Leading Indicators: Not So Good After All

When I did my first evaluation of the Composite Leading Index issued by DOS, it was based on a relatively short sample (from 2005-2008). The results were generally good, with the Leading index approximating movements in real GDP. I've since found a more complete source for the index series, covering 1999-2009. Redoing my forecast models for GDP yielded some disappointing results.

First, fitting the regression became more complicated, with diagnostics revealing serial correlation and heteroscedasticity, which required fitting an ARIMA model (with heteroscedasticity-corrected standard errors) to the data rather than a purely deterministic one. Second, while the Leading Index showed a good fit for most of the sample, the relationship basically breaks down right at the beginning of the downturn in 2008:4. The following shows the model using seasonally adjusted rGDP:


LOG(RGDP_2005R_SA) = 0.365*LOG(IND_LEAD) + 9.992 + [AR(1)=0.973] (R2=0.992)

...and using seasonal dummies:


LOG(RGDP_2005_R) = 0.349*LOG(IND_LEAD) + 10.051 + 0.023*D2 + 0.052*D3 + 0.046*D4 + [AR(1)=0.973] (R2=0.992)

In both models, the residual breaks sharply from the standard error line, which represents about 70% of probable observations assuming a normal distribution. The seasonally adjusted model indicates that the fall in 2009:1 rGDP was a 4 standard error event (a probability of 0.003%), while the seasonal dummy model indicates a 2.8 standard error event (a probability of 0.2%).

In short, this downturn is something out of the ordinary even for a recession. The Leading Index is in fact currently out of step with the economy, and would overstate GDP if used as a basis for forecasting. In comparison, the residual in the short 2000 recession was barely out of the 1 standard error line, and that only in the second model.

To be fair, the Leading Index is not constructed as a direct representative of movements in GDP, but rather an indicator of turning points in economic activity. Also, my data only covers from 1999-2009, which really is just one full business cycle. On that basis, there may be some hope that the DOS forecast of a turnaround in the economy in 2009:3 may come true. But I'm not holding my breath.