Core Projects - AI & Digital Tools for Financial Decision-Making and Compliance

Traditional financial education programs generally improve knowledge but often fail to produce lasting changes in financial behavior. The emergence of large language models (LLMs) such as ChatGPT creates an opportunity for more personalized financial learning. This study aims to examine whether interactive LLM-based learning enhances financial literacy and financial decision-making more effectively than alternative methods. The study consists of a between-subject experiment with five groups: two chatbot-based interventions (GPTproxy, a general-purpose LLM assistant; and FinTutor, an LLM financial tutor), web search, text reading, and a baseline that skips training. Participants complete financial literacy tests and behavioral tasks before and after a dedicated learning phase, in which treated groups use their assigned tool to study core financial topics. This design allows for the evaluation of how each method affects three outcomes: financial literacy, saving preferences, and understanding of risk diversification. Finally, we aim to assess whether chatbot interactions outperform the other learning tools, and whether FinTutor is more effective than GPTproxy.

Tax compliance is critical for the functioning of modern economies, yet it is often undermined by discrepancies between individuals' stated support for taxation policies and their actual compliance behavior. This research explores the connection between individuals' stated support for redistribution and tax policies and their actual compliance with those policies, focusing on how self-awareness, memory, and personal beliefs influence behavior. Drawing on behavioral economics, the research tests the hypothesis that the act of expressing support for taxation policies can create a binding psychological commitment, thereby improving subsequent compliance. The study will assess the relationship between stated preferences and actual behavior by measuring the amount participants are willing to donate from their own earnings in a redistribution task. The expected results include identifying the conditions under which preference elicitation can effectively increase compliance with redistributive policies, and understanding the role of self-consistency, recall, and social norms in shaping individuals' actions. The findings will inform the design of low-cost, scalable interventions that could improve tax compliance, which in turn could strengthen public goods provision, foster trust in institutions, and support collective action on societal challenges such as inequality and climate change. By bridging the gap between expressed beliefs and actual behavior, this study aims to enhance both fiscal policy effectiveness and democratic legitimacy.

The digitalization of finance has transformed how investors access and process information. Digital investment platforms present investors with multiple screen-based sources of information, requiring them to allocate limited attention across competing disclosures. Amid growing concerns that sustained exposure to digital technologies may affect attention spans, we investigate how investors allocate attention to investment-related information on digital finance platforms and whether digital-native investors—defined based on their exposure to digital technologies rather than on demographic attributes—exhibit distinct attentional patterns. In our experiment, digital finance investors view investment opportunities on a screen, while their gaze is tracked by state-of-the-art webcam-based remote eye tracking technology. This allows us to obtain direct measures of attention, thereby objectively identifying how much information investors attend to and which types of information attract greater attention. The results will shed light on the cognitive processes underlying investor information processing in digital finance.