Given the surprisingly strong retail sales report on July 16, I expected at least some positive action in the Nowcast forecast. Nope. GDPNow, did react.
GDPNow Reaction History
Nowcast Reaction History
The largest move on any economic report since July 6 is -0.07 percentage points.
Implications
- The Nowcast model accurately predicted 24 of 24 economic report since July 6.
- The GDPNow model failed to predict numerous economic reports over the same timeframe.
How likely is the first item?
I looked back further. The last time Nowcast moved a “full” 0.1 percentage point on any forecast was a -0.12 percentage points on the June 19 housing permits report.
The running score for the Nowcast model to predict the economic report’s impact on that model is a whopping 35 consecutive economic data points that it tracks. Wow!
Let me ask again: How likely is that?
I have commented on the non-volatility of Nowcast on numerous occasions but this has now reached the point of absurdity.
Margins of Error
The margins of error of both GDPNow and Nowcast is one full percentage point.
These are not the final forecasts, but if they were, the next GDP number would need to be in the range of 3.5% to 3.7% for both models be within their stated confidence levels.
Related Articles
- Retail Sales Surge Continues With Huge Buying Spree
- GDPNow Forecast Surges to 4.5% on Retail Sales, Monthly Treasury Statement
- Housing Starts Unexpectedly Plunge 12.3% in June, Permits Down 2.2%
Mike “Mish” Shedlock
The other thing that seems apparent is that the GDPNow model gets adjusted much more often, which probably accounts for some of it’s volatility. The end result of all the adjustments, though, should be an even better model.
The final GDPNow has been more accurate. Nowcast looks at more things than GDPNow, some of them admittedly useless.
There is another, more likely, possible implication from the failure to move with in response to the various data points, other than that the Nowcast correctly predicted the data. That is that the Nowcast model doesn’t include some of those data points at all, or, at the very least, it’s not very sensitive to those data points.
In the final analysis, the proof is in the pudding. Which model has been more accurate in it’s final prediction? Which has been more accurate with it’s prediction mid-quarter? From memory, the Nowcast has been more accurate, with the GPDNow forecast moving wildly through the month, with predictions at times that are way off the mark.