Suitability of using technical indicator-based Strategies as potential strategies within intelligent trading systems

Abstract

The potential of machine learning to automate and control nonlinear, complex systems is well established. These same techniques have always presented potential for use in the investment arena, specifically for the managing of equity portfolios. In this paper, the opportunity for such exploitation is investigated through analysis of potential simple trading strategies that can then be meshed together for the machine learning system to switch between. It is the eligibility of these strategies that is being investigated in this paper, rather than application. In order to accomplish this, the underlying assumptions of each trading system are explored, and data is created in order to evaluate the efficacy of these systems when trading on data with the underlying patterns that they expect. The strategies are tested against a buy-and-hold strategy to determine if the act of trading has actually produced any worthwhile results, or are simply facets of the underlying prices. These results are then used to produce targeted returns based upon either a desired return or a desired risk, as both are required within the portfolio-management industry. Results show a very viable opportunity for exploitation within the aforementioned industry, with the Strategies performing well within their narrow assumptions, and the intelligent system combining them to perform without assumptions.

Classification of intraday S&P500 returns with a Random Forest

“…The remaining features are technical indicators, including the changes in the 1-, 3-, 5-and 10-day momentums, the relative strength index (RSI), the moving average convergence-divergence (MACD), moving averages (MA) and Bollinger bands (Di Lorenzo, 2013;Hurwitz & Marwala, 2011). The technical indicators MACD, RSI and Bollinger bands are transformed into trading signals instead of just using their values directly.…”

Optimising a targeted fund of strategies using genetic algorithms

“…In order to ensure enough facets are available to trade between, data is generated with specific underlying patterns [6], and is then overlaid with an Energybased noise function [6]. These datasets (or pseudodata) are then traded on with a simple trading strategy, based upon their underlying facets [6].…”

“…can also be modified such that the first term returns no error if the performance exceeds the bound set by the target (so returns zero if [D][ '] is less than T, or if [D][R'] is greater than T, creating equations(5)and(6).…”