Weird Data Says This is a Recession

Mark Sheppard is a current Economics PhD student at CUNY Graduate Center. He is an alumnus of both the Harris School of Public Policy (MPP ’21) and Georgetown University (MA ’19). He is also a former Executive Board member of the Chicago Policy Review.

As the Federal Reserve raises interest rates to mitigate inflation, with some economists warning that the policy could inform downward pressure, potentially even a recession, it is informative to unpack what constitutes a recession. The formal definition of a recession has undergone some revisions lately. The National Bureau of Economic Research (NBER) which takes responsibility for this critical business cycle determination, recently declared that the United States is not in a recession, largely by changing what constitutes a recession. NBER explicitly broadened the definition of a recession to allow for a more holistic review. After which the White House, and even the Wikipedia page was updated; however, there is some unconventional data that is less optimistic.

There are multiple indicators for recessions, each one uses a different underlying dataset and as a result each one tells a different story. NBER then makes a holistic decision based on the available research. Traditionally GDP, gross domestic product, serves as the go-to metric, or more specifically the growth rate of GDP. The Federal Reserve publishes GDP data quarterly, and two quarters of negative GDP growth spelled triggered a recession declaration.

Google Trends data, which aggregates search results of hundreds of millions of users, has shown a record spike in the terms “recession” and “inflation.” Since the advent of internet search engines, these datasets correlate surprisingly high with actual recessions and price changes. Despite representing very large samples, macroeconomists seldom use this data when forecasting the economic outlook of consumers. Importantly, Google Trends data often acts as a leading indicator, even when compared against other real-time indicators, presumably because the data captures a fairly clean survey of consumer sentiment.

This definitional change from the NBER represents far more than a semantic difference. In 1983 the Bureau of Labor Statistics updated the definition of inflation, removing housing data from the calculations; which in hindsight may have blinded experts in the lead up to forecasting the 2008 financial crisis. So, the implications of a definitional change can have very real downstream empirical consequences.

Image created by Mark Sheppard

This was problematic because that created a policy lag, as lawmakers would have to wait six months before there was an official recession declaration. The Sahm Rule, which is a real time recession indicator based on unemployment data, provides valuable insight, not because the measure exactly agrees with the Google Trends data, but rather for the technical innovation that the Sahm Rule represents. Economist Claudia Sahm created the Sahm Rule, which looks at a ratio of year-over-year unemployment, to determine a recession, in large part to correct for the policy lag built-in to previous indicators. By tweaking the underlying metric, Claudia Sahm was able to present a real time indicator. Even though the Sahm Rule is one of the quickest indicators used, the Google Trends data still leads.

Image created by Mark Sheppard

Traditionally, macroeconomic policymaking, as a sub-discipline, heavily relies on established datasets from a limited set of institutions, like unemployment figures from the Bureau of Labor Statistics. Inversely microeconomists tend to show greater flexibility adopting new or experimental data to tease out causal or corollary relationships. The tradeoff is that macroeconomic policy is more careful in overreacting but potentially more limited in the ability to assess. While there is value in the constancy of legacy datasets, there is also deep merit to novel metrics.

Novel, or as I like to think of them: weird metrics can often be more flexible instruments than their traditional counterparts. There are countless examples of strange datasets being informative during crisis incidents in economics adjacent fields, when other measures lagged too far behind the need for intervention. At the height of the pandemic, negative reviews for Yankee Candle were used as an early signal for COVID-19 rates because loss of a smell was an early symptom. When Venezuela’s currency collapsed, due to hyperinflation, Bloomberg developed the Café con Leche index, which tracked the cost of a cup of coffee as a stand-in for the change in consumer prices. This type of data proves invaluable in a crisis, as traditional metrics lag behind. During hurricanes due to the destruction inherent to the storm, many weather measurement systems can be destroyed or damaged. To compensate for this the Federal Emergency Management Agency (FEMA) developed the Waffle House Index, which is an informal metric named after the Waffle House restaurant chain to determine the effect of a storm and the likely scale of assistance required for disaster recovery. These datasets might seem like toy case studies, but they work, and specifically they work when more staid metrics fail to meet the moment.

Image created by Mark Sheppard

Obviously, macroeconomists should not make Google Trends data the sole source of empirical business cycle analysis; while obviously some survey data is utilized in the field, economists still underappreciated qualitative data, but in this case economists should not blanketly dismiss what amount to surveys with hundreds of millions of observations either. If a record number of Americans are searching for answers regarding a potential recession that clearly provides a critical signal about the state of the economy, even if a full recession is never declared. NBER presumably punted on determining whether this is a recession because there are some encouraging economic signs, though the traditional metrics point towards a downturn. The question remains of whether in this definitional update if economists are still missing some unconventional signs, or if they’re simply looking at the wrong thing, and more importantly what happens if we’re wrong.