Enhancing the accuracy of stock return movement prediction in Indonesia through recent fundamental value incorporation in multilayer perceptron

Abstract

Purpose – The study aims to explore how integrating recent fundamental values (RFVs) from conventional accounting studies enhances the accuracy of a machine learning (ML) model for predicting stock return movement in Indonesia. Design/methodology/approach – The study uses multilayer perceptron (MLP) analysis, a deep learning model subset of the ML method. The model utilizes findings from conventional accounting studies from 2019 to 2021 and samples from 10 firms in the Indonesian stock market from September 2018 to August 2019. Findings – Incorporating RFVs improves predictive accuracy in the MLP model, especially in long reporting data ranges. The accuracy of the RFVs is also higher than that of raw data and common accounting ratio inputs. Research limitations/implications – The study uses Indonesian firms as its sample. We believe our findings apply to other emerging Asian markets and add to the existing ML literature on stock prediction. Nevertheless, expanding to different samples could strengthen the results of this study. Practical implications – Governments can regulate RFV-based artificial intelligence (AI) applications for stock prediction to enhance decision-making about stock investment. Also, practitioners, analysts and investors can be inspired to develop RFV-based AI tools. Originality/value – Studies in the literature on ML-based stock prediction find limited use for fundamental values and mainly apply technical indicators. However, this study demonstrates that including RFV in the ML model improves investors’ decision-making and minimizes unethical data use and artificial intelligence-based fraud.

Keywords

• Fundamental value
• Artificial intelligence
• Multilayer perceptron
• Stock prediction
• Market efficiency
• Machine learning