Behavioral finance has pivoted to AI and fintech since 2017: a systematic review of 166 studies shows AI/ML, sentiment analytics and crypto research now dominate, but the literature is fragmented and lacks causal, real‑time and human‑vs‑algorithm comparisons.

Main Finding

Technology—especially AI, machine learning, social-media sentiment analytics, and crypto/blockchain infrastructure—has become central to behavioral finance research since 2017. The systematic review and bibliometric analysis of 166 Scopus-indexed studies (1991–2025) shows emergent research clusters where technological intermediaries and algorithmic processes both mediate and amplify investor behavioral biases, and identifies systematic gaps for comparative, real‑time and algorithm‑human interaction studies.

Key Points

• Sample and timeframe: 166 Scopus-indexed publications spanning 1991–2025.
• Methods: PRISMA-based systematic review combined with bibliometric mapping using Biblioshiny and VOSviewer.
• Major thematic clusters identified:
  • Fintech platforms and digital trading systems that mediate bias in real time.
  • Robo‑advisory services and investor interactions with algorithmic decision systems.
  • AI/ML prediction of behavioral patterns and automated detection of biases.
  • Sentiment analysis and social‑media analytics as proxies for investor psychology.
  • Cryptocurrency and blockchain markets where algorithmic trading and information asymmetry intensify behavioral distortions.
• Temporal trend: marked acceleration in technology-centric behavioral finance literature after 2017, concentrated in AI-enabled sentiment modeling and crypto market analytics.
• Content gaps and future directions:
  • Direct comparisons of human versus algorithmic trading decisions and outcomes.
  • Empirical study of investor behavior when interacting with robo advisors and opaque AI systems.
  • Causal investigation of whether technology shapes, mitigates, or amplifies biases.
  • Development and validation of real‑time, data‑driven bias detection models using big data, ML, and behavioral tracking.
• Limitations noted by the study: reliance on Scopus-indexed literature (coverage bias), heterogeneity of methods across studies, and potential lag between technological diffusion and published research.

Data & Methods

• Review protocol: PRISMA-based systematic review to identify, screen and include relevant literature (resulting set = 166 papers).
• Bibliometric tools:
  • Biblioshiny (R/ Bibliometrix) for descriptive bibliometric indicators and temporal analyses (publication trends, authorship, journals).
  • VOSviewer for mapping co‑occurrence, co‑citation and thematic clusters (visualization of research clusters and keyword networks).
• Analytic outputs:
  • Network maps of keywords/authors/institutions to identify emergent clusters.
  • Temporal analyses showing growth in technology‑oriented subfields post‑2017.
  • Content analysis of themes and identification of conceptual and empirical gaps.
• Data scope and constraints: study limited to Scopus-indexed publications up to 2025; heterogeneity in study designs in the underlying literature prevented extensive meta‑analytic effect aggregation.

Implications for AI Economics

• Market microstructure and efficiency:
  • AI and algorithmic decision systems change information diffusion and trading dynamics, potentially reducing some frictions while creating new, algorithmically amplified biases and feedback loops.
  • Research should quantify how algorithmic strategies affect price discovery, volatility, liquidity, and tail risk in markets with heterogeneous human-algorithm interactions.
• Model risk and systemic risk:
  • Widespread use of similar AI models and sentiment signals can increase correlation of behavior across agents, raising systemic risk; regulators and modelers need diagnostics for model‑homogeneity risk.
• Forecasting and risk management:
  • Real‑time ML-based bias detectors (sentiment shifts, attention surges) can be integrated into risk systems to improve early‑warning signals for mispricing and herding.
  • Comparative evaluation of human vs. algorithmic portfolio decisions can inform when automation improves outcomes and when it introduces novel risks.
• Policy and regulation:
  • Findings motivate policies on transparency, auditability and accountability of robo advisors and AI trading systems (disclosure, stress testing, and algorithmic impact assessments).
  • Market surveillance should incorporate AI-driven sentiment and behavioral analytics to detect manipulation and information cascades in social-media‑influenced markets, especially crypto.
• Distributional and welfare effects:
  • Technology can democratize access (lower costs, robo advice) but may also exacerbate informational asymmetries between sophisticated algorithmic traders and retail investors—research should quantify welfare and distributional impacts.
• Research agenda for AI economics:
  • Controlled field and lab experiments comparing human vs algorithmic decision-making under identical information and interfaces.
  • Causal identification of how interface design, explainability, and disclosure affect investor reliance on and reaction to AI recommendations.
  • Development and benchmarking of real‑time bias detection algorithms using multi‑source (order‑book, social media, news, behavioral tracking) data and open evaluation datasets.
  • Policy simulations assessing regulatory interventions (transparency, circuit breakers, algorithmic diversity requirements) on market stability and welfare.

Summary: The reviewed literature maps a rapid convergence of behavioral finance and AI/fintech research since 2017, highlights both mitigating and amplifying roles of technology on biases, and calls for comparative, real‑time and policy‑oriented research to manage the economic and systemic consequences of algorithmically mediated investor behavior.