You can keep scrolling left until your chart runs out of bars but when is enough?
Have you picked an arbitrary number of periods to test, such as 24 bars, 365 bars, 52 bars…?
An arbitrary number of sample trades such as 20 trades, 50 trades, 100 trades?
Or have you just decided that when you can’t be bothered to do any more back testing, that’s enough?
And what is more relevant, recent results or historical results?
I know that in the end, it doesn’t really matter because nothing predicts the future. Even if your back test has glowing results, that doesn’t mean the future will give you something similar, such is the will of the markets…
Speaking as an inexperienced newbie, I would suggest that it’s more important to be as objective as possible in back testing different systems. That is, for each system the same extent of backtesting for each should be applied whether that’s one month or one decade. As we all know us humans are terribly good at finding what we want to see - if one has invested time and effort into creating a trading system then we will always tend to see the areas in price history where it would have worked and gloss over the parts where it would not have done well.
I generally prefer 1000+ sample trades across different years and different pairs. I normally trade 11 pairs (majors + Yen crosses), so will backtest each pair, from 2001 to 2012. If the vast majority of pairs show a significantly positive profit factor (generally 1.2+) then I’m usually happy. It would suggest that my backtest is robust. I can’t remember where I read it, but sample sizes above 1,000 show significant diminishing returns in accuracy. I know that Newspoll and other polling companies survey approx 1,000 to 1,200 voters every month to get a snapshot of the political intentions of 20 million Australians.
I’ve heard different opinions. Mark Douglas, the author of Trading in the Zone, suggests a sample size of only 20 trades, which seems REALLY low. I hope a good statistician will chime in here.
During my thinking time in the shower, I came to the conclusion that no amount of back testing can guarantee positive results for the future. I could waste my time back testing forever but as long as I get a good sample of positive trades and negative trades with an overall positive result, that’s all one can ask for. Hedging my bets on multiple currencies pairs can’t hurt either.
1-2 years of quality data is generally what I use. I find it easier to breakdown and analyze things like sessions, seasonality and economic factors that might’ve skewed the results(against/for me). The system will have to keep evolving around a dynamic market so I don’t feel the need to run a static backtest across to much data.
Years of data and trade numbers can give very different results depending on what time frame you want to trade.
dan82au’s 2 years of data will give a huge number of possible trade on a 1 minute chart, but Kevin LaCoste and JJ6845616’s 10+ years may not have very high trade numbers if trading a monthly time frame.
The smaller the number of trades you use the more likely you are to encounter ‘curve fitting’ (getting very good fit to historical pattens but potentially poorer results in future).
Backtesting 10 years of a D1 system is a whole different prospect to testing 10 years of a M5 system.
Even with just the M5 system someone scalping 20 times a day will have a whole different testing perspective from someone who is swinging one trade a day.
Lets not even get into wether it’s a high probability low reward or low probability high reward system or some kind of mix thereof.
If you scroll back to where the bars end then you can go to the history center and download even more data (at least on MT4).
Of course the only person who can really answer this is you, you’re the one that has to be happy and confident with the system you are trading.