Financial Risk Management
Introduction to Market Risk Management Following P. Jorion, Financial Risk Management Chapter 11
Daniel HERLEMONT
Old ways to measure risk notional amounts sensitivity measures (duration, Greeks) scenarios ALM, DFA assume linearity do not describe probability
Daniel HERLEMONT
Page 1 1
1938
Bonds duration
1952
Markowitz mean-variance
1963
Sharpe’s CAPM
1966
Multiple risk-factors
1973
Black-Scholes option pricing
1983
RAROC, risk adjusted return
1986
Limits on exposure by duration
1988
Risk-weighted assets for banks; exposure limits by Greeks
1993
VaR endorsed by G-30
1994
Risk Metrics
1997
CreditMetrics, CreditRisk+
Daniel HERLEMONT
How much can we lose?
Everything correct, but useless answer.
How much can we lose realistically?
Daniel HERLEMONT
Page 2 2
What is the current Risk?
Bonds
duration, convexity
Stocks
volatility
Options
delta, gamma, vega
Credit
rating
Forex
target zone
Total
?
Daniel HERLEMONT
Standard Approach
Daniel HERLEMONT
Page 3 3
Modern Approach
Financial Institution
Daniel HERLEMONT
Definition VaR is defined as the predicted worst-case loss at a specific confidence level (e.g. 99%) over a certain period of time.
Daniel HERLEMONT
Page 4 4
Definition (Jorion) VaR is the maximum loss over a target horizon such that there is a low, prespecified probability that the actual loss will be larger.
Daniel HERLEMONT
VaR 1 0.8 0.6 0.4
VaR1%
1%
0.2
Profit/Loss -3
-2
-1
1
2
3
Daniel HERLEMONT
Page 5 5
Meaning of VaR A portfolio manager has a daily VaR equal $1M at 99% confidence level. This means that there is only one chance in 100 that a daily loss bigger than $1M occurs,
under normal market conditions.
VaR 1%
Daniel HERLEMONT
Returns
year
1% of worst cases Daniel HERLEMONT
Page 6 6
Main Ideas
A few well known risk factors Historical data + economic views Diversification effects Testability Easy to communicate
Daniel HERLEMONT
History of VaR
80’s - major US banks - proprietary 93 G-30 recommendations 94 - RiskMetrics by J.P.Morgan 98 - Basel SEC, FSA, ISDA, pension funds, dealers Widely used and misused!
Daniel HERLEMONT
Page 7 7
FRM-99, Question 89
What is the correct interpretation of a $3 overnight VaR figure with 99% confidence level? A. expect to lose at most $3 in 1 out of next 100 days B. expect to lose at least $3 in 95 out of next 100 days C. expect to lose at least $3 in 1 out of next 100 days D. expect to lose at most $6 in 2 out of next 100 days
Daniel HERLEMONT
VaR caveats VaR does not describe the worst loss VaR does not describe losses in the left tail VaR is measured with some error
Daniel HERLEMONT
Page 8 8
Other Measures of Risk The expected left tail loss
The standard deviation
The semi-standard deviation
Daniel HERLEMONT
Risk Measures 1 0.8 0.6 0.4 0.2
Profit/Loss -3
-2
-1
1
2
3
Daniel HERLEMONT
Page 9 9
Properties of Risk Measure Monotonicity (XR(Y)) Translation invariance R(X+k) = R(X)-k Homogeneity R(aX) = a R(X) (liquidity??) Subadditivity R(X+Y) ≤ R(X) + R(Y) the last property is violated by VaR!
Daniel HERLEMONT
No subadditivity of VaR
Daniel HERLEMONT
Page 10 10
Daniel HERLEMONT
FRM-98, Question 22 Consider arbitrary portfolios A and B and their combined portfolio C. Which of the following relationships always holds for VaRs of A, B, and C? A. VaRA+ VaRB = VaRC B. VaRA+ VaRB ≥ VaRC C. VaRA+ VaRB ≤ VaRC D. None of the above
Daniel HERLEMONT
Page 11 11
Confidence level low confidence leads to an imprecise result. For example 99.99% and 10 business days will require history of 100*100*10 = 100,000 days in order to have only 1 point.
Daniel HERLEMONT
Time horizon long time horizon can lead to an imprecise result. 1% - 10 days means that we will see such a loss approximately once in 100*10 = 3 years. 5% and 1 day horizon means once in a month. Various subportfolios may require various horizons.
Daniel HERLEMONT
Page 12 12
Time horizon When the distribution is stable one can translate VaR over different time periods.
VaR (T days ) = VaR (1 day ) T This formula is valid (in particular) for iid normally distributed returns.
Daniel HERLEMONT
FRM-97, Question 7 To convert VaR from a one day holding period to a ten day holding period the VaR number is generally multiplied by: A. 2.33 B. 3.16 C. 7.25 D. 10
Daniel HERLEMONT
Page 13 13
Basel Rules horizon of 10 business days 99% confidence interval an observation period of at least a year of historical data, updated once a quarter
Daniel HERLEMONT
Basel Rules MRC Market Risk Charge = MRC SRC - specific risk charge, k ≥3.
k 60 MRCt = Max ∑ VaRt −i , VaRt −1 + SRCt 60 i =1 VaRt = VaRt (1d , 99%) × 10 Daniel HERLEMONT
Page 14 14
FRM-97, Question 16 Which of the following quantitative standards is NOT required by the Amendment to the Capital Accord to Incorporate Market Risk? A. Minimum holding period of 10 days B. 99% one-tailed confidence interval C. Minimum historical observations of two years D. Update the data sets at least quarterly
Daniel HERLEMONT
VaR system
Risk factors
Portfolio
Historical data
positions
Model
Mapping
Distribution of risk factors
VaR method
Exposures
VaR
Daniel HERLEMONT
Page 15 15
FRM-97, Question 23 The standard VaR calculation for extension to multiple periods also assumes that positions are fixed. If risk management enforces loss limits, the true VaR will be: A. the same B. greater than calculated C. less then calculated D. unable to determine
Daniel HERLEMONT
FRM-97, Question 9 A trading desk has limits only in outright foreign exchange and outright interest rate risk. Which of the following products can not be traded within the current structure? A. Vanilla IR swaps, bonds and IR futures B. IR futures, vanilla and callable IR swaps C. Repos and bonds D. FX swaps, back-to-back exotic FX options
Daniel HERLEMONT
Page 16 16
Stress-testing scenario analysis stressing models, volatilities and correlations developing policy responses
Daniel HERLEMONT
Scenario Analysis Moving key variables one at a time Using historical scenarios Creating prospective scenarios The goal is to identify areas of potential vulnerability.
Daniel HERLEMONT
Page 17 17
FRM-97, Question 4 The use of scenario analysis allows one to: A. assess the behavior of portfolios under large moves B. research market shocks which occurred in the past C. analyze the distribution of historical P&L D. perform effective back-testing
Daniel HERLEMONT
FRM-98, Question 20 VaR measure should be supplemented by portfolio stress-testing because: A. VaR measures indicate that the minimum is VaR, they do not indicate the actual loss B. stress testing provides a precise maximum loss level C. VaR measures are correct only 95% of time D. stress testing scenarios incorporate reasonably probable events.
Daniel HERLEMONT
Page 18 18
FRM-00, Question 105 VaR analysis should be complemented by stress-testing because stress-testing: A. Provides a maximum loss in dollars. B. Summarizes the expected loss over a target horizon within a minimum confidence interval. C. Assesses the behavior of portfolio at a 99% confidence level. D. Identifies losses that go beyond the normal losses measured by VaR.
Daniel HERLEMONT
Page 19 19