Open Access
Article
SciPap-2417
Behavioral Biases in Derivatives Markets: A Machine Learning Approach to Identifying Non-Rational Investor Behavior
by
G Prasanna Kumar 1,* 
, Valluri Venkata Rao 2 and Parla Suresh 3
1 Department of Management Studies, Pragati Engineering College, Surampalem, Kakinada District 533437, India
2 School of Management, MITS Deemed to be University, Madanapalle, Andhra Pradesh 517326, India
3 School of Business, Aditya University, Surampalem, Andhra Pradesh 533437, India
* Authors to whom correspondence should be addressed.
Abstract: Derivatives markets combine leverage, nonlinearity, and short holding horizons, creating conditions in which behavioral distortions may translate into amplified losses and market stress. This study develops an interpretable, episode-based machine-learning framework to detect bias-consistent trading behavior in exchange-traded derivatives and to translate micro-level signals into a market-level monitoring indicator. Anonymized order- and trade-level records are transformed into position episodes and merged with contract characteristics and microstructure controls. Theory-aligned proxy rules generate weak labels for disposition/loss aversion (time-to-close asymmetry, roll-loser patterns), overconfidence (turnover, leverage-at-entry, short-gamma exposure), lottery demand (low-delta preference), and herding (crowding/flow correlation). Models are trained using time-respecting splits with nested cross-validation and randomized hyperparameter search, and evaluated on a strictly out-of-time test window. Tree-based methods provide moderate but stable discrimination (e.g., Gradient Boosting ROC-AUC ≈ 0.62) with acceptable calibration, while explanation analysis indicates dominance of loss-related and leverage/turnover signals over market-state variables. Aggregating episode probabilities yields a daily Bias Index that is positively associated with next-session realized volatility and supports an alert rule capturing a meaningful share of volatility spikes with manageable false alarms. Limitations include proxy-label noise, market-specific coverage, and associative (non-causal) inference. Practical value arises from transparent scoring and channel attribution that can support targeted nudges, suitability prompts, and surveillance prioritization in periods of elevated behavioral risk.
Keywords: Behavioral Finance, Machine Learning, Derivatives, Investor Biases, Options Market, Interpretable Ai.
JEL classification: C45 - Neural Networks and Related Topics, C55 - Large Data Sets: Modeling and Analysis, C58 - Financial Econometrics, G12 - Asset Pricing • Trading Volume • Bond Interest Rates, G13 - Contingent Pricing • Futures Pricing, G41 - Role and Effects of Psychological, Emotional, Social, and Cognitive Factors on Decision Making in Financial Markets
Received: 9 November 2025 / Revised: 22 May 2026 / Accepted: 22 May 2026 / Published: 22 June 2026
