Tuesday, February 2, 2010

More on the uncertainty surrounding 2010 GDP forecasts

This is a follow-up to the previous post, in which I discussed the fact that my GDP model produces a very large range of possible 2010 GDP growth, depending on the number of lags (past observations) of the data.

To illustrate this, I backward tested my model to estimate year-ahead GDP growth as of each January since 2003, in what is called "pseudo out of sample" forecasts. It simply means that I put my model in the situation it would have been at the time, and thus, it cannot use data not known as of that moment. For example, for the 2004 GDP growth forecast, only data published through January 2004 can be used.

I had my model produce forecasts using 8 lags (the past 8 months of data) to 38 lags. Then, I calculated the standard deviation of these yearly forecasts:

To summarize, what this tells is that the uncertainty surrounding the 2010 figure is about five times larger than usual.

The conclusion: either my model is misspecified (which I don't rule out, but it's been very good on a pseudo out of sample exercise), or, fading monetary and fiscal stimulus, combined with an uncertain outlook for private investment and final demand, makes it very difficult to have a strong opinion on 2010 GDP. And it could of course be both.

No comments:

Post a Comment