Computational Agency In the Portfolio Society: How AI Literacy Reshapes Household Financial Decisions In China

Background and Purpose: In the contemporary “portfolio society” where proactive financial market participation becomes a hallmark of daily life, the intersection of technological advancement and household asset allocation is reshaping social stratification, as financial returns increasingly dictate class mobility. Bridging digital capital and Actor-Network Theory with household finance, this study investigates the association between AI literacy and household financial investment behavior in China. The primary objective is to determine how AI literacy develops “computational agency”, the capacity to navigate complex financial markets via human-computer interaction, and how this agency reshapes financial decision-making.

Methods: The study utilizes a nationally representative sample of 11,225 observations from the 2023 China Household Finance Survey (CHFS), which encompasses 29 provinces and 258 counties across China and provides comprehensive demographic, employment, and financial data. The independent variable, AI literacy, is measured using a 5-point ordinal scale that assesses individuals’ understanding and practical use of AI. The dependent variable is household financial market participation, measured as a binary indicator of whether the household possesses financial products. To unpack computational agency, we evaluate three mediating variables: risk attitude, self-rated financial literacy, and financial attentiveness (all measured on 3-point scales). Controlling for individual (e.g., age, education), household (e.g., income, debt), and regional characteristics, we employ Probit models and Average Marginal Effects (AME) for baseline analysis, alongside Coarsened Exact Matching (CEM) for dealing with endogeneity problems, and mediation tests by Sobel and Bootstrap. Finally, we employed a sub-sample test to investigate the heterogeneity of higher education, debt, and income.

Results: Descriptive statistics reveal a limited financial market participation rate of 13.3% among the sampled households. Baseline estimations demonstrate a significant positive association between AI literacy and investment behavior: a one-unit increase in AI literacy corresponds to a 0.9% higher probability of participating in financial markets. Mechanistic analyses confirm that this association operates through computational empowerment, which reconstructs households' preferences, endowments, and attention. Specifically, high AI literacy significantly enhances risk tolerance, improves financial knowledge acquisition, and optimizes the allocation of cognitive bandwidth toward financial information. Furthermore, heterogeneity analyses utilizing education, income, and debt as moderators reveal an inherent duality. While AI literacy fosters epistemic inclusion by predominantly benefiting individuals lacking higher education, its advantages remain tethered to material constraints, primarily increasing market participation among middle-to-high-income and indebted households.

Conclusions and Implications: This research contributes to the digital equity and financial capability framework in social work by illustrating how AI endows both challenges and reinforces traditional social stratification. The study concludes that AI literacy is not a neutral equalizer; rather, its empowering potential is tethered to existing material resources. For policy and practice, these findings suggest that addressing the algorithmic divide requires more than technical training, and it necessitates comprehensive interventions that mitigate the underlying structural and material inequalities of the digital era.