Abstract: Background risk is conventionally measured from the dispersion of realized earnings within groups of similar workers. Households act on the risk they believe they face, and the two are nearly unrelated: across matched cells, realized earnings-tail asymmetry explains one percent of the variation in the asymmetry workers report about their own earnings. They also behave differently. Realized group tail risk, measured on 177,000 matched worker pairs at cell reliabilities near 0.9, shows no association with participation, holdings, or the risky share. Perceived tail risk does: workers who see more downside than upside in their own earnings move retirement assets toward safety, those who see more upside move toward risk, and the perceived variance, entered alongside, predicts nothing. The relation holds within a worker as beliefs change, survives controlling for elicited risk tolerance, and replicates in a separate account. Portfolio levels track scale; portfolio actions track shape.

Abstract: Using transaction-level data on 189,530 Chinese retail investors over the 2013–2016 boom and bust, we measure experience as counts of large realized gains and losses and find that it reshapes the disposition effect, the tendency to sell winners and hold losers, in an asymmetric way. A two-standard-deviation increase in accumulated large losses raises the propensity to sell winners rather than losers by about 23% of the baseline effect, while an equivalent increase in gains lowers it by about 21%. Composition-matched simulations show the asymmetry is not a mechanical artifact of building experience from past trades. To explain both the baseline effect and this asymmetry, we build a memory-based recall model in which similarity-weighted retrieval of past outcomes guides selling. A single retrieval channel reproduces all three patterns without shifting preferences.

Abstract: Heterogeneous discounting is a workhorse explanation for wealth inequality among similar households, but models calibrate its distribution for want of direct measurement. Two Japanese panels ask an identical intertemporal-choice question annually for fifteen years (56,050 elicitations). I measure that distribution: a persistent component with a standard deviation of 10.6 percentage points, stable for decades, behind noise that leaves any single elicitation with reliability one-half. Correcting the noise—averaging a household’s answers, or instrumenting one wave with another—doubles the slope of financial wealth and securities participation on the elicited rate; per standard deviation of the persistent component, the corrected associations are 3.9 percentage points of participation and about a third of financial wealth, and they replicate in a cohort recruited a decade later. An illustrative life-cycle accounting fed the corrected component attributes 70% of late-career wealth divergence to it; fed the one-shot moment, a third.

Abstract: Using the 2018–2022 Survey of Income and Program Participation, I measure pension and retirement lump-sum distributions that are not rolled into another retirement account. Involuntary job separation is associated with an increase of about one percentage point in the annual non-rollover probability, relative to a control-group mean of 0.8 percent. The estimate is similar across covariate adjustment, matching, state-by-year fixed effects, and reason-restricted samples—including on one common sample—although the design is observational and contains 26 treated cash-out events. The association is concentrated among retirement-account owners with little pre-separation liquid wealth. Among 4,551 reason-coded separator person-years, involuntary separators and voluntary quitters have similar lump-sum receipt rates; among 87 receipts, estimates suggest that involuntary separators are less likely to roll over, but exact and unweighted inference is inconclusive. End-of-year liquid assets show no detectable average decline, while annual measurement precludes inference about within-year sequencing. Complementary PSID evidence documents consumption and net-worth losses after displacement. The findings connect job loss to retirement-account leakage and motivate evaluation of rollover defaults, emergency savings, and displacement-contingent liquidity policies.

Abstract: When manufacturing expands in a market economy, marriage tends to come forward: industrial work raises earnings and with them marriageability. That channel requires wages to respond to local labor demand. We ask what happens where they do not. In China before the 1990s jobs were assigned, wages followed national grids, and the work unit allocated housing, held the hukou, and issued the certificate required to register a marriage. Linking a newly assembled city–year panel of firm registrations to marriage histories in the China Family Panel Studies, and using China’s Third Front industrialization program to generate variation in local industrial composition, we find the opposite sign. Among people still unmarried at twenty, a one-standard-deviation increase in the goods-producing intensity of their city lowers the probability of having married by 25 by 1.7 percentage points. The association runs through state and collective employment and is absent for privately owned goods production, which points to administrative allocation rather than to earnings.

Abstract: Do labor-market conditions at school leaving shape how households allocate saving decades later? I link 4,120 women in Japan’s JPSC to job-openings ratios at predicted graduation. Better entry conditions raise regular first employment and raise the later share of saving designated for the spouses—most precisely for the wife—rather than for family-common, child, or other purposes; the wife–husband split is unchanged, and the husband’s conditions carry no detectable effect. Wife-designated flows predict growth in her individually held assets, though treatment effects on those stocks are not established. National-cohort estimates are stable in sign but sensitive to persistent-process inference; within-year prefecture variation is directionally consistent and imprecise.