[matt1]

A better question is how do you know your algorithm is successful? If it predicts correctly 10 days in a row, have you got a winner? It's like those guys that advertise by saying "We outperformed the market by 6.2% on average the last four years." Put enough monkeys in front of a trading terminal and you're going to have some that outperform the market by 6.2%; doesn't mean you should invest your money with them.

[gaius]

Have you ever wondered how every fund manager has a fund in the top 25%? It's easy, they launch several funds in one sector, then after a few years merge them all into the top one, and use its historical performance for the marketing materials not the average of them all.

[mattmaroon]

You can't. You can only know your algorithm was successful. The market is changing, so even if you could prove, given variance and sample size, that your algorithm was +EV and not merely lucky, you would only be proving that it was +Ev and not merely lucky. Not that it will be in the future.

[joshu]

That's why you look at sharpe ratio, rather than absolute return.