Reconsidering 'trends and random walks in macroeconomic time series'

We employ a Bayesian perspective to identify the type of prior needed to support the inference that most macroeconomic time series follow random walks. For many of the series considered by Nelson and Plosser (1982) the required prior involves assigning very low probability to trendstationary alternatives. When this prior is relaxed trend-stationarity is generally supported, thus the unit root inference seems inappropriate for these series: despite Nelson and Plosser's results indicating that macroeconomic time series are not inconsistent with the random walk hypothesis, our results indicate that for most series the trend-stationarity hypothesis is much more likely.