C2 RISK DIVISION EDF CEA Inria School Systemic Risk and Quantitative Risk Management
EDF CEA INRIA School Systemic Risk and Quantitative Risk Management Regulatory rules evolutions and internal models implementation Pierre Gaye Deputy head of Risks on Maket Activities P.2
Monitoring Risks Evolving rules _ The most popular risks for the bank _ Market risk _ Credit risk _ Operational risk _ Liquidity but also legal, compliance, reputation, commercial,. _ 2 main frameworks _ Internal monitoring _ Regulatory capital or ratios (Bale 1,2, 2.5, 3 ) but also rules as Dodd Franck, EMIR, _ and now, retroaction of these risks in the pricing theory _ CSA, CVA, DVA, FVA, P.3
Regulatory framework Evolving rules _ Pre crisis _ Market risk : losses due to MtM variation _ VAR : 3 times 99% percentile of 10 days all portfolio MtM variation _ Credit risk : losses due to counterparties default _ RWA : high level percentile of 1 year losses distribution in a 1 factor model _ Post crisis _ Market risk : adding VAR in stressed period and credit migration and defaults _ SVAR : 3 times 99% percentile of 10 days MtM portfolio variation during the worst historical period _ IRC : 99.9% percentile of 1 year credit portfolio MtM variation due to rating migration up to default _ Credit risk : adding VAR on CVA variation for derivative products _ CVA VAR : 99% percentile of 10 days CVA variation due to market spread shocks un heteroclite set of rules which as been defined to increase bank capital to overcome internal models deficiencies. Perhaps fundamental review of trading book would redesign a coherent framework. P.4
_ A well defined problem Internal model implementation Value at risk distribution _ Basle 2 Computation of unexpected loss A high level quantile of reduced to _ Basle 3 Computation of «market» expected loss CVA simplified formula _ Credit process Computation of worst case exposure A high level quantile of In 3 cases : assessing the marginal distribution of exposure value with or without margin calls P.5
Context Portfolio value distributions _ MtM distribution, EPE, ENE and PFE illustration 1 Illustration calcul EPE et ENE 0.8 0.6 0.4 MtM portefeuille 0.2 0 0 1 2 3 4 5 6-0.2-0.4-0.6 Temps Quantile 99% D1=Distribution MtM horizon 1 an D2 D4 Profil EPE Profil ENE P.6 6
_ From market parameters to exposure _ Market parameter joint diffusion processes Should be conditional to counterparty default event Technical approach Exposure value computation _ Deal repricing Non linear function : full valorization Deal ageing and path dependencies _ Exposure computation Netting or not netting Collateral joint simulation Simple but non linear transformation of MtM No analytical solution : only Monte Carlo simulation available P.7
_ How to represent joint behavior of thousands of market parameters? _ Historical or Risk neutral? Banking book : classical credit process Trading book : pricing and distribute Market parameters modeling Parameterization _ Diffusion processes Absolute levels at each time horizon Short term variations for margin calls Brownian motion without or with mean reversion, jumps _ Joint distributions Correlation/Cointegration Differentiate short term/long term correlations _ Necessary factorial type model Dimension reduction Operational flexibility P.8
Market parameters modeling Calibration _ Thou shalt never underestimate risks _ Non adequation of normal distribution Fix 2 objectives Non minimizing PFE Non minimizing EE Conservatism on volatility level Distribution réelle Distribution modélisée : distribution normale avec volatilité fractile Pay off Distribution normale avec volatilité standard 120,00% 100,00% 80,00% 60,00% 40,00% _ Trend levels and Correlations instability and neutrality 20,00% 0,00% - 2,5-2 - 1,5-1 - 0,5 0 0,5 1 1,5 2 2,5 _ Historical period choice 3 years or 10 years? _ Statistical process or expert process? P.9
_ Making difference between first and second order risks _ A specialized pricing function library Using FO pricing function and distribution would need 5 million hours Specific implementation necessary to reduce computation time by a few magnitude order Could be at precision cost Technical approach Some necessary simplifications _ Market state representation Tens of thousands of market parameters Thousands of issuers : stock prices and spread curves Number of tenors on interest, spread or volatility curves Smiles on volatity surfaces A necessary simplification Memory management True systematic risk dimension probably of only a few hundreds of risk factors No diffusion of second order risk parameters P.10
_ How many computations for what precision level? _ Number of scenarios - 1000 Individual counterparty precision level : 10% Large number of counterparties Total EEPE or RWA : probably less than 1% _ Number of time steps - 100 Individual operation with continuous profile Could be highly bumpy Maturity mismatch Significant cash flow payment intermediate coupon or maturity settlement _ Margin call simulation Deals settling during holding period MTA simulation Probabilité Technical approach Monte Carlo simulation control 40,0% 35,0% 30,0% 25,0% 20,0% 15,0% 10,0% 5,0% Distributions théoriques des estimateurs EE et PFE 0,0% -0,200 0,300 0,800 1,300 1,800 2,300 2,800 Nombre d'écarts type Moyenne 1000 scenarios EE moy MtM=-0.5 1000 scenarios EE moy MtM=0 1000 scenarios EE moy MtM=0 4000 scenarios EE moy MtM=0.5 1000 scenarios PFE 95% 1000 scenarios PFE 99% 1000 scenarios PFE 99% 4000 scenarios PFE 99.5% 1000 scenarios Impact studies demonstrate possible optimization but generally low impact P.11
Backtesting _ An unobservable distribution _ Comparing samples of realized exposure to the 1y modelized distribution How to get independent observations? Non overlapping historical periods _ Extrapolate short term behavior on the long term At least backtest short term MtM variations like VAR _ Test full distribution By construction generation of normal distribution Small size of samples large confidence interval _ Sanity checks MtM replication at t=0 Short term MtM variations P.12
Conclusions _ Overall a global robust system _ Quite prudent market parameter representation Individual performance test _ Much more difficult testing of correlations Keep only significant ones _ But very stable along impact studies _ With a few remaining challenges _ Taking into account exotics _ Taking into account market/credit correlations _ With renewed interest with CVA _ Technical enhancements predictable _ Model confrontation through pricing P.13
Systemic Risk _ Globally quite conservative measures _ IRC= 20% of counterparts are in default _ VAR+SVAR+3 time multiplier _ Credit : 8 to 10% of equity toward exposure which will double with CVA VAR _ Are they enough to avoid Grands Risques? _ NBT _ Complementary measures : concentrations, WWR (country risk) _ Reintroducing systemic risks _ CVA VAR hedges : procyclacity and hedge provider _ CCPs _ But system will adapt _ Banks prefer to exchange IAs P.14