Two-Stage Sector Rotation Methodology Using Machine Learning and Deep Learning Techniques

TL;DR

This paper introduces a two-stage machine learning approach for sector rotation, predicting ETF prices with RNNs and ranking sectors to outperform traditional portfolios.

Contribution

It presents a novel two-stage methodology combining feature selection, RNN-based price prediction, and sector ranking for improved investment strategies.

Findings

Echo State Networks outperform other RNN models in prediction accuracy.

The methodology beats equally weighted portfolios in long-term performance.

Model robustness is confirmed across different lookback and lookahead windows.

Abstract

Market indicators such as CPI and GDP have been widely used over decades to identify the stage of business cycles and also investment attractiveness of sectors given market conditions. In this paper, we propose a two-stage methodology that consists of predicting ETF prices for each sector using market indicators and ranking sectors based on their predicted rate of returns. We initially start with choosing sector specific macroeconomic indicators and implement Recursive Feature Elimination algorithm to select the most important features for each sector. Using our prediction tool, we implement different Recurrent Neural Networks models to predict the future ETF prices for each sector. We then rank the sectors based on their predicted rate of returns. We select the best performing model by evaluating the annualized return, annualized Sharpe ratio, and Calmar ratio of the portfolios that includes the top four ranked sectors chosen by the model. We also test the robustness of the model performance with respect to lookback windows and look ahead windows. Our empirical results show that our methodology beats the equally weighted portfolio performance even in the long run. We also find that Echo State Networks exhibits an outstanding performance compared to other models yet it is faster to implement compared to other RNN models.