Estimating Volatilities and Correlations Following Options, Futures, and Other Derivatives, 5th edition by John C. Hull Chapter 17
Daniel HERLEMONT
Standard Approach to Estimating Volatility
Define σn as the volatility per day between day n-1 and day n, as estimated at end of day n-1 Define Si as the value of market variable at end of day i Define ui= ln(Si/Si-1) σ n2 = u =
1 m ∑ (u − u ) 2 m − 1 i =1 n − i
1 m ∑u m i =1 n − i
Daniel HERLEMONT
Page 1 1
Simplifications Usually Made
Define ui as (Si-Si-1)/Si-1 Assume that the mean value of ui is zero Replace m-1 by m This gives
σ 2n =
1 m 2 ∑ u m i =1 n − i
Daniel HERLEMONT
Statistical properties 2 ^2 2 ERQM =E (σ n − σ )^ 2 = σ 4 n
If prices follow a lognormal process
Example: with 50 historical price, the relative error on variance estimate is still about 20% Daniel HERLEMONT
Page 2 2
Efficient estimators with Highs and Lows Parkinson Rogers Satchell Garman Klass
Exemple: with n=50, relative error is about 10%, to be compared with the 20% error
Daniel HERLEMONT
Volatility based on Open/Close/High/Low Source: Malik Magdon-Ismail, Amir F. Atiya Volatility Estimation Using High, Low, and Close Data
Daniel HERLEMONT
Page 3 3
Difficulty in evaluating and comparing volatility models is due to the fact that volatility is not directly observable. An approach to solve this particular problem is to compare the volatility forecasts, with squared returns
Daniel HERLEMONT
Weighting Scheme
Instead of assigning equal weights to the observations we can set
∑
σ 2n =
m i =1
α i u n2− i
where m
∑α
i
=1
i =1
Daniel HERLEMONT
Page 4 4
ARCH(m) Model
In an ARCH(m) model we also assign some weight to the long-run variance rate, VL:
σ n2 = γ V L +
∑
m i =1
α i u n2 − i
where m
γ+
∑α
i
=1
i =1
Daniel HERLEMONT
EWMA Model
In an exponentially weighted moving average model, the weights assigned to the u2 decline exponentially as we move back through time This leads to
σ n2 = λσ n2 −1 + (1 − λ ) u n2−1
Daniel HERLEMONT
Page 5 5
Attractions of EWMA
Relatively little data needs to be stored We need only remember the current estimate of the variance rate and the most recent observation on the market variable Tracks volatility changes RiskMetrics uses λ = 0.94 for daily volatility forecasting
Daniel HERLEMONT
GARCH (1,1) In GARCH (1,1) we assign some weight to the long-run average variance rate
σ n2 = γ V L + α u n2−1 + βσ 2n −1 Since weights must sum to 1 γ + α + β =1
Daniel HERLEMONT
Page 6 6
GARCH (1,1) continued
Setting ω = γV the GARCH (1,1) model is
σ 2n = ω + α u n2−1 + βσ 2n −1 and
VL =
ω 1− α − β
Daniel HERLEMONT
Example
Suppose
σ n2 = 0.000002 + 0.13un2−1 + 0.86σ n2−1 The long-run variance rate is 0.0002 so that the longrun volatility per day is 1.4%
Daniel HERLEMONT
Page 7 7
Example continued
Suppose that the current estimate of the volatility is 1.6% per day and the most recent percentage change in the market variable is 1%. The new variance rate is 0.000002 + 013 . × 0.0001 + 0.86 × 0.000256 = 0.00023336
The new volatility is 1.53% per day
Daniel HERLEMONT
GARCH (p,q)
p
σ
2 n
=ω +
∑α u i
i =1
q
2 n−i
+
∑β σ j
2 n− j
j =1
Daniel HERLEMONT
Page 8 8
Other Models
We can design GARCH models so that the weight given to ui2 depends on whether ui is positive or negative We do not have to assume that the conditional distribution is normal (e.g. Student residuals)
Daniel HERLEMONT
Variance Targeting
One way of implementing GARCH(1,1) that increases stability is by using variance targeting We set the long-run average volatility equal to the sample variance Only two other parameters then have to be estimated
Daniel HERLEMONT
Page 9 9
Maximum Likelihood Methods
In maximum likelihood methods we choose parameters that maximize the likelihood of the observations occurring
Daniel HERLEMONT
Example 1
We observe that a certain event happens one time in ten trials. What is our estimate of the proportion of the time, p, that it happens? The probability of the outcome is
10p(1− p)9 We maximize this to obtain a maximum likelihood estimate: p=0.1
Daniel HERLEMONT
Page 10 10
Example 2
Estimate the variance of observations from a normal distribution with mean zero 1 − ui2 Maximize: ∏ exp 2v i = 1 2 πv n ui2 or: − ln( v ) − ∑ v i =1 1 n This gives: v = ∑ ui2 n i =1 n
Daniel HERLEMONT
Application to GARCH
We choose parameters that maximize
ui2 ∑ − ln(vi ) − vi i =1 n
Daniel HERLEMONT
Page 11 11
How Good is the Model?
The Ljung-Box statistic tests for autocorrelation We compare the autocorrelation of the ui2 with the autocorrelation of the ui2/σ σ i2 Is σi2 a predictor of ui+12 ? Perform a regression of σi2 versus ui+12
Daniel HERLEMONT
Forecasting Future Volatility A few lines of algebra shows that
E[σ 2n + k ] = VL + (α + β) k (σ 2n − VL ) The variance rate for an option expiring on day m is 1 m
m −1
∑ E [σ
2 n+k
]
k =0
Daniel HERLEMONT
Page 12 12
Correlations
Define ui=(Ui-Ui-1)/Ui-1 and vi=(Vi-Vi-1)/Vi-1 Also σu,n: daily vol of U calculated on day n-1 σv,n: daily vol of V calculated on day n-1 covn: covariance calculated on day n-1 σu,n σv,n) The correlation is covn/(σ
Daniel HERLEMONT
Correlations continued
Under EWMA β covn-1 covn = (1− (1−β) un-1vn-1+β Under GARCH (1,1) covn = ω + α un-1vn-1+β β covn-1
Daniel HERLEMONT
Page 13 13