Friday, April 27. 2007
Even more curvefitting
Real GDP vs 1,1-GARCH fit on CFNAI-M3 (1Q lag), OECD CLI (1Q lag) and KRWI (unlagged), 1967-2007 (1987-2007 shown)
Given the mild interest in my previous runs, both the braindead OLS vs the Conference Board's leading indicators and the 1,1-GARCH fit on the Kasriel Recession Warning Indicator, I ran another. I fit quarterly real GDP at annual rates against the CFNAI three-month average and the OECD Composite Leading Indicators quarterly change at annual rates, both lagged one quarter, and the Kasriel Recession Warning Indicator as a coincident dummy variable.
GARCH estimates using the 159 observations 1967:3-2007:1(I have the full table here.)
Dependent variable: Real GDP qAR
QML standard errors
VARIABLE COEFF STDERR T STAT P-VALUE
constant 3.4 0.2 14.8 0.0 ***
CFNAIm3 (lag) 1.4 0.4 3.7 0.0 ***
CLI qAR (lag) 0.1 0.0 3.0 0.0 ***
KRWI dummy -1.8 0.8 -2.3 0.0 **
alpha(0) 0.2 0.3 0.6 0.6
alpha(1) 0.1 0.1 1.1 0.3
beta(1) 0.8 0.1 7.5 0.0 ***
So, a principal-components measure of economic activity and the OECD's indicators would appear to lead real production to some extent*, while Kasriel's indicator (which is available well in advance even of preliminary GDP) seems to coincide. While this fit and the table include the punk 07Q1 number, excluding it does not really change the coefficients of the forty-year fit, though the KRWI loses one *'s worth of significance, since at least on a coincident basis it "worked" in Q1.
Now, despite enjoying a nice regression fit as much as the next fellow, I tend to have relatively little patience for their use to make anything but very general predictions. After all, had I left out the KRWI, or run it lagged one quarter instead of coincident, out-of-sample runs from data through Q4 would have predicted 3.0% and 3.1%, respectively for Q1, instead of the relatively close-to-reality 1.4% that came out of this model. Real Q1 GDP, lest you missed it, was 1.3%.
That said, the CFNAI and OECD CLI are in the can for Q1 and it is not hard to believe the KRWI could continue to register through the end of Q2. If you believe it will, and that this regression is not useless numbercrunching, and that Kasriel's indicator is worthwhile, then the model predicts a 0.9% GDP for Q2.
I think I'll have to check back later. For now, despite greatly having entertained myself, I remain skeptical.
* Naturally, these series coincide more than they lead. This is not a dissertation, it's fun with gretl.
Comments
Display comments as
(Linear | Threaded)
Two things.
The drop in statistical significance you note when Q1 is excluded is surely partly due to the good match between the series in question in Q1. The drop in significance may also be partly due to a reduction in degrees of freedom.
The other think is that, while the model looks for a 0.9% pace of growth in Q2, there is some evidence (see John Kitchener's write-up of his current GDP model) that what the economy given in period 1, it taketh away in period 2. That is to say, if there is a downside miss vs the model (or trend) in period 1, then the best-odds forecast is an upside miss in the next period. Put in that context, the 0.9% estimate cries out for massaging upward.
The drop in statistical significance you note when Q1 is excluded is surely partly due to the good match between the series in question in Q1. The drop in significance may also be partly due to a reduction in degrees of freedom.
The other think is that, while the model looks for a 0.9% pace of growth in Q2, there is some evidence (see John Kitchener's write-up of his current GDP model) that what the economy given in period 1, it taketh away in period 2. That is to say, if there is a downside miss vs the model (or trend) in period 1, then the best-odds forecast is an upside miss in the next period. Put in that context, the 0.9% estimate cries out for massaging upward.
We agree on the source of the coefficient significance "improvement": the (coincident) KRWI fits the Q1 GDP disappointment. As I said, it "worked" in Q1, and I remain skeptical, since I am not a monetarist. Reducing DF by 1 probably wasn't an issue, though, since the full sample includes 159 quarters.
I agree, there is a taketh-away tendency in the GDP series. That partly is an artifact of its construction, since a high/low reading in Qn will be harder/easier to beat in Qn+1, and partly reflects economic verity, insofar as an inventory build/drawdown tends to encourage the opposite in a subsequent period.
Recessions are special: during a recession, weakness begets more weakness. That's why I have been scratching around to figure out whether it's possible I am wrong about the soft landing. If we are in the process of landing hard, everything normal changes for a while.
I agree, there is a taketh-away tendency in the GDP series. That partly is an artifact of its construction, since a high/low reading in Qn will be harder/easier to beat in Qn+1, and partly reflects economic verity, insofar as an inventory build/drawdown tends to encourage the opposite in a subsequent period.
Recessions are special: during a recession, weakness begets more weakness. That's why I have been scratching around to figure out whether it's possible I am wrong about the soft landing. If we are in the process of landing hard, everything normal changes for a while.
Quicksearch
Categories
Syndicate This Blog
Blog Administration
Powered by
Recession probability index, 1967-2007 Although the sluggish GDP growth rates of the past year have produced quite an obvious move up the recession probability index, it is still far from the point at which we would conclude that a recession has likely s
Tracked: Apr 27, 22:11