Is it considered a viable trading strategy to do the opposite of a consistently losing model?

That is, whenever the model says short, you go long, and vice versa.

Disclaimer: I would never do this. I am just interested in the opinion of other members of this community.


If you do this, you would destroy the value of the statistical tests that you performed on the backtest. You had a hypothesis that the strategy would make money, but the hypothesis was rejected. You cannot say "I will accept the hypothesis that the opposite strategy is successful"; no statistician would agree with this conclusion. In that case, you might as well test strategies at random and trade them in whatever direction (direct or opposite) they seem to work, but it would be an unsound procedure from a statistical point of view, with low chance of success in out of sample data. Some people do that, but I would not consider it valid statistical trading.

What you could do is formulate a new hypothesis, but that hypothesis cannot be accepted yet. It would need to be tested ON NEW DATA, not the data that you used in the backtest. Perhaps, you would monitor the market for a while and see how the opposite strategy does. At some point, you may have enough data to conclude that the opposite strategy works and you can trade it for real.

| improve this answer | |
  • $\begingroup$ Thanks for the insightful answer @alex-c. What is the hypothesis has binary outcomes, that is the strategy's hypothesis was that the prices would converge to value x but the back-testing results showed that the prices actually diverge from value x? Also, I don't see how testing it on new live data would make a difference. Every model is prone to failure when tested live, even if the hypothesis was correct and the back-test results were promising. Can you please elaborate why new data is so important? $\endgroup$ – Moataz Elmasry Sep 4 '17 at 15:59
  • $\begingroup$ @MoatazElmasry there is an enormous amount of literature on the importance of out of sample validation, just look on Google, you'll get more than I imagine you'll feel like reading, including many answers on this and related sites. $\endgroup$ – will Sep 5 '17 at 7:35
  • $\begingroup$ @will out of sample validation doesn't necessarily require new data. Also, I am not doing any training on my model, so out of sample validation techniques like cross-validation don't make sense. $\endgroup$ – Moataz Elmasry Sep 5 '17 at 7:40
  • $\begingroup$ this is an old post, but the core of the question remains unanswered, it being: would it work? and if not, why? saying that it doesn't work just because it doesn't fit your hypothesis is a bit of a lazy argument imo, and doesn't answer how it might fail. $\endgroup$ – k.c. sayz 'k.c sayz' Oct 11 '19 at 7:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.