Say you've written a strategy. During back-testing, you find out that the strategy is losing money on most of the positions it enters into. Is it viable to use the strategy but instead do the exact opposite of what the strategy tells you in terms of positioning? I understand that such a behavior requires further investigation, but if we ignore that, can you rely on such a strategy?
Yes, just write down a losing strategy, then do the opposite, and you'll be left with a consistently profitable strategy that'll shoot to the moon. ... Seriously though, in my very limited experience, tweaking stuff and doing the opposite may sometimes get you further along the path. It's just that it rarely is the one and only answer you need, but maybe one part of 10-100 answers you need after you've discarded 100-1000 times as many ideas. Note that this doesn't in any way imply brute-forcing your way to any sort of consistency either. Trying too much is the path to overfitting your system.
I agree with your statement that this could potentially constitute over-fitting. But for example, wouldn't you say that a losing model that expects convergence could also be seen as a winning model that expects divergence?
HaHa, many of us have gone through this "I know, I'll just fade myself." No, find a strategy that does work... fading a bad one is not a viable methodology.
Can you explain why? A lot of people don't seem to understand that doing the opposite of a losing strategy != profitable strategy.
Many have actually done this and failed. If this were a realistic dynamic, the goal would be to find the worst strategy you can come up with and do the opposite. Don't believe me?...try it. It's not hard to develop a strategy with positive expectancy. The hard part is trading it with focus, patience, discipline and courage.
I think the problem is consistency rather than the approach. I would have no problem fading a strategy if this turns a negative expectancy into a positive one - the problem is finding a strategy that is consistently bad so that after reversing it and adding slippage/commissions you'll end up with a positive one. In reality you'll end up with two losers cause neither one is consistent enough. Do a basic backtest on something like Take Stock X - if it hits the lower bollinger band (20,2) go long and sell after 5 days. That's a pretty basic strategy - so if there would be any benefit to this entry signal you would make money. Backtest it on a broad universe (e.g. 2-300 stocks for 5 years). You'll see this is a losing strategy after you add commissions and slippage (5$ per trade on a 10k trade unlevered). Now do the opposite - go short after a stock hits the lower bollinger band (exact same parameters, sell 5 days later, 5$ slippage ...) You'll end up with another losing strategy. The problem is that neither approach (selling or buying) is not consistent enough to make money either way.
Effectively op is asking: if opposite of right is wrong then why not opposition of wrong is right. Simple answer: There are millions of ways to go wrong , but only one way to do it right..... 1+1 is only 2 not 3 or 4 or 5....not 105...
It is equally, essentially, difficult to find an equity system bad enough such that reversing it will make it profitable. It is equally, essentially, difficult to find an equity system with a 30% win rate as it is to find a system with a 70% win rate. It is equally, essentially, difficult to find an equity system with a 0% win rate as it is to find a system with a 100% win rate.
You will need to know why the strategy is losing money. Fees... slippage... It could be it's a 50/50 strategy with 50/50 payout but adding fees and slippage makes both ways a losing strategy. Or you might have tight stops that will get triggered either way, also making both a losing strategy...