Making Sense of Recession Probabilities

In late April, the U.S. Bureau of Economic Analysis (BEA) announced the preliminary estimate of gross domestic product for the first quarter of 2025, which showed GDP declined 0.3%. Even prior to this, Google web searches of the term “recession” hit peak popularity.Google Trends provides search interest for a term “relative to the highest point on a chart for a given region and time.” The web search interest values for the term “recession” in the United States for March and April 2025, relative to the entire history of search data at the time of collection (Jan. 1, 2004, to May 7, 2025), were 49 and 57, respectively. This is the second highest sum of search interest for two consecutive months in the data. News articles mentioning recession often cited market analysts’ “probability of recession.”For example, see J.P. Morgan’s May 19 article, “The probability of a recession has now fallen below 50%,” or Reuters’ April 7 article, “Goldman Sachs raises odds of US recession to 45%, second hike in a week.” How do economists and Wall Street analysts generate these so-called recession probabilities, and what do they mean?

Defining a Recession

Before discussing recession probabilities, we will first define a recession. According to the arbiter of U.S. recession dates, the National Bureau of Economic Research’s (NBER) Business Cycle Dating Committee, recessions are pervasive and prolonged contractions in economic activity across many sectors. Hence, weakness in a few industries would generally not be characterized as a recession. In fact, to determine a recession, the NBER uses several economic data series, including nonfarm payroll employment, real personal income less transfers, real personal consumption expenditures, wholesale-retail sales adjusted for price changes, and industrial production.

The NBER’s objective is to determine the chronology of turning points in the business cycle—the beginning (peak of overall economic activity) and end (trough) of recessions. The NBER turning points are not revised, but the data used to label recessions are frequently revised. Thus, the NBER typically has waited eight to 22 months before making its final call about the timing of recessions.For more information on business cycle dating at the NBER, visit its website.

The NBER’s methods contrast with characterizations of recessions often seen in the media that are constructed from a single variable in real time. Two popular definitions of a recession used in the media are two consecutive quarters of negative GDP growth and the Sahm rule, which signals the start of a recession when the three-month moving average of the U.S. unemployment rate rises by 0.5 percentage points or more relative to its lowest point in the previous 12 months. Both characterizations can be thought of as rules of thumb to predict whether the NBER will eventually call a recession for a given forecast horizon (current or future).

Interpreting Recession Probabilities

Like the aforementioned rules of thumb, a recession probability assigns a likelihood that the NBER will (eventually) label the relevant (current or future) period a recession. Unlike the rules of thumb, recession probabilities are generated from sophisticated statistical models that incorporate many economic and financial variables. These models can assess the probability of various kinds of events. For instance, a model can measure the probability that the U.S. is currently in a recession, or it can measure the probability that the U.S. will experience a recession in the next 12 months.

Recession probabilities can be interpreted similarly to meteorological forecasts. The National Weather Service defines the probability of precipitation as the likelihood of more than 0.01 inches of rain in a designated area during a specified time. Likewise, one can think of a recession probability as the likelihood that the NBER will define the current (or future) period as a recession based on the current economic data. Both newly released data and revisions to past data can affect the probability of a recession. Depending on the model, the probability is usually obtained from a regression of the NBER recession dates on economic and financial data.

FRED, the Federal Reserve’s online economic database, houses two recession indicators, both of which estimate the odds that the economy has entered a recession in recent months.The Smoothed U.S. Recession Probabilities series is based on the probability of a recession in a given month, but it’s released on a two-month lag. For example, the most recent value (released May 1) is the probability that a recession began in March 2025. The Sahm rule has a shorter lag in its release; the rule’s most recent value (released May 2) is for April 2025. The Smoothed U.S. Recession Probabilities series from economists Marcelle Chauvet and Jeremy Max Piger is estimated using a statistical model that inputs revised data. The Real-time Sahm Rule Recession Indicator is computed from the aforementioned rule of thumb and uses real-time data on the unemployment rate.Real-time means data as they are first released. This does not include revisions that are made to data at a later time. These recession probabilities do not reflect the views or analysis of the St. Louis Fed or the Federal Reserve System; rather, they represent the types of rules and models that are regularly used to assess the likelihood that the U.S. is (or will be) in recession.

Instead of focusing on the current period, we examine the historical performance of these two indicators. In other words, we want to see whether the probability of recession is high during an NBER-defined recession (shaded in gray) and low otherwise.

Notice that the real-time Sahm indicator (the second figure) rises only after the NBER turning point, a reflection that the unemployment rate is a lagging indicator. Moreover, there are instances in which the Sahm indicator rises very close to the suggested threshold (e.g., November 1967 and September 1976), although these do not count as false positives.The results of the unemployment rates for July, August and September 2024 caused the Sahm indicator to reach 0.53, 0.57 and 0.50 percentage points, respectively. However, the indicator has since fallen below 50, and Claudia Sahm, the indicator’s creator, indicated that the rise may have been a false alarm. The smoothed recession probabilities series (the first figure) uses revised data over the whole sample to construct its estimate, meaning that, like the NBER, it uses data well after the event to infer the timing of a recession.

Conclusion

What goes into a forecast (or nowcast) determines the predicted probability of recession. The variety of models and forecast horizons and the use of different financial and economic data lead to a wide range of predictions. Unlike the series in FRED, we generally do not know how the various models presented in the media have performed in the past. Thus, while recession probabilities can provide a snapshot of current financial and economic conditions, relying on them to form judgments about whether we are experiencing a recession in real-time—especially while lacking complete and revised data—can be perilous.

Notes

1. Google Trends provides search interest for a term “relative to the highest point on a chart for a given region and time.” The web search interest values for the term “recession” in the United States for March and April 2025, relative to the entire history of search data at the time of collection (Jan. 1, 2004, to May 7, 2025), were 49 and 57, respectively. This is the second highest sum of search interest for two consecutive months in the data.
2. For example, see J.P. Morgan’s May 19 article, “The probability of a recession has now fallen below 50%,” or Reuters’ April 7 article, “Goldman Sachs raises odds of US recession to 45%, second hike in a week.”
3. For more information on business cycle dating at the NBER, visit its website.
4. The Smoothed U.S. Recession Probabilities series is based on the probability of a recession in a given month, but it’s released on a two-month lag. For example, the most recent value (released May 1) is the probability that a recession began in March 2025. The Sahm rule has a shorter lag in its release; the rule’s most recent value (released May 2) is for April 2025.
5. Real-time means data as they are first released. This does not include revisions that are made to data at a later time.
6. The results of the unemployment rates for July, August and September 2024 caused the Sahm indicator to reach 0.53, 0.57 and 0.50 percentage points, respectively. However, the indicator has since fallen below 50, and Claudia Sahm, the indicator’s creator, indicated that the rise may have been a false alarm.