Putting numbers on vigor (#2: Projecting debt service)

Just at the end of August, at the end of my month-long investigation of debt service, FRED updated the data series. We looked at it. They added the data for 2016 Q1, and not much else changed.

I finished the "bowl" again, using the update.

Graph #1: Household Debt Service as a Percent of DPI, with projection thru 2025

Excel's second order polynomial makes a nice bowl shape. The other trendline options give a straight line or a sagging line or a line that goes off in a direction that doesn't seem right. The second order polynomial feels right. But it's a weird combination of meticulous calculation and gut feel that determines how the prediction turns out.


The trendline equation doesn't know about changes that might arise in the economy that could send the projection off in a different direction. That smooth red curve on Graph #1 will continue as far into the future as I am willing to show it. That's obviously not realistic. The trend will break eventually.

This is like the St. Louis Fed's New Characterization of the economic outlook, by James Bullard and others. They write

The concept of a single, long-run steady state to which the economy is converging is abandoned, and is replaced by a set of possible regimes that the economy may visit.

They are saying they don't know about the future, and they are now only predicting the next 2.5 years. Because they expect that trends could change.

So don't read too much into that long red line. At some point, the trend will break for sure. Maybe we'll go looking for that break point in a later part of this series.


Looking at Graph #1, the bowl seems a little lopsided. It's a little wider on the right, the red side. Between 2008 and 2012 the fall was fast. The increase after 2016 seems slower. Just by eye. Maybe that makes sense: People were faster to get rid of debt than they are to take on new debt. That could influence the debt service graph.

But the situation was different in the 1990s. That bowl looks to me to be a little wider on the left. People were reducing debt (and debt service) at a leisurely rate on the downslope of the bowl. But on the upslope, the change was less leisurely. People were excited about the economy picking up, maybe.

I think that's the case now, too. People have been hesitating, and that made for a pretty good lag. It seems like the whole of President Obama's second term was lag. But people are itching now, itching for better times. And I think people are going to start spending more, which will create the better times we're itching for. And if that happens, debt service will go up faster than Graph #1 shows.


This slower increase after 2016 on the graph, this is new. I didn't really notice it before, not in Graph #7 of the 21st, and not in Graph #3 of the 27th.

Now that I look, maybe it is there. For a better look, I put all three projections together on one graph:

Graph #2: Three Estimates of Future Debt Service

Nah, the first two run close. This new projection is a lot slower to increase.

Now I'm wondering why.

The new one, green on Graph #2, uses data from after the update. We looked at that, and there was not much changed. They just added the one data point for 2016 Q1.

That new data point was lower than the one before it. Maybe that threw the trend off? I don't know. Here is the graph that shows both versions of the data series:

Graph #3: Household Debt Service before (blue) and after (red) the update

As I said before, the new data value is down slightly from the previous number, and not much else is different.

Think that one extra data point, being low like that, could have thrown off the trend line so much?

Graph #4: Four Estimates of Future Debt Service

Yup. Graph #4 is a copy of Graph #2 but with the black line added. The black line uses the same recently-revised data as the green line, except I eliminated the data item they added for 2016 Q1.

So the difference between the green and black lines is entirely due to the 2016 Q1 item being "down slightly" from the previous number. Wow.


// the Excel file