際際滷

際際滷Share a Scribd company logo

Intra Horizon Risk 2010

Download as PPTX, PDF
Y
yamanote

Looking at the risk within an investment period rather than just at the end of the period. Evaluating the effect on VaR of non-Normal distributions, jumps, and drift

Convert to study guideBETA

Transform any presentation into a summarized study guide, highlighting the most important points and key insights.

1 of 42
Downloaded 10 times
Intra-Horizon RiskNick WadeSeptember 2010
OverviewCriticisms of tracking error / VaRTail shapeTime dimensionIntroduction of first-passage probabilityDerivation in Normal, no drift caseSidetrack 1: time-diversification argumentPossible ExtensionsDrift, jumps, stochastic volatilityIntroduce Jump ModelsReview three common jump models, empirical evidenceSidetrack 2: predictability of jumpsSolving first-passage probability for jump modelsEmpirical VaR intra-horizon versus end-of-horizon for three jump modelsConclusion:Implementable intra-horizon VaR / t.e. measure under assumptionsRoadmap for intra-horizon risk under more flexible distributions
Criticisms of VaR/tracking errorTail Shape: says nothing about size of loss, only probability of lossTime Dimension: says nothing about losses prior to horizon, recovery etcDistributional assumptions: doesnt allow for jumps, non-Normal behaviorTime-scaling issues
Tail shapeVaR result is independent of tail shapeFor measures that specifically include tail risk see:Tail-conditional expectation, worst conditional expectationArtzner, Delbaen, Eber and Heath (1997, 1999)Expected shortfall, CVaR etcAcerbi, Tasche (2001)
Time DimensionVaR/tracking error says nothing about the profit and loss distribution over the trading horizonBasel Committee (1996)Denote intra-horizon riskLargely overlooked topic, few papers:Stultz (1996)Kritzman and Rich (2002)Boudoukh, Richardson, Stanton, Whitelaw (2004)Restrictive Assumptions: Brownian motion, no jumpsNote: Basel Committee (1996) also recognizes jumps are relevant for risk management purposes.Bakshi, Panayatov (forthcoming JFE)
MotivationReturnRisk BandTimeTracking error or VaR are concerned about the likely distribution of returns at the end of some investment period. In this example we have scored a goal the final return is within the risk bands we set ourselves.
Motivation IIReturnRisk BandTimeSo FiredDifficult MeetingNice Bonus
Another ViewReturnRisk BandTimeIn this example there are five return paths. Only one path ends up outside our risk bands (and that on the right side) so by EOH measures the probability of breaching the band is 1/5. If we look throughout the horizon, 3 out of 5 paths breach the barrier at some point during the horizon and hence the intra-horizon probability of breach is 3/5
Tracking Error, VaREnd of Horizon MeasureCharacterizes the distribution of returns at the end of an investment periodSays nothing about:Maximum loss at some probability during the periodProbability of recovery following a loss during the periodor anything else about the return path
Risk horizon should match investment horizonIt should be obvious, but probably isnt, that our forecast horizon for our risk model should match our forecast horizon for our alphas, and implicitly therefore our holding periodRosenberg & Guy (1975)What Rosenberg neglects is the very human characteristic of second-guessing the process mid-period & the idea that we might be leveraged and therefore unable to put our feet on the wheel and wait for it all to turn out right in the end
Who Cares?Survival Analysis Leveraged Investor: if there is a floor below which I cannot go, I should care greatly what the probability of hitting that floor is during the periodMonitored Investor if there is a good chance I will lose a client if performance is worse than a particular hurdle at any time during the year, I should care about the probability of hitting that hurdle
Who Cares IILoan agreement collateralSecurities lending collateralCapital adequacyRetirement/Endowment funding a liability intra-horizon
Focus for TodayLooking at the time-dimensionWhat happens to the value of a position/portfolio during an investment horizonCompare results to standard end-of-horizon measureWe are using end-of-horizon VaR calculated using the Normal distribution, zero drift, and no jumps as our reference point. All multipliers are with respect to that measure.
Not a new issueWe can think about the issue in two ways:The size of a loss given a probability (i.e. VaR)Turn that on its head: the probability of hitting a given loss level (barrier)Barrier Options since 1960sMerton, Reiner, Rubenstein (1973)analytical closed form pricing for barrier options(same problem: need to figure out probability of hitting a barrier during an investment horizon)Hirsa and Madan (2003)Kudryavtsev and Levendorskii (2006)
Concept First-PassageBy using the concept from statistics of first-passage time (also known as first-hitting time) we can estimate the probability that our return will hit a particular hurdle e.g. -10% during the investment period, or estimate the maximum loss $X during the period at some given confidence level e.g. 99%
CaveatsThis does nothing to fix the issue of strategy risk drift in the mean - to be discussed separately todayThis measure is equally exposed to estimation error as with the regular tracking error / VaR measure[and equally benefits just the same from measures to address estimation error]We are saying nothing about asymmetry of preferencesWe are staying away from saying anything about benchmarks being suitable or relative versus absolute riskTo first get an intuitive tractable solution we assume Normality and zero drift, although we later relax those assumptions
Conclusions First(After this slide you can check your Blackberry and not miss anything important)Intra-Horizon VaR (VaR-I) is larger than End-of-Horizon VaRNormal, no jumps: VaR-I = 1.107 * VaR1Non-Normal, allowing jumps: that multiplier can be up to 2.642Non-Normal, jumps, stochastic vol, drift it can get worse1 Feller (1971)2Bakshi and Panayatov (forthcoming, JFE)
Probability of loss at EOHAssumptionsNormalStationary voli.i.d.(so we can use sqrt T)This is just the difference between the cumulative % loss and the cumulative expected return divided by the cumulative standard deviation, then apply N() to convert this standardized distance from the mean into a probability estimate.(Note I am cheating already and including a drift term)
VaRBy rearranging the previous equation and solving for L (the loss) we get:
Probability of Loss Intra HorizonSecond part never zero or negativeImplies IH loss estimate > EOH estimate (always)IH P(loss) rises as investment horizon expands, whereas EOH P(loss) declines as investment horizon expands=> time-diversification argument
[Side Track 1]: Time DiversificationSamuelson (1963) argued against time diversification:Although EOH p(loss) declines as inv hor incr., that benefit is offset by increasing magnitude of potential lossOthers have argued that time diversifies risk because the result does not depend on the magnitude of the potential lossWH risk presents another facet to that argument:Risk increases even if investors only care about probability of loss
Solving the ProblemFirst show path-independent (i.e. end of horizon) probability of lossThen use principle of reflection to convert path-dependent (i.e. within horizon) probability of loss into an equivalent path-independent (i.e. end of horizon) measureReview implications and assumptions
Path-independent probability of return r after time T Assume continuous returns are normally distributed:Probability that cumulative return = rWhere:n is the number of years in Tr is the continuous cumulative return is the standard deviation of the continuous returns
Principal of ReflectionAReturnBTimeInvoking the principal of reflection; the path-dependent probability of breaching a barrier and subsequently recovering to A [solid line] is the same as the path-dependent probability of breaching the barrier and then descending to B, which is the same as the path-independent probability of ending up at B, because B cannot be reached without breaching the barrier at some point.
The Main PointJoint Prob (breach barrier b, and recover to A) = prob (end at B)The reflection principle allows us to convert a path-dependent probability into a path-independent probability under some assumptions
Set up the probabilitiesProb (breach barrier) = 1 prob( never breach)= 1 joint prob(end up above, and min always above)= 1 prob(end above) + joint prob(end above, and min <=barrier)= prob(end below) + joint prob(end above, and min <=barrier)By principal of reflection: the last term is equal to the path independent probability of reaching a return r below the barrier
Assumptions so farZero driftNormal distribution / Brownian motionStationary volatilityNo jumps
Implications so farWithin-horizon risk increases with time, adding fuel to the debate against time-diversification, because second term always positiveEven assuming Normal distribution, stationary volatility, no jumps, and no drift, the 1% VaR-I is 10.7% bigger than 1% VaRFeller (1971)At 2.5% its 14.4% bigger, and at 5% its 19.2% biggerBoudoukh, Richardson, Stanton, Whitelaw (2004)
Relaxing the zero drift assumptionGirsanovs theoremUseful transformation: using Girsanovs theorem we can transform a Brownian motion with drift into a Brownian motion without drift to make the maths easier.The first-passage probability for a Brownian motion with drift is well known. See for example Karatzas and Shreve (1991)
The Impact of DriftFor comparative purposes, look at what happens when we introduce drift of 10% and 15% respectively. Boudoukh, Richardson, Stanton, Whitelaw (2004)
What about jumps?Basel 1996: recognizes jumps are importantother evidence:Evidence for jumps:Bakshi, Cao, Chen (1997)Bates (2000)Anderson, Benzoni, Lund (2002)but difficulty explaining market crashes with return jumps and diffusive volatilityEvidence for large jump risk premiaPan (2002)Evidence for jumps in return and in volatilityEraker, Johannes, Polson (2003)For two examples of models with return and volatility jumps seeDuffie, Pan, Singleton (2000)
Jump ModelsJump processes:Merton (1976)Carr, Geman, Madan, Yor (2002)Carr and Wu (2003)Two groups:finite jumps (i.e. infrequent)infinite jumps (i.e. lots)
Sidetrack 2: predictability of jumpsStrong evidence for predictability of jumps:(For US markets) the VIX can predict both jump arrival and sizeJohannes, Kumar, Polson (1999)Strong evidence of jump clusteringImplies that Merton (1976) model is not useful as it cannot capture persistence in the data i.e. jump clustering
The ApproachSome analytical characterizations are known, particularly for stable processes with one-sided jumps:Kyprianou (2006)In general, need to resort to PIDE approach [partial integro-differential equations]Solve using finite difference methodsOr Monte CarloAtiya & Metwally (2002) very fast approach, around 100 times faster than pure MC by leveraging a Brownian bridge to reduce the number of points that need to be calculated.Beyond this point I am out of my depth and will be visible only by bubbles on the surface
Some resultsCompared to standard Normal VaR JD = Mertons jump-diffusion model CGMY is the two-sided pure-jump Levy model of Carr, Geman, Madan, and Yor FMLS is the finite-moment log-stable model of Carr and Wu(Bakshi and Panayatov)
ImplicationsVaR with jumps is bigger than N() VaRWe suspected as much beforehandThe choice of model makes a differenceBut do they all fit the data equally well?VaR-I is consistently greater than VaRWe knew that beforehand, based on the known result with the Normal, no drift caseVaR-I can be more than double standard VaRAnd this goes some way toward justifying the Basel multipliers
Stochastic Volatility and JumpsJust a tasteStochastic volatility in a VaR context:Eberlein, Kallsen, and Kristen (2002)Two-dimensional PIDEs for models with both jumps and stochastic volatility:Feng and Linetsky (2006)Can be applied to first-passage calculations.(let me know how you get on)
Related MeasuresLeaning heavily once again on the barrier-option analogy we could also consider:A risk range rather than a single loss level (double barrier)A time-varying barrier or rangeA barrier that varies as a function of volatilityE.g. Lo & Hui (2007), pricing double barriers where volatility, dividend yield, and the barriers are stochastic.
ConclusionsIntra-Horizon Risk is important, and neglectedIf you are a buy-side portfolio manager you can probably implement and use N() no drift today. If you are a levered investor or a trading desk, follow the yellow brick road of references laid out today, put some work into it, and be safe.
References IAcerbi, C., Tasche, D., 2001. Expected Shortfall: a natural coherent alternative to Value at Risk. Working paper.Anderson, T., Benzoni, L., Lund, J., 2002. Towards an empirical foundation for continuous time equity return models. Journal of Finance 57, 1239-84.Artzner, P., Delbaen, F., Eber, J., Heath, D., 1997. Thinking Coherently. Risk 10, 68-71Artzner, P., Delbaen, F., Eber, J., Heath, D., 1999. Coherent Measure of Risk. Mathematical Finance 9, 203-228.Atiya, A. F., Metwally, S. A. K., 2002. Efficient Estimation of First Passage Time Density Function for Jump Diffusion Processes. SIAM Journal on Scientific Computing 2002 (online version revised 2004)Bakshi, G., Panayatov, G., (forthcoming). First-Passage probability, Jump Models, and Intra-Horizon Risk. Journal of Financial Engineering (forthcoming)Basel Committee, 1996. Overview of the amendment to the capital accord to incorporate market risks. Working paper. Basel Committee on Banking Supervision.
References IIBates, D., 2000. Post 87 Crash fears in S&P 500 futures options. Journal of Econometrics 94, 181-238.Boudoukh, J., Richardson, M., Stanton, R., Whitelaw, R., 2004. Max var: long-horizon value-at-risk in a mark-to-market environment. Journal of Investment Management 2, 1-6.Carr, Geman, Madan, Yor, 2002.Carr, Wu, 2003.Duffie, D., Pan, J., Singleton, K., 2000. Transform analysis and asset pricing for affine jump diffusions. Econometric 68, 1343-1376.Eraker, B., Johannes, M., Polson, N., 2003. The Impact of Jumps in Volatility and Return. The Journal of Finance 58, 1269-1300.Feller, W., 1971. An Introduction to Probability Theory and Its Applications, Vol. II. John Wiley & Sons, New York.Johannes, M., Kumar, R., Polson, N., 1999. State Dependent Jump Models: How do US Equity Indices Jump? Online.
References IIIKaratzas, I., Shreve, S., 1991. Brownian Motion and Stochastic Calculus, second edition. Springer (New York), Graduate Texts in Mathematics #113, 1991, 196-197.Kritzman, M., Rich, D., 2002. The mismeasurement of risk. Financial Analysts Journal 58, 91-99.Lo, C.F., Hui, C.H., 2007. Valuing double barrier options with time dependent parameters by Fourier series expansion. Published online February 2007.Merton, Reiner, Rubenstein, 1973Merton, 1976Jorion, P. 1997. Value at Risk. Chicago, IL: Irwin.Rosenberg B. and Guy J. The Prediction of Systematic Risk Berkeley Research Program in Finance, Working Paper 33, February 1975.Samuelson, P.A. 1963. Risk and Uncertainty: A Fallacy of Large Numbers. Scientia, vol. 57, no. 6 (April/May): 108-113Stultz, R., 1996. Rethinking risk management. Journal of Applied Corporate Finance 9, 8-24

More Related Content

Intra Horizon Risk 2010

  • 2. OverviewCriticisms of tracking error / VaRTail shapeTime dimensionIntroduction of first-passage probabilityDerivation in Normal, no drift caseSidetrack 1: time-diversification argumentPossible ExtensionsDrift, jumps, stochastic volatilityIntroduce Jump ModelsReview three common jump models, empirical evidenceSidetrack 2: predictability of jumpsSolving first-passage probability for jump modelsEmpirical VaR intra-horizon versus end-of-horizon for three jump modelsConclusion:Implementable intra-horizon VaR / t.e. measure under assumptionsRoadmap for intra-horizon risk under more flexible distributions
  • 3. Criticisms of VaR/tracking errorTail Shape: says nothing about size of loss, only probability of lossTime Dimension: says nothing about losses prior to horizon, recovery etcDistributional assumptions: doesnt allow for jumps, non-Normal behaviorTime-scaling issues
  • 4. Tail shapeVaR result is independent of tail shapeFor measures that specifically include tail risk see:Tail-conditional expectation, worst conditional expectationArtzner, Delbaen, Eber and Heath (1997, 1999)Expected shortfall, CVaR etcAcerbi, Tasche (2001)
  • 5. Time DimensionVaR/tracking error says nothing about the profit and loss distribution over the trading horizonBasel Committee (1996)Denote intra-horizon riskLargely overlooked topic, few papers:Stultz (1996)Kritzman and Rich (2002)Boudoukh, Richardson, Stanton, Whitelaw (2004)Restrictive Assumptions: Brownian motion, no jumpsNote: Basel Committee (1996) also recognizes jumps are relevant for risk management purposes.Bakshi, Panayatov (forthcoming JFE)
  • 6. MotivationReturnRisk BandTimeTracking error or VaR are concerned about the likely distribution of returns at the end of some investment period. In this example we have scored a goal the final return is within the risk bands we set ourselves.
  • 7. Motivation IIReturnRisk BandTimeSo FiredDifficult MeetingNice Bonus
  • 8. Another ViewReturnRisk BandTimeIn this example there are five return paths. Only one path ends up outside our risk bands (and that on the right side) so by EOH measures the probability of breaching the band is 1/5. If we look throughout the horizon, 3 out of 5 paths breach the barrier at some point during the horizon and hence the intra-horizon probability of breach is 3/5
  • 9. Tracking Error, VaREnd of Horizon MeasureCharacterizes the distribution of returns at the end of an investment periodSays nothing about:Maximum loss at some probability during the periodProbability of recovery following a loss during the periodor anything else about the return path
  • 10. Risk horizon should match investment horizonIt should be obvious, but probably isnt, that our forecast horizon for our risk model should match our forecast horizon for our alphas, and implicitly therefore our holding periodRosenberg & Guy (1975)What Rosenberg neglects is the very human characteristic of second-guessing the process mid-period & the idea that we might be leveraged and therefore unable to put our feet on the wheel and wait for it all to turn out right in the end
  • 11. Who Cares?Survival Analysis Leveraged Investor: if there is a floor below which I cannot go, I should care greatly what the probability of hitting that floor is during the periodMonitored Investor if there is a good chance I will lose a client if performance is worse than a particular hurdle at any time during the year, I should care about the probability of hitting that hurdle
  • 12. Who Cares IILoan agreement collateralSecurities lending collateralCapital adequacyRetirement/Endowment funding a liability intra-horizon
  • 13. Focus for TodayLooking at the time-dimensionWhat happens to the value of a position/portfolio during an investment horizonCompare results to standard end-of-horizon measureWe are using end-of-horizon VaR calculated using the Normal distribution, zero drift, and no jumps as our reference point. All multipliers are with respect to that measure.
  • 14. Not a new issueWe can think about the issue in two ways:The size of a loss given a probability (i.e. VaR)Turn that on its head: the probability of hitting a given loss level (barrier)Barrier Options since 1960sMerton, Reiner, Rubenstein (1973)analytical closed form pricing for barrier options(same problem: need to figure out probability of hitting a barrier during an investment horizon)Hirsa and Madan (2003)Kudryavtsev and Levendorskii (2006)
  • 15. Concept First-PassageBy using the concept from statistics of first-passage time (also known as first-hitting time) we can estimate the probability that our return will hit a particular hurdle e.g. -10% during the investment period, or estimate the maximum loss $X during the period at some given confidence level e.g. 99%
  • 16. CaveatsThis does nothing to fix the issue of strategy risk drift in the mean - to be discussed separately todayThis measure is equally exposed to estimation error as with the regular tracking error / VaR measure[and equally benefits just the same from measures to address estimation error]We are saying nothing about asymmetry of preferencesWe are staying away from saying anything about benchmarks being suitable or relative versus absolute riskTo first get an intuitive tractable solution we assume Normality and zero drift, although we later relax those assumptions
  • 17. Conclusions First(After this slide you can check your Blackberry and not miss anything important)Intra-Horizon VaR (VaR-I) is larger than End-of-Horizon VaRNormal, no jumps: VaR-I = 1.107 * VaR1Non-Normal, allowing jumps: that multiplier can be up to 2.642Non-Normal, jumps, stochastic vol, drift it can get worse1 Feller (1971)2Bakshi and Panayatov (forthcoming, JFE)
  • 18. Probability of loss at EOHAssumptionsNormalStationary voli.i.d.(so we can use sqrt T)This is just the difference between the cumulative % loss and the cumulative expected return divided by the cumulative standard deviation, then apply N() to convert this standardized distance from the mean into a probability estimate.(Note I am cheating already and including a drift term)
  • 19. VaRBy rearranging the previous equation and solving for L (the loss) we get:
  • 20. Probability of Loss Intra HorizonSecond part never zero or negativeImplies IH loss estimate > EOH estimate (always)IH P(loss) rises as investment horizon expands, whereas EOH P(loss) declines as investment horizon expands=> time-diversification argument
  • 21. [Side Track 1]: Time DiversificationSamuelson (1963) argued against time diversification:Although EOH p(loss) declines as inv hor incr., that benefit is offset by increasing magnitude of potential lossOthers have argued that time diversifies risk because the result does not depend on the magnitude of the potential lossWH risk presents another facet to that argument:Risk increases even if investors only care about probability of loss
  • 22. Solving the ProblemFirst show path-independent (i.e. end of horizon) probability of lossThen use principle of reflection to convert path-dependent (i.e. within horizon) probability of loss into an equivalent path-independent (i.e. end of horizon) measureReview implications and assumptions
  • 23. Path-independent probability of return r after time T Assume continuous returns are normally distributed:Probability that cumulative return = rWhere:n is the number of years in Tr is the continuous cumulative return is the standard deviation of the continuous returns
  • 24. Principal of ReflectionAReturnBTimeInvoking the principal of reflection; the path-dependent probability of breaching a barrier and subsequently recovering to A [solid line] is the same as the path-dependent probability of breaching the barrier and then descending to B, which is the same as the path-independent probability of ending up at B, because B cannot be reached without breaching the barrier at some point.
  • 25. The Main PointJoint Prob (breach barrier b, and recover to A) = prob (end at B)The reflection principle allows us to convert a path-dependent probability into a path-independent probability under some assumptions
  • 26. Set up the probabilitiesProb (breach barrier) = 1 prob( never breach)= 1 joint prob(end up above, and min always above)= 1 prob(end above) + joint prob(end above, and min <=barrier)= prob(end below) + joint prob(end above, and min <=barrier)By principal of reflection: the last term is equal to the path independent probability of reaching a return r below the barrier
  • 27. Assumptions so farZero driftNormal distribution / Brownian motionStationary volatilityNo jumps
  • 28. Implications so farWithin-horizon risk increases with time, adding fuel to the debate against time-diversification, because second term always positiveEven assuming Normal distribution, stationary volatility, no jumps, and no drift, the 1% VaR-I is 10.7% bigger than 1% VaRFeller (1971)At 2.5% its 14.4% bigger, and at 5% its 19.2% biggerBoudoukh, Richardson, Stanton, Whitelaw (2004)
  • 29. Relaxing the zero drift assumptionGirsanovs theoremUseful transformation: using Girsanovs theorem we can transform a Brownian motion with drift into a Brownian motion without drift to make the maths easier.The first-passage probability for a Brownian motion with drift is well known. See for example Karatzas and Shreve (1991)
  • 30. The Impact of DriftFor comparative purposes, look at what happens when we introduce drift of 10% and 15% respectively. Boudoukh, Richardson, Stanton, Whitelaw (2004)
  • 31. What about jumps?Basel 1996: recognizes jumps are importantother evidence:Evidence for jumps:Bakshi, Cao, Chen (1997)Bates (2000)Anderson, Benzoni, Lund (2002)but difficulty explaining market crashes with return jumps and diffusive volatilityEvidence for large jump risk premiaPan (2002)Evidence for jumps in return and in volatilityEraker, Johannes, Polson (2003)For two examples of models with return and volatility jumps seeDuffie, Pan, Singleton (2000)
  • 32. Jump ModelsJump processes:Merton (1976)Carr, Geman, Madan, Yor (2002)Carr and Wu (2003)Two groups:finite jumps (i.e. infrequent)infinite jumps (i.e. lots)
  • 33. Sidetrack 2: predictability of jumpsStrong evidence for predictability of jumps:(For US markets) the VIX can predict both jump arrival and sizeJohannes, Kumar, Polson (1999)Strong evidence of jump clusteringImplies that Merton (1976) model is not useful as it cannot capture persistence in the data i.e. jump clustering
  • 34. The ApproachSome analytical characterizations are known, particularly for stable processes with one-sided jumps:Kyprianou (2006)In general, need to resort to PIDE approach [partial integro-differential equations]Solve using finite difference methodsOr Monte CarloAtiya & Metwally (2002) very fast approach, around 100 times faster than pure MC by leveraging a Brownian bridge to reduce the number of points that need to be calculated.Beyond this point I am out of my depth and will be visible only by bubbles on the surface
  • 35. Some resultsCompared to standard Normal VaR JD = Mertons jump-diffusion model CGMY is the two-sided pure-jump Levy model of Carr, Geman, Madan, and Yor FMLS is the finite-moment log-stable model of Carr and Wu(Bakshi and Panayatov)
  • 36. ImplicationsVaR with jumps is bigger than N() VaRWe suspected as much beforehandThe choice of model makes a differenceBut do they all fit the data equally well?VaR-I is consistently greater than VaRWe knew that beforehand, based on the known result with the Normal, no drift caseVaR-I can be more than double standard VaRAnd this goes some way toward justifying the Basel multipliers
  • 37. Stochastic Volatility and JumpsJust a tasteStochastic volatility in a VaR context:Eberlein, Kallsen, and Kristen (2002)Two-dimensional PIDEs for models with both jumps and stochastic volatility:Feng and Linetsky (2006)Can be applied to first-passage calculations.(let me know how you get on)
  • 38. Related MeasuresLeaning heavily once again on the barrier-option analogy we could also consider:A risk range rather than a single loss level (double barrier)A time-varying barrier or rangeA barrier that varies as a function of volatilityE.g. Lo & Hui (2007), pricing double barriers where volatility, dividend yield, and the barriers are stochastic.
  • 39. ConclusionsIntra-Horizon Risk is important, and neglectedIf you are a buy-side portfolio manager you can probably implement and use N() no drift today. If you are a levered investor or a trading desk, follow the yellow brick road of references laid out today, put some work into it, and be safe.
  • 40. References IAcerbi, C., Tasche, D., 2001. Expected Shortfall: a natural coherent alternative to Value at Risk. Working paper.Anderson, T., Benzoni, L., Lund, J., 2002. Towards an empirical foundation for continuous time equity return models. Journal of Finance 57, 1239-84.Artzner, P., Delbaen, F., Eber, J., Heath, D., 1997. Thinking Coherently. Risk 10, 68-71Artzner, P., Delbaen, F., Eber, J., Heath, D., 1999. Coherent Measure of Risk. Mathematical Finance 9, 203-228.Atiya, A. F., Metwally, S. A. K., 2002. Efficient Estimation of First Passage Time Density Function for Jump Diffusion Processes. SIAM Journal on Scientific Computing 2002 (online version revised 2004)Bakshi, G., Panayatov, G., (forthcoming). First-Passage probability, Jump Models, and Intra-Horizon Risk. Journal of Financial Engineering (forthcoming)Basel Committee, 1996. Overview of the amendment to the capital accord to incorporate market risks. Working paper. Basel Committee on Banking Supervision.
  • 41. References IIBates, D., 2000. Post 87 Crash fears in S&P 500 futures options. Journal of Econometrics 94, 181-238.Boudoukh, J., Richardson, M., Stanton, R., Whitelaw, R., 2004. Max var: long-horizon value-at-risk in a mark-to-market environment. Journal of Investment Management 2, 1-6.Carr, Geman, Madan, Yor, 2002.Carr, Wu, 2003.Duffie, D., Pan, J., Singleton, K., 2000. Transform analysis and asset pricing for affine jump diffusions. Econometric 68, 1343-1376.Eraker, B., Johannes, M., Polson, N., 2003. The Impact of Jumps in Volatility and Return. The Journal of Finance 58, 1269-1300.Feller, W., 1971. An Introduction to Probability Theory and Its Applications, Vol. II. John Wiley & Sons, New York.Johannes, M., Kumar, R., Polson, N., 1999. State Dependent Jump Models: How do US Equity Indices Jump? Online.
  • 42. References IIIKaratzas, I., Shreve, S., 1991. Brownian Motion and Stochastic Calculus, second edition. Springer (New York), Graduate Texts in Mathematics #113, 1991, 196-197.Kritzman, M., Rich, D., 2002. The mismeasurement of risk. Financial Analysts Journal 58, 91-99.Lo, C.F., Hui, C.H., 2007. Valuing double barrier options with time dependent parameters by Fourier series expansion. Published online February 2007.Merton, Reiner, Rubenstein, 1973Merton, 1976Jorion, P. 1997. Value at Risk. Chicago, IL: Irwin.Rosenberg B. and Guy J. The Prediction of Systematic Risk Berkeley Research Program in Finance, Working Paper 33, February 1975.Samuelson, P.A. 1963. Risk and Uncertainty: A Fallacy of Large Numbers. Scientia, vol. 57, no. 6 (April/May): 108-113Stultz, R., 1996. Rethinking risk management. Journal of Applied Corporate Finance 9, 8-24

Editor's Notes

  • #15: Hirsa and Madan pricing options under Levy pure jump (i.e. infinite jump)K&amp;L same
  • #30: Notation: B= barrier level , S= Current position/portfolio level,