Adaptive Autoregressive SP500 Trading System

Introduction

This article features an adaptive 'trading system' based on the good old autoregressive model. The system trades the very short-term daily trends of the SP 500 stock market index. The three terms (weights) of the model are modified walking-forward bar by bar by the swarm adaptation engine and they each range from -0.333 to +0.333. Thus, the system is highly dependent on the swarm adaptation engine.

The model is very basic. The predicted change in price (PD) is calculated as follows:

PD = Term1 * (Price(t) - Price(t-1)) + Term2 * (Price(t-1) - Price(t-2)) + Term3 * (Price(t-2) - Price(t-3))

If PD is positive then the system goes long and vice versa. The simplicity of the model and the uniform ranges across the terms means there is little opportunity for curve fitting prior to the system being run. Although, there was one optimization step taken. Initially the system was run with a performance lookback of 250 trading days. The equity curve was looking inconsistent so the run was stopped and the performance lookback was increased to 1,000 trading days. No periods for the performance lookback were tested between 250 and 1000 trading days.

System Settings

The trading system simulation was run using BioComp Dakota and SP 500 stock index data from 1980 to present. This period featured a number of very different market regimes.

The Delta Period is set to 1 which means bar by bar changes in price are used. If the Delta Period was set to 2 then the first change in price would be calculated by taking the last closing value minus the closing value two trading days ago and the second change in price would be calculated by taking the closing value one trading day back minus the closing value 3 trading days back.

Performance is calculated using the proportion of perfect while in position (PPIP) equity engine. The Performance Lookback period is set to 1,000 trading days and the system trades on the current close of the market. A system of this nature could be used to trade leveraged mutual funds that closely track the S&P 500 index. Note, I am not suggesting that this would be appropriate for this particular system.

Trading Simulation Results

The equity curve was reasonably consistent as the system managed to adapt reasonably well over various changes in market regimes. It was interesting to see what quadrants the tradebots occupied while the system was running. Changes in market regimes were easily identified.

The percent of perfect was reasonable at 8.9%. However, the system only had a slight edge on the market with approximately 50% profitable trades. Perhaps a slightly more sophisticated model, along the same lines, would result in a more substantial edge.

Regards,
James