I have been testing several trading systems. All of them are profitable on the long side, and fail on the short side. Survivorship bias filters: I limited my data to the past 3 years (weekly) instead of to the last 11 years as during the first tests. However, I still get the same results. Any explanations of what is going on/how to fix it? PS. My code is fine.
This is common. Most markets have a long term long bias that is difficult to overcome by most systems.
Easy solution for this problem, Program your system to only execute trades in the direction of the long term trend...
You may find that Cheesy's solution helps but doesnt solve the problem if the market's downward moves have different characteristics to the up moves (many do). This can be caused by laws, different participants (who goes short vs who goes long) and different psychological responses. Why not use this system for Up only maybe with an overall trend filter to keep it out during bear or early bear periods and seek a different system (or faster responding version of the same system) for Down only.
I think that if you use a long term time frame like a daily trend, trade on a shorter time frame like the hourly or 4 hour or whatever, the system should not execute the short term down trends if the major trend is still up. Unless you are looking at monthly charts, that trend you have will change relativly soon and you will have to program it to take shorts only.. What time frame charts are you using to execute your trade?
With commodity futures, I find that the long trades perform much better than the short trades. However I also find that when you take all trades (long and short), the result is better still. Here is a representative example of this kind of behavior. Notice that run #3, which takes all trades, beats runs #1 and #2. (You may need to right click and Save Target As ... onto your hard disk, then open it with Windows Picture and FAX viewer. Sometimes these image files display very badly inside a web browser window.)
If you are dealing with more than price in your coding, then you can consider a possible cause that you have not sufficiently taken into account. Look at the part of your code that deals with accumulation and distribution. You may have it backwards for the beginning of shorts or you may have just ommitted the effect. Hori demonstrates this for you quite nicely with his results. In hori's case he fails to deal with more than price.
axcel, I have a similar problem with some of my strategies on the e-minis. Is is valid to combine one strategy that does well going long with a totally different strategy going short to form a good trading system? (given that both long and short strategies have back tested to be profitable). Does a trading system need to have symmetry to be profitable in the future?
On this question the only opinion that matters is yours. If you like it, do it. If you don't, don't. You can ask yourself the question: is the long-only system good enough that I would be comfortable trading it all by itself, standalone? You could also consider whether the short-only system is good enough that you'd be comfortable trading it all by itself. If the answers are yes and yes, you may decide you feel extra good trading the two standalone systems, simultaneously. But this is for you to decide.