Happily Under-Insured Ever After? The Role of Beliefs in Household Specialization

Ana Costa-Ramón (Zurich), Ursina Schaede (Tufts), Michaela Slotwinski (Zurich & Neuchâtel) and Johannes Stupperich (Stanford) — “Happily Under-Insured Ever After? The Role of Beliefs in Household Specialization,” presented by Schaede at the NBER Summer Institute, Gender in the Economy, on 28 July 2026 (Day 1, afternoon). No discussant; the presenter waived the ten-minute no-questions rule about five minutes in. Written from the 24 July 2026 draft.

Household specialization is a portfolio decision, and it is a strange one.

One member of the couple stops accumulating market human capital and starts accumulating something that is only valuable inside this particular household. The position is concentrated, illiquid, and exposed to a single counterparty. It pays well as long as the couple stays together and badly otherwise. Any finance person shown this position would ask what it is hedged with and at what assumed default probability.

The assumed probability is where this paper lives. Psychologists established in the 1990s that people are systematically over-optimistic about their own separation risk relative to national base rates. And across many countries women face far worse financial outcomes than men after divorce — which suggests the couple never smoothed consumption across that state for both of them. Put those together and you get the paper’s hypothesis: women over-specialize into the home because they are pricing the divorce state too cheaply.

The obvious problem is that beliefs about divorce are correlated with everything. So: shock them.

The experiment

About 1,500 Swiss female teachers, all currently in a relationship, all mothers, all part-time — the average works two days a week. Switzerland has high maternal labor force participation and overwhelmingly part-time hours, and its pension is closely proportional to lifetime salary, so hours forgone today are retirement income forgone later.

Teachers were chosen for a specific reason that is worth admiring. The design needs women who could increase their hours if they wanted to, and the experiment was run during a teacher shortage — so demand-side hurdles are about as low as they get. If nothing moves here, it is not because nobody would hire her.

The treatment is a documentary-style video with real protagonists, designed to make both the possibility and the consequences of separation relatable, with some subtle pointers toward mitigating the financial risk. The comparison is a control group; outcomes come from follow-up surveys and, crucially, from the employer’s administrative personnel records a year later.

What moves

Two insurance channels, and both respond.

Financially, treated women score 0.3 standard deviations higher on a savings index, and are 55 percent more likely — relative to the control mean — to sign up for an incentivized tool that computes the compensation payments owed within the couple based on work history and future plans. That is not a survey answer; it is a costly action.

On labor supply, treated women increase work hours by about 2.5 percent over the control mean in the employer’s records.

Worth reading that p-value rather than the abstract. Both groups increase hours — this is a population of mothers whose children are getting older — and the treatment roughly doubles the increase, but the difference is significant only at 10 percent. The heterogeneity is where the effect is concentrated and cleanest: women who were least insured against separation at baseline raise hours by 6.5 percent relative to control. Which is the right pattern if the mechanism is insurance, since those are the women with the most exposure to hedge.

The theory is a two-period household model with a lower-earning wife and a higher-earning husband, choosing savings and the wife’s labor market share, with an adjustment cost $\tau$ standing in for human capital depreciation or habit formation. With probability $\phi$ the couple separates in period 2, savings split equally, each spouse keeps their own income — the Swiss regime.

The prediction that makes the paper is asymmetric. Raise $\phi$ and the wife’s labor market investment always weakly rises, but the savings response is ambiguous — because labor supply is targeted insurance (her period-2 income accrues to her alone) while savings get split at divorce and are therefore effectively taxed. Labor supply is such an efficient hedge that it can crowd out saving. And on welfare: correcting over-optimistic beliefs raises the couple’s ex-ante expected utility, while shifting expected consumption utility from the higher earner to the lower one. It is a Pareto improvement in expectation that the husband pays for.

Two objections that improved the paper in real time

The first came almost immediately: separation risk is not exogenous to specialization. If a woman invests more in domestic life, that itself changes the probability the marriage survives, so the thing being shocked is endogenous to the thing being measured. Schaede’s answer is that separation cannot be observed in the administrative data, but a large follow-up battery on relationship satisfaction shows that if anything the treatment makes relationships more stable — and the treated group is much more likely to have sought a conversation with their partner about it.

That answer sets up the second objection, which was sharper. If the intervention lowers the actual divorce probability while the model says the mechanism is a higher believed probability, the moving parts point in opposite directions. Why not model the belief as being about how insured I am rather than about how likely divorce is?

Schaede took the point and had a partial answer ready: the paper carries a parameter for how much income each spouse keeps at separation, and shocking beliefs about that delivers the same predictions. Which is the honest resolution — the treatment is a video about what separation is like, and what it plausibly changes is not “will this happen to me” so much as “what would it be like if it did, and am I ready.” The paper’s title says under-insured, not over-optimistic, and the model would be tighter if it followed its own title.

And there is a limitation the authors state themselves: nobody knows what the correct separation belief is. It is unobservable to the econometrician and to the woman. So this cannot be a paper about correcting an error toward a known truth. It is a paper showing that beliefs are load-bearing — move them, and a real, administratively recorded, retirement-relevant decision moves with them.

(The finding that should unsettle you is not the 2.5 percent. It is the 55 percent take-up of a tool that merely calculates what one spouse would owe the other. These are educated women in a country with well-defined marital property law, and more than half again as many of them wanted the number once someone made the scenario concrete. They were not refusing to insure. Nobody had told them the premium.)