Households perceive income as more persistent and less dispersed
BDE Paper

Households perceive income as more persistent and less dispersed

Households perceive more persistent income and less uncertainty than their income histories imply, Bank of Spain research finds. The gap changes simulated precautionary saving, wealth accumulation and protection of consumption against shocks.

Beliefs diverge from income histories

Henrique S. Basso, Olympia Bover, Julio Galvez and Laura Hospido compare expectations and realized income in the Spanish Survey of Household Finances.

Their sample covers 3,468 households and 8,823 household-wave observations from 2014, 2017, 2020 and 2022.

Respondents allocate ten probability points across five intervals for income growth over the next twelve months.

The authors estimate state-dependent autoregressive and nonlinear quantile models, accounting for coarse responses and concentration in the middle interval.

Baseline autoregressive persistence averages 0.9998 in expectations versus 0.6853 in realized income; innovation dispersion averages 0.0411 versus 0.4722. Perceived dynamics also display weaker nonlinear dependence on current income.

Smaller buffers weaken consumption protection

Correcting classical measurement error at a reliability ratio of 0.80 narrows the persistence gap by about 46 percent and the dispersion gap by about 57 percent.

Reporting corrections and horizon checks do not eliminate the differences.

In the life-cycle model, consumption insurance is 0.781 under realized dynamics, 0.730 under subjective dynamics and 0.779 after focal-response correction.

It falls to 0.653 when households save according to corrected subjective beliefs but experience realized shocks; that counterfactual assumes biased beliefs.

Different measures, unresolved explanation

The authors do not establish whether the gap reflects biased beliefs or superior private information.

Their recovery of distributions within survey intervals depends on model restrictions.

They therefore argue for distinguishing perceived risk from realized dynamics, not replacing one dataset with the other.

Report an error