Pseudo‑panel methods and an example of application to Household Wealth data

Résumé En

Pseudo-panel methods are an alternative to using panel data for estimating fixed effects models when only independent repeated cross-sectional data are available. They are widely used to estimate price or income elasticities and carry out life-cycle analyses, for which long-term data are required, but panel data have limits in terms of availability over time and attrition. Pseudo-panels observe cohorts, i.e. stable groups of individuals, rather than individuals over time. Individual variables are replaced by their intra-cohort means. Due to the linearity of this transformation, the linear model with individual fixed effect corresponds to its pseudo-panel data counterpart. The individual fixed effect is replaced by a cohort effect and the model is particularly simple to estimate if the cohort effect can be itself considered as a fixed effect. The criteria for forming the cohorts must therefore take into account a number of requirements. It must obviously be observable for all the individuals and form a partition of the population (each individual is classified into exactly one cohort); beyond this, it must correspond to a characteristic of the individuals that will not change over time (e.g. year of birth). Finally, the size of the cohorts results from a trade-off between bias and variance. It must be large enough to limit the extent of measurement error on intra-cohort variable means, that generates bias and imprecise estimators of the model parameters. However, increasing the size of the cohorts decreases the number of cohorts observed, which makes estimators less precise. The extension to non-linear models is not direct and only introduced here. Finally, the article provides an application to the French Household Wealth Survey (enquête Patrimoine).