Wondering if vehicle miles traveled (VMT) is a good predictor of recessions. Stop being weird. First look at the role of VMT during the recession, compared to heavy truck sales (advised at some point by calculated risk) and the eponymous Sahm rule (real-time version).
figure 1: 12-month growth rate for vehicle miles traveled, nsa (teal), heavy truck sales, sa (tan) and Sahm rule metrics – real time (black). Sahm’s rule is the 3-month moving average of the unemployment rate relative to the lowest unemployment rate over the past 12 months. The dashed orange line represents the threshold for the Sahm rule metric. Dates of peak-to-trough recessions as defined by NBER are shaded in gray. Source: FHA via FRED, Census via FRED, FRED, and NBER.
It’s hard to see, but the 12-month change in VMT was down a few months ago (at levels below pre-pandemic levels), while heavy truck sales were up year-over-year. Sahm’s rule is close to zero (0.5 percentage points are needed to break the threshold).
How do these variables predict recessions (peaks and valleys, as defined by NBER BCDC, not the amorphous definition some use)? Here are the probabilistic regression results for the period 1970M01-2022M10/M11 (I transformed the Sahm rule variable into a dummy variable (Sahmruleindex) that takes a value of 1 if the Sahm rule variable exceeds the 0.5 threshold).
Note that the year-over-year growth rate of VMT does not explain much of the recession according to McFadden R2. In fact, the year-on-year growth rate of heavy truck sales can best explain the problem. The Sahm’s rule index has a median explanatory power of 18%. Sam’s Rule should indicate that a recession has begun, not necessarily that a recession is currently underway.
Below I plot the predicted probabilities associated with estimated VMT and heavy truck sales probabilities, and the implications of Sahm’s rule (not estimated).
figure 2: Probit estimates recession probabilities based on the 12-month growth rates of VMT (teal), heavy truck sales (tan), and the implications of Sahm’s rule interpretation (black). Dates of peak-to-trough recessions as defined by NBER are shaded in gray. Source: NBER and authors’ calculations.
Since it is difficult to see what is going on, especially in the recent recession, I focus on the period beginning before the Great Recession (corresponding to Figure 1).
image 3: Probit estimates of recession probability based on 12-month growth rates for VMT (teal), heavy truck sales (tan), and implications (details) explained by Sahm’s rule (black). Dates of peak-to-trough recessions as defined by NBER are shaded in gray. Source: NBER and authors’ calculations.
Now, given the extreme nature of VMT behavior during the pandemic, VMT would do better if I truncated the sample at 2019M12. However, the McFadden R2 only rises to 12%, still not breaking the 50% threshold for the 2007-09 recession (but would have predicted the 1996 recession).
So I’m still skeptical Mr. Kopits’ assertion (in rebuttal Macroduck’s Reviews), the VMT is a magnet for recession statements, namely:
…so you’re saying that 1.1 million jobs were added while both gas consumption and VMT were down? So not only are those extra 1.1 million workers not driving to work, but those with jobs are also driving less. It is possible, of course. But if I had to decide between the CES and the HH survey, the data is more consistent with the HH survey. But that’s not what Mengzi did. Neither do you. But I did, and as it turns out, the inference seems to be correct.
My view is that VMT is particularly unreliable from a formal point of view, and given recent developments, including working from home, it is reasonable to assume that there is a structural rupture in the employer-VMT relationship.
To account for potential disruptions in this employment-VMT relationship, I converted the data monthly to quarterly and regressed on the first log difference. Stability around the pandemic was rejected by the recursive residual one-step Chow test.
Figure 4: Residuals from previous step of recursion from indicated regression (blue, right scale), 95% interval (red dashed line, right scale), probability of no change (blue circle, left scale). Dates of peak-to-trough recessions as defined by NBER are shaded in gray. Source: Author’s calculations using EViews.
Total:
- Given the structural breaks identified in the data, VMT should not be relied upon to infer employment.
- There are few signs that the economy is in recession (as defined by the NBER Business Cycle Dates Committee) through November 2022 or the first half of 2022.
Of course, random observers can define the recession any way they want to make their claims valid.That is, one can A recession in which the Michigan index falls below a certain value is called a recession. But this is not in the spirit of the business cycle literature.
