Nearly Redundant Parameters and Measures of Persistence in Economic Time Series

Many economic time series are nonstationary, apparently exhibiting persistent cycles around a smoothly growing trend. First-differencing removes the nonstationarity, but creates another problem: near parameter redundancy. This paper explores the serious bias and wide dispersion of parameter estimates created by standard Box-jenkins methods applied to weakly trend-reverting series, and shows that this bias can be substantially reduced by the estimation of appropriately specified unobserved components models.