Lumpy Investment and Filtering Techniques

Abstract

In this paper, we propose the use of the Markov-switching filter to identify investment spikes. In using Markov-switching filter, we apply a first-order two-state Markov-switching mean model to the investment rates de-trended using Hodrick and Prescott's (1997) filter. A Gibbs-sampling procedure is used to produce the marginal posterior distributions of unobserved state variables and model parameters. Among other advantages, this filter allows us to identify multi-year investment spikes. Some investment projects are so large that they last more than one year. Thus, a single annual accounting period would not necessarily reflect the total expenditure necessary to complete the project. Furthermore, even a year-long project need not start at the beginning of an accounting year nor reach completion by the end of an accounting year. We estimate the filter using Compustat data over the period 1988 to 2007 for 504 firms without any missing values in the period. We find that some 86% of firms have lumpy investment using the filter with the 5% level of significance. We also categorize about 12% of firm-years in the sample as having investment spikes.