Is there such thing is too BIG a backtesting sample size?

Beginner Questions
1

I’ve been backtesting baseline indicators on Tradingview using NNFX rules. But then I found out that someone had set up a script that does this automatically, and each backtest goes back to the beginning of 2010.

I know that market environments change over time and certain indicators will work better at some times than they do at others, but I was still struck by how the larger sample size affected winning percentages.

For example, when I backtested the Kijun Sen at length setting of 45, I found that for GBP/CHF it generated a winning signal on GBP/CHF 65% of the time over the past (roughly) two years. However, according to the script, the same indicator/setting/pair only wins 44.07% of the time over the last decade.

If you were comparing indicators and settings, which number would you consider more telling/important? Is this smaller sample size too small, or does the larger sampling take too much data from a period of time that is outdated or irrelevant to today’s market?

2

Hi, if you use more data, the results of the backtest are more reliable and the strategy is more resistant to market changes.

Period data that you believe may be out of date or irrelevant at the moment may be current in the future.
Just look at the current market situation and the ATR indicator, three months ago no one would have thought that the situation with the virus would increase volatility on the currency market as in 2008. Regards Greg

1 Like
3

Great points! Thank you!

4

It seems to me that one month is enough to test any strategy. The main thing is that the strategy or approach is right for you and you are well oriented on how to use it in practice…

1 Like
5

If the rate is 70% or higher, then you can take the job. If the indicators are unstable, then you should either check the settings or remove it from your trading plan.