The pass-through of beer taxes to prices: evidence from state and federal tax changes.

Shrestha, Vinish ; Markowitz, Sara

I. INTRODUCTION

Alcohol excise taxes are a common form of taxation used by the federal government and state governments alike, serving the dual purpose of raising revenue and protecting the public health through reduced consumption. (1) From a policy perspective, it is crucial to understand how alcohol taxes are passed through to prices if higher taxes are to be levied. Standard economic theory predicts that taxes are at most fully passed through in a competitive market where firms are price takers. However, the market for beer in the United States can be best categorized as an oligopoly, and beer stands out as the most prevalent type of alcohol consumed in the country. (2) Theories in imperfect competition suggest various possibilities of tax shifting including over- and under-shifting to prices. In oligopolistic markets, as Fullerton and Metcalf (2002) discuss, an over-shifting of tax can occur because of both the existence of market power and the strategic behavior of firms. Under-shifting is also possible when the burden of higher taxes is shifted to labor in the form of lower wages or to other factors of production. Ultimately, the amount of the tax pass-through is an empirical question, and therefore, the main goal of this study is to provide new estimates of the magnitude of the beer tax pass-through to prices in the United States.

When legislators consider changing beer taxes, precise and recent estimates of the pass-through rate are required to understand how the tax change will affect retail prices. To our knowledge, there are only a few existing studies that empirically estimate the pass-through rate of alcohol taxes to prices in the United States. Using quarterly state-level data from 1982 to 1997, Young and Kwapisz (2002) find that taxes on beer, spirits, and wine are over-shifted to retail prices. Specifically, the authors estimate that a $1 increase in the tax per barrel of beer increases the retail beer price by a range of $1.67-1.88. Kenkel (2005) uses the 2002 tax hike in Alaska to perform before-after analyses and concludes that alcohol taxes are more than fully passed through to prices. Kenkel finds that the average pass-through rate for off-premise consumption is 2.14, 2.03, and 1.57 for Budweiser, Bud Light, and Heineken, respectively. Although a novel study in various ways, the lack of a control group and the study's focus just on Alaska raises questions regarding the external validity of the results. Stehr (2007) uses price data from the third quarter editions of the American Chamber of Commerce Research Association (ACCRA) from 1990 to 2004 and finds that the magnitude of the pass-through of beer taxes to prices is 0.94. This point estimate suggests that beer taxes are almost fully passed through to prices and a confidence interval around the estimate includes values greater than one. However, Stehr's study only includes state and year fixed effects in the model as additional control variables and excludes other potential determinants of both alcohol prices and taxes. Also, it is unclear as to why the study only uses the third quarter editions of ACCRA, when tax changes may occur any time during the year. Reflecting on the existing studies in the United States, it must be emphasized that the most recent data used ends in 2004. Since then, the beer industry has experienced two major mergers of leading producers--SABMiller and Coors in 2007 followed by Anheuser-Busch and InBev in 2008. Given this restructuring within the industry, new estimates of the tax pass-through rate are warranted.

In terms of studies outside the United States, Bergman and Hansen (2010) analyze the tax pass-through rate in Denmark using the liquor tax cut in 2003, the beer tax increase in 1997, and the beer tax cut in 2005. The authors find evidence of over-shifting during the incidence of tax hikes and under-shifting during the events of tax cuts. Carbonnier (2013) studies differences between the ad valorem taxes and per unit excise taxes in France and finds that the per unit excise taxes are over-shifted whereas ad valorem taxes are under-shifted to prices.

The objective of this study is to provide new estimates of beer tax pass-through to prices. We focus on beer because of its popularity among alcoholic beverages consumed in the United States. First, to evaluate whether beer taxes are fully, under-, or over-shifted to prices, we use state-level increases in beer taxes in relatively recent years (2000-2014) by implementing a difference-in-differences strategy. It must be noted that six states increased beer taxes after the two major mergers in 2007 and 2008. This allows us to obtain a pass-through estimate following the restructuring of the beer industry. However, given that the industry structure has itself remained oligopolistic for the past few decades, we supplement our state-level analysis by examining the federal beer tax increase in 1991, when taxes on a six-pack of beer nearly doubled from 16 to 30 cents. (This was the most recent federal tax change.) We then compare the results obtained from the federal tax hike with our findings from the state-level tax increases. Due to the lack of a valid control group, the effects of the federal increases in beer taxes have not been widely assessed. We use a regression discontinuity design to estimate the pass-through of the federal tax increase. The state-level analysis has the advantage of using the most recent data available and allows for a difference-in-differences framework. The federal tax change has the advantage of being much larger than the typical state-level change, which aids in the identification of the magnitude of the tax pass-through rate.

The findings from the state-level difference-in-differences analysis that focuses on the years 2000-2014 suggest that a 10-cent tax increase on a six-pack of beer is associated with an increase in prices by 17 cents (in real terms). In this framework, obtaining unbiased estimates requires that there are no state-specific characteristics influencing prices that are correlated with taxes. Our results are robust to controlling for the possible confounders and state-specific linear time trends. Similarly, results focusing on the federal beer tax hike indicate that a 10-cent tax increase per six-pack of beer increases retail prices by 19-22 cents (in real terms). The magnitudes of our findings from both analyses show that beer taxes are more than fully passed-through to prices. We find similar results when using nominal values of price and tax. The results also suggest that the recent mergers in the brewing industry did not significantly change the pass-through rate.

II. THE U.S. BEER INDUSTRY AND TAX PASS-THROUGH

Since the end of Prohibition, the number of independent brewers in the United States has continuously declined and the major beer companies such as Anheuser-Busch InBev and MillerCoors have gained significant market share. Brewing companies in the United States underwent a "beer war" from the 1960s to the mid-1980s, which was accentuated by fierce competition in the marketing and advertisement sectors, including the production of new brands and tough price competition. The number of independent macro-brewers declined from 766 in 1935 to about 20 firms in 2012 (Gokhale and Tremblay 2012). Iwasaki et al. (2008) and Gokhale and Tremblay (2012) both find that the beer industry became more competitive, but still classified as concentrated, as the firms became fiercely competitive within the oligopoly structure from the 1960s to mid-1980s. This situation remained constant throughout the 1990s. We refer the reader to Tremblay and Tremblay (2005) and George (2009) for more details on the history of the beer industry.

The macro-brewing sector of the United States beer industry emerged into an oligopolistic structure by 1970. Although the concentration ratio is not the sole determinant of a specific market structure, the four-firm concentration ratio ([CR.sub.4]) increased from about 44% in 1970 to 92.8% in 2003 (Tremblay et al. 2005; Gokhale and Tremblay 2012). As reported by Tremblay et al. (2005), CR4 surpassed 40% in 1968, which is the cut-off established by Scherer and Ross (1990) and Shepherd (1990) to distinguish a competitive market from an imperfectly competitive one. Greer (1998) characterizes the beer industry in the 1990s as a tight oligopoly.

Recently, the beer industry experienced two major mergers with SABMiller and Coors merging in late 2007 and InBev buying Anheuser-Busch in 2008. In 2008, InBev was the second largest brewing company in the world, and Anheuser-Busch was a leading American brewer holding 48.5% of U.S. beer sales (Anheuser-Busch 2008). At the time of the merger, the annual sales revenue of Anheuser-Busch InBev was estimated to be more than $36 billion a year, surpassing the leading brewer, SABMiller. These mergers further concentrated the industry. Gokhale and Tremblay (2012) report increases in the CR4 and Herfindahl-Hirschman Index (HHI) from 92.03 and 3672, respectively, in 2006 to 93.59 and 4255, respectively, in 2009. (At the time of the writing of this paper, there are talks of Anheuser-Busch InBev taking over SABMiller, the world's second largest brewer. (3))

Figures 1A and 1B show trends in beer prices and taxes at the federal and state levels over time. A spike in beer prices after the merger between Anheuser-Busch and InBev in 2008 is clearly seen in Figure 1B. However, as we explain in detail below, the price shown in this figure is the price of Heineken, a competing brand that was not part of the merger. This price increase was implemented by Heineken at the end of 2008 and although the timing of this price increase occurs just after the merger, the company states that the price increase was part of a strategy to offset recession-induced changes in sales volume and higher input costs. In fact, in 2009, the other major brewers announced price increases as well for the same reasons (Heineken 2010; Rooney and Clifford 2009). It is interesting to note the lack of a noticeable price increase after the 2007 merger between SABMiller and Coors. Ashenfelter et al. (2015) analyze the merger of SABMiller and Coors and find that the offsetting effects of the increased concentration and increased efficiency in production resulted in no change in average price.

The few existing studies focusing on the alcohol tax pass-through in the United States predate these major mergers. Given these major changes in the beer industry, it is desirable to estimate the magnitude of the beer tax pass-through to prices using recent data to see whether the mergers affected the firms' behaviors regarding pass-through rates.

The amount and incidence of a tax pass-through depend on the elasticity of demand, elasticity of supply, and market structure. Taxes are at most fully passed through to prices in a competitive market. The term "fully passed through" means that the ratio of the price change to the tax change equals 1. However, varying degrees of tax shifting are possible in an imperfectly competitive market, including the possibility of over- and under-shifting. Over-shifting in an oligopoly setting can be mainly accredited to the existence of market power and the strategic behavior of firms (Fullerton and Metcalf 2002). Since firms recognize that a tax increase will force a price increase and a reduction in quantity demanded, they may seek to raise prices even more than the amount of the tax increase to compensate for lost revenue. Under a constant elasticity demand function with a demand elasticity less than zero, over-shifting will always occur and the value increases as demand becomes less elastic. Even in the absence of an oligopoly, over-shifting is possible in an industry facing decreasing costs due to scale economies (Besley and Rosen 1999).

[FIGURE 1 OMITTED]

Focusing on the event of a cigarette tax change, Barzel (1976) indicates that an over-shifting of cigarette taxes to prices can be explained by a change in the quality of the product. As cigarette taxes are imposed on a pack, the producers respond by packing longer cigarettes to evade taxes. In this case, prices are influenced by increases in both taxes and costs. However, this concept is infeasible in explaining why beer taxes may be over-shifted to prices because taxes vary with the amount of alcohol to some extent. (4) Moreover, there is no evidence of increased ethanol content in the beer brands used in this study.

As mentioned before, under-shifting of taxes to retail prices is also possible in imperfect competition. Given that the consumer price is unchanged, or is changed less than the amount of the tax increase, prices received by suppliers will fall and the burden of the tax is passed backward to the factors of production, in general--land, labor, and capital (Fullerton and Metcalf 2002). Theoretically, under-shifting occurs if both the cost and demand functions are linear. Bergman and Hansen (2010) provide empirical evidence of under-shifting. The authors focus on Denmark and use episodes of tax cuts for liquor and soft drinks in 2003 and beer in 2005 to show that under-shifting occurs during tax cuts. Ally et al. (2014) find that retailers in the United Kingdom respond to increases in alcohol taxes by under-shifting their cheaper products, whereas taxes are over-shifted for expensive products. Specifically, the authors find that among beers, ciders, spirits, and wine, the level of under-shifting is greatest for beer and wine.

It is worthwhile to mention that the competitive nature of the production and retail sectors may differ within the beer industry. The structure of the retail sector varies across states with some states implementing a state-run retail distribution, some having a license retail system, and some having a combination of the two. The question of whether the U.S. retail sector can be classified as imperfect competition has been overlooked so far and the answer will depend on where and how beer is sold in each state. However, a number of papers in the industrial organization literature claim that retail trade in general is non-competitive. For example, Hall (1988) conducts an industry-specific analysis to examine the status of marginal cost and price in the United States. Hall begins with a notion that a competitive firm will have marginal cost equal to market price. He develops a method for jointly testing the equality of price and marginal cost and constant returns. In a competitive market, the ratio of price to marginal cost should equal 1. Hall finds that the ratio of price to marginal cost is 2.355 for the retail trade sector, suggesting the existence of imperfect competition. Several papers included in a review by Anderson (1990) find a positive and significant relationship between grocery store prices or profits and market concentration. These studies claim that local grocery stores are imperfectly competitive. The caveat to these findings is that positive correlation can be driven by the existence of higher costs in more competitive markets, rather than market power.

Another issue that needs to be discussed is the party upon whom (production, distribution, or retail) the federal or state taxes are levied. The textbook treatment of tax incidence states that that party responsible for remitting government taxes is independent of who bears the tax burden. However, as the ability to pass taxes may vary at different points down the supply chain (across economic agents), the levy might matter significantly. Although such analyses focusing on alcohol have not been conducted, in the context of diesel fuel taxation, Kopczuk et al. (2013) finds that diesel fuel taxes are passed through to retail prices to a greater extent when the point of tax collection is prime suppliers rather than retailers. The federal taxes on beer are levied on manufacturers and there exists variation among parties responsible for remitting state-level beer taxes. (5) Considering all the factors involved in the tax pass-through, we would like to emphasize that estimating the magnitude of the beer tax pass-through to retail price remains an empirical question.

III. ESTIMATION STRATEGIES

A. Difference-in-Differences Estimation--Evidence from State-Level Tax Changes

As a first estimation strategy, we implement a difference-in-differences model by using within state changes in beer taxes to identify the effect of taxes on prices. The model used is a reduced form price equation with taxes and other exogenous determinants of demand, costs, and strategic considerations as right-hand-side variables. The basic specifications are given below:

(1a) [P.sub.tc] = [[alpha].sub.1] + [[alpha].sub.2][Tax.sub.ts] + [[alpha].sub.3][X.sub.ts] + [[lambda].sub.t] + [[tau].sub.c] + [[rho].sub.q] + [[epsilon].sub.tc]

(1b) [P.sub.tc] = [[alpha].sub.1], + [[alpha].sub.2][Tax.sub.ts] + [[alpha].sub.3]Post Merger * [Tax.sub.ts] + [[alpha].sub.4][X.sub.ts] + [[lambda].sub.t] + [[tau].sub.c] + [[rho].sub.q] + [[epsilon].sub.tc]

where [P.sub.tc] represents the beer price at time t in city/county c; [Tax.sub.ts] indicates the state-level beer tax at time t in state s; and Post Merger represents the time from 2009 to 2014, the years after the two major brewery mergers. X is a vector comprised of time-varying state-specific variables affecting the market for beer that includes: (1) the percentage of each state's population living in dry counties and the number of liquor outlets per capita (capturing accessibility and costs of obtaining alcohol); (2) the percentage of each state identifying with certain religions (Latter-day Saints, Catholics, and Southern Baptists) which may affect the demand for beer; (3) the presence of 0.08% blood alcohol content (BAC) laws and an indicator for a Sunday alcohol sales ban (state laws that may discourage alcohol use); (4) the number of brewing jobs and brewing establishments (proxies for the potential strength of the beer industry within the state and ongoing state sentiment toward alcohol); and (5) the state unemployment rate and state real per capita income. The variables [y.sub.t] and [c.sub.t] denote year and city fixed effects, respectively; [q.sub.t] represents quarterly dummies, and etc is the error term.

Including city and year fixed effects in the models can be viewed as an extension of a difference-in-differences framework that allows for multiple treatment and control groups and multiple time periods (Wooldridge 2001). The inclusion of city fixed effects captures time-invariant heterogeneity across cities including measures of the marginal cost of production such as the distance to the closest macro brewery. The unbiased estimates of higher beer taxes in the setting given above require that there exist no contemporaneous state-level trends that are correlated with increases in beer prices and taxes. As an alternate specification, we include state-specific linear time trends in the model to capture the unobserved time-varying linear changes that may affect both prices and taxes. However, caution should be taken as state-specific linear time trends may absorb much of the variation in the data, leading to insufficient identification. In additional regressions, we omit the year fixed effects and include the HHI to reflect changes in the industry concentration over time. Regressions are estimated with OLS and the standard errors are clustered at the state level.

B. Regression Discontinuity Design--How Did the Federal Beer Tax Hike Affect Beer Prices?

To support the identification of the pass-through of beer taxes to prices obtained by using state-level changes on beer taxes, we examine the effect of beer tax changes on prices by using the federal tax hike of 1991. Although the federal tax increase of 1991 was large and almost doubled the tax rate, it was a nationwide change. This disallows the formation of any obvious control group. As a form of nonexperimental design, a regression discontinuity (RD) design can be used to evaluate the effect of the federal beer tax increase. The RD design is implemented in this case by considering the exact quarter of the federal beer tax hike as a threshold or a cut-off point and quarters before and after the quarter of tax increase as a rating variable. The RD design uses observations narrowly around the threshold to predict the counterfactual beer prices and relies on the assumption that other determinants of beer prices do not demonstrate a jump around the threshold. The average treatment effect at the cut-off point (k) can be written as:

(2) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

which equals E\[Y.sub.i](Y) - [Y.sub.i](0)|X = k]. The idea is to use average beer prices just before the federal tax increase as a counterfactual (denied treatment) and compare it to beer prices just after the cut-off point (received treatment). To evaluate the effect of the federal beer tax increase on beer prices, we estimate the following regression:

(3) [P.sub.tc] = [[alpha].sub.1] + [[alpha].sub.2][period.sub.t] + [[alpha].sub.3]f (.) + [[alpha].sub.4][X.sub.ts] + [e.sub.tc]

where [P.sub.tc] represents the price of a six-pack of beer at time t in city/county c; [period.sub.t] is a dummy variable indicating whether an observation belongs to the period after the federal beer tax increase, and f(*) is a smooth function of time away from the event of the federal beer tax increase. We primarily estimate Equation (3) semiparametrically by using a local linear estimation and a triangular kernel. To access the robustness of the results, we use alternate bandwidths and various polynomial forms. Specifications with and without the covariates are estimated.

The RD design hinges on two main assumptions: (1) the rating variable-quarters away from the event of the federal tax increase-is not influenced by the federal tax change of 1991; and (2) the assignment of treatment (increase in federal tax) is likely to be exogenous. To access the validity of these assumptions, we visually inspect the trend in several baseline characteristics. Any discrete jump in a control variable poses concerns regarding the validity of the underlying assumptions. A main concern is the recession of the early 1990s in which we expect to see a significant jump in the unemployment rate mirroring the federal tax hike of 1991. All specifications control for state-level unemployment rates. One possible reason for raising the federal alcohol taxes in 1991 may have been to raise government revenue during the recession. Ruhm (1995) suggests that alcohol consumption decreases during recessions and reduces tax revenues.

IV. DATA

Our two separate analyses described above utilize two different time frames: (1) the pass-through of the federal beer tax increase in 1991 uses data for the years 1985 to 1995; and (2) the pass-through of the relatively recent state-level excise taxes uses data from 2000 to 2014. We further refine the pass-through of state-level excise taxes to show the effects before and after the major mergers by comparing estimates from 2000 to 2006 and 2009 to 2014. We use the quarterly retail price of a six-pack of beer at the city/county level as reported in the Council for Community and Economic Research (C2ER) cost of living index (formerly known as the American Chamber of Commerce Research Association or ACCRA). The C2ER collects quarterly beer prices for approximately 300 communities including cities and counties. Note that the prices include state, federal, and local excise taxes but do not include sales taxes, therefore our results should be interpreted as the pass-through for excise taxes only.

Prior to the fourth quarter of 1989, ACCRA/C2ER collected prices on six-packs (12 oz. containers) of Budweiser and Schlitz. Between the fourth quarter of 1989 and the fourth quarter of 1999, prices for Budweiser and Miller were recorded. ACCRA then switched to collecting prices of Heineken beginning in the first quarter of 2000.6 To account for the switch in brand from Schlitz to Miller in the fourth quarter of 1989, the prices are brand adjusted. The average price of a six-pack of beer in the third quarter of 1989 was $3.33 and the average price was $3.35 in the fourth quarter of 1989. The U.S. beer price index, a part of the U.S. Bureau of Labor Statistics' Consumer Price Index, increased by 0.59% between these two quarters. Hence, we assume that had the brands remained the same, the average beer price would have been $3.35 ($3.33*1.0059) in the fourth quarter of 1989. Since the average beer price was in fact $3.35, this suggests that change of brand beginning in the fourth quarter of 1989 should not make much of a difference in our analysis (we discuss this issue in greater detail below). For our analyses pertaining to state-level tax changes that begin in 2000, no brand adjustment is necessary as we focus on the retail prices of Heineken. The beer taxes used are equivalent to a six-pack and are converted to 1984 dollars by using the BLS Consumer Price Index (CPI).

The federal and state-level beer taxes come from the Brewers Almanac (2012) and the respective taxes are corroborated using tax data reported by the Tax Foundation and NIAAA's Alcohol Policy Information System. While conducting analyses using state-level taxes, we focus on the time period 2000-2014 and use state-level changes in beer taxes (per six-pack). The earlier time period is not included because of the aforementioned brand change in 2000. The time frame 2000-2014 allows us to focus solely on Heineken, a brand that was not included in the major mergers of brewers. During 2000-2014, nine states experienced nominal tax changes. It is possible that this small number of nominal changes makes it difficult to identify the magnitude of the pass-through and we interpret the results with this caveat in mind.

In estimating the tax pass-through rate, it is questionable as to whether real or nominal values should be used. Strands of literature in Macroeconomics and Industrial Organization indicate that changes in nominal prices in response to shocks (such as increases in costs and the money growth rate) vary with the rate of inflation. Taylor (2000) argues that the responsiveness of changes in nominal prices as a result of increases in costs has declined with the rate of inflation over time in the United States and other developed countries. At a low rate of inflation, a relatively large fraction of buyers experience only a single price (Head et al. 2010). Any increases in costs will create a large increase in price dispersion, which strengthens the search intensity (from the buyer's side) and reduces market power. This limits market power and reduces the intensity with which sellers can influence price changes as a result of shocks. In contrast, during inflation, changes in costs will create a lower degree of price dispersion, which reduces the search intensity. Hence, prices may be more responsive to shocks during times of inflation. Considering these issues, and taking inflation into account, we examine the pass-through rate in real terms. However, to examine how manufacturers change prices as a result of tax changes, separate from the effects of inflation, we also perform some analyses by using nominal beer prices and taxes.

Figures 1A and IB show the trends in real beer prices and taxes in 1984 dollars (both federal and state-level) for the years 1985-1995 and 2000-2014, respectively. Figure 1A displays prices for Budweiser and Schlitz/Miller Light and Figure IB pertains to Heineken. Taken together, both figures highlight three main points: (1) real beer prices and taxes declined over time; (2) on average, state-level beer taxes are historically lower than the federal tax; and (3) the federal tax in 1991 is mirrored by a sharp increase in beer prices.

As mentioned above, the models include a number of variables to account for the determinants of demand, costs, and strategic considerations. We include the number of outlets licensed to sell alcohol per capita and the percentage of a state's population living in dry counties. These measures of the accessibility of alcohol capture both the retail market availability and ongoing sentiments regarding drinking. Data for liquor outlets and percentage living in dry counties come from the Adams Liquor Handbook (various years). We also show results from specifications that include the HHI in order to help describe the degree of concentration in the industry. The data for the HHI are extracted from Gokhale and Tremblay (2012). The data for state-level alcohol control policies (a blood alcohol concentration level of 0.08% and bans on Sunday alcohol sales) are obtained from the Alcohol Policy Information System and are used to reflect states' attitudes regarding use and misuse of alcohol. Following the findings of Ruhm (1995) that alcohol consumption is procyclical, we include annual state-specific unemployment rates and real per capita income (converted to 1984 dollars using the CPI). Data for the state unemployment rate come from the Bureau of Labor Statistics and data for per capita income are extracted from the Bureau of Economic Analysis. Due to religious beliefs, states having a high concentration of certain religions may also have strong anti-drinking sentiments. To account for such sentiments, we include the rates of adherence (per 1,000 population) for Southern Baptist, Latter-day Saints and Catholic religions. These data come from the Association of Religious Data Archives, which publishes adherence rates for the years 1990, 2000, and 2010. Data for the remaining years are interpolated. Finally, in analyses focusing on state-level beer tax changes, we include the number of brewing jobs in the state per 10,000 population and the number of brewing establishments per 10,000 population to proxy for the potential strength of the beer industry and to again capture some of the unobserved state sentiment toward alcohol. (7) The data for these variables are from the annual Brewers Almanac.

Summary statistics for the variables used in this study are shown in Table 1. The table shows that taxes have eroded over time due to inflation. The average real taxes for a six-pack of beer were 29 cents between 1985 and 1995, whereas the average fell to 25 cents between 2000 and 2014. Of note is that the proportion of states adapting a BAC level of 0.08 increased from 9.8% in the early period to 87.9% in the later years. This can be attributed to states lowering their BAC limits from 0.10 to comply with federal incentives for a 0.08 BAC level established by Congress in 2001. By 2004 all 50 states had passed a 0.08 BAC law.

V. PASS-THROUGH RESULTS

A. Results from the State-Level Difference-in-Differences Framework

Table 2 presents results from the state-level difference-in-differences framework using data from 2000 to 2014. Model (1) in Table 2 includes the full set of control variables except for state-specific linear time trends, and Model (2) adds the state-specific linear time trends. Models (3) and (4) include the HHI instead of year fixed effects, and Model (4) adds state-specific linear time trends to Model (3).8 Additionally, to capture the time invariant heterogeneity across C2ER cities/countries, all specifications include these area fixed effects.

The coefficients on real beer taxes in all models shown in Table 2 indicate that beer taxes are more than fully passed through to prices. Models (1) and (2) show that a $1 increase in the beer tax is associated with a rise in the beer price by $ 1.74 and $1.73, respectively. In other words, a 10-cent increase in the beer tax is associated with an increase in the beer price of 17 cents. Both coefficients are significant at the 1% level. Once the HHI is used instead of year fixed effects, the coefficient on the real beer tax in Model (3) increases to 3.40. However, the inclusion of state-specific linear time trends in Model (4) reduces the coefficient and indicates that a $1 increase in the beer tax per six-pack is associated with an increase in beer prices by $2.19 (significant at the 1% level). This suggests that a 10-cent increase in the beer tax is associated with an increase in retail prices of 22 cents. Our preferred specification is the one given by Model (2), which includes state-specific linear time trends. The inclusion of state-specific linear time trends accounts for unobserved linear factors in a state which potentially can be correlated with both beer prices and taxes.

Although the coefficients presented are all statistically different from zero, a more interesting test is whether the values are greater or less than one, that is, are the taxes under- or over- shifted. To answer this, we present p values at the bottom of Table 2 that pertain to a one-tailed hypothesis test of the null that [[alpha].sub.2] [less than or equal to] 1 versus the alternative that [[alpha].sub.2] > 1. In all models, we reject the null hypothesis in favor of the alternative that taxes are more than fully passed through (the pass-through coefficient is greater than 1) at the 5% level. To further test the robustness of our findings, we have conducted analyses with beer prices averaged at a state-level rather than using prices reported at the city-level. The resulting coefficients are extremely similar to the results shown in Table 2 and are available upon request.

Table 3 is similar to Table 2, except that Table 3 presents the pass-through estimates in nominal terms, using nominal beer prices and taxes. The results show the direct effect of increases in the beer taxes passed to prices, separate from the effects of inflation. Similar to the estimates from Table 2, the coefficients on the nominal beer tax in Table 3 indicate that beer taxes are more than fully passed through as prices. Using our preferred specification, Model (2) with linear time trends, a $1 (or 10 cent) increase in the tax per six-pack is associated with an increase in beer prices of $1.78 (18 cents) and the coefficient is significant at the 1% level. Referring to the p values at the bottom of Table 3, the pass-through estimates across all models in Table 3 are statistically different than 1 at the 5% level.

Six states changed their beer tax after the merger between Anheuser-Busch and InBev in July 2008. To investigate whether beer tax increases after the merger had different effects as compared to the pre-merger era, we estimate Equation (lb) including the interaction term between the post-merger indicator and beer taxes. In this equation, [[alpha].sub.2] shows the effect of higher beer taxes on prices in the pre-merger era, [[alpha].sub.3] indicates whether the incidence of beer tax pass-through is different in the post-merger era compared to the pre-merger era, and [[alpha].sub.2] + [[alpha].sub.3] demonstrates the effect of higher beer taxes on prices after the merger.

Table 4 shows the results after estimating Equation (lb). Models (1) and (2) present pass-through estimates using real beer prices and real beer taxes, whereas Models (3) and (4) use nominal beer prices and taxes. The coefficients on real beer taxes in Models (1) and (2) suggest that taxes are more than fully passed through to prices in the pre-merger era. The coefficient on the interaction term (between post-merger and real beer tax) is positive in Model (2), suggesting that the tax pass-through increased slightly after the merger. However, the coefficient is imprecisely estimated. The estimates presented at the bottom of Table 4 suggest that beer taxes remain more than fully passed through to prices after the merger. In our preferred specification, Model (2), the pre-period pass-through effect is 1.81 and increases slightly to 2.01 in the post-period. The F-Statistic obtained from the joint test between the interaction term and the beer tax reveals that the estimates are statistically significant at the 1% level. In addition, the post-merger pass-through estimate obtained from Model (2) is statistically greater than 1 (p value =.02). Similar results are obtained using nominal instead of real values in Models (3) and (4). The pre-merger pass-through effect shown in Model (4) is 1.82, whereas the post-merger effect is 2.00. This shows that although tax pass-through estimates may have increased slightly after the merger, such changes are not statistically significant at the conventional levels.

B. Results from the Federal Tax Increase Using a RD Design

Figures 2A and 2B show scatter plots and fitted local polynomial lines of the real beer price, real beer tax, percent dry, liquor outlets per capita. the proportion of states with BAC of 0.08%, the proportion of states with a Sunday sales ban, the unemployment rate, and real per capita income. Figure 2A shows significant discontinuities in both beer prices and taxes after the increase in the federal beer tax, whereas other baseline variables provide no such evidence of discontinuity, with the exception that the unemployment rate shows a visible jump, as expected, due to the recession of the early 1990s. (9)

Table 5 shows the RD estimates of the tax pass-through (in real terms). Panel A presents estimates without including the control variables shown in Table 1. Models (1) and (2) in Panel A show estimates from a local linear model with bandwidths of 9.43 and 8.1, which are optimal bandwidths estimated by using the method derived by Imbens and Kalyanaraman (2012) and cross validation, respectively. The coefficients on the after-tax change in Models (1) and (2) suggest that beer prices increased by $0,204 and $0,242, respectively. These coefficients are both significant at the 1% level. The tax change per six-pack due to the federal tax hike in 1991 was 11 cents per six-pack in 1984 dollars (14 cents in nominal terms). Our calculation suggests that the increase in the federal tax in 1991 was more than fully passed through as beer prices with a 10-cent increase in the beer taxes increasing real beer prices by 19-22 cents. Models (3)-(7) present OLS estimates after including different functional forms of the running variables as reported at the bottom of the table. Here, the sample considered is nine quarters before and after the quarter of the federal beer tax increase. (10) The OLS estimates are similar to the estimates obtained from the local linear regression indicating that beer taxes are more than fully passed through to prices. To further assess the robustness of the findings in Panel A, Panel B presents OLS estimates after including the full set of state-specific variables listed in the notes. The estimates obtained in Panel B also suggest that beer taxes are over-shifted to prices and are similar in magnitude to the estimates in Panel A. (11)

[FIGURE 2 OMITTED]

Table 6 shows the results after using nominal beer prices as the dependent variable. The local linear estimates in Models (1) and (2) show that the beer price increased by $0,255 (with a bandwidth of 6.997) and $0,278 (with a bandwidth of 8.1), respectively, after the federal tax increase. The calculation suggests that a 10-cent increase in the beer tax increased beer prices by 18-20 cents in nominal terms. This range is similar to that of Table 5.

There is the possibility that the beer companies anticipated the increase in the federal beer tax and acted proactively by stockpiling beer and adjusting prices before the tax increase. In this case, the pass-through estimate of the federal beer tax would be understated. However, Figure 2A shows that real beer prices decreased steadily before the federal tax hike, with the lowest prices seen just before the tax hike. This figure provides some evidence that the anticipation of the federal beer tax increase did not increase beer prices beforehand. In addition, we ran some placebo models using different quarters in 1990 as the date of the treatment and found statistically insignificant coefficients with magnitudes close to zero. Results are available upon request.

VI. ROBUSTNESS CHECKS

One issue that may influence our results regards border crossing. That is, retailers may respond differently in terms of the pass-through when faced with competition from retailers located in neighboring states with lower excise tax rates. To account for this possibility, we run models that eliminate border areas. To do so, we first identify the counties in which the ACCRA/C2ER cities are located. We then calculate the minimum distance from the centroid of the county to the nearest state border. We conduct our analyses by excluding counties within 30 miles of a bordering state. This eliminates 101 cities from the analysis. The results of this robustness check are presented below in Table A1, Panel A. The findings are similar to the results presented in Table 2. We also test models that exclude counties within 10 and 20 miles of the border and results remain similar (results available upon request). As a second method, we use all cities but include a measure of neighboring states' beer taxes in the models. The border tax is weighted by the distance from the centroid of the county located close to the border. The results from this robustness exercise are presented in Table Al, Panel B. The coefficients on the real beer tax are again similar in magnitude to the coefficients shown in Table 2. The results from these robustness exercises suggest that cross-border shopping is not driving the main results of this study.

Another issue of concern regards sales taxes on beer. Recall that the C2ER prices include excise taxes but do not include sales taxes, so our main results are interpreted as the pass-through for excise taxes only. However, some states impose additional ad valorem taxes on alcohol products so it is useful to add in a measure of these taxes in order to form a more complete picture of the pass-through rates. Information on ad valorem beer taxes and general sales taxes are gathered from the Alcohol Policy Information System and the Tax Foundation. The rules vary by state with some states imposing no sales taxes on alcohol, others applying the general sales tax to alcohol products, and still others adding an alcohol-specific ad valorem tax. We include a separate variable that contains the percentage of price imposed as a sales/ad valorem tax in the models shown in Table Al, Panel C. The coefficients on the real beer (excise) taxes are positive and similar in magnitude to the coefficients provided in Table 2, suggesting that the main result of our study is robust to the inclusion of ad valorem taxes.

To further access the validity of our results, we also implement a first difference regression where changes in beer prices are regressed on tax changes and first differences of time-varying variables, including leads and lags of beer taxes. This tests for any delay or anticipation effects. The results from such specifications are similar to our main findings, suggesting that beer taxes are more than fully passed through to retail prices (results not shown but available upon request).

VII. DISCUSSION AND CONCLUSION

This paper adds to the literature by providing new estimates of the beer tax pass-through using relatively recent state-level changes in beer taxes. The mergers of leading beer producers--SABMiller and Coors in 2007 followed by Anheuser-Busch and InBev in 2008--warrants new estimates of the beer tax pass-through. To support the identification of the tax pass-through obtained by using state-level changes, this study also examines the effect of the federal beer tax increase of 1991 (when the tax per drink almost doubled) on retail prices.

Using within variation in state-level tax changes, we find that a 10-cent increase in beer taxes increases prices by about 17 cents. We find no statistical evidence suggesting that the mergers changed the pass-through of beer taxes to prices--beer taxes remain more than fully passed through as prices in the post-merger years although the magnitudes increase slightly after the mergers to 20 cents. Our estimates are comparable with the findings from Young and Kwapisz's (2002) study, which finds evidence of over-shifting, suggesting that 10-cent increases in beer taxes increases beer prices by a range of 17-19 cents.

The findings obtained from the RD design that analyzes the federal beer tax increase in 1991 indicate that a 10-cent increase in federal beer taxes raises beer prices by 19-22 cents. The point estimates of the federal tax pass-through are only slightly larger than those found for state taxes. The stability of the estimates is remarkable, especially given the differences in time periods understudy (early 1990s vs. 2000-2014), and the different beer brands examined.

The beer industry in the United States provides a unique situation for studying tax pass-through behaviors. Once highly competitive, it is now characterized as an oligopoly, but includes fierce brand competition within. Recent mergers have further concentrated the industry. Occasional changes in excise tax rates make it possible to analyze tax pass-through rates over time and to see how concentration affects firm behaviors in this regard. Our main conclusion is that despite the changes within the industry, the pass-through rates remain remarkably stable over time and firms are able to over-shift the tax burden to consumers.

ABBREVIATIONS

ACCRA: American Chamber of Commerce Research Association

BAC: Blood Alcohol Content

C2ER: Council for Community and Economic Research

CPI: Consumer Price Index

CR4: Four-firm Concentration Ratio

HHI: Herfindahl-Hirschman Index

RD: Regression Discontinuity APPENDIX TABLE A1 State-Level Tax Pass-Through Estimates from Alternative Specifications (1) (2) Panel A: Excluding Counties Close to Border N = 12,154 Real beer tax 1.741 *** 1.83 *** (six-pack) (0.494) (0.316) Panel B: Controlling for Across Border Taxes N = 15,040 Real beer tax 1.745 *** 1.726 *** (six-pack) (0.431) (0.314) Neighboring state's real beer -0.031 -0.011 tax (weighted by distance (0.050) (0.043) to the bordering state) Panel C: Effect of State-Level Beer Taxes Including Sales Tax N= 15,040 Real beer tax (six-pack) 1.741 *** 1.722 *** (0.433) (0.319) Sales tax -0.005 0.003 (0.008) (0.007) HHI No No Year fixed effects Yes Yes Area fixed effects Yes Yes Linear time trend No Yes (3) (4) Panel A: Excluding Counties Close to Border N = 12,154 Real beer tax 3.275 *** 2.069 *** (six-pack) (0.693) (0.377) Panel B: Controlling for Across Border Taxes N = 15,040 Real beer tax 3 421 *** 2.185 *** (six-pack) (0.709) (0.413) Neighboring state's real beer -0.025 -0.004 tax (weighted by distance (0.057) (0.046) to the bordering state) Panel C: Effect of State-Level Beer Taxes Including Sales Tax N= 15,040 Real beer tax (six-pack) 3 4 *** 2.186 *** (0.705) (0.415) Sales tax -0.001 0.003 (0.008) (0.008) HHI Yes Yes Year fixed effects No No Area fixed effects Yes Yes Linear time trend No Yes Notes: Robust standard error clustered at the state level in parentheses. The dependent variable is the price for a six- pack of Heineken, converted to 1984 dollars. All specifications include percent dry, the religion variables, liquor outlets, 0.08 BAC, Sunday alcohol ban, unemployment rate, real income per capita, brewing jobs, brewing establishments, indicators for quarter, and in Models 3 and 4, the yearly Herfindahl-Hirschman Index. * Significant at 10%; ** significant at 5%; *** significant at 1% in a two-tailed test. TABLE A2 Federal-Level Tax Pass-Through Estimates in Real Terms, Years Limited to 1990-1995 Local Linear Local Linear OLS Models (1) (2) (3) Panel A. Without Controls After 0.2413 *** 0.2413 *** 0.1830 *** (0.0330) (0.0330) (0.0118) [R.sup.2] .0678 Panel B. With Full Controls After 0.2072 *** (0.0343) [R.sup.2] .2531 Function of Running Variable Included Running variable No Running * after federal No change Running squared No Running squared * after No federal change N 6,937 6,937 4,153 OLS OLS Models (4) (5) Panel A. Without Controls After 0.2276 *** 0.2335 *** (0.0127) (0.0193) [R.sup.2] .0703 .0704 Panel B. With Full Controls After 0.2571 *** 0.2475 *** (0.0351) (0.0376) [R.sup.2] .2578 .2578 Function of Running Variable Included Running variable Yes Yes Running * after federal No Yes change Running squared No No Running squared * after No No federal change N 4,153 4,153 OLS OLS Models (6) (7) Panel A. Without Controls After 0.2249 *** 0.3500 *** (0.0183) (0.0276) [R.sup.2] .0709 .0726 Panel B. With Full Controls After 0.2421 *** 0.3643 *** (0.0395) (0.0488) [R.sup.2] .2578 .2593 Function of Running Variable Included Running variable Yes Yes Running * after federal Yes Yes change Running squared Yes Yes Running squared * after No Yes federal change N 4,153 4,153 Notes: Robust standard errors are reported in parentheses. Panel A reports results without including control variables. Models (1) and (2) use data from 1990 to 1995. Models (3)-(7) use data starting in 1990 and ending 9 quarters away from the quarter of the increase in the federal beer tax. Observations in the first quarter of 1991 (during the quarter of the federal tax change) are not included to induce the clarity of the RD design. Models (1) and (2) show local linear regression estimates using a triangle kernel where the bandwidths are estimated by using the method derived from Imbens and Kalyanaraman (2012) and cross validation, respectively. Panel B reports OLS results with state-level control variables: percentage dry, liquor outlets, 0.08 BAC, Sunday alcohol ban, unemployment rate, real income per capita, and the yearly Herfindahl-Hirschman Index. * Significant at 10%; ** significant at 5%; *** significant at 1% in a two-tailed test.

doi: 10.1111/ecin.12343

Online Early publication April 6, 2016

REFERENCES

Ally, A. K., Y. Meng, R. Chakraborty, P. W. Dobson, J. S. Seaton. J. Holmes, C, Angus, Y. Guo, D. Hill-McManus, A. Brennan, and P. S. Meier. "Alcohol Tax Pass-Through Across the Product and Price Range: Do Retailers Treat Cheap Alcohol Differently?" Addiction, 109(12), 2014, 1994-2002.

Anderson, K. B. A Review of Structure-Performance Studies in Grocery Retailing. Washington, D.C.: Bureau of Economics, Federal Trade Commission, 1990.

Anheuser-Busch. "InBev and Anheuser-Busch Agree to Combine, Creating the Global Leader in Beer with Budweiser as Its Flagship Brand." 2008. Accessed January 24, 2016. http://anheuser-busch.com/index .php/inbev-and-anheuser-busch-agree-to-combinecreating-the-global-leader-in-beer-with-budweiser-asits-flagship-brand/.

Ashenfelter, O., D. Hosken, and M. Weinberg. "Efficiencies Brewed: Pricing and Consolidation in the U.S. Beer Industry." The Rand Journal of Economics, 46(2), 2015, 328-61.

Barzel, Y. "An Alternative Approach to the Analysis of Taxation." Journal of Political Economy, 84(6), 1976, 1177-97.

Bergman, M. U., and N. L. Hansen. "Are Excise Taxes on Beverages Fully Passed Through to Prices? The Danish Evidence." Working Paper, University of Copenhagen, Denmark, 2010.

Besley, T., and H. S. Rosen. "Sales Taxes and Prices: An Empirical Analysis." National Tax Journal, 52(2), 1999, 157-78.

Bray, C. "Deadline Extended on Anheuser-Busch InBev Offer for SABMiller." 2015. Accessed January 26, 2016. http://www.nytimes.com/2015/10/29/business/ dealbook/anheuser-busch-inbev-sabmiller-beer-dealtalks.html.

Carbonnier, C. "Pass-through of Per Unit and Ad Valorem Consumption Taxes: Evidence from Alcoholic Beverages in France." The BE Journal of Economic Analysis and Policy, 13(2), 2013, 837-63.

Chaloupka, F., M. Grossman, and H. Saffer. "The Effects of Price on Alcohol Consumption and Alcohol-Related Problems." Alcohol Research and Health, 26(1), 2002, 22-34.

Cook, P., and G. Tauchen. "The Effect of Liquor Taxes on Heavy Drinking." The Bell Journal of Economics, 13(2), 1982, 379-90.

Fullerton, D., and G. E. Metcalf. "Tax Incidence." Handbook of Public Economics, 4, 2002, 1787-872.

George, L. M. "National Television and the Market for Local Products: The Case of Beer." Journal of Industrial Economics, 57(1), 2009, 85-111.

Gokhale, J., and V. J. Tremblay. "Competition and Price Wards in the U.S. Brewing Industry." Journal of Wine Economics, 7(02), 2012, 226-40.

Greer, D. F. "Beer: Causes of Structural Change," in Industry Studies, edited by L. L. Duetsch. New York: Prentice Hall, 1998, 85-115.

Grossman, M., F. J. Chaloupka, H. Saffer, and A. Laixuthai. "Effects of Alcohol Price Policy on Youth: A Summary of Economic Research." Journal of Research on Adolescence, 4(2), 1994, 347-64.

Hall, R. E. "The Relation between Price and Marginal Cost in U.S. Industry." Journal of Political Economy, 96, 1988, 921-47.

Head, A., A. Kumar, and B. Lapham. "Market Power, Price Adjustment, and Inflation." International Economic Review, 51(1), 2010, 73-98.

Heineken. "Heineken N.V. Reports 18% Organic Net Profit Growth for 2009, Triples Free Operating Cash Flow." 2010. Accessed January 26, 2016. http://www .theheinekencompany.com/media/media-releases/ press-releases/2010/02/1387272.

Imbens, G., and K. Kalyanaraman. "Optimal Bandwidth Choice for the regression Discontinuity Estimator." The Review of Economic Studies, 79, 2012, 933-59.

Iwasaki, N., B. J. Seldon, and V. J. Tremblay. "Brewing Wars of Attrition for Profit (and Concentration)." Review of Industrial Organization, 33(4), 2008, 263-79.

Kenkel, D. "New Estimates of the Optimal Tax on Alcohol." Economic Inquiry, 34(2), 1996, 296-319.

--. "Are Alcohol Tax Hikes Fully Passed Through to Prices? Evidence from Alaska." American Economic Review, 95(2), 2005, 273-7.

Kopczuk, W., J. Marion, E. Muehlegger, and J. Slemrod. "Do the Laws of Tax Incidence Hold? Point of Collection and the Pass-Through of State Diesel Taxes." Working Paper No. w19140, National Bureau of Economic Research, 2013.

LaVallee, R. A., and H.-y. Yi. Apparent Per Capita Alcohol Consumption: National, State, and Regional Trends, 1977-2009. Bethesda, MD: National Institute on Alcohol Abuse and Alcoholism, 2011.

Markowitz, S., and M. Grossman. "The Effects of Beer Taxes on Physical Child Abuse." Journal of Health Economics, 19(2), 2000, 271-82.

Rooney, B., and C. Clifford. "Brace Yourself: Beer Prices Are Going Up." 2009. Accessed January 26, 2016. http://money.cnn.com/2009/08/25/news/companies/ anheuser_busch/.

Ruhm, C. J. "Economic Conditions and Alcohol Problems." Journal of Health Economics, 14(5), 1995, 583-603.

Scherer, F. M., and D. Ross. Industrial Market Structure and Economic Performance. Boston: Houghton Mifflin Company, 1990.

Shepherd, W. G. The Economics of Industrial Organization. Englewood Cliffs, NJ: Prentice Hall, 1990.

Shrestha, V. "Estimating the Price Elasticity of Demand for Different Levels of Alcohol Consumption among Young Adults." American Journal of Health Economics, 1(2), 2015,224-54.

Stehr, M. "The Effect of Sunday Sales Ban and Excise Taxes on Drinking and Cross-Border Shipping for Alcoholic Beverage." National Tax Journal, 60, 2007, 85-105.

Taylor, J. B. "Low Inflation, Pass-Through, and the Pricing Power of Firms." European Economic Review, 44(7), 2000, 1389-408.

Tennessee Senate Republican Caucus. "List of Some of the Laws that Will Become Effective July 1, 2013." 2013. Accessed January 24, 2016. http://www.tngopsenate .com/list-of-some-of-the-laws-that-will-becomeeffective-july-1-2013/.

Tremblay, V. J., and C. H. Tremblay. The US Brewing Industry: Data and Economic Analysis. Cambridge, MA: MIT Press, 2005.

Tremblay, V. J., N. Iwasaki, and C. H. Tremblay. "The Dynamics of Industry Concentration for U.S. Micro and Macro Brewers." Review of Industrial Organization, 26(3), 2005, 307-24.

Wooldridge, J. M. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press, 2001.

Young, D. J., and A. B. Kwapisz. "Alcohol Taxes and Beverage Prices." National Tax Journal, LV, 2002, 57-73.

(1.) Estimates of elasticities vary but recent work by Shrestha (2015) finds a price elasticity of -0.2. He also finds that heavy drinkers reduce their alcohol consumption in response to higher alcohol prices. In addition, many studies have showed that higher alcohol taxes are associated with reductions in alcohol-related outcomes including mortality, crime, and health problems. See for example, Cook and Tauchen (1982), Grossman et al. (1994), Kenkel (1996), Markowitz and Grossman (2000), Chaloupka, Grossman, and Saffer (2002).

(2.) According to a report from the National Institute of Alcohol Abuse and Alcoholism (NIAAA), beer accounted for more than 50% of the total per capita ethanol consumption in 2009 (LaVallee and Yi 2011).

(3.) See Bray (2015).

(4.) Many states calculate taxes based on the volume of alcoholic beverage, and few calculate off the weight.

(5.) For example, the point of collection of beer taxes in Utah is at the wholesale level, in Washington it is at the producer level, in Illinois and Nebraska it is at the manufacturer or distributor level. One reason for conducting state-level and federal analyses separately is due to variation in the parties on which federal and state taxes are levied.

(6.) ACCRA/C2ER ceased reporting beer prices for the fourth quarter of the year starting in 2007. Hence, fourth quarters for 2007 through 2014 are excluded from our analyses.

(7.) Brewing establishments are defined as the number of breweries and small brewers. Data for brewing jobs and brewing establishments are only available for the years 2004, 2006, 200S, 2010, and 2012. Values for other years are interpolated.

(8.) All models include Tennessee, although we note that Tennessee experienced a dramatic switch in its tax structure in mid-2013. Originally Tennessee had a price-based tax structure that included a 17% tax on price at the wholesale level. On July 1, 2013, Tennessee's Beer Reform Act went into effect, and changed the tax structure from a price based system to taxes based on volume. Additional analyses (not shown) are conducted by excluding Tennessee. The results remain similar. For details see Tennessee Senate Republican Caucus (2013).

(9.) The unemployment rate is included in some of the RD models. See Panel B of Table 5. Religion variables, brewing jobs, and brewing establishments are excluded since there is little to no variation in these variables around the window of the tax change.

(10.) Different numbers of quarters away from the date of the federal tax change are used to test the robustness of our results, including 6- and 12-month windows. The results are robust and are available upon request.

(11.) As mentioned in the Data section, ACCRA/C2ER reported prices of Budweiser and Schlitz before the fourth quarter of 1989 and switched into Miller instead of Schlitz starting in the fourth quarter. To investigate whether the change in brand affects the RD results, we perform a robustness check that only uses data after the brand change. The results from this exercise are presented in Table A2, which is structured similarly to Table 5. The coefficients are similar in both tables, providing evidence that the change in brand from Schlitz to Miller does not influence the pass-through results for the federal tax increase.

Shrestha: Assistant Professor, Department of Economics, Towson University, Towson, MD 21252. Phone 410-704-2956, Fax 410-704-3424, E-mail [email protected]

Markowitz: Assistant Professor, Department of Economics, Emory University and NBER, Atlanta, GA 30322. Phone 404-712-8167, Fax 404-727-4639, E-mail [email protected] TABLE 1 Summary Statistics Federal Tax Analysis 1985-1995 A = 11,999 Variable Mean SD Real beer price of Budweiser and Miller 2.726 0.303 (six-pack in 1984 dollars) Nominal beer price of Budweiser and Miller 3.568 0.562 Real beer price of Heineken (six-pack in 1984 dollars) Nominal beer price of Heineken Real beer tax (in 1984 dollars) federal + state 0.292 0.12 Real federal beer tax (in 1984 dollars) 0.18 0.044 Nominal federal beer tax 0.241 0.081 Real state-level beer tax (in 1984 dollars) Nominal state-level beer tax Percentage dry 4.886 10.957 Liquor outlets per capita 0.001 0.001 BAC 0.08 0.098 0.297 Sunday alcohol sales ban 0.317 0.465 Unemployment rate 6.138 1.523 Real per capita income (in $1,000) 13.767 1.651 Latter-day Saints (per 1,000 population) Catholic (per 1,000 population) Southern Baptist (per 1.000 population) Brewing jobs (per 10,000 population) Brewing establishments (per 10,000 population) HHI (divided by 100) 26.908 2.065 State Tax Analysis 2000-2014 A = 15,040 Variable Mean SD Real beer price of Budweiser and Miller (six-pack in 1984 dollars) Nominal beer price of Budweiser and Miller Real beer price of Heineken 3.888 0.32 (six-pack in 1984 dollars) Nominal beer price of Heineken 7.828 0.763 Real beer tax (in 1984 dollars) federal + state 0.247 0.078 Real federal beer tax (in 1984 dollars) Nominal federal beer tax Real state-level beer tax (in 1984 dollars) 0.08 0.07 Nominal state-level beer tax 0.169 0.148 Percentage dry 4.318 10.016 Liquor outlets per capita 0.001 0.001 BAC 0.08 0.879 0.326 Sunday alcohol sales ban 0.29 0.454 Unemployment rate 5.912 1.906 Real per capita income (in $1,000) 17.433 2.382 Latter-day Saints (per 1,000 population) 23.11 80.453 Catholic (per 1,000 population) 166.186 103.849 Southern Baptist (per 1.000 population) 96.463 97.479 Brewing jobs (per 10,000 population) 1.516 2.1 Brewing establishments (per 10,000 population) 0.087 0.078 HHI (divided by 100) 38.807 3.207 Note: The time period 1985-1995 excludes the first quarter of 1991 when the federal beer tax increase occurred. C2ER cities/counties, all specifications include these area fixed effects. TABLE 2 State-Level Tax Pass-Through Estimates in Real Terms (1) (2) Real beer tax (six-pack) 1.741 *** 1.725 *** (0.430) (0.315) Percentage dry -0.009 -0.017 *** (0.007) (0.004) Latter-day Saints (per 1,000 population) -0.021 *** -0.012 ** (0.007) (0.005) Catholics (per 1,000 population) -0.001 0.001 (0.001) (0.001) Southern Baptists (per 1,000 population) -0.002 -0.004 (0.005) (0.003) Liquor outlets per capita 27.360 9.539 (36.687) (20.405) BAC 0.08 -0.026 0.018 (0.030) (0.024) Sunday alcohol sales ban -0.012 0.006 (0.079) (0.055) Unemployment rate 0.027 * 0.008 (0.014) (0.009) Brewing jobs (per 10,000 pop.) 0.023 -0.014 (0.015) (0.018) Brewing establishments (per 10,000 pop.) -0.767 -0.493 (0.547) (0.491) Real per capita income (in $1,000) 0.028 0.032 (0.018) (0.020) HHI [rho] Value ([H.sub.0]: pass-through .046 .013 [less than or equal to] 1) Year fixed effects Yes Yes Area fixed effects Yes Yes Linear time trend No Yes N 15,040 15,040 [R.sup.2] .620 .664 (3) (4) Real beer tax (six-pack) 3 397 *** 2187 *** (0.703) (0.415) Percentage dry -0.008 -0.012 *** (0.006) (0.003) Latter-day Saints (per 1,000 population) -0.028 *** -0.014 *** (0.007) (0.003) Catholics (per 1,000 population) -0.001 0.001 (0.001) (0.001) Southern Baptists (per 1,000 population) -0.004 -0.002 (0.003) (0.002) Liquor outlets per capita 8.301 -1.205 (39.701) (27.089) BAC 0.08 -0.092 *** -0.036 * (0.025) (0.019) Sunday alcohol sales ban -0.007 0.030 (0.075) (0.056) Unemployment rate 0.032 *** 0.036 *** (0.006) (0.005) Brewing jobs (per 10,000 pop.) 0.032 ** -0.015 (0.016) (0.024) Brewing establishments (per 10,000 pop.) -1.328 ** -0.063 (0.498) (0.509) Real per capita income (in $1,000) -0.003 0.035 * (0.015) (0.020) HHI -0.021 *** -0.002 (0.007) (0.006) [rho] Value ([H.sub.0]: pass-through .001 .003 [less than or equal to] 1) Year fixed effects No No Area fixed effects Yes Yes Linear time trend No Yes N 15,040 15,040 [R.sup.2] .600 .652 Notes: Robust standard error clustered at the state level in parentheses. The dependent variable is the price for a six-pack of Heineken, converted to 1984 dollars. All specifications include indicators for quarter. * Significant at 10%; ** significant at 5%; *** significant at 1% in a two/tailed test. TABLE 3 State-Level Tax Pass-Through Estimates in Nominal Terms (1) (2) Nominal beer tax 2 110 *** 1.782 *** (0.496) (0.280) Percentage dry -0.023 -0.036 *** (0.014) (0.009) Latter-day Saints (per 1,000 population) -0.037 *** -0.019 * (0.013) (0.010) Catholics (per 1,000 population) -0.001 0.0001 (0.003) (0.002) Southern Baptists (per 1,000 population) -0.002 -0.009 (0.008) (0.006) Liquor outlets per capita 41.367 16.072 (71.492) (44.549) BAC 0.08% -0.035 0.045 (0.055) (0.048) Sunday alcohol sales ban -0.045 0.024 (0.145) (0.100) Unemployment rate 0.059 ** 0.018 (0.029) (0.019) Brewing jobs (per 10.000 pop.) 0.045 -0.022 (0.033) (0.037) Brewing establishments (per 10,000 pop.) -1.546 -0.976 (1.050) (1.019) Real per capita Income (in $ 1,000) 0.070 * 0.053 (0.037) (0.041) HHI p Value ([H.sub.0]: pass-through .015 .004 [less than or equal to] 1) Year fixed effects Yes Yes Area fixed effects Yes Yes Linear time trend No Yes N 15,040 15,040 [R.sup.2] .717 .749 (3) (4) Nominal beer tax 2.275 *** 1.728 *** (0.465) (0.232) Percentage dry -0.022 -0.040 *** (0.017) (0.007) Latter-day Saints (per 1,000 population) -0.025 ** -0.021 *** (0.012) (0.007) Catholics (per 1,000 population) 0.0002 -0.0001 (0.003) (0.002) Southern Baptists (per 1,000 population) -0.002 -0.008 * (0.008) (0.004) Liquor outlets per capita 41.823 18.285 (75.780) (51.998) BAC 0.08% 0.058 -0.041 (0.051) (0.040) Sunday alcohol sales ban -0.086 0.024 (0.151) (0.102) Unemployment rate 0.079 *** 0.048 *** (0.014) (0.010) Brewing jobs (per 10.000 pop.) 0.006 -0.021 (0.032) (0.043) Brewing establishments (per 10,000 pop.) 0.056 -0.814 (1.220) (1.023) Real per capita Income (in $ 1,000) 0.218 *** 0.076 * (0.043) (0.042) HHI 0.083 *** 0.017 (0.010) (0.011) p Value ([H.sub.0]: pass-through .004 .001 [less than or equal to] 1) Year fixed effects No No Area fixed effects Yes Yes Linear time trend No Yes N 15,040 15.040 [R.sup.2] .691 .745 Notes: Robust standard error clustered at the state level in parentheses. The dependent variable is the nominal price for a six-pack of Heineken. All specifications include indicators for quarter. * Significant at 10%; ** significant at 5%; *** significant at 1% in a two-tailed test. TABLE 4 State-Level Tax Pass-Through Before and After Beer Company Mergers (1) (2) Real beer tax (six-packs) 1.876 *** 1.809 *** (0.355) (0.297) Post-merger * real beer tax 0.146 0.201 (0.220) (0.353) Nominal beer tax Post-merger * nominal beer tax Percentage dry -0.009 -0.016 *** (0.008) (0.005) Latter-day Saints (per 1.000 population) -0.022 *** -0.012 ** (0.007) (0.005) Catholics (per 1,000 population) -0.001 0.001 (0.001) (0.001) Southern Baptists (per 1,000 population) -0.000 -0.002 (0.004) (0.003) Liquor outlets per capita 27.871 10.534 (38.829) (21.119) BAC 0.08 -0.023 0.020 (0.029) (0.025) Sunday alcohol sales ban -0.022 -0.012 (0.081) (0.055) Unemployment rate 0.031 ** 0.012 (0.014) (0.009) Brewing jobs (per 10,000 pop.) 0.031 * -0.003 (0.017) (0.019) Brewing establishments (per 10,000 pop.) -0.626 -0.358 (0.605) (0.634) Real per capital income (in $1,000) 0.026 0.029 (0.017) (0.023) Effect after the mergers 2.022 *** 2.01 *** F-Statistics 14.3 21.79 Year fixed effects Yes Yes Area fixed effects Yes Yes Linear time trend No Yes N 13,253 13,253 [R.sup.2] .632 .676 (3) (4) Real beer tax (six-packs) Post-merger * real beer tax Nominal beer tax 2.256 *** 1.828 *** (0.378) (0.290) Post-merger * nominal beer tax -0.131 0.174 (0.226) (0.342) Percentage dry -0.022 -0.034 *** (0.015) (0.010) Latter-day Saints (per 1.000 population) -0.038 *** -0.020 * (0.014) (0.010) Catholics (per 1,000 population) -0.001 0.001 (0.003) (0.002) Southern Baptists (per 1,000 population) 0.001 -0.004 (0.008) (0.006) Liquor outlets per capita 38.235 18.411 (77.634) (46.853) BAC 0.08 -0.032 0.049 (0.055) (0.049) Sunday alcohol sales ban -0.053 -0.010 (0.147) (0.098) Unemployment rate 0.071 ** 0.026 (0.030) (0.019) Brewing jobs (per 10,000 pop.) 0.072 ** 0.004 (0.034) (0.037) Brewing establishments (per 10,000 pop.) -1.321 -0.725 (1.175) (1.327) Real per capital income (in $1,000) 0.065 * 0.048 (0.035) (0.046) Effect after the mergers 2.125 *** 2.002 *** F-Statistics 17.91 25.35 Year fixed effects Yes Yes Area fixed effects Yes Yes Linear time trend No Yes N 13,253 13,253 [R.sup.2] .738 .769 Notes: Robust standard error clustered at the state level in parentheses. The dependent variable is the price for a six-pack of Heineken. The analyses exclude years 2007 and 2008 (merger years). All specifications include indicators for quarter. * Significant at 10%; ** significant at 5%; *** significant at 1% in a two-tailed test. TABLE 5 Federal-Level Tax Pass-Through Estimates in Real Terms Local Linear Local Linear OLS Models (1) (2) (3) Panel A. Without Controls After 0.204 *** 0.242 *** 0.1544 *** (0.0263) (0.0355) (0.0107) [R.sup.2] .0678 Panel B. With Full Controls After 0.1596 *** (0.0188) [R.sup.2] .3378 Function of Running Variable Included Running variable No Running * after federal No change Running squared No Running squared * after No federal change N 11,999 11,999 5,514 OLS OLS Models (4) (5) Panel A. Without Controls After 0.2042 *** 0.2054 *** (0.0118) (0.0127) [R.sup.2] .0702 .0702 Panel B. With Full Controls After 0.1717 *** 0.1645 *** (0.0243) (0.0222) [R.sup.2] .3381 .3385 Function of Running Variable Included Running variable Yes Yes Running * after federal No Yes change Running squared No No Running squared * after No No federal change N 5,514 5,514 OLS OLS Models (6) (7) Panel A. Without Controls After 0.2095 *** 0.2358 *** (0.0131) (0.0176) [R.sup.2] .0705 .0708 Panel B. With Full Controls After 0.1635 *** 0.2114 *** (0.0243) (0.0230) [R.sup.2] .3385 .3395 Function of Running Variable Included Running variable Yes Yes Running * after federal Yes Yes change Running squared Yes Yes Running squared * after No Yes federal change N 5,514 5,514 TABLE 6 Federal-Level Tax Pass-Through Estimates in Nominal Terms Local Local Linear Linear OLS Models (1) (2) (3) Panel A. Without Controls After 0.255 *** 0.278 *** 0.5500 *** (0.0388) (0.0479) (0.0135) [R.sup.2] .3336 Panel B. With Controls After 0.3205 *** (0.0258) [R.sup.2] .5137 Other Controls Running variable No Running * No after federal change Running squared No Running squared * No after federal change N 11,999 11,999 5,514 OLS OLS OLS Models (4) (S) (6) Panel A. Without Controls After 0.2949 *** 0.2789 *** 0.2820 *** (0.0157) (0.0167) (0.0173) [R.sup.2] .3583 .3601 .3601 Panel B. With Controls After 0.2285 *** 0.2056 *** 0.1957 *** (0.0338) (0.0308) (0.0336) [R.sup.2] .5201 .5218 .5221 Other Controls Running variable Yes Yes Yes Running * No Yes Yes after federal change Running squared No No Yes Running squared * No No No after federal change N 5,514 5,514 5,514 OLS Models (7) Panel A. Without Controls After 0.2947 *** (0.0234) [R.sup.2] .3601 Panel B. With Controls After 0.2271 *** (0.0326) [R.sup.2] .5222 Other Controls Running variable Yes Running * Yes after federal change Running squared Yes Running squared * Yes after federal change N 5,514 Notes: Robust standard errors are reported in parentheses. Panel A reports results without including control variables. Models (1) and (2) use data from 1985 to 1995. Models (3)-(7) use data 9 quarters before and after the date of the federal beer tax increase. Observations in the first quarter of 1991 (the quarter of the federal tax change) are not included to induce the clarity of the RD design. Models (1) and (2) show local linear regression estimates using a triangle kernel where the bandwidths are estimated by using the method derived by Imbens and Kalyanaraman (2012) and cross validation, respectively. Panel B reports OLS results with state-level control variables: percent dry, liquor outlets, 0.08 BAC, Sunday alcohol ban, unemployment rate, real income per capita, and the yearly Herfindahl-Hirschman Index. * Significant at 10%; ** significant at 5%; *** significant at 1% in a two-tailed test.