During the last expansion, the US economy reduced its unemployment rate from a peak of 10% in July 2019 to 3.5% in April 2020. Through the last years of that expansion there was an ongoing debate about how much slack was left in the US labour market (Williamson 2015). By 2015, some saw the US labour market as being close to full employment even if the employment-to-population ratio remained clearly below previous peaks. Structural factors such as ageing were mentioned to support this conclusion (Tracy et al. 2015). In August 2015, the CBO estimated the natural rate of unemployment to be around 5.06%, a rate that was very close to the actual unemployment rate that month (CBO 2015). Despite this pessimistic prediction, in the following years unemployment continued its decline and the employment to population ratio never stopped increasing. As a result, by the end of 2019 estimates of the capacity of the US labour market to absorb more workers had been revised upwards substantially. And then a global pandemic happened.
Imagine the following counterfactual scenario. What if the global pandemic associated with COVID-19 had happened earlier – in the summer of 2015? A recession would have started with the labour market immediately deteriorating and we would have never had an opportunity to witness years of a much tighter version of the labour market. Would our views on the potential of the labour market be the same as today if the expansion had ended in 2015? Very likely.
In a recent paper (Fatas 2021), I argue that many expansions in the US economy fit this pattern: they end too early. Because of this, estimates of the natural rate of unemployment, or any concept related to full employment, might be too pessimistic because we have limited opportunity to see a fully performing labour market. As a result, decades when recessions happened often, such as in the 1970s, display higher than normal unemployment rates, leading to higher estimates of the natural rate of unemployment.
The main fact that supports this conclusion is simple: the US economy does not display significant periods of low and stable unemployment rate what we might call a high-pressure economy (Figure 1). During expansions, the labour market is always improving until we hit the next recession. We do not observe long periods where the improvement stops, and unemployment hovers around a number that resembles a natural, long-run equilibrium state. If the 2009-2020 expansion that lasted for more than ten years did not produce such a pattern, how long of an expansion would be needed to create one?
Figure 1 US unemployment rate
Another way to restate the same fact is that low unemployment is a strong predictor of recessions. While this is expected, because of the mean reverting property of unemployment, the extent to which this happens in the US is unusual because of the absence of periods of stable low unemployment.1 Interestingly, we do not observe the same pattern in all countries. In some economies, sustained periods of stable and low unemployment means that low unemployment is a much weaker predictor of recessions.
Why are expansions cut short?
One obvious hypothesis is that external shocks happen too often. The oil price shock in 1979 can be seen as cutting short the expansion that had started in April 1975. The 2020 global pandemic might be the reason why we have missed a period of low and stable unemployment after 2019.
The alternative explanation is that there is something inherently unstable about periods of tight labour markets. Imbalances build and they become the seed for the next recession. There is plenty of evidence in the academic literature that fits this narrative: inflation or credit growth tend to accelerate as the expansion matures (Kiley 2018, Adrian et al. 2019, Schularick and Taylor 2012). But if this is the main reason, we need to acknowledge the role of policy errors. Is monetary policy letting inflation get out of control, forcing central banks to switch to disinflationary policies with an associated recession? Or why is macroprudential policy not able to manage imbalances in the financial system?
The missing ingredient: The low speed of recovery
External shocks or imbalances can trigger a recession and can explain why expansions might end too early. But to explain the recurrent pattern in US labour markets, we need something else: a slow speed of labour market recoveries. The best example is the last expansion (2009–2020). It ended because of a global pandemic but it was ongoing for more than ten years. Why did it take ten years for the labour market to go from 10% to 3.5%? What if we had achieved unemployment of under 4% by 2015 – could the US have enjoyed low and stable unemployment for five years? The same could be said about the build-up of imbalances. Why do they build faster than the time it takes for labour markets to heal?
The work of Hall and Kudlyak (2021) has highlighted a very important stylised fact about the US labour market: labour market recoveries are slow and too linear relative to the predictions of standard economic theory. There are two set of related stylised facts that support this conclusion and that can be seen in Figure 2.
- Labour market recoveries are very similar across all expansions, regardless of how bad the preceding recession was. This means that if we start an expansion with a high unemployment rate (as in 2009), it will take longer to get to the same low unemployment rate (i.e. a longer expansion will follow). The ongoing expansion is an outlier, with a speed of recovery that is much faster than any previous expansion.
- Within any expansion, the speed of recovery does not change much as time passes. Theory predicts that the decline in unemployment should be much faster in early years when unemployment rates are high or employment rates are low. In Fatas (2021), I compare this pattern to the predictions of a simple search model. I show that the implied effect of labour market slack on the speed of recovery is five to ten times smaller in the data than in the model.
Figure 2 Labour market recovery during US expansions
To explain these two stylised facts we need models of the labour market with significant frictions to the process of hiring that slows down the initial phase of the recovery. For example, theories of congestion in the labour market such as Mercan et al. (2021) are much better at matching the data.
The plucking model with a twist
Our framework fits well with the ‘plucking model’ of the business cycle where recessions are shocks that take the economy away from potential (Friedman 1993, Dupraz et al. 2019). But our results add a couple of twists to this framework. First, the economy spends little time close to potential. Statistically, when the economy is close to potential, the likelihood of a recession increases rapidly. Second, regardless of the magnitude of the shock, the economy always recovers at a similar rate, and at a rate that is too slow compared to what standard theories predict. With a slow recovery, the race between healing the labour market is lost to the external shocks and imbalances that generate the next recession.
One way to highlight the relevance of these additional insights is to go back to the issue of measurement of the natural rate of unemployment or full employment. We can think about the unemployment rate at the end of an expansion phase as an approximation of full employment. But in Fatas (2021) I show that 50% of the variation of that measure can be explained by two variables: how high the unemployment rate was at the beginning of the preceding recession, and how long the expansion was. This fits well with the notion of a slow and stable speed of recovery during expansions that gets interrupted too early by the next recession. As a result, unemployment rates at the end of an expansion are history-dependent, which makes the interpretation of these numbers as indicators of the natural rate of unemployment problematic.
From a policy point of view, the recipe is straightforward. Finding ways to increase the speed of labour market recovery either through improving the efficiency of labour market matches or by providing policies that better support aggregate demand in the early phase of the expansion could potentially allow the labour market to spend more years close to full employment. And this needs to be done while macroprudential polices ensure that imbalances do not build faster than the speed at which the labour market heals.
CBO – Congressional Budget Office (2015), An Update to the Budget and Economic Outlook: 2015 to 2025, August.
Fatas, A (2019), “The 2020 (US) Recession”, VoxEU.org, 12 March.
Fatas, A (2021), “The Elusive State of Full Employment”, CEPR Discussion Paper 16535.
Adrian, T, N Boyarchenko, and D Giannone (2019), "Vulnerable Growth", American Economic Review 109: 1263–89.
Dupraz, S, E Nakamura, and J Steinsson (2019), "A plucking model of business cycles", NBER Working Paper 26351.
Friedman, M (1993), "The plucking model of business fluctuations revisited", Economic Inquiry 31: 171–177.
Hall, R E and M Kudlyak (2021), "Why Has the US Economy Recovered So Consistently from Every Recession in the Past 70 Years?", NBER Macroeconomics Annual.
Kiley, M T (2018), "Unemployment Risk", Finance and Economics Discussion Series, Board of Governors of the Federal Reserve System.
Mercan, Y, B Schoefer, and P Sedláček (2021), "A Congestion Theory of Unemployment Fluctuations", NBER Working Paper 28771.
Schularick, M and A M Taylor (2012), "Credit booms gone bust: Monetary policy, leverage cycles, and financial crises, 1870-2008", American Economic Review 102: 1029–61.
Tracy, J, R Rich, S Kapon, and E Fu (2015), “Mind the Gap: Assessing Labor Market Slack”, Liberty Street Economics, 25 August.
Williamson, S D (2015), “How Tight Is the Labor Market?”, Federal Reserve Bank of St. Louis On The Economy Blog, 2 November.
1 Using this logic, I suggested in 2019 that the US economy was getting close to a recession (Fatas 2019). The prediction turned out to be right, but for reasons that had nothing to do with previous cycles: a global pandemic resulted in the end of that expansion.