Vol 24, No 4; Article by Meera Sharma; December 2012
This paper studies the impact of the revision by the Basel Committee in 2011 of one of the quantitative standards for internal models of market risk. The revision allows calculation of the Value-at-Risk (VaR) using a weighting scheme on historical data (or data lengths less than a year) as long as the resultant capital charge is at least as conservative as that using un-weighted data. This study applies the historical simulation method to calculate VaR for ten year long daily returns data of Indian stocks and the INR/USD exchange rate. Data lengths of 1000, 750, 500 and 250 days are used in combination with data lengths of 190 and 125 days to evaluate impact of the revision. The VaRs generated by each data length individually and by the combinations of long and short data lengths are evaluated using four tests, namely the regulatory back test; and hypotheses tests of unconditional coverage, independence, and conditional coverage of VaR exceptions.
The combination of short and long historical observation periods improves the performance of long period VaRs in regulatory back tests, resulting in lower regulatory penalties in both data series. The combination also improves performance in hypothesis tests of unconditional coverage. The inability of historical simulation, using both long and short data lengths, to deal with exception clustering is highlighted by their failure of tests of independence. All the VaR methods, taken individually and in combination, fail the test of conditional coverage in case of the equity data. However, the combinations between 1000 and 750 VaRs and 190 and 125 day VaRs pass the test of conditional coverage in case of forex rate data, showing some potential in combination VaRs to deal with exception clustering.
