Online Learning for Technical Strategy Portfolios and Statistical Arbitrage

Summary

This paper summary describes an online learning method based on adversarial experts for selecting parameters in a zero-cost portfolio of technical trading strategies. The method combines a collection of historically tested strategies and studies their dynamics using daily and intraday data from the Johannesburg Stock Exchange. It also introduces a hypothesis test for assessing whether the aggregate portfolio exhibits statistical arbitrage, and compares the approach with benchmark portfolio algorithms while accounting for trading costs and slippage.

The reported findings are cautious: the daily strategy did not pass the statistical arbitrage test after costs, and the intraday strategy was not shown to be statistical arbitrage after costs. The summary also says the authors estimate backtest overfitting probability nonparametrically to examine generalization error. It offers no detailed data description, parameter settings, or full test results, so the summary alone is not enough to reproduce the analysis or judge performance beyond the reported conclusions.

Key ideas

• An adversarial-experts online learning algorithm tunes an aggregate portfolio of technical strategies.
• The study examines both daily and intraday data from the Johannesburg Stock Exchange.
• Its statistical arbitrage test accounts for costs, and neither sampling frequency is reported as establishing post-cost statistical arbitrage.
• The paper also considers slippage, benchmark comparisons, and nonparametric estimates of backtest overfitting.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.