# Earnings Losses after Job Loss

This is a companion web app to the paper "Understanding the Sources of Earnings Losses After Job Displacement: A Machine-Learning Approach" [paper ] by Andreas Gulyas and Krzysztof Pytka.

This app estimates 11 year cumulative earnings losses conditional on individual characteristics. Please choose below the worker and firm characteristics, and press the button Estimate Earnings Loss below to obtain the earnings losses.

The estimates are based on the machine-learning methodology described in our paper. The underlying data comes from Austrian social security records from 1984 - 2017. The continuous worker and job characteristics are categorized in deciles according to the overall distribution of male workers employed on 1st of January. Please see the paper for a detailed definition of the variables. Due to computational constraints, fewer options are available. Please see the paper for more detailed results.

Conceptually, we estimate the conditional average treatment effect using the following difference-in-difference design: $y_{it} = \tau(\mathbf{z}_i) \mathbf{1}(t\geq t^*)\times D_i + \theta(\mathbf{z}_i) D_i + \gamma_t(\mathbf{z}_i) + \epsilon_{it},$ where $y_{it}$ is annual earnings, $D_i$ is an indicator equal to one for a displaced persons, $t^*$ the displacement year $t$ the current year, and $\mathbf{z}_i$ a vector of individual characteristics. In this app we report the 11 year cumulative earnings changes as measured by $11 \cdot \tau(\mathbf{z}_i)$

Estimate earnings losses conditional on worker and job characteristics.

*We follow the earnings loss literature and condition on workers with at least 2 years of job tenure and firm size greater then 30. Because the decile categories are defined for the overall employed population, earnings losses cannot be estimated for low values of job tenure and firm size. See the paper for details about the sample construction.