The Uncertain Unit Root in the U.S. Poverty Rate

The U.S. poverty rate, like many other aggregate economic time series, shows considerable persistence. It is logical to consider the model involving a unit root to provide a good description of the data generation process for the poverty rate. We pretest for unit roots in annual U.S. poverty rate data for the postwar period to examine its long-run features given the importance of a unit root for economic forecasting, macroeconometric cointegration modeling and Granger causality testing. Applying a number of available test procedures for pretesting on U.S. postwar poverty rate data, we find results that both support and contradict the claim that the poverty rate is a difference stationary process. The poverty rate data are found to be consistent with a unit root hypothesis when the alternative is I(0) with a linear trend. But the null hypothesis of a unit root is convincingly rejected when the alternative of I(0) with a broken trend line for a break at an endogenous point in time is considered. The estimate of the break in the trend corresponds to an information technology shock during the early 1970s.