APPENDIX: Example calibration to market
implied volatilities using the analytical weighted Monte Carlo approach
[this page | pdf | references | back links]
Return
to Abstract and Contents
Next
Section
APPENDIX: Example calibration
to market implied volatilities using the analytical weighted Monte Carlo approach
As at some date in the past a leading commercial vendor’s
risk system included the following standard deviations and correlations within
its covariance matrix risk model for the following four UK equities:
|
Security
|
Standard Deviation (%pa)
|
Correlation with
|
|
1
|
0.29
|
0.33
|
0.30
|
|
AL
|
16.92
|
0.29
|
1
|
0.35
|
0.29
|
|
BLT
|
28.56
|
0.33
|
0.35
|
1
|
0.45
|
|
AVZ
|
36.64
|
0.30
|
0.29
|
0.45
|
1
|
|
BAY
|
32.85
|
1
|
0.29
|
0.33
|
0.30
|
The predicted tracking error of a model portfolio consisting
of 35% Alliance and Leicester, 35% BHP Billiton, 15% Amvescap and 15% British
Airways versus an equally weighted benchmark of these four stocks using this
risk model was 5.35%pa.
As at approximately the same date the (at-the-money) implied
volatilities for (call) options on these equities were approximately as follows:
|
Security
|
Implied Volatility (%pa)
|
|
AL
|
22
|
|
BLT
|
31
|
|
AVZ
|
30
|
|
BAY
|
27
|
Calibrating simultaneously to these four pieces of market information
using the analytical weighted Monte Carlo approach gives a calibrated
covariance matrix as shown below. The calibrated volatility of each individual
security now matches its implied volatility. There are also some changes to
individual correlations. The tracking error of the model portfolio using this
calibrated covariance matrix is 4.77%pa.
|
Security
|
Standard Deviation (%pa)
|
Correlation with
|
|
AL
|
BLT
|
AVZ
|
BAY
|
|
AL
|
22
|
1
|
0.35
|
0.33
|
0.30
|
|
BLT
|
31
|
0.35
|
1
|
0.33
|
0.26
|
|
AVZ
|
30
|
0.33
|
0.33
|
1
|
0.36
|
|
BAY
|
27
|
0.30
|
0.26
|
0.36
|
1
|
If we were also to calibrate to an ‘index’ implied
volatility, then the individual security implied volatilities would remain the
same, since they are already calibrated to the market. What instead would
happen is that correlations between securities would change. There is no listed
option relating to a basket of these four stocks. If the above model fully
matched market implieds then the ‘index’ implied volatility would be 19.3%pa.
But suppose that there was an observable ‘index’ implied volatility and that it
was 15%pa. The covariance matrix would then be as follows. The tracking error
of the model portfolio would increase to 5.73%pa.
|
Security
|
Standard Deviation (%pa)
|
Correlation with
|
|
AL
|
BLT
|
AVZ
|
BAY
|
|
AL
|
22
|
1
|
0.13
|
0.09
|
0.06
|
|
BLT
|
31
|
0.13
|
1
|
0.03
|
-0.05
|
|
AVZ
|
30
|
0.09
|
0.03
|
1
|
0.12
|
|
BAY
|
27
|
0.06
|
-0.05
|
0.12
|
1
|
Conversely, if our ‘index’ basket actually had an implied
volatility of 21%pa then the covariance matrix would become:
|
Security
|
Standard Deviation (%pa)
|
Correlation with
|
|
AL
|
BLT
|
AVZ
|
BAY
|
|
AL
|
22
|
1
|
0.46
|
0.45
|
0.41
|
|
BLT
|
31
|
0.46
|
1
|
0.46
|
0.40
|
|
AVZ
|
30
|
0.45
|
0.46
|
1
|
0.47
|
|
BAY
|
27
|
0.41
|
0.40
|
0.47
|
1
|
The average correlation between individual securities is
quite sensitive to divergent movements between index implied volatility and
average single security implied volatility, as is the tracking error of the
model portfolio, which would now be 4.30%pa.
As we might also expect, each individual security
calibration point disproportionately affects the volatility of that particular
security. For example, suppose we only calibrated the original risk model to
one implied volatility, namely the one for Amvescap. The calibrated
volatilities would then be as follows:
|
|
Volatility prior to calibration (%pa)
|
Volatility post calibration (%pa)
|
Change
|
|
AL
|
16.92
|
16.62
|
-2%
|
|
BLT
|
28.56
|
27.97
|
-2%
|
|
AVZ
|
36.64
|
30.00
|
-18%
|
|
BAY
|
32.85
|
31.75
|
-3%
|
NAVIGATION LINKS
Contents | Prev