using fat tails to model gray...
TRANSCRIPT
![Page 1: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/1.jpg)
<#>
© 2008 Morningstar, Inc. All rights reserved.
Using Fat Tails to Model Gray Swans
× Paul D. Kaplan, Ph.D., CFAVice President, Quantitative ResearchMorningstar, Inc.
![Page 2: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/2.jpg)
2
Swans: White, Black, & Gray
× The Black Swan is a metaphor for a rare event with extreme impact× “All swans white” accepted truth until first black one seen× Example: Stock market crash on 10/19/87× Such events tell us that our models are seriously flawed× Term popularized in books by Nassim Taleb
× Gray Swans are “events of considerable nature which are far too big for the bell curve, which are predictable, and for which one can take precautions” Benoit Mandelbrot (inventor of fractal geometry)
× “We seem to have a once-in-a-lifetime crisis every three or four years” Leslie Rahl (founder of Capital Market Risk Advisors)
![Page 3: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/3.jpg)
3
If the Bell Curve is Such a Bad Model, Why Do We Use It?
× Central Limit Theorem× Sums of independent & identically distributed (i.i.d.)
random variables with finite variance tend towardsa normal distribution, regardless of underlying distribution
× Application to forecasting cumulative wealth× WT = W0(1+R1)(1+R2)…(1+RT)× lnWT = lnW0 +ln(1+R1)+ln(1+R2)+…+ln(1+RT)× So if log-returns are i.i.d., with finite variance,
log of cumulative wealth tends to a normal distribution & cumulative wealth tends to a lognormal distribution
![Page 4: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/4.jpg)
4
What if Variance is Infinite?
× In the 1960s, Mandelbrot & his student, Eugene Fama, explored a model in which extreme events occur at realistic frequencies
× Log of return relative, ln(1+R), has infinite variance× Generalized Central Limit Theorem concludes that log of
cumulative wealth has stable distribution× Stable distribution have very fat tails
× Until recently, Mandelbrot-Fama work largely ignored× Hard math× Most portfolio theory does not work with infinite variance × Difficult to estimate model from data
![Page 5: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/5.jpg)
5
Parameters of Stable Distributions
× Alpha – Fatness of Tails× 0<alpha ≤ 2 (normal)× if alpha ≤ 1, mean of distribution infinite
× Beta – Skewness (if alpha<2)× (fully left skewed) -1 ≤ beta ≤ 1 (fully right skewed)× if beta = 0, distribution symmetric
× Gamma – Scale× Gamma>0× If alpha=2, gamma2 = variance/2
× Delta – Location× if alpha>1, delta = mean of distribution
![Page 6: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/6.jpg)
6
Alpha & The Fatness of Tails
0
1
-5 -4 -3 -2 -1 0 1 2 3 4 5
Alpha = 0.5
Alpha = 1.0Alpha = 1.5
Alpha = 2.0
![Page 7: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/7.jpg)
7
Beta & Skewness
0-5 -4 -3 -2 -1 0 1 2 3 4 5
Beta = -0.75Beta = +0.75Beta = 0
![Page 8: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/8.jpg)
8
Scaling Property of Gamma & Delta
1
2
3
4
5
6
7
8
9
10
11
12
1 2 3 4 5 6 7 8 9 10 11 12
Delta
Gam
ma
Alpha = 0.5Alpha = 1.0Alpha = 1.5Alpha = 2.0
![Page 9: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/9.jpg)
9
What About Kurtosis?
× How much of the variance is due to infrequent extreme deviations from the mean, rather than frequent modest deviations
× Often described as a measure of fat tailness× No variance, no kurtosis either× Stable distributions have fatter tails than distributions with finite
kurtosis
![Page 10: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/10.jpg)
10
Normal Distributions: Thin Tails
0%
5%
10%
15%
20%
25%
30%
35%
40%
1 2 4 8 16 32 64 128 256
Given Loss of at Least x
Prob
abili
ty o
f Los
s of
at L
east
2x
![Page 11: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/11.jpg)
11
Finite Kurtosis: Dieting Tails
0%
5%
10%
15%
20%
25%
30%
35%
40%
1 2 4 8 16 32 64 128 256
Given Loss of at Least x
Prob
abili
ty o
f Los
s of
at L
east
2x
![Page 12: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/12.jpg)
12
Stable Distributions: Fat Tails
0%
5%
10%
15%
20%
25%
30%
35%
40%
1 2 4 8 16 32 64 128 256
Given Loss of ar Least x
Prob
abili
ty o
f Los
s of
at L
east
2x
![Page 13: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/13.jpg)
13
Alpha & Conditional Tail Probability
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Alpha
Prob
abili
ty o
f Los
s B
eing
at L
east
Tw
ice
of W
hat I
s K
now
n
![Page 14: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/14.jpg)
14
Log-Stable Model of Returns & Cumulative Wealth
× Lognormal model: ln(1+R) has a normal distribution× Log-stable model: ln(1+R) has a stable distribution× Parameters:
T1/alpha•gamma√T•sigma/√2gammasigma/ä2
T•deltaT•mudeltamu
beta-beta-
alpha2alpha2
StableNormalStableNormal
lnWTln(1+R)
![Page 15: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/15.jpg)
15
EnCorr Application: Histogram Overlay in Analyzer
× Procedure× Calculate historical log-returns× Fit parameters of stable distribution to historical log-returns× Draw resulting distribution of returns over histogram
× Results× Distribution curve that fits data better than lognormal model× Example: Monthly returns on S&P 500 Jan. 1926 – Dec. 2007
0.00660.0084Delta0.02830.0366Gamma
-0.4491-Beta1.66532.000Alpha
Log-StableLognormalParameter
![Page 16: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/16.jpg)
16
Histogram of S&P 500 with Lognormal Overlay
0
20
40
60
80
100
120
140
160
180
200
-29% -21% -13% -5% 3% 11% 19% 27% 35% 43%0
20
40
60
80
100
120
140
160
180
200
-0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4
![Page 17: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/17.jpg)
17
Histogram of S&P 500 with Log-Stable Overlay
0
20
40
60
80
100
120
140
160
180
200
-29% -21% -13% -5% 3% 11% 19% 27% 35% 43%0
20
40
60
80
100
120
140
160
180
200
-30% -20% -10% 0% 10% 20% 30% 40%
![Page 18: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/18.jpg)
18
Applications in EnCorr Optimizer: Forecasting & Simulation
× Use log-stable model in place of lognormal model× Setting the four parameters
× Use lognormal model value for E[ln(1+R)] for delta× Fit parameters of stable distribution to historical log-returns× Use fitted values of alpha & beta× Annualize fitted values of gamma using scaling property
![Page 19: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/19.jpg)
19
Example: Lognormal Model of S&P 500
× Set delta using SBBI data× Historical equity premium = 7.05%× Recent Treasury yield = 4.50%× E[R] = 7.05% + 4.50% = 11.56%× Historical standard deviation = 19.97%× Lognormal model gives E[ln(1+R)] = 0.0936 = delta
× Set alpha & beta from fit of monthly historical data× Alpha = 1.6653× Beta = -0.4491
× Set gamma by annualizing monthly fitted value× GammaA = 121/alpha•gammaM = 0.1260
![Page 20: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/20.jpg)
20
Wealth Forecasting Using Lognormal Model
-0.693150 5 10 15 20 25 30 35 40 45 50
Years into Future
Wea
lth In
dex
0.5
1
2
5
10
20
50
100
200
500
1000
2000
5000
5th Percentile
25th Percentile
50th Percentile
75th Percentile
95th Percentile
![Page 21: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/21.jpg)
21
Wealth Forecasting Using Log-Stable Model
-0.693150 5 10 15 20 25 30 35 40 45 50
Years into Future
Wea
lth In
dex
0.5
1
2
5
10
20
50
100
200
500
1000
2000
5000
5th Percentile
25th Percentile
50th Percentile
75th Percentile
95th Percentile
![Page 22: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/22.jpg)
22
Monte Carlo Simulation with Log-Stable Distributions
× Example: Drawdown Problem× Start with $1,000,000× Invest in S&P 500× Assume returns follow log-Stable distribution× Withdraw $80,000 per year for 50 years
× Use Monte Carlo Simulation× Run 1,000 simulations of 50 years each× Calculate percentiles of wealth for each year× Calculate probability of not running out of money each year
![Page 23: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/23.jpg)
23
Simulated Wealth Paths: 10th Percentile
40 5 10 15 20 25 30 35 40 45 50
Future Year
Wea
lth
LognormalLog-Stable
$10,000
$20,000
$50,000
$100,000
$200,000
$500,000
$1,000,000
![Page 24: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/24.jpg)
24
Probability of Not Running Out of Money
50%
55%
60%
65%
70%
75%
80%
85%
90%
95%
100%
0 5 10 15 20 25 30 35 40 45 50
Future Year
Prob
abili
ty o
f Not
Run
ning
Out
of M
oney
Lognormal
Log-Stable
![Page 25: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/25.jpg)
25
Summary
× Significant market events occur far more frequently than predicted by the lognormal model of returns
× The basis of the lognormal normal model is the Central Limit Theorem which assumes that variance is finite
× If variance is infinite, the log-stable model follows from the Generalized Central Limit Theorem
× Stable distributions have parameters for fatness of tails, skewness, scale, & location
× The log-stable model will be an alternative to the lognormal model in the EnCorr Analyzer & Optimizer
![Page 26: Using Fat Tails to Model Gray Swansmorningstardirect.morningstar.com/clientcomm/LogStableDistributio… · Using Fat Tails to Model Gray Swans × Paul D. Kaplan, Ph.D., CFA Vice President,](https://reader034.vdocument.in/reader034/viewer/2022051805/5ff373c19826cc770c5402e3/html5/thumbnails/26.jpg)
26