Yesterday I finished working on a system that took me almost a year to develop. It is fully automated end-of-day stock system, based on pullback type entry for long trades and support penetration for short (both entries use next day market orders ), with trades lasting about ten days. Results shown are on a 150 Nasdaq stock portfolio,investing 10% capital per long and 20% per short position, with a round turn slippage of 0.8%, IB commissions, and margin loan rate included, trading round lots. What is interesting is that the results are more or less the same ( 5% +/- ) and consistent (on quarterly basis) on virtually any portfolio I tried it on, as long as the portfolio is large enough to disperse the risk (more than 100 stocks). It doesn't matter if it's a nasdaq or nyse stock, large or small cap, with good or bad fundamentals...as long as it has enough stocks, the result is the same on every portfolio. I tested it on more than 20 different nasdaq ,nyse, mixed portfolios that do not include any of the stocks that I used to develop the system... tell me what do you think... Is it possible to achieve these results in live trading, are they what I can hope for when I start trading it for real? Since initial position size will be 100-300 shares, opened with market on open order. Is it possible to achieve even better results than simulated? Are slippage settings an overkill ?
Looks Good!, Now put it to work real time with real money and talk to us if results mimic what you backtested, even for six months of real game. Cheers!
If I read this correctly was the drawdown as high as 52.77%? Could you discuss this...I noticed an 80% plus win rate... Can you identify why your short trading is representitive of such a high drawdown...could your results help you refine your system? (note your high ulcer index for the shorts) Could you define the "wealth lab score"? I am not familiar with that. Is 36.07 a good score? What is possible? Have you considered dialing down the exposure on the short trades and/or filtering them better? (not big changes, just slight changes will do the trick as you do not want to over optimize). Try to get the DD to <30% even if you lose a little yield... Michael B.
Looks sweet. Put it to the test and see what she's made out of. This system design stuff really gets me going, it's been a while since I did any programming. But I really want start spending some time on this aspect of trading. Just thinking about it gets me excited.
Looks Great! I had a system like that once, but it bought future bars. I was in a state of shock for a day or so. But the results were too good anyways, I should have known from the start.
Hello, There are a few issues that you should consider adressing before implementing this system. 1. It sounds like the system does not act with the same rules long as short. A lack of symmetry in the rules allows you greater ability to curve-fit results. Hence, the short results (especially) are suspect. 2. Your data set is over 7 years long. Since you have more long trades than short trades, survivorship bias may be an issue (although it is unlikely change results to the full extent of the gains) 3. You have fixed percentages of the portfolio invested in positions. In a time series test, this can increase your ability to curve-fit results due to the varying leverage. 4. Your tests on other stock groups may not be completely independent, and will certainly not be if your system uses any market based variables or requires extreme price changes (which are to some extent systematically based). 5. Your maximum drawdowns were relatively recent. Has something changed? I suggest the following tests: 1. Implement the long side rules for going long, and the inverse of the long side rules for shorting. 2. Implement the short side rules for going long, and the inverse of the short side rules for shorting. ---The market does not act symmetrically for all inefficiencies, so this may not be appropriate, depending on the logic of your system, use your judgement--- 3. Hard to avoid survivorship without a database free of it. One way is to get the average beta of the stocks, the average return of the stocks, and find the market-adjusted returns of the stocks. The difference between this and zero is a very rough approximation of survivorship bias impact on a 100% invested portfolio. 4. Limit the number of positions to something that is usually smaller than the number of positions that the system wants to take (so you have a relatively constant fraction of your portfolio invested). To do this, you may want to test a very large number of stocks (the entire NYSE, entire NASDAQ, or the S&P 500). Alternatively: Position size = portfolio value / number of positions, and rebalance daily. 5. Look to see if similar strategies had poor results recently, and run tests every month or so to see if anything strange is going on. Also, test for a significant difference in results between the early years and the past year or so. 6. Run a test on only the past two years. 7. A chart of the results would be good. You may PM me if you have any questions. Best of luck, Morgan
OK, maybe you've found the "Holy Grail", the secret to riches.....maybe not..... not enough info.......what are your slippage asssumptions in your testing? I've seen many, many methods put out "cha-ching" results in testing, but when you factor in a realistic slippage assumption, the gold mine turns into a porta-potty....
Michael, your observations about drawdown were a real concern for me also when I first saw the results for the portfolio in question, but that is the isolated case for this portfolio because my short strategy on this particular combination of stocks kind of breaks down(everything that can go wrong does that at the worst possible time...). I uploaded it for you and others to review it just because of that fact, on other portfolios it has much better results... ...like the one's below (note that it is tested on completely different portfolio of stocks, somewhat larger with slightly different position sizes to account for larger stock universe, also note that the exposure is normal). Wealth-Lab score is a risk adjusted return measure which I don't take into account too seriously because it is strongly affected by return. I tried to make other measures like Win% and Payoff ratio (avg win/avg loss) as high as possible and as constant as possible over different stocks, while keeping drawdown under control.