The Endless Metrics Recession Indicator

In the years since the financial crisis, market observers have made numerous calls that the economy is entering or has already entered a recession. After a decade, all of those claims have turned out to be false.

To be fair, the post-2008 recovery has been historically long, choppy, and weak compared to previous expansionary periods in U.S. history. There have also been several really close calls, where real GDP was negative for a quarter or was almost zero over two quarters. That would have made for a really weak recession, by the definition of two consecutive quarters of negative growth, but a recession nonetheless.

Recessions are binary, either there is one or there isn't. However, as the previous commentary suggests, some nuance to the story can be helpful for analysis. So, the chart below was developed to provide just that.

The Endless Metrics Recession Indicator

How it's built

The Endless Metrics Recession Indicator ("EMRI") was built by modeling various economic, market, and interest-rate indicators against recessionary periods as denoted by the National Bureau of Economic Research ("NBER").

For any given recession, there are leading, lagging, and coincident indicators. For example, the stock market usually falls before a recession because corporate earnings and those closest to the lines of business start seeing issues. Unemployment is usually a lagging indicator as businesses don't lay people off right away as it's not always immediately obvious when a recession has begun. Unemployment levels also take much longer to decrease than they do to increase (easier to fire en masse than to spend time to find and hire good people). Finally, GDP is often considered a coincident indicator, even though it is released and revised after a given quarter has ended.

What it means

By taking many factors like the ones just mentioned and combining them through a model, an analyst can quantify a wide-range of data in an organized way. In this case, the result is a historical time series metric that can be analyzed to assess whether there is a recession or not. For the metric, a higher value indicates conditions more like a recession.

Since recessions are binary, a cut-off maker was optimized to provide a yes or no to whether or not a given quarter is a recession. If the metric is above the time series, it is indicating that yes, there is a recession. Otherwise, nope, it's all good.

When does it work?

The chart has shaded regions that indicate recessions, so it passes the eye test. Numerical analysis backs this up as well: with 256 quarters (64 years) of observations, the model matched 252 with the NBER for a 98.4% historical accuracy. It hit every single quarter of recession and had four quarters of false positives. These quarters were clustered in with actual recessions, as opposed to randomly throughout an expansion.

So, that is almost as much accuracy as one could hope for and any future model enhancements will have a high bar to improve upon.

Why it's useful

NBER-based recessionary periods are usually denoted significantly after a recession has passed. A bunch of really smart people have to come together, debate, look at data, look at data revisions, and collectively agree to binary recession-or-not dates. That takes time! In early 2008, some people may have seen recession signs but it wasn't obvious to everyone. It would've been helpful to sell stocks at that time if cash was needed as opposed to early 2009.

Models and methodology also help provide rigor to the endless stream of data out there. It is hard to digest and come to conclusions. A model such as this effectively provides an opinion. Someone might say, "Hey, we are in a recession it' s time to panic!" That statement can be challenged against the model that has an almost perfect historical accuracy. There will always be someone crying recession. If a recession is being discussed and the model confirms that, it provides stronger evidence for decision making.

Models can also evolve and adapt to external changes and internal weaknesses. For example, as new data sources become available, they can be integrated if useful for prediction. For this model, it is mostly coincidental, it doesn't give as accurate a prediction about the future. However, the model can be improved to do just that. It's a stepping stone.

So, when is the next one?

As mentioned, the model will give an opinion about recessions before the NBER but mostly coincidentally to current economic conditions. Right now, the model does not see a recession.

However, in coming quarters, some leading indicators suggest trends that will provide conditions more conducive to a recession. This makes sense logically; the expansion since the financial crisis has been long! While an expansion may not die of old age, typical business cycle trends make it harder to stay alive.

Future posts will check in on the performance of the EMRI, including an opinion for 2019Q1 in a few weeks. That being said, things look good. At least, they should and the model should agree - it certainly doesn't feel like a recession yet, does it?

For any questions, comments, or inquiries related to this article or any other on this site please reach out to: