Welcome to NexusFi: the best trading community on the planet, with over 150,000 members Sign Up Now for Free
Genuine reviews from real traders, not fake reviews from stealth vendors
Quality education from leading professional traders
We are a friendly, helpful, and positive community
We do not tolerate rude behavior, trolling, or vendors advertising in posts
We are here to help, just let us know what you need
You'll need to register in order to view the content of the threads and start contributing to our community. It's free for basic access, or support us by becoming an Elite Member -- see if you qualify for a discount below.
-- Big Mike, Site Administrator
(If you already have an account, login at the top of the page)
Is there any formula to estimate the minimum number of trades in a backtest for different timeframes so that the results are statistically significant? For example, for 1-min forex data, how many trades should the backtest generate to make any sense? Any pointers to articles will be appreciated.
Can you help answer these questions from other members on NexusFi?
Generally 30 trades is what people look at "statistically significant" however i like to see hundreds. You can use the standerd error formula to find out how much error there is and determine if you have enough samples. 5% or less standard error is usually good.
Indeed from statistics a minimum of 30 trades is the minimum to show some statistical signifcance. The goal of a long enough timespan is to account for different market states.
But generally speaking, I think you can break it down to 3 areas:
For example, 150 trades over the last 5 months would ( for me) have enough trades, but the timespan is too short. 50 trades over the past 3 years has ( again for me) a long enough timespan, but too little trades. One could also ask yourself if so little number of trades are tradable in itself. On the other hand, I don't need something to work for the past 10 years, if this ever exists at all. Last, If I would trade micro size I would risk far more, while starting outright with futures I would be far more careful.
Last you need to take ‘the jump from the diving board’ at some time, you will always have uncertainty. Just start with smallest amount or unit possible, and increase size gradually only and only after succes, while keeping a very sharp eye on developing drawdown to know when to pull the plug.
One of my worst enemies are my own false assumptions
Be careful about using one minute bars and be sure to take into account the spread. It's better to use actual tick data and capture the actual spread at the time of the trades. Most forex pairs spread change depending on the time of day, news, etc...
Mark Douglas suggests 20 samples as adequate to test the win rate of a trade strategy. You can develop a rolling average spreadsheet that calculates the win rate of the most recent 20 samples (discard 1st result as you include 21st result etc.) thus always providing the most recent market results for the strategy.
I think the most important part of that article is right in the beginning:
Perhaps one of the most frequent questions asked of a statistician is,
"How many measurements should be included in the sample?"
Unfortunately, there is no correct answer without additional information (or assumptions).
I like that because it reminds me that there is no correct answer. Everything you do in trading has to be taken with a degree of uncertainty, and you have to try to plan accordingly.
Good answer. I have been doing many tests and it appears that the sample size depends on actually on itself. meaning that the sample size depends on the quality of sample and not on quantity.