Confidence That Pays Off

Abstract

A persistent minority of US households reports negative net worth at every life-cycle stage. Standard structural models explain part of the dispersion through income shocks, asset-price exposure, and consumption commitments — but they leave a meaningful share of the within-group heterogeneity unexplained. The behavioural-household-finance literature has long suspected that non-cognitive capital is the missing factor: psychological resources that govern how households plan, persist under stress, and execute long-horizon financial decisions.

This paper isolates **self-efficacy** as formalised by Bandura — the household head's belief in their own capacity to influence future outcomes through deliberate action — as the non-cognitive construct best placed to explain the residual. The hypothesis runs through dual-process theory: self-efficacy is what supports goal-directed financial behaviour when the environment imposes high cognitive load, distinct from generalised optimism or confidence.

The empirical setting is the PSID 2016 Well-Being and Daily Life Supplement linked to the PSID Family-level wealth and Marriage History files (n ≈ 8,341 supplement respondents; analytic sample restricted to household heads and partners aged 30 and above). The existing analysis applies a serious classical identification stack: Heckman selection correction (addressing voluntary-completion of the supplement), instrumental-variable estimation (using exogenous shifters of self-efficacy that do not run through net worth), structural equation modelling, OLS, logit, probit, and quantile regression. The triangulated estimates recover a robust, economically meaningful association between self-efficacy and household net worth that survives the linear-control specifications.

We propose extending the existing analysis with a cross-fitted Double-ML estimator that partials out the full high-dimensional covariate vector without parametric assumptions, and a Causal Forest CATE estimator that converts the population-average effect into household-specific estimates by life-cycle stage. The proposed methodological contribution, once implemented, would be the first Double-ML estimate of a psychological-capital construct on a household balance-sheet outcome, and would identify the life-cycle stages — particularly those characterised by recent layoffs or divorces — where non-cognitive interventions deliver the largest financial-resilience return.

Data & Methods