Some highly touted soft measures never match economic data except by accident. Consumers sentiment is the prime example.
Yet, once again, most of the investment community believes in the strong “second-half recovery” theory instead of looking for reasons why soft data is not going to harden.
Record Gap Between Hard and Soft Data
Davies says “The inclusion of survey or soft data in nowcasting models is crucial to improve the accuracy of the signals produced by these models in real time.”
He concludes that hard data understates growth and expects a much stronger second quarter.
John Hussman has a different conclusion. Hussman proposes that this late in the cycle, soft data is typically noise.
Soft survey-based measures tend to be most informative when they uniformly surge coming out of recessions. In contrast, during late-stage economic expansions, positive disparities in soft measures tend to be false signals that are resolved in favor of harder measures.
The current positive divergence is particularly likely to be misleading. The charts of survey-based measures below demonstrate why. What’s striking about survey-based economic measures is that their 5-year rolling correlation with actual subsequent economic outcomes has plunged to zero in recent years (and periodically less than zero), meaning that these measures have been nearly useless or even contrary indicators of subsequent economic outcomes. The first chart shows the correlation between these survey measures and subsequent 3-month growth in employment and industrial production. Note that a correlation of 1.0 is perfect correlation, a correlation of -1.0 is a perfect inverse correlation, and a correlation of zero implies no relationship at all.
In the chart above, the green bars show U.S. economic recessions. The blue employment line shows the steepest collapse in the reliability of survey-based data in post-war history. While it’s not always the case, one can also see a tendency for these measures to be least reliable (having the lowest correlation with subsequent economic outcomes) just before U.S. recessions.
What’s also interesting is that in recent years, these survey-based measures have also become less correlated with prior changes in economic activity. So at least in recent years, these measures have lost their correlation with both past and subsequent economic activity. These surveys may be measuring something, but whatever it is, it’s weakly related to actual economic fluctuations.
Of course, these optimistic economic surveys are helping to feed optimism among investors of a coming economic renaissance, which we view as wholly inconsistent with the arithmetic of likely GDP growth in the coming years (see in particular Stalling Engines: The Outlook for Economic Growth). That inconsistency is likely to play out over time, but in the short-run, bullish investor sentiment is driving economic sentiment, which is helping to drive bullish investor sentiment. Discussing this feedback loop with Bill Hester, he remarked, “We’re living in a huge echo chamber.”
Crackings the Shells
I can see how soft data might smooth out forecasts. But what is it really measuring now?
The sentiment measures are a joke. Close scrutiny proves sentiment measures stock market performance, not expected measures of consumer spending.
Although everyone touts the ISM report, what about Markit’s PMI report? The Markit and ISM reports are two soft measures of the same thing. While ISM estimates 4.3% first quarter GDP, Markit suggests 1.7%.
Thus, not even the soft data is uniformly strong.
Auto Sales and Plunge in GDPNow
In regards to auto sales and the plunge in the GDPNow estimate I received this reply from Pat Higgins, the creator of GDPNow:
Thanks for forwarding me your blog link and the positive feedback. Regarding your question, almost all of the decline on April 5 was due to auto sales rather than the ISM Nonmanufacturing Index. If one uses the April 4th estimate of the model’s dynamic factor with the April 5th estimate of auto sales, the model forecasts for real business equipment investment and real PCE goods good are only about 0.1 percentage point higher than their April 5th levels provided in the GDPNow spreadsheet. Real PCE goods growth fell from 2.5 percent to 1.2 percent on April 5th and real equipment investment growth fell from 9.7 percent to 6.4 percent on the same day. These subcomponent declines were responsible for most of the model nowcast decline on the 5th.
Finally, it is a well-understood phenomenon that economic tops are a rounded affair while bottoms happen quickly (with governments and the Fed typically both stimulating).
It is during these lengthy topping processes where soft measures are most likely to be wrong.
I side with Hussman.
Mike “Mish” Shedlock