August 23rd, 2020, 5:01 am
Two methods are common to calculate VaR (Full-revaluation & Taylor method ). For the diffusion risks such as IR,FX,EQ, CSR ( credit spread risk ),COM(commodity), historical simulation (full revaluation ) VaR is the most used method. You need at least 310 historical observations at the risk factor level. For most risk factors apply relative shocks, for IR (negative ) apply absolute shocks. A combination of both methodologies applies to volatility surfaces for different asset classes ( strike, moneyness, delta, maturity and tenor ), apply different shocks to the wings & chest. For vanilla Fixed income portfolios where only first-order effects are dominant (i.e mostly delta risk ), to cut time, use delta of the portfolio as an approximation to your P&L instead of full -revolution VaR (Taylor method ), and get the VaR from that. For other risks that are not captured in your diffusion risks such as all trading book positions that are subject to own funds requirement for specific interest rate risk, except securitization positions and n-th to default baskets, use a type of credit VaR model in banking we call it the Incremental risk charge (IRC), captures losses that arise due to 2 types of risk ( migration risk & outright default ), over a one-year capital horizon, and 99.9% confidence level ( this model requires a lot of work and collaboration with your credit risk economic capital teams, banking book IRRBB teams and such to get the input to the model, e.g, issuer-issuer correlation matrix for underlying issuers, methodology for Probability of default, etc These processes are very much involved and require an assessment of the liquidity of risk factors that goes into your VaR model because illiquid risk factors have to be captured in another model that works as an add-on to your overall market risk capital charge. Anyhow, this is just a high-level taste of how to think about building a VaR model to capture market risk for trading book positions. The actual work of arriving at the shocks per risk factor, full-revaluation of positions to arrive at P&L vectors, aggregation at desk level and trading book, backtesting, regulatory statistical tests to satisfy robustness of VaR model to regulators, etc.. among others requires continuous tweaks. Also, given that facts on the ground show more occurrence of extreme events (jumps) or shocks, is VaR with the usual assumptions a good proxy for actual risk processes currently we are seeing in the Market!!