In our long, professional career we have come to realise time and again that not all backtests are the same. Often we missed the necessary care. On examination, differences were found in comparison to the results we obtained with our methodology. The adjacent notes give a brief overview of what is important in backtesting.
Historical data of impeccable quality is a challenge. To ensure quality, we make samples and compare historical time series from several data providers. In principle, we only use data from providers that we consider to be professional and trustworthy. These includes, but is not limited to companies, such as Bloomberg, Refinitv and Siblis Research.
Buy and sell orders are created as limit orders. This ensures, that if a market has adequate liquidity, the slippage is reduced to a minimum.
In order to be comparable to a price index such as the S&P 500, payments such as dividends are not counted. Profits from the investment strategy are reinvested.
An essential prerequisite for a realistic backtest is the consideration of the so-called survivorship bias [link to survivorship bias in the glossary]. All changes in the composition of the index have of course been carefully considered in all investment strategies. The backtest corresponds at all times to the (earlier) composition of the respective index or benchmark.
Practical transaction costs are considered in all backtests. These are relatively high, i.e. set above the exchange fees and accordingly offer enough buffer. Note: For Xetra trading on Deutsche Boerse, for example, the transaction costs - depending on the trading volume - are between 0.36 and 0.90 basis points (as of January 1, 2020). 100 basis points correspond to 1%.
Unlike an index, investment strategies are not always 100% invested. Interest is not paid on the capital not invested. This may not play a role in the current quasi-interest-free period but would have positive effects in the backtest of the past decades. We do not use this additional income when calculating the performance.
