quantitative finance - vosvrdaweb
Post on 07-Jun-2022
6 Views
Preview:
TRANSCRIPT
Quantitative Finance
Conditional Heteroskedastic Models
Miloslav S. VosvrdaDept of Econometrics
ÚTIA AV ČR
Robert EngleMV1
Professor of FinanceMichael Armellino Professorship in the
Management of Financial Services
Joined Stern 2000Phone: (212) 998-0710Fax : (212) 995-4220
Email: rengle@stern.nyu.eduOffice: KMEC 9-6244 West Fourth Street
Suite 9-62 New York, NY 10012-1126
Snímek 2
MV1 Prof Engle je nositelem Nobelovy ceny za ekonomii pro rok 2003 Miloslav Vosvrda; 23.3.2004
Recent Awards
The Bank of Sweden Prize in Economic Sciencesin
Memory of Alfred Nobel 2003for
methods of analyzing economic time series with time-varying volatility (ARCH)
The Pivotal Interest in almost all
financial applications
is
The Predictabilityof
Price Changes
The Pivotal Requirement in almost all
financial applications
Any Volatility ModelMust Be Capable to
Forecast the Volatility
is
Financial Time SeriesThe volatile behavior in financial markets is usuallyreferred to as the “volatility”. Volatility has become a very important concept in different areas in financial theory and practice. Volatility is usually measured by variance, or standard deviation. The financial marketsare sometimes more volatile, and sometimes less active. Therefore a conditional heteroskedasticity and stylized facts of the volatility behavior observed in financial time series are tuypical features for each financial market.
ARCH, GARCH
This one is a class of (generalized) autoregressive conditional heteroskedasticitymodels which are capable of modeling of time varying volatility and capturing many of the stylized facts of the volatility behavior observed in financial time series.
Stylized Facts1) Non-Gaussian, heavy-tailed, and skewed
distributions. The empirical density function of returns has a higher peak around its mean, butfatter tails than that of the corresponding normaldistribution. The empirical density function istailer, skinnier, but with a wider support than thecorresponding normal density.Figure
NEXT
2) Volatility clustering (ARCH-effects).Figure3) Temporal dependence of the tail behavior.
6 4 2 0 2 4 60
0.2
0.4
0.6
dnormx 0, 0.7,( )
dt x 3,( )
x Back
Back
Series and Conditional SD
0.02
0.03
0.04
0.05
Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q11984 1985 1986 1987 1988 1989 1990 1991 1992
0.1
Conditional SD
-0.1
0.0
Original Series
Val
ues
4) Short- and long-range dependence.
5) Asymmetry-Leverage effects. There is evidencethat the distribution of stock returns is slightlynegatively skewed. Agents react more strongly to negative information than to positive information.Figure
Next
Back
0 200 400 600 800 100020
0
20
40
60EGARCH(1,1)
54.403re
turn
s
time
10.126−
rn
1 103×1 n
Capital Market Price Models• The Random Walk Model (RWM)• Martingale (MGL)• Conditional Heteroscedastic Model (CHM)
– Models of the first category• ARCH• GARCH
– IGARCH– EGARCH– CHARMA
• ARCH-M• GARCH-M
– Models of the second category• Stochastic volatility model Next
The Random Walk Model(RWM)
the price observed at the beginning of time tS t
( ) ( ) 2an error term with =0 and Var values of the independent of each other
t t tE εε ε ε σ=
1t t tS S ε−= + 1t t tS S ε−− =t
t jS1jε
=
= ∑RWM was first hypothesis about how financial prices move. Fama (1965) compiled EMH.
A Prediction by the RWM
( ) ( )1 11 , , ...h h h hˆ
hS E S S S S+ −= =
( ) ( )( ) ( )
2 1
1 2 1
ˆ 2 , ,...ˆ, ,... 1
h h h h
h h h h h h
S E S S S
E S S S S Sε
+ −
+ + −
= =
= + = =
( )ˆh hS l S=
The 1-step ahead forecast at the forecast origin h is
The 2-step ahead forecast at the forecast origin h is
For any l > 0 forecast horizon we have
Therefore, the RWM is not mean-reverting
Forecast Error
( ) 1h h l he l ε ε+ += + +
( ) 2hVar e l l εσ=⎡ ⎤⎣ ⎦
l→∞
The l-step ahead forecast error is
So thatwhich diverges to infinity as
The RWM is not predictable.
RWM with drift
( )1t tE S Sµ = − −
1t t tS Sµ ε−= + +RWM with drift for the price S
t
Var t 2
1i
iεε σ⎛ ⎞
=
=∑⎜ ⎟⎝ ⎠
t is
where
Then S S t01
t
t jj
µ ε=
= + +∑and
Therefore the conditional standard deviationgrows slower than the conditional expectation
2t εσ
Le Roy’s critique of RWMThe critique of RWM by Le Roy (1989) led to some serious questions about RWM as the theoretical model of financial markets. The assumption that price changes are independent was found to be too restrictive. Therefore was, after broad discussion, suggested the following model
1t t tt
t
S S DrS
+ − +=
wherer is a returnD is a dividend
We assume that ( ) is a constantt tE r rℑ =Taking expectations at time t of the both sides
1t t tt
t
S S DrS
+ − +=
( )1
1t t t
t
E S DS
r+ + ℑ
=+
( )1 1 1t t t t th S h S D+ + +⋅ = ⋅ +
and we get
We assume that reinvesting of dividends isand t t tx h S= ⋅
Back
Thus
( ) ( ) ( )( ) ( )1 1t t t
1 1 1 1t t t t t t t t+ + + +E x E h S h S D
r h S r x
ℑ = ⋅ ℑ = ⋅ + =
= + ⋅ = +i.e. that is a martingalex . For , is a submartingale, because
t0r > x
For , is a supermartingale, because tx
t
( )E x x1t t t+ ℑ ≥
0r <
( )E x x1t t t+ ℑ ≤
Very Important DistinctionA stochastic process following a RWM is morerestrictive than the stochastic process thatfollows a martingale.A financial series is known to go throughprotracted both quiet periods and periods of turbulence. This type of behaviorcould be modeled by a process with conditionalvariances. Such a specification is consistent with a martingale, but not with RWM.
Martingale processes lead to non-linear stochastic processes that are capable of modeling higher conditional moments. Such models are called
Conditional Heteroskedastic Models.These ones are very important for a modeling of the volatility. The volatility is an very important factor in options trading. The volatility means the conditional variance of underlying asset return
( ) ( )( ) ( )221 1 1,where t t t t t t t t tVar r E r E rσ µ µ− − −= ℑ = − ℑ = ℑ
Back
Conditional HeteroskedasticModel (CHM)
( ) ( )( ) ( )22 21 1 1t t t t t t t tVar r E r Eσ µ ε− − −= ℑ = − ℑ = ℑ
( )E r 1t t t
01 1
t t tr
p q
t i t i i t ii i
r
µ ε
µ
µ µ φ θ ε
−
− −= =
= +
= ℑ
= + ⋅ + ⋅∑ ∑
Let rt is a stationary ARMA(p,q) model, i.e.,
We distinguish two categories of CHM• The first category: An exact function to a
governing of the evolution of the st2 is used
• The second category: A stochastic equation to a describing of the st
2 is usedThe models coming under the first category are models of type ARCH or GARCH. Thesemodels may catch three out of the five stylizedfeatures, namely 1), 2), and 4).
The models coming under the second category are models of type Stochastic Volatility Model.
Back
ARCH modelThis type of model was introduced by Engle in the paper "Autoregressive Conditional Heteroskedasticity with Estimates of the Varianceof U.K. Inflation," Econometrica 50 (1982),pp. 987-1008, for modeling the predictive variance for U.K. inflationrates.
ARCH modelThe basic idea t t tr µ ε= +
t tr tε µ= −
2 2 20 1 1
t tσ η= ⋅ innovations
t t m t mσ α α ε α ε− −= + + +
( ) ( ), 0 , 1tt t
0 10 , 0 , , 0m
E V a rη η ηα α α
= =
> ≥ ≥
( ) ( )0,1 o rt tN tη η ⋅∼ ∼
ARCH(1)
0 1
t t t
0, 0ε σ ηα α= ⋅
2 20 1 1t t
> >
σ α α ε −= +
( )0 1/ 1 .α α−
so that
This process is stationary if, and only if, Unconditional variance is equal
1 1.α <
The fourth moment is finite if 213 1.α <
( )0,1t Nη ∼
2εσ
The kurtosis is so given by ( ) ( )2 21 13 1 / 1 3 .α α− −
0 200 400 600 800 10005
0
5
10ARCH(1)
time
retu
rns r n
n
µn β0 β1 n⋅+:=rn µn εn α0 α1 εn 1−( )2⋅+⎡
⎣⎤⎦⋅+:=
α0 0.1:= α1 0.1:= β0 0.02:= β1 0.007:=
0 100 200 300 400 50010
5
0
5ARCH(1)
time
retu
rns
4.75
5.327−
rn µn−
5001 n
0 200 400 600 800 10000
0.5
1
1.5ARCH(1) Volatility
time
vola
tility
1.273
0.1
α0 α1 εn 1−( )2⋅+
σε
1 103×1 n
0 1 2 3 4 5 60.5
0
0.5
1
r-mi(r-mi)^2+Stadard deviation-Stadard deviation
Autocorrelation Function1
0.05−
A i
AA i
σε
σε−
60 i
0 1 2 3 4 5 60.5
0
0.5
1
r-mi(r-mi)^2+Stadard deviation-Stadard deviation
Partial Autocorrelation Function1
0.069−
Pi
PP i
σε
σε−
60 i
Simulated ARCH(1) errorse(
t)
0 100 200 300 400 500
-1.0
-0.5
0.0
0.5
1.0Simulated ARCH(1) volatility
sigm
a(t)
0 100 200 300 400 500
0.2
0.4
0.6
0.8
1.0
Sample Quantiles: min 1Q median 3Q max
-0.9662 -0.1044 0.005786 0.1096 1.225
Sample Moments: mean std skewness kurtosis
0.002704 0.2315 0.1384 7.016
Number of Observations: 500
moment rr 0, 4,( ) 6.214=
moment rrr 0, 4,( ) 2.402 10 3×=
moment rr 0, 3,( ) 1.85=
moment rrr 0, 3,( ) 108.796=
The fourth
The fourth
The third
The third
Daily Stock Returns of FORD
Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1
1984 1985 1986 1987 1988 1989 1990 1991 1992
-0.1
5-0
.10
-0.0
50.
000.
050.
10
Series : ford.s
Lag
AC
F
0 2 4 6 8 10 12
0.0
0.2
0.4
0.6
0.8
1.0
Lag
AC
F
0 2 4 6 8 10 12
0.6
0.8
0.4
0.2
0.0
1.0
Series : ford.s^2
The kurtosis exceeds 3, so that the unconditional distribution of ε is fatter tailed than the normal.
Testing for ARCH effects.This test is constructed on a simple LagrangeMultiplier (LM) test. The null hyphothesis is that:There are no ARCH effects,i.e.,
The test statistic is
1 0mα α= = =
2 2~ ( )LM N R pχ= ⋅
Test for ARCH Effects: LM Test
Null Hypothesis: no ARCH effects
Test Statistics:FORD
Test Stat 112.6884p.value 0.0000
Dist. under Null: chi-square with 12 degrees of freedom
Total Observ.: 2000
A practical problem with this model is that, withm increasing, the estimation of coefficients oftenleads to the violation of the non-negativity of thea’s coefficients that are needed to ensure that conditional variance st is always positive. A natural way to achieve positiveness of the conditional variance is to rewrite an ARCH (m) model as
( )20 , 1 , 1 , 1 1, , , TT
t m t m t m t t t mσ α ε ε− − − − −= + Ε ΩΕ Ε = …
ForecastingForecasts of the ARCH model can be obtained recursively. Consider an ARCH(m) model. At forecastorigin h, the 1-step ahead forecast of is
( )2 2 20 1 11h h m h mσ α α ε α ε + −= + + +
( ) ( )2 2 2 20 1 2 22 1h h h m h mσ α α σ α ε α ε + −= + + + +
21hσ +
( ) ( )2 20
1
m
h i hi
l l iσ α α σ=
= + −∑
( )2 2 if 0h h l il i l iσ ε + −− = − ≤where
The 2-step ahead forecast of is22hσ +
The l-step ahead forecast of is2h lσ +
0 200 400 600 800 1000 12005
0
5
10
15ARCH(1) Forecasting
time
retu
rns
rn
rnn
rrnn
rrrnn
n nn,
Weakness of ARCH Models• The model assumes that positive and negative shocks have
the same effects on volatility because it depends on the square of the previous shocks.
• The ARCH model is a little bit restrictive because for a finite fourth moment is necessary to be This constraint becomes complicated for higher order of the ARCH.
• The ARCH model provides only way to describe the behavior of the conditional variance. It gives no any new insight for understanding the source of variations of a financial time series.
• ARCH models are likely to overpredict the volatility because they respond slowly to large isolated shocks .
[ )2 0,1/ 3 .α ∈1
Although the ARCH model is simple, it often requires many parameters to adequately describethe volatility process. To obtain more flexibility, the ARCH model was generalised to GARCH model. This model works with the conditional variance function.
Back
GARCH model
2 2 20
1 1
,t t t t t tm s
t i t i j t ji j
r µ ε ε σ η
σ α α ε β σ− −= =
= + = ⋅
= + +∑ ∑
( ) ( )
( )0
max ,
( ) 1, and 0 ,i i i
, 0, 10, 0, 0,
t t t
i j
m s
E Varη η ηα α β
= =
> ≥ ≥
1
0i
j
for i m
for j s
α β α+ < = >=∑
>=β
A Connection to the ARMA processFor better underestanding the GARCH model is usefulto use the following transform:
2 2 2 2 2 2 so that , and t t t t t t t i t i t iξ ε σ σ ε ξ σ ε ξ− − −= − = − = −
( )( )max ,
2 20
m s s
t i i t i t j t j
Substitute these expressins into GARCH equation andwe get as follows
1 1i j
ε α α β ε ξ β ξ− −= + + + −∑ ∑= =
tξ
( )0, cov , 0, for 1.t t t jE jξ ξ ξ −= = ≥
The process is a martingale difference series, i.e.,
The process
GARCH(max(m,s),s)
is an application of the ARMA idea to the squared series . Thus2
tε
( )( )max ,m s s
2 20
1 1t i i t i t j t j
i j
ε α α β ε ξ β ξ− −= =
= + + + −∑ ∑
( )( )max ,
20
1
/ 1m s
t i ii
Eε α α β=
⎛ ⎞= − +⎜ ⎟⎜ ⎟
⎝ ⎠∑
GARCH(1,1)
( )221 1 11 2 0tε α β−− − + >
21tε −
•A large gives rise to a large . A large tends to be followed by large generating the well-known behavior of volatility clustering in financial time series.•If then kurtosis
•The model provides a simple parametric function for describing the volatility evolution.
2tσ2 2
1 1t torε σ− −2tε
( )( ) ( )2 221 1 1 1 13 1 /1 2 3tα β ε α β−− + − − + >
2 2 20 1 1 1 1t t tσ α α ε β σ− −= + +
( )0 1 1 1 10 , 1 0 , 1 0 , 1α α β α β> > ≥ > ≥ + <
0 200 400 600 800 10005
0
5
10GARCH(1,1)
time
retu
rns
8.883
1.229−
rn
1 103×1 n
GARCH(1,1) Volatility3
0.5
1
1.5
2
2.5
0 200 400 600 800 10000
vola
tility
2.765
0.403
δn
1 103×1 ntime
Forecasting by GARCH(1,1)
( )20
2 21 1
1h
h h
σ α
α ε β σ
=
+ +
We assume that the forecast origin is h. For 1-step ahead forecast For 2-step ahead forecast
For l-step ahead forecast
( )( ) ( )
20
21 1
2
1h
h
σ α
α β σ
= +
+
( )( )( )
( ) ( )1
0 1 1 12 21 1
1 1
11
1
l
lh hl
α α βσ α β σ
α β
−
−− +
= + +− −
Therefore ( )2 0
1 11h ll ασα β→∞⎯⎯⎯→
− −
The multistep ahead volatility forecasts of GARCH(1,1) model converge to the unconditional
t
infinity provided that Var(εt) exists.variance of ε as the forecast horizon increases to
0 200 400 600 800 1000 12005
0
5
10
15
original realizationforecasting+standard deviation-standard deviation
GARCH(1,1) Forecasting
time
retu
rns
13.82
2.887−
r n
rnn
rr nn
rrr nn
1.1 103×1 n nn,
Mean Equation: ford.s ~ 1
Conditional Variance Equation: ~ garch(1, 1)
Coefficients:
C 7.708e-004A 6.534e-006
ARCH(1) 7.454e-002GARCH(1) 9.102e-001
0.02
0.03
0.04
0.05
Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q11984 1985 1986 1987 1988 1989 1990 1991 1992
Conditional SD
-0.1
0.0
0.1
Original Series
Valu
esSeries and Conditional SD
-0.1
5-0
.10
-0.0
50.
000.
050.
10
Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q41984 1985 1986 1987 1988 1989 1990 1991 1992
ford.mod11
Valu
esSeries with 2 Conditional SD Superimposed
0.0
0.2
0.4
0.6
0.8
1.0
0 10 20 30
ACF
Lags
ACF of Observations
0.0
0.2
0.4
0.6
0.8
1.0
0 10 20 30
ACF
Lags
ACF of Squared Observations
-0.05
0.00
0.05
0 20 40 60
CCF
Lags
Cross Correlation
-0.1
5-0
.10
-0.0
50.
000.
050.
10
Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q41984 1985 1986 1987 1988 1989 1990 1991 1992
residuals
Res
idua
lsGARCH Model Residuals
0.01
50.
025
0.03
50.
045
0.05
5
Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q41984 1985 1986 1987 1988 1989 1990 1991 1992
volatility
Con
ditio
nal S
DGARCH Volatility
-6-4
-20
24
Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q4 Q2 Q41984 1985 1986 1987 1988 1989 1990 1991 1992
residuals
Stan
dard
ized
Res
idua
ls
GARCH Standardized Residuals
0.0
0.2
0.4
0.6
0.8
1.0
0 10 20 30
ACF
Lags
ACF of Std. Residuals
0.0
0.2
0.4
0.6
0.8
1.0
0 10 20 30
ACF
Lags
ACF of Squared Std. Residuals
-6
-4
-2
0
2
4
-2 0 2
QQ-Plot
10/13/1989
12/23/1991
10/19/1987
Quantiles of gaussian distribution
Stan
dard
ized
Res
idua
ls
QQ-Plot of Standardized Residuals
garch(formula.mean = ford.s ~ 1, formula.var = ~ garch(1,1))Mean Equation: ford.s ~ 1Conditional Variance Equation: ~ garch(1, 1)Conditional Distribution: gaussian--------------------------------------------------------------
Estimated Coefficients:--------------------------------------------------------------
Value Std.Error t value Pr(>|t|) C 7.708e-004 3.763e-004 2.049 0.02031225A 6.534e-006 1.745e-006 3.744 0.00009313
ARCH(1) 7.454e-002 5.362e-003 13.902 0.00GARCH(1) 9.102e-001 8.762e-003 103.883 0.00
AIC(4) = -10503.79BIC(4) = -10481.39
Normality Test:--------------------------------------------------------------
Jarque-Bera P-value 364.2 0
Shapiro-Wilk P-value 0.9915 0.9777
Ljung-Box test for standardized residuals:--------------------------------------------------------------
Statistic P-value Chi^2-d.f. 14.82 0.2516 12
Ljung-Box test for squared standardized residuals:--------------------------------------------------------------
Statistic P-value Chi^2-d.f. 14.04 0.2984 12
Lagrange multiplier test:--------------------------------------------------------------Lag 1 Lag 2 Lag 3 Lag 4 Lag 5 Lag 6 Lag 72.135 -1.085 -2.149 -0.1347 -0.9144 -0.2228 0.708
Lag 8 Lag 9 Lag 10 Lag 11 Lag 12 C
-0.2314 -0.6905 -1.131 -0.3081 -0.1018 0.9825
TR^2 P-value F-stat P-value14.77 0.2545 1.352 0.2989
Test for Residual Autocorrelation
Test for Autocorrelation: Ljung-Box
Null Hypothesis: no autocorrelation
Test Statistics: Test Stat 14.8161p.value 0.2516
Test for Residual^2 Autocorrelation
Test for Autocorrelation: Ljung-Box
Null Hypothesis: no autocorrelation
Test Statistics: Test Stat 14.0361p.value 0.2984
0 200 400 600 800 10000.6
0.4
0.2
0
0.2
0.4GARCH(1,1)
time
retu
rns
0.395
0.538−
rn
1 103×1 n
0 100 200 300 400 5004
2
0
2
4GARCH(1,1) Volatility
time
vola
tility
3.055
3.114−
δnk
5001 nk
0 200 400 600 800 1000 12000.6
0.4
0.4
0.2
0
0.2
0.6
original realizationforecasting+standard deviation-standard deviation
GARCH(1,1) Forecasting
time
retu
rns
0.459
0.538−
r n
r nn
rr nn
rrr nn
1.1 10 3×1 n nn,
Weakness of GARCH Models
This model encounters the same weakness as theARCH model. In addition, recent empirical studies ofhigh-frequency financial time series indicate that the tail behavior of GARCH models remains too short even with standardised Student t-innovations.
Back
ARCH-M modelThe return of a security may depend on its volatility.If we take conditional deviation as a measure for risk,it is possible to use risk as a regressor in returnsmodeling. To model such a phenomenon, one may consider the ARCH-M model, where M meansin mean. This type model was introduced by Engle in the paper "Estimation of Time Varying Risk Premia inthe Term Structure:the ARCH-M Model," (with David Lilien and Russell Robins), Econometrica 55 (1987): 391-407.
A simple ARCH-M(1) model
2 2 20 1 1
t t t t
t t m t m
r cµ σ η
σ α α ε α ε− −
= + ⋅
= + + +The parameter c is called the risk premium parameter. A positive c indicates that the return is positivelyrelated to its past volatility.
0 200 400 600 800 10002
0
2
4
6
8ARCH-M(1)
time
retu
rns
7.079
0.08−
rn
1 103×1 n
Back
GARCH-M ModelA simple GARCH(1,1)-M model
2
2 2 20 1 1 1 1
t t t t
t t t
t t t
ε σ ηr cµ σ ε= + ⋅ +
σ α α ε β σ− −
= ⋅
= + +
The parameter c is called the risk premium parameter.A positive c indicates that return is positively related to its past volatility. By formulation we can see thatthere are serial correlations in the return series rt.
0 200 400 600 800 10005
0
5
10GARCH-M(1,1)
time
retu
rns
8.959
1.181−
rn
1 103×1 n
Back
IGARCHIGARCH models are unit-root (intergrated) GARCHmodels. This type model was introduced by Engle and Bollerslev in the paper “Modelling the Persistence of ConditionalVariances," Econometric Reviews,5, (1986), pp. 1-50.A key feature of IGARCH models is that the impactof past squared shocks for onis persistent.
2 2t i t i t iζ ε σ− − −= − 0i > 2
tε
IGARCH(1,1)An IGARCH(1,1) model can be written as
where is defined as before and
2 2 20 1 1 1 1
, ,t t t t t t
t t t
r µ ε ε σ η
σ α β σ α ε− −
= + =
= + +
tε 11 0.β> >
0 200 400 600 800 100010
5
0
5
10
15IGARCH(1,1)
time
retu
rns
10.025
6.151−
rn
1 103×1 n
Forecasting by IGARCH(1,1)When repeated substitutions in
givewhere h is the forecast origin. The effect of on futurevolatilities is also persistent, and the volatility forecasts form astraight line with slope . The case of is ofparticular interest in studying the IGARCH(1,1) model. Thevolatility forecasts are simply for all forecast horizons.
( ) ( ) ( )2 20 1 1 1h hl lσ α α β σ= + + −
1 1 1α β+ =
( ) ( ) ( )2 201 1 , 1,h hσ σ α= + − ≥
( )2 1hσ
0α0 0α =
( )2 1hσ
0 200 400 600 800 1000 120010
0
10
20
IGARCH(1,1) Forecasting40
30
time
retu
rns
31.071
6.151−
rn
rnn
rrnn
rrrnn
1.1 103×1 n nn,
Back
FARIMAGeneral IGARCH(p,q) process is called FARIMA(p,d,q):
, ,t t t t t tr µ ε ε σ η= + =2 2 2 2 2
0 1 1 1 1 1 1 1( ) (1 ) ( ) 2t t t tσ tα β σ ε α ε α β ε− − −= + − + + + + −
EGARCH ModelThis model allows to consider asymmetric effectsbetween positive and negative asset returns throughthe weighted innovation
( ) ( )t t t tg Eη θη γ η η⎡ ⎤= + −⎣ ⎦where θ and γ are real constant.
Leverage Effects
Negative shocks tend to have a larger impact on volatility than positive shocks. Negative shocks tend to drive down the shock price, thus increasing the leverage ( i.e. the debt-equity ratio) of the shock and causing the shock to be more volatile. The asymmetricnews impact is usually refferred to as the leverage effect.
( )t tEη η− ( ) 0.tE g η =⎡ ⎤⎣ ⎦
,rt t t t t tµ ε ε= + σ η= ⋅
Because is iid sequence then
So EGARCH(m,s) can be written as
( )2 10 1
1
1exp1
ss
t tmm
L L gL L
β βσ α ηα α −
⎛ ⎞+ + += +⎜ ⎟− − −⎝ ⎠
EGARCH(1,0)
( )20 1
1
,
1exp1
t t t t t t
t tgL
ε ε σ η= ⋅r µ= +
σ α ηα −
⎛ ⎞= +⎜ ⎟−⎝ ⎠
( )( ) ( )( ) ( )
if 0
if 0t t t
tt t t
Eg
E
θ γ η γ η ηη
θ γ η γ η η
⎧ + − ≥⎪= ⎨− − <⎪⎩
( )tg ηThe asymmetry of can easily be seen as
( )( )
( )
1t−12
12 21 0 1
112
1
exp if 0
exp 1 2 /
exp if 0
t
tt t
tt
t
εθ γ εσ
σ σ α α γ πεθ γ εσ
−
−
−
−−
−
⎧ ⎡ ⎤⎪ ⎢ ⎥+ ≥⎪ ⎢ ⎥⎪ ⎣ ⎦⎡ ⎤= − − ⎨⎣ ⎦ ⎡ ⎤⎪
⎢ ⎥− <⎪⎢ ⎥⎪ ⎣ ⎦⎩
and
0 200 400 600 800 100020
0
20
40
60EGARCH(1,1)
time
retu
rns
54.403
10.126−
rn
1 103×1 n
Back
garch(formula.mean = hp.s ~ 1, formula.var = ~ egarch(1,1), leverage = T, trace = F)
Mean Equation: hp.s ~ 1Conditional Variance Equation: ~ egarch(1, 1)
Coefficients:
C 0.000313A -1.037907
GARCH(1) 0.886652LEV(1) -0.133998
ARCH(1) 0.227878
CHARMAThis model uses random coefficients to producea conditional heteroscedasticity. The CHARMA model has the following form
( )20, ,t ηη σ∼ ( ) 1 ,..., Tt t mtδ δ δ=
1 1 2 2
,...
t t t
t t t t t mt t m t
r µ εε δ ε δ ε δ ε η− − −
= += + + + +
whereis a sequence of iid random vectors with mean zeroand non-negative definite covariance matrix Ώ and
tδ is independent of tη
CHARMA(2)
tε
1 1 2 2,t t t t t t t t tr µ ε ε δ ε δ ε η− −= + = + +
The CHARMA model can easily be generalized sothat the volatility of may depend on someexplanatory variables. Let be m explanatoryvariables available at time t. Consider the model
1
mit t
x=
, 11
,m
t t t t it i t ti
r xµ ε ε δ η−=
= + = +∑where and are sequences of random vectors and random variables .
tη ( ) 1 ,..., Tt t mtδ δ δ=
0 200 400 600 800 10005
0
5
10
15CHARMA(2)
time
retu
rns
11.25
3.262−
rn
1 103×2 n
Then the conditional variance of istε
( ) ( )2 21, 1 , 1 1, 1 , 1,..., ,...,
Tt t m t t m tx x x xησ σ − − − −= + Ω
is a diagonal matrixΩwhere
Back
Stochastic Volatility Model
( )2
2 20 1 1
, ,t t t t t t
t t t
r gµ ε ε= + = σ η
σ α α σ υ−= + +
( )0,1iidtη ∼
( )20,t iid υυ σ∼Back
t tη υ⊥where
The Routes Ahead
• High frequency volatility• High dimension correlation• Derivative pricing• Modeling non-negative processes• Analysing conditional simulations by
Least Squares Monte Carlo
Thank you very muchfor your attention
top related