CYCLICAL FLUCTUATIONS IN AGGREGATE ECONOMIC ACTIVITIES

Cyclical Fluctuations in Aggregate Economic Activities

Cyclical Fluctuations in Aggregate Economic Activities

Introduction

The reconstruction of the historical data of the US main economic indicators after 9/11 has allowed economic historians to characterize US business cycles since 1870. Based on the dating rules provided by the National Bureau of Economic Research (NBER) there is an accepted general belief that the business cycle before the Second World War was decidedly more severe than the post-war cycle. Many authors have emphasized this finding.

Cyclical Fluctuations in Aggregate Economic Activities

Directly related to this issue are the historical changes in the timing of business cycles (duration from peak to trough, and trough to peak) that have also been the subject of recent controversies. Since, unfortunately, no employment or GNP series for pre-war years can be truly comparable to post-war series; all historical estimates presented so far have been built on highly debatable assumptions. For instance, Watson (1994) presents evidence that the change in timing of NBER cycles indeed reflects changing definitions rather than changing economic behaviour. Consequently, a general agreement about business cycle severity and duration before the Second World War seems hard to reach in the near future, and discrimination between the NBER and Romers's estimates is difficult to establish on quantitative terms.

The main intention of this paper is to describe and forecast both characterizations (NBER and Romer's) of the pre-war US reference cycles, using the methodology developed by García-Ferrer and Queralt (1998) for the post-Second World War GNP data. This methodology makes use of two sources of information. First, information about the historical duration of US business cycles in order to decompose cyclical behaviour into one or more of the following categories: short-, medium-, and long-term cycles. Note that this information is different for NBER and Romer's characterizations. Second, for both approaches, the real US GNP series is used as reference activity, that is, its specific cyclical behaviour is used for matching the reference dates of the pre-war US economy.

The empirical analysis of the business cycle regularities involves the controversial issue of detrending (e.g. Canova, 1998). This paper openly recognizes a subjective view on what constitutes a trend. Our results are based on a particular class of unobserved component models represented in the state-space form and estimated by recursive application of the Kalman filter. These models allow us to easily obtain cycles of different length by selecting a parameter known as noise to variance ratio or NVR. Based on a empirical framework, where business cycles are viewed as the sum of cycles of different length, we argue that if we decompose an output series into cycles of different lengths we can obtain a good anticipation of business cycle turning points for both NBER and Romer's reference dates, not only historically, but also in a true ex-ante forecasting exercise.

The paper is organized as follows: the next section describes briefly the US GNP quarterly series from 1875.1 to 1940.4 and both the NBER and Romer's criteria to establish the reference ...