Yothin Jinjarak, Rashad Ahmed, Sameer Nair-Desai, Weining Xin, Joshua Aizenman 06 July 2020
The COVID-19 shock hit the euro area in March 2020, mushrooming into a major pandemic that tested the medical, social, and economic capabilities of euro area countries (Bénassy-Quéré and Weder di Mauro 2020). Within two months, the enormity of the health and economic threats became clear. The virus wreaked havoc on both domestic and external demand in France, Italy, and Spain, while contractions in Germany (and smaller euro area economies) were also sizeable, but less severe. As a result, COVID-19 triggered costly containment policies, the collapse of aggregate demand and international trade, and a sharp drop in the GDP of OECD countries. In March 2020, the ECB activated the Pandemic Emergency Purchase Programme (PEPP), tuned to deal with the evolving dire financial and fiscal needs of the euro area (Lane 2020).1 In Jinjarak et al. (2020), we take stock of daily data from Janurary to June 2020, evaluating the impact of COVID-19 dynamics, and the ECB’s and countries’ fiscal policies on the patterns of sovereign spreads in the euro area. We apply a case study multi-stage methodology, comparing the impact of prevailing market factors against that of COVID-19 dynamics, the ECB’s fiscal policies, and countries’ fiscal policies on the sovereign spreads of euro area countries. We also assess the overall financial and fiscal adjustments to collapsing demand.
In the first stage, we estimate a dynamic factor model on the pre-COVID period data. Our outcome variable is the daily change in the log Credit Default Swap (CDS) spread of euro area countries. We control for two key factors. A global factor (capturing the common component of sovereign risk fluctuations at the global level), and a regional factor (capturing common fluctuations within the euro area). As a result, each country has its own unique global and regional ‘betas,’ which capture country-specific systematic exposure to aggregate global and regional risk, respectively.2 We estimate the model over the pre-COVID period from 01 January 2014 through to 30 June 2019. We use the remaining six months of 2019 to validate the ‘out-of-sample’ efficacy of our model (prior to the COVID shock in 2020). Finally, the sovereign spread ‘COVID residuals’ are constructed by comparing the realised change in log CDS each day (after the estimation period to the model-expected value), given the true realisations of the factors and lagged log CDS change.
Figure 1 plots ratios of public debt to GDP for euro area countries against their estimated global betas and regional betas (estimated from the factor regressions with daily data). The public debt-to-GDP measure is an average over annual data, spanning from 2014 to 2018. The association between public debt levels and global betas are weak, while public debt levels are positively associated with regional betas across selected countries.3
Figure 1 Public debt/GDP versus global and regional betas of sovereign CDS spreads
Note: Public debt measures are the 2014-2018 average. Betas are estimated during the pre-COVID period, Jan 2014 – June 2019.
Figure 2 shows the correlation between COVID-related fiscal stimulus announced in 2020 and country-specific global and regional betas (estimated over 2014-2019). Regional betas are significantly and negatively associated with COVID-related fiscal stimulus size across the euro area. That is, systematically riskier countries issued less stimulus in realtion to GDP. These findings support the view that the risk associated with limited fiscal space is ‘priced’ into euro area sovereign spreads during ‘normal times,’ and country risk possibly constrained (to some degree) the size of the COVID-19 stimulus that various euro area countries were able to deploy.
Figure 2 Announced 2020 COVID-related fiscal stimulus versus global and regional betas
Note: COVID-19 fiscal stimulus data taken from the IMF COVID policy tracker. Betas are estimated during the pre-COVID period, Jan 2014 – June 2019.
After estimating the two-factor model from 1 January 2014 to 30 June 2019, we extrapolate the model based on realised values from 1 July 2019 through to 15 June 2020. The upper-left panel of Figure 3 traces the euro area average cumulative log CDS change over this period (solid line), against the progression implied by the model (dashed). It should be noted that the model performs reasonably well in tracing the realised values from 1 July 2019 to 31 December 2019 (the ‘validation period’). However, in March 2020, the realised values diverged from the model-implied values. This was triggered by panic over the COVID-19 pandemic. Hence, at the aggregate euro area level, the factor model alone could not explain all of the variation in CDS adjustment due to the COVID-19 shock. Sovereign CDS spread widening ceased almost immediately around 18 March 2020 (the first vertical line), when the ECB announced the PEPP. However, the divergence between actual and model-implied changes persisted. The subsequent vertical line represents 4 June 2020, when the ECB announced the doubling of the programme.
Figure 3 Euro area sovereign CDS spreads, July 2019 – June 2020
Note: The dashed line in the upper-left figure reflects model-implied values. Vertical lines reflect May 18th and Jun 4th. The lower-right figure, Actual-Fitted, reflects the cumulative sovereign spread COVID residual, where the model was estimated to Jan 2014 – June 2019 data.
The upper-right panel of Figure 3 charts the cross-sectional dispersion of CDS spreads over the same period, highlighting the sharp rise in volatility amid the COVID-19 panic in March 2020. The lower charts compare high mortality versus low mortality euro area countries (by the end of April 2020). Figure 4 compares GIIPS versus non-GIIPS, and GIIPS versus Core.
Figure 4 GIIPS versus other euro area countries
Note: Vertical lines reflect May 18th and June 4th. RHS figures, Actual-Fitted, reflects the cumulative COVID residual, where the model was estimated to Jan 2014-June 2019 data. Core defined as Germany, France, Belgium, Netherlands.
Tracing the realised evolution of sovereign CDS spreads of high- versus low-mortality countries in the euro area,4 the trends between the two groups were entirely parallel up until around May 18th. At that point, we see a persistent gap that emerged between high mortality and low mortality countries (Figure 3, lower-left panel). The gap persists when comparing the cumulative COVID residual between these two groups (Figure 3, lower-right panel), suggesting that this divergence is possibly being driven by COVID-specific risks rather than traditional determinants like fiscal space or systematic risk.
The GIIPS (Greece, Italy, Ireland, Portugal, Spain) group cumulative sovereign spread COVID residuals (Figure 4, upper-right panel) are far more volatile than non-GIIPS. Both groups saw similar spikes following the initial COVID-19 panic in March 2020. However, the GIIPS group reverted sharply upon the 18 May 2020 announcement of the ECB PEPP. In contrast, non-GIIPS CDS, on average, rose sharply and remained high. This pattern goes against the conventional view that fiscally fragile countries would realise wider and more persistent credit spreads. Both fragile and resilient countries saw comparative widening in their sovereign CDS spreads.
In the second stage, we separate the out-of-sample COVID-19 pandemic period in 2020 (from January to May 2020) into three subsamples; January to February, March, and April to May. We find that March is the period during which the realised values of daily CDS spread change diverged the most from the model-implied values, while the factor model does well in tracing the realised values before, and after, that period (Figure 5). Additionally, daily CDS spread changes were the most volatile during March. Therefore, we focus on the sample in March 2020 and examine whether COVID-19-specific indicators may account for the variation in CDS adjustment that is not explained by the dynamic factor model.
Figure 5 Euro area average sovereign spread COVID residual
Note: COVID Residual: the difference between the actual CDS adjustment and the change implied by the model, at both the individual country and aggregate EZ levels over the pandemic period. Realized (solid) and fitted (dashed, factor model estimated on 2014-2019 data from Equation ) daily euro area average CDS changes, separated by 2020 time periods.
To do this, we estimate a panel model examining the relationship between the sovereign spread COVID-19 residual and a set of COVID-specific variables. COVID-specific variables are grouped into three categories: (1) mortality outcomes per capita,5 (2) economic activity,6 and (3) policy interventions.7,8 We also control for country- and time-fixed effects. We find that new mortality rates and new mortality growth rates are positively and significantly associated with COVID-19 residuals across all three specifications, and explain the greatest share of variation in COVID-19 residuals among all three categories. This finding implies that debt pricing during the pandemic may have been significantly impacted by country-specific mortalities. Countries that saw higher new mortality rates, or new mortality growth rates, were likely to see a wider divergence in realised sovereign CDS spread dynamics from model-implied values.9 In sum, country-specific mortality outcomes (especially daily new mortality dynamics and country-specific fiscal responses) help account for the variation in CDS spread dynamics that is left unexplained by the dynamic factor model.
Using a slightly modified specification, we compare the explanatory power of the dynamic factor model predictions and COVID-19-specific variables.10 We find that model-implied values no longer trace realised CDS spread changes during the pandemic.11 However, new daily mortalities and the growth rate of new mortalities are both positively and significantly correlated with daily CDS spread changes across all specifications.12 Country-specific announcements of fiscal responses to the pandemic were significantly associated with daily CDS spread changes. Those countries which announced fiscal responses (and thus increased their debt burden) were more likely to experience greater daily CDS spread changes.
Importantly, our results show that according to our set of potential explanatory variables, COVID-19-specific factors explain the greatest share of variation in sovereign CDS spread dynamics during the COVID-19 pandemic period.13 In Figure 6, we chart the aggregate (average) CDS spread dynamics for euro area countries during the pandemic by plotting the aggregate realised values and model-implied values. Surprisingly, the aggregate model-implied values from our regressions (which control for COVID-specific factors) trace the realised values almost perfectly, such that their lines coincide with each other.
Figure 6 COVID-related risks and factors accounted for euro area average CDS
Note: Solid lines reflect the realized daily average euro area CDS spreads changes. Dashed lines reflect predicted average euro area CDS spreads changes implied by the specifications  and  of Table 2, respectively.
Hence, we conclude that COVID-19-specific factors play a crucial role in explaining the divergence of sovereign CDS spread dynamics during the pandemic, which we term: ‘COVID dominance’. The implementation of the PEPP and other programmes substantially reduced the dispersion of euro area sovereign spreads induced by COVID-19-specific risks and associated policies. During the first waves of the pandemic (January to June 2020), these policies prevented self-fulfilling runs on sovereign and corporate debt, thereby freeing the funding resources needed to fight the medical and economic consequences of the pandemic. As a result, these policies increased the fiscal space of the GIIPS (and other indebted countries), supporting expansionary fiscal policy needed to fund the medical and economic struggles associated with COVID-19.
Belz, S, J Cheng, D Wessel, D Gros and A Capolongo (2020), “What’s the ECB doing in response to the COVID-19 crisis?”, CEPS, 26 May.
Bénassy-Quéré, A and B Weder di Mauro (2020), Europe in the Time of Covid-19, a VoxEU.org eBook, CEPR Press.
Cheng, J, D Skidmore and D Wessel (2020), “What’s the Fed doing in response to the COVID-19 crisis? What more could it do?”, Brookings, 12 June.
Hale, T, S Webster, A Petherick, T Phillips and B Kira (2020), “Oxford COVID-19 Government Response Tracker”, Blavatnik School of Government.
Lane, P (2020), “Pandemic central banking: the monetary stance, market stabilisation and liquidity”, Remarks at at the Institute for Monetary and Financial Stability Policy Webinar, 19 May.
Jinjarak, J, R Ahmed S Nair-Desai, W Xin and I Aizenman (2020), “Pandemic Shocks and Fiscal-Monetary Policies in the Eurozone: COVID-19 Dominance During January – June 2020”, NBER Working Paper 27185.
1 On March 18th, 2020 the ECB activated a new QE line of 750 billion Euro Pandemic Emergency Purchase Programme (PEPP), targeting national and regional government bonds, including Greek sovereign debt, and various private sector bonds. On June 4, 2020, the ECB almost doubled the PEPP, increasing its size up to 1350 billion Euros. These new policies have increased sharply the ECB’s balance sheet to about half of EZ pre-COVID GDP. The pandemic dynamics in the US induced in March 2020 massive fiscal stimulus, augmented by the expansion of QE policies to a wide spectrum of economic activities, and the provision of ample liquidity to foreign countries via swap lines and new repo facilities. The size of these innervations dwarfs the Fed’s policies during the GFC. These policies include up to $2.3 trillion in lending to support households, employers, financial markets, and state and local governments. In addition, the Fed activated its international swap lines at low interest rate to Canada, England, the Eurozone, Japan, and Switzerland, and extended the maturity of those swaps. It has also extended the swaps to the central banks of Australia, Brazil, Denmark, Korea, Mexico, New Zealand, Norway, Singapore, and Sweden. The Fed is also offering dollars to central banks that don’t have an established swap line through a new repo facility called FIMA (for “foreign and international monetary authorities”). The Fed will make overnight dollar loans to the central banks, taking U.S. Treasury debt as collateral. See Cheng et al. for more on the Federal Reserve’s policies (2020), and Belz et al. (2020) for the ECB’s policy response to the COVID crisis.
2 It is important to note that the model is heterogeneous, allowing for regression estimates across countries to vary. The main assumption we make is that the estimated factor ‘betas’ capture the most important determinants which influenced sovereign spreads over the 2014-2019 period.
3 If Greece is excluded due to its remarkably large debt/GDP, the positive correlation between regional betas and fiscal space strengthens sharply, turning significant at the 10% level, confirming the notion that riskier CDS spreads are associated with fiscal fragility.
4 High mortality-per-capita countries: Belgium, Spain, Italy, France, Netherlands. Low mortality-per-capita countries: Slovakia, Latvia, Cyprus, Greece, Lithuania. We consider mortality-per-capita as of the end of April 2020.
5 We include daily new mortality rate (per 1,000,000 population), daily new mortality growth rate, total mortality rate (per 1,000,000 population), and total mortality growth rate.
6 We include daily mobility measure in terms of driving (reported by Apple) and daily growth rates of policy Stringency Indices). Lower mobility levels or stricter government non-pharmaceutical interventions may signal greater economic contraction, which may increase the debt financing burden and thus impact debt pricing during the COVID-19 pandemic.
7 We include dummy variables indicating the date of country-specific key fiscal policy announcement, the date of European Commission’s fiscal policy announcement, the date of European Central Bank’s Pandemic Emergency Purchase Programme (PEPP) announcement, and the date of the Federal Reserve’s monetary policy announcement In order to capture key policy announcements, we aggregated a set of variables from numerous datasets for a set of key fiscal, monetary, and miscellaneous policies, across individual countries, the European Union as a whole, the European Central Bank, and the Federal Reserve. These variables captured whether or not an action or proposal was made by a given nation/institution on a specific date in the sample. Thus, we did not control for the size or number of policies on any given day, and only if the date corresponded with the announcement of at least one key policy. With the exception of the Federal Reserve (whose major announcements related to reductions in the interest rate along with fiscal spending), we restricted our analysis of key fiscal policies to those which provided “millions” or “billions” of local currency units in spending.
8 The primary data sources used to construct these policy announcement variables are listed here: Yale COVID-19 Financial Response Tracker; Harvard Global Policy Tracker; Bruegel COVID-19 National Dataset; IMF Policy Responses to COVID-19; OECD COVID-19 Action Map; St. Louis Federal Reserve; and the European Parliament.
9 In contrast, mobility measures and the Stringency Index growth do not seem to correlate with COVID residuals. Among all policies, country-specific fiscal policy announcements have a significantly positive association with COVID residuals, indicating that countries that increase their debt burdens were likely to see larger discrepancy in CDS spread dynamics.
10 We do this by treating realized log changes in CDS spreads as the outcome variable, while augmenting the panel regression with our model-implied values from values of daily CDS spread change generated from the dynamic factor model equation on the RHS along with COVID-specific variables. Essentially, we take apart the two components that make up the COVID residual.
10 These results are fully consistent with fragile EZ countries, like the GIIPS, exhibiting relatively low CDS adjustment given their fiscal space compared to other EZ countries, as we showed in the previous section.
11 Consistent with time-series evolution of the COVID residuals shown in the previous section (Figures 3), countries that had higher level of daily new mortality rates or higher new mortality growth rates were likely to realize more severe daily CDS spread changes.