Statistical power of intervention analyses: simulation and empirical application to treated lumber prices
Abstract
Timber product markets are subject to large shocks deriving from natural disturbances and policy shifts. Statistical modeling of shocks is often done to assess their economic importance. In this article, I simulate the statistical power of univariate and bivariate methods of shock detection using time series intervention models. Simulations show that bivariate methods are several times more statistically powerful than univariate methods when underlying series are nonstationary and potentially involved in cointegrating relationships. In an empirical application to detect the long-run price impacts of the voluntary phase-out of chromated copper arsenate in pressure-treating southern pine lumber for residential applications, I find the multivariate methods to be more powerful as well. I identify highly significant long-run price increases of 11% for two of three treated southern pine dimension lumber price series evaluated using multivariate approaches. The univariate method detected a long-run increase only for the third product, and the statistical significance was weak, although comparable, in magnitude to the first two products.