Why Does GDPNow Model Sometimes Move Counter to Economic Releases?

I am sometimes puzzled as to why the Atlanta Fed GDPNow Model moves counter to what one would normally expect following economic news releases.

For example, on December first, construction spending hit an 8-year high. On the same day the ISM was a huge disappointment.

I expected the net overall result of those to be a small net minus, with construction adding to GDP and ISM subtracting.

I commented on that idea in GDPNow Forecast for 4th Quarter Dips to 1.4% Following Weak ISM and Strong Construction Reports. Here are a few snips.
Construction Spending Beats Estimates

The consensus estimate for construction spending was +0.6% but the actual month-over-month reading was +1.0%.

Construction Contribution to GDP

As noted earlier today, this morning's ISM report was a disaster. (See Manufacturing ISM Contracts; Lowest Reading Since June 2009; Glimmers of Hope Extinguished). But the construction report looked good.

Combined with ISM spending, I would have expected the net result of today's reports to be an overall small minus.

Actual GDPNow Forecast Results were dramatically different.

GDPNow Forecast Sinks 0.4 Percentage Points to 1.4 Percent.

Construction Subtracts From GDP vs. Prior Estimate

Compared to the November 27 report that included new home sales, construction appears to have subtracted approximately 0.09 percentage points and ISM roughly another 0.31 percentage points. Other reports in between may have contributed but those should be the main factors.

I have a question into the Atlanta Fed regarding construction contribution to GDP.
My Question to Atlanta Fed
Construction Spending was up 1% month-over-month today. I would have expected this to add to your GDPNow forecast.

Instead it subtracted.

Bloomberg Econoday made this comment: “The housing data in this report are not only favorable but the pop higher in related permits, in data previously released with the housing starts & permits report, hints at further gains ahead. This report points to a solid fourth-quarter contribution from construction.”

Instead it appears the construction report subtracted 0.09 vs. the prior estimate. Can you explain?

Reply from Atlanta Fed

I received a reply from Patrick Higgins at the Atlanta Fed on December 8. I missed his reply as it somehow went into my spam bucket.

Higgins writes ...
Hi Mike,

This far away from the GDP release, the forecasts of nonresidential structures and residential investment depend importantly on an estimate of a dynamic activity factor very similar to the Chicago Fed National Activity Index (CFNAI). As of December 1, the estimate of that factor for November (and the forecast for December) was heavily influenced by the ISM Manufacturing Index since there was very little data for November released at that point. Like the CFNAI, the dynamic factor has mean 0, standard deviation 1. As of the GDPNow release on Nov 25, the model had forecasted values for the dynamic factor of -0.06 for November and -0.05 for December (i.e. both were essentially 0, or trend). These changed to -0.72 for November and -0.31 December after the data on December 1. Since then, the estimates of the dynamic factors for November and December have returned close to their November 25 readings. You can see these values in row 5 of the tabs Residential and NonresStructures of the Model. You can also see how they feed into forecasts of the monthly data but clicking the formula bar for any of the teal shaded cells. For example, cell FV9 of tab Residential.

If you simply replace the December 1st estimates of the dynamic factor with the estimates as of last GDPNow release on November 25th, the December 1 nowcast for residential investment growth becomes 6.5% [instead of 4.5%] and the December 1 nowcast for nonresidential structures investment growth becomes -0.4% [instead of -1.1%]. So the nowcasts for these components depend importantly on the dynamic factor even though that is not a direct input into GDP. Once we get closer to the GDP release, the dynamic factor becomes less important as the model relies more heavily on actual released data.

Thanks Pat.

That may be a bit complicated for most to follow, but it provides a nice explanation as to why the GDPNow  model moves counter to what one might otherwise conclude on release of the data.

By the way, I have found the folks at the Atlanta Fed as well as the BLS very helpful in answering questions like these.

Mike "Mish" Shedlock