Introduction

The study of dynamics of economic resources is usually based on statistical reduced- form models which notionally decompose individual resources into “permanent” and “transitory” components modelled using time-series methods (e.g. Abowd and Card, 1989). Inter-temporal changes are interpreted agnostically as shocks, usually assumed to be symmetrically and identically and normally distributed. Recent advances allow non-linear shocks propagation (e.g. Meghir and Pistaferri, 2004, or Arellano et al., 2014), parameters with latent heterogeneity (Browning et al., 2010), and models incorporating interactions between wages, hours and job changes (Altonji et al., 2013). Despite such recent advances, Guvenen et al. (2015) argue that “the workhorse model in the literature … fails to match most of the prominent features” of the processes of economic resources. For example, the density of observed year-to-year earnings changes is not normal but “spikey” (many individuals have no changes), heavy-tailed (many individuals experience very large changes), and asymmetric (income losses outweigh income changes). The problem of the standard approach is not surprising as it does not tie earnings changes explicitly to observable events in the individual’s life (such as job losses or job changes, or illness). Moreover, data limitations usually restrict researchers to short observation windows on individuals, and measurement errors in surveyed self-reported incomes confound the effects of genuine shocks. Finally, we note that existing methods allow to estimate the model parameters that describe the stochastic properties of resource dynamics, but they fail to deliver estimates of the individual trajectories of the latent permanent components. Hence the standard model poorly describes an individual’s resource trajectory.

In this WP, we seek to overcome these methodological and empirical challenges. The new SOEP-RV dataset enables us not only to describe life-time earnings trajectories and the distributional properties of changes over time using complete (measurement- error free) income histories, but also to tie them to events in the individuals’ life (e.g. unemployment and health shocks) and the household context using our survey-based contextual data. Where appropriate, we will also complement the analysis using the administrative VVL database in order to increase the number of observations: We dispose of complete and completed income histories of individuals (from the age of 15) who retired in the years 2004, 2005, 2007, and 2010.

Tasks

We will consider the following tasks:

1. Risk factors over the life-cycle
2. The shape of earnings changes
3. Reduced form models of earnings process