With the first GDP estimate for Q4 released, I can now compare the actual performance of U.S. GDP with my model's forecast for 2009. A picture is worth a thousand words, so here's a chart:
This is what my model would have predicted in January of last year (well not exactly as I am using revised data. I believe if I used vintage data the prediction would have been less accurate, although in the same ball park).
Note: the green bands represent the 50%, 75% and 90% confidence bands around the central forecast.
Despite improvements in recent months, my model is still calling for (much) below-consensus growth in 2010. However, because of the extremely wild swings the underlying series have experienced, which are due to unprecedented (in the estimation sample) economic dislocations, I am afraid the model could be being stretched beyond its capabilities.
(Technical note:) What is happening precisely is that, depending on the used number of lags of the data, the forecasts can be extremely different. Lags are how many months of data I let my model use: for example, I could use the past 12 months observations of money supply. I had determined previously that 30 months of observations was a good number of lags, but in any case, if I used 12 months, 24 months or 36 months, the forecasts were not very different. This is no longer the case, and it's a big problem.
However, I believe it is also a reminder that 2010 risks being quite surprising to both bulls and bears. It was easy to forecast a recovery last year when the largest stimulus in history was kicking in. It was also easy to forecast the recession (pardon my lack of humility), when the housing bubble popped. 2010 is a much less obvious story: the contribution from stimulus is going to fade and subtract from growth pretty soon, and private demand seems to be recovering extremely slowly.
Getting Involved in Bitcoin
3 years ago