Rethinking Impulsivity in Consumer Behavior: An Adaptive Learning Perspective

Abstract

Impulsive consumption is often interpreted as self‐control failure. This study examines whether impulsive choice can also arise as an adaptive exploration strategy under uncertainty. Methodologically, the study uses agent‐based computational simulation, formalizing sequential consumer decisions as a Markov decision process and training policies with proximal policy optimization. Five model architectures compare regret specifications and social‐cognitive inputs from the theory of planned behavior, including subjective norms and self‐efficacy. All behavioral data are synthetic; no human participants or market dataset are used. Training budgets differ by analysis: the baseline and exploration analyses use 50,000 steps, the dedicated ablation and psychological plausibility experiments use 200,000 steps, and a separate clustering check uses 10 parameter‐heterogeneous agents trained for 50,000 steps each. Behavioral outputs are examined using SHAP (SHapley Additive exPlanations) attribution, parameter sensitivity analysis, and K‐means clustering. The full model yields four interpretable behavioral profiles, while the separate agent check identifies three broader groups, providing limited structural correspondence rather than an exact replication of four consumer types. Regret‐related inputs and social‐cognitive variables alter simulated decision patterns, and the exploration analysis shows increasing context dependence during training. These results motivate a computational account in which impulse and emotional regulation interact during sequential choice. They do not establish evolutionary adaptation, human effect‐size replication, or differentiation of initially identical consumers. The contribution is a process‐level extension of TPB‐inspired modeling that generates hypotheses for longitudinal consumer research. Applications to segmentation and intervention require empirical validation and context‐specific calibration.