AI-DRIVEN BEHAVIORAL ECONOMICS IN EMERGING MARKETS: MODELING INVESTOR BIASES AND MARKET ANOMALIES IN NEPAL'S STOCK EXCHANGE (NEPSE)

Abstract

Artificial Intelligence (AI) is having a notable impact on global capital markets, influencing everything from transactions to investor behavior.The convergence of AI and behavioral economics is becoming a defining force in global capital markets, impacting both transaction dynamics and investor behavior. However, the emerging markets like the Stock Exchange of Nepal (NEPSE) are still relatively under-researched in this joint space where informational asymmetries exist, there is not enough institutional investor participation and regulation is fragmented and subject to cognitive biases. This study involves the application of machine learning, natural language processing (NLP), and structural equation modeling (SEM/PLS- SEM) to primary investor survey data as well as secondary NEPSE market indices to detect, measure, and mitigate biases of investors associated with anomalies in the NEPSE market using AI driven analytical models. A sequential explanatory mixed method was used. The questionnaire, consisting of a 5-point Likert scale questionnaire and semi-structured interviews with 22 active NEPSE institutional experts, were used to collect the quantitative data from 547 active investors in NEPSE. The secondary data consisted of day-to-day returns of NEPSE index from 2019-2025. Results confirm that overconfidence bias (beta = 0.431, p < 0.001), herding behavior (beta = 0.389, p < 0.001), loss aversion (beta = 0.312, p < 0.001), and anchoring bias (beta = 0.278, p < 0.01) significantly predict suboptimal investment decisions. The accuracy of investor sentiment classification achieved by the AI-based sentiment analysis with the BERT model is 91.7%. There is a moderation effect between herding and investment decisions when introducing the use of AI and a partial mediation of the relation of herding and loss aversion effect when introducing financial risk propensity.The use of AI in herding and the use of financial risk propensity partially mediate the relationship between herding and loss aversion. NEPSE shows non-random distribution and time varying inefficiency which is consistent with the premise of the Adaptive Market Hypothesis (AMH). It is a study that combines the AI model of behavioral econometrics with direct psychology data of investors from the primary market to Nepal, and proposes a Behavioral Anomaly Detection Framework (BADF) combining the elements of Prospect Theory, Adaptive Market Hypothesis, and Human centered AI Design Principles.