Three months ago, I tried to construct an advance estimate for 2009Q1 GDP growth, based on the GDP releases for the first two months and the LFS release for the third month. Here is what I came up with:
The 2009Q1 number came in at -5.4%, which put me three standard deviations away. That's pretty embarrassing as far as forecast errors go, but I can at least comfort myself with the fact that at least I published a standard deviation and not just a point estimate. AFAICT, no-one else did.
Below the fold, I'm going to go through the exercise again for 2009Q2.
Update: A glitch has been fixed, giving a number that looks more like what everyone else is publishing.
There are two reasons why my 2009Q1 forecast went awry. The first is that the relationship between the monthly and quarterly GDP series was weaker than usual:
If you take the quarterly averages of the monthly data, you get an estimate for 2009Q1 GDP growth that is 1.3 ppt less than the quarterly number. The average discrepancy has a mean of 0.1 and a standard error of 0.8, so the model was unlucky there. The other is that StatsCan revised some of the monthly numbers for GDP.
Notwithstanding, here is what you get if you apply the same methodology to the updated data:
The model generates a predictive density with mean of -3.4%, a standard deviation of 0.5%, and with an interquartile range of [-3.7,-3.0].
eta: In the original post, I forgot to update the first two months of the quarter, so I had reported an estimate based on the average of January, February and June. That number was -1%, which - as brendon noted in the comments - was way out of line with his and other forecasts.
1% would certainly be a shock - My small-scale model is telling me -3.5%, the Bank of Canada is forecasting -3.5% and consensus from the Banks (or rather the consenus from the Global Insight spreadsheets they all seem to use) is also close to -3%. These estimates are 4 standard deviations from your model, but then again, when are we in the forecasting business ever right?
Posted by: brendon | August 05, 2009 at 12:32 AM
sorry - meant to say 4 stdev from your model's mean estimate.
Posted by: brendon | August 05, 2009 at 12:33 AM
I think what's driving my number is that June saw an increase in hours worked, so the model forecasts an increase in GDP for that month.
Posted by: Stephen Gordon | August 05, 2009 at 07:14 AM
Apparently I was wrong - the post has been edited to correct a glitch...
Posted by: Stephen Gordon | August 05, 2009 at 01:10 PM
hey that looks better (in that it fits with my forecast).
Posted by: brendon | August 05, 2009 at 02:03 PM