A string of recent VoxEU contributions has drawn attention to the accelerating economic divergence across subnational regions in many countries (e.g. Iammarino et al. 2018, Rosés and Wolf 2018, Goodhart and Venables 2020). The debate on underlying causes and possible remedies is far from settled. But one unifying feature is that most contributions have abstracted from monetary policy as a potential factor – with notable exceptions including Haldane (2016), Cœuré (2018) and Haldane (2020).
This abstraction, in turn, clashes with the active role that monetary policy has played in steering aggregate economic conditions over the past decade and with the increasingly widespread recognition that monetary policy may trigger relevant distributional effects, for instance at the household level (e.g. Ampudia et al. 2018).
In a recent paper, we contribute to closing this gap by exploiting granular data on economic activity at the city- and county-level to shed light on the impact of monetary policy on regional inequality in the euro area (Hauptmeier et al. 2020).
Heterogeneity across space and time
At first glance, these data confirm the pattern of pronounced and intensifying economic divergence that has been highlighted also in other contexts.1 In fact, the dispersion is particularly accentuated at this disaggregated level, with the coefficient of variation for per-capita GDP across cities and countries exceeding that across euro area countries by around 50%.
Moreover, economic fortunes have drifted further apart in the upper and lower parts of the distribution, especially in the aftermath of the global financial and economic crisis of 2007/08. This crisis triggered a steep fall in euro area economic activity, followed by a double-dip recession, both of which are clearly visible at the mean of the regional per capita GDP distribution (Figure 1). Vastly different trajectories emerge for the outer parts of the distribution however. For instance, regions at the 95th percentile on average experienced a solid recovery after 2009, whereas per capita GDP at the 5th percentile just continued drifting down after the crisis and only showed a mild turnaround in the last two sample years.
Against this background, the regional level not only presents itself as an interesting testing ground for understanding the geographical patterns of monetary policy transmission, but also emerges as a crucial issue in its own right for policy-makers interested in tackling regional heterogeneity.
Figure 1 Evolution of GDP per capita
Source: Hauptmeier, Holm-Hadulla and Nikalexi (2020).
Notes: The lines show the percentiles of the regional per-capita GDP distribution in the euro area. The percentiles have been normalised to 100 in 2008.
The uneven incidence of monetary policy shocks
To study whether monetary policy tends to mitigate or aggravate the trend towards greater regional inequality, we set up a panel local projections framework, estimated with the quantile fixed effects estimator of Machado and Silva (2019). Our identification strategy makes use of the fact that our outcome variable is measured at a more granular level than those entering the central bank reaction function. This in turn allows us to control for aggregate factors which, given the explicit macroeconomic stabilization mandates of central banks, typically make it difficult to disentangle the cause and effect of observed co-movements between monetary policy indicators and the economy.
The estimates point to pronounced heterogeneity in the regional patterns of monetary policy transmission, which intensifies in monotonous fashion towards the outer tails of the distribution. For instance, Figure 2 compares the dynamic response of regional output to a monetary policy shock for the 5th and 95th percentile of the distribution. In both parts of the distribution, output contracts after the shock, but the contraction is much sharper at the lower end. As a consequence, the peak output losses at the 5th percentile are around one third higher than at the 95th percentile.
Moreover, this gap widens over time. In fact, while output in the upper parts of the distribution fully recovers, the contraction proves persistent in the lower parts, with the estimates essentially moving sideways after the first year and still remaining close to the trough after five years.
Figure 2 Impact of monetary policy on output in the tails of the distribution
Source: Hauptmeier, Holm-Hadulla and Nikalexi (2020).
Notes: Vertical axis refers to impact of 100 basis point policy rate hike on regional GDP (in %). Horizontal axis refers to horizon of IRF (in years). Solid lines denote point estimates and shaded areas denote 90% confidence bands. Estimates are based on quantile fixed effects regressions.
Sources and implications of hysteresis
The result of monetary policy shocks triggering long-lived adjustments in the output of poorer regions is striking. First, it implies that monetary policy has long-lasting effects on regional inequality, with tightening shocks aggravating and easing shocks mitigating it. Second, it contrasts with the common notion of monetary policy merely causing transitory adjustments in the real economy and, as such, it adds to a long-standing debate on potential sources of long-term monetary non-neutrality. This debate has enjoyed a revival in the aftermath of the 2007/08 financial crisis (Yellen 2016), with the most recent contribution by Jordà et al. (2020) also documenting lasting effects of monetary policy on GDP in a panel of advanced economies.
To better understand these patterns, we also explore the forces leading transitory policy shocks to morph into durable adjustments in the economy, a phenomenon often referred to as ‘hysteresis’. In view of a broad literature originating from Blanchard and Summers (1986), it appears natural to include labour markets in the list of usual suspects here. However, recent findings by Jordà et al. (2020) challenge this impulse as they document hysteresis in the capital stock and total factor productivity, but not in labour.
Drilling down into the anatomy of hysteresis in our data, we confirm labour markets as a dominant, but not the only source of the long-lived real effects of monetary policy shocks. To see this, Figure 3 presents the estimated effects five years after the shock, separately for regional employment and labour productivity, for different parts of the per-capita GDP distribution. It shows that the monetary policy effect after five years is still statistically significant and economically relevant for both variables in the lower parts of the distribution. At the same time, employment hysteresis is more pronounced and more broad-based across the distribution and, in fact, it even extends to the sample mean. By contrast, productivity hysteresis concentrates in the lower tails of the distribution and the coefficient already becomes statistically indistinguishable from zero in the 20th to 30th percentile range.
Taken together, the heterogeneous incidence of hysteresis implies that monetary policy has a long-lasting impact on regional inequality in that tightening shocks aggravate and easing shocks mitigate it; and this pattern is particularly pronounced for regional employment outcomes.
Figure 3 Impact on employment and labour productivity at the end of the estimation horizon
Source: Hauptmeier, Holm-Hadulla and Nikalexi (2020).
Notes: LHS panel shows impacts of a 100 basis point rate hike on employment at horizon h = 5 (in %) for the decile ranges from: 0-10; 20-30; 60-70; and 90-100 as indicated in the legend; RHS panel shows corresponding impacts on labour productivity. Diamonds indicate point estimates and bars indicate 90% confidence intervals.
COVID-19 and interregional inequality – some indicative evidence
A natural question is whether our empirical framework yields any insights into how the economic fallout from the ongoing COVID-19 crisis is likely to shape the trends in regional inequality. One potential avenue is to exploit the link that our model establishes between economic activity at the national and at the regional level. In particular, national GDP is included among the set of explanatory variables and our model yields quantile-specific coefficients on all regressors.
So, starting from the presumption that the COVID-19 crisis will trigger a pronounced drop in aggregate GDP, we can use these coefficients to gauge the resultant declines in different parts of the regional distribution. Importantly, this exercise takes a rather narrow perspective: first, since national GDP serves as a control variable in our model, the coefficients merely capture dynamic correlation patterns, rather than causal relationships; second, the regional incidence of the ongoing public health emergency is likely to evolve in a complex fashion, which our model does not incorporate but which is likely to also shape its economic consequences; third, the exercise is based on a partial perspective that abstracts from certain important variables, e.g. potential policy responses at the euro area level.
With these caveats in mind, Figure 4 shows the typical pattern in regional GDP for the bottom and top 10% of the distribution after a 1% contraction in national GDP. In both parts, regional GDP is highly synchronised with national GDP, as visible from the coefficients near -1 in the first and second year after the contraction. But the point estimates then diverge and GDP in the upper part of the distribution returns to its initial level already in the second year after the national GDP contraction, in contrast to the lower part, where it takes two additional years. To the extent these historical regularities provide a decent guide for future adjustment patterns, this suggests that regions in the lower parts of the distribution take longer to recover from the deterioration in aggregate conditions. This, in turn, would imply a temporary increase in regional inequality.
Figure 4 Dynamic relationship between regional and national GDP
Source: Hauptmeier, Holm-Hadulla and Nikalexi (2020).
Notes: Vertical axis refers to coefficient of national GDP (scaled to a 1% contraction). Horizontal axis refers to horizon of IRF (in years). Diamonds denote point estimates and ranges denote 90% confidence bands.
References
Ampudia, M, D Georgarakos, J Slacalek, O Tristani, P Vermeulen and G Violante (2018), “Monetary Policy and Household Inequality”, ECB Working Paper No. 2170.
Blanchard, O J and L H Summers (1986), “Hysteresis and the European Unemployment Problem”, NBER Macroeconomics Annual 1: 15–78.
Cœuré, B (2018), “The Local Impact of the ECB’s Monetary Policy”, European Central Bank.
Goodhart, C and A Venables (2020), “Regenerating the Cities that Were Left Behind”, VoxEU.org, 17 January.
Haldane, A (2016), “Whose Recovery?” Bank of England.
Haldane, A (2020), “Central Banks and Regional Inequality”, VoxEU.org, 7 February.
Hauptmeier, S, F Holm-Hadulla, K and Nikalexi (2020),“Monetary Policy and Regional Inequality”, ECB Working Paper No. 2385.
Iammarino, S, A Rodríguez-Pos and M Storper (2018), “Regional Inequality in Europe: Evidence, Theory and Policy Implications”, VoxEU.org, 13 July.
Jordà, Ò, S R Singh and A M Taylor (2020), “The Long-Run Effects of Monetary Policy”, Federal Reserve Bank of San Francisco.
Machado, J A and J S Silva (2019), “Quantiles via Moments”, Journal of Econometrics 213(1): 145–173.
Rosés, J and N Wolf (2018), “The Return of Regional Inequality: Europe from 1900 to Today”, VoxEU.org, 14 March.
Yellen, J L (2016), “Macroeconomic Research After the Crisis. The Elusive ‘Great’ Recovery: Causesand Implications for Future Business Cycle Dynamics”, 60th Annual Economic Conference, Board of Governors of the Federal Reserve System Speeches.
Disclaimer
The views expressed here are those of the authors and do not necessarily reflect those of the European Central Bank.
Endnote
1 The data range from 1999 (the start of the euro) to 2015 (the last year of data availability) and cover the cities and countries in the founding member countries in the euro area (excluding Luxembourg) and Greece. The source is the Regional European Database of Cambridge Econometrics, which is based on the Nomenclature of Territorial Units for Statistics (NUTS) and combines data from Eurostat’s REGIO database and the European Commission’s AMECO database. Our analysis refers to NUTS3 regions, which offer the highest level of disaggregation for which information on economic activity is available.
