Backtesting Strategies

HI all, I’m not sure if this is the place for this question, but I didn’t find a better topic on my glance of available headings. I have a question about backtesting. So, traditionally, it is recommended a strategy is backtested for two years, but I’m assuming that’s using the daily time frame. If you trade a smaller timeframe (I prefer the 15 minute), should you backtest a 15 min strategy for two years? A daily strategy would have about 500 points in a two year backtest. A similar sample size for a 15 minute backtest would be a week. How long should one backtest a strategy? Is it a length of time, or is it a certain number of data points? Any advice on this would be greatly appreciated. Happy Trading!!!

Fair point. However, just as not every day in 2 years is the same as every other day, even in a single 25-hour period, many 15-minute periods will not be highly comparable. The day consists of 4 contiguous market sessions, - London, London + New York, New York and Asia. That’s 4, not 3 as is often stated.

Can you compare the first or second or third London 15-minute period with a random 15 minutes in the middle of the Asian session? Highly unlikely.

Thank you, that’s a very interesting point. One of the things I was considering was the possibility that, depending on economic reports given in a week, strategies could swing drastically. In a daily, these swings would be averaged out across the day’s trading. I suppose on a shorter scale, more attention would need to be paid to reports like this.
Do you think insights could be found if you were to look at the four sessions over a few weeks (let’s say 20 days of trading, limiting to just those sessions of action, or the session you’d primarily be trading)?

Yes, I think some useful data would come from such a limited study. Lot of traders only trade during Lon or Lon+NY but unfortunately we are all stuck with charts that rigidly respect time, not market sessions - they’re very accurate but its the wrong target. Targeted data is much more useful than random stuff.

At the end of the day, recognise all you need to do is know enough to make more than you lose, not to know everything or even to know more and more.

ideally, you might want to backtest a strategy under a range of different market conditions

maybe easier said than done

the problem with the approach of testing, say, a 15-minute-chart strategy for a short period of time (some weeks), even with a large number of trades, is that the outcomes might be heavily influenced by “how the market happened to be behaving over those few weeks”, which might be very limiting

2 Likes

Yes, that would be the problem. Determining “live” vs “dead” markets for testing would be needed. I guess I’d need to develop a definition for those terms in my study. Thanks for the feedback.

1 Like

2 years are the absolute minimum for a backtest.
You need also a significant number of trades.
After 5 years of experience with algo trading I would never test on less than 4 years and 300 trades for a significant sample.

Hi, What will happen to your strategy if volatility doubles in the near term? Regards Greg

I’m learning through my backtesting that my strategy would get decimated if violatility changes in the near term. I have done a deep look at my strategy and have decided based on feedback from this forum to expand my timeframe.

2 Likes

Try to use higher timeframes and/or learn to use ATR .