Dependência na duração dos ciclos de negócios brasileiros: uma análise para os anos 1947-2017

Abstract

This paper aims to investigate whether there is dependence on the length of business cycles in Brazil. This concept is related to the hypothesis that as the phases of the cycle become longer, its probabilities of reaching the end may increase (positive dependence) or decrease (negative dependence). For this purpose, a series of quarterly product covering the period between 1947 and 2017 is built, using three different data sources. The series is submitted to the BBQ dating algorithm (HARDING and PAGAN, 2003), to identify peaks (points of maximum) and valleys (points of minimum), and to obtain the duration of expansions, recessions and complete cycles. Alternatively, a growth cycle approach is also adopted, which treats the phases as deviations from the growth trend, using the HP-filter (HODRICK and PRESCOTT, 1997). The dependence tests on duration are performed by two approaches: i) a logit model following Ohn, Taylor and Pagan (2004), estimated with the binary series of failures and durations of recessions and expansions; ii) a survival analysis, using the Weibull risk function, as suggested by Castro (2010), applied to recessions, expansions and complete cycles. The results of the first approach point to the presence of a positive duration dependence for recessions and expansions in terms of trend deviations, with average duration of three years. Additionally, it is detected positive duration dependence in expansions after 1980, when considering the BBQ dating. Our survival analysis also indicates positive duration dependence of the complete cycles and to a lesser extent of the expansions

(BBQ dating), with cycles with average duration of approximately six years. For the trend- deviation cycles, there are pieces of evidence pointing towards positive duration dependence

in recessions, expansions, and complete cycles. These results imply that longer phases of the Brazilian cycles are more likely to end, which can contribute to the anticipation in the development of countercyclical policies for the next phase, as the present phase exceeds its expected average duration. On the other hand, it is unlikely that a phase will come to an end shortly after its beginning, which requires greater attention to variables such as unemployment and decline in economic activity during deceleration phases and the increase in inflation in expansions, as these effects may last more than a short amount time.