Tweaking The T3 Trading System by Steve Burns
Have you ever wished that your filter-based trading system adapted to shifts and nuances in data trends? You might want to take a look at this.
Conventional averaging or filter-based trading systems have no provisions for adapting to data trend dynamics. Their fitting constants — the parameters in their formulas — are predetermined and remain constant for all data. Analysts routinely overfit these systems to past data, with resulting disappointment when the systems are used.
An alternative to the nonadaptive filter/overfitting dilemma is to incorporate dynamics into trading system-fitting parameters. An intelligent adaptive filter is one that learns from the data and adjusts parameter values.
I use an adaptive filter based on a least-squares regression fitting. The r-squared of the regression ranges from zero to 1, with zero being no success and 1 being perfect success. R-squared works well as an adaptive fitting parameter because its value is determined by the trending data. Here’s how I used it.