Gross Fixed Capital Formation in the Euro Area During the COVID-19 Pandemic

This paper examines the impact of the COVID-19 pandemic on gross fixed capital formation across the euro area. The empirical analysis suggests that the intensity of the lockdown measures to contain the spread of the virus and the country-specific structure of the economy along with other traditional drivers, in particular falling output, can explain a large part of the contraction. The bold policy response at the national and EU level mitigated the impact of COVID-19 and supported the recovery. The faster-than-expected rebound in economic activity suggests that the negative economic impact of the pandemic will be more contained than initially feared. However, uncertainty over future health developments remains high, especially given the risks of new more transmissible variants.

Following the outbreak of the COVID-19 pandemic, gross fi xed capital formation (GFCF) in the euro area fell very rapidly in the fi rst and second quarter of 2020, much faster than at the height of the global fi nancial crisis. The sharp contraction in GFCF prompted many commentators to highlight the risks that the pandemic could lead to another period of subdued investment growth similar to the one following the global fi nancial crisis, when it took about ten years 1 to return to its pre-crisis level. 2 However, GFCF recovered (although only partially) at a much faster pace than in the wake of the fi nancial crisis ( Figure 1). The multifaceted and sizable policy response at the national and EU level mitigated the impact of the 1 In the national accounts (ESA, 2010), gross fi xed capital formation covers machinery, equipment, buildings and structures, as well as cultivated biological resources and intellectual property products. 2 After the global fi nancial crisis, the loss of capital stock was the main drag on potential output growth (ECB, 2020).
crisis and the plunge in GFCF at the onset turned out to be short-lived. Investment bounced back forcefully in the context of very strong (and temporarily held back) demand and favourable fi nancing conditions (European Commission, 2021a, 2021c. Public investment also picked up considerably.
This paper examines how the COVID-19 pandemic aff ected investment across the euro area. First, unlike previous investigations that have tended to focus on the impact of the COVID-19 crisis on overall GDP, it assesses the impact of the crisis and lockdown measures on GFCF. Second, this paper estimates the sensitivity of GFCF to lockdown measures over time and across countries supporting the idea of ongoing learning from experiences and gradual adaptation, which includes greater digitalisation. Third, it provides an assessment of the upside and downside risks for GFCF from COVID-19.

Gross fi xed capital formation during the COVID-19 pandemic
Following the COVID-19 shock, gross fi xed capital formation contracted by around 23% between the fourth quarter of 2019 and the second quarter of 2020. Over the same period, GDP fell by 15% and the decline in investment was the second largest cause for this overall contraction (following the drop in consumption). This contraction was much larger than the one recorded following the outbreak of the global fi nancial crisis ( Figure 2). 3 What was extraordinary about the decline in 2020 was that it all happened Investment Figure 1 Gross fi xed capital formation and GDP in the euro area Source: Eurostat.

Figure 2
Gross fi xed capital formation in the euro area during COVID-19 and global fi nancial crisis in just two quarters -mainly due to the tightening of lockdown measures to contain the spread of the pandemic (see below).
Lower investments in machinery and equipment (excluding the very volatile intellectual property products data) accounted for most of the fall in the second quarter of 2021 ( Figure 2), but it rebounded strongly in the third quarter of 2020. By contrast, dwellings and other buildings and structures contributed less to the contraction and they had recovered their pre-crisis levels by the fi rst quarter of 2021. Investment in intangibles, such as research and development, 4 fell less than investment in machinery and equipment.
At the institutional sector level, the fall in private investment was partly compensated by a symmetric rise in public investment as euro area governments pledged substantial public investment to support the recovery from the pandemic. This was in stark contrast to the period following the global fi nancial crisis (Figure 3), which saw euro area governments cutting back on public investment with the aim of hastening the consolidation of public finances.
The depth of the decline in GFCF between the fourth quarter of 2019 and the second quarter of 2020 varied widely within the euro area, ranging from just below 1% in Finland to 80% in Ireland ( Figure 4). Intellectual property -one key and growing component of GFCF -has been particularly volatile in Ireland, Estonia, Cyprus and Luxembourg (see the right-hand side in Figure 4). 4 The volatile Ireland data are excluded.
Part of these cross-country diff erences in investment growth can be attributed to diff erences in the intensity of the lockdown measures (the second quarter of 2020 in Figure 5). As restrictions on movement were lifted between the end of the second and the third quarter of 2020, GFCF rebounded in that third quarter. Lockdown measures were tightened again in the fourth quarter of 2020 on the back of renewed pressures on the member states' health systems; the economic impact of the sec- Changes in gross fi xed capital formation since the onset of COVID-19 Notes: Data on GFCF for IE, CY, LU and EE show very strong volatility in the intellectual property investment component. Total growth (bullet) measures the compound growth rate (i.e. multiplicative). Given the large size of the growth rates, adding quarterly growth rates (coloured bars) is only a rough approximation of the total growth rate between the fi rst quarter of 2020 and second quarter of 2021.
ond lockdown, however, was more contained than that of the fi rst one.

COVID-19 related drivers of gross fi xed capital formation
The literature suggests a strong negative relationship between governments' lockdown measures and GDP (including its components). This negative impact increases with the intensity of measures (e.g. IMF, 2020; Niermann and Pitterle, 2021), the importance of tourism in the economy and lower quality of governance (e.g. Sapir, 2020). However, over time, economic activity became less sensitive to lockdown measures as fi rms and households adapted to the new environment (see Figure 5 and the empirical result below).
Early evidence suggested that higher uncertainty in the initial phase of the COVID-19 crisis (European Commission, 2021d; Gayer et al., 2021) took a toll on business investment. For example, surveying about 13,500 fi rms across the EU in 2020, the EIB (2020) reports that about 80% of EU fi rms considered uncertainty to be an impediment, with some 50% of fi rms even considering it a major impediment. 5 Gieseck and Rujin (2020) report that heightened uncertainty could have accounted for around onefi fth of the decline in activity by the fi rst half of 2020, with a particularly strong impact on fi xed capital formation.
At the beginning of the COVID-19 crisis, fi nancing conditions tightened given the overall uncertainty of the scale and duration of the crisis. However, the increase was shortlived (see Figure 6), following a strong monetary policy response, which prevented fi nancing conditions from tightening in a pro-cyclical way (Lane, 2020). Further fi nancial relief was provided under various state credit guarantee 5 See EIB (2021). European Commission (2020a) reports that fi rms expected a contraction of 4.5% in capital expenditure in 2020 with more than 40% of participants indicating negative expectations. Investment programmes that supported solvable fi rms' access to finance for investment (European Commission, 2020b).
Monetary and supervisory authorities supported the financing of investments in several ways. The ECB's monetary policy response mainly consisted of additional asset purchases including via the pandemic emergency purchase programme, ample liquidity provision (mostly via targeted long-term refi nancing operations), and easing of collateral standards, while maintaining the deposit facility rate at a record low of -0.5% (since September 2019). At the same time, several national macro-prudential authorities reduced countercyclical capital and systemic risk buff ers, while the Single Supervisory Mechanism allowed banks to meet part of their core capital requirements with non-core capital instruments. ed investments via several measures including following the activation of the general escape clause of the Stability and Growth Pact. These measures included emergency spending on health care, short-time work schemes, grants, loan guarantees, loan repayments moratoria, tax deferrals, 7 liquidity support and the roll-out of a vaccination programme.

Empirical results
The impact of the COVID-19 pandemic on quarterly growth in gross fi xed capital formation across the euro area is estimated via a panel error correction model. The model relates investment to output, the past change in capital stock that requires investment to off set capital depreciation, 8 fi nancing costs, a news-based measure of uncertainty 9 and the equity-to-book ratio. To account for the impact of the pandemic, this base model is augmented to include lockdown measures using the Oxford stringency index, 10 a pandemic dummy (equal to 1 for the length of the pandemic since the second quarter of 2020 11 ) that captures the net impact of other factors including fi scal and monetary policy responses. 12 This section reports estimation results for a panel error correction model, covering 15 euro area member states 13 from the fi rst quarter of 2002 to the second quarter of 2021. 14 First, the equilibrium relationship is estimated between the level of gross fi xed capital formation (I) and the level of real GDP, the fi nancing cost 7 And in some countries the introduction of temporary suspensions of bankruptcy proceedings. 8 Net capital stock data with quarterly frequency are interpolated from AMECO annual capital stock series OKND. 9 Uncertainty is measured by the Economic Policy Uncertainty index based on newspaper articles regarding policy uncertainty. However, part of the impact of rising uncertainty may also be captured by other explanatory variables such as the pandemic dummy and lockdown measures. 10 The Oxford COVID-19 stringency index varies between 1 (very loose) and 100 (very tight). It includes several dimensions: (i) lockdown and closure measures; (ii) economic response and (iii) health system measures (see Halle et al., 2020). 11 Complemented with a dummy for the fi rst quarter of 2020 as the fi rst weeks of this quarter were not yet aff ected by the pandemic. 12 A dummy equal to 1 for the length of the pandemic since the second quarter of 2020, complemented with a dummy equal to 1 for the fi rst quarter of 2020. 13 IE, EE, CY and LU are not included as they show strong variability in the intellectual property products component. 14 The main data sources are Eurostat National Accounts and Sectoral Accounts, Oxford COVID-19 Government Response Tracker project and AMECO.
with the subscripts i and t referring to the countries and quarters respectively, and whereby β 1 , β 3 > 0 while β 2 , β 4 , β 6 < 0 and the sign of β 5 is ambiguous as it covers a whole range of transmission channels. ECT is the error correction term used in the second step of the regression analysis. Table 1 shows that the point estimates all have the expected sign. The Pedroni and Kao panel cointegration test suggests that the null hypothesis of no co-integration can be rejected at a high level of confidence. 15 The real user cost of capital is measured as with IR measured as the bank lending rate, τ the rate of capital depreciation, PC the price of capital, and P the price of output. The expected price change is assumed to be equal to the observed past change. 16 The price/book ratio for the Europe STOXX 600 index is taken as a proxy for the Tobin Q. 17 A dummy equal to 1 from the fi rst quarter of 2020 to the second quarter of 2021, and zero during other periods.
Next, the short-to medium-term dynamics are estimated with pooled generalised least squares, 18 using instrumental variables, 19 i.e.
(  were performed, indicating that the qualitative nature of these results is broadly unchanged if (i) a stricter version of the Oxford indicator that focuses only on mobility restrictions is considered (V1-lockdown), (ii) investment in dwellings is excluded (V1-dwellings), (iii) the error correction term is estimated excluding pandemic related variables (V1-technical), 20 (iv) replacing the change in the lockdown measures by its level did not change the signifi cance of the point estimates, (v) estimation period is limited to the pre-pandemic period (V1-pre 2020), (vi) not enough degrees of freedom are available to obtain stable estimates for some important COVID-19 related factors such as the vaccination rate that took off in the fi rst quarter of 2021.

Lockdown measures
The econometric results suggest that quarterly growth in GFCF decreases with the tightening of lockdown measures. This statistically signifi cant fi nding suggests that a ten-point tightening in the Oxford stringency index leads on average to a contraction of about 2.5 percentage points in GFCF quarter on quarter growth (V1 in Table 2).
18 Allowing for correlation between the random components across member states. 19 Including lagged explanatory variables and the policy variables excluding its cyclical component estimated via the Hodrick-Prescott fi lter. 20 In all variants, except V2-technical, the error correction term (ECT) for the entire sample is estimated based on an equilibrium equation (1) as reported in Table 1. For V2-technical, the error correction terms are obtained re-estimating equation (1) for a sample ending in the fourth quarter of 2019, and fi tting the error correction term from the fi rst quarter of 2020 to second quarter of 2021 using observed explanatory variables and point estimates of the re-estimated equation 1.

Investment
The sensitivity of GFCF to the lockdown measures (V2 in Table 2) 21 decreases over time. This perhaps refl ects learning from experiences and gradual adaptation, which includes greater digitalisation. Along these lines, earlier research demonstrates that the impact of the second and third wave on turnover in the various countries was substantially diff erent from that of the fi rst wave, as turnover reductions were relatively subdued in the member states that suff ered most in the fi rst wave.
The sensitivity of GFCF also diff ers across member states. It is the strongest in Italy and the weakest in Malta and Finland (V3 and Figure 7). 22 Such cross-country diff erences in responsiveness to the lockdown measures might refl ect differences in economic structure such as the share of tourism and contact-intensive sectors in the economy (Coutinho et al., 2021). Figure 8 confi rms that the responsiveness to the lockdown measures increases with the size of contactintensive sectors (as a share of total gross value added). In turn, these lockdown measures lowered private consumption and exports, thereby putting additional downward pressure on GDP and consequently also on investment.

The policy response
The pandemic dummy is found statistically signifi cant (see V1 in Table 2). As such, the dummy captures the role of various factors including the response of monetary and fi scal policy during the COVID-19 crisis. To disentangle GFCF's support of the policy response, the base model 21 V2 allows the point estimate of the lockdown measures to vary across the six quarters during which the pandemic was hitting the euro area. 22 V3 allows for the point estimate of the lockdown measures to vary across the 15 euro area member states in the sample.
(V1) is augmented with a proxy for the monetary and fi scal policy interventions (see V4 in Table 2) while keeping the parameter of the confi nement measures constant over time and across member states and keeping a dummy to capture all other COVID-19 related factors.
The change in the ECB balance sheet (as measured by the change in total liabilities during the pandemic) is used as a proxy for the monetary policy related intervention. As for the fi scal policy response, it is measured by general government net lending (as a share of GDP).
The signifi cant positive point estimate for monetary policy suggests that it supported investment through the normalisation of fi nancial market conditions and the provision of credit to the banking sector at favourable rates that helped banks to grant loans to solvable fi rms. 23 Interestingly, both the monetary policy and fi nancing conditions positively aff ect GFCF. As the latter refl ects mostly market risk premia, the eff ect of the ECB policy measures are already somehow captured by the fi nancing condition 23 Caveat: keeping the coeffi cients fi xed over time and per country may imply that the lower sensitivity of households and fi rms to lockdown measures during the second phase of the COVID-19 crisis is not captured. As a result, there is a risk of overestimating the impact of the policies.

Figure 7
Responsiveness to a change in lockdown measures across the euro area Note: Based on V4 in Table 2. Point estimate signifi cance *** p<0.001, ** p<0.05 and * p<0.
Source: Authors' estimates. The signifi cant negative point estimate for the public budget balance suggests that the increase in headline defi cit supported investment by countering the downward impact of the pandemic shock on aggregate demand (Bellia et al., 2021). 24 Figure 9 provides an overview of the contribution of the various drivers of GFCF during the COVID-19 crisis.

Long-term impacts of COVID-19
Upside risks The pandemic accelerated investment in ICT infrastructure 25 to accommodate the rise in online work and digital sales. The McKinsey Global Institute Report (2021) expects such changes will have the potential to increase 24 On the combined eff ect of monetary and fi scal policy following the outbreak of the pandemic. 25 Bellmann et al. (2021) report that almost 30% of the surveyed German companies reported that the pandemic accelerated the introduction of digital technologies.

Figure 9
Decomposition of the changes in gross fi xed capital formation during COVID-19 Notes: Model estimation based on variant V4 in Table 2   annual productivity growth by about one percentage point up to 2024. Also notable, investment in intellectual property products (e.g. investment in software and research and development) held up better than investment in machinery and equipment. This might be because the exchange of intellectual property products involves less physical interaction.
The pandemic also disrupted the functioning of global value chains (GVCs). The fear of a repeat of a pandemic may then strengthen the incentives to bring production closer to home, 26 thus requiring additional investment. But the available evidence on the impact of COVID-19 on GVCs is somewhat ambiguous. 27 At the same time, such reshoring may limit countries' opportunities to exploit their comparative advantages thereby lowering the return on capital and incentives to invest.

Downside risks
Available evidence suggests that much of the initially feared long-run COVID-19 crisis damage has been avoided thanks to the bold policy response at the national and EU level. However, there remain some risks that might dampen investment going forward, especially in case of a re-intensifi cation of the pandemic (ECB, 2020; IMF, 2021).
If the emergency policy support measures for fi rms are lifted too abruptly, it might contribute to an increase in corporate distress. This in turn may intensify the fi nancing constraints on investment. For example, OECD (2021) expects insolvencies to increase signifi cantly in the next two years, particularly in high-contact services sectors, admittedly from artifi cially low levels.
At the same time, the continuation of support policies could carry the risk of locking capital and labour in unproductive sectors, hindering business dynamism over the medium to long term (Claeys et al., 2021;Ebeke et al., 2021). Nevertheless, preliminary evidence suggests that this eff ect remains modest (Helmersson et al., 2021;Cros et al., 2021), and that business creation has rebounded since the second quarter of 2021 (Eurostat, 2021).

Conclusion
This paper suggests that lockdown measures to limit the spread of the virus had a strong adverse impact on gross fi xed capital formation across the euro area. The impact varied across countries and over time refl ecting partly cross-country diff erences in economic structure and gradual learning and adaptation by economic agents.
The strong rebound in investment in a context of very strong (and temporarily held back) demand, favourable fi nancing conditions and supportive public investments (European Commission, 2021c) provides reasons for optimism. However, it is still too early to assess the longterm impact of the COVID-19 crisis on GFCF. Available evidence suggests that much of the long-run damage initially feared might have been avoided thanks to the bold policy response at the height of the pandemic and the comprehensive recovery strategy that has ensued.