Looking at alternative data about the US economy

No point in updating a model during the government shut down

The U.S. government shutdown has stopped the flow of official data. Economic information, it turns out, is not “essential.” Press reports focused on the delay of the employment release (due out on Friday, October 3). For The Nowcast, however, the lack of official data means that there’s not much point in updating the model. This problem will grow if, as seems likely, the shutdown continues. And not only will we lack new data for projecting GDP; GDP itself won’t be published. The advance release of GDP for Q3 is due to be released on October 30, so if the shutdown continues beyond then, forecasting the “current quarter” will become even more difficult.

While we wait for economic data, let’s start to look at some alternatives. What will signposts do business and economic policymakers have available?

The new GDP: Industrial production

The main alternative to GDP is industrial production, (IP) which is estimated by the Federal Reserve, which is not affected by the shut down. While industrial production covers only manufacturing, utilities (mainly energy production) and mining, that’s a significant share of the economy. And for purposes of looking at business cycles, it might be enough.

A simple regression suggests a high correlation between changes in IP and changes in GDP. (The equation starts in 2013 because econometric tests indicated that a structural break occurred about that time. A dummy for the Covid change in 2020Q1 was not significant, although it does improve the fit. GDPR is real GDP; INDPRO is industrial production)

Keep in mind that this equation uses revised, not vintage data. But it does suggest that IP could be a useful proxy for GDP (and for the state of the economy).

However, remember that structural break? the past ten years have seen a disconnect between industrial production (IP) and GDP.

IP has always been more volatile than GDP. But since about 2014, IP has stopped trending upwards, unlike GDP. That suggests using some caution in reading IP as a proxy for the state of the economy.

Worse, IP itself depends on both private and public sector estimates of physical production. Private sector sources will continue publishing, but the quality of the Fed’s IP data will be compromised to the extent that government data is no longer available.

Employment data

The commentary this week focused on the missing employment data, and no wonder. This data is some of the highest quality and most timely data on the state of the economy that we have. And, of course, it’s importance can’t be underestimated, since ensuring stable employment is a key target of policymaking, and even written into the Fed’s mandate.

I’ve found (so far) two candidates for proxying the state of the labor market.

First, ADP, the payroll company, has been publishing estimates of private sector employment based on its internal data. ADP claims to cover about 26 million US workers (out of the total of 160 million workers total, and 135 million private sector workers). That’s a hefty share of the total workforce. ADP’s customers do not reflect all US employers—for one thing, they tend to be larger than average. But, in theory, the ADP numbers should be a useful substitute for the official BLS estimates. That may be the case, although a comparison graph also gives some reason to be skeptical.

In many months, ADP is very close to the BLS number. But then we get certain months (October 2024, August 2025) where ADP simply misses completely. Note that this not vintage data (the data as originally published), so may not fully represent the experience in real time of using ADP to project or understand the BLS establishment data. Differences in seasonal adjustment, in particular, may affect what we see here.

On average, ADP has historically been a pretty good predictor of the BLS private sector jobs number (ADP doesn’t have, or at least doesn’t publish, information on the public sector.) The regression below, which has dummy variables for each month in the early stages of Covid, suggests that ADP is worth considering as an indicator of the state of the labor market.

One caveat: the residuals from this equation are consistently negative in the past two years (notice that the Durbin Watson statistic is pretty low for an equation in differences). In other words, ADP has been consistently overpredicting private sector employment recently. That’s concerning, and something to keep in mind when ADP releases their numbers.

A more esoteric, but interesting measure of the labor market is data from Google trends. Here’s a comparison of the actual unemployment rate with Google searches for “unemployment” in the United States.

Although the broad outlines are correct, a regression confirms that we need to use some care in interpreting the Google trends data.

The low Durbin Watson suggests severe misspecification (i.e., something’s missing in this regression). Not surprising, when you notice the gradual rise in the unemployment rate over the recent past is not matched with any growth in searching for “unemployment.”

That doesn’t make the Google Trends data useless, but it does appear to be less helpful than the ADP data in proxying for BLS employment measures. Of course, the above is just suggestive. Researchers have, in fact, been attempting much more sophisticated methods for connecting Google trends to unemployment data. See here, here, here, and here for examples. I imagine that some researchers are now in high gear to publish a proxy labor market measure. The problems with Google search that I’ve highlighted may well be fixable. Stay tuned!

There no replacement for official data

In the short run—and in the long run as well—the US government has substantial resources and a unique ability to reach into data sources. The private sector will never be able (or, at least, willing to spend the money) to replace those data sources. And while current US government policymakers may be satisfied with losing the data in the short run to achieve their political goals, that won’t hold so much as time goes on—and particularly if the economy goes south. But for now, those of us interested in understanding the economy will have to manage with whatever is available, and hope for a speedy end to the government shut down.

Oh, and I’m sure you’ll see many additional private sector sources cited if the shut down continues for a long time. I’ll try to cover more of these (and data sources for other concepts, like inflation) in the weeks ahead. But one rule: without a reasonably long time series to validate the new measure against the government’s “official” measure, the new data essentially amounts to a guess. Unfortunate, but true. Save your excitement for measures that can be validated over a long time.