Challenges In XVA Pricing And Risk - MathWorks

Transcription

Challenges in xVA Pricing and RiskAndrew McClelland, PhDDirector of Quantitative ResearchNumerixMatlab Computational Finance ConferenceNYC 2017September 28, 2017Andrew McClelland, NumerixxVA Pricing and Risk1/13

Presentation OutlineQuick overview of Numerix and why we prioritize xVAWhat is xVA? Some intuition for how things come togetherComputational di culties encountered in xVAMulti-factor hybrid models and American Monte Carlo1Matlab for complicated exposures modeling and aggregations1Used when referring to Least-Squares Monte Carlo, a la Longsta andSchwartz ('01), in the context of exposures calculations.Andrew McClelland, NumerixxVA Pricing and Risk2/13

Numerix Overview 12 founded in '96, headquartered in NYC, 250 employeesNumerixIt has origins as a pricing & risk library for derivativesOriented towards exotic & multi-factor derivatives,eg .PRDCsManaging risk for exotics is a very complicated processRequires sophisticated & delicate models, methods, calibrations2https://www.numerix.com.Andrew McClelland, NumerixxVA Pricing and Risk3/13etc .

Numerix Overview 13 founded in '96, headquartered in NYC, 250 employeesNumerixIt has origins as a pricing & risk library for derivativesOriented towards exotic & multi-factor derivatives,eg .PRDCsManaging risk for exotics is a very complicated processRequires sophisticated & delicate models, methods, calibrations3https://www.numerix.com.Andrew McClelland, NumerixxVA Pricing and Risk3/13etc .

Pricing Work owrate curvesMarketDatavol surfacesFX levelsforward ratesCalibratedModelshort rate factorslocal/stochastic volExoticValuationMonte CarloPDElattice/treeStandard pricing work ow. First, market data such as volatility surfaces aresourced, transformed and cleansed. Second, a pricing model is calibrated to the marketdata. Third, a pricing method such as Monte Carlo is invoked to compute the price.Figure:Andrew McClelland, NumerixxVA Pricing and Risk3/13

Risk Work owrate curvesMarketMarketDataData'vol surfacesFX levelsforward ratesCalibratedCalibratedModelModel'short rate factorslocal/stochastic volExoticExoticValuationValuation'Figure:Monte CarloPDElattice/treeStandard risk work ow. Risk factors among market data and model elements,eg . yield nodes and model parameters, are identi ed as risk factors. Sensitivities againstthese factors are computed and are hedged using vanilla instruments.Andrew McClelland, NumerixxVA Pricing and Risk3/13

Risk Work owrate ation'vol surfacesFX levelsforward ratesCalibratedCalibratedModelModel'short rate factorslocal/stochastic volExoticExoticValuationValuation'Figure:Monte CarloPDElattice/treeStandard risk work ow. Risk factors among market data and model elements,eg . yield nodes and model parameters, are identi ed as risk factors. Sensitivities againstthese factors are computed and are hedged using vanilla instruments.Andrew McClelland, NumerixxVA Pricing and Risk3/13

Risk Work owrate ation'vol surfacesFX levelsforward ratesCalibratedCalibratedModelModel'short rate factorslocal/stochastic volExoticExoticValuationValuation'Figure:Monte CarloPDElattice/treeStandard risk work ow. Risk factors among market data and model elements,eg . yield nodes and model parameters, are identi ed as risk factors. Sensitivities againstthese factors are computed and are hedged using vanilla instruments.Andrew McClelland, NumerixxVA Pricing and Risk3/13

Numerix Overview 2After crisis exotics volumes dropped but xVAs,eg .CVA, emergedBig challenge for banks and managing xVA resembles managing exoticsNice use case for how our clients use a mix of NX and MatlabContent should be interesting for this audience alsoSits within larger issue ofholisticAndrew McClelland, Numerixmodeling of trade impactxVA Pricing and Risk4/13

Numerix Overview 2After crisis exotics volumes dropped but xVAs,eg .CVA, emergedBig challenge for banks and managing xVA resembles managing exoticsNice use case for how our clients use a mix of NX and MatlabContent should be interesting for this audience alsoSits within larger issue ofholisticAndrew McClelland, Numerixmodeling of trade impactxVA Pricing and Risk4/13

Numerix Matlab InterfaceFigure: Example of a CVA calculation using the Matlab interface for NX in combinationwith native Matlab functionality.Andrew McClelland, NumerixxVA Pricing and Risk4/13

Numerix Overview 2After crisis exotics volumes dropped but xVAs,eg .CVA, emergedBig challenge for banks and managing xVA resembles managing exoticsNice use case for how our clients use a mix of NX and MatlabContent should be interesting for this audience alsoSits within larger issue ofholisticAndrew McClelland, Numerixmodeling of trade impactxVA Pricing and Risk4/13

Numerix Overview 2After crisis exotics volumes dropped but xVAs,eg .CVA, emergedBig challenge for banks and managing xVA resembles managing exoticsNice use case for how our clients use a mix of NX and MatlabContent should be interesting for this audience alsoSits within larger issue ofholisticAndrew McClelland, Numerixmodeling of trade impactxVA Pricing and Risk4/13

What is xVA?Traditional valuations involve the risk-neutral valuation formulaClassical Valuationh RTie t e t r (u) du C(T )V (t) E45Many ways to derive this formula, equilibrium , arbitrage ,etc .6 is a bit restrictive but mimics bank practiceHedging approachClassical hedging assumptions ignore certain frictions eg .counterparty credit risk, funding spreads & collateral4eg . Cox, Ingersol and Ross ('85).eg . Harrison and Kreps ('79).6eg . Black and Scholes ('73) or Burgard and Kjaer ('11).5Andrew McClelland, NumerixxVA Pricing and Risk5/13

What is xVA?Traditional valuations involve the risk-neutral valuation formulaClassical Valuationh RTie t e t r (u) du C(T )V (t) E78Many ways to derive this formula, equilibrium , arbitrage ,etc .9 is a bit restrictive but mimics bank practiceHedging approachClassical hedging assumptions ignore certain frictions eg .counterparty credit risk, funding spreads & collateral7eg . Cox, Ingersol and Ross ('85).eg . Harrison and Kreps ('79).9eg . Black and Scholes ('73) or Burgard and Kjaer ('11).8Andrew McClelland, NumerixxVA Pricing and Risk5/13

Valuations via Hedging ArgumentsCDS HedgeCounterpartySwap HedgeCounterpartyVanillaSwapsStructured ianClearing HouseFigure:Basic considerations within hedging derivation of valuation rule.Andrew McClelland, NumerixxVA Pricing and Risk5/13

What is xVA?Traditional valuations involve the risk-neutral valuation formulaClassical Valuationh RTie t e t r (u) du C(T )V (t) EMany ways to derive this formula, equilibrium10 , arbitrage11 ,etc .12 is a bit restrictive but mimics bank practiceHedging approachClassical hedging assumptions ignore certain frictions eg .counterparty credit risk, funding spreads & collateral10eg . Cox, Ingersol and Ross ('85).eg . Harrison and Kreps ('79).12eg . Black and Scholes ('73) or Burgard and Kjaer ('11).11Andrew McClelland, NumerixxVA Pricing and Risk5/13

Valuations via Hedging ArgumentsCDS HedgeCounterpartySwap HedgeCounterpartyVanillaSwapsStructured ianClearing HouseModern considerations within hedging derivation of valuation rule, includingcounterparty credit risk, funding costs, initial margin requirements.Figure:Andrew McClelland, NumerixxVA Pricing and Risk5/13

Valuations via Hedging ArgumentsCDS HedgeCounterpartySwap HedgeCounterpartyVanillaSwapsStructured anClearing HouseModern considerations within hedging derivation of valuation rule, includingcounterparty credit risk, funding costs, initial margin requirements.Figure:Andrew McClelland, NumerixxVA Pricing and Risk5/13

Valuations via Hedging ArgumentsCDS HedgeCounterpartyCDSContractsSwap HedgeCounterpartyVanillaSwapsStructured anClearing HouseModern considerations within hedging derivation of valuation rule, includingcounterparty credit risk, funding costs, initial margin requirements.Figure:Andrew McClelland, NumerixxVA Pricing and Risk5/13

Valuations via Hedging ArgumentsCDS HedgeCounterpartyCDSContractsSwap HedgeCounterpartyVanillaSwapsStructured SwapBankTreasuryFundingCustodianClearing HouseClienttounccgAdinFuninitiInalrgMaModern considerations within hedging derivation of valuation rule, includingcounterparty credit risk, funding costs, initial margin requirements.Figure:Andrew McClelland, NumerixxVA Pricing and Risk5/13

Valuations via Hedging ArgumentsCDS HedgeCounterpartyCDSContractsSwap HedgeCounterpartyVanillaSwapsStructured SwapBankTreasuryFundingCustodianClearing HouseClienttounccgAdinFuninitiInalrgMaModern considerations within hedging derivation of valuation rule, includingcounterparty credit risk, funding costs, initial margin requirements.Figure:Andrew McClelland, NumerixxVA Pricing and Risk5/13

What is CVA?Counterparty credit risk credit valuation adjustment"13 or CVABasically upfront cost of dynamically hedging counterparty defaultMust be added to classical value, or cooked into deal paymentsCredit Valuation AdjustmentCVA(t)et E Lifetime Discounted Counterparty Loss(t, T ) IFRS13 (accounting standard) requires CVA be recorded in statementsBasel III extended capital framework to include a CVA charge13eg . Du e and Canabarro ('04) and Gregory ('15).Andrew McClelland, NumerixxVA Pricing and Risk6/13

What is CVA?Counterparty credit risk credit valuation adjustment"14 or CVABasically upfront cost of dynamically hedging counterparty defaultMust be added to classical value, or cooked into deal paymentsCredit Valuation AdjustmentCVA(t)et E Lifetime Discounted Counterparty Loss(t, T ) IFRS13 (accounting standard) requires CVA be recorded in statementsBasel III extended capital framework to include a CVA charge14eg . Du e and Canabarro ('04) and Gregory ('15).Andrew McClelland, NumerixxVA Pricing and Risk6/13

What is CVA?Counterparty credit risk credit valuation adjustment"15 or CVABasically upfront cost of dynamically hedging counterparty defaultMust be added to classical value, or cooked into deal paymentsCredit Valuation Adjustment ZetCVA(t) ET Discounted Counterparty Loss(s) dstIFRS13 (accounting standard) requires CVA be recorded in statementsBasel III extended capital framework to include a CVA charge15eg . Du e and Canabarro ('04) and Gregory ('15).Andrew McClelland, NumerixxVA Pricing and Risk6/13

What is CVA? Counterparty credit risk credit valuation adjustment"16 or CVABasically upfront cost of dynamically hedging counterparty defaultMust be added to classical value, or cooked into deal paymentsCredit Valuation Adjustment ZetCVA(t) ETe Rstr (u) du λ(s)(V (s)) dstIFRS13 (accounting standard) requires CVA be recorded in statementsBasel III extended capital framework to include a CVA charge16eg . Du e and Canabarro ('04) and Gregory ('15).Andrew McClelland, NumerixxVA Pricing and Risk6/13

What is CVA? Counterparty credit risk credit valuation adjustment"17 or CVABasically upfront cost of dynamically hedging counterparty defaultMust be added to classical value, or cooked into deal paymentsCredit Valuation Adjustment ZetCVA(t) ETe Rstr (u) du λ(s)(V (s)) dstIFRS13 (accounting standard) requires CVA be recorded in statementsBasel III extended capital framework to include a CVA charge17eg . Du e and Canabarro ('04) and Gregory ('15).Andrew McClelland, NumerixxVA Pricing and Risk6/13

CVA Complications: Counterparty AggregationCrucial to note that CVA is computed at theAssume swaps with valuesV1swapandV2swapcounterpartylevelwith counterpartyPortfolio exposure is not the sum of individual swap exposuresPortfolio Exposure Aggregation(V pfolio ) (V1swap V2swap ) 6 (V1swap ) (V2swap ) ClearlyV2swap 0Thus thecould o setincrementalV1swap 0,andvica versaCVA involves whole counterparty portfolioWhy xVA calcs are so expensive. some xVAs involve many c/parties!Andrew McClelland, NumerixxVA Pricing and Risk7/13

A Practical CVA ComputationEvaluate the CVA formula via discretization and Monte CarloUsep 1, . . . , Ppaths andi 0, · · · , TtimestepsCVA CalculationCVA(t0 ) 1PT XPXλp (si )e Rsti rp (u) du(Vp (si )) sii 1 p 1Xp,i , eg .1Simulate paths state variables2Revalue portfolio along all paths3Evaluaterates, hazards, FXpfolioVp,i V1swap,p,i · · ·(Vp,i ) , weight by discount factors & hazards and aggregateAndrew McClelland, NumerixxVA Pricing and Risk8/13

A Practical CVA ComputationEvaluate the CVA formula via discretization and Monte CarloUsep 1, . . . , Ppaths andi 0, · · · , TtimestepsCVA CalculationCVA(t0 ) 1PT XPXλp (si )e Rsti rp (u) du(Vp (si )) sii 1 p 1Xp,i , eg .1Simulate paths state variables2Revalue portfolio along all paths3Evaluaterates, hazards, FXpfolioVp,i V1swap,p,i · · ·(Vp,i ) , weight by discount factors & hazards and aggregateAndrew McClelland, NumerixxVA Pricing and Risk8/13

CVA Simulation StepsShort Rate 03012345678910Horizon (Yrs)Example of simulations involved in a CVA calculation. Portfolio consists of a5Y swap and a 10Y swap.Figure:Andrew McClelland, NumerixxVA Pricing and Risk8/13

CVA Simulation Steps10Y Swap zon (Yrs)Example of simulations involved in a CVA calculation. Portfolio consists of a5Y swap and a 10Y swap.Figure:Andrew McClelland, NumerixxVA Pricing and Risk8/13

CVA Simulation Steps5Y Swap zon (Yrs)Example of simulations involved in a CVA calculation. Portfolio consists of a5Y swap and a 10Y swap.Figure:Andrew McClelland, NumerixxVA Pricing and Risk8/13

CVA Simulation StepsPortfolio Horizon (Yrs)Example of simulations involved in a CVA calculation. Portfolio consists of a5Y swap and a 10Y swap.Figure:Andrew McClelland, NumerixxVA Pricing and Risk8/13

CVA Simulation StepsPositive Portfolio Horizon (Yrs)Example of simulations involved in a CVA calculation. Portfolio consists of a5Y swap and a 10Y swap.Figure:Andrew McClelland, NumerixxVA Pricing and Risk8/13

CVA Simulation StepsPositive Portfolio Horizon (Yrs)Example of simulations involved in a CVA calculation. Portfolio consists of a5Y swap and a 10Y swap.Figure:Andrew McClelland, NumerixxVA Pricing and Risk8/13

Main Challenges in CVAPortfolios can span multiple risk factors,Building models is di cult and buildingeg .swaps in many CCYshybridmodels is more soRequire correlation structures for (latent) variables & joint calibrationComputingVp,ican be anO(P 2 )problem for exoticsIn practice this can be extremely expensive and we try to avoidAmerican MC (AMC) is a powerful regression-basedAndrew McClelland, NumerixxVA Pricing and Risk9/13O(P)algorithm

American Monte CarloEtymology from its use in pricing American (early-exercise) optionsIdea is to use the originalPpaths to build future values via projectionFuture Values by Regressione t [C(T )] V (t) Ee t [ (t, T )] 0C(X (T )) V (X (t)) (t, T ), Ey f (x) υ, E[υ] 0Evaluated via regression withSelection of basis functionsf (·) β0 φ0 (·) β1 φ1 (·) · · ·φ(·)crucial, especially in high dimensionsDi cult to implement for generic products with scripted payo sAndrew McClelland, NumerixxVA Pricing and Risk10/13

American Monte CarloEtymology from its use in pricing American (early-exercise) optionsIdea is to use the originalPpaths to build future values via projectionFuture Values by Regressione t [C(T )] V (t) Ee t [ (t, T )] 0C(X (T )) V (X (t)) (t, T ), Ey f (x) υ, E[υ] 0Evaluated via regression withSelection of basis functionsf (·) β0 φ0 (·) β1 φ1 (·) · · ·φ(·)crucial, especially in high dimensionsDi cult to implement for generic products with scripted payo sAndrew McClelland, NumerixxVA Pricing and Risk10/13

American Monte CarloEtymology from its use in pricing American (early-exercise) optionsIdea is to use the originalPpaths to build future values via projectionFuture Values by Regressione t [C(T )] V (t) Ee t [ p (t, T )] 0C(Xp (T )) V (Xp (t)) p (t, T ), Ey f (x) υ, E[υ] 0Evaluated via regression withSelection of basis functionsf (·) β0 φ0 (·) β1 φ1 (·) · · ·φ(·)crucial, especially in high dimensionsDi cult to implement for generic products with scripted payo sAndrew McClelland, NumerixxVA Pricing and Risk10/13

American Monte CarloEtymology from its use in pricing American (early-exercise) optionsIdea is to use the originalPpaths to build future values via projectionFuture Values by Regressione t [C(T )] V (t) Ee t [ p (t, T )] 0C(Xp (T )) V (Xp (t)) p (t, T ), Ey f (x) υ, E[υ] 0Evaluated via regression withSelection of basis functionsf (·) β0 φ0 (·) β1 φ1 (·) · · ·φ(·)crucial, especially in high dimensionsDi cult to implement for generic products with scripted payo sAndrew McClelland, NumerixxVA Pricing and Risk10/13

American Monte CarloEtymology from its use in pricing American (early-exercise) optionsIdea is to use the originalPpaths to build future values via projectionFuture Values by Regressione t [C(T )] V (t) Ee t [ p (t, T )] 0C(Xp (T )) V (Xp (t)) p (t, T ), Eyp f (xp ) υp , E[υp ] 0Evaluated via regression withSelection of basis functionsf (·) β0 φ0 (·) β1 φ1 (·) · · ·φ(·)crucial, especially in high dimensionsDi cult to implement for generic products with scripted payo sAndrew McClelland, NumerixxVA Pricing and Risk10/13

American Monte Carlo Example0.25American MC for a BermudanCashflow(T)Value(t)0.20.150.10.050-0.08 -0.06 -0.04 -0.0200.020.040.060.080.10.12Short Rate(t)Example of AMC for a simple Bermudan swaption. Pathwise cash ows areprojected onto the short rate using a Hull White ('90) model.Figure:Andrew McClelland, NumerixxVA Pricing and Risk10/13

American Monte CarloEtymology from its use in pricing American (early-exercise) optionsIdea is to use the originalPpaths to build future values via projectionFuture Values by Regressione t [C(T )] V (t) Ee t [ p (t, T )] 0C(Xp (T )) V (Xp (t)) p (t, T ), Eyp f (xp ) υp , E[υp ] 0Evaluated via regression withSelection of basis functionsf (·) β0 φ0 (·) β1 φ1 (·) · · ·φ(·)crucial, especially in high dimensionsDi cult to implement for generic products with scripted payo sAndrew McClelland, NumerixxVA Pricing and Risk10/13

General Bermudan Swaption ScriptExample of a Numerix payo script for a Bermudan swaption. All DISCOUNTINGvariables are subjected to a regression as the algorithm works backward in time.Figure:Andrew McClelland, NumerixxVA Pricing and Risk10/13

Where Does Matlab Come In? 1Clients using Numerix have theirownviews on how to aggregateInstead of writing NX scripts & functions to do the maths use MatlabExpose intermediate calcs to allow user logicMany assumptions on counterparty behavior and default correlationseg . wrong-way risk" or WWR: exposures correlate with default probSubjective and clients want control over default prob trajectoriesAndrew McClelland, NumerixxVA Pricing and Risk11/13

Where Does Matlab Come In? 2Also complications re. collateral arrangements and marginingRatings triggers can force lower collateral thresholds and min transfersEarly-terminations as function of market conditions or credit health?Similarly so for rolling of shorter-term tradesCollateral rehypothecation across counterparties for funding o setsAndrew McClelland, NumerixxVA Pricing and Risk12/13

What Does the Future Hold?The really di cult calculations relate to portfolio-level e ectsLooking for capital impacts (KVA) and initial margin impacts (MVA)18Capital and initial margin requirements are becoming more onerousPricing these and/or optimizing against them is crucialAlso bene ts to fully tracking cost of collateral transformation18See eg . Basel FRTB (BCBS365) and Basel-IOSCO (BCBS317).Andrew McClelland, NumerixxVA Pricing and Risk13/13

Counterparty credit risk ) credit valuation adjustment" 16 or CVA Basically upfront cost of dynamically hedging counterparty default Must be added to classical value, or cooked into deal payments Credit Valuation Adjustment CVA (t) Ee t Z T t e R s t r(u)du (s)(V(s)) ds IFRS13 (accounting standard) requires CVA be recorded in statements