On May 1, the GDPNow Model came up with an initial forecast of 4.3% for second quarter GDP.
My first reaction was “here we go again”. That was quickly followed by “I’ll take the under, way under” bet.
Let’s take a look at the latest forecast, why I think the model is wrong, and some interesting charts of GDPNow forecasts vs what actually happened.
I emailed Pat Higgins, the creator of GDPNow about the initial forecast. He could not say too much because of an FOMC blackout, but here is his email.
The March growth rates for each of private permanent-site residential, private nonresidential, and state+local construction spending put in place were each lower than the model was predicting on April 27th in the final model nowcast for the first-quarter. But all three of these growth rates for February were revised up from the previous construction spending report. I can’t do a deeper analysis on the implications for the GDPNow forecast for the second quarter because of the blackout on FOMC communications until this Friday. Hope this helps a little though.
Weather Related GDP Again
Prior to seeing the initial GDPNow forecast I had emailed Pat my comment on construction spending: “The second quarter is not off to a good start. However, the GDP models may go haywire once again by not factoring in weather-related phenomenon.”
GDPnow Initial and Final Forecasts vs Actual
I created the above chart from the “tracking archives” spreadsheet from GDPNow downloadable data. I added a few columns (Initial, Final Actual) to the sheet.
My friends at Advisor Perspectives helped with data labels.
In the above chart, Actual means the Advance GDP estimate, not the final revised GDP number.
GDPNow Track Record
In Investigating Weather-Related Effects on Construction Spending I posted this chart that explains the jump in the initial GDPNow forecast.
This was a strong report and even stronger than it looks at the first glance. However, I caution everyone to not read too much into this strength.
December was unusually cold. January was unusually warm as was February. Seasonal adjustments do not factor in weather.
Second Warmest February in 123 Years!
Please consider Assessing the U.S. Climate in February 2017.
During February, the average contiguous U.S. temperature was 41.2°F, 7.3°F above the 20th century average. This ranked as the second warmest February in the 123-year period of record. Nearly one-quarter of the U.S. was record warm in February. Only February 1954 was warmer for the nation at 41.4°F. Between December 2016 and February 2017, the average temperature across the contiguous U.S. was 35.9°F, 3.7°F above average, the sixth warmest winter on record.
Estimate Way High
My comment on the construction data was “Before anyone gets too giddy over these numbers, bear in mind very unusual seasonality factors. Builders build when the weather permits, and the weather was very cooperative this Winter.”
Other evidence suggests the same thing.
- Auto Sales Puke Again: Year-Over-Year Totals: GM -6%, Ford -7.2%, Toyota -4.4%, Fiat-Chrysler -7.0%
- Consumer Spending Flat, PCE Inflation Weakest Showing In 16 Years, Rate Hike Odds Rising.
- Regional Lender Loan Crash: Nearly Every Major Regional Bank Missed Lending Estimate.
- Subprime Credit Card Losses Bite Capital One: Income Down 20%, Charge-Offs Up 30%.
Auto sales make up approximately 20% of consumer spending. It was the massive slide in auto sales in the first quarter that caused me to dramatically lower my GDP estimate long before anyone else did.
The April auto sales report (link #1 above) came out today. It is the first hard data point for the quarter.
Consumer spending for March (link #2 above) came out yesterday.
I do not have an estimate for the second quarter yet, as one data point is insufficient. But once again I think weather-related events, this time in housing, got the quarter off to another amazing pie-in-the-sky forecast.
Think Outside the Model
I have learned much from the Atlanta Fed model. In addition, Pat Higgins has been very generous with his time, answering questions.
Instead of blaming the model, learn the model’s weaknesses and temper the model forecasts with a bit of human analysis.
Think outside the model. You can only do that if you understand situations when the model is highly likely to be wrong, like now!