Estimating hysteresis effects
Francesco Furlanetto, Ørjan Robstad, Pål Ulvedal, Antoine Lepetit 09 November 2020
Macroeconomists are used to decomposing movements in real economic activity into an upward trend and transitory fluctuations around the trend, interpreted as business cycles. According to this conventional view, the trend is determined by supply-side factors, such as developments in technology and changes in labour supply, while the business cycle is mostly driven by shocks to the components of aggregate demand and monetary policy. This trend-cycle decomposition is embedded in the standard toolkit of modern macroeconomic analysis. On the one hand, dynamic stochastic general equilibrium (DSGE) models imply that demand factors have either no effect at all or only a small transitory effect on the productive capacity of the economy, depending on the parameterisation and the definition of potential output (Blanchard 2018). On the other hand, structural vector autoregressions (SVAR) are often estimated imposing the identification scheme used in the seminal paper by Blanchard and Quah (1989), which assumes the presence of one (and only one) shock with potentially permanent effects on output and one (and only one) shock with zero long-run effects on output. The former is traditionally interpreted as a supply shock while the latter is seen as a demand shock. The ‘independence assumption’ that productive capacity is independent of monetary policy, and more generally of demand factors, has become the dominant paradigm in macroeconomics and is the basis of the inflation-targeting framework used by most central banks (Blanchard 2018).
One alternative (and minority) view, popularised by Blanchard and Summers (1986), states that fluctuations in demand (and large recessions in particular) may have a permanent effect on the productive capacity of the economy through hysteresis effects. Economic developments in Europe in the 1980s seemed to support this view since unemployment was stabilising at a higher level following every recession. However, the long period of stability referred to as the Great Moderation was interpreted by many economists as supportive of the conventional view, and research on hysteresis largely disappeared. The idea that recessions may have long-run effects has re-emerged in the aftermath of the Great Recession as estimates of potential output have been lowered continuously over several years. While these revisions could also reflect lower pre-crisis trends in productivity growth and labour supply somewhat masked by the boom in the pre-Great Recession period (Fernald et al. 2017, among others), Summers (2014) interpreted them as evidence of hysteresis and stated that “[a]ny reasonable reader of the data has to recognise that the financial crisis has confirmed the doctrine of hysteresis more strongly than anyone could have anticipated“.
In a recent paper (Furlanetto et al. 2020) we run a horse-race between the conventional view and the hysteresis view by proposing a simple extension of the SVAR framework proposed by Blanchard and Quah (1989). We allow for (without imposing) hysteresis effects to play a role in economic dynamics. More specifically, we disentangle two shocks (rather than one) with potentially permanent effects on economic activity: a traditional supply shock and a more novel demand shock that are separately identified on the basis of the short-run co-movement between output and prices. Similarly, we also decompose the transitory shock into two components: a demand and a supply shock, both with zero long-run effect on output. In practice, we combine long-run and sign restrictions to identify a SVAR using the state-of-the-art methodology proposed by Arias et al. (2018). We estimate our model over the sample 1983:Q1 to 2019:Q4 and focus our attention on the demand shock with potentially permanent effects on output. Notably, we evaluate its importance for economic fluctuations and its transmission mechanism. The more important this shock is, the larger are the deviations from the independence assumption and the larger is the role for hysteresis effects.
Figure 1 Variance decomposition
Our main result is on the relevance of hysteresis effects. Figure 1 plots the forecast error variance decomposition of the model. The vertical axis indicates the share of the variance explained by the different shocks, while the horizontal axis indicates the horizon. We find that demand shocks with potentially permanent effects are important in the US: they explain almost 50% of long-run output fluctuations. While not dominant, such an important role for demand shocks in the long-run highlights that the traditional interpretation of the Blanchard-Quah shock with permanent effects as a supply shock is not warranted. Both supply-side and demand-side factors with long-run effects are needed to jointly explain data on output, prices, employment, and investment. Interestingly, as shown in Figure 2, we also find that hysteresis effects are concentrated in the three recessions included in our sample (1990-1991, 2001, and 2007-2009). Consistent with this observation, these effects are less important, yet not negligible, if the model is estimated on a shorter sample ending just before the Great Recession.1
Figure 2 Historical decomposition of the growth rate in GDP per capita
Our second result relates to the transmission mechanism of hysteresis effects. In Figure 3, we present impulse response functions to contractionary demand and supply shocks with permanent effects. To investigate the transmission mechanism in more detail we first decompose the long-run effect on output into an effect on employment and an effect on output per worker. We find that a decline in demand propagates almost exclusively through employment. Output per worker barely responds to the shock. In contrast, output per worker drops sharply and durably in response to a supply shock, thereby showing that that the two shocks with permanent effects transmit to the economy in meaningfully different ways.
Figure 3 IRFs of the two shocks with potentially permanent effects on output
We now dig deeper in the analysis and study the transmission of demand shocks with potentially permanent effects to more macroeconomic variables. To expand the variable set we estimate impulse responses using local projection methods. On the employment side, we find that a decline in demand triggers an increase in unemployment, a decline in participation, and an increase in applications (and awards) for disability insurance. These results are consistent with standard hysteresis channels based on the skill depreciation and reduced employability of the long-term unemployed, or on the loss of morale of workers as a result of the poor state of the labour market.
As for labour productivity, we uncover that the negligible effects on output per worker are the result of competing forces. On the one hand, a decline in demand leads to a large and persistent decrease in investment (both R&D and total). Through this channel, capital intensity and labour productivity should decline. On the other hand, contractionary demand shocks hit less productive workers disproportionately, which pushes up labour productivity according to a standard composition effect. We demonstrate this last point by considering the response of several variables to a contractionary demand shock with permanent effects. First, both utilisation-adjusted TFP, which controls for changes in the stock of capital, and a measure of labour quality increase following the shock. Second the employment share of workers with a bachelor’s degree or higher increases, meaning that employment for skilled workers declines less than for other workers. Third, the number of employed workers performing routine (and arguably less productive) tasks, as a ratio of total employment, is negatively affected, in keeping with the fact that job polarisation takes place mainly in recessions and generates jobless recoveries, as shown by Jaimovich and Siu (2020).
These results are compatible with the view that recessions (that are mainly driven by demand shocks) are periods of intense reallocation in which the least productive workers and units are disproportionately affected. This finding echoes the idea that recessions have a cleansing effect on the productive system, an idea attributed to Schumpeter and Hayek and revived by Caballero and Hammour (1994).
All in all, using a simple extension of the Blanchard-Quah decomposition over the period 1983-2019, we find evidence that demand-driven recessions (and large recessions in particular) have long-lasting effects, are associated with prolonged declines in investment and employment, but do not materially affect labour productivity. We believe these empirical findings will prove useful in informing the policy discussion and distinguishing between alternative theories of hysteresis. They may also be of particular relevance today as economists and policymakers are trying to assess the long-run scarring effects of the current COVID-19 crisis.
Authors’ note: The views expressed in this column are those of the authors and should not be attributed to Norges Bank or to the Board of Governors of the Federal Reserve System.
Arias, J, J Rubio-Ramirez and D Waggoner (2018), “Inference based on structural vector autoregressions identified with sign and zero restrictions: Theory and applications”, Econometrica 86(2): 685-720.
Blanchard, O (2018), “Should we reject the natural rate hypothesis?”, Journal of Economic Perspectives 32(1): 97-120.
Blanchard, O and D Quah (1989), “The dynamic effects of aggregate demand and aggregate supply”, American Economic Review 79(4): 655-673.
Blanchard, O and L Summers (1986), “Hysteresis and the European unemployment problem”, NBER Macroeconomics Annual 1: 15-78.
Caballero, R and M Hammour (1994), “The cleansing effect of recessions”, American Economic Review 84(5): 1350-68.
Fernald, J, J Stock, R Hall and M Watson (2017), “The disappointing recovery of output after 2009”, Brookings Papers on Economic Activity 48: 1-58.
Furlanetto, F, Ø Robstad, P Ulvedal and A Lepetit (2020), “Estimating hysteresis effects”, Norges Bank Working Paper 13/2020.
Jaimovich, N and H Siu (2020), “Job polarization and jobless recoveries”, Review of Economics and Statistics 102(1): 129-147.
Summers, L (2014), “Fiscal policy and full employment”, Speech at Center on Budget and Policy Priorities Event on Full Employment.
1 Interestingly, demand shocks with permanent effects play a negligible role if the model is estimated over the period 1949:Q1 to 1982:Q4, a time when recoveries were not ‘jobless’.