- Email: [email protected]
Spillovers from the beer market to U.S. cigarette demand
Economic Modelling 55 (2016) 292–297
Contents lists available at ScienceDirect
Economic Modelling journal homepage: www.elsevier.com/locate/ecmod
Spillovers from the beer market to U.S. cigarette demand☆ Rajeev K. Goel a,⁎, James E. Payne b, James W. Saunoris c a b c
Department of Economics, Illinois State University, Normal, IL 61790-4200, USA J. Whitney Bunting College of Business, Georgia College & State University, Milledgeville, GA 31061, USA Department of Economics, Eastern Michigan University, Ypsilanti, MI 48197, USA
a r t i c l e
i n f o
Article history: Accepted 18 February 2016 Available online xxxx Keywords: Cigarettes Beer Demand Smoking Smuggling Elasticity Regulations U.S.
a b s t r a c t We study the cross-effects of the beer market on U.S. cigarette demand. The extant literature has mainly focused on the cigarettes and (hard) liquor relationship with inconclusive findings on substitution or complementarity. Our results show cigarettes and beer serve as complements as supported through beer price (tax) and nonprice (regulation) channels. We also find negative and elastic cigarette demand and positive income elasticity. Border effects, both intranational and international, as well as habit-formation effects are significant, while the effects of cigarette advertising and income inequality are insignificant. Policy implications are discussed. Published by Elsevier B.V.
1. Introduction The demand for addictive products has interested scholars and lawmakers for quite some time; in particular, cigarettes and alcohol have garnered quite a bit of attention in the literature.1 Overtime, the literature has recognized the spillover effects associated with smuggling activities.2 These spillovers, however, are potentially multidimensional. For instance, spillovers can be (i) geographic whereby smuggling (both casual and organized) takes place across jurisdictions to exploit price differentials (mainly due to excise tax differences); or (ii) they could be driven by cross-product effects, in which demand changes in one product, via tax/regulatory changes (Fleenor, 1998; Warner, 1982) or socio-economic factors (Aristei and Pieroni, 2009) have an impact on the demand for other products (via smuggling or substitution). Obviously, as noted by Lanoie and Leclair (1998), Gallet (1999), and Bates et al. (2015), lawmakers need a careful accounting of all spillovers in order to design effective cessation and taxation policies. This study provides a state-level analysis of cigarette demand in the U.S., focusing on the spillovers from the beer market. This ☆ We would like to thank two referees for useful comments. ⁎ Corresponding author. E-mail addresses: [email protected] (R.K. Goel), [email protected] (J.E. Payne), [email protected] (J.W. Saunoris). 1 See Gallet and List (1998), Chaloupka and Warner (2000), and U.S. Department of Health and Human Services (2000). 2 See ACIR (1985); Baltagi and Levin (1986); Coats (1995), and Thursby and Thursby (2000).
http://dx.doi.org/10.1016/j.econmod.2016.02.022 0264-9993/Published by Elsevier B.V.
interdependence takes into account the abovementioned spillover effects. Specifically, we examine the cross-price elasticities of cigarette demand with regard to beer taxes/prices and the effect of border prices to account for geographic spillovers. The related literature has almost exclusively focused on the interdependence between cigarettes and hard liquor without a clear cut finding of substitution or complementarity across samples from various countries.3 In contrast, the present work focuses on cigarette demand and its responsiveness to the beer market (i.e., prices and regulations). In recent years, beer drinking in the U.S. has been increasing while consumption of hard liquor has been decreasing.4 Furthermore, media advertising of beer seems more acceptable and prevalent than hard liquor advertising (although the Internet has undermined media restrictions). Finally, although both smoking and drinking are addictive, given the qualitative differences in their secondary effects, each face different regulations on consumption, sale/marketing, and transportation. For example, (i) unlike cigarettes, there are restrictions on the sale of alcohol as it may not be sold on a particular day or certain times during the day; (ii) the transportation of cigarettes is relatively free whereas beer/alcohol cannot be transported in opened containers or opened multipacks; and (iii)
3 See Fogarty (2010) for a literature survey; Goel and Morey (1995) for the U.S.; Pierani and Tiezzi (2009) for Italy; and Tauchmann et al. (2013) for Germany; also see Clements et al. (2010). 4 See http://www.gallup.com/poll/174074/beer-americans-adult-beverage-choice-year. aspx, https://www.brewersassociation.org/statistics/national-beer-sales-production-data/
R.K. Goel et al. / Economic Modelling 55 (2016) 292–297
alcohol does not face government-mandated restrictions like the ban on cigarette broadcast advertising (see Gallet, 1999). Indeed, these regulations have differing effects on the demand for the two products and, consequently, on related spillovers. Section 2 describes the model to be employed and the data. Section 3 discusses the empirical results while Section 4 provides concluding remarks. 2. Model and data We begin by following the literature in specifying the basic model for cigarette demand (see, for example, Chaloupka and Warner, 2000), with the main novelty lying in consideration of beer market spillovers, in general form as follows: Cit ¼ f PCit ; INCit ; PBit ; BPCit ; BPBit ; CANADAi ; MEXICOi
ð1Þ
where i = 1,…,48 represents the 48 contiguous U.S. states and t = 2005,…,2014 denotes the time period of our analysis (Alaska and Hawaii are excluded because they do not have any U.S. contiguous states). Further, superscripts C and B, respectively, denote cigarettes and beer. Cigarette consumption, C, is cigarette sales (20-packs) per capita; cigarette price, PC, is the average retail cigarette price (cents/pack); and INC denotes personal disposable income per capita in thousands of dollars. As such, it is hypothesized that higher cigarette prices lower consumption, while greater income makes cigarettes more affordable. In addition, related studies typically add one or two other controls depending upon their focus.5 In our case the focus is on cigarette–beer demand interdependence and related border spillovers. Obtaining data on the beer market, however, is considerably more challenging. Therefore, in the absence of readily available cross-state retail beer prices, we proxy beer prices, PB, by state beer taxes.6 We use beer taxes, both own, PB, and in border states, BPB, to determine whether beer market price affects cigarette demand. Higher beer prices (taxes) would reduce cigarette demand if the two products are viewed as complements in consumption. In addition, given the somewhat stringent regulations in the beer market, some smokers who would normally consider beer and cigarettes as complements may be dissuaded from purchasing cigarettes due to the increasing transaction costs. We consider both the price and non-price (regulatory) effects of the beer market on cigarette demand. The geographic effects are incorporated by including cigarette prices in border states, BPC, and by identifying states with foreign borders. For instance, Maine shares its U.S. border with New Hampshire and international border with Canada.7 Cigarettes are taxed at the federal, state, and sometimes even at the local level (Orzechowski and Walker, 2014). While federal excise taxes apply uniformly to all states, there are substantial cross-state differentials in other taxes and these differentials provide inducements for individual smokers and organized crime syndicates to engage in trafficking, as discussed by ACIR (1985), Fleenor (1998), and Warner (1982).8 Higher border prices/taxes would increase a state's consumption or sales. The variables, BPC and BPB, are the spatial
5 See Chaloupka and Warner (2000) and U.S. Department of Health and Human Services (2000) for extensive literature reviews. 6 In the absence of consistent state-level data on beer prices, we proxy beer prices with state beer taxes. Further, we follow the literature in taking cigarette sales to denote cigarette consumption. Finally, while the related data are available for additional years, our choice of the sample period is partly driven by capturing the post-WHO Framework Convention on Tobacco Control period (http://www.who.int/fctc/en/). 7 Since Alaska and Hawaii do not have any contiguous U.S. border states, they were dropped from the analysis. 8 In terms of non-price regulatory variations, some regulations such as requirements relating to health warning labels on cigarette packages and broadcast advertising bans are uniform across states, while public place smoking bans vary across states and in many instances are even imposed by local governments (see Goel, 2013).
293
lags of PC and PB, respectively.9 Following the spatial econometrics literature, we use two weight matrices based on geographic distance to define “neighborliness”—contiguity and inverse distance (Anselin, 1988). To construct the N × N (48 × 48) spatial weight matrix, we compute the ijth element for contiguous neighbors as wij = 1 if state i and j share a land border and zero otherwise, and inverse distance weights are calculated as wij ¼ d1ij where d is the Euclidean distance between state i and state j. The full weight matrix is an NT × NT block diagonal matrix with T (number of time periods) copies of the N × N matrix along the diagonal. Each NT × NT weight matrix is pre-multiplied by the variable of interest to create its spatial lag. Contiguity is likely to capture both casual (by commuters or consumers crossing state borders on weekends) and organized (by crime syndicates) smuggling, while inverse distance would mainly capture organized smuggling across more distant states. CANADA and MEXICO, respectively, identify states sharing foreign borders with Canada and Mexico (Connelly et al., 2009). Specifically, CANADA is defined by a dummy variable equal to 1.0 for states bordering Canada: Idaho, Maine, Michigan, Minnesota, Montana, New Hampshire, New York, North Dakota, Vermont, and Washington. MEXICO is defined by a dummy variable equal to 1.0 for states bordering Mexico: Arizona, California, New Mexico, and Texas. The two foreign borders of the United States are somewhat qualitatively different with the Canadian border being substantially longer but much more porous.10 As discussed previously, the broader literature on the cigarette–alcohol relation has failed to find a robust relation. Hopefully, our analysis will shed light on the interdependence. The variables defined above were obtained from a variety of sources. The cigarette sales per capita and average retail cigarette price were obtained from the Tax Burden on Tobacco. Personal disposable income per capita is from the U.S. Bureau of Economic Analysis, 2014 and state beer tax representing price from the Tax Foundation, 2015. The border state prices for cigarettes and beer are constructed as noted above. Details about the variables, summary statistics, and data sources are provided in Table 1. 3. Empirical results We use the two-step efficient GMM to estimate Eq. (1), which provides efficient estimates in the presence of unknown forms of heteroskedasticity (Baum et al., 2003). Because PC and spatial lags (BPB and BPC) are likely endogenous, we instrument these variables using cigarette taxes, CT, and spatial lags (up to the third order) of the exogenous variables, i.e., INC and CT. Cigarette taxes are a significant component of cigarette prices and many states frequently raise these excise taxes to raise revenues and control smoking (Orzechowski and Walker, 2014). State cigarette tax data were obtained from Tax Burden on Tobacco. Rejection of the Kleibergen and Paap (2006) rk LM statistic and insignificance of the Hansen's J statistic support this instrument choice. 3.1. Price spillovers from the beer market to U.S. cigarette demand Table 2 reports the baseline results associated with Eq. (1) along with two measures of border state spillover effects using spatial contiguity and inverse distance. Given the logarithmic form of key variables, the corresponding coefficients represent elasticities. Consistent with theory, the price elasticity of cigarette demand is negative, while the income elasticity is positive. The coefficient estimates are fairly stable
9 See Gallet (2006) for an alternate spatial focus that accounts for health information and supply aspects. 10 We are considering only foreign land borders, although some smuggling from/to nations in close proximity to the United States (e.g., Bahamas and, with the lifting of the trade embargo, Cuba) might also be taking place.
294
R.K. Goel et al. / Economic Modelling 55 (2016) 292–297
Table 1 Variable definitions, summary statistics and data sources. Variable
Description [mean; standard deviation; (minimum, maximum)]
Log of cigarette sales (20-packs) per capita. [59.77; 26.39; (15.4, 185.2)]a Log of average retail cigarette price (cents/pack). [225.17; 50.41; (149.22, 429.16)]a BPC Spatial lag of PC. Contiguity weights: [224.04; 43.32; (159.96, 379.25)]; Inverse distance weights: [225.42; 30.06; (183.23, 302.38)]a B P Log of state beer tax (dollars/gal). [3.72; 2.96; (0.26, 15.57)]b BPB Spatial lag of PB. Contiguity weights: [3.63; 2.34; (1.33, 14.88)]; Inverse distance weights: [3.70; 0.57; (2.89, 5.66)]b CT Log of state cigarette tax (cents/pack). [54.61; 36.96; (2.56, 193.39)]a INC Log of personal disposable income per capita (000s of dollars). [16.41; 2.19; (12.21, 22.31)]c INEQUALITY State income inequality; measured using a variant of the Gini coefficient, with higher values denoting greater inequality. Data available up to 2013 [0.61; 0.03; (0.54, 0.71)]d CIGADV Log of total cigarette advertising expenditures in dollars per capita. Data are aggregate across states and available up to 2012. [1.95; 2.19; (0.10, 13.06)]e BeerREG1 Dummy variable equal to one if the state prohibits grocery store beer sales, zero otherwise. These states include Alaska, Delaware, North Dakota, Pennsylvania, Rhode Island, and Wyoming.f BeerREG2 Dummy variable equal to one if the state is an alcohol beverage control state. These states include Alabama, Maryland, Minnesota, and Utah.f Male Percent of population that is male. [49.28; 0.64; (48.23, 51.25)]g Youth Percent of population under the age of 19. [13.69; 1.00; (11.30, 17.83)]g Non-white Percent of state population that is non-white. [16.74; 9.45; (2.71, 39.78)]g CANADA Dummy variable for states bordering Canada: Idaho, Maine, Michigan, Minnesota, Montana, New Hampshire, New York, North Dakota, Vermont, Washington. MEXICO Dummy variable for states bordering Mexico: Arizona, California, New Mexico, Texas. C PC
Notes: Summary statistics (based on raw data) use annual observations for 48 contiguous states (2005–2014). Observations = 480. CPI (1982–84 = 100) used to deflate monetary variables. Data sources: a Tax Burden on Tobacco. b Tax Foundation, 2015 c U.S. Bureau of Economic Analysis, 2014. d Frank (2009). e Federal Trade Commission Cigarette Report for 2012, 2015. f https://en.wikipedia.org/wiki/Alcohol_laws_of_the_United_States. g http://wonder.cdc.gov/bridged-race-v2014.html.
across the alternate spatial weights (i.e., contiguity and inverse distance). However, as noted by Goel and Nelson (2012), in recent years, cigarette demand has become elastic, and this is borne out by our results. A number of factors are likely contributing to this greater priceresponsiveness of cigarette demand, including health awareness, smoking restrictions in public places, and availability of substitute products (smuggled or contraband cigarettes). Quantitatively, a 10% increase in cigarette prices would reduce consumption by 13%, while similar increases in income would raise cigarette consumption by 9%. The positive coefficient on BPC is consistent with the notion that higher border prices increase a state's cigarette sales via organized or casual smuggling. The border cigarette price effects are about three times more pronounced with inverse distance (Models 2.3–2.4) than with contiguity (Models 2.1–2.2), signifying that spillovers from cigarette smuggling extend beyond a state's immediate borders.11 Turning our focus to the demand interdependence between cigarettes and beer, the two appear as complements, signifying that consumers often consume the two together. However, the magnitude of the effect is rather modest—a 10% increase in PB decreases C by about
11 Interestingly, a 10% increase in cigarette prices across own and border states would reduce cigarette consumption by approximately 8% (=10*(−1.266 + 0.456)) (Model 2.1); however, when one considers the effect of a 10% increase over a broader geographic region using inverse distance the combined effect becomes statistically insignificant. We thank a referee for pointing this out.
Table 2 Spillovers from the beer market to U.S. cigarette demand: baseline models. Dependent variable: cigarette consumption, C. Independent variables
P
C
PB BPC BPB INC
Contiguity (2.2)
(2.3)
(2.4)
−1.266⁎⁎⁎
−1.266⁎⁎⁎ (0.087) −0.031⁎⁎
−1.250⁎⁎⁎
−1.250⁎⁎⁎ (0.073) −0.022⁎⁎
(0.016) 0.456⁎⁎⁎
(0.016) 0.456⁎⁎⁎
(0.011) 1.335⁎⁎⁎
(0.011) 1.335⁎⁎⁎
(0.104) 0.247⁎⁎⁎ (0.070) 0.865⁎⁎⁎ (0.093)
(0.104) 0.247⁎⁎⁎ (0.070) 0.865⁎⁎⁎ (0.093) −0.422⁎⁎⁎
(0.344) 0.108 (0.194) 0.888⁎⁎⁎ (0.093)
(0.344) 0.108 (0.194) 0.888⁎⁎⁎ (0.093) −0.745⁎⁎⁎
(0.087) −0.031⁎⁎
CANADA
Kleibergen–Paap rk Wald F statistic Hansen's J statistic
(0.073) −0.022⁎⁎
(0.118) −0.534⁎⁎⁎ (0.043)
MEXICO
Observations R-squared Kleibergen–Paap rk LM statistic
Inverse distance
(2.1)
(0.078) −0.593⁎⁎⁎ (0.035)
480 480 480 480 0.981 0.981 0.986 0.986 3.044⁎⁎⁎ 3.044⁎⁎⁎ 5.750⁎⁎⁎ 5.750⁎⁎⁎ [0.004] [0.004] [0.000] [0.000] 17.11 17.11 37.82 37.82 6.580 [0.160]
6.580 [0.160]
1.054 [0.901]
1.054 [0.901]
Notes: Time and state dummies and a constant are included but not reported. Two-step GMM estimates are reported with robust standard errors in parentheses and probability values in brackets. The critical values for the Kleibergen–Paap rk Wald F statistic based on test size and bias are in Stock and Yogo (2005). Excluded instruments for PC, BPB, and BPc include CT and spatial lags up to the third order of INC and CT. ⁎ Significance at p b 0.1 level. ⁎⁎ Significance at p b 0.05 level. ⁎⁎⁎ Significance at p b 0.01 level.
0.3 to 0.4%. Furthermore, border beer prices, BPB, significantly affect cigarette sales, especially in contiguous states (Models 2.1 and 2.2). Higher beer prices in bordering states increase a state's cigarette sales as consumers crossing state borders to take advantage of lower beer prices also take the opportunity to purchase complementary cigarettes.12 Interestingly, the magnitude of the border beer price effect on cigarette consumption is about six to seven times greater than own beer price effect (comparing the coefficients on PB and BPB in Models 2.1 and 2.2). It could be the case that bordering residents that cross the border to buy cheaper cigarettes are also buying beer (and vice versa).13 Finally, states that border Canada and Mexico (Model 2.2 and 2.4) have lower cigarette consumption, either due to substitution from cheaper smuggled cigarettes or greater monitoring. These findings are generally in line with ones with earlier data (Connelly et al., 2009); however, the significant (negative) effect of the Canadian border is a new revelation (Model 2.4). Our alternate consideration of border weights uniquely provides some interesting insights. The coefficient on BPB is significant only in contiguous states, while the magnitude of BPC is about a third smaller in contiguous states (Models 2.1, 2.2, versus 2.3, 2.4). Understandably, the border effects of cigarette prices are more pronounced than those of beer prices. In light of the differential transportation restrictions, cross-border beer smuggling is mainly across immediate neighbors, while that of cigarettes can span across distant states as well (Lovenheim, 2008). 12 The generality of these findings to other nations must await availability of comparable data. 13 Availability of individual survey level data would enable one to account for other important individual attributes (education, employment, marital status, gender, religion, etc.) that might crucially dictate smoking and drinking behaviors.
R.K. Goel et al. / Economic Modelling 55 (2016) 292–297
3.2. Price and non-price spillovers from the beer market to U.S. cigarette demand To check the robustness of our findings and to broaden the focus of the present work by incorporating non-price (regulatory) effects of the beer market along with price effects, we conduct a series of robustness checks. These include considering habit-formation effects, advertising, and income inequality. All models in Table 3 include the two regulatory variables, BeerREG1 and BeerREG2, and the estimated Eq. (1) now has the following expanded form Cit ¼ g PCit ; INCit ; PBit ; BPCit ; BPBit ; CANADAi ; MEXICOi ; BeerREG1; BeerREG2
ð2Þ Together with beer taxes, these regulatory variables might be seen as accounting for both price and non-price spillovers from the beer market to cigarette demand. In choosing the set of beer regulations to consider, we settled on a set that could be readily and consistently quantified across states. Specifically, these include (i) BeerREG1: a
Table 3 Spillovers from the beer market to U.S. cigarette demand: robustness checks. Dependent variable: cigarette consumption, C. Independent variables
Beer sales Habit regulations effects
Cigarette Income advertising inequality
(3.1)
(3.2)
(3.3)
(3.4)
BPC
−1.266⁎⁎⁎ (0.087) −0.031⁎⁎ (0.016) 0.456⁎⁎⁎
−0.954⁎⁎⁎ (0.119) −0.020⁎ (0.012) 0.452⁎⁎⁎
−1.332⁎⁎⁎ (0.107) −0.046 (0.032) 0.428⁎⁎⁎
−1.292⁎⁎⁎ (0.093) −0.037⁎⁎ (0.019) 0.426⁎⁎⁎
BPB
(0.104) 0.247⁎⁎⁎
(0.096) 0.102 (0.085) 0.653⁎⁎⁎ (0.114) −0.254⁎⁎⁎ (0.071) −0.278⁎⁎⁎
(0.140) 0.300⁎⁎⁎
(0.115) 0.270⁎⁎⁎
(0.081) 0.822⁎⁎⁎ (0.107) −0.259⁎⁎⁎ (0.097) −0.460 (0.360) −0.236 (0.582) 0.063 (0.129)
(0.072) 0.882⁎⁎⁎ (0.108) −0.307⁎⁎⁎ (0.072) −0.475⁎⁎⁎
PC PB
INC CANADA MEXICO BeerREG1 BeerREG2
(0.070) 0.865⁎⁎⁎ (0.093) −0.340⁎⁎⁎ (0.071) −0.452⁎⁎⁎ (0.056) −0.180⁎⁎⁎ (0.052) 0.082 (0.077)
C (lagged)
(0.069) −0.142⁎⁎⁎ (0.050) 0.114 (0.081) 0.334⁎⁎⁎
(0.066) −0.207⁎⁎⁎ (0.065) 0.068 (0.083)
(0.089) CIGADV
0.012 (0.248)
INEQUALITY Observations R-squared Kleibergen–Paap rk LM statistic Kleibergen–Paap rk Wald F statistic Hansen's J statistic
480 0.981 3.044 [0.004] 17.11
384 0.989 0.994 [0.159] 7.946
384 0.982 2.863 [0.008] 15.71
0.223 (0.185) 432 0.981 3.322 [0.005] 16.57
6.580 [0.160]
5.507 [0.239]
7.678 [0.104]
5.643 [0.227]
Notes: Time and state dummies and a constant are included but not reported. The spatial aspects considered are contiguous states. Two-step GMM estimates are reported with robust standard errors in parentheses and probability values in brackets. The critical values for the Kleibergen–Paap rk Wald F statistic based on test size and bias are in Stock and Yogo (2005). Excluded instruments for PC, BPB, and BPC include CT and spatial lags up to the third order of Income and CT. C (lagged) is the one year time lag of C and is instrumented using the year two time lag of C. ⁎ Significance at p b 0.1 level. ⁎⁎ Significance at p b 0.05 level. ⁎⁎⁎ Significance at p b 0.01 level.
295
dummy variable equal to 1.0 identifying states that prohibit grocery store beer sales. These states include Alaska, Delaware, North Dakota, Pennsylvania, Rhode Island, and Wyoming; and (ii) BeerREG2: a dummy variable equal to 1.0 if the state is an alcohol beverage control state. These states include Alabama, Maryland, Minnesota, and Utah. The results in Model 3.1 show the effect of BeerREG1 to be negative and significant, while the effect of BeerREG2 was statistically insignificant. Thus, restrictions on grocery store beer sales have negative spillovers on cigarette demand and this finding reinforces the complementarity between beer and cigarettes.14 Furthermore, the reinforcing price and non-price spillovers from the beer market is a significant finding that merits consideration by policymakers. The addictive nature of cigarettes has been well-recognized and has led to the emphasis on early interventions in smoking cessation programs. To account for habit-formation effects, we include lagged cigarette consumption as a regressor in the Model 3.1, and the corresponding results are in Model 3.2 of Table 3.15 The two-step GMM estimation procedure is again employed and the excluded instruments for PC, BPB, and BPC include CT and spatial lags up to the third order of INC and CT; and C (lagged) is instrumented using the year two time lag of C.16 The Hansen's J statistic is statistically insignificant and the Kleibergen–Paap rk LM statistic is significant. These suggest that the instruments are largely valid.17 The coefficient on lagged cigarette consumption is positive and statistically significant, reinforcing earlier findings about the presence of a habit formation effect (Baltagi and Levin, 1986). Both price and income elasticities of cigarette demand retain the earlier signs and statistical significance, albeit with somewhat smaller magnitudes. The complementarity between cigarettes and beer holds and is significant. Further, as before, the spatial spillovers from border cigarette prices are significant. However, the effects of border beer prices are not significant, and the effect of beer regulations, BeerREG1, is negative and significant as in Model 3.1. The results with regard to foreign borders are in line with the results in Table 2. Cigarette advertising can increase cigarette demand or at least counter the influences of anti-smoking campaigns. However, over the years, various regulatory mandates have changed the advertising avenues (e.g., broadcast ban on cigarette advertising), while technical changes have opened new advertising avenues (e.g., the Internet and video games). We consider the effects of cigarette advertising by including real per capita cigarette advertising expenditures (CIGADV) obtained from the Federal Trade Commission Cigarette Report for 2012, 2015 in Model 3.3. The resulting coefficient, while positive, failed to achieve statistical significance. This insignificance might partly be due to a single measure failing to account for qualitatively different types of advertising and partly for the lack of consideration of the lagged structure of advertising (Goel, 2011).18 These aspects, while potentially important, seem tangential to the focus of the present work. A noteworthy result, however, is that in this case, both beer price and non-price effects are significant. Besides the level of income, its dispersion or inequality might also affect the demand for cigarettes (Connelly et al., 2010). Greater income inequality might make cigarettes unaffordable for some, inducing them
14 Stehr (2007) found that states that lifted bans on Sunday alcohol sales experienced an increase in sales. 15 See Baltagi and Levin (1986) and Aristei and Pieroni (2009) for a similar treatment of habit effects. An alternate strand of this literature considers rational addiction by including lead consumption – see, for example, Bask and Melkersson (2004). 16 Besides consumption, cross-ownership and co-sponsorship of marketing activities by cigarette and beer companies would affect retail prices of both products (Jiang and Ling, 2011). Investigation of comprehensive holdings of all companies and their subsidiaries is beyond the scope of the current research. 17 The use of the lagged dependent variable as a regressor complicates the instrument choice and we tried several different variations. 18 The effects of advertising might also vary significantly across nations—see, for example, Clements et al. (1985) and McLeod (1986).
296
R.K. Goel et al. / Economic Modelling 55 (2016) 292–297
to switch to cheaper substitutes. Further, low-income smokers might smoke more when they view smoking as complementary to leisure. We follow Frank (2009) to measure income inequality by using a variant of the Gini coefficient with higher values denoting greater income inequality. Our results fail to find a significant effect of income inequality (INEQUALITY) on cigarette demand (Model 3.4), and this finding is in line with previous cross-section results for the United States (Connelly et al., 2010).19 3.3. Additional considerations: Demographic factors and time effects Table 4 incorporates demographic aspects by including age (Youth), gender (Male) and race (Non-white). These factors vary across states and can have a significant effect on smoking (see U.S. Department of Health and Human Services, 2000). Variants of Models 2.1 and 3.1, incorporating demographic aspects, are presented in Models 4.1 and 4.3. Results show that males had higher cigarettes consumption, ceteris paribus. The effects of youth and race were statistically insignificant. Further, to gain insights into time patterns of smoking, Models 4.2 and 4.4 explicitly report the coefficients on year dummies from Models 2.1 and 3.1 (year 2005 is excluded). All the time dummies were negative and statistically significant. In terms of magnitude, the absolute value of time effects was higher over time, signifying a decreasing trend in smoking. Overall, it seems reasonable to conclude that the baseline results are qualitatively robust to considerations of smoking habit formation effects, advertising, and income inequality.20 The non-price effects, especially those captured by BeerREG1, are negative and significant in most cases.
Table 4 Spillovers from the beer market to U.S. cigarette demand: demographic factors and time effects. Dependent variable: cigarette consumption, C. Independent variables
(4.1)
(4.2)
(4.3)
(4.4)
−1.260⁎⁎⁎
−1.260⁎⁎⁎
BPC
(0.088) −0.035⁎⁎ (0.016) 0.422⁎⁎⁎
−1.266⁎⁎⁎ (0.087) −0.031⁎⁎ (0.016) 0.456⁎⁎⁎
(0.088) −0.035⁎⁎ (0.016) 0.422⁎⁎⁎
−1.266⁎⁎⁎ (0.087) −0.031⁎⁎ (0.016) 0.456⁎⁎⁎
BPB
(0.120) 0.258⁎⁎⁎
(0.104) 0.247⁎⁎⁎
(0.120) 0.258⁎⁎⁎
(0.104) 0.247⁎⁎⁎
(0.063) 0.601⁎⁎⁎ (0.149) −0.400 (0.268) −0.472⁎⁎
(0.070) 0.865⁎⁎⁎ (0.093) −0.422⁎⁎⁎
(0.063) 0.601⁎⁎⁎ (0.149) −0.414⁎⁎⁎
(0.070) 0.865⁎⁎⁎ (0.093) −0.340⁎⁎⁎
(0.118) −0.534⁎⁎⁎ (0.043)
(0.073) −0.486⁎⁎⁎ (0.127) 0.178⁎⁎⁎ (0.054) −0.038 (0.031) 0.014 (0.014) −0.284⁎⁎⁎
(0.071) −0.452⁎⁎⁎ (0.056)
P
C
PB
INC CANADA MEXICO Male Youth Non-white BeerREG1 BeerREG2 Year (2006) Year (2007) Year (2008) Year (2009)
4. Concluding remarks This study adds to the literature by examining demand interdependence between cigarette demand and the beer market, and by studying geographic and cross-product spillovers. Our analysis using data for the 48 contiguous U.S. states from 2005 to 2014 shows cigarette demand price elasticity to be negative and price elastic, and the income elasticity to be positive. It seems that repeated cigarette excise tax increases over the years have moved cigarette demand to the elastic portion of the demand curve. Further, in line with the literature (Baltagi and Levin, 1986), cigarette demand shows some inertia from habit formation, while the effects of advertising and income inequality are insignificant. Finally, both price and non-price effects of the beer market have a significant (negative) bearing on the cigarette demand. In regard to cross-product spillovers, cigarettes and beer turn out to be complements.21 The broader literature on smoking-drinking interdependence has failed to find a conclusive relation. For instance, Goel and Morey (1995) find cigarettes and liquor to be substitutes for the U.S., while Pierani and Tiezzi (2009) find complementarity between cigarettes and alcohol in the case of Italy. This result has obvious value for policy coordination in reducing the social costs from these products (Bates et al., 2015). 19 In a study using Canadian data, Latif (2014) alternately considers the effect of economic recessions on smoking and drinking. 20 We also considered extensions to take account of institutional and regulatory differences across states. For instance, we considered accounting for the four states (Florida, Minnesota, Mississippi, and Texas) that were not part of the Master Settlement Tobacco Agreement and entered separate deals with tobacco companies. They might spent settlement dollars differently and with fewer restrictions. A dummy variable identifying these states, when used as an additional regressor, produced similar results. These results are not reported but are available upon request. 21 Given comparable data on beer consumption and price, one could also estimate a beer demand equation and see whether beer consumers view cigarettes as complementary. Specifically, for now, consistent data on retail beer prices across states are not readily available in the public domain. Further, availability of micro-level data would enable consideration personal attributes of consumers.
(0.199) 0.178⁎⁎⁎ (0.054) −0.038 (0.031) 0.014 (0.014)
Year (2010) Year (2011) Year (2012) Year (2013)
−0.028⁎⁎ (0.012) −0.074⁎⁎⁎
(0.090) −0.014 (0.307)
−0.180⁎⁎⁎ (0.052) 0.082 (0.077) −0.028⁎⁎
(0.013) −0.112⁎⁎⁎ (0.012) −0.147⁎⁎⁎ (0.013) −0.092⁎⁎⁎
(0.012) −0.074⁎⁎⁎ (0.013) −0.112⁎⁎⁎ (0.012) −0.147⁎⁎⁎ (0.013) −0.092⁎⁎⁎
(0.035) −0.121⁎⁎⁎ (0.038) −0.188⁎⁎⁎ (0.036) −0.256⁎⁎⁎
(0.035) −0.121⁎⁎⁎ (0.038) −0.188⁎⁎⁎ (0.036) −0.256⁎⁎⁎
(0.034) (0.034) −0.304⁎⁎⁎ −0.304⁎⁎⁎ (0.036) (0.036) Observations 480 480 480 480 R-squared 0.981 0.981 0.981 0.981 Kleibergen–Paap rk LM statistic 4.524⁎⁎⁎ 3.044⁎⁎⁎ 4.524⁎⁎⁎ 3.044⁎⁎⁎ [0.001] [0.004] [0.001] [0.004] Kleibergen–Paap rk Wald F 20.54 17.11 20.54 17.11 statistic Hansen's J statistic 7.853⁎ 6.580 7.853⁎ 6.580 [0.097] [0.160] [0.097] [0.160] Year (2014)
Notes: Time and state dummies and a constant are included in all models, and time effects are reported in Models 4.2 and 4.4. The spatial aspects considered are contiguous states. Two-step GMM estimates are reported with robust standard errors in parentheses and probability values in brackets. The critical values for the Kleibergen–Paap rk Wald F statistic based on test size and bias are in Stock and Yogo (2005). Excluded instruments for PC, BPB, and BPc include CT and spatial lags up to the third order of INC and CT. ⁎ Significance at p b 0.1 level. ⁎⁎ Significance at p b 0.05 level. ⁎⁎⁎ Significance at p b 0.01 level.
Finally, turning to geographic spillovers, the border effects are studied using alternate spatial weights, taking account of both domestic and foreign borders. The effects of both neighboring cigarette prices and neighboring beer prices are significant. The spatial effects of cigarette prices might extend beyond immediate neighbors. This revelation seems new to the extant literature. Some of the differences between the relative geographic effects might be driven by the nature of the two products, related transportation and storage costs, and
R.K. Goel et al. / Economic Modelling 55 (2016) 292–297
regulation, such as varying minimum age restrictions for consumption and purchase.22 While cross-border smuggling of tobacco products has been recognized for quite some time (see ACIR, 1985), the problem does not seem to have yet been contained. There are several implications of our findings for public policy. First, the significance of the habit effects reinforces the need for early intervention programs in preventing smoking initiation. Second, the relatively elastic cigarette demand increases possibilities of smoking reduction through excise taxes, while undermining their revenuegenerating potential. Third, the complementarity between cigarettes and beer suggests policy coordination to internalize cross-product externalities. While earlier findings for the U.S. show liquor and cigarettes to be substitutes (Goel and Morey, 1995), this study finds beer and cigarettes to be complements. Fourth, policymakers should account for both price (tax) and non-price (regulatory) spillovers from the beer market to the cigarette market. Both policies, as in our analysis, could be reinforcing each other. Finally, policies to check cross-border spillovers need to pay attention to both domestic and international borders and to the prices and regulations of related products (in this case beer). These policies are relatively more difficult to coordinate across international borders. References Advisory Commission on Intergovernmental Relations (ACIR), 1985. Cigarette Tax Evasion: A Second Look. Advisory Commission on Intergovernmental Relations, Washington, DC. Anselin, L., 1988. Spatial Econometrics: Methods and Models. Kluwer Academic Publishers, Dordrecht. Aristei, D., Pieroni, L., 2009. Addiction, social interactions and gender differences in cigarette consumption. Empirica 36, 245–272. Baltagi, B.H., Levin, D., 1986. Estimating dynamic demand for cigarettes using panel data: the effects of bootlegging, taxation and advertising reconsidered. Rev. Econ. Stat. 68, 148–155. Bask, M., Melkersson, M., 2004. Rationally addicted to drinking and smoking? Appl. Econ. 36, 373–381. Bates, L.J., Cesur, R., Santerre, R.E., 2015. Short-run marginal medical costs from booze and butts: evidence from the states, South. Econ. J. 81, 1074–1095. Baum, C.F., Schaffer, M.E., Stillman, S., 2003. Instrumental variables and GMM: estimation and testing. Stata J. 3, 1–31. Chaloupka, F.J., Warner, K.E., 2000. The economics of smoking. In: Culyer, A.J., Newhouse, J.P. (Eds.), Handbook of Health Economics vol. 1. North-Holland, Amsterdam, pp. 1539–1627. Clements, K.W., McLeod, P.B., Selvanathan, E.A., 1985. Does advertising affect drinking and smoking? Department of Economics, University of Western Australia, Discussion Paper # 85.02 January Clements, K.W., Lan, Y., Zhao, X., 2010. The demand for marijuana, tobacco and alcohol: inter-commodity interactions with uncertainty. Empir. Econ. 39, 203–239. Coats, R.M., 1995. A note on estimating cross-border effects of state cigarette taxes. Natl. Tax J. 48, 573–584. Connelly, R.T., Goel, R.K., Ram, R., 2009. Demand for cigarettes in the United States: effects of prices in bordering states and contiguity with Mexico and Canada. Appl. Econ. 41, 2255–2260.
22
https://www.dor.ms.gov/info/faqs/TobaccoBeerandAlcohol.html
297
Connelly, R.T., Goel, R.K., Ram, R., 2010. Income inequality and cigarette consumption: evidence from the United States. Appl. Econ. Lett. 17, 423–426. Federal Trade Commission Cigarette Report for 2012, 2015. issued https://www.ftc.gov/ reports/federal-trade-commission-cigarette-report-2012. Fleenor, P., 1998. How excise tax differentials affect interstate smuggling and cross-border sales of cigarettes in the United States. Tax Found. Background Pap. 26. Fogarty, J., 2010. The demand for beer, wine and spirits: a survey of the literature. J. Econ. Surv. 24, 428–478. Frank, M., 2009. Inequality and growth in the United States: evidence from a new statelevel panel of income inequality measures. Econ. Inq. 47, 55–68. Gallet, C.A., 1999. The effect of the 1971 advertising ban on behavior in the cigarette industry. Manag. Decis. Econ. 20, 299–303. Gallet, C.A., 2006. Health information and cigarette consumption: supply and spatial considerations. Empirica 33, 35–47. Gallet, C.A., List, J.A., 1998. Elasticities of beer demand revisited. Econ. Lett. 61, 67–71. Goel, R.K., 2011. Advertising media and cigarette demand. Bull. Econ. Res. 63, 404–416. Goel, R.K., 2013. Where to snuff the puff?: relative effectiveness of U.S. smoking control policies. J. Socio-Econ. 44, 97–102. Goel, R.K., Morey, M.J., 1995. The interdependence of cigarette and liquor demand, South. Econ. J. 62, 451–459. Goel, R.K., Nelson, M.A., 2012. Cigarette demand and effectiveness of U.S. smoking control policies: state-level evidence for more than half a century. Empir. Econ. 42, 1079–1095. Jiang, N., Ling, P.M., 2011. Reinforcement of smoking and drinking: tobacco marketing strategies linked with alcohol in the United States. Am. J. Public Health 101, 1942–1954. Kleibergen, F., Paap, R., 2006. Generalized reduced rank tests using the singular value decomposition. J. Econ. 133, 97–126. Lanoie, P., Leclair, P., 1998. Taxation or regulation: looking for a good anti-smoking policy. Econ. Lett. 58, 85–89. Latif, E., 2014. The impact of recession on drinking and smoking behaviours in Canada. Econ. Model. 42, 43–56. Lovenheim, M.F., 2008. How far to the border?: the extent and impact of cross-border casual cigarettes smuggling. Natl. Tax J. 61, 7–33. McLeod, P.B., 1986. Advertising bans, tobacco and cigarette consumption. Econ. Lett. 20, 391–396. Orzechowski, W., Walker, R.C., 2014. The Tax Burden on Tobacco: Historical Compilation vol. 49. Orzechowski and Walker, Arlington, VA. Pierani, P., Tiezzi, S., 2009. Addiction and interaction between alcohol and tobacco consumption. Empir. Econ. 37, 1–23. Stehr, M., 2007. The effect of Sunday sales bans and excise taxes on drinking and crossborder shopping for alcoholic beverages. Natl. Tax J. 60, 85–105. Stock, J.H., Yogo, M., 2005. Testing for weak instruments in linear IV regression. In: Andrews, D.W.K., Stock, J.H. (Eds.), Identification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg. Cambridge University Press, Cambridge, pp. 80–108. Tauchmann, H., Lenz, S., Requate, T., Schmidt, C.M., 2013. Tobacco and alcohol: complements or substitutes? Empir. Econ. 45, 539–566. Tax Foundation, 2015. State Sales, Gasoline, Cigarette, and Alcohol Tax Rates by State. pp. 2000–2014 (Washington, DC). http://taxfoundation.org/article/state-salesgasoline-cigarette-and-alcohol-tax-rates. Thursby, J.G., Thursby, M.C., 2000. Interstate cigarette bootlegging: extent, revenue losses, and effects of federal intervention. Natl. Tax J. 53, 59–77. U.S. Bureau of Economic Analysis, 2014. Statistical Abstract of the United States. U.S. Department of Health and Human Services, 2000. Reducing Tobacco Use: A Report of the Surgeon General. Centers for Disease Control and Prevention, Atlanta, GA. Warner, K.E., 1982. Cigarette excise taxation and interstate smuggling: an assessment of recent activity. Natl. Tax J. 35, 483–490.
Contents lists available at ScienceDirect
Economic Modelling journal homepage: www.elsevier.com/locate/ecmod
Spillovers from the beer market to U.S. cigarette demand☆ Rajeev K. Goel a,⁎, James E. Payne b, James W. Saunoris c a b c
Department of Economics, Illinois State University, Normal, IL 61790-4200, USA J. Whitney Bunting College of Business, Georgia College & State University, Milledgeville, GA 31061, USA Department of Economics, Eastern Michigan University, Ypsilanti, MI 48197, USA
a r t i c l e
i n f o
Article history: Accepted 18 February 2016 Available online xxxx Keywords: Cigarettes Beer Demand Smoking Smuggling Elasticity Regulations U.S.
a b s t r a c t We study the cross-effects of the beer market on U.S. cigarette demand. The extant literature has mainly focused on the cigarettes and (hard) liquor relationship with inconclusive findings on substitution or complementarity. Our results show cigarettes and beer serve as complements as supported through beer price (tax) and nonprice (regulation) channels. We also find negative and elastic cigarette demand and positive income elasticity. Border effects, both intranational and international, as well as habit-formation effects are significant, while the effects of cigarette advertising and income inequality are insignificant. Policy implications are discussed. Published by Elsevier B.V.
1. Introduction The demand for addictive products has interested scholars and lawmakers for quite some time; in particular, cigarettes and alcohol have garnered quite a bit of attention in the literature.1 Overtime, the literature has recognized the spillover effects associated with smuggling activities.2 These spillovers, however, are potentially multidimensional. For instance, spillovers can be (i) geographic whereby smuggling (both casual and organized) takes place across jurisdictions to exploit price differentials (mainly due to excise tax differences); or (ii) they could be driven by cross-product effects, in which demand changes in one product, via tax/regulatory changes (Fleenor, 1998; Warner, 1982) or socio-economic factors (Aristei and Pieroni, 2009) have an impact on the demand for other products (via smuggling or substitution). Obviously, as noted by Lanoie and Leclair (1998), Gallet (1999), and Bates et al. (2015), lawmakers need a careful accounting of all spillovers in order to design effective cessation and taxation policies. This study provides a state-level analysis of cigarette demand in the U.S., focusing on the spillovers from the beer market. This ☆ We would like to thank two referees for useful comments. ⁎ Corresponding author. E-mail addresses: [email protected] (R.K. Goel), [email protected] (J.E. Payne), [email protected] (J.W. Saunoris). 1 See Gallet and List (1998), Chaloupka and Warner (2000), and U.S. Department of Health and Human Services (2000). 2 See ACIR (1985); Baltagi and Levin (1986); Coats (1995), and Thursby and Thursby (2000).
http://dx.doi.org/10.1016/j.econmod.2016.02.022 0264-9993/Published by Elsevier B.V.
interdependence takes into account the abovementioned spillover effects. Specifically, we examine the cross-price elasticities of cigarette demand with regard to beer taxes/prices and the effect of border prices to account for geographic spillovers. The related literature has almost exclusively focused on the interdependence between cigarettes and hard liquor without a clear cut finding of substitution or complementarity across samples from various countries.3 In contrast, the present work focuses on cigarette demand and its responsiveness to the beer market (i.e., prices and regulations). In recent years, beer drinking in the U.S. has been increasing while consumption of hard liquor has been decreasing.4 Furthermore, media advertising of beer seems more acceptable and prevalent than hard liquor advertising (although the Internet has undermined media restrictions). Finally, although both smoking and drinking are addictive, given the qualitative differences in their secondary effects, each face different regulations on consumption, sale/marketing, and transportation. For example, (i) unlike cigarettes, there are restrictions on the sale of alcohol as it may not be sold on a particular day or certain times during the day; (ii) the transportation of cigarettes is relatively free whereas beer/alcohol cannot be transported in opened containers or opened multipacks; and (iii)
3 See Fogarty (2010) for a literature survey; Goel and Morey (1995) for the U.S.; Pierani and Tiezzi (2009) for Italy; and Tauchmann et al. (2013) for Germany; also see Clements et al. (2010). 4 See http://www.gallup.com/poll/174074/beer-americans-adult-beverage-choice-year. aspx, https://www.brewersassociation.org/statistics/national-beer-sales-production-data/
R.K. Goel et al. / Economic Modelling 55 (2016) 292–297
alcohol does not face government-mandated restrictions like the ban on cigarette broadcast advertising (see Gallet, 1999). Indeed, these regulations have differing effects on the demand for the two products and, consequently, on related spillovers. Section 2 describes the model to be employed and the data. Section 3 discusses the empirical results while Section 4 provides concluding remarks. 2. Model and data We begin by following the literature in specifying the basic model for cigarette demand (see, for example, Chaloupka and Warner, 2000), with the main novelty lying in consideration of beer market spillovers, in general form as follows: Cit ¼ f PCit ; INCit ; PBit ; BPCit ; BPBit ; CANADAi ; MEXICOi
ð1Þ
where i = 1,…,48 represents the 48 contiguous U.S. states and t = 2005,…,2014 denotes the time period of our analysis (Alaska and Hawaii are excluded because they do not have any U.S. contiguous states). Further, superscripts C and B, respectively, denote cigarettes and beer. Cigarette consumption, C, is cigarette sales (20-packs) per capita; cigarette price, PC, is the average retail cigarette price (cents/pack); and INC denotes personal disposable income per capita in thousands of dollars. As such, it is hypothesized that higher cigarette prices lower consumption, while greater income makes cigarettes more affordable. In addition, related studies typically add one or two other controls depending upon their focus.5 In our case the focus is on cigarette–beer demand interdependence and related border spillovers. Obtaining data on the beer market, however, is considerably more challenging. Therefore, in the absence of readily available cross-state retail beer prices, we proxy beer prices, PB, by state beer taxes.6 We use beer taxes, both own, PB, and in border states, BPB, to determine whether beer market price affects cigarette demand. Higher beer prices (taxes) would reduce cigarette demand if the two products are viewed as complements in consumption. In addition, given the somewhat stringent regulations in the beer market, some smokers who would normally consider beer and cigarettes as complements may be dissuaded from purchasing cigarettes due to the increasing transaction costs. We consider both the price and non-price (regulatory) effects of the beer market on cigarette demand. The geographic effects are incorporated by including cigarette prices in border states, BPC, and by identifying states with foreign borders. For instance, Maine shares its U.S. border with New Hampshire and international border with Canada.7 Cigarettes are taxed at the federal, state, and sometimes even at the local level (Orzechowski and Walker, 2014). While federal excise taxes apply uniformly to all states, there are substantial cross-state differentials in other taxes and these differentials provide inducements for individual smokers and organized crime syndicates to engage in trafficking, as discussed by ACIR (1985), Fleenor (1998), and Warner (1982).8 Higher border prices/taxes would increase a state's consumption or sales. The variables, BPC and BPB, are the spatial
5 See Chaloupka and Warner (2000) and U.S. Department of Health and Human Services (2000) for extensive literature reviews. 6 In the absence of consistent state-level data on beer prices, we proxy beer prices with state beer taxes. Further, we follow the literature in taking cigarette sales to denote cigarette consumption. Finally, while the related data are available for additional years, our choice of the sample period is partly driven by capturing the post-WHO Framework Convention on Tobacco Control period (http://www.who.int/fctc/en/). 7 Since Alaska and Hawaii do not have any contiguous U.S. border states, they were dropped from the analysis. 8 In terms of non-price regulatory variations, some regulations such as requirements relating to health warning labels on cigarette packages and broadcast advertising bans are uniform across states, while public place smoking bans vary across states and in many instances are even imposed by local governments (see Goel, 2013).
293
lags of PC and PB, respectively.9 Following the spatial econometrics literature, we use two weight matrices based on geographic distance to define “neighborliness”—contiguity and inverse distance (Anselin, 1988). To construct the N × N (48 × 48) spatial weight matrix, we compute the ijth element for contiguous neighbors as wij = 1 if state i and j share a land border and zero otherwise, and inverse distance weights are calculated as wij ¼ d1ij where d is the Euclidean distance between state i and state j. The full weight matrix is an NT × NT block diagonal matrix with T (number of time periods) copies of the N × N matrix along the diagonal. Each NT × NT weight matrix is pre-multiplied by the variable of interest to create its spatial lag. Contiguity is likely to capture both casual (by commuters or consumers crossing state borders on weekends) and organized (by crime syndicates) smuggling, while inverse distance would mainly capture organized smuggling across more distant states. CANADA and MEXICO, respectively, identify states sharing foreign borders with Canada and Mexico (Connelly et al., 2009). Specifically, CANADA is defined by a dummy variable equal to 1.0 for states bordering Canada: Idaho, Maine, Michigan, Minnesota, Montana, New Hampshire, New York, North Dakota, Vermont, and Washington. MEXICO is defined by a dummy variable equal to 1.0 for states bordering Mexico: Arizona, California, New Mexico, and Texas. The two foreign borders of the United States are somewhat qualitatively different with the Canadian border being substantially longer but much more porous.10 As discussed previously, the broader literature on the cigarette–alcohol relation has failed to find a robust relation. Hopefully, our analysis will shed light on the interdependence. The variables defined above were obtained from a variety of sources. The cigarette sales per capita and average retail cigarette price were obtained from the Tax Burden on Tobacco. Personal disposable income per capita is from the U.S. Bureau of Economic Analysis, 2014 and state beer tax representing price from the Tax Foundation, 2015. The border state prices for cigarettes and beer are constructed as noted above. Details about the variables, summary statistics, and data sources are provided in Table 1. 3. Empirical results We use the two-step efficient GMM to estimate Eq. (1), which provides efficient estimates in the presence of unknown forms of heteroskedasticity (Baum et al., 2003). Because PC and spatial lags (BPB and BPC) are likely endogenous, we instrument these variables using cigarette taxes, CT, and spatial lags (up to the third order) of the exogenous variables, i.e., INC and CT. Cigarette taxes are a significant component of cigarette prices and many states frequently raise these excise taxes to raise revenues and control smoking (Orzechowski and Walker, 2014). State cigarette tax data were obtained from Tax Burden on Tobacco. Rejection of the Kleibergen and Paap (2006) rk LM statistic and insignificance of the Hansen's J statistic support this instrument choice. 3.1. Price spillovers from the beer market to U.S. cigarette demand Table 2 reports the baseline results associated with Eq. (1) along with two measures of border state spillover effects using spatial contiguity and inverse distance. Given the logarithmic form of key variables, the corresponding coefficients represent elasticities. Consistent with theory, the price elasticity of cigarette demand is negative, while the income elasticity is positive. The coefficient estimates are fairly stable
9 See Gallet (2006) for an alternate spatial focus that accounts for health information and supply aspects. 10 We are considering only foreign land borders, although some smuggling from/to nations in close proximity to the United States (e.g., Bahamas and, with the lifting of the trade embargo, Cuba) might also be taking place.
294
R.K. Goel et al. / Economic Modelling 55 (2016) 292–297
Table 1 Variable definitions, summary statistics and data sources. Variable
Description [mean; standard deviation; (minimum, maximum)]
Log of cigarette sales (20-packs) per capita. [59.77; 26.39; (15.4, 185.2)]a Log of average retail cigarette price (cents/pack). [225.17; 50.41; (149.22, 429.16)]a BPC Spatial lag of PC. Contiguity weights: [224.04; 43.32; (159.96, 379.25)]; Inverse distance weights: [225.42; 30.06; (183.23, 302.38)]a B P Log of state beer tax (dollars/gal). [3.72; 2.96; (0.26, 15.57)]b BPB Spatial lag of PB. Contiguity weights: [3.63; 2.34; (1.33, 14.88)]; Inverse distance weights: [3.70; 0.57; (2.89, 5.66)]b CT Log of state cigarette tax (cents/pack). [54.61; 36.96; (2.56, 193.39)]a INC Log of personal disposable income per capita (000s of dollars). [16.41; 2.19; (12.21, 22.31)]c INEQUALITY State income inequality; measured using a variant of the Gini coefficient, with higher values denoting greater inequality. Data available up to 2013 [0.61; 0.03; (0.54, 0.71)]d CIGADV Log of total cigarette advertising expenditures in dollars per capita. Data are aggregate across states and available up to 2012. [1.95; 2.19; (0.10, 13.06)]e BeerREG1 Dummy variable equal to one if the state prohibits grocery store beer sales, zero otherwise. These states include Alaska, Delaware, North Dakota, Pennsylvania, Rhode Island, and Wyoming.f BeerREG2 Dummy variable equal to one if the state is an alcohol beverage control state. These states include Alabama, Maryland, Minnesota, and Utah.f Male Percent of population that is male. [49.28; 0.64; (48.23, 51.25)]g Youth Percent of population under the age of 19. [13.69; 1.00; (11.30, 17.83)]g Non-white Percent of state population that is non-white. [16.74; 9.45; (2.71, 39.78)]g CANADA Dummy variable for states bordering Canada: Idaho, Maine, Michigan, Minnesota, Montana, New Hampshire, New York, North Dakota, Vermont, Washington. MEXICO Dummy variable for states bordering Mexico: Arizona, California, New Mexico, Texas. C PC
Notes: Summary statistics (based on raw data) use annual observations for 48 contiguous states (2005–2014). Observations = 480. CPI (1982–84 = 100) used to deflate monetary variables. Data sources: a Tax Burden on Tobacco. b Tax Foundation, 2015 c U.S. Bureau of Economic Analysis, 2014. d Frank (2009). e Federal Trade Commission Cigarette Report for 2012, 2015. f https://en.wikipedia.org/wiki/Alcohol_laws_of_the_United_States. g http://wonder.cdc.gov/bridged-race-v2014.html.
across the alternate spatial weights (i.e., contiguity and inverse distance). However, as noted by Goel and Nelson (2012), in recent years, cigarette demand has become elastic, and this is borne out by our results. A number of factors are likely contributing to this greater priceresponsiveness of cigarette demand, including health awareness, smoking restrictions in public places, and availability of substitute products (smuggled or contraband cigarettes). Quantitatively, a 10% increase in cigarette prices would reduce consumption by 13%, while similar increases in income would raise cigarette consumption by 9%. The positive coefficient on BPC is consistent with the notion that higher border prices increase a state's cigarette sales via organized or casual smuggling. The border cigarette price effects are about three times more pronounced with inverse distance (Models 2.3–2.4) than with contiguity (Models 2.1–2.2), signifying that spillovers from cigarette smuggling extend beyond a state's immediate borders.11 Turning our focus to the demand interdependence between cigarettes and beer, the two appear as complements, signifying that consumers often consume the two together. However, the magnitude of the effect is rather modest—a 10% increase in PB decreases C by about
11 Interestingly, a 10% increase in cigarette prices across own and border states would reduce cigarette consumption by approximately 8% (=10*(−1.266 + 0.456)) (Model 2.1); however, when one considers the effect of a 10% increase over a broader geographic region using inverse distance the combined effect becomes statistically insignificant. We thank a referee for pointing this out.
Table 2 Spillovers from the beer market to U.S. cigarette demand: baseline models. Dependent variable: cigarette consumption, C. Independent variables
P
C
PB BPC BPB INC
Contiguity (2.2)
(2.3)
(2.4)
−1.266⁎⁎⁎
−1.266⁎⁎⁎ (0.087) −0.031⁎⁎
−1.250⁎⁎⁎
−1.250⁎⁎⁎ (0.073) −0.022⁎⁎
(0.016) 0.456⁎⁎⁎
(0.016) 0.456⁎⁎⁎
(0.011) 1.335⁎⁎⁎
(0.011) 1.335⁎⁎⁎
(0.104) 0.247⁎⁎⁎ (0.070) 0.865⁎⁎⁎ (0.093)
(0.104) 0.247⁎⁎⁎ (0.070) 0.865⁎⁎⁎ (0.093) −0.422⁎⁎⁎
(0.344) 0.108 (0.194) 0.888⁎⁎⁎ (0.093)
(0.344) 0.108 (0.194) 0.888⁎⁎⁎ (0.093) −0.745⁎⁎⁎
(0.087) −0.031⁎⁎
CANADA
Kleibergen–Paap rk Wald F statistic Hansen's J statistic
(0.073) −0.022⁎⁎
(0.118) −0.534⁎⁎⁎ (0.043)
MEXICO
Observations R-squared Kleibergen–Paap rk LM statistic
Inverse distance
(2.1)
(0.078) −0.593⁎⁎⁎ (0.035)
480 480 480 480 0.981 0.981 0.986 0.986 3.044⁎⁎⁎ 3.044⁎⁎⁎ 5.750⁎⁎⁎ 5.750⁎⁎⁎ [0.004] [0.004] [0.000] [0.000] 17.11 17.11 37.82 37.82 6.580 [0.160]
6.580 [0.160]
1.054 [0.901]
1.054 [0.901]
Notes: Time and state dummies and a constant are included but not reported. Two-step GMM estimates are reported with robust standard errors in parentheses and probability values in brackets. The critical values for the Kleibergen–Paap rk Wald F statistic based on test size and bias are in Stock and Yogo (2005). Excluded instruments for PC, BPB, and BPc include CT and spatial lags up to the third order of INC and CT. ⁎ Significance at p b 0.1 level. ⁎⁎ Significance at p b 0.05 level. ⁎⁎⁎ Significance at p b 0.01 level.
0.3 to 0.4%. Furthermore, border beer prices, BPB, significantly affect cigarette sales, especially in contiguous states (Models 2.1 and 2.2). Higher beer prices in bordering states increase a state's cigarette sales as consumers crossing state borders to take advantage of lower beer prices also take the opportunity to purchase complementary cigarettes.12 Interestingly, the magnitude of the border beer price effect on cigarette consumption is about six to seven times greater than own beer price effect (comparing the coefficients on PB and BPB in Models 2.1 and 2.2). It could be the case that bordering residents that cross the border to buy cheaper cigarettes are also buying beer (and vice versa).13 Finally, states that border Canada and Mexico (Model 2.2 and 2.4) have lower cigarette consumption, either due to substitution from cheaper smuggled cigarettes or greater monitoring. These findings are generally in line with ones with earlier data (Connelly et al., 2009); however, the significant (negative) effect of the Canadian border is a new revelation (Model 2.4). Our alternate consideration of border weights uniquely provides some interesting insights. The coefficient on BPB is significant only in contiguous states, while the magnitude of BPC is about a third smaller in contiguous states (Models 2.1, 2.2, versus 2.3, 2.4). Understandably, the border effects of cigarette prices are more pronounced than those of beer prices. In light of the differential transportation restrictions, cross-border beer smuggling is mainly across immediate neighbors, while that of cigarettes can span across distant states as well (Lovenheim, 2008). 12 The generality of these findings to other nations must await availability of comparable data. 13 Availability of individual survey level data would enable one to account for other important individual attributes (education, employment, marital status, gender, religion, etc.) that might crucially dictate smoking and drinking behaviors.
R.K. Goel et al. / Economic Modelling 55 (2016) 292–297
3.2. Price and non-price spillovers from the beer market to U.S. cigarette demand To check the robustness of our findings and to broaden the focus of the present work by incorporating non-price (regulatory) effects of the beer market along with price effects, we conduct a series of robustness checks. These include considering habit-formation effects, advertising, and income inequality. All models in Table 3 include the two regulatory variables, BeerREG1 and BeerREG2, and the estimated Eq. (1) now has the following expanded form Cit ¼ g PCit ; INCit ; PBit ; BPCit ; BPBit ; CANADAi ; MEXICOi ; BeerREG1; BeerREG2
ð2Þ Together with beer taxes, these regulatory variables might be seen as accounting for both price and non-price spillovers from the beer market to cigarette demand. In choosing the set of beer regulations to consider, we settled on a set that could be readily and consistently quantified across states. Specifically, these include (i) BeerREG1: a
Table 3 Spillovers from the beer market to U.S. cigarette demand: robustness checks. Dependent variable: cigarette consumption, C. Independent variables
Beer sales Habit regulations effects
Cigarette Income advertising inequality
(3.1)
(3.2)
(3.3)
(3.4)
BPC
−1.266⁎⁎⁎ (0.087) −0.031⁎⁎ (0.016) 0.456⁎⁎⁎
−0.954⁎⁎⁎ (0.119) −0.020⁎ (0.012) 0.452⁎⁎⁎
−1.332⁎⁎⁎ (0.107) −0.046 (0.032) 0.428⁎⁎⁎
−1.292⁎⁎⁎ (0.093) −0.037⁎⁎ (0.019) 0.426⁎⁎⁎
BPB
(0.104) 0.247⁎⁎⁎
(0.096) 0.102 (0.085) 0.653⁎⁎⁎ (0.114) −0.254⁎⁎⁎ (0.071) −0.278⁎⁎⁎
(0.140) 0.300⁎⁎⁎
(0.115) 0.270⁎⁎⁎
(0.081) 0.822⁎⁎⁎ (0.107) −0.259⁎⁎⁎ (0.097) −0.460 (0.360) −0.236 (0.582) 0.063 (0.129)
(0.072) 0.882⁎⁎⁎ (0.108) −0.307⁎⁎⁎ (0.072) −0.475⁎⁎⁎
PC PB
INC CANADA MEXICO BeerREG1 BeerREG2
(0.070) 0.865⁎⁎⁎ (0.093) −0.340⁎⁎⁎ (0.071) −0.452⁎⁎⁎ (0.056) −0.180⁎⁎⁎ (0.052) 0.082 (0.077)
C (lagged)
(0.069) −0.142⁎⁎⁎ (0.050) 0.114 (0.081) 0.334⁎⁎⁎
(0.066) −0.207⁎⁎⁎ (0.065) 0.068 (0.083)
(0.089) CIGADV
0.012 (0.248)
INEQUALITY Observations R-squared Kleibergen–Paap rk LM statistic Kleibergen–Paap rk Wald F statistic Hansen's J statistic
480 0.981 3.044 [0.004] 17.11
384 0.989 0.994 [0.159] 7.946
384 0.982 2.863 [0.008] 15.71
0.223 (0.185) 432 0.981 3.322 [0.005] 16.57
6.580 [0.160]
5.507 [0.239]
7.678 [0.104]
5.643 [0.227]
Notes: Time and state dummies and a constant are included but not reported. The spatial aspects considered are contiguous states. Two-step GMM estimates are reported with robust standard errors in parentheses and probability values in brackets. The critical values for the Kleibergen–Paap rk Wald F statistic based on test size and bias are in Stock and Yogo (2005). Excluded instruments for PC, BPB, and BPC include CT and spatial lags up to the third order of Income and CT. C (lagged) is the one year time lag of C and is instrumented using the year two time lag of C. ⁎ Significance at p b 0.1 level. ⁎⁎ Significance at p b 0.05 level. ⁎⁎⁎ Significance at p b 0.01 level.
295
dummy variable equal to 1.0 identifying states that prohibit grocery store beer sales. These states include Alaska, Delaware, North Dakota, Pennsylvania, Rhode Island, and Wyoming; and (ii) BeerREG2: a dummy variable equal to 1.0 if the state is an alcohol beverage control state. These states include Alabama, Maryland, Minnesota, and Utah. The results in Model 3.1 show the effect of BeerREG1 to be negative and significant, while the effect of BeerREG2 was statistically insignificant. Thus, restrictions on grocery store beer sales have negative spillovers on cigarette demand and this finding reinforces the complementarity between beer and cigarettes.14 Furthermore, the reinforcing price and non-price spillovers from the beer market is a significant finding that merits consideration by policymakers. The addictive nature of cigarettes has been well-recognized and has led to the emphasis on early interventions in smoking cessation programs. To account for habit-formation effects, we include lagged cigarette consumption as a regressor in the Model 3.1, and the corresponding results are in Model 3.2 of Table 3.15 The two-step GMM estimation procedure is again employed and the excluded instruments for PC, BPB, and BPC include CT and spatial lags up to the third order of INC and CT; and C (lagged) is instrumented using the year two time lag of C.16 The Hansen's J statistic is statistically insignificant and the Kleibergen–Paap rk LM statistic is significant. These suggest that the instruments are largely valid.17 The coefficient on lagged cigarette consumption is positive and statistically significant, reinforcing earlier findings about the presence of a habit formation effect (Baltagi and Levin, 1986). Both price and income elasticities of cigarette demand retain the earlier signs and statistical significance, albeit with somewhat smaller magnitudes. The complementarity between cigarettes and beer holds and is significant. Further, as before, the spatial spillovers from border cigarette prices are significant. However, the effects of border beer prices are not significant, and the effect of beer regulations, BeerREG1, is negative and significant as in Model 3.1. The results with regard to foreign borders are in line with the results in Table 2. Cigarette advertising can increase cigarette demand or at least counter the influences of anti-smoking campaigns. However, over the years, various regulatory mandates have changed the advertising avenues (e.g., broadcast ban on cigarette advertising), while technical changes have opened new advertising avenues (e.g., the Internet and video games). We consider the effects of cigarette advertising by including real per capita cigarette advertising expenditures (CIGADV) obtained from the Federal Trade Commission Cigarette Report for 2012, 2015 in Model 3.3. The resulting coefficient, while positive, failed to achieve statistical significance. This insignificance might partly be due to a single measure failing to account for qualitatively different types of advertising and partly for the lack of consideration of the lagged structure of advertising (Goel, 2011).18 These aspects, while potentially important, seem tangential to the focus of the present work. A noteworthy result, however, is that in this case, both beer price and non-price effects are significant. Besides the level of income, its dispersion or inequality might also affect the demand for cigarettes (Connelly et al., 2010). Greater income inequality might make cigarettes unaffordable for some, inducing them
14 Stehr (2007) found that states that lifted bans on Sunday alcohol sales experienced an increase in sales. 15 See Baltagi and Levin (1986) and Aristei and Pieroni (2009) for a similar treatment of habit effects. An alternate strand of this literature considers rational addiction by including lead consumption – see, for example, Bask and Melkersson (2004). 16 Besides consumption, cross-ownership and co-sponsorship of marketing activities by cigarette and beer companies would affect retail prices of both products (Jiang and Ling, 2011). Investigation of comprehensive holdings of all companies and their subsidiaries is beyond the scope of the current research. 17 The use of the lagged dependent variable as a regressor complicates the instrument choice and we tried several different variations. 18 The effects of advertising might also vary significantly across nations—see, for example, Clements et al. (1985) and McLeod (1986).
296
R.K. Goel et al. / Economic Modelling 55 (2016) 292–297
to switch to cheaper substitutes. Further, low-income smokers might smoke more when they view smoking as complementary to leisure. We follow Frank (2009) to measure income inequality by using a variant of the Gini coefficient with higher values denoting greater income inequality. Our results fail to find a significant effect of income inequality (INEQUALITY) on cigarette demand (Model 3.4), and this finding is in line with previous cross-section results for the United States (Connelly et al., 2010).19 3.3. Additional considerations: Demographic factors and time effects Table 4 incorporates demographic aspects by including age (Youth), gender (Male) and race (Non-white). These factors vary across states and can have a significant effect on smoking (see U.S. Department of Health and Human Services, 2000). Variants of Models 2.1 and 3.1, incorporating demographic aspects, are presented in Models 4.1 and 4.3. Results show that males had higher cigarettes consumption, ceteris paribus. The effects of youth and race were statistically insignificant. Further, to gain insights into time patterns of smoking, Models 4.2 and 4.4 explicitly report the coefficients on year dummies from Models 2.1 and 3.1 (year 2005 is excluded). All the time dummies were negative and statistically significant. In terms of magnitude, the absolute value of time effects was higher over time, signifying a decreasing trend in smoking. Overall, it seems reasonable to conclude that the baseline results are qualitatively robust to considerations of smoking habit formation effects, advertising, and income inequality.20 The non-price effects, especially those captured by BeerREG1, are negative and significant in most cases.
Table 4 Spillovers from the beer market to U.S. cigarette demand: demographic factors and time effects. Dependent variable: cigarette consumption, C. Independent variables
(4.1)
(4.2)
(4.3)
(4.4)
−1.260⁎⁎⁎
−1.260⁎⁎⁎
BPC
(0.088) −0.035⁎⁎ (0.016) 0.422⁎⁎⁎
−1.266⁎⁎⁎ (0.087) −0.031⁎⁎ (0.016) 0.456⁎⁎⁎
(0.088) −0.035⁎⁎ (0.016) 0.422⁎⁎⁎
−1.266⁎⁎⁎ (0.087) −0.031⁎⁎ (0.016) 0.456⁎⁎⁎
BPB
(0.120) 0.258⁎⁎⁎
(0.104) 0.247⁎⁎⁎
(0.120) 0.258⁎⁎⁎
(0.104) 0.247⁎⁎⁎
(0.063) 0.601⁎⁎⁎ (0.149) −0.400 (0.268) −0.472⁎⁎
(0.070) 0.865⁎⁎⁎ (0.093) −0.422⁎⁎⁎
(0.063) 0.601⁎⁎⁎ (0.149) −0.414⁎⁎⁎
(0.070) 0.865⁎⁎⁎ (0.093) −0.340⁎⁎⁎
(0.118) −0.534⁎⁎⁎ (0.043)
(0.073) −0.486⁎⁎⁎ (0.127) 0.178⁎⁎⁎ (0.054) −0.038 (0.031) 0.014 (0.014) −0.284⁎⁎⁎
(0.071) −0.452⁎⁎⁎ (0.056)
P
C
PB
INC CANADA MEXICO Male Youth Non-white BeerREG1 BeerREG2 Year (2006) Year (2007) Year (2008) Year (2009)
4. Concluding remarks This study adds to the literature by examining demand interdependence between cigarette demand and the beer market, and by studying geographic and cross-product spillovers. Our analysis using data for the 48 contiguous U.S. states from 2005 to 2014 shows cigarette demand price elasticity to be negative and price elastic, and the income elasticity to be positive. It seems that repeated cigarette excise tax increases over the years have moved cigarette demand to the elastic portion of the demand curve. Further, in line with the literature (Baltagi and Levin, 1986), cigarette demand shows some inertia from habit formation, while the effects of advertising and income inequality are insignificant. Finally, both price and non-price effects of the beer market have a significant (negative) bearing on the cigarette demand. In regard to cross-product spillovers, cigarettes and beer turn out to be complements.21 The broader literature on smoking-drinking interdependence has failed to find a conclusive relation. For instance, Goel and Morey (1995) find cigarettes and liquor to be substitutes for the U.S., while Pierani and Tiezzi (2009) find complementarity between cigarettes and alcohol in the case of Italy. This result has obvious value for policy coordination in reducing the social costs from these products (Bates et al., 2015). 19 In a study using Canadian data, Latif (2014) alternately considers the effect of economic recessions on smoking and drinking. 20 We also considered extensions to take account of institutional and regulatory differences across states. For instance, we considered accounting for the four states (Florida, Minnesota, Mississippi, and Texas) that were not part of the Master Settlement Tobacco Agreement and entered separate deals with tobacco companies. They might spent settlement dollars differently and with fewer restrictions. A dummy variable identifying these states, when used as an additional regressor, produced similar results. These results are not reported but are available upon request. 21 Given comparable data on beer consumption and price, one could also estimate a beer demand equation and see whether beer consumers view cigarettes as complementary. Specifically, for now, consistent data on retail beer prices across states are not readily available in the public domain. Further, availability of micro-level data would enable consideration personal attributes of consumers.
(0.199) 0.178⁎⁎⁎ (0.054) −0.038 (0.031) 0.014 (0.014)
Year (2010) Year (2011) Year (2012) Year (2013)
−0.028⁎⁎ (0.012) −0.074⁎⁎⁎
(0.090) −0.014 (0.307)
−0.180⁎⁎⁎ (0.052) 0.082 (0.077) −0.028⁎⁎
(0.013) −0.112⁎⁎⁎ (0.012) −0.147⁎⁎⁎ (0.013) −0.092⁎⁎⁎
(0.012) −0.074⁎⁎⁎ (0.013) −0.112⁎⁎⁎ (0.012) −0.147⁎⁎⁎ (0.013) −0.092⁎⁎⁎
(0.035) −0.121⁎⁎⁎ (0.038) −0.188⁎⁎⁎ (0.036) −0.256⁎⁎⁎
(0.035) −0.121⁎⁎⁎ (0.038) −0.188⁎⁎⁎ (0.036) −0.256⁎⁎⁎
(0.034) (0.034) −0.304⁎⁎⁎ −0.304⁎⁎⁎ (0.036) (0.036) Observations 480 480 480 480 R-squared 0.981 0.981 0.981 0.981 Kleibergen–Paap rk LM statistic 4.524⁎⁎⁎ 3.044⁎⁎⁎ 4.524⁎⁎⁎ 3.044⁎⁎⁎ [0.001] [0.004] [0.001] [0.004] Kleibergen–Paap rk Wald F 20.54 17.11 20.54 17.11 statistic Hansen's J statistic 7.853⁎ 6.580 7.853⁎ 6.580 [0.097] [0.160] [0.097] [0.160] Year (2014)
Notes: Time and state dummies and a constant are included in all models, and time effects are reported in Models 4.2 and 4.4. The spatial aspects considered are contiguous states. Two-step GMM estimates are reported with robust standard errors in parentheses and probability values in brackets. The critical values for the Kleibergen–Paap rk Wald F statistic based on test size and bias are in Stock and Yogo (2005). Excluded instruments for PC, BPB, and BPc include CT and spatial lags up to the third order of INC and CT. ⁎ Significance at p b 0.1 level. ⁎⁎ Significance at p b 0.05 level. ⁎⁎⁎ Significance at p b 0.01 level.
Finally, turning to geographic spillovers, the border effects are studied using alternate spatial weights, taking account of both domestic and foreign borders. The effects of both neighboring cigarette prices and neighboring beer prices are significant. The spatial effects of cigarette prices might extend beyond immediate neighbors. This revelation seems new to the extant literature. Some of the differences between the relative geographic effects might be driven by the nature of the two products, related transportation and storage costs, and
R.K. Goel et al. / Economic Modelling 55 (2016) 292–297
regulation, such as varying minimum age restrictions for consumption and purchase.22 While cross-border smuggling of tobacco products has been recognized for quite some time (see ACIR, 1985), the problem does not seem to have yet been contained. There are several implications of our findings for public policy. First, the significance of the habit effects reinforces the need for early intervention programs in preventing smoking initiation. Second, the relatively elastic cigarette demand increases possibilities of smoking reduction through excise taxes, while undermining their revenuegenerating potential. Third, the complementarity between cigarettes and beer suggests policy coordination to internalize cross-product externalities. While earlier findings for the U.S. show liquor and cigarettes to be substitutes (Goel and Morey, 1995), this study finds beer and cigarettes to be complements. Fourth, policymakers should account for both price (tax) and non-price (regulatory) spillovers from the beer market to the cigarette market. Both policies, as in our analysis, could be reinforcing each other. Finally, policies to check cross-border spillovers need to pay attention to both domestic and international borders and to the prices and regulations of related products (in this case beer). These policies are relatively more difficult to coordinate across international borders. References Advisory Commission on Intergovernmental Relations (ACIR), 1985. Cigarette Tax Evasion: A Second Look. Advisory Commission on Intergovernmental Relations, Washington, DC. Anselin, L., 1988. Spatial Econometrics: Methods and Models. Kluwer Academic Publishers, Dordrecht. Aristei, D., Pieroni, L., 2009. Addiction, social interactions and gender differences in cigarette consumption. Empirica 36, 245–272. Baltagi, B.H., Levin, D., 1986. Estimating dynamic demand for cigarettes using panel data: the effects of bootlegging, taxation and advertising reconsidered. Rev. Econ. Stat. 68, 148–155. Bask, M., Melkersson, M., 2004. Rationally addicted to drinking and smoking? Appl. Econ. 36, 373–381. Bates, L.J., Cesur, R., Santerre, R.E., 2015. Short-run marginal medical costs from booze and butts: evidence from the states, South. Econ. J. 81, 1074–1095. Baum, C.F., Schaffer, M.E., Stillman, S., 2003. Instrumental variables and GMM: estimation and testing. Stata J. 3, 1–31. Chaloupka, F.J., Warner, K.E., 2000. The economics of smoking. In: Culyer, A.J., Newhouse, J.P. (Eds.), Handbook of Health Economics vol. 1. North-Holland, Amsterdam, pp. 1539–1627. Clements, K.W., McLeod, P.B., Selvanathan, E.A., 1985. Does advertising affect drinking and smoking? Department of Economics, University of Western Australia, Discussion Paper # 85.02 January Clements, K.W., Lan, Y., Zhao, X., 2010. The demand for marijuana, tobacco and alcohol: inter-commodity interactions with uncertainty. Empir. Econ. 39, 203–239. Coats, R.M., 1995. A note on estimating cross-border effects of state cigarette taxes. Natl. Tax J. 48, 573–584. Connelly, R.T., Goel, R.K., Ram, R., 2009. Demand for cigarettes in the United States: effects of prices in bordering states and contiguity with Mexico and Canada. Appl. Econ. 41, 2255–2260.
22
https://www.dor.ms.gov/info/faqs/TobaccoBeerandAlcohol.html
297
Connelly, R.T., Goel, R.K., Ram, R., 2010. Income inequality and cigarette consumption: evidence from the United States. Appl. Econ. Lett. 17, 423–426. Federal Trade Commission Cigarette Report for 2012, 2015. issued https://www.ftc.gov/ reports/federal-trade-commission-cigarette-report-2012. Fleenor, P., 1998. How excise tax differentials affect interstate smuggling and cross-border sales of cigarettes in the United States. Tax Found. Background Pap. 26. Fogarty, J., 2010. The demand for beer, wine and spirits: a survey of the literature. J. Econ. Surv. 24, 428–478. Frank, M., 2009. Inequality and growth in the United States: evidence from a new statelevel panel of income inequality measures. Econ. Inq. 47, 55–68. Gallet, C.A., 1999. The effect of the 1971 advertising ban on behavior in the cigarette industry. Manag. Decis. Econ. 20, 299–303. Gallet, C.A., 2006. Health information and cigarette consumption: supply and spatial considerations. Empirica 33, 35–47. Gallet, C.A., List, J.A., 1998. Elasticities of beer demand revisited. Econ. Lett. 61, 67–71. Goel, R.K., 2011. Advertising media and cigarette demand. Bull. Econ. Res. 63, 404–416. Goel, R.K., 2013. Where to snuff the puff?: relative effectiveness of U.S. smoking control policies. J. Socio-Econ. 44, 97–102. Goel, R.K., Morey, M.J., 1995. The interdependence of cigarette and liquor demand, South. Econ. J. 62, 451–459. Goel, R.K., Nelson, M.A., 2012. Cigarette demand and effectiveness of U.S. smoking control policies: state-level evidence for more than half a century. Empir. Econ. 42, 1079–1095. Jiang, N., Ling, P.M., 2011. Reinforcement of smoking and drinking: tobacco marketing strategies linked with alcohol in the United States. Am. J. Public Health 101, 1942–1954. Kleibergen, F., Paap, R., 2006. Generalized reduced rank tests using the singular value decomposition. J. Econ. 133, 97–126. Lanoie, P., Leclair, P., 1998. Taxation or regulation: looking for a good anti-smoking policy. Econ. Lett. 58, 85–89. Latif, E., 2014. The impact of recession on drinking and smoking behaviours in Canada. Econ. Model. 42, 43–56. Lovenheim, M.F., 2008. How far to the border?: the extent and impact of cross-border casual cigarettes smuggling. Natl. Tax J. 61, 7–33. McLeod, P.B., 1986. Advertising bans, tobacco and cigarette consumption. Econ. Lett. 20, 391–396. Orzechowski, W., Walker, R.C., 2014. The Tax Burden on Tobacco: Historical Compilation vol. 49. Orzechowski and Walker, Arlington, VA. Pierani, P., Tiezzi, S., 2009. Addiction and interaction between alcohol and tobacco consumption. Empir. Econ. 37, 1–23. Stehr, M., 2007. The effect of Sunday sales bans and excise taxes on drinking and crossborder shopping for alcoholic beverages. Natl. Tax J. 60, 85–105. Stock, J.H., Yogo, M., 2005. Testing for weak instruments in linear IV regression. In: Andrews, D.W.K., Stock, J.H. (Eds.), Identification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg. Cambridge University Press, Cambridge, pp. 80–108. Tauchmann, H., Lenz, S., Requate, T., Schmidt, C.M., 2013. Tobacco and alcohol: complements or substitutes? Empir. Econ. 45, 539–566. Tax Foundation, 2015. State Sales, Gasoline, Cigarette, and Alcohol Tax Rates by State. pp. 2000–2014 (Washington, DC). http://taxfoundation.org/article/state-salesgasoline-cigarette-and-alcohol-tax-rates. Thursby, J.G., Thursby, M.C., 2000. Interstate cigarette bootlegging: extent, revenue losses, and effects of federal intervention. Natl. Tax J. 53, 59–77. U.S. Bureau of Economic Analysis, 2014. Statistical Abstract of the United States. U.S. Department of Health and Human Services, 2000. Reducing Tobacco Use: A Report of the Surgeon General. Centers for Disease Control and Prevention, Atlanta, GA. Warner, K.E., 1982. Cigarette excise taxation and interstate smuggling: an assessment of recent activity. Natl. Tax J. 35, 483–490.