I have make a strategy and do its back tests. I find that for most stocks, it runs well; but a little not well; How shall I do? Maybe I should selected all 'well' stocks and only trade them for this strategy? or I should do more works? Thank you!
^ This. Products can be idiosyncratic, each with their own 'character'. Trade it on the good ones. Just keep an eye on it.
That's not a good recommendation at all. This is the worst kind of curve fitting. If one could just select stocks the strategy is performing on, then producing high returns would be incredibly easy. Stocks aren't separate products either. Try to figure out what causes the differences in performance, is your strategy perhaps biased towards bullish or bearing stocks? Does it only do well in high volatility?
Exactly. Use your results to determine WHY your strategy performs well with certain instruments, then follow up with further study. Look at all conditions. What, why, where, how?
I think it depends what sort of strategy you're running. MMs for example operate differently in AAPL (expensive, relatively cheap spread, liquid) vs. BAC (cheap, relatively wider spread, liquid) vs. less liquid stocks. There are also products in the equity space. VXX for example, has totally different dynamics than {set of non volatility equity products}. If you're just talking about medium or low frequency strategies on S&P 500 or Russell components, you're probably right.
Of course. He did mention stocks however, not volatility or even ETFs. I was thinking more of low and medium frequency trades indeed as you rarely get experienced people designing HFT systems asking questions like that.
Thank your for the ideas and explanations; The key point is that the strategy should be confirmed which situation it is skilled on , which not and why?; after that the situation selected items should be added into the strategy; so the strategy should fit well almost all of stocks; for this course maybe allocating different parameters for different types of stocks is ok? for example (Close(i)-Close(i-1))/Close(i)>0.2 for stock A and >0.3 for Stock B.
I don't autotrade. So take my opinion with a grain of salt... From what I read (and I read alot), autotraders usually agree that backtests aren't always a good indication of future performance. That walk-forward testing is far superior. And real money, far superior to that. Assuming you have a substantial edge in a couple products, and assuming you have autotraded live before, with real money - I would trial the strategy on the favorable products with 1/5th regular size, and see how it performs. No mention was made of look-back period for the back-test, the slippage+spread+commissions factored into the tests, trading hours (liquidity) during the test etc. These all come into play. If all lined up, that's what I would do. Trade and proceed with caution assuming the edge is substantial. Any sign of edge deterioration or substantial capital loss, go back to the drawing board and reassess. The reason I say that is a trader can get lost with endless backtests and shadow-boxing. Obviously an edge has to be real, but backtests have their flaws that only become apparent with real-money, live, walk-forward testing. All depends on your experience level, how comfortable a trader is autotrading, and can they afford risk capital.