The latest economic numbers have boosted our nowcasts for Q3 GDP, which is scheduled for official release from the government on October 26. Meantime, it’s getting easier to anticipate some improvement over Q2’s weak 1.3% growth.

Here’s a review of how today’s nowcasts compare with recent history and two widely cited predictions (via The Wall Street Journal’s survey of economists and the latest estimate from the National Association for Business Economics):

For additional perspective, consider how our nowcasts have evolved in recent weeks as new macro numbers have been published:

Every guesstimate of what’s coming, no matter how refined, is suspect, of course, and a healthy dose of caution is necessary here. The 2.9% nowcast for Q3 GDP via the 10-factor model is almost certainly overstating the possibilities for what we’ll learn from the Bureau of Economic Analysis at the end of the month. If so, some of the excess optimism is probably due to the fact that this particular model draws on a longer span to make its estimates. When I start the series of Q4 nowcasts, I’ll probably refine the 10-factor model so that it focuses exclusively on the quarter-over-quarter trends, if only for more direct comparison with the other models. For now, to maintain continuity with what I’ve been publishing for Q3 nowcasts, I’ll keep all as is through through the end of the month.
Keep in mind that the four-factor model, which sticks with the latest quarter vs. the previous quarter for analyzing the data, also improved recently. That implies that the prospects generally have ticked up for expecting a higher Q3 growth rate vs. Q2’s dismal pace. The 10-factor model may be an anomaly, but the change for the better overall seems to resonate in the numbers.
Ah-ha, you say—the ARIMA and VAR nowcasts are unchanged. True enough, but there was no new information for updating these models (that’s a function of how these models are run). By contrast, the 4- and 10-factor GDP nowcasts are higher because the additional reports since we last crunched the numbers on October 8 have been positive. Indeed, September reports for retail sales, industrial production, and new residential housing construction have dispatched a trio of positive data points. As a result, the 4- and 10-factor models that are sensitive to these indicators have perked up.
In fact, a broad read on the economy, based on the numbers published so far, supports the view that growth is holding up, and perhaps even strengthening a bit. It’s always dangerous to put too much faith in any one model or specific point forecast, but the numbers in the final weeks before the Q3 GDP number hits the streets suggest that the economy rebounded somewhat in the July-to-September period vs. the previous quarter.
Here’s a brief profile of how each of The Capital Spectator’s nowcasts are calculated:
4-Factor Nowcast. This estimate is based on a multiple regression of quarterly GDP in history relative to quarterly changes for four key economic indicators: real personal consumption expenditures, real personal income less government transfers, industrial production, and private non-farm payrolls. This model compares the data on a quarterly basis, looking for relationships with GDP within each quarter from the early 1970s to the present. The four independent variables are updated monthly and so the nowcast is revised as new data is published. In effect, this model is telling us what the data trends in the current quarter imply for the quarter’s GDP growth.
10-Factor Nowcast. This model also uses a multiple regression framework for historical data from the early 1970s and updates the estimates as new numbers arrive, but with two key differences relative to the 4-factor model above. First, this model uses more factors—10 in all. In addition to the data quartet used in the 4-Factor model, the 10-Factor nowcast also incorporates the following series:
• ISM Manufacturing PMI Composite Index
• housing starts
• initial jobless claims
• the stock market (S&P 500)
• crude oil prices (spot price for West Texas Intermediate)
• the Treasury yield curve spread (10-year Note less 3-month T-bill)
The second difference is that the 10-factor model analyzes relationships across a longer span of time by considering the average of changes across the trailing one-, two-, three-, and four-quarter comparisons. The intuition here is that there may be influences on GDP that predate activity in the current quarter, and that those influences come from a broader set of economic trends. If so, the 10-factor model will do a better job of capturing those signals relative to the 4-factor model.
ARIMA Nowcast. The econometric engine for this nowcast is known as an autoregressive integrated moving average. The technique is using only real GDP’s history, dating from the early 1970s onward, for anticipating the current quarter’s change. As the most recent quarterly GDP number is revised, so too is the ARIMA nowcast, which is calculated in R software via Professor Rob Hyndman’s “forecast” package, which optimizes the prediction model based on the data set’s historical record.
VAR Nowcast. The vector autoregression model looks to several data series in search of interdependent relationships for estimating GDP. I use the four variables in the 4-factor model noted above to generate VAR nowcasts of GDP. As new data is published, so too is the VAR nowcast. The basic idea here is to let the data specify the model’s parameters. The data sets are based on historical records from the early 1970s, using the “vars” package for R to crunch the numbers.