Quality versus Quantity: Evidence from Machine Learning Option Pricing

This study investigates the relative importance of data quality versus quantity in machine learning (ML)–based option pricing within the controlled Black–Scholes framework. Using simulated datasets, we analyze how distortions of volatility or option prices affect model performance across varying dataset sizes. Our findings reveal that ML models are robust to volatility noise but extremely sensitive to the quality of price data. Deterioration in price data integrity severely degrades predictive accuracy, and this damage cannot be compensated by increasing sample size, except partially for gradient-based methods and only under symmetric, zero-mean distortions. The message is clear: One should prioritize data quality over sheer volume, even when applying advanced ML models. This has important implications for real-world trading and risk management strategies.