How social ties could carry household shocks into credit markets

A household loss could affect credit markets through more than immediate asset sales. Families might help relatives, draw down investments or become less willing to hold risky bonds. But tracing that possibility requires knowing where household responses travel, whose investment demand changes and how readily other investors absorb the adjustment.

In an arXiv preprint, Faisal Quaiyyum of the University of Chicago Booth School of Business develops a framework for connecting these steps. The paper combines observed spending associations, a mathematical model and simulations. These provide different kinds of evidence: the spending results establish a spatial relationship, while the credit-price prediction gains demonstrate what is possible under specified assumptions.

Spending beyond recorded storm footprints

The main empirical analysis covers 48 storm declarations from 2020–2024, with 74,779 county–declaration observations across 1,574 recipient counties. It combines Opportunity Insights spending data assigned by cardholder residence, the Social Connectedness Index, FEMA incident information and NOAA property-damage reports.

Recipient counties lie outside expanded, report-based storm footprints. Their exposure measures social connections to damaged counties, weighted by those counties’ shares of recorded event damage. This is not simply the percentage of friends affected. Because damage shares sum to one, the index describes where recorded destruction occurs rather than preserving a storm’s total monetary severity.

With incident-aligned county reports, a one-point increase in this damage-share-weighted link index is associated with approximately 0.353 percent lower spending four weeks after onset, relative to the preceding week. The regressions account for measured local weather, geographic exposure, production links and other controls, alongside county and declaration effects.

Alternative damage-record constructions produce similar negative estimates, and all 48 leave-one-declaration-out estimates remain negative. Statistical precision nevertheless depends on how dependence is handled: the primary estimate’s calendar-block bootstrap p-value is 0.055, compared with 0.024 under declaration or physical-episode grouping.

These checks do not establish a uniquely social causal mechanism. Social ties could still proxy for evacuation, shared news or other recipient-specific shocks. Later spending estimates are smaller and imprecise, so the four-week association does not establish a persistent decay rate.

Why financial importance changes over time

The model represents transmission through a social graph using two evolving components: an immediate information or attention response and a delayed household-demand response. Each combines spatial patterns that fade at different rates under the assumed diffusion law. Their behavioral interpretations motivate the equations rather than being derived from household preferences.

The key financial step weights household responses by how strongly desired credit holdings change at a fixed price. Those weights are not equivalent to wealth, existing bond ownership or network connectedness. Two counties with similar connections can therefore have different financial relevance, and that relevance can change with the prediction horizon.

Under the model’s nondegeneracy conditions, no single fixed network score reproduces relative financial exposure at every horizon. This need not mean that county rankings always reverse; changing relative magnitudes is enough.

The market mechanism is conditional. Holding intermediary capital and expected payoff conditions fixed, reduced household demand raises the compensation required to hold credit. The paper connects that compensation to the excess bond premium—the part of corporate bond spreads not explained by expected default—through an assumed measurement relation. Capital changes and other pricing channels can alter the total response.

Crucially, lower spending does not establish lower credit demand. Cutting consumption can reduce the need to sell assets, while portfolio rebalancing can occur without any spending change. Desired reductions in holdings also differ from executed sales after prices adjust.

What the prediction gains demonstrate

Recovering the financially relevant response requires independently measured demand sensitivities and household observations that capture the combination those sensitivities select. Comparisons within a storm remove responses common to all recipients. Recovering an absolute aggregate effect therefore also requires an independent aggregate anchor or restrictions supplying that missing component.

Simulations implement these requirements on the observed social network, using hypothetical household accounts, aggregate measurements and simulated fund data. With household loading scales and the market mapping estimated, the dynamic operator reduces error on the propagated credit-premium component by about 30 percent relative to direct, one-step social exposure.

That result is not an actual bond-market forecasting record. The simulated transmission follows the operator’s own diffusion law and includes favorable financial assumptions. The observed spending data do not pin down that law, and held-out predictions of residual state spending contrasts perform close to a zero-response benchmark. The paper’s next empirical step is therefore essential: measure portfolio adjustment alongside spending before claiming a demonstrated household-to-credit-price transmission mechanism.