Consumption and Portfolio Choice over the Life Cycle under Extremely Leptokurtic Distribution of Earnings Changes

Guvenen, Karahan, Ozkan, and Song (2014) sheds new light on the nature of idiosyncratic idiosyncratic income risk. They document the following facts: First, earnings changes display extreme leptokurtosis, meaning that compared to a normal distribution (with the same standard deviation), most earnings changes are very close to zero but few changes are extremely large. Second, there is enormous dispersion in the variance of earnings shocks across individuals: the top 10% most volatile individuals have an average standard deviation of shocks that is 6 times larger than the least volatile 10%. Third, individuals face a negatively skewed distribution of income shocks (with the left tail of the distribution being longer than the right tail). Furthermore, the distribution of earnings shocks varies systematically across different income and age groups along these dimensions. Lastly, the persistence of earnings shocks varies with i) the magnitude of the shock (small ones being more persistent than large shocks), ii) the sign of the shock (positive shocks being more persistent than negative ones), and iii) the percentile of the individual in the income distribution, with positive shocks being almost permanent (transitory) for income poor (rich) households, and negative shocks being almost completely transitory (permanent) for them. All of these features of the data have interesting implications for the consumption savings decisions of households as well as the portfolio allocation between risky and riskless as well as between liquid and illiquid assets. To study the implications of these features of individual earnings dynamics for the consumption decisions of households as well as for their portfolio allocation, we develop and estimate an incomplete markets life cycle model of portfolio choice.Abstract We proceed in two steps. As many of the facts described above are inconsistent with the vast majority of income processes in the literature, we start by providing an alternative. We estimate a parsimonious income process and show that it can match many features of the data. We then use the model to study if some of these features can help us explain several puzzles about the portfolio allocation of households.