How to Effectively Backtest Your Strategies

This article is written by Søren Beissenherz Lanng who is the founder and CEO of TickCOM.

It is no secret that the majority of the markets have been very directional over the past 8 years; stocks and bonds have been in a rally since 2008, while gold and oil have been in a bear trend.

This has had a great impact on the development of automated strategies, since any long strategy for almost any stock will work very well when backtested using the past eight years of historical data. Perhaps if moving into very short timeframes we can achieve a sort of balanced strategy in terms of the long/short ratio, but the performance will still be influenced by medium and long-term trends.

Backtesting the aforementioned markets therefore requires much attention and care, especially if optimizing a strategy.

However, strategies for stocks on a 15 minute timeframe would not be possible to backtest for years to come! Stocks traditionally have a very strong correlation, and the same goes for bonds- meaning that we cannot even test a strategy across different instruments on these markets.

This will likely have an impact on the automated trading strategies that are currently running in the stocks and bonds markets. These have been working for the past 8 years, but at some point will fail.

There is an alternative and solution to this current paradox within strategy development and backtesting.

The alternative is to move to currency pairs, where we have a high count of instruments, some of which are correlated, some not. If, for example, a strategy is tested across ten different non-correlated currency pairs and gives similar results, with equity curves also behaving equally, this strategy is clearly a robust one.

The use of different currency pairs, rather than stocks and bonds, is a very powerful tool for developing and backtesting automated trading strategies. It is much more informative and useful to test a strategy using 3 months of 1 minute data across 10 different currency pairs, than to backtest a strategy on one instrument over 8 years of 1 minute data.

Testing across currency pairs means that we do not need to backtest for years back, testing a market behavior that is long gone. We know that a strategy works until the market’s behavior changes, and then it doesn’t work anymore.

And so, the faster we develop strategies, the more recent the data we can use- and the higher is the probability that we have developed a strategy that will also work next week.