Introduction
In the previous article I mentioned that I had one final improvement to make to my implementation of the K-NN algorthim that is used for predicting the trend of oscillators. The improvement was simple – I didn’t want to select nearest neighbors that were close to others in a chronological sense. Only the fittest within a ‘sliding’ time period, that relates to the depth parameter, is now selected. The SP trading system signal featured in this article is looking pretty good.
Trading System Settings
The trading system settings are slightly different to those described in the previous article and some of the trade bot parameter names have changed. Parameter Trading Delay is now named Signal Delay. Signal Delay was the original name for this parameter and it is a more appropriate name because it relates to more than just the trading delay (I will explain why in another article). The Detrend Method parameter has been changed back to its original name of Price Transform. Yep, I spend too much time thinking about naming things. Detrend Method wasn’t appropriate because there won’t be many actual detrend routines associated with it. Some of the newly implemented price transformations do some long term detrending of price, but that is about it.
Dakota Data Settings
This trading simulation / test run uses ratio adjusted SP futures data provided by Pinnacle Data Corp. From what I have seen, using different back-adjustment methods does not affect the simulation results to any significant degree. This is a good thing because as a general rule, the more sensitive a trading system is to the data, the more likely it is that the system will fail.
Dakota Bots and Swarm Settings
The Prediction Window has been set to three trading days for this test run. Previously the CPredictor was set to predict the trend of the WaveOsc2 indicator two trading days into the future. The Price Transform method is set to one meaning that the raw price series was transformed using arithmetic returns.
- Dakota Equity Engine Settings
The proportion of perfect trading is the metric used to report each trade bots performance to the swarm adaptation routine. The performance is measured over one thousand trading days or approximately four years.
Trading Simulation Results
The screen images that follow describe the results of running the trading simulation.
- Dakota Price, Signal and Equity Charts
- Dakota Trade Report
The trading simulation results are similar to the results of the system featured in the previous article. At this point I am focusing on documentation etc. to make the latest round of developments available to others. When that is complete I will write up some results of modeling other markets using the CPredictor.
Regards,
James
