Carbon risk and return prediction: Evidence from the multi-CNN method

Abstract

This paper investigates the carbon risk and its role in stocks’ return prediction by identifying the carbon risk information implied in feature engineering. We predict the stock returns with different neural networks, construct the investment portfolio according to the predicted returns and reflect the returns of stocks with different carbon risks through the relevant evaluation of the investment portfolio. Our Multi-CNN method can best collect information on different relationship types and make full use of graph structure data to identify carbon risks. With or without carbon factor, the stock market performance of high-carbon industry is better than that of medium-carbon industry, and the performance of low-carbon industry is the worst. Moreover, our finding is consistent in both Chinese and American markets. Investment should pay attention to carbon risk and requires corresponding carbon risk premium.