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)
When you are backtesting a strategy, how many trades is a good sample size? How many trades is considered enough to rely on? So far I have backtested 20 trades on a new strategy I am developing. I don't have any fancy backtesting software, I am just manually filling out an excel spreadsheet based on analysis of the chart.
So far, after 20 trades into backtesting, my strategy is profitable, and I am now temped to trade it live with real money, starting small, with just 1 contract in case of failure. My backtesting is complete with stop loss placement so I feel like its ready to go with risk mitigation, but want some input from the community here.
I am planning to keep backtesting my strategy another 10 trades at least but how far back should I go? Is 30 trades enough? Backtesting this way is very time consuming. What do you think?
Can you help answer these questions from other members on NexusFi?
market have various regimes/periods
- calm versus volatile
- trending versus ranging
depending of the type of stratergy and the type of market it best
performs in, you need more or less back testing. you want to be
sure that favorable and less favorable periods were tested. it's
ok, not to make money if the type of market is not good, but you
don't want to go bust during that period, that is what you want
to test..
if your approach is very systematic, then a automation is recommended
i did that for several members and quite often it turns out that a
strategy that looks manually (quite good on excel on paper or screen)
does not do well, when doing systematically...
if it's simple, automate it and run it through a year of data..
it well learn you a lot
This is not the way to backtest. I recommend you get Tradestation. You can also paper trade your idea in Ninjatrader.
There is no hard and fast answer as to how many trades. It will depend on how big the edge is (bigger edge, less trades), how many variables you have (more variables more trades), and considerations around your beliefs or Bayesian reasons as to why a system works. The number 30 I believe comes from statistical studies around confidence interval and if I recall the Chi square distribution.
I would typically look for at least 30 trades per year and 10 years of performance data. But, this may not be the way that every system would be evaluated. With less trades, you would need to go back further in history and if you have more trades you might need less history or if you have Bayesian beliefs then that might influence how far back in history you go, as well.
Thanks for the reply.
I am looking at only the 30 minute chart, on CL crude oil.
1 round trip trade per day based and analizing the last 48 hours, with analysis of the same timeframe each time. Trade execution goal is to complete the round-trip trade within about a 12 hour trading window.
So far I have tested the last 20 trading days.
I don't really know if it is possible to automate because I am using subjective/relative analysis of trend and momentum of the last 48 hours to give me my entry signal. I have my criteria but it is not numerical, it is basically current momentum and trend relative to historical momentum and trend.
I guess since oil has been uptrending lately I should test it back when it was downtrending huh?
@ClearTrades I think, in that case, trading it with Ninjatrader market replay makes sense and paper trading it, i.e. real-time sim account, will be most helpful.
The heuristic of using 30 samples comes from confidence levels between the normal and student t distribution when sampling.
It has absolutely nothing to do with trading or most other real world problems.
IMO you need to think in more of a Bayesian way than in using the tools for sampling from known distributions with all kinds of assumptions on the variable.
If you take 1 sample, then another, then another each new sample is giving you information about the system.
10 samples with 7 winners of course does not tell you this is a "winning" system but it also isn't meaningless.
The main problem though is like rleplae mentioned. If we know anything for certain with trading it is that the distributions will change.
Don't get too hung up on sampling from stale data and last year's distributions. Like right now I am not even sure how much value samples have from a few months ago when fed fund futures had hugely different distributions than today.
30 trades is not even in the ballpark. You want well over 100. If you can't be bothered to even do that before throwing money at it, then you will have some very quick and hard lessons with this trading game.
Keep testing for some days. You'll be glad you did, I'll bet this strategy will revert to being no better than random. Strategies that work are not easy to come by. They take a lot of time to find. You will spend a lot of time, which will seem like wasted time as you churn through idea after idea that doesn't work out.
Save your money and your mental capital for those strategies that you have throughly tested out.
Use only MR(Market replay) method for backtesting for at least 12-months back. This is the only reliable backtest method. Anything else is fake and fails.
Thanks-