At yesterday's close, 'NDX OU' generated an entry signal which I took. In May, I did a cursory review of my models performance over the past few years. This model, due to its poor performance was put on the watch list for possible discontinuation.
This morning, I did a much more detailed performance analysis of this model. I found the results to be so poor, that I closed the trading position pre-market, and have discontinued trading the model.
I find the Profit Factor (PF) statistic to be one of the most telling of a model's performance. During the Historical Build Mode, the PF was 3.68. This means that for every $1.00, the model booked as a trading loss, it booked $3.68 as a trading gain. In other words, for every 1 step back, it made 3.68 steps forward. For my style of trading, I may accept a PF>=2.00, but I really prefer a PF>=3.00.
So the PF of 3.68 during the Historical Build Mode was quite acceptable. During the Historical Test Mode, the PF was 1.64. This is definitely below my minimal threshold, but I decided to trade the model anyway thinking that perhaps this lower performance was just a mild setback or a basing period.
During the Real Time Trading Mode, the PF was 0.74. Anything below 1.00 represents negative performance. So the PF has been in a steady decline from the Historical Build Mode, to the Historical Test Mode, to the Real Time Trading Mode. This is also true for all the other stats shown in the table above.
I should have done this analysis prior to putting the trade on yesterday. Closing out the trade pre-market led to a 0.22% loss. The money management stop for this model is currently at 1.00%. So I sold it at a 0.22% loss to prevent the additional 0.78% loss. I don't know if this particular trade would have gotten stopped out, but I have lost my confidence in the model so I just took this minimal loss and have moved on.
The lesson here for me is to go beyond the cursory performance review of the models that are on the discontinuation watch list, and undertake a more detailed performance review like the one above. Not having done so prior to this morning cost me 0.22%.
Though this model is now discontinued, it can be brought back on line if it can be rebuilt. My intuition strongly suggests that this model is dead.
This morning, I did a much more detailed performance analysis of this model. I found the results to be so poor, that I closed the trading position pre-market, and have discontinued trading the model.
I find the Profit Factor (PF) statistic to be one of the most telling of a model's performance. During the Historical Build Mode, the PF was 3.68. This means that for every $1.00, the model booked as a trading loss, it booked $3.68 as a trading gain. In other words, for every 1 step back, it made 3.68 steps forward. For my style of trading, I may accept a PF>=2.00, but I really prefer a PF>=3.00.
So the PF of 3.68 during the Historical Build Mode was quite acceptable. During the Historical Test Mode, the PF was 1.64. This is definitely below my minimal threshold, but I decided to trade the model anyway thinking that perhaps this lower performance was just a mild setback or a basing period.
During the Real Time Trading Mode, the PF was 0.74. Anything below 1.00 represents negative performance. So the PF has been in a steady decline from the Historical Build Mode, to the Historical Test Mode, to the Real Time Trading Mode. This is also true for all the other stats shown in the table above.
I should have done this analysis prior to putting the trade on yesterday. Closing out the trade pre-market led to a 0.22% loss. The money management stop for this model is currently at 1.00%. So I sold it at a 0.22% loss to prevent the additional 0.78% loss. I don't know if this particular trade would have gotten stopped out, but I have lost my confidence in the model so I just took this minimal loss and have moved on.
The lesson here for me is to go beyond the cursory performance review of the models that are on the discontinuation watch list, and undertake a more detailed performance review like the one above. Not having done so prior to this morning cost me 0.22%.
Though this model is now discontinued, it can be brought back on line if it can be rebuilt. My intuition strongly suggests that this model is dead.
Posts
No comments:
Post a Comment