Learning zero-cost portfolio selection with pattern matching

TL;DR

This paper introduces a zero-cost portfolio selection method using pattern matching and adversarial learning, demonstrating its effectiveness on JSE data and its potential for intraday trading despite transaction costs.

Contribution

It extends adversarial agent-based learning to zero-cost portfolios with a quadratic approximation, applying it to real market data and comparing its performance to standard benchmarks.

Findings

Speed advantage from analytic solutions of mutual fund separation

Strategies remain profitable after transaction costs and slippage

Patterns in JSE data can be exploited collectively without predictability of individual assets

Abstract

We consider and extend the adversarial agent-based learning approach of Gy{\"o}rfi {\it et al} to the situation of zero-cost portfolio selection implemented with a quadratic approximation derived from the mutual fund separation theorems. The algorithm is applied to daily sampled sequential Open-High-Low-Close data and sequential intraday 5-minute bar-data from the Johannesburg Stock Exchange (JSE). Statistical tests of the algorithms are considered. The algorithms are directly compared to standard NYSE test cases from prior literature. The learning algorithm is used to select parameters for agents (or experts) generated by pattern matching past dynamics using a simple nearest-neighbour search algorithm. It is shown that there is a speed advantage associated with using an analytic solution of the mutual fund separation theorems. It is argued that the expected loss in performance does not undermine the potential application to intraday quantitative trading and that when transactions costs and slippage are considered the strategies can still remain profitable when unleveraged. The paper demonstrates that patterns in financial time-series on the JSE can be systematically exploited in collective but that this does not imply predictability of the individual asset time-series themselves.