Does the decline in the 12-month moving average VMT herald a recession?

mr. Stephen Kopitz writes:

But overall, the 12mm-based VMT is turning a little early or right at the start of a recession.

I take a series of unseasonally adjusted vehicle mileage going back to 1970 and define a moving average variable over 12 months. I’m thinking of defining a dummy variable that takes on the value 1 whenever that moving average term goes negative. This is shown as the blue bar in the chart below, with NBER recession dates shaded in dark grey:

Figure 1: Months when the 12-month moving average VMT is declining (blue bar) and recession dates as determined by the NBER (dark gray bars). Source: DOT FHA via FRED, NBER and author’s calculations.

Using a one-lag probit model with a 12-month negative moving average of the dummy VMT model as a predictor, I predict recessions as follows:

Figure 1: The implied probability of a recession using a dummy variable takes on a value of one when the moving average of the moving average of 12-month vehicle miles traveled is negative, one month behind (pink), and the NBER determines recession peak and trough dates (gray bar). Source: NBER and author’s calculations.

The McFadden R2 from this regression is 0.11. The share of correct recession forecasts is 48% (using a threshold of 30% for recession forecasting). The output of the regression is at the end of the post.

Obviously, there are many false calls precisely because there are numerous cases where the 12-month moving average of VMT declines without an accompanying NBER-defined recession (of course, one can define recessions in such a way that the VMT variable accurately predicts ; this would mean defining recessions as months after the 12-month moving average VMT decline!)

What if we use the actual 12 month moving average change (one month behind)? Then we get this picture.

Figure 3: Implied probability of a recession using a moving average of 12-month vehicle mileage one month behind (sky blue) and NBER-defined peak-to-trough dates (gray bar). Source: NBER and author’s calculations.

The McFadden R2 from this regression is 0.04. The share of correct recession forecasts is 16% (again using the 30% threshold). The output of the regression is at the end of the post.

Again, not a particularly good recession predictor as defined by the NBER.

Application:

The VMT variable is a reasonable indicator of matches. The probit regression has a McFadden R2 of 7% with 18% correct recession calls using the 30% threshold.

Figure 4: The implied probability of a recession using a moving average of 12 months of vehicle mileage simultaneously (green) and NBER-defined recession peak-to-trough dates (gray bar). Source: NBER and author’s calculations.