Bootstrapping Estimates of

the CER Model

Econ 424/Amath 462
Eric Zivot
Summer 2013 Summer 2013
Updated: J uly 23, 2013
Eric Zivot 2006
Data for Examples
returns.z
0
.
0
0
.
2
s
b
u
x
-
0
.
4
-
0
.
2
s
0
.
2
-
0
.
2
0
.
0
0
m
s
f
t
-
0
.
4
0
.
0
0
0
.
0
5
0
.
1
0
-
0
.
1
5
-
0
.
0
5
0
s
p
5
0
0
Eric Zivot 2006
T = 100 months
1993 1994 1995 1996 1997 1998 1999 2000
Index
Estimated Standard Errors
> se.muhat = sigmahat.vals/sqrt(nobs)
> rbind(muhat.vals,se.muhat)
sbux msft sp500 p
muhat.vals 0.0277 0.0275 0.01253
se.muhat 0.0135 0.0106 0.00378
i 2h i 2h l / ( b /2) > se.sigma2hat = sigma2hat.vals/sqrt(nobs/2)
> rbind(sigma2hat.vals,se.sigma2hat)
sbux msft sp500
sigma2hat.vals 0.01845 0.01141 0.00143
se.sigma2hat 0.00261 0.00161 0.00020
> se.sigmahat = sigmahat.vals/sqrt(2*nobs) g g / q ( )
> rbind(sigmahat.vals,se.sigmahat)
sbux msft sp500
sigmahat.vals 0.1358 0.1068 0.0378
Eric Zivot 2006
g
se.sigmahat 0.0096 0.0075 0.0026
R function sample() p ()
# random permutations of the index vector 1:5
> sample(5)
[1] 1 3 2 5 4
> sample(5)
[1] 4 2 3 5 1
# random sample of size 5 from MSFT return with
replacement
> sample(MSFT, 5, replace=TRUE)
[1] -0.02904 0.12130 -0.01890 -0.15332 -0.14627
Eric Zivot 2006
Brute Force Bootstrap
Same idea as Monte Carlo Simulation but instead of
generating randomdata froman assumed distribution generating random data from an assumed distribution,
you generate random data by sampling with replacement
from the observed data
# bootstrap distribution for
> B = 999 # why use 999?

> muhat.boot = rep(0, B)

> nobs = length(MSFT)
> for (i in 1:B) { > for (i in 1:B) {
+ boot.data = sample(MSFT, nobs, replace=TRUE)
+ muhat.boot[i] = mean(boot.data)
}
Eric Zivot 2006
}
Brute Force Bootstrap
# bootstrap bias
> mean(muhat.boot) - muhat.MSFT
[1] -0.0005643
# bootstrap SE
Bootstrap SE is very close
t l ti SE
# p
> sd(muhat.boot)
[1] 0.01045
to analytic SE
# analytic SE
> sigmahat.MSFT/sqrt(length(MSFT))
[1] 0.01068
Eric Zivot 2006
Brute Force Bootstrap Brute Force Bootstrap
Histogram of muhat.boot
2
0
0
0
5
Normal Q-Q Plot
F
r
e
q
u
e
n
c
y
1
0
0
1
5
0
0
.
0
3
0
.
0
4
0
.
0
5
m
p
l
e

Q
u
a
n
t
i
l
e
s
F
r
e
0
5
0
0
.
0
0
0
.
0
1
0
.
0
2
S
a
m
p
l
muhat.boot
-0.01 0.01 0.03 0.05
0
-3 -2 -1 0 1 2 3
Theoretical Quantiles
par(mfrow=c(1,2)) pa ( o c( , ))
hist(muhat.boot, col="slateblue1")
abline(v=muhat.MSFT, col="white", lwd=2)
qqnorm(muhat.boot)
li ( h t b t)
Eric Zivot 2006
qqline(muhat.boot)
par(mfrow=c(1,1))
R Package boot R Package boot
Implements a variety of bootstrapping functions
Background material is book by Davidson and
Hinkley, Bootstrap Methods and Their
A li ti C b id U i it P 1997 Application, Cambridge University Press, 1997.
Main functions are:
b t() b t t li d f ti boot() bootstrap user supplied function
boot.ci() compute bootstrap confidence
i t l interval
Eric Zivot 2006
Example: Bootstrapping sample mean Example: Bootstrapping sample mean
# function for bootstrapping sample mean
mean.boot = function(x, idx) { mean.boot function(x, idx) {
# arguments:
# x data to be resampled
# id ector of scrambled indices created # idx vector of scrambled indices created
# by boot() function
# value:
# ans mean value computed using resampled
# data
ans = mean(x[idx]) ans mean(x[idx])
ans
}
Eric Zivot 2006
Example: Bootstrapping sample mean
> MSFT.mean.boot = boot(MSFT, statistic = mean.boot, R=999)
> class(MSFT.mean.boot)
[1] "b t [1] "boot
> MSFT.mean.boot
Number of bootstrap samples
ORDINARY NONPARAMETRIC BOOTSTRAP
Call: Call:
boot(data = MSFT, statistic = mean.boot, R = 999)
Bootstrap Statistics :
original bias std. error
t1* 0.02756 -0.00013 0.01052
Eric Zivot 2006
Sample mean
Bootstrap estimate of
bias
Bootstrap estimate of SE
Histogram of t
0
4
0
0
.
0
5
0
.
0
6
s
i
t
y
0
3
0
0
.
0
3
0
.
0
4
D
e
n
s
1
0
2
0
0
.
0
1
0
.
0
2
t
*
0
1
0
.
0
1
0
.
0
0
0
t*
0.00 0.04 -3 -1 1 2 3
-
0
Quantiles of Standard Normal
Eric Zivot 2006
> plot(MSFT.mean.boot)
Compare Bootstrap Statistics with Analytic Formulas
ORDINARY NONPARAMETRIC BOOTSTRAP
Call:
boot(data = MSFT, statistic = mean.boot, R = 999)
Bootstrap Statistics :
original bias std. error
t1* 0.02756 -0.00013 0.01052 t1 0.02756 0.00013 0.01052
# compare boot SE with analytic SE
> se.muhat.MSFT = sigmahat.MSFT/sqrt(length(MSFT))
h t S > se.muhat.MSFT
[1] 0.01068
Eric Zivot 2006
Bootstrap Confidence Intervals Bootstrap Confidence Intervals
> boot ci(MSFT mean boot conf = 0 95 type = > boot.ci(MSFT.mean.boot, conf = 0.95, type =
+ c("norm","perc"))
BOOTSTRAP CONFIDENCE INTERVAL CALCULATIONS
Based on 999 bootstrap replicates
CALL :
boot.ci(boot.out = MSFT.mean.boot, conf = 0.95, type =
c("norm", "perc")) c( norm , perc ))
Intervals :
Level Normal Percentile
95% ( 0 0071 0 0483 ) ( 0 0065 0 0471 ) 95% ( 0.0071, 0.0483 ) ( 0.0065, 0.0471 )
Calculations and Intervals on Original Scale
Eric Zivot 2006
Example: Bootstrapping Sample SD Example: Bootstrapping Sample SD
# function for bootstrapping sample standard deviation
sd.boot = function(x, idx) {
# arguments:
# x data to be resampled
# idx vector of scrambled indices created by
# b t() f ti # boot() function
# value:
# ans sd value computed using resampled data
ans = sd(x[idx]) ans = sd(x[idx])
ans
}
Eric Zivot 2006
Example: Bootstrapping Sample SD Example: Bootstrapping Sample SD
> MSFT.sd.boot = boot(MSFT, statistic = sd.boot, R=999)
> MSFT.sd.boot
ORDINARY NONPARAMETRIC BOOTSTRAP
Call:
boot(data = MSFT, statistic = sd.boot, R = 999)
Bootstrap Statistics :
original bias std. error
t1* 0.1068 -0.00145 0.01078
# compare boot SE with analytic SE based on CLT
> se.sigmahat.MSFT = sigmahat.MSFT/sqrt(2*length(MSFT))
> se.sigmahat.MSFT
Eric Zivot 2006
g
[1] 0.00755
Histogram of t
4
0
0
.
1
3
0
.
1
4
t
y
3
0
1
1
0
.
1
2
D
e
n
s
i
t
2
0
9
0
.
1
0
0
.
t
*
0
1
0
0
.
0
8
0
.
0
9
t*
0.08 0.12
0
-3 -1 1 2 3
Quantiles of StandardNormal
Eric Zivot 2006
> plot(MSFT.sd.boot)
t Quantiles of Standard Normal
Example: Boostrapping Normal VaR Example: Boostrapping Normal VaR
ValueAtRisk.boot = function(x, idx, p=0.05, w=100000) {
# t d t t b l d # x.mat data to be resampled
# idx vector of scrambled indices created by
# boot() function
# p probability value for VaR calculation # p probability value for VaR calculation
# w value of initial investment
# value:
# ans Value-at-Risk computed using resampled data # ans Value at Risk computed using resampled data
q = mean(x[idx]) + sd(x[idx])*qnorm(p)
VaR = (exp(q) - 1)*w
VaR
}
Eric Zivot 2006
Example: Boostrapping Normal VaR Example: Boostrapping Normal VaR
> MSFT.VaR.boot
ORDINARY NONPARAMETRIC BOOTSTRAP
C ll Call:
boot(data = MSFT, statistic = ValueAtRisk.boot, R = 999)
Bootstrap Statistics :
original bias std. error
t1* -13769.40 210.2801 1886.953
Sample VaR estimate
Bootstrap SE
Eric Zivot 2006
p
Histogram of t
0
2
0
t
y
0
.
0
0
0
-
1
2
0
0
0
D
e
n
s
i
t
0
.
0
0
0
1
0
-
1
6
0
0
0
t
*
0
0
0
0
0
0
-
t*
-20000 -14000 -8000
0
.
0
0
0
-3 -1 1 2 3
-
2
0
0
0
Quantiles of StandardNormal
Eric Zivot 2006
> plot(MSFT.VaR.boot)
t Quantiles of Standard Normal
Example: Boostrapping Normal VaR Example: Boostrapping Normal VaR
> boot.ci(MSFT.VaR.boot, conf=0.95, type=c("norm", "perc"))
BOOTSTRAP CONFIDENCE INTERVAL CALCULATIONS
Based on 999 bootstrap replicates
CALL CALL :
boot.ci(boot.out = MSFT.VaR.boot, conf = 0.95, type =
c("norm", "perc"))
Intervals :
Level Normal Percentile
95% (-17678, -10281 ) (-17212, -10009 )

2 ( )
boot
SE
* *
.025 .975
[ , ] q q
Eric Zivot 2006