Should a strategy backtested against three years of tick data continue to produce positive results?


The proof is what it does with 750 randomly selected trading days, over and over again.

In other words how does it do over a three year period of time, starting with any day randomly selected with a random time step. Based on several thousand "seasons" of data, what was the best it performed and what was the worst. Now you will get a better understanding of the capabilities of the system.

Otherwise you may have discovered a method of explaining the past, not a system of predicting the future.


No.

If you have found a mechanism for producing a consistent income, based on past data, then I expect others will too... in which case I would expect the actions of those other traders to lessen your future win rate.


No.

Forex markets are probably one of the most efficient markets out there. The assumption essentially is that the markets are inefficient, which is a very dangerous assumption.

Furthermore, the three years of testing is not enough. This problem is similar to 20 tests with a confidence level of 95%. Approximately one of the tests will show a statistical significance. See https://xkcd.com/882/.

If you try enough many strategies on the past 3 years, you will with certainty invent a strategy that has beat the market. However, this will not mean that the strategy will beat the market in the future.

If you have a priori assumption that the strategy might be good, and 3 years of testing shows it is good, then I would at least consider running the same tests for 10 years, 20 years or 30 years of data. But remember that in any case, you're betting against the efficient market hypothesis, which is a very dangerous bet.

Furthermore, as Peter K noticed, if you invent a strategy that outperforms the market, chances are others will invent it too and thus eat away your excess returns.

No matter what you do, do not take positions that might get you into a debt problem. So, selling short is an obvious no-no. You must have a bound for the maximum losses that the positions you have taken might get you into. A stop-loss mechanism may not be good enough, as it may not react quickly enough.