,

433
00:21:49.345 --> 00:21:51.685
for my family to insure me and so on.

434
00:21:51.785 --> 00:21:54.965
And that's why that creates a big,

435
00:21:55.625 --> 00:21:57.405
misrepresentation of the nature of risk.

436
00:21:57.565 --> 00:22:01.525
I want. Big policy implication is somewhat

437
00:22:01.565 --> 00:22:02.605
of a technical one,

438
00:22:03.355 --> 00:22:06.735
and it's the following that the Gaussian assumption

439
00:22:07.555 --> 00:22:10.605
is really the, the bread

440
00:22:10.605 --> 00:22:14.325
and butter, it's like the salt in every model

441
00:22:15.305 --> 00:22:19.425
of individual behavior that is used

442
00:22:19.565 --> 00:22:23.985
to develop policies to judge the consequence

443
00:22:24.005 --> 00:22:28.945
of policies from retirement policies to unemployment,

444
00:22:28.945 --> 00:22:31.345
insurance, social welfare policies,

445
00:22:31.525 --> 00:22:34.105
to really anything that I can think of.

446
00:22:35.085 --> 00:22:37.465
And if we have the wrong model,

447
00:22:37.705 --> 00:22:39.345
I know this sounds technical, right?

448
00:22:39.805 --> 00:22:42.185
But you know, you know,

449
00:22:43.085 --> 00:22:46.305
we may just be happy driving our car without worrying about,

450
00:22:46.445 --> 00:22:48.905
you know, what, what's happening in the back, right?

451
00:22:49.205 --> 00:22:50.665
How does the, the gas

452
00:22:50.665 --> 00:22:53.145
that I put in there cause some combustion

453
00:22:53.145 --> 00:22:55.225
and that turns into a heat and energy?

454
00:22:55.765 --> 00:22:57.505
We may not worry about that,

455
00:22:57.805 --> 00:23:01.905
but the truth is, right, if the gasoline is, is you know,

456
00:23:02.045 --> 00:23:05.105
not the right type for your car, your car won't go anywhere.

457
00:23:05.725 --> 00:23:10.045
So similarly, I think I would say,

458
00:23:10.505 --> 00:23:15.165
we really need to up our game as, as, as, as far as the

459
00:23:15.805 --> 00:23:20.285
modeling framework that goes to evaluate the outcomes,

460
00:23:20.385 --> 00:23:22.005
the consequence of policies.

461
00:23:22.865 --> 00:23:25.405
And, this is not the only paper.

462
00:23:25.745 --> 00:23:27.925
Uh, I, like I said, I've been working in this area

463
00:23:27.985 --> 00:23:32.005
for 15 years, and so some of the results here first,

464
00:23:32.185 --> 00:23:35.965
or the results were in my papers more than 10 years ago.

465
00:23:36.425 --> 00:23:40.605
So people have looked, for example, at something very,

466
00:23:41.265 --> 00:23:43.245
very technical.

467
00:23:43.355 --> 00:23:46.525
It's called like the meridian taxation literature, okay?

468
00:23:46.985 --> 00:23:50.365
Uh, very high end, cutting edge theoretical work.

469
00:23:50.825 --> 00:23:52.645
And in that framework, they asked

470
00:23:53.075 --> 00:23:57.355
what should be the optimal tax for the top 1%, okay?

471
00:23:57.775 --> 00:24:00.435
And what they found is that if you take Gaussian income

472
00:24:00.575 --> 00:24:03.875
shocks, the number is around like 20%.

473
00:24:04.575 --> 00:24:06.635
If you actually take this income process,

474
00:24:07.105 --> 00:24:08.315
it's more than 40%.

475
00:24:09.115 --> 00:24:11.495
So these are two very different pictures, right?

476
00:24:11.915 --> 00:24:15.175
So others have looked at the implications for asset pricing,

477
00:24:15.475 --> 00:24:17.695
you know, others have looked at for retirement saving.

478
00:24:18.355 --> 00:24:22.575
Um, so I would say it may not sound like a very,

479
00:24:22.795 --> 00:24:26.135
catchy kind of policy implication in addition to the fact

480
00:24:26.135 --> 00:24:29.175
that yes, the shocks can be very big,

481
00:24:29.715 --> 00:24:33.215
but at the same time, we do incorporate,

482
00:24:33.635 --> 00:24:36.785
that's actually an important point in our modeling.

483
00:24:36.845 --> 00:24:39.305
We do something that is not always done,

484
00:24:39.405 --> 00:24:43.145
but should always be done, which is that you,

485
00:24:43.365 --> 00:24:46.985
we are allowing for what we call an income floor by

486
00:24:46.985 --> 00:24:48.425
that time mean we assume that

487
00:24:48.935 --> 00:24:50.865
when you have very long tailed shocks,

488
00:24:51.285 --> 00:24:54.225
people literally can get a shock that reduce their income

489
00:24:54.225 --> 00:24:56.145
to zero, and they can stay there for several years

490
00:24:57.635 --> 00:24:58.695
in some modeling.

491
00:24:59.235 --> 00:25:02.455
We just let that be, and we don't let them have zero income,

492
00:25:02.475 --> 00:25:04.375
but we put like a very low income for that.

493
00:25:04.995 --> 00:25:07.335
But in reality, so what we did in our paper,

494
00:25:07.675 --> 00:25:10.855
we took the most recent estimates from a paper in

495
00:25:10.855 --> 00:25:14.375
Econometrica two years ago, about the total value

496
00:25:14.435 --> 00:25:17.895
of welfare, benefits, like, you know,

497
00:25:18.295 --> 00:25:19.375
a rental support

498
00:25:19.635 --> 00:25:21.615
and, you know, food stamps,

499
00:25:21.935 --> 00:25:23.255
unemployment insurance and so on.

500
00:25:23.675 --> 00:25:26.695
And then we, we, we put a minimum income floor.

501
00:25:26.715 --> 00:25:29.455
We said that even if the model predicts

502
00:25:29.455 --> 00:25:31.015
that your income goes down to zero,

503
00:25:31.605 --> 00:25:33.935
your consumption will not be close to zero.

504
00:25:33.935 --> 00:25:36.175
Because in reality, you know, the government,

505
00:25:36.315 --> 00:25:39.055
the society is going to put something that turn.

506
00:25:39.155 --> 00:25:42.095
If we hadn't done that, the 40% number I mentioned

507
00:25:42.095 --> 00:25:44.735
to you would've been like a, you know, like something

508
00:25:44.735 --> 00:25:46.255
that is, not even credible.

509
00:25:46.395 --> 00:25:48.735
It would be like a 90% of your income you would be willing

510
00:25:48.735 --> 00:25:49.735
to give up because

511
00:25:49.735 --> 00:25:51.815
of the technicality in the way our models work.

512
00:25:52.395 --> 00:25:56.655
So I would say for policy, income shops are very big,

513
00:25:56.715 --> 00:25:58.695
but that minimum threshold we are putting

514
00:25:59.805 --> 00:26:02.015
reduces the welfare costs substantially.

The Cost of Business Cycles: Bob Lucas's Hypothetical Inquiry

00:00:00 - Need to go back a little bit to 1987 when the very famous, macro economist, Bob Lucas, asked a question, about what we now call the cost of business cycles, right? We talked so much about recessions financial crisis, and, he had done 20 years of work on that, but then he got convinced that economic growth, studying growth as opposed to fluctuations could be more important. So he asked this hypothetical question. He said, suppose that this, hypothetical,

00:00:37 - and he said that I can eliminate all recessions. You know, the economy will grow at a constant 2% rate along the trend line. how much would you be willing to pay me to the, to the US public? And this is a simple calculation as it turns out, you know, when, when he calculated under different assumptions, he found the number to be like literally ridiculously small. It was like 0.1% of consumption per year.

00:01:08 - as if like people didn't care about this fluctuations. Now, when we go through them, we find them to be, you know, really big, deal. But when you look in the context of the entire history, they were really overwhelmed by the trend growth rate. So a, a large literature emerged after that trying to overturn this because he himself included, wrote papers on business cycles for 20 years. And people said, no, no, no, you just, you know,

Reassessing the Cost of Economic Fluctuations

00:01:39 - I will edit to my model, it'll be a bigger number. That fact, it'll be a bigger number. So to make a long story short, over 15 years after his paper, maybe hundreds of papers were written on this, and, Bob Lucas gave this presidential address to the American Economic Association in 2003. And he said, well, you know, we have all this work that challenged my paper. And he kind of wished that some of them could be true, because, you know, he invested so much in that.

00:02:11 - Based on all the new research, the number was not 0.1% of consumption per year. So if you think about, you know, like, income per capita being $60,000, that is something like, you know, $600 a year, right? So it's something so small. The updated number was 0.01%, it was even smaller. So again, that caused an uproar, but there's so much work on that that

00:02:42 - that is much more closer to us, right? You know, if you look at overall wage fluctuations that a worker experiences during their lifetime, these aggregate fluctuations, the reason that calculation turned out to be so small is because the volatility, which we measure by, say, the standard deviation of GDP growth, okay, is something like 1% per year, or at max it is 2% per year.

Understanding Income Volatility and Risk

00:03:14 - But if you look at the individuals like you and me, and you know, many others out there, if you calculate the same statistic of volatility or individual income, in the US in many countries actually that we have data on, it is close to 50%, not one or 2%. It's not like our income is not fluctuating, but it is fluctuating for many reasons that are not the recessions.

00:03:45 - And we realize we have never asked the same question that Bob asked, Bob Lucas asked for this much bigger risk. So that is motivation for the paper to ask. Again, suppose an alien kind of scientist came from, you know, Mars, maybe we'll meet them if we go there and said, how much are you willing to pay me? And our answer turns out to be substantial, that actually even for a moderate risk conversion,

00:04:17 - of their annual income to eliminate these fluctuations. This paper builds, it's like a continuation or outgrowth of my, research agenda in the last maybe 15 years. actually I worked on income risk throughout my career, but in the last 15 years, what has changed, has been the availability of big data. And this will surprise no one, right? We hear it everywhere. And in economics, it's, this started about maybe 15

Leveraging Data Insights Post-Great Recession

00:04:50 - And, I view it as a silver lining, so to speak, of the great recession that a lot of government officials, statistics, statistical agencies, what they realized is we are sitting on this mountain of really good quality, earnings, histories of individuals, that's the Social Security Administration or IRS tax agencies and so on. And pretty much every developed country has these records,

00:05:23 - So after the great recession when, you know, it was so painful and so long, and there was a pushback against economists about, you know, kind of, why didn't you tell us, you know, this was gonna happen or tell us what to do now? And the economists went back and said, well, we don't know. Like there are so many questions whose answers we didn't know because we simply couldn't access the data that we needed to do research. So that led to an effort throughout the world actually,

00:05:54 - So when you have, to give you an Example, we went from, data sets on earnings from surveys of a thousand people or 2000 people to data sets on entire populations, like 150 million people in the labor force in the US over 40 years. So this was not an order of magnitude. This was like, you know, several orders of magnitude jump instantly. What that allowed was, in a way it is like,

Clarity Through Data: Rethinking Income Risk Assumptions

00:06:26 - So the word looks like, a lot of, you see the word behind this really dirty glass, and somebody comes and just clears your glasses or they give you kind of, that was my, experience about 10 years ago when I realized that I, I couldn't read as well. And you put them on and you are like, my God, like, the world is so much sharper. And what you thought was like a horse actually turns out to be, you know, like, I don't know, like,

00:06:58 - So we got this much clearer idea about how income risks looked like. And one outgrowth of that was when we couldn't see things very clearly. The data was small, there was measurement error, and a lot of other issues. We had to substitute data with our assumptions. We had to impose assumptions on top of the data to be able to learn anything from it. And a very common assumption, almost universal assumption

00:07:29 - years ago, was that the, the income shocks were Gaussian, they were bell shaped basically. And there were a, a few kind of, clues that they may, that may not be the best assumption, but we did not have the quality of data to, to, to push that line. So with these big data sets, I said, let's go crazy. You know, like, why do we, let's remove all the assumptions

Reassessing Income Risk: Moving Beyond Gaussian Assumptions

00:08:00 - And the non-parametric work, which is a paper published, three years ago in Econometrica with, who is also part of this project and, and some other economist, we found like, some, some features of income shocks that looked nothing like Gaussian. Like you couldn't be as far from Gaussian almost as you, you know, you want, we had like huge peaks in the middle. We had very long tails, meaning the probability of income,

00:08:33 - was something like 80 times larger in the data than what you get. under Gaussian. The probability of your income going down by 90% was, again, like 150 times higher than what a gian assumption implied. And the asymmetric came part of that, that one thing we saw was that, you know, income shocks were the, the, the common assumption was what we call like an auto aggressive process, meaning

00:09:03 - so if your income goes up today, it tends to revert back to, you know, the mean, but this was symmetric, for positive shocks and negative shocks. And what we found out was that that was not really a good description, especially when you look at like workers at the low end, with low incomes, when they got a negative shock, you know, those shocks were actually quite, transitory, meaning they were not lasting very long.

Understanding Economic Shocks: Positive vs. Negative Outcomes

00:09:36 - they were really pers persistent, and the opposite was true at the top. So we found a number of differences like this. And when you think about the real life positive and negative shocks, we call both of them shocks, but you know, a positive shocks usually is associated with a good outcome, right? You get a promotion, you know, you find a much better job, you finish college and you find your first job.

00:10:06 - with negative outcomes. You know, you lose your job, your industry gets a very big shock, you get sick, you cannot work. So fundamentally, these are different animals, so to speak, right? And we want to kind of treat them as such. So what we did in this paper, we took, we took a fairly, general, a very rich, what we call a lifecycle model,

00:10:36 - as realistically as possible. There's another class of models that, you know, if somebody's not doing this kind of work when they first hear, they're surprised about it. they are infinite horizon models. So it assumes that people live forever. I wish that was true. but in reality, we live a certain number of years, then we retire, and then sadly, at some point we die and we leave some bequest to our children, and children get some bequest we save for retirement.

Income Risk and Social Safety Nets

00:11:07 - We cannot borrow, you know, too smooth a a, a big income fall all the time, right? There's a limit to that. But at the same time, there's government insurance, right? You know, the, the, the society, the social safety net does not allow people to fall too far down when they get an income risk, right? You can think about unemployment, insurance, welfare system, but you can also think about like, you know, charities and, and churches and families and many other informal sources of, of this, insurance.

00:11:38 - because the reason this matters is if you don't allow for the way people can mitigate these shocks, right? So if my income goes down by 50% this year, if I am, I am forced to reduce my spending by 50%, that will hurt a lot, right? Because for most people, a lot of our incomes are spoken for. We have what are called like commitment goods, right?

00:12:09 - I already have, you know, health insurance and so on. So I, what do I do? Do I sell my house? Do I sell my car? But if I can smooth that by borrowing or if the government helps me, that's a complete different game. So to measure. So that's why this question is far more harder to answer then what Bob Lucas asked. Now, in 1987, of course we are, you know, like 30 plus years later, we have better tools, better computers and so on.

Exploring Income Stability and Risk Aversion

00:12:43 - And one thing we did was to ask the same questions. Suppose I can eliminate all of these fluctuations, and your income just grows at the average level for the entire population, which follows this nice, you know, hump shaped kind of curve. Your income grows up to age 45, 50 and then slightly declines, you know, before retirement. And it turned out like the number I mentioned that you know,

00:13:11 - with a very moderate risk aversion, by the way. So there's this concept of risk aversion, right? That that determines how much people dislike, you know, risk. And you can get inflated numbers if you assume that people are super risk averse. And that's kinda an easy way to get a sensational number. But at the same time, if you assume that people are so risk averse, then people should not be crossing the street, okay? So we know their behavior. And so we took a number pretty on,

00:13:43 - and we still found that people would be willing to, to give up 40% of their income. And the second question we asked was, what if we did not have this new evidence on income risk? That's why it's in the title, right? The title says, you know, non GA earnings risk because until a few years ago, most work empirical work in this area that modeled income risk assumed Gaussian shocks, and we found that you would be understating even.

Underestimating Income Risk: The Shift from Gaussian to Non-Gaussian Models

00:14:17 - because we are taking Gaussian shocks now, but we are matching the same target in the data. So they are in a way inferred from the same data. But in one case, we, use the older approach of imposing gian distribution in the new one. We don't, and we found that you'll be understating, like the, the income risk that you get is about 40% can be 40% or less. So it's 40% or 40%.

00:14:51 - for non Gaussian income risk, gives you an estimate of the cost that is about two to four times higher than under the Gaussian risk. I don't know if like the listeners remember, this was before even like when the great recession happened. it started with the failures of a few banks, right? The first one I think was in the spring,

00:15:20 - it was bear terms, or 2008. And then a while later came Lehman Brothers, right? That also went down. And, but there's a story before that. In 1997, there was another hedge fund called the Long-term Capital Management, LTCM, that went bankrupt. And, the interesting thing about that hedge fund was that it was founded, one of the founders was a Nobel Prize winning economist from the

Understanding Gaussian Distribution and Financial Crises

00:15:52 - black Shaw formula. And so people were like, wow, you know, like the smartest economist, have this hedge fund and it goes bankrupt. And, in 1997, the answer they gave was, they said, what happened? and the story is long, I won't go into details, but there was like, I think Russia in, had put like a moratorium on their debt payments. And so it was something really remote kind of

00:16:21 - And I might be mixing this, whether it was the Thailand, you know, Christ in 97, it's one of the two, but the response that was given was, what happened was a 40 standard deviation event. So what does that mean? So a Gaussian distribution is something that is very, very common in nature. It is the most common distribution that we see, and everybody knows it as like the bell curve, right? That's a smooth, like bell shaped.

00:16:53 - a feature called standard deviation, which is how wide the tails of the distribution is. In other words, it tells you compared to the mean, what is the likelihood of getting an outcome that is say, 20% higher than the mean? Suppose that the distribution standard deviation or sigma in this language is 20%, okay, hundred G in distribution events that are plus one standard deviation, that is 20%

Non-Gaussian Risks in Economic Predictions

00:17:26 - And in between they have about a two thirds probability. So about 65% of the time, you'll not observe an outcome that is 20% above the mean or 20% below the, or larger than that, okay? You can do two sigma, two standard deviation. Ask the same question. What is the likelihood that you'll observe an outcome 40% or higher than the mean if the standard deviation is 20%?

00:17:57 - Now, if you go to 40 standard deviations, okay, that's what they meant, they said, so 40 Sigma event, what they meant was this was an event that had almost like zero point, you know, put 20 zeros and then a one, okay? Even that is an overstatement, actually. So they said this was something that should have never happened. Now, when the great recession came, and you know, a IG and others, basically these, these insurance companies were struggling and they were selling before the crisis.

00:18:31 - that they were basing some of their pricing or most of their pricing on the Gaussian assumption again, okay? So what that means is that they had these numbers that I have in mind that, plus two standard deviation or larger event has only a two, 3% probability, three standard deviation has this much probability. The problem is the world, some key variables in the world are not Gaussian.

Understanding Non-Gaussian Income Fluctuations

00:19:02 - usually like in, you know, complicated derivative securities, auction price and so on. But what we did in our work is like what we kind of discovered in a way, it, by the way, we are not the first one. There's actually a very, very nice paper in two year 2000 by Akia and Keen who emphasized this, but they, you know, it wasn't followed up, too much. So let's move now to non Gaussian. What does that mean? Okay, I gave you a number at the beginning.

00:19:33 - of individual income growth is about 50%. Now, under Gaussian, the standard deviation has a second interpretation, which is, it is the size of the average shock, okay? Which means in a typical year, a typical person can expect a plus one or minus one standard deviation shock. But I don't know you, but if I, I look at myself and you know, my sister, my, you know, my friends, I don't see that many people whose income fluctuate

00:20:04 - That is because the data is not Gaussian. So what does it look like? It looks like this almost like a cone, okay? That is, basically very, very concentrated in the middle, the probability distribution. By that we mean most of the time most people only experience like a five, 10, 15% income change year to year. But again, compared to the Gaussian, what is different?

Rethinking Income Risk: Beyond Gaussian Assumptions

00:20:36 - It's because the standard deviation takes into account the entire distribution. And that is what I said a few minutes ago, that the tail events, okay, that LTCM and Bear Stearns and everybody else said, this is a 40 standard deviation event. No, like it wasn't, it was actually those tail events are much more likely, as I mentioned before, the vol, that your income will go down by 90, 95%

00:21:08 - what the GA in distribution tells you. So if you work with that distribution, you are vastly over underestimating the risk that's out there, the tail risk that's out there. And, you know, we are human beings. It, it doesn't take too much convincing to, realize that that's exactly the risk that we hate. Because if I, my income goes down by 10%, I can manage, I can reduce my consumption by 10%,

00:21:36 - But if it goes down by 90%, 95%, and if that shock is persistent, then it's very hard for myself to insure myself, for the society to insure me to, for my family to insure me and so on. And that's why that creates a big, misrepresentation of the nature of risk. I want. Big policy implication is somewhat of a technical one, and it's the following that the Gaussian assumption

Rethinking Policy Models: The Importance of Accurate Frameworks

00:22:10 - and butter, it's like the salt in every model of individual behavior that is used to develop policies to judge the consequence of policies from retirement policies to unemployment, insurance, social welfare policies, to really anything that I can think of. And if we have the wrong model,

00:22:39 - But you know, you know, we may just be happy driving our car without worrying about, you know, what, what's happening in the back, right? How does the, the gas that I put in there cause some combustion and that turns into a heat and energy? We may not worry about that, but the truth is, right, if the gasoline is, is you know, not the right type for your car, your car won't go anywhere. So similarly, I think I would say,

00:23:15 - modeling framework that goes to evaluate the outcomes, the consequence of policies. And, this is not the only paper. I, like I said, I've been working in this area for 15 years, and so some of the results here first, or the results were in my papers more than 10 years ago. So people have looked, for example, at something very,

Revisiting Taxation: Gaussian vs Non-Gaussian Income Shocks

00:23:43 - It's called like the meridian taxation literature, okay? very high end, cutting edge theoretical work. And in that framework, they asked what should be the optimal tax for the top 1%, okay? And what they found is that if you take Gaussian income shocks, the number is around like 20%. If you actually take this income process, it's more than 40%. So these are two very different pictures, right?

00:24:15 - you know, others have looked at for retirement saving. so I would say it may not sound like a very, catchy kind of policy implication in addition to the fact that yes, the shocks can be very big, but at the same time, we do incorporate, that's actually an important point in our modeling. We do something that is not always done, but should always be done, which is that you,

00:24:46 - that time mean we assume that when you have very long tailed shocks, people literally can get a shock that reduce their income to zero, and they can stay there for several years in some modeling. We just let that be, and we don't let them have zero income, but we put like a very low income for that. But in reality, so what we did in our paper, we took the most recent estimates from a paper in Econometrica two years ago, about the total value

Impact of Minimum Income Floors on Welfare Costs

00:25:18 - a rental support and, you know, food stamps, unemployment insurance and so on. And then we, we, we put a minimum income floor. We said that even if the model predicts that your income goes down to zero, your consumption will not be close to zero. Because in reality, you know, the government, the society is going to put something that turn. If we hadn't done that, the 40% number I mentioned to you would've been like a, you know, like something

00:25:59 - It would be like a 90% of your income you would be willing to give up because of the technicality in the way our models work. So I would say for policy, income shops are very big, but that minimum threshold we are putting reduces the welfare costs substantially.

Faculti AI can make mistakes

Research

Policy

Analysis