Alternative to Buy-and-Hold: Predicting Indices Direction and Improving Returns Using a Novel Hybrid LSTM Model

Abstract

Predicting stock direction is challenging as stock price time series are extremely noisy. Moreover, the widely accepted efficient market hypothesis states that it is impossible to consistently generate excess returns than the market over a long-term horizon. Hence, the best approach for investors is thought to be the passive buy-and-hold strategy in indices. However, some researchers suggest that the market does have a predictable component. This paper’s objective is to provide investors with an alternative predictive system that generates an excess return over the classical buy-and-hold strategy and reduces risk. The authors propose an alternative investing model, Average True Range (ATR) and Momentum-based Long Short-Term Memory online (a-m-LSTM-o), that innovatively uses the technical indicators with Long Short-Term Memory (LSTM). Further, this study experiments with multiple other LSTM investing models using daily indices data of the world’s top five economies. Based on the prediction, the proposed model’s return is 172%, which is significantly higher than the buy-and-hold return of 139%, and it also has a lower drawdown of −49% compared to −51% for the buy-and-hold strategy. Hence, the authors suggest that the proposed model may be a good alternative to the passive approach of the investors.