understanding tail risk=1based on work with nic...
TRANSCRIPT
![Page 1: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/1.jpg)
Understanding Tail Risk1
Laura Veldkamp
New York University
1Based on work with Nic Kozeniauskas, Julian Kozlowski, Anna Orlik and VenkyVenkateswaran.
1/20
![Page 2: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/2.jpg)
Why Study Information Frictions?
Every expectation, mean, variance, covariance is conditioned on someinformation set. Information assumptions pervade every model.
Preferences, technologies and budget constraints have beenexhaustively studied. Information is less explored.
The finance sector is all about collecting, processing, transmitting andselling data (information).
Problem: We can’t observe information. How to discipline?Solutions:
I Count news stories, analyst reports, textual analysis, big dataprocessing.
I Model information choice: Information differs from preferences becausewe have some control over what we learn.
I Treat agents like econometricians (today). They see what we see.
2/20
![Page 3: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/3.jpg)
Introduction to Tail Risk
Asset prices reflect risk. Risk depends on underlying variance of anoutcome and on how much one knows.
Many models shock tail risk, uncertainty, firm-specific risk ...But where do these come from?
Tail Risk: Prob[yt+1 < α|It ]
Uncertainty: Stdev of a forecast (error) conditional on It .
Ut = Std [yt+1|It ] =
√E[(yt+1 − E (yt+1|It))2 |It
]Firm-specific (micro) risk:
∫(yit − yt)
2di .
Higher-order uncertainty:∫
(E [yt |Iit ]− E [yt ])2di .
3/20
![Page 4: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/4.jpg)
Two Possible Sources of Shocks
1 Actual variance of some data-generating distribution changes.I Jurado, Ludvigson, Ng (2015): Find two large increases in macro
variance.I Tail risk? Hard to measure changes.
2 Conditional variance changes because our beliefs about thedistribution change.
I Why would beliefs change if the true distribution is the same?I We must not know the distribution and learn about it.
4/20
![Page 5: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/5.jpg)
How Do We Learn About Distributions?
1 A Bayesian parametric approach
2 A classical econometrics, non-parametric approach.
In both cases,
We’ll use macro data and standard econometric tools to estimate adistribution and then re-estimate it each period with new data. Ouragents do the same.
Changes in variance and tails of this distribution are a key source ofshocks.
5/20
![Page 6: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/6.jpg)
Bayesian Approach: What Distribution to Estimate?
Key feature: Agents estimate tail probabilities.
A normal distribution fixes these → no Ut action.Need parameters that govern higher moments (skewness).
I Can capture skewness in GDP data (-0.3)I Key for our forecasts to resemble SPF forecast data
Solution: Take a linear hidden state model (Kalman filter system) anddo an exponential twist.
A form of g-and-h transformation used in statistics for Bayesiandistribution fitting (Headrick ’10).
6/20
![Page 7: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/7.jpg)
Forecasting Exercise
We estimate this:
yt = c + b exp(−St − σεt)St = ρSt−1 + σSξt
where εt and ξt ∼ iid N(0, 1). yt = GDP growth.
Use real-time GDP data (1968-2013, Philly Fed) to estimate.
Begin with prior beliefs estimated from 1947-68 data.
Observe each quarter of data and apply Bayes’ Law.I Metropolis-Hastings + change-of-measure→ distributions of parameters.
How big are uncertainty changes? Ut = Std [yt+1|It ].
7/20
![Page 8: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/8.jpg)
8/20
![Page 9: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/9.jpg)
Result 1: Large Uncertainty Shocks
known parameters estimated paramsnormal skewed normal skewed
Std dev of Ut 0 0.05% 0.48% 1.50%
VolatilityNormalSkewed
VolatilityNormalSkewed
1970 1980 1990 2000 2010
−0.15
−0.1
−0.05
0
0.05
0.1
Figure: Uncertainty (Ut) in linear and skewed models, in mean-zero, logdeviations from trend.
Parameter learning + Skewness = Large uncertainty shocks.
9/20
![Page 10: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/10.jpg)
What Explains Large Shocks? Tail Risk.
Tail Riskt ≡ Prob[yt+1 ≤ −6.8%|y t ] (1-in-100 year event)Correlation(BSw ,Ut) is 75% (both detrended).
UncertaintyBlack Swan Risk
1970 1980 1990 2000 2010
−0.4
−0.2
0
0.2
Most changes in uncertainty come from re-estimating tail risk.
10/20
![Page 11: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/11.jpg)
Why Is Tail Risk Volatile?
Extreme event probabilities are very sensitive to small revisions inskewness.
Skewness keeps fluctuating because it is hard to learn.
−20 −13.4 0 10 200
2
4
6
8
10
12
14
Growth rate (%)
Density
Skew −0.46
Skew −0.67
−13.4 −11.70
0.025
11/20
![Page 12: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/12.jpg)
Tail Risk amplifies uncertainty in bad times
Skewness can be represented as a concave function of a normal.
State0
∆at
State Uncertainty
Skewness, which governs tail risk, amplifies macro uncertainty in badstates.
12/20
![Page 13: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/13.jpg)
Tail Risk Also Creates Forecast Bias
E [yt+1|y t , θ] is mean GDP growth = 2.68%.E [yt+1|y t ] is average growth forecast = 2.29% in data, = 2.27% in model.
Lemma: Suppose y = g(x) where g is concave and x ∼ N(µ, σ). µ and σare unknown, with unbiased beliefs. Then mean > forecast.
. Mean Forecast
E[St+1|yt, θ]
Jensen effect
E[yt+1|yt, θ]
GDP
Growth (y)
State (S) E[St+1|yt]
Additional Jensen effect
from model uncertainty
E[Yt+1|yt, θ]
E[Yt+1|yt]
Forecaster believes
f(St+1|yt)
GDP
Growth (y)
State (S)
When we estimate parameter uncertainty and skewness, we matchthe bias in professional forecasts.
13/20
![Page 14: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/14.jpg)
Approach 2: Non-parametric, Classical Estimation
Consider an iid shock, φt , with unknown pdf g
Information set: finite history of shock realizations φt−snt−1s=0
The Gaussian kernel density estimator
gt (φ) =1
ntκ
nt−1∑s=0
Ω
(φ− φt−s
κ
)
Key property: Beliefs are martingales
Et [gt+1|It ] ≈ gt → Persistence
Next: use this mechanism to create persistence (long run risk).
14/20
![Page 15: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/15.jpg)
US Real GDP: Stagnation
7.50
8.00
8.50
9.00
9.50
10.00
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
0.13
15/20
![Page 16: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/16.jpg)
Tail Risk Stayed High
Why do some recessions have persistent effects ?
Because they cause us to re-assess macro risk.
Suggestive evidence from financial markets: A tail risk index
110
112
114
116
118
120
122
124
126
128
1301990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
Years
Note: Constructed from out-of-the-money put options on S&P 500
16/20
![Page 17: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/17.jpg)
Economic Model - based on Gourio (AER, 2012)
Representative household with Epstein-Zin preferences over, Ct − ζ L1+γt1+γ
A continuum of firms, indexed by i
Production: yit = Akαit l1−αit
Aggregate capital quality shocks: kit = φt kit φt ∼ g (·) iid
Idiosyncratic shocks (iid): Πit = vit [yit + (1− δ)kit ] ,∫vitdi = 1
Debt has a tax advantage and a default cost.
Labor hired in advance, before observing shocks.
17/20
![Page 18: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/18.jpg)
Capital Quality Shocks
Key feature: Increase left tail risk, post-2009
1950 1960 1970 1980 1990 2000 20100.8
0.9
1
1.1
1.2
0.8 0.85 0.9 0.95 1 1.05 1.1
Density
0
10
20
30
40
2007
2009
0.8 0.85 0.90
1
2
We do:
Calibrate model,
feed in this data through 2007, normalize ‘07 outcome to 0.
observe effects of 08-09 shocks,
take random draws from the 2009 distribution (report avg outcome).18/20
![Page 19: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/19.jpg)
Stagnation: Model and Data
2010 2020 2030-0.3
-0.2
-0.1
0
0.1
Capital quality shock
2010 2020 2030-0.3
-0.2
-0.1
0
0.1
GDP
ModelDataNo learning
2010 2020 2030-0.3
-0.2
-0.1
0
0.1
Investment
2010 2020 2030-0.3
-0.2
-0.1
0
0.1
Labor
Without belief revisions, a steady recovery to initial level19/20
![Page 20: Understanding Tail Risk=1Based on work with Nic ...people.stern.nyu.edu/lveldkam/pdfs/NorthLectureVeldkamp.pdf · Understanding Tail Risk1 Laura Veldkamp New York University 1Based](https://reader033.vdocument.in/reader033/viewer/2022042018/5e75cc6bff223409d777b57d/html5/thumbnails/20.jpg)
Conclusions
Obviously, no one knows the true distribution of shocks. Simple,displined tools to replace rational expectations hypothesis.
New data permanently reshapes our assessment of macro risks,especially tail risks because data on tails is scarce.
Changes in tail risk provide a unified theory of uncertainty, risk,sentiment shocks and belief biases.
→ A new persistence mechanism / source of long-run risk.→ A new source of price fluctuations→ A new risk factor
How much of business cycle fluctuations, asset pricing puzzles orother phenomena could learning about tail risk explain?
20/20