Produção Industrial no Brasil: uma Análise de Dados em Tempo Real

The recent literature on real-time data analysis has shown that several economic activity measures go through important revisions over time, limiting the use of those measures. We organize a real-time data set for industrial production (IP) in Brazil and assess the revisions of the series of IP growth, acceleration and gap. We show that the revisions of IP month-over-month growth and three-moving average growth are substantial (average absolute revision of 0.9 p.p. and 0.4 p.p., respectively), although those revisions become less important as the aggregation period increases (for example, for twelve-month growth). Furthermore, we show that the revisions of the IP acceleration (change in the IP growth rate) are also relevant. In order to analyze the revision of the IP-based output gap, we use three methods to extract the trend: HP filter, linear trend and quadratic trend. The output gap revisions are substantial in all methods, with average absolute revisions from 1.3 p.p. to 3.2 p.p. In two out of the three methods, the revision implied changes in the output gap sign in 30 percent or more of the time. In general, both the IP data revision and the sample size increase were relevant sources of output gap revisions, although the results suggest predominance of the revisions stemming from the latter.