By Robert Waldmann
I am fairly sure it is a coincidence, but current US GDP is not disappointingly low compared to an almost incredibly crude forecast based on data from before 1990.
This graph shows the natural log of US real GDP and a quadratic trend estimated using data from before 1990.
According to this graph, current GDP is not anomalously low. Rather, GDP in the 90s and 00s was anomalously high. I am almost tempted to take it seriously, because, with the benefit of hindsight, it seems that GDP was driven up by the dot.com and housing bubbles during those periods.
If taken literally, the trend implies that GDP growth declines by -0.029 percent per year per year.
I was interested in the extent to which deviations from quadratic trends estimated with pre 1990 data might be useful when forecasting. I don't have any particular conclusion.
Here is a crude regression:
- linvr is log real investment minus a quadratic trend estimated with pre 1990 data
- lconsr is log real consumption minus a quadratic trend estimated with pre 1990 data
- the dependent variable d4linvr is linvr minus linvr lagged one year
- l4linvr is linvr lagged one year
- l8linvr is linvr lagged two years
- l4lconsr is lconsr lagged one year
- l8lconsr is lconsr lagged two years
The coefficients were estimated with overlapping 4 quarter intervals, so the standard errors were corrected for serial correlation. Based on this regression it is possible to forecast the change in real investment out of sample giving forecasts pd4linvr.
The regression suggests that investment reverts towards the quadratic trend and also that high growth of consumption is followed by high growth of investment.
Using only post 1990 out of sample data, the forecasts are significantly correlated with actual changes in real investment.
Here is the scatter of out of sample forecasts and outcomes:
Note that the scales are different.
I don't think I have to type that I tried this first with the annual change in log GDP as the dependent variable. That gave similar results within the pre 1990 sample, but the out of sample forecast growth rates weren't significantly correlated with the out of sample actual growth rates.
I'm not sure what to make of my empirical work with 1960s era econometric techniques. I do find the suggestion that investment is strongly mean reverting (as in related to bubbles which burst) highly plausible.