Title: Long-term forecasting of stock returns: Avoid overly complex machine learning and prioritize benchmarking Authors:  Parastoo Mousavi - Bayes Business School, City St George\'s, University of London (United Kingdom) [presenting]
Jens Perch Nielsen - City, University of London (United Kingdom)
Tatiana Franus - Bayes Business School, City, University of London (United Kingdom)
Abstract: Machine learning is increasingly the default choice for data analysis, often regarded as the only solution. The value of incorporating simple, intuitive models is argued when forecasting long-term stock returns. We show that by focusing on manual optimization, especially in data-constrained environments, simpler models can outperform automated machine learning methods. Our findings highlight the critical role of human oversight in financial forecasting problems and challenge the assumption that automated approaches always deliver superior results.