Value added taxes raise prices exponentially – studies prove

V. European VAT Changes A. Graphical Evidence First, we provide some graphical and nonparametric evidence by examin- ing the unconditional mean of the price index using the full sample of EU VAT changes in the 3 months before and after a VAT reform. In each case, we normalize the series to 100 in the month preceding the reform. Figure 7A plots the unweighted average price of all commodities sepa- rately for VAT increases and decreases. The figure shows that prices in- crease discontinuously in the month following a VAT increase but de- crease less in response to a VAT decrease. Importantly, this asymmetry is not driven by any subset of commodities. Indeed, when we plot disag- gregated versions of figure 7A by two-digit COICOP groups (in figs. E20– E22; figs. E11–E23 are available online), we find that all commodities exhibit this asymmetric pass-through, with the exception of communica- tion (COICOP group number 8) and furnishings, household equipment, and routine household maintenance (COICOP group number 5).19 B. Fixed-Effects Estimates To estimate the pass-through to prices of VAT increases and decreases, we first follow the approach of Evans, Ringel, and Stech (1999), who estimate the pass-through of cigarette taxes using different tax changes across US states over time. We run the following fixed-effects regression: D logðpict Þ 5 b0D logð1 1 tict Þ k510 (2) 1 o bkDlogð11ticðt1kÞÞ1Dlt 1gDXct 1Deict, k5210,k≠0 19 Possibly because of sample size, trends do not appear to be parallel for two COICOP categories: (1) alcoholic beverages, tobacco, and narcotics and (2) transport. While pass- through is asymmetric in both of these cases, the violation of the parallel assumptions sug- gest that we should be cautious in interpreting these two figures. 4456 journal of political economy FIG. 7.—Asymmetric response of prices to VAT changes. A, Response of the uncondi- tional means of prices to VAT increases and decreases. Lower panels plot the coefficients from the fixed-effects regression (2) for VAT increases (B) and VAT decreases (C ) on the full sample of reforms and include 10 month leads and lags. where i denotes the commodity, c the country, and t the month in which the price is observed, lt time fixed effects, pict the price, tict the VAT rate, and εict the error term. We control for a given country’s nominal interest rate, GDP per capita, and unemployment rate with Xct. For each of xt ∈ flogðpictÞ,logð11tictÞ,lt,Xct,eictg,Dxt isequaltoxt 2xt21. In equation (2), b0 identifies the pass-through of a VAT change in the month when the change occurs: for example, if b0 5 0, then the price does not respond to a VAT change, and if b0 5 1, the price responds one-to-one to a VAT change. The second term of the equation estimates any forward- or backward-looking responses of prices to changes in VAT rates; b25, for example, identifies the response of prices at time t to VAT changes that will occur at time t 1 5. The fixed-effects regression generalizes an event-study design with mul- tiple periods, commodities, and countries, and its main identification as- sumption is the same as that for difference-in-differences regressions: ab- sent the tax change, there would have been no change in the prices of the treated relative to the untreated commodities. Figure 7A shows a sharp change in prices at the time of the reform, with no pretrends and no evi- dence of anticipatory behavior, which lends support to this identification [chart omitted] what goes up may not come down 4457 assumption. Identification relies on within-country-specific commodity variation in VAT rates over time. In order to estimate the pass-through rate of VAT increases and de- creases, we estimate equation (2) separately on two samples. The first sample includes all observations treated by a VAT increase along with all untreated observations and is used to estimate the pass-through rate of VAT increases. Similarly, the second sample includes all observations treated by a VAT decrease along with all untreated observations and is used to estimate the pass-through rate of VAT decreases.20 Therefore, commodity-country pairs that never undergo a VAT change are included in both samples. The results of these fixed-effects regressions are reported in table 1.21 Columns (1) and (2) of table 1 correspond to VAT increases and decreases, respectively. The first row of each regression (labeled b0) shows the pass- through of the VAT change to prices 1 month after the reform. In partic- ular,b1i showstheresponseofpricestoVATchangesimonthsafterthere- form, while b2i shows the response of prices i months before the reform. Figures 7B and 7C plot the coefficients from regressions for VAT increases and decreases, respectively, and show that the pass-through to prices of VAT increases is equal to 34% while that of VAT decreases is equal to 6% one month after the reform; both are statistically significant. Importantly, there are no significant price responses in any months within a 10 month window around the VAT increases and decreases. We perform several ro- bustness checks, including running specification (2) without controls but with time fixed effects (table E8; tables E5–E23 are available online) separately on reforms that are classified as temporary and permanent (ta- bles E9 and E10), in periods of above- and below-median growth (ta- bles E11 and E12), on reforms that are not concurrent with other tax changes (table E13), with country-commodity specific inflation controls (table E14), and including more leads and lags (figs. E23A and E23B). We find similar levels of asymmetry in each case. In addition, we examine the relation- ship between the timing of VAT changes with underlying economic con- ditions and find that the two are not correlated. These correlations are discussed in detail in appendix B. 20 Untreated observations refer to commodity-country pairs that never undergo a VAT change as well as commodity-country pairs that are treated by a VAT change at a different time than the treatment period. 21 In a recent working paper, Abraham and Sun (2018) show that event study specifica- tions can be problematic under certain conditions. In our case, these concerns would ap- ply if there were heterogeneous effects across groups of VAT changes occurring at the same time. Given that our evidence is robust across a wide range of specifications as well as in several subsamples and also when plotting unconditional means in fig. 7A, this concern is unlikely to apply to our setting. 4458 journal of political economy TABLE 1 Pass-Through Estimates Using Fixed-Effects Regression (Full Sample) D log Price Increase Decrease (1) (2) .34 .064 (.063) (.030) .020 .025 (.019) (.017) .030 .026 (.019) (.015) .020 2.044 (.027) (.021) .015 2.005 (.016) (.028) 2.043 2.004 (.017) (.022) .049 2.020 (.034) (.022) 2.011 2.008 (.027) (.020) Yes Yes .014 .014 386,557 342,792 b0 b11 b22 b12 b23 b13 b24 b14 Time fixed effects R2 Observations Note.—The coefficients reported in this table show the pass-through of VAT increases and decreases to prices, estimated using specification (2) on the full sample of re- forms. Column 1 shows the estimates for VAT increases, and col. 2 those for VAT decreases. Standard errors are clustered by month and are in parentheses. b0 measures the pass-through of the VAT change at the time of the re- form, and bi measures price changes i months away from the reform. C. Matching Estimates In this section, we supplement our fixed-effects estimation with a match- ing approach that allows us to better mirror the setting of the Finnish ex- periment and to more directly interface with the theoretical predictions derived in section II. This approach also allows us to evaluate and alleviate concerns that in the preceding analysis we are never able to observe the effect of VAT increases and decreases at the same time, for the same com- modity, and in the same country. Our matching procedure follows Imbens (2015), and in particular that paper’s approach to analyzing the Imbens, Rubin, and Sacerdote (2001) lottery data (in sec. 6.1). First, and motivated by our theoretical framework from section II, we choose a set of key characteristics used to estimate pro- pensity scores. These characteristics include time, measures of economic activity, the magnitude of the VAT change, and the type of commodity what goes up may not come down 4459 TABLE 2 Pass-Through Estimates: Matched Sample With Trimming D log Price Increase Decrease (1) (2) .29 .042 (.10) (.047) Without Trimming D log Price b0 b11 b22 b12 b23 b13 b24 b14 Time fixed effects R2 Observations .021 (.019) .031 (.019) .021 (.027) .016 (.016) 2.042 (.018) .051 (.034) 2.0092 (.027) .025 (.017) .026 (.015) 2.044 (.021) 2.0050 (.028) 2.0047 (.022) 2.020 (.022) 2.0089 (.020) Increase (3) .33 (.066) .020 (.019) .030 (.019) .020 (.027) .015 (.016) 2.042 (.018) .049 (.034) 2.0096 (.027) Decrease (4) .056 (.037) .025 (.017) .026 (.015) 2.044 (.021) 2.0050 (.028) 2.0046 (.022) 2.020 (.022) 2.0089 (.020) Yes 384,559 342,453 386,080 342,477 Yes Yes .013 .014 .014 .014 Yes Note.—This table reports the pass-through estimates of VAT increases and decreases to prices estimated using specification (2) on matched reforms. Columns 1 and 2 implement the matching estimator on a trimmed sample using the approach outlined in Imbens (2015). Propensity scores are estimated using logit. Columns 3 and 4 report the estimates for the matched sample without trimming. VAT increases and decreases are matched on time, type of commodity (COICOP digit 2), size of VAT change, and GDP growth. In addi- tion, we match the preincrease VAT rate to the postdecrease VAT rate and the postincrease VAT rate to the predecrease VAT rate. Columns 1 and 3 show the estimates for VAT increases, and cols. 2 and 4 those for VAT decreases. Standard errors are clustered by month and are in parentheses. b0 measures the pass-through of the VAT change at the time of the reform, and bi measures price changes i months away from the reform. (two-digit COICOP code). Most importantly, guided by results from sec- tion II, we also match preincrease VAT rates to postdecrease VAT rates and predecrease VAT rates to postincrease VAT rates. Second, we use the es- timated propensity scores, based on these key characteristics, to trim the sample as suggested by Crump et al. (2009). Finally, in this trimmed sam- ple, we match the VAT increases and decreases, using the matching esti- mator outlined in section 5.4 of Imbens (2015).22 We then estimate equa- tion (2) on the matched sample of VAT changes. Estimates for pass-through from this matched sample, summarized in table 2, show that our finding of asymmetry is robust to matching. Indeed, 22 We cluster our standard errors as we do in our main specification. Bootstrapping yields similar standard errors, with all VAT increase pass-through estimates significant at least at the 1% level. However, we do not report these here, following cautions in Abadie and Imbens (2008) against using bootstrap with matching estimators. 4460 journal of political economy these estimates reveal asymmetries that are similar to those estimated us- ing the full (unmatched) sample of reforms. In appendix A, we also for- mally test the ability of our matching procedure to increase overlap in co- variates. We follow Imbens (2015), who uses normalized differences in order to assess overlap. Our matching procedure substantially improves overlap and reduces all normalized differences. While there is no formal threshold for normalized differences and while our matching procedure does not reduce all normalized differences to zero, they are well in line with those estimated by Imbens (2015) using the Imbens, Rubin, and Sa- cerdote (2001) lottery data. Because our matched sample mirrors the Finnish experiments, our pass-through estimates can be directly compared to the predictions of stan- dard incidence theory discussed in section II. We conclude that our find- ing of asymmetric pass-through in the matched sample is difficult to rec- oncile with static incidence theory. We provide more details on how the matching procedure is implemented in appendix A, and include formal tests of pre- and postmatching overlap in observables and further robustness checks. Note that a limitation of our approach is that we cannot match on prereform price levels because our data do not allow us to condition on price levels in a consistent way across all EU countries and commodities. This limitation implies that we are im- plicitly assuming that, conditional on matching on the key characteristics outlined above, any remaining variation in price levels has no meaningful effect over the asymmetry. For this reason, a stronger test of the asymmetry would also match on price levels. D. Long-Term Persistence of the Asymmetry In subsection IV.A we showed that asymmetry in the pass-through of VATs persisted for several years after the Finnish hairdressing reforms were en- acted. In this section, we provide evidence suggesting that this persistence is not a peculiarity of Finnish hairdressers: we observe it in other sectors and countries as well. In order to analyze longer-run horizons, it is impor- tant to focus attention on relatively large VAT changes and sectors in which prices are relatively stable. Absent these conditions, the effect of VAT changes on prices might be masked by natural variation in prices and in- flation that may be quantitatively significant in the long run. In order to study this we start by considering two case studies that in- volve large VAT changes. First, the European Commission approved an ap- plication to reclassify sit-down restaurants from the standard VAT rate to the reduced VAT rate.23 Both France and Finland took advantage of this 23 Following a campaign promise by then French president Jacques Chirac, France ap- plied for an authorization to reclassify sit-down restaurants from the standard to the re- duced VAT rate in 2002. The application was approved for all member states in January 2009. what goes up may not come down 4461 new law, leading to a 14 percentage point VAT cut for French sit-down res- taurants and a 9 percentage point cut for Finnish ones, in 2009. While the VAT rate did not revert to its original level as in the Finnish hairdressing experiment, we can exploit smaller increases in the reduced VAT rate: 1.5 and 3 percentage point increases in France, in 2012 and 2014, and a 1 percentage point increase in Finland, in 2013. Figures 8A and 8B show that the asymmetric pass-through persisted over several years both in Fin- land and in France. This evidence is to be interpreted cautiously, in part because the VAT changes are not of the same size and it is conceivable that the asymmetric pass-through is in part due to that. Second, and more comprehensively, Hungary cut its standard VAT rate from 25% to 20% in January 2006 and increased it from 20% to 25% in July 2009. These changes were enacted as part of a set of campaign prom- ises preceding the 2006 parliamentary elections. Figure 8C shows the re- sponse of commodities that were subject to the standard rate in Hungary FIG. 8.—Long-term persistence of the asymmetry. A, Response of Finnish sit-down res- taurants to a 9 percentage point VAT decrease and a 1 percentage point VAT increase com- pared to a control group of Norwegian sit-down restaurants. B, Response of French sit- down restaurants to a 14 percentage point VAT decrease and 1.5 and 3 percentage point VAT increases relative to a control group of Italian restaurants. We also include a hypothet- ical counterfactual that uses the VAT increase pass-through for VAT decreases. C, Response of all commodities subject to the standard VAT rate in Hungary (excluding diesel and gas- oline) to a 5 percentage point VAT decrease and a 5 percentage point VAT increase relative to a control group consisting of neighboring countries. D, Cumulative pass-through rates of all VAT increases and decreases in our sample over 20 months. [chart omitted] 4462 journal of political economy compared to a set of control countries.24 We find not only strong evidence of asymmetry but also that the asymmetry persisted over several years after the VAT rate was returned to 25%. Because the standard VAT rate applies to a wide range of commodities, this finding suggests that the long-term persistence of asymmetry may arise in more than a small set of specific sectors. Finally, we can assess the long-term persistence of the asymmetry in the full sample of reforms. To do so, we first reestimate equation (2) with 20 leads and 20 lags (we plot the resulting coefficients in figs. E23A and E23B). We find no substantial lagged responses, even 20 months after the changes take place. Second, we add up all of the lag coefficients and plot the cumulative pass-through rate for VAT increases and decreases over a 20 month period following the VAT changes in figure 8D. While both cu- mulative pass-through rates appear to decrease over time, the magnitude of the asymmetry—which is given by the difference between the two cu- mulative pass-through rates—remains relatively constant and shows no signs of convergence. However, in spite of the two time series being rela- tively stable, the precision of the cumulative pass-through estimates mechan- ically decreases over time, making any long-run statements past 18 months only suggestive.25 VI. Conclusion In this paper, we show that prices respond asymmetrically to VAT changes. First, prices respond more to VAT increases than to VAT decreases. Sec- ond, this asymmetric response results in an asymmetric pass-through of VAT changes to profits and markups. Third, the asymmetry persists over the long run. These patterns are inconsistent with the predictions of standard, static incidence models, and our findings therefore call into question the em- pirical relevance of these models for an important and widely employed tax. It is important for future research to examine how general these find- ings are and, in particular, to what degree the types of asymmetries we doc- ument arise with other taxes. More broadly, our findings suggest that it may be important to develop and empirically assess alternative models of incidence that can account for these types of asymmetries. One poten- tially promising avenue is to move beyond the static setting of classical 24 We included every commodity subject to the standard VAT rate, with the exception of diesel and gasoline because of strong volatility. Details of the list of commodities and con- trol group countries can be found in app. D (apps. C–E are available online). 25 We cannot reject that the cumulative VAT increase and decrease pass-through rates are equal after 18 months. We report the level difference between the two cumulative pass-through rates and the corresponding significance levels in table E15. what goes up may not come down 4463 models, and evaluate to what degree dynamic models of tax incidence can generate asymmetries like the ones we document.