Hi Jerry, I liked some of your suggestions. I have started working with Profit 8. With Profit 8 you can build models using the R^2 performance metric. In prior versions of Profit there were mostly profit based performance metrics (many different ones). Using R^2 with random data partitioning is working very well for me. I was thinking along the lines of creating some 'artificial' price series to help identify when a given market is trending or oscillating as well as identify lower risk trades. I haven't thought this through yet, but I will be pursuing this because the benefits are potentially very substantial. Constructive criticism or just bouncing ideas around is of great value in my opinion. Kind Regards, James
Jerry, I looked up your records and I now finally understand your thinking. You have (had) an NGO (NeuroGenetic Optimizer) license from eleven years ago. While we still sell the NGO (one went out the door today as a matter of fact), what we are discussing here is Profit and Dakota, not generalized regression modeling tools, and thus your comment "If BioComp no longer markets such products than as you point out a comparative benchmark is impossible." is correct in this regard. Not wanting to be too commercial, but I think its appropriate in this context that you take a read at the product page to compare capabilities of the offerings: http://www.biocompsystems.com/products/index.html You won't see the NGO there, but the EMS III does use it. The NGO has been somewhat replaced by iModel, iUnderstand and Process Modeler using "Mesh" modeling rather than neural. Presently that whole suite circa 2000 is being reworked using our Intellect 3.0 architecture. Profit does use the Mesh modeling regression engine, but the technology is only a part of creating timing signals. Thanks, Carl
Signal for Thu Oct 30, 2008 MOC: Short Trading Instructions for Thu Oct 30, 2008 Sell 2 ESZ8 MOC Open Position: Long 1 ESZ8 @ 835.25 Closed Net Profit / Loss: ($2,292.50) Note: ($amount) indicates a loss. Hypothetical trading history is attached. Kind Regards, James
I am going to have to put today's signal on hold until I do some checks. A couple of others that run my systems have the signal as remaining long today. It is possible for some divergence to occur due to differences in data. I will update here as soon as I can. Kind Regards, James
Carl, Thanks I thought of this too. I know I haven't had any 'problems' with my data, but the time of day that we update our data can have an impact because of fixes to incorrect prices. Given that I am outnumbered as to the 'correct' signal for today. I have backed up my systems and I am restoring them from Oct 9 and will re-run them several times. We don't get divergences very often so I want to take this action to ensure I am posting the best that I can. I could be the noise injection as you say, either way this is the best action to take. Thanks, James
There is no change to the signal for today. I am confident that my data is correct and restoring the systems as at Oct 9, 2008 and bringing them up to date produces the same signal. Kind Regards, James
Signal for Fri Oct 31, 2008 MOC: Short Trading Instructions for Fri Oct 31, 2008 None Open Position: Short 1 ESZ8 @ 961.00 Closed Net Profit / Loss: $3,995.00 Note: ($amount) indicates a loss. Hypothetical trading history is attached. Kind Regards, James
Congrats on being back in the black again, hope it stays that way! (the addition of stops should help)
Thanks! The plan with the stops is to adjust the number of contracts traded as well as the size of the stop-loss depending on the volatility. Greater volatility means less contracts traded with a greater stop-loss level. The goal will be to produce a smooth equity curve. I have a good understanding of why different copies of my system have diverged slightly and I am doing something to reduce the potential for divergence and increase the degree to which the overall system can adapt at the same time. Basically, instead of having 45 Dakota sub-systems with 1 Dakota meta-system I am going to have 45 sub-systems, 10 meta-systems and 1 master system that is simply the average of the signals from the 10 meta-systems. Each of the 10 meta-systems will be similar to the current single meta-system, however, the parameter ranges will be wider. I may end up needing more than 10 meta-systems. Kind Regards, James