The current profits in the account would be more than triple what I have now. The average gain per contract traded is still "only" $35, but the win/loss $ ratio is now 1.64. The percentage of trades that win is unchanged at 61%. The t-statistic for trades in NQ is now the unbelievably significant value of 7.48 (currently a respectable 2.12) and the Sharpe ratio of weekly returns on the account is 4.24 (currently 1.42). Obviously lower levels of loss mitigation will lead to results somewhere in between these two extremes. This shows the immense importance of recognizing losing trades fast. The difficulty is that many trades that are down a little eventually win. So maybe this is unrealistic. So I repeated this exercise assuming all losses are the size they really were unless I lost more than $200 per contract. All those losses are truncated at $200 per contract:
The equity curve looks pretty similar. Now profits are around twice their actual level. The average gain per contract traded is $24 and the win/loss $ ratio is 1.08. The t-statistic for trades in NQ is 4.83 and the Sharpe ratio of weekly returns on the account is 3.73. I think this is a realistic goal. Holding onto more winners longer is obviously desirable too, but just cutting losses could be easier. In theory at least.
2 comments:
Were there any winning trades that you would have bailed out of because they dropped to a $200 loss before eventually making money?
Of course the real problem is that setting any parameters based on historic data may not pay off if the future isn't the same shape as the past. Trading patterns always repeat, until they don't.
BTW, if your model is performing better than you are, is there any way to automate you're trading so it exactly trades the model?
*Were there any winning trades that you would have bailed out of because they dropped to a $200 loss before eventually making money?*
$200 per contract is a 0.5% move in the NDX so looking at the model itself there would have to be. My standard stop in the model algorithm is a move in the market of 1.25%. Though recently 1% would probably be better as volatility has been lower. OTOH most of the really bad losing trades where I lost more than $200 per contract occurred when either I made a trade against the model or with the model but the model signal was negated. In these cases stopping the trade should be an easy thing to do in theory.
*BTW, if your model is performing better than you are, is there any way to automate you're trading so it exactly trades the model?*
In the long-term that is something I would like to do. The model is based on running an econometric model together with the kind of rules that can be programmed into typical automated trading systems. So it couldn't be totally automated using the standard platforms. There is also a little judgment still in evaluating the econometric model output occasionally though We could develop an algorihm for that. In the longer term I am interested in working with a fund management firm on developing something that uses the model. We would need a model that could trade S&P futures as the NASDAQ futures are too thin a market outside the regular trading hours - a case of where small traders have an advantage over the big firms. So my goals in the near term are getting more disciplined in actually trading the model and developing a better S&P strategy.
Post a Comment