Marie Juliette Lizzardi (Marche Polytechnic University)
Claudia Pigini and Francesco Valentini
This paper presents an approach to address the \textit{limited mobility bias} in employer-employee wage decompositions by introducing a grouped fixed-effects version of the AKM framework. Although the traditional AKM model decomposes wage variation into worker and firm fixed effects, its estimation can suffer significant biases in sparse networks, due to insufficient job transitions. To mitigate this issue, we extend the grouped firm-effects methodology \citep[][, ECTA, 87(3):699–739]{bonhomme2019distributional}, modeling firm heterogeneity as discrete via a data-driven k-means clustering of firms into classes based on wage distributions. We then use this partitioning in the estimation of a standard linear additive model akin to the AKM set-up. However, differently from the BLM framework, our approach relaxes some restrictive sample assumptions - allowing for firms to appear in the dataset with limited temporal presence or to employ only movers - in finite settings. We prove asymptotic validity for this generalized clustering approach under weak dependence conditions. By applying this method to Italian administrative data, we show substantial improvement in network connectivity, significantly reducing the bias from limited mobility.