A Comparative Evaluation of Fiscal Stabilization Strategies during the Covid-19 Pandemic with Germany as a Reference Point

2.1 A Temporary VAT Cut as Unconventional Fiscal Policy Measure

The temporary reduction of the VAT rate is an unconventional form of fiscal policy. D’Acunto, Hoang, and Weber (2018, 519) define unconventional fiscal policy as “… policies that generate an increasing path of consumption taxes that result in households’ higher inflation expectations and negative real interest rates.” Thus, unconventional fiscal policy relies on households as agents of transmission for fiscal policy (D’Acunto, Hoang, and Weber 2022). Private spending is incentivized through the channel of lowering real interest rates, either by a pre-announced permanent increase of consumption taxes, or – as in the case of the German VAT cut – by temporarily reducing the tax rate for the duration of a pre-specified period. Assuming that monetary policy is kept constant (or constrained by the zero lower bound), a change in inflation expectations should translate directly to a change in the perceived real interest rate. The reduction of real interest rates induces an intertemporal substitution effect via the consumption Euler equation, leading consumers to save less in the current period and to bring forward planned future expenditures to benefit from the relatively lower prices induced by the VAT cut.

However, temporary VAT reductions have rarely been used in the past. Therefore, little is known about the usefulness and relative effectiveness of these measures compared to conventional fiscal policy. In fact, there is only one known case where a temporary VAT cut was adopted to stimulate demand. In December 2008, the UK reduced their regular VAT rate from 17.5 % to 15 % for a duration of 13 months (see, e.g. Barrell and Weale 2009; Blundell 2009; Chirakijja et al. 2010; Crossley, Low, and Wakefield 2009; Crossley, Low, and Sleeman 2014). Blundell (2009) and Crossley, Low, and Wakefield (2009) document a relatively small income effect of this temporary consumption tax reduction in the UK (as compared to conventional fiscal policy).^[4]

Moreover, we believe that the dynamic nature of the pandemic and the overall high level of uncertainty regarding potential demand and supply-side restrictions sets the most recent global economic crisis apart from previous recessions. This is in line with Guerrieri et al. (2022) who find that global health shocks, which lead to restrictions on public life and asymmetrically affect economic activity in specific sectors, result in lower fiscal multipliers compared to supply shock-induced recessions in which sectors are affected equally. Therefore, one must be cautious to apply the existing evidence on government spending multipliers during recessions (including studies of the UK VAT cut) to the Covid-19 crisis.

2.2 Determinants of the Efficacy of a Temporary VAT Cut

According to Ramey (2021), successful fiscal stabilization policy should be salient, comprehensible, and actionable.^[5] Following this line of thought, we discuss some of the factors that might affect the efficacy of a temporary VAT cut as an unconventional fiscal policy measure in general and specifically during the Covid-19 pandemic.

3.1 Step 1: Pre-selection Based on Post-treatment Characteristics

To interpret the treatment effect in a reasonable way, it is crucial to illustrate the assumptions we make about the counterfactual German economy. In our setup, the weighted control group cannot simply represent Germany without the VAT cut as most of the potential donor pool countries also implemented large-scale fiscal measures to counter the consequences of the Covid-19 pandemic. To tackle these challenges, the matching procedure is applied on a carefully pre-selected pool of donor countries. Accordingly, the ideal synthetic control group should feature the same expenditures for the fiscal package (as % of GDP) as Germany and should only differ in a way that it did not implement the temporary VAT rate reduction. Put differently, we construct a counterfactual Germany based on the assumption that if Germany had not chosen the VAT cut as a specific policy to boost consumption, it would have been replaced by a different fiscal measure with the same expenditures (as % of GDP).^[9]

Our starting point is a pool consisting of OECD countries and additional advanced East Asian non-OECD economies. In a first step, we select only those countries with a comparable fiscal package size during the treatment period. Furthermore, we remove those countries that also used wide-ranging temporary VAT rate reductions (e.g. Cyprus or Ireland) from the pool of potential donors. These countries are not suited for an analysis of the relative efficacy of the German VAT cut in comparison to countries with other fiscal measures of similar size. This exclusion yields a donor pool of 24 countries and is used for the benchmark estimations (specification Full in Table 1).^[10]

To reduce the level of heterogeneity in the donor pool with respect to important post-treatment characteristics, we narrow it down along two different dimensions. Following our first exclusion criterion, we keep only those countries with a similar fiscal package size in both quarters of the treatment period (specification Excl. 1a in Table 1).^[11] With our second exclusion criterion, we remove those countries that used comparable, but not full-scale VAT cut policies for non-medical products during the treatment period (specification Excl. 1b in Table 1). Lastly, we combine those two dimensions and leave only those countries in the donor pool with a similar fiscal package size during the treatment period, but no (partial) VAT rate reduction (specification Excl. 2 in Table 1).

The variation of the donor pool composition allows us to analyze the relative effectiveness of the different fiscal responses in the countries that make up the counterfactual twin economy. This potential to gather evidence on the cost-effectiveness of the different fiscal measures is further exploited by various validation exercises outlined in Section 3.3. Finally, we compare Germany and the donor pool countries with respect to their lockdown stringency, Covid anxiety, and consumer confidence since post-treatment differences in these variables across countries might confound the interpretation of the treatment effect.

3.2 Step 2: Matching on Pre-treatment Characteristics

Relying on the four different specifications, we apply a reweighting scheme on the pre-selected donor countries so that systematic pre-treatment differences between Germany and the control units are removed. We achieve this by employing a matching procedure where we construct a single artificial twin economy (“synthetic control”) from the donor pool that closely resembles the German economy just before the onset of the pandemic (2019q4) with respect to the covariates we suspect to influence the likelihood of being selected into treatment and that might, at the same time, influence the outcome variable of interest (see also the discussion in, e.g. Gutmann, Neuenkirch, and Neumeier 2021; Neuenkirch and Neumeier 2016). These variables are GDP per capita, the shares of private consumption and government spending relative to GDP, as well as the growth rates of real GDP and private consumption to the same quarter in the previous year and three years ago.^[12] The synthetic control is obtained by reweighting the group of donor pool countries so that covariate balance between the treated and untreated units is achieved.^[13] Put differently, we remove systematic differences between the treated and the control units and ensure that these are on virtually the same trajectory before the imposition of treatment, in particular due to the usage of the three-year growth rates.

By matching on these covariates, we obtain a synthetic control that is highly similar to Germany regarding its pre-treatment characteristics, allowing us to attribute the differences in the observations of our variable of interest between the two groups to the effect of the treatment. Hence, we mimic a random selection into treatment and control for the influence of these confounders on our estimated ATT:

The vector x contains the matching covariates. The weights for the construction of the synthetic control are obtained via entropy balancing, a pre-processing method designed by Hainmueller (2012) that employs a reweighting scheme to achieve covariate balance in observational studies with binary treatments. As we match on the last observation before the outbreak of the Covid-19 pandemic, we construct a synthetic twin economy that closely resembles Germany in terms of its macroeconomic situation just before the health shock hit all (developed) countries worldwide. Put differently, the weights are chosen to create a doppelganger that is as similar as possible to Germany. At the same time, the weights should be as close as possible to uniform base weights.

3.3 Extensions and Robustness Tests

Apart from estimating the ATT of Germany’s VAT rate reduction, we conduct several alternative analyses to evaluate the robustness of our results and to increase the credibility of our causal interpretation. These validation exercises also help us to gain insights about and to foster the discussion on the relative effectiveness of the different emergency fiscal policy measures.

First, we use placebo tests to validate our findings. In most settings, no effect should be found in the absence of treatment (Eggers, Tuñon, and Dafoe 2021). This means that we should not detect a significant difference between Germany and its doppelganger before the treatment was in place. Such a test is particularly important in our setting, since we match on the last observation before the onset of the Covid-19 pandemic (2019q4), but our treatment period is 2020q3–2020q4. Hence, we carry out placebo tests for 2020q1 and 2020q2 to check whether Germany’s consumer spending was affected differently by the pandemic shock than that of the synthetic control.^[14] The same tests are carried out for 2021q1 and 2021q2. However, as a temporary tax cut works through an income effect and an intertemporal substitution effect, we expect a part of the consumer spending that would normally happen in 2021 to be brought forward to the period with the tax cut in place.

Second, as suggested by Abadie (2021) and implemented in Abadie, Diamond, and Hainmueller (2015), Barber and West (2022), and Cho (2020), we conduct a “leave-one-out” analysis, where one country at a time is excluded from the donor pool and the analysis is repeated, all else being equal. Should the exclusion of a certain country from the donor pool lead to substantial changes in the direction or magnitude of the estimated treatment effect (while the alternative twin economy remains a good match for Germany), we investigate the economic situation and the policy response of the excluded country during the treatment period more closely. This exercise is used to gather evidence on the cost-effectiveness of the different fiscal measures and to provide policy advice on which measure(s) provided the greatest “bang for the buck.”

4.1 Descriptive Statistics

The upper panel of Table 2 shows the descriptive statistics for Germany and an equally weighted average of the 24 countries in the full donor pool prior to the pandemic (in 2019q4). There are no statistically significant differences for the variables GDP per capita, the consumption to GDP ratio, and the one-year growth rates of real GDP and consumption. However, Germany appears to be on a different growth and consumption trajectory compared to the average donor pool country, while also having a different level of government spending.

The lower panel of Table 2 illustrates how closely the synthetic control resembles Germany’s economy when relying on the broadest donor pool of 24 countries as there are no longer significant differences in the means of the two groups. Despite the small sample size and our choice to match on seven covariates, the weights obtained by the entropy balancing procedure allow us to construct synthetic controls that are successful in replicating Germany’s macroeconomic situation and trajectory before the pandemic. This also holds for the three other donor pools, with only a minor difference in the three-year growth rate of real GDP remaining in the case of the pre-selection based on the fiscal package size (Excl. 1a) and the strict pre-selection (Excl. 2).^[15] However, this should not be a concern as Germany is on virtually the same consumption trajectory when compared to its doppelganger (see also Figure 1 below).

Figure 1:

Consumption growth rates: Germany versus synthetic control group. Figure shows the trajectories of the year-over-year consumption growth rates for Germany (observed values, blue line) and the synthetic control group (weighted average, red line) based on the four donor pools. “Excl. 1a: Package Size” removes countries whose fiscal package size was outside the range of ±5.5 pp in one quarter of the intervention period (2020q3–2020q4). “Excl. 1b: Similar Policy” removes countries that used comparable, but not full-scale VAT cut policies for non-medical products during the treatment period. “Excl. 2: Strict” combines the two restrictions. The grey line indicates 2019q4, that is, the observation used in the entropy balancing algorithm. The blue bar highlights the treatment period (2020q3–2020q4).

Table 3 shows the composition of Germany’s doppelganger for all four donor pools. The five countries with the largest weights are highlighted in bold. Without any further pre-selection (column Full), Denmark (27.5 %), the Netherlands (14.2 %), Greece (12.1 %), Canada (9.3 %), and Japan (7.9 %) are most fitting to replicate the macroeconomic situation of Germany in 2019q4. Canada remains among the most important donors for all specifications. When restricting the donor pool, other countries like Australia, Austria, or France are stepping in and receive larger weights.

This changing composition of the donor pool is of particular interest for our analysis of the relative efficacy of the German VAT rate cut. While all countries tried to dampen the adverse effects of Covid-19 on the economy, their responses were different regarding the measures used and the size of the stimulus package. For instance, Denmark, Canada, and Japan focused on direct transfers and payments to households to stimulate the economy, whereas Greece also adopted temporary VAT rate reductions for several sectors of their economy (e.g. tourism).^[16] Hence, the ATTs based on different donor pools (in the next subsection) might reveal interesting insights. This will be complemented by a “leave-one-out” analysis in Section 4.3 where we check if the exclusion of a particular country from the full donor pool (substantially) changes the ATT.

4.2 Baseline Results

Figure 1 shows, based on the different donor pools, the trajectories of Germany (blue line) and its twin economy (red line) for the period 2015q1–2021q2. With each specification of the donor pool, Germany and the weighted control group were virtually identical in their consumption growth rate in the last observation before the outbreak of the pandemic (indicated by the grey vertical lines) and also on the same consumption growth trajectory before 2019q4. This corroborates the efficacy of the entropy balancing algorithm in harmonizing the pre-treatment macroeconomic conditions in Germany and its doppelganger.

Table 4 shows the ATTs for the period 2020q1–2021q2 for the four different donor pools alongside standard errors in parentheses and p-values in brackets. In the case of the full donor pool and the group that excludes other countries that adopted partial VAT rate reductions during the treatment period (Excl. 1b), the placebo tests for the first two quarters of 2020 are not significant, indicating that consumer spending initially reacted very similarly to the Covid-19 shock in both groups. The estimated effect of the actual intervention is significantly negative in 2020q4 if no countries (−1.77pp, column Full) or only those countries with a similar measure (−2.22pp, Excl. 1b) are excluded from the donor pool, indicating that a VAT rate cut (and not only the large-scale one implemented in Germany) was relatively ineffective compared to the policy response of the twin economy.

The relative inefficiency of the German VAT rate cut becomes even more striking when considering the actual sizes of the fiscal packages (as % of GDP) in Germany and in the four synthetic control groups (see Table A1 in the Appendix). Indeed, the size of the fiscal package in Germany is numerically larger in 2020q3 and significantly larger in 2020q4 when compared to the weighted averages of the full donor pool and the specification based on Excl. 1b. Hence, we find negative treatment effects for these two specifications in 2020q4 despite smaller expenditures in the synthetic control groups.

In the group where only those countries with a similar fiscal stimulus size are included (Excl. 1a), both placebo tests are significant at the 5 % level with a positive ATT and in the group based on the strict pre-selection (Excl. 2), we observe a significant positive ATT for 2020q2. This implies that Germany – when compared to its doppelganger based on these two donor pools – was initially hit less severely by the Covid shock, as indicated by a smaller decline in consumption. For both specifications, we find no statistically significant estimates of the ATT in the treatment period.

The results for the placebo tests in 2021q1 and 2021q2 appear most striking. Irrespective of the specification, these indicate that Germany’s consumption did not recover nearly as well as that of its doppelganger in the first half of 2021 as we observe a relative difference in the consumption growth rate (to the previous year’s quarter) between −4.96pp and −7.28pp. There are several possible explanations for this result. As previously illustrated in Sections 2 and 3, negative estimates for the placebo tests in the post-treatment period are not too surprising, as we expect that Germany’s consumers brought forward some purchases to benefit from the VAT rate reduction. This implies, however, that consumption in Germany would have decreased to even lower levels in the second half of 2020 without the VAT cut policy. A part of this large difference might also be explained by a level effect and the positive placebo tests in the first half of 2020 as some of the other countries had more catching-up potential in the post-treatment period. Nevertheless, this only holds to some extent for specifications Excl. 1a and Excl. 2 and the magnitude of the difference between Germany’s consumption growth rate and that of its twin economy still appears noticeable.

4.3 Extensions and Robustness Tests

4.3.2 Leave-One-Out Analysis

Figure 2 shows the estimated ATTs of Germany’s VAT rate reduction for 2020q3 (left panel) and 2020q4 (right panel) alongside 95 % confidence intervals. The top ATTs are from the previous subsection, followed by the ATTs from the “leave-one-out” analysis (where the label represents the excluded country), and an ATT where Canada, Denmark, Japan, Sweden, and the United States are jointly excluded from the full donor pool.^[21] The ATTs in blue (red) are (not) significant at the 5 % level and the dashed lines indicate the point estimates from the analysis without further pre-selection (specification Full) as a reference point.

Figure 2:

Average treatment effects: robustness tests. Figure shows the estimated ATTs of Germany’s VAT rate reduction for 2020q3 (left panel) and 2020q4 (right panel) alongside 95 % confidence intervals. The names below the ATTs refer to the countries omitted in the leave-one-out analysis. ATTs in blue (red) are (not) significant at the 5 % level. The dashed lines indicate the ATTs from the analysis without further pre-selection (specification full) as a reference point.

In 2020q3, the ATT is quite robust to changes in the composition of the donor pool. For all but two specifications, the ATT is close to zero and insignificant, although the effect is larger (and even slightly positive) when comparing Germany only to those countries with a similar fiscal package size or those remaining after the strict pre-selection. The only noticeable difference can be found when Denmark is excluded from the benchmark donor pool. In this case, the estimated ATT for Germany turns positive, indicating that the negative effect of the health shock on the Danish economy was less pronounced.

For 2020q4, we obtain significantly negative ATTs in most of the specifications. Noticeable exceptions are – as already documented in Table 4 – the ATTs according to donor pools with a similar fiscal package size (Excl. 1a) and based on strict exclusion (Excl. 2). Of particular interest are the ATTs when excluding Canada, Denmark, Japan, Sweden, or the United States as these are no longer significantly negative at the 5 % level. Hence, these five countries (in particular, Denmark and Japan) appear to drive the overall negative effect of Germany’s ATT in 2020q4.

As a consequence, we remove these five countries jointly from the full donor pool and reestimate the ATT. When relying on such a restriction, the relative effect of Germany’s temporary VAT cut turns significantly positive in 2020q3 (1.86pp, p-value < 0.01) and 2020q4 (1.04pp, p-value = 0.04). Put differently, the fiscal packages of Canada, Denmark, Japan, and the United States – that mostly focused on direct transfers to stimulate consumption – appear to be particularly effective, which is why we provide a detailed discussion of these and the “special” situation in Sweden in Section 5.

5.1 Denmark

How did Denmark respond to the Covid shock and what did it do differently than Germany? According to Andersen (2021), Denmark’s health outcomes during the pandemic and the economic situation before the pandemic were favorable in comparison to most OECD countries, which might have positively affected the resilience of private consumption.^[22] However, we suspect Denmark’s successful stimulus to be largely owed to the “unfreezing” of holiday pay. To offer a quick relief solution for households and to dampen the economic downturn, Danish authorities allowed workers to request a part of their frozen holiday pay in Autumn 2020 and the rest at the beginning of 2021. According to the Danish Ministry of Employment (2021), a total of 31.3 billion DKK of holiday money after taxes was paid out to Danish workers in 2020 (approximately 1.4 % of GDP). We conjecture that a substantial share of those payments was promptly used for consumption purposes since these transfers were only made on request. Hence, Denmark employed a relatively easy-to-understand (potential) transfer payment that could particularly be helpful for financially constrained consumers. However, only the coincidental change in the Danish holiday pay system made this “clever” measure possible.^[23]

5.2 Canada and the United States

In Canada and the United States, policymakers reacted to the (large) drop in consumption induced by the Covid shock with conventional fiscal policy measures.^[24] Governments offered large financial support to citizens via direct transfer payments that especially targeted low- and middle-income households.^[25] For the US, the aggregate MPC out of the stimulus income was estimated at 0.35 in the month after receiving the transfer (Baker et al. 2020). Similarly, survey results from Coibion, Gorodnichenko, and Weber (2020) indicate an MPC of around 0.4 between April and July 2020. Most strikingly, the MPC of individuals with large balances on their bank accounts was significantly smaller at 0.1 compared to liquidity-constrained individuals with less than 100 USD on their bank accounts, whose MPC was estimated to be above 0.4 (Baker et al. 2020).

5.3 Japan

The Japanese response was special as it was the only case of large unconditional transfer payments to stimulate consumption. Recipients were given a cash transfer of 100,000 YEN, which is equivalent to 42 % of the median monthly income of full-time workers. These cash transfers were distributed between May and August 2020 irrespective of age, income, and employment status of recipients (Kaneda, Kubota, and Tanaka 2021). It is not surprising that this policy resulted in a lower average MPC out of transfer income as compared to the US, since these transfers were completely untargeted and households with high income and high liquidity are not expected to react substantially to a one-time increase in income. In fact, the estimated MPC ranges between 0.1 (Hattori, Komura, and Unayama 2021, during the first month after the reception) and 0.16 (Kaneda, Kubota, and Tanaka 2021, for the period of six weeks after the reception). Similar to the US stimulus cheques, both papers find that liquidity constrained households reacted more to the cash transfers than unconstrained recipients.^[26] Hence, this indicates a lower cost-effectiveness of the Japanese unconditional transfers compared to the US (and Canadian) stimulus cheques or, put differently, larger multipliers for targeted transfers. Yet, the unconditional lump-sum transfer likely caused fewer administrative costs than the targeted ones.

5.4 Sweden

Compared to most advanced economies, Sweden adopted loose lockdown measures and mostly continued “business as usual.” Accordingly, we believe that the better performance of the Swedish economy can be largely attributed to consumers facing fewer limitations to spend. Moreover, Swedish consumers were quite optimistic (with respect to economic and health-related indicators) throughout the pandemic and the overall decline of private consumption was less pronounced compared to Germany and most other countries in the donor pool.^[27]