Abstract

This study investigates the interconnectedness of economic sectors in emerging markets and applies a hybrid machine learning approach, CNN-LSTM, to predict sector index returns in Indonesia, Malaysia, and Thailand. Our findings reveal the interdependence among asset classes within each market and demonstrate that CNN-LSTM can extract non-linear relationships better than traditional statistical analyses. We find that while the CNN-LSTM approach improves prediction accuracy in most sectors, the noisy path of the interconnected structure reduces forecasting performance in Malaysia. Moreover, sectors that are statistically unconnected to the network failed to achieve precise predictions in the Thailand stock market. However, these same sectors provided improved prediction scores in the Malaysian stock market. This indicates that the effectiveness of CNN-LSTM in capturing sectoral interdependencies may vary depending on the market context and the degree of sectoral connectivity. Our study contributes to the literature by highlighting the importance of non-linear relationships in predicting asset returns in emerging markets. Our findings suggest that incorporating the interactions between asset returns, and especially connections between economic activities, should be used for testing the asset pricing model. Our study offers a unique perspective by focusing on the interconnectedness of economic sectors and the use of machine learning techniques. These insights can inform investment decisions and contribute to the broader literature on asset pricing in emerging markets.

You must be a member to download the full article.