In late summer of 1997 the most rigorous and extensive backtesting ever done on options, futures and swaps began at the London Clearing House, using FHS. Professors Giovanni Barone-Adesi and Kostas Giannopoulos oversaw the project. Les Vosper, an experienced senior analyst at the LCH, was responsible for auditing the model. David, our software developer, was a major component of the team. Stavros had designed and developed the database. Other in-house analysts joined the team for shorter periods to help with the auditing.
A database with historical prices for all the financial futures that were trading at that time on LIFFE, thousands of options written on them, as well positions of about one hundred fifty daily member accounts were retrieved from the archives and were pooled on a local server. In addition they were plain vanilla swaps in eight currencies with up to 5000 in each one. On each currency historical interest rates on the following maturities were collected, ON,1W,1M, 2M, 3M, 6M, 9M, 12M, 1.5Y, 2Y, 2.5Y, 3Y, 3.5Y, 4Y, 4.5Y, 5Y, 5.5Y, 6Y, 6.5Y, 7Y, 7.5Y, 8Y, 8.5Y, 9Y, 9.5Y, 10Y.
After creating the database an extensive auditing on the historical price series of the futures and the options began. Rolling time series of returns, on each maturity, were created from the historical futures prices. For each of the 120 series of returns on futures a rigorous econometric analysis was carried out to select the best model from a set of GARCH and other conditional time series models. The criteria were set under both the prism of econometric tests and risk management goals. A similar econometric analysis was carried out on the rates for each currency and maturity.
The econometric fitting was carried on out of sample information, using historical data ending a day before the first day of the backtesting period. During the backtesting the conditional mean and variance equations were recursively fed with updated returns to generate the simulated pathways for each underlying asset for the next ten days. Algorithms were developed to test automatically the validity of the econometric model on each day of the backtesting period and for each underlying asset. There were checks for structural breaks and other extreme irregularities to assess the need to recalibrate model parameters. Other algorithms checked the FHS-simulated scenarios for both contracts and member accounts or swaps books. They reported any suspect values and the results were verified with simulations carried out on independent software, such as RATS and excel. Irregularities and suspicious prices were either due to code errors or failures of the econometric models. Once the bug was fixed the backtesting re-started. For each day for a two year period, for about one hundred fifty accounts and swaps books with all possible combinations of stochastic or constant implied volatility and for stochastic or hedged FX rates. David was constantly modifying the software to audit the data and write new testing features. Nothing was left to chance.