How far to the border? The extent and impact of cross - border casual cigarette smuggling.

E[Q | [P.sub.h] = [p.sub.h] [P.sub.h] = [p.sub.b], ln(D) = ln(d)] - E[Q | [P.sub.h], = [P.sub.b], ln(D) = ln(d)]/ E[Q | [P.sub.h] = [p.sub.h], [P.sub.b] = [p.sub.b], ln(D) = 0] - E[Q | [P.sub.h] = [P.sub.b], ln(D) = ln(d)]

If everyone behaves as if they live on the border, so E[Q | [P.sub.h] = [p.sub.h], [P.sub.b] = [p.sub.b], ln(D) = ln(d)] = E[Q | [P.sub.h] = [p.sub.h], [P.sub.b] = [p.sub.b], ln(D) = 0], then the above equation implies 100 percent smuggling. If, on the other hand, everyone behaves as if they purchase from their home state (meaning that the price difference is zero), then E[Q | [P.sub.h] = [p.sub.h], [P.sub.b] = [p.sub.b], ln(D) = ln(d)] = E[Q | [P.sub.h] = [P.sub.b], ln(D) = ln(d)], and there will be zero smuggling. The smuggling percentage is the ratio of these two quantities. Another way to proceed would be to use the parameter estimates from Table 6 to identify the parameters in equation [4] and calculate P([S.sub.i] = 1). Since I assume a linear probability model for smuggling, this procedure can create estimates outside of the range 0,1. Equation [10] can be thought of as a rescaling of P([S.sub.i] = 1) to be between 0 and 1. I am essentially determining the extent to which individuals behave as if they live in the home state and face only the border price, or live in the home state and face only the home state price. I perform this calculation only for the full demand model, as the statistic does not have the same interpretation if applied to the intensive or extensive margins. Results are presented in Panel C of Table 7.

I find evidence of large amounts of cross-border purchases. Depending on the specification, the preceding calculation implies between 13.1 and 25.1 percent of consumers in MSAs purchase cigarettes in a lower-price state or reservation. (33) The estimates including Native American Reservations are much larger due to the reduction in traveling distance and price when these jurisdictions are included (see Table 5). The estimates in Table 7 are population-weighted averages over all MSAs. It is important to note these percentages can only be generalized to the United States as a whole if the distribution of distance with respect to lower-price borders for MSAs are representative of the distribution for non-MSAs. It is unclear whether the preceding estimates are smaller or larger than they would be for the United States as a whole, and the reader is urged to use caution when applying these estimates out of sample.

[FIGURE 1 OMITTED]

Figure 1 presents a simulation of smuggling percentage for different distances at the mean level of all variables aside from distance. The parameter estimates used were those from Table 6, Panel C, column iv. The figure represents how smuggling changes by distance for the average consumer in the sample. The percent smuggling ranges from a high of 100 for those who live on the border to zero for those who live more than 77 miles from the border. While the shape of the figure is imposed by the assumption of a log-linear relationship between distance and smuggling, it is interesting to note that my estimates imply a good deal of smuggling behavior occurs outside of 25 miles, which is the cutoff assumed by Chaloupka (1991). Further, the assumption of 100 percent smuggling within a 20-mile band by Lewit et al. (1981) and Lewit and Coate (1982) appears to fit the data poorly. By allowing smuggling behavior to vary log linearly with respect to distance, my model and parameter estimates yield a more complete picture of cross-state purchasing behavior than previous studies.

Under the assumption cross-state purchasers smoke the same amount as those who purchase cigarettes in their home state, the smuggling percentage also can be interpreted as the proportion of consumed cigarettes that are purchased in border localities. My estimates imply consumers who smuggle will smoke more than those who do not. Thus, the smuggling percentage represents a lower bound on the percentage of cigarettes that are casually smuggled. When interpreted in this manner, these estimates are large, particularly in light of previous estimates of casual smuggling under one percent (Stehr, 2005). (34)

There are some sources of validation for this finding in the state of New York. The Center for a Tobacco-Free New York conducted a survey and found 25 percent of New York State residents purchased cigarettes on a Native American Reservation (FACT Alliance, 2005). Further, the New York Association of Convenience Stores found Western New York cigarette sales dropped between 25 and 50 percent after the 2000 tax increase (FACT Alliance, 2005). There is also anecdotal evidence of high volumes of casual smuggling: when South Dakota increased its cigarette excise tax by one dollar in January 2007, Larchwood Mini Mart in Iowa reported its January cigarette sales tripled total sales for 2006. One consumer reported she makes the 20-mile trip from Sioux Falls once or twice a week (Efrati, 2007).

Together with the average price differences listed in Tables 2 and 5, the distance distributions are consistent with the large predicted smuggling amounts. Although the mean of distance is 93 miles excluding Native American Reservations and 68 miles including Native American Reservations, the median of these variables is 65 and 45 miles, respectively. In the 2001-2002 CPS supplements, the median person living in an MSA lived approximately 49 miles from a lower-price border state or reservation. The average per-pack price difference faced by consumers was forty-five cents (a little over 12 percent of the average real home state price). As the average smoker smoked 15 cigarettes per day (three-fourths of a pack), she would save $123.19 per year by purchasing all of her cigarettes in a border locality and not changing her smoking behavior. This is a fairly substantial amount of average savings given most individuals need only travel 50 miles or less one to two times a year to realize them. (35) The large amount of casual smuggling implied by the empirical estimates is consistent with many consumers taking advantage of the substantial savings from purchasing in lower-priced jurisdictions.

Table 8 presents similar information to Table 7 broken down by state for the full model. The estimates are derived from column iv of Table 6, so they exclude Native American Reservations but include a year trend. Note these estimates are averages of the various statistics over MSAs within a state weighted by the number of observations that constitute each MSA-specific mean, not state-level estimates. Distance is still measured at the MSA level as this is the level of observation in the study. Table 8 illustrates the large differences across states in the responsiveness of consumption to changes in the home state price as well as in the percent of consumers who engage in casual smuggling. These results underscore the importance of accurately accounting for smuggling incentives in cigarette demand models; the "naive" elasticity estimate of -0.326 in Table 6, Panel C, column ii provides a poor estimate of the home state price elasticity in many states.

The casual smuggling estimates presented in Table 8 vary from a high of 63 percent in Washington, D.C. to a low of zero percent in Delaware, Idaho, Kentucky, Missouri, New Hampshire, and New Mexico. The large value for Washington, D.C. occurs because it is three miles from Virginia, and there is an average difference of eighty-cents per pack between the two locations. Given the location of their MSAs with respect to lower-price borders, at least 25 percent of consumers in Arkansas, Massachusetts, Maryland, New Jersey, Rhode Island, and West Virginia are estimated to engage in smuggling activity. The home state price elasticities reflect these differences, with the low-smuggling states being more home price elastic than the high smuggling states. Similar patterns emerge for the impact of smuggling on smoking. (36)

Using the MSA-specific estimates of the percent of consumers who casually smuggle combined with information on the closest lower-price locality, I calculate the net percent change in sales for each state due to cross-border purchasing activity. (37) Results are reported in the final column of Table 8 and suggest that there are clear winners and losers from the existence of interstate price differentials. At the extreme, New Hampshire sales double because they are the lowest-tax state in New England. Virginia, Indiana, Kentucky, and Delaware also gain substantial sales from cigarette tax evaders. Conversely, Maryland, Kansas, Massachusetts, and Illinois lose significant sales (and thus tax revenue) due to the availability of lower-price cigarettes in nearby jurisdictions. These results imply that in the states with large quantities of smuggling and inelastic home state price elasticities, cigarette taxes are ineffective at both reducing smoking of residents and providing substantial tax revenue to the home state. Instead, these taxes often serve to export both consumers and tax revenues to nearby states.

Discussion

The most striking finding in this analysis is that, on average, consumption is non-responsive to variation in the home state price. What the state average results in Table 8 make clear, however, is the substantial heterogeneity in home state price responsiveness that varies according to the geographic distribution of each state's population. Thus, in MSAs that are far from lower-price borders, the home state price elasticity is negative, whereas for those close to the border, my estimates imply a positive home state price elasticity.