There are several EAs (Expert Advisors) in the market claiming profitable results by showcasing backtest data. I have tried tons of free and paid EAs with amazing backtest results. According to the backtest results, these EAs provide impressive returns and very low drawdowns, even when tested over several years of data, yielding positive outcomes.
However, when the same EA is applied to a real account with identical settings, the results are completely different—it blows the account in a day, a week, or at most a month. Are backtest data unreliable or illegitimate?
If you select any period of time and optimise the EA over it, it will give you the best possible return for that period. This is called overfitting and means the back test over that period will be great, but for any other data you have no idea and it often sucks.
The best way to test a system is to optimise it over a period, say 3 months and then back test it over a much longer period and see if the results match. But I don’t know any systems where optimising parameters results in long term profitability.
EA’s being produced for sale are typically backfitted (made specifically to have good backtest results without that meaning anything, going forward).
If you’re betting on horses, it’s dead easy to come up with a backfitted system which you can prove was successful historically (“when the 3.30 race at Ascot on Tuesday was won by the favourite, the chances of the favourite in the 5.00 race at Epsom are twice as high”). It may be true, historically, but it’s also valueless garbage, just like EA’s advertised for sale.
It’s much easier to make money by selling EA’s than it is by trading.
Correlation is not causation.
If you had an EA that was actually profitable, would you want to sell copies of it on the web for a few hundred dollars?!
Mine aren’t. I wouldn’t be able to make a living without them.
I couldn’t possibly ever develop a method with an edge without knowing how to backtest validly; I’d just be guessing, like everyone else who ends up losing money.
But if you buy an EA on the strength of “backtested results” the chances of those being unreliable moving forward (however accurate they were historically) are about 100%, for exactly the reasons explained just above.
I know. But they’re all backfitted garbage for people who trade on hope and guesswork and dream of automation and endless profit without proper trading education and understanding, aren’t they?
If you had an EA that was actually profitable, would you want to sell copies of it on the web for a few hundred dollars?!