1. Market Risk Modelling By A.V. Vedpuriswar July 31, 2009

2. 1 Volatility

3. 2 Basics of volatility  Volatility is a huge issue in risk management.  Volatility is a key parameter in modelling market risk  The science of volatility measurement has advanced a lot in recent years.  Here we look at some basic concepts and tools.

4. 33 Estimating Volatility  Calculate daily return u 1 = ln S i / S i-1  Variance rate per day  We can simplify this formula by making the following simplifications. u i = (S i – S i-1) / S i-1 ū = 0m-1 = m If we want to weight

5. 44 Estimating Volatility  Exponentially weighted moving average model means weights decrease exponentially as we go back in time.  n 2 = 2 n-1 + (1 - ) u 2 n-1 = [  n-2 2 + (1- )u n-2 2 ] + (1- )u n-1 2 = (1- )[u n-1 2 + u n-2 2 ] + 2  n-2 2 = (1- ) [u n-1 2 + u 2 n-2 + 2 u n-3 2 ] + 3  2 n-3  If we apply GARCH model,  n 2 = Y V L +  u n-1 2 +  2 n-1 V L = Long run average variance rate Y +  +  = 1.If Y = 0,  = 1-,  =, it becomes exponentially weighted model.  GARCH incorporates the property of mean reversion.

6. 55 Problem  The current estimate of daily volatility is 1.5%. The closing price of an asset yesterday was $30. The closing price of the asset today is $30.50. Using the EWMA model, with λ = 0.94, calculate the updated estimate of volatility.

7. 66 Solution h t = λ σ 2 t-1 + ( 1 – λ) r t-1 2  λ=.94  r t-1 =ln[(30.50 )/ 30] =.0165  h t =(.94) (.015) 2 + (1-.94) (.0165) 2  Volatility =.01509 = 1.509 %

8. 7 Greeks

9. Introduction  Greeks help us to measure the risk associated with derivative positions.  Greeks also come in handy when we do local valuation of instruments.  This is useful when we calculate value at risk. 8

10. 99 Delta  Delta is the rate of change in option price with respect to the price of the underlying asset.  It is the slope of the curve that relates the option price to the underlying asset price.  A position with delta of zero is called delta neutral.  Delta keeps changing.  So the investor’s position may remain delta neutral for only a relatively short period of time.  The hedge has to be adjusted periodically.  This is known as rebalancing.

11. 10 Gamma  The gamma is the rate of change of the portfolio’s delta with respect to the price of the underlying asset.  It is the second partial derivative of the portfolio price with respect to the asset price.  If gamma is small, it means delta is changing slowly.  So adjustments to keep a portfolio delta neutral can be made only relatively infrequently.  However, if gamma is large, it means the delta is highly sensitive to the price of the underlying asst.  It is then quite risky to leave a delta neutral portfolio unchanged for any length of time.

12. 11 Theta  Theta of a portfolio is the rate of change of value of the portfolio with respect to change of time.  Theta is also called the time decay of the portfolio.  Theta is usually negative for an option.  As time to maturity decreases with all else remaining the same, the option loses value.

13. 12 Vega  The Vega of a portfolio of derivatives is the rate of change of the value of the portfolio with respect tothe volatility of the underlying asset.  High Vega means high sensitivity to small changes in volatility.  A position in the underlying asset has zero Vega.  The Vega can be changed by adding options.  If V is Vega of the portfolio and V T is the Vega of the traded option, a position of –V/ V T in the traded option makes the portfolio Vega neutral.  If a hedger requires the portfolio to be both gamma and Vega neutral, at least two traded derivatives dependent on the underlying asset must usually be used.

14. 13 Rho  Rho of a portfolio of options is the rate of change of value of the portfolio with respect to the interest rate.

15. 14 Problem  Suppose an existing short option position is delta neutral and has a gamma of － 6000. Here, gamma is negative because we have sold options. Assume there exists a traded option with a delta of 0.6 and gamma of 1.25. Create a gamma neutral position.

16. 15 Solution  To gamma hedge, we must buy 6000/1.25 = 4800 options.  Then we must sell (4800) (.6) = 2880 shares to maintain a gamma neutral and original delta neutral position.

17. 16 Problem  A delta neutral position has a gamma of － 3200. There is an option trading with a delta of 0.5 and gamma of 1.5. How can we generate a gamma neutral position for the existing portfolio while maintaining a delta neutral hedge?

18. 17 Solution  Buy 3200/1.5=2133 options  Sell (2133) (.5)=1067 shares

19. 18  Suppose a portfolio is delta neutral, with gamma = - 5000 and vega = - 8000. A traded option has gamma =.5, vega = 2.0 and delta = 0.6. How do we achieve vega neutrality?  Problem

20. 19  To achieve Vega neutrality we can add 4000 options.  Delta increases by (.6) (4000) = 2400  So we sell 2400 units of asset to maintain delta neutrality.  As the same time, Gamma changes from – 5000 to ((.5) (4000) – 5000 = - 3000.

21. 20  Suppose there is a second traded option with gamma = 0.8, vega = 1.2 and delta = 0.5.  if w 1 and w 2 are the weights in the portfolio,  - 5000 +.5w 1 +.8w 2 = 0 - 8000 + 2.0w 1 + 1.2w 2 = 0  w 1 = 400 w 2 = 6000.  This makes the portfolio gamma and vega neutral.  Now let us examine delta neutrality.  Delta = (400) (.6) + (6000) (.5) = 3240  3240 units of the underlying asset will have to be sold to maintain delta neutrality. 

22. 21 Value at Risk

23. Introduction  Value at Risk (VAR) is probably the most important tool for measuring market risk.  VAR tells us the maximum loss a portfolio may suffer at a given confidence interval for a specified time horizon.  If we can be 95% sure that the portfolio will not suffer more than $ 10 million in a day, we say the 95% VAR is $ 10 million. 22

24. 23  Average revenue = $5.1 million per day  Total no. of observations = 254.  Std dev = $9.2 million  Confidence level = 95%  No. of observations < - $10 million = 11  No. of observations < - $ 9 million = 15 Illustration

25.  Find the point such that the no. of observations to the left = (254) (.05) = 12.7  (12.7 – 11) /( 15 – 11 )=1.7 / 4≈.4  So required point = - (10 -.4)=- $9.6 million  VAR = E (W) – (-9.6) = 5.1 – (-9.6) = $14.7 million  If we assume a normal distribution,  Z at 95% ( one tailed) confidence interval = 1.645  VAR = (1.645) (9.2) =$ 15.2 million

26. 25 Problem  The VAR on a portfolio using a one day horizon is USD 100 million. What is the VAR using a 10 day horizon ?

27. 26 Solution  Variance scales in proportion to time.  So variance gets multiplied by 10  And std deviation by √10  VAR = 100 √10 = (100) (3.16) =316  ( σ N 2 = σ 1 2 + σ 2 2 ….. = Nσ 2 )

28. 27 Problem  If the daily VAR is $12,500, calculate the weekly, monthly, semi annual and annual VAR. Assume 250 days and 50 weeks per year.

29. 28 Solution Weekly VAR ＝ (12,500) (√5) ＝ 27,951 Monthly VAR ＝ ( 12,500) (√20) ＝ 55,902 Semi annual VAR ＝ (12,500) (√125) ＝ 139,754 Annual VAR ＝ (12,500) (√250) ＝ 197,642

30. 29 Variance Covariance Method

31. 30 Problem  Suppose we have a portfolio of $10 million in shares of Microsoft. We want to calculate VAR at 99% confidence interval over a 10 day horizon. The volatility of Microsoft is 2% per day. Calculate VAR.

32. 31 Solution  σ= 2% =(.02)(10,000,000) = $200,000  Z (P =.01) =Z (P =.99) = 2.33  Daily VAR =(2.33) (200,000) = $ 466,000  10 day VAR = 466,000 √10 = $ 1,473,621 Ref : Options, futures and other derivatives, By John Hull

33. 32 Problem  Consider a portfolio of $5 million in AT&T shares with a daily volatility of 1%. Calculate the 99% VAR for 10 day horizon.

34. 33 Solution  σ =1%=(.01) (5,000,000)= $ 50,000  Daily VAR=(2.33) (50,000)= $ 116,500  10 day VAR=$ 111,6500 √10= $ 368,405

35. 34 Problem  Now consider a combined portfolio of AT&T and Microsoft shares. Assume the returns on the two shares have a bivariate normal distribution with the correlation of 0.3. What is the portfolio VAR.?

36. 35 Solution  σ 2 =w 1 2 σ 1 2 + w 2 2 σ 2 2 + 2 ῤPw 1 W 2 σ 1 σ 2  =(200,000) 2 + (50,000) 2 + (2) (.3) (200,000) (50,000)  σ =220,277  Daily VAR = (2.33) (220,277) = 513,129  10 day VAR = (513,129) √10=$1,622,657  Effect of diversification = (1,473,621 + 368,406) – (1,622,657) = 219,369

37. 36 Monte Carlo Simulation

38. 37 What is Monte Carlo VAR?  The Monte Carlo approach involves generating many price scenarios (usually thousands) to value the assets in a portfolio over a range of possible market conditions.  The portfolio is then revalued using all of these price scenarios.  Finally, the portfolio revaluations are ranked to select the required level of confidence for the VAR calculation.

39. 38 Step 1: Generate Scenarios  The first step is to generate all the price and rate scenarios necessary for valuing the assets in the relevant portfolio, as well as the required correlations between these assets.  There are a number of factors that need to be considered when generating the expected prices/rates of the assets: – Opportunity cost of capital – Stochastic element – Probability distribution

40. 39 Opportunity Cost of Capital  A rational investor will seek a return at least equivalent to the risk-free rate of interest.  Therefore, asset prices generated by a Monte Carlo simulation must incorporate the opportunity cost of capital.

41. 40 Stochastic Element  A stochastic process is one that evolves randomly over time.  Stock market and exchange rate fluctuations are examples of stochastic processes.  The randomness of share prices is related to their volatility.  The greater the volatility, the more we would expect a share price to deviate from its mean.

42. 41 Probability Distribution  Monte Carlo simulations are based on random draws from a variable with the required probability distribution, usually the normal distribution.  The normal distribution is useful when modeling market risk in many cases.  But it is the returns on asset prices that are normally distributed, not the asset prices themselves.  So we must be careful while specifying the distribution.

43. 42 Step 2: Calculate the Value of the Portfolio  Once we have all the relevant market price/rate scenarios, the next step is to calculate the portfolio value for each scenario.  For an options portfolio, depending on the size of the portfolio, it may be more efficient to use the delta approximation rather than a full option pricing model (such as Black-Scholes) for ease of calculation.  Δoption = Δ(ΔS)  Thus the change in the value of an option is the product of the delta of the option and the change in the price of the underlying.

44. 43 Other approximations  There are also other approximations that use delta, gamma (Γ) and theta (Θ) in valuing the portfolio.  By using summary statistics, such as delta and gamma, the computational difficulties associated with a full valuation can be reduced.  Approximations should be periodically tested against a full revaluation for the purpose of validation.  When deciding between full or partial valuation, there is a trade- off between the computational time and cost versus the accuracy of the result.  The Black-Scholes valuation is the most precise, but tends to be slower and more costly than the approximating methods.

45. 44 Step 3: Reorder the Results  After generating a large enough number of scenarios and calculating the portfolio value for each scenario: – the results are reordered by the magnitude of the change in the value of the portfolio (Δportfolio) for each scenario – the relevant VAR is then selected from the reordered list according to the required confidence level  If 10,000 iterations are run and the VAR at the 95% confidence level is needed, then we would expect the actual loss to exceed the VAR in 5% of cases (500).  So the 501st worst value on the reordered list is the required VAR.  Similarly, if 1,000 iterations are run, then the VAR at the 95% confidence level is the 51st highest loss on the reordered list.

46. 45 Formula used typically in Monte Carlo for stock price modelling

47. 46 Advantages of Monte Carlo  This method can cope with the risks associated with non- linear positions.  We can choose data sets individually for each variable.  This method is flexible enough to allow for missing data periods to be excluded from the VAR calculation.  We can incorporate factors for which there is no actual historical experience.  We can estimate volatilities and correlations using different statistical techniques.

48. 47 Problems with Monte Carlo  Cost of computing resources can be quite high.  Speed can be slow.  Random Numbers may not be all that random.  Pseudo random numbers are only a substitute for true random numbers and tend to show clustering effects.  Quasi-Monte Carlo techniques have been developed to produce quasi-random numbers that are more uniformly spaced.

49. 48  Monte Carlo is based on random draws from a variable with the required probability distribution, often normal distribution.  As with the variance-covariance approach, the normal distribution assumption can be problematic.  Monte Carlo can however, be performed with alternative distributions.  Model risk is the risk of loss arising from the failure of a model to sufficiently match reality, or to otherwise deliver the required results.  For Monte Carlo simulations, the results (value at risk estimate) depend critically on the models used to value (often complex) financial instruments.

50. 49 Historical Simulation

51. 50 Introduction  Historical simulation is one of the three most common approaches used to calculate value at risk.  Unlike the Monte Carlo approach, it uses the actual historical distribution of returns to simulate the VAR of a portfolio.  Use of real data, coupled with ease of implementation, has made historical simulation a very popular approach to estimating VAR.

52. 51 Few assumptions  Historical simulation avoids the assumption that returns on the assets in a portfolio are normally distributed.  Instead, it uses actual historical returns on the portfolio assets to construct a distribution of potential future portfolio losses.  From this distribution, the VAR can be read.  This approach requires minimal analytics.  All we need is a sample of the historic returns on the portfolio whose VAR we wish to calculate.

53. 52 Steps  Collect data  Generate scenarios  Calculate portfolio returns  Arrange in order.

54. 53 Problem % ReturnsFrequencyCumulative Frequency - 1611 - 1412 - 1013 - 725 - 516 - 439 - 3110 - 1212 0315 1116 2218 4119 6120 7121 8122 9123 11124 12126 14227 18128 21129 23130 What is VAR (90%) ?

55. 54  10% of the observations, i.e, (.10) (30) = 3 lie below -7 So VAR = -7 Solution

56. 55 Advantages  Simple  No normality assumption  Non parametric

57. 56 Disadvantages  Reliance on the past  Length of estimation period  Weighting of data  Data issues

58. 57 Comparison of different VAR modeling techniques

59. 58 Simulation vs Variance Covariance  Simulation approaches are preferred by global banks due to: – flexibility in dealing with the ever-increasing range of complex instruments in financial markets – the advent of more efficient computational techniques in recent years – the falling costs in information technology  However, the variance-covariance approach might be the most appropriate method for many smaller firms, particularly when : – they do not have significant options positions – they prefer to outsource the data requirement component of their risk calculations to a company such as RiskMetrics – significant savings can often be made by using outsourced volatility and correlation data, compared to internally storing the daily price histories required for simulation techniques

60. 59 Model Validation

61. 60 Basel Committee Standards  Banks that prefer to use internal models must meet, on a daily basis, a capital requirement that is the higher of either: – the previous day's value at risk – the average of the daily value at risk of the preceding 60 business days multiplied by a minimum factor of three  VAR must be computed on a daily basis.  A one-tailed confidence interval of 99% must be used.  The minimum holding period should be 10 trading days.  The minimum historical observation period should be one year.

62. 61  Banks should update their data sets at least once every three months.  Banks can recognize correlations within broad risk categories.  Provided the relevant supervisory authority is satisfied with the bank's system for measuring correlations, they may also recognize correlations across broad risk factor categories.  Banks' internal models are required to accurately capture the unique risks associated with options and option-like instruments.  The Basel Committee has also specified qualitative factors that banks must meet before they are permitted to use internal models.

63. 62  The Basel Committee prescribes an increase in capital requirements if, based on a sample of 250 observations (a one-year observation period), the VAR model underpredicts the number of exceptions (losses exceeding the 99% confidence level).  For such purposes, three 'zones' have been distinguished by the Committee.  Green Zone : 0-4 exceptions  Yellow zone : 5-9 exceptions  Red zone : 10 or more exceptions

64. 63 Stress Testing

65. 64 Introduction  Stress testing involves analysing the effects of exceptional events in the market on a portfolio's value.  These events may be exceptional, but they are also plausible.  And their impact can be severe.  Historical scenarios or hypothetical scenarios can be used.

66. 65 Two approaches to Stress testing  Single-factor stress testing (sensitivity testing) involves applying a shift in a specific risk factor to a portfolio in order to assess the sensitivity of the portfolio to changes in that risk factor.  Multiple-factor stress testing (scenario analysis) involves applying simultaneous moves in multiple risk factors to a portfolio to reflect a risk scenario or event that looks plausible in the near future.

67. 66 Conducting Stress Tests  From a computational viewpoint, stress testing can be thought of as a variant of simulation methods.  It merely uses a different technique to generate scenarios.  Once scenarios have been developed, the next step is to analyze the effect of each scenario on portfolio value.  This can sometimes be done in the same way as a simulation to calculate VAR.  Stress tests can typically be run by inputting the stressed values of the risk factors into existing models and recalculating the portfolio value using the new data.

68. 67 Extreme Value Theory  EVT is a branch of statistics dealing with the extreme deviations from the mean of statistical distributions.  The key aspect of EVT is the extreme value theorem.  According to EVT, given certain conditions, the distribution of extreme returns in large samples converges to a particular known form, regardless of the initial or parent distribution of the returns.

69. 68 EVT Parameters  This distribution is characterized by three parameters – location, scale and shape (tail).  The tail parameter is the most important as it gives an indication of the heaviness (or fatness) of the tails of the distribution.  The EVT approach is very useful because the distributions from which return observations are drawn are very often unknown.  EVT does not make strong assumptions about the shape of this unknown distribution.