Podcast
Important Information:
A New Way of Detecting Recessions
Important Information:
John O'Trakoun and Adam Scavette discuss their new approach to gauging the health of the economy and the likelihood that a recession is taking place before other indicators would detect it. O'Trakoun is a senior policy economist at the Federal Reserve Bank of Richmond and Scavette is a community development economic advisor at the Federal Reserve Bank of Philadelphia.
Related Links
Transcript
Tim Sablik: My guests today are John O'Trakoun and Adam Scavette. John is a senior policy economist at the Richmond Fed and Adam is an economic advisor at the Philadelphia Fed. John and Adam, welcome back to the show.
John O'Trakoun: Thanks for having us.
Adam Scavette: Glad to be back.
Sablik: Our discussion for today is all about predicting and identifying recessions. You both recently created a new tool for forecasting recessions, the Scavette-O'Trakoun-Sahm-style Indicator, or SOS Indicator, which congrats on that acronym title.
We'll be getting into the details of your measure in a moment. But I wanted to start our conversation by talking about a tension that's at the heart of this type of forecasting, a tension between accuracy and timeliness. For a forecast to be useful, it ideally needs to arrive before the event it is predicting happens. But typically, the earlier warning you try to provide, the more difficult it is to be accurate because you might not have all the data you need. How did you think about this trade-off when you started your project?
O'Trakoun: That's a great question, and it gets to one of the key trade-offs in the business of forecasting. For our project, which tries to develop an indicator of recessions, we tried to make contributions on both dimensions at the same time.
To address timeliness, we used weekly data on unemployment insurance claims. That gives us one of the highest frequency measures of labor market conditions that we have available. It can also give us earlier recession signals compared to some other indicators which are based on monthly data, like the unemployment rate.
On the accuracy piece, we considered some of the ways that standard economic data could be less accurate. For example, some of the most important data on the state of the economy, like monthly nonfarm payrolls and retail sales, can be revised in later months as additional data are collected. This could also affect the headline unemployment rate, which might change as a-result of updating seasonal factors or benchmarking to administrative data on unemployment insurance. Another accuracy issue might emerge because some important data is obtained from surveys. Respondents could potentially misinterpret a question or choose not to respond at all. Because the data underlying our indicator are already based on administrative data rather than surveys, it can partially mitigate some of these accuracy concerns.
Sablik: We'll be digging into more of those as we go on in this conversation.
As the name of your SOS Indicator indicates, it's based on another popular recession forecasting tool, which is the Sahm rule. Adam, what is the Sahm rule?
Scavette: The Sahm rule is a recession indicator that relies on the monthly U.S. unemployment rate to determine whether the nation has entered a recession. It was developed by economist Claudia Sahm when she was working at the Federal Reserve Board of Governors back in 2019. The Sahm rule states that when the three-month moving average of the national unemployment rate is a half a percentage point or more above its low over the prior 12 months, we are in the early months of a recession.
Sablik: How well has the Sahm rule performed when it comes to predicting recessions?
O'Trakoun: Well, the Sahm rule has generally been a pretty reliable recession indicator throughout history, but that doesn't mean it's infallible. For example, a Dallas Fed economist found in 2024 that there's been at least four instances since 1950 when the rule was triggered, but a recession didn't happen.
Last year, we saw the Sahm rule triggered in July, at least based on incoming real-time unemployment data. But later on, a recession didn't happen in 2024 as far as we know. Subsequent data revisions reversed that triggering, but it was a reminder that even though we call it a rule, it's really more like a statistical regularity and not a hard and fast law of nature.
Sablik: I think there's a common joke about economists predicting nine out of the last five recessions, or something to that effect. [Laughs]
How is your indicator different from the Sahm rule or other recession forecasting tools? Maybe another way of asking that [is] what are the shortcomings of other tools that you're trying to address with yours?
Scavette: The main shortcoming of the Sahm rule and other survey-based recession indicators that we really try to address is the presence of survey bias. The unemployment rate relies on a survey of about 60,000 households where people are asked to report their employment status. There's noisiness associated with the sampling variability in the survey, which has become more problematic over time through a few different channels such as non-response and misclassification.
For example, the Bureau of Labor Statistics noted that fewer people have been responding to the survey over time, which results in higher rates of non-response bias. Specifically, the inclusion of citizenship questions in the survey has resulted in higher non-response rates in states with larger shares of noncitizen and Hispanic populations. Additionally, during the COVID-19 pandemic, the Bureau of Labor Statistics noted that the unemployment rate was affected by misclassification due to laid-off workers reporting they were employed but absent from their jobs.
Unlike the Sahm rule, which uses the unemployment rate, our SOS Indicator uses unemployment insurance claims, which are based on administrative filings data aggregated across state agencies. Therefore, instead of relying on a survey that asks workers about their employment status, we use the underlying data based on when workers actually file for the unemployment insurance. We think this is a much more reliable data source.
Additionally, unemployment insurance claims are published by the U.S. Department of Labor at a weekly frequency, which John mentioned earlier. So, the SOS is available to the public at a much higher frequency than recession indicators that rely on any monthly rates.
O'Trakoun: I'll just chime in here that there are other rules out there that try to predict recessions based on data that doesn't come from surveys, like financial market data on interest rates. Similar to our indicator, these are available at high frequency. Actually, because you can look at things like Treasury yields every second of the day, they're even higher frequency than our indicator.
Many of these rules are based on the statistical regularity that we tend to see an inverted yield curve prior to recessions. That means that short-run interest rates like the two-year Treasury are higher than long-run interest rates like the 10-year Treasury yield. In typical times, that might be a sign of recession because it might mean the Fed will cut rates in the future due to weak economic conditions. But in the COVID recovery, we had an inverted yield curve because we had high inflation, which was expected to decline. That would have allowed the Fed to cut interest rates as inflation came down. That caused an inverted yield curve but not necessarily a recession.
And, in comparison to these financial market indicators, which are kind of one step removed from real economic activity, our indicator is based on real labor market activity.
Sablik: As you mentioned earlier, John, these kinds of indicators are not infallible. How does the SOS Indicator perform when it comes to predicting past recessions?
O'Trakoun: The data we use — the insured unemployment rate — only goes back to 1971, which means our indicator has a little over 50 years of history. Over that time period, it's never been wrong, but that doesn't mean it's going to be that way forever. Inevitably, sooner or later, this indicator is going to be wrong and it's going to flag a recession that just doesn't happen. That's because, in the business of forecasting, there is no such thing as perfect. Being humble about our ability to understand the state of the economy is the forecaster's most important tool, above and beyond having any particular rule or fancy econometric model.
Sablik: The SOS Indicator is now available on the Richmond Fed's website and, as we've been talking about, it'll be updated weekly. What is it showing now? And I should mention that we're recording this on April 1 — happy April Fool's Day.
Scavette: As of March 27, the SOS Recession Indicator is not signaling a recession. There haven't been any major swings in national unemployment insurance claims filings in recent months, so the indicator has remained pretty stable at a low reading of 0.004 since late last year.
O'Trakoun: This is a good time for me to emphasize to our listeners that throughout this podcast, we've been using words like forecasting and prediction pretty casually. But our indicator is technically what's called a coincident indicator. That means it's not about where we're headed tomorrow. It's about whether we're in a recession today.
The only real sense in which the SOS indicator is predictive is that it can take some time for the NBER business cycle dating committee — this is the entity which officially declares whether the U.S. economy is in a recession — to reach a consensus about the start and end dates of a recession and announce it to the public.
Sablik: That's a really great point to keep in mind.
John and Adam, do you have any plans for expanding this work or adjusting the indicator over time?
O'Trakoun: For now, we're using the indicator as currently published to monitor the state of the labor market. But any good forecaster has to keep in mind that sometimes the behavior of data can shift based on changes in structural trends, like demographics or technology, or other factors. These might lead us to adjust and revise the way we calculate the series if they cause the insured unemployment rate to behave differently relative to the U.S. business cycle.
And, our indicator is only available if the underlying data are available, which is not always guaranteed. For example last year, the Bureau of Economic Analysis discontinued some data on employment by full-time and part-time status across industries, across states, counties and metros. Data availability conditions are changing all the time and we change our process accordingly so that we can deliver on our core mandate of maximum employment and price stability the best we can.
Sablik: John and Adam, thank you so much for joining me today.
