I was having a discussion with someone who runs a fund based on automated trading. I came up with a pretty good algorithm for trading. First he tested it, and then recommended I learn how to do backtesting on the computer. His feedback was that it works in the short run, but it has to work well in the long run to adopt it...I understand that level of conservatism. As for learning it, I went through that rite of passage this weekend, learned to code (somewhat) and started tweaking the algorithm. Now it works great (seems to). But here's my long-winded question. Suppose you develop an algorithm that works really great in the short term (like over the last 6 months). But over the last few years, it would have lost money. Get that? I has blistering returns in back-testing now, but if I'd have started using it a few years ago, I would have lost money. I know that some methods work and then stop working. My question to you is what would you do? How good would a method or algorithm have to be in the short run for you to adopt it, even if it didn't work well in the long run? Just curious, SM
Seems to me on the topic of trading systems that, just like it is safe to have uncorrelated instruments/stocks/... in one's portfolio to mitigate risks and reduce volatility, wouldn't it be safer to have several, uncorrelated trading *systems* available, running every day on live data and therefore generating forward-simulation data on equity returns day in day out, such that as a trader one would be able to select and run the best performing system over a period of time (the idea here is not to switch gear every hour or even day ^^) - as opposed to always running the same single trading system, at the risk of hitting a significant DD waiting at the corner... Wadda think ya'all...?
That sounds like a classic description of a curve-fitted system. Any long-biased system would have done well in the past 6 months but how will it do when the market trends down? In general if you a system that works well only in certain market conditions (trending vs non-trending for example) then you really haven't solved the problem - since you still have the issue of determining what type of market you are currently in without the benefit of hindsight.
Good feedback. Thank you. It is for a trending market, though it identifies shorting opportunities, so I can play a trending market both ways. But non-trending is a tougher nut to crack. I've been trying to tweak it without watering it down for chop. Based on this feedback, I think it will be a good idea to come up with something totally new for a non-trending market. Then I'll run both simultaneously. This should work as long as either method doesn't get it's butt kicked in the wrong type of market...like if maybe it breaks even. Maybe thats the trick...one algorithm for chop, and one for trending that will identify whether you should go long or short. Neither will lose (much) in the wrong environment. Then play both. Makes sense... SM