2 months, 6 months, 1 year, 2 years? :33:
Given that 2011 had an abnormal amount of natural disasters would you say it would be a bad year to use for back testing?
2 months, 6 months, 1 year, 2 years? :33:
Given that 2011 had an abnormal amount of natural disasters would you say it would be a bad year to use for back testing?
I would say around 10.000 trades…this should cover a bad/good years
According to me 6 months back testing is sufficient. We also can go for back testing with number of trades.
Hello Johnny, I have tested several methods for trading for two years, I have obtained success rates of these methods and 2012 will be the year in which I’ll try those methods in a real account
Good luck
According to Mark Douglas, author of Trading in the Zone, a sample size of at least 20 trades would be sufficient.
If you use this site, sample size calculator, you’ll need approximately 400 trades to achieve 95% confidence with a confidence interval of 5 (so you can expect the real world result to be +/- 5 of the result from your sample). Obviously the higher confidence you seek, the greater the sample size. Fiddle with the calculator to find a sample size you’re comfortable with.
The number of trades required depends on your win:loss rate and the TP:SL ratio.
If your WL rate is 90% but your TP:SL is 1/10 (eg you are scalping) then 20 winning trades in a row is highly likely - buy entirely compatible with a longrun losing strategy.
If however WL rate is 60% and TP:SL is 2/1, then 20 winning trades in a row is highly unlikely. Indeed the ratio of win pips is such that you’re probably fine
Of course when you back test you don’t know your WL ratio - but you do know the TP:SL ratio. I would use the latter as a guide. If you have a good TP:SL then you cab trust a smaller back test sample.
Following on KingKaivar’s comment, it’s not a question of time but trades. The more frequently the system trades the less time required. As a rule of thumb, though, you want to make sure the back test period covers all the major market conditions - trending up/down, ranging, high/low volatility.
As for 2011, the only reason to question results from the year is if there are only 1 or 2 trades that were heavily influenced by one-off events which represent the majority of the system’s performance. But that’s something you need to review in any system test.
Well considering the well known truth that “past performance is not indicative of future results”, the backtesting approach is merely a confidence-building exercise. So whatever magical number of months/years it takes for you to feel good about your system, that’s how far you should go with your backtesting.
I like to backtest for quite a while, using all the data I have available, and looking at the trades without compounding, just to see how the system works in different conditions… ( but I guess thats much easyer with automated trading then manual trading)
To my mind system testing is as much about understanding how a system works in different market conditions as anything else.
I personally back test for at least 5 years if it is a brand new system. Usually 10 years using a commercial testing program, which is always a great tool to have. Then if i am just making some modifications to an existing system then 2-3 years is fine. Obviously the larger the sample size the better. Using a program made for backtesting makes the testing easier and faster. Personally its something i like to do, it builds confidence in your system as well in your ability to execute that system. Also who can argue that watching 10 years of PA for any single currency never helped anyone get a feel for how it moves.
I have gone back up to 2 years on every new trading system. But I have learn that just because it worked back then don’t
mean it will work today. Understanding the trend, support and resistance levels and the overall market conscience, and you’ll be successful.