The goal of this project was to identify or develop an EST scriptbot that will perform well when the majority of signals are noisy and poor. EST is short for ensemble signal trader. EST scriptbots read in and process trading signals produced by other systems. Version 1.90 of the EST ScriptBot Library contains scriptbot EST-MinRiskAdjPPIP-Filter that was developed with this goal in mind.

EST-MinRiskAdjPPIP-Filter uses an adaptive voting scheme based on signals that meet or exceed a minimum level of risk adjusted proportion of perfect over the lookback period. The proportion of perfect while in position (PPIP) is calculated for each day in the lookback period. The standard deviation of PPIP values is then calculated and a multiple is subracted from the PPIP value corresponding to the last trading day to arrive at the risk adjusted PPIP value.

A summary of the parameter values for the systems built using EST-MinRiskAdjPPIP-Filter follow:

• The minimum level of risk adjusted PPIP is set to zero.
• The adapted lookback period ranges from 250 to 500 trading days.
• The risk adjustment factor is set at 2.0 standard deviations in PPIP.
• The voting thresholds are set to 50%. That is, no adaptive voting is applied.

Ten random signals and three ‘good’ signals from the ATS SP Dakota Systems were used to test the effectiveness of scriptbot EST-MinRiskAdjPPIP-Filter. The random systems output either -1 or +1 to ensure they have a significant impact. Reverse adjusted SP futures data provided by Pinnacle Data Corp. has been used for the traded security. All systems have been lined up so that signals begin about 1/2/1999 and end on 2/24/2010. All systems use a trading delay of 1 trading day.

First of all I will present the equity curves generated by systems built using the average of the random,  random plus good and the good signals. This illustrates what we are up against.

The equity curve produced by averaging the random signals:

Equity Curve - Average of the Random Signals

The equity curve produced by averaging the good and random signals:

Equity Curve - Average of the Good and Random Signals

The equity curve produced by averaging the good signals:

Equity Curve - RiskAdjPPIP Filter Applied to the Good and Random Signals

Now that we have our points of reference, I will present the equity curves produced by applying scriptbot EST-MinRiskAdjPPIP-Filter to the good plus random signals and the good signals.

The equity curve produced by applying scriptbot EST-MinRiskAdjPPIP-Filter to the good and random signals:

Equity Curve - RiskAdjPPIP Filter Applied to the Good Signals

The equity curve produced by applying scriptbot EST-MinRiskAdjPPIP-Filter to the good signals:

Equity Curve - RiskAdjPPIP Filter Applied to the Good Signals

The table below summarizes the results.

Summary of Results

Scriptbot EST-MinRiskAdjPPIP-Filter did help reduce the influence of the random signals. The results were a little disappointing because I was hoping to achieve similar statistics to those obtained by averaging the good signals. However, applying scriptbot EST-MinRiskAdjPPIP-Filter to the good signals produced a better than expected premium over the average of the good signals. Overall, it appears as though EST-MinRiskAdjPPIP-Filter is an effective EST scriptbot.

Regards,

James