Combining CNN and Grad-CAM for profitability and explainability of investment strategy: Application to the KOSPI 200 futures

The use of AI in financial markets is no longer a special case but a universal phenomenon. Fund managers are seeking to improve returns with AI, and financial institutions are striving to improve work efficiency through AI. While most financial AI papers focus on better results (accuracy or profitability), recent trends suggest that simply introducing AI into financial markets is no longer desirable. Major countries are establishing guidelines that require not only efficiency but also transparency (explainability) and responsibility when financial institutions use AI. In this study, Grad-CAM (gradient-weighted class activation map) was applied to the convolutional neural network (CNN) model to identify the model’s important features (explainability) for decision-making. For empirical analysis, KOSPI 200 futures contract data were used, and returns using the important features extracted from Grad-CAM were compared with those from the benchmark strategies. As a result of backtesting in 2021, the proposed strategy showed higher returns and lower volatility than the benchmark strategies. In addition, the proposed model in this study indicated that profitability and explainability can be satisfied at the same time in financial markets. The proposed model appears to help fund managers use AI more responsibly.