1) Are you trading with any stops? Would hypothetical system performance be improved with a stop? If so, what would that stop be?
My trading partner uses relatively tight stops with a trade entry / re-entry strategy that is independent of the trading signal generated by the Dakota systems. I will not reveal his strategies on any public forum.
2) If your hypothetical trades are taken without stops, honestly, could a trader psychologically and economically handle the substantial risk (of a market meltdown or meltup overnight), such trades are exposing you to?
This depends on your personal risk / reward profile. Michael Bryant's Market System Analyzer application is a great tool for helping you work out how many contracts to trade etc.
3) Is there another market(s) you would add, in order to provide diversification?
When I have the resources I would like to model some FX markets. I have just about finished a largish project and will move onto modeling intraday trends next.
4) What size account do you believe is necessary to execute these hypothetical trades?
This depends on the other strategies (such as stop-loss and other trade entry / exit rules) as well as your target return and maximum drawdown at a given confidence level. I am not licensed to offer advise of this nature.
5) Are you using one piece of information about the market, that is, an adaptive swarm based on a single indicator? Wouldn't additional market information, from other time frames, other markets and other indices improve your swarm's performance? If so, what additional information would you want to consider?
System Sp_z0020 is an ensemble signal trader that uses signals exported from 39 Dakota systems. The 39 Dakota systems are based on the WaveOscillator, TrendOscillator, CCI, Stochastic, Cooks Probability, etc. indicators. Some of the systems are intermarket as well (use the VIX, NY-ARMS, stock prices etc.)
6) How many trades, and what length of time did you use to develop this adaptive swarm system?
I used 9 years of data to develop each of the 45 swarms that I currently have running. At least 1 year of walk-forward out-of-sample data was retained for cross validation. Less than 5% of all Dakota systems that passed a series of tests for robustness over the 9 year modeling period were rejected based on performance over the walk-forward OOS period.
7) Do you have confidence that your adaptive swarm is not based on "chance"? What evidence can your provide to this community that factors such as the "data mining fallacy" have not played a major role in swarm choice and swarm performance?
I have a reasonably high level of confidence because I apply a series of tests for robustness and have rejected less than 5% of swarms developed based on the out-of-sample performance. If my rejection rate based on the OOS period was high then I would not have confidence. I don't offer any evidence.
Kind Regards,
James
