RDP 2005-07: The Australian Business Cycle: A Coincident Indicator Approach 7. Conclusion

The results in this paper suggest that coincident indices based on the recently developed techniques of Stock and Watson (1999, 2002a, 2002b) and Forni et al (2000, 2001) for estimating approximate factor models with many series are useful tools for studying the Australian business cycle. The quarterly indices are quite robust to the selection of variables used in their construction, the sample period used in estimation, and the number of factors included. Somewhat surprisingly, we find that increasing the number of factors beyond the first does not substantially change the shape of the cycle, but often makes the indices noisier (less persistent). So, while a handful of factors may be required to provide an adequate representation of the data panel, it is not clear that as many factors are needed to form a coincident index. In contrast, the monthly indices are sensitive to the number of factors included in the indices. Two factors seemingly capture different economic cycles so that an index based on only one of these presents a very different impression of the business cycle to one based on a combination of the two. The monthly indices also seem to be fairly robust to the composition of the panel of data.

The coincident indices provide a much smoother representation of the cycle in economic activity than do standard national accounts measures. To the untrained eye, quarterly changes in GDP appear to be largely white noise, at least in the early part of the sample. However, the quarterly coincident indicators are highly persistent and display the type of long swings that one would expect from a measure of the business cycle. Since the coincident indices are essentially a weighted average of the growth rates of the panel of data, this highlights the benefits of assessing the business cycle using a wide range of data series, and using statistical criteria to weight them together.

Notably, the indices do not display the marked decline in volatility evident in Australian quarterly GDP growth, suggesting this decline may overstate the reduction in the volatility of economic activity and at least partially reflect improvements in the measurement of GDP. One consequence of the high volatility of quarterly GDP growth before 1980 is that it identifies many recessions. Some of these appear to be spurious, the result of noise at a time of low, but probably not negative, growth. In contrast, because they present a smoother perspective of the business cycle in the 1960s and 1970s, the coincident indices identify fewer recessions in this period than does GDP. Over the past 45 years, the coincident indices locate three recessions – periods when there was a widespread downturn in economic activity. The three recessions occurred in 1974–1975, 1982–1983 and 1990–1991. These recessions break the past 45 years into four expansions, with two long expansions of over 12 years each, bracketing two shorter expansions of around 7 years each.

It is obviously difficult to offer general conclusions about factor modelling based on data from just one country. However, our results appear to strengthen the finding of Inklaar et al (2003) who show (using European data) that relatively small numbers of appropriately selected series may be able to provide similar results to factor models using much larger panels. A second conclusion might be that a coincident index can often be constructed using just one factor, but this is dependent on the panel of data.