Analyzing the impact of Wal-Mart Supercenters on local food store sales.

The entry of discount mass merchandisers into the grocery business is part of a rapid consolidation occurring in the grocery industry in recent years. Supermarket News estimates that the top five retail grocery chains now account for nearly 40% of U.S. sales. At the top of this list is Wal-Mart. Operating 1,980 Supercenters as of January 31, 2006, Wal-Mart's 2005 share of the nation's retail grocery market was estimated to range from 15% to 20%.

Supercenters are Wal-Mart's fastest growing store format. With more than fifty departments including a full-line grocery section and average size of 187,000 square feet, these stores compete with a wide range of existing retailers in the markets they enter. The greatest competitive pressures from the expansion of Supercenters occur to existing grocery stores. (1) Nearly 500 of these stores are located in counties with an urban population of fewer than 20,000 people. These are primarily rural trade centers in which retail trade is akin to a zero-sum game. Unless population or incomes are growing substantially, there is a relatively fixed amount of money to be spent in the retail sector.

A new large store will capture considerable trade, which must come at the expense of other merchants in the trade area (Blair and Kumar 1997). This article examines the impact of Wal-Mart Supercenters on grocery stores sales in local markets in Mississippi. The findings suggest that Wal-Mart Supercenters located in nonmetropolitan counties capture, on average, 17% of the existing grocery market within the first two years of operation. In metropolitan counties, Supercenters capture about 4% of existing grocery stores' sales one year after entry.

There has been a recent surge in academic research evaluating impacts of the growth of Wal-Mart stores at both the local and national level. This research shows both costs and benefits associated with Wal-Mart's growth, which are distributed across consumers, workers, business owners, and taxpayers in unequal ways. Several recent studies analyzing the impacts on employment and wages in the retail sector after Wal-Mart's entry find that retail employment and earnings decline as a result of Wal-Mart (Basker 2005a: Dube, Eidlin, and Lester 2005: Neumark, Zhang, and Cicarella 2005). Consumers, in contrast, appear to benefit from Wal-Mart's entry in the form of lower prices. Studies focusing on consumer impacts have found that a Supercenter's entry reduces grocery prices. Not only do Supercenters offer lower prices, but their entry may have the indirect effect of lowering prices at competing stores. Estimates of this indirect effect range from 3% overall to as high as 13% for specific items (Basker 2005b; Hausmann and Leibtag 2005).

In general, Wal-Mart's effect on existing local retailers is negative, although some complementary businesses may benefit from Wal-Mart's presence (Irwin and Clark 2006). For grocery stores, competition from Wal-Mart discount stores (not Supercenters) is relatively limited (Gruidl and Kline 1992; Stone 1995; Artz 1999). Most impacted are sales of product lines that compete directly with the discount store such as paper products, health and beauty aids, and cleaning supplies. These products typically account for approximately 25% to 30% of grocery store sales. Supercenters, however, pose a more serious competitive threat since they offer a full line of grocery items.

To date, there has been little research on Supercenters' impact on existing grocers due to their newness and a lack of reliable data on grocery sales. Previous studies that have relied on sales tax receipts for data on retail sales have not fully measured changes in grocery sales occurring after Wal-Mart entry because in most states, food items are not taxable. Therefore, changes in sales of grocery stores, in which generally more than 70% of the items are not subject to sales tax, are not fully reflected in these data. The research that does exist focuses on metropolitan areas for which data are more likely available and finds little evidence that Supercenters affect the grocery retail concentration in larger metropolitan areas (Franklin 2001).

This article analyzes changes in food store sales following the opening of one or more Wal-Mart Supercenters in local markets in Mississippi. Unlike most states, all food items are subject to the sales tax in Mississippi; therefore these data allow us to account fully for food sales. While overall retail sales might rise with the entry of Supercenters, without corresponding population and income growth, some portion of sales are merely being redistributed from existing stores to the Supercenter. Since food items sold in Supercenters are reported in the general merchandise category and not in the food stores' category, these data allow us to examine distributional changes occurring among retail merchandise categories. In addition, we analyze nonmetropolitan markets where much of Wal-Mart's growth has occurred and where any negative impacts of Supercenter entry on existing stores are likely to be most severe.

Data

This analysis focuses on the impact of a new Wal-Mart Supercenter on existing food store sales in the host market area. By October 2005, there were fifty-one Wal-Mart Supercenters in forty counties in Mississippi. The first Supercenters opened in the state in September 1992. Opening dates for Supercenters in Mississippi were obtained from Wal-Mart's website and from local newspaper archives. (2)

Sales tax data reflecting food and beverage sales were collected for all 82 Mississippi counties from fiscal year 1990 to 2005. (3) The food and beverage category reported by the state includes not only grocery stores, but also restaurants and drinking establishments. Previous studies have found that restaurant sales in the host town may increase following the addition of a Wal-Mart store. As such, the impact on this category could comprise two offsetting effects: an increase in restaurant sales and a decrease in grocery sales. We include data from County Business Patterns on the number of eating and drinking establishments in our analysis to control for growth in the number of restaurants.

Figure 1 provides a preliminary look at the potential impact of Supercenters on food and beverage sales, plotting the average per capita sales ratios for host and non-host counties from 1990 to 2005. Without controlling for other factors such as income growth, the figure suggests that around the time of entry of the first Supercenters in the state, trends in per capita food and beverage sales across the two groups began to diverge. By 2005, the average non-host county experienced a gain in per capita sales of roughly 50%, whereas the average host county's per capita sales rose only half as much, roughly 23%. Meanwhile, average population and income in host counties, two major drivers of retail sales growth, outpaced that in non-host counties. (4) The following analysis attempts to quantify the divergence in per capita retail food sales that appears in figure 1 and to determine if it is statistically significant after controlling for other county characteristics and for the likelihood that Wal-Mart locates its Supercenters in higher-growth counties.

[FIGURE 1 OMITTED]

Estimation Strategy

We use a difference-in-differences estimation strategy to examine the impact of Wal-Mart Supercenters on changes in the local market's food store sales. The difference-in-differences method compares outcomes in host counties before and after the addition of a Supercenter as well as comparing these changes with a control group of counties without a Wal-Mart Supercenter.

Equation (1) provides the basis for our empirical specification.

(1) [S.sub.it] = [X'.sub.it][[beta].sub.0] + [[beta].sub.t], + [[gamma].sub.i] [[delta].sub.w] + [W.sub.it] + [[epsilon].sub.it].

Let [S.sub.it] be food store sales for county i at time t, [X.sub.it] be a set of covariate controls that capture differences across counties, [[beta].sub.t] be a time effect that does not vary across counties, and [[gamma].sub.i] be a county effect that is fixed over time. The presence of a Wal-Mart Supercenter is indicated by the variable [W.sub.it] that is measured as the number of Wal-Mart Supercenters open in the county at time t, weighted by the number of years open. (5) The coefficient on [W.sub.it] measures the effect of the Supercenter on sales. The reference group in this case is counties that do not have a Supercenter at time t. Time is measured in years; the change from period 0 to period 1 represents a change from one year to the next. Differencing equation (1) yields

(2) [S.sub.i], - [S.sub.it-1] = ([[beta].sub.t], - [[beta].sub.t-1]) + ([X.sub.it] - [X.sub.it-1])'[[beta].sub.0] + [[delta].sub.w]([W.sub.it], - [W.sub.it]-1) + ([[epsilon].sub.it] - [[epsilon].sub.it-1]),

where ([W.sub.it] - [W.sub.it-1]) measures the change in intensity of Wal-Mart's Supercenter presence and the county fixed effects disappear. Since the impact of a new Supercenter may occur over a number of years, and not just in the first year of opening, we modify equation (2) to incorporate lags,

(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where k = -3, -2 ... 2, 3 indexes three years prior to the Supercenter's entry through the third year after entry. Additional control variables included to account for the observed differences between host and non-host counties are contained in [X.sub.it]. These include county population from the U.S. Census, income data from the Bureau of Economic Analysis, the number of restaurants from County Business Patterns, and the presence of casinos in a county Along with the entry of Wal-Mart Supercenters in Mississippi, thirty casinos opened in seven counties between 1992 and 1999. We include the number of casinos in county i at time t, weighted by the number of years open, to control for the potential influence this development may have on food and beverage sales growth in a county. (6)

The difference-in-differences estimation strategy relies on the assumption that the host county growth would have been the same as the non-host county growth but for the entry of the Supercenter (Angrist and Krueger 1999). One concern is the comparability of the non-host county control group. Non-host counties are considerably smaller on average than host counties in terms of population and per capita food sales (table 1). However, if we compare beginning-of-the-period growth rates in food sales, the dependent variable, the samples look quite similar (figure 2).

[FIGURE 2 OMITTED]

A related concern is endogeneity of the treatment. Wal-Mart does not make location decisions randomly, but rather strategically If Wal-Mart is systematically choosing high growth areas, then we might observe a positive impact of the Supercenter on food store sales even when the true effect is negative. As a result, estimates from ordinary least squares (OLS) will be biased toward zero. Recent studies on the impact of Wal-Mart's entry on wages and employment provide evidence of this effect, finding larger, negative impacts of Wal-Mart entry on retail wages using instrumental variables estimation than are obtained from OLS (Basker 2005a; Dube, Eidlin, and Lester 2005; Neumark, Zhang, and Cicarella 2005). Estimates from OLS can be interpreted as a lower bound on the estimates of the impact.

To address the endogeneity of timing of Supercenter openings, we exploit the fact that many Mississippi counties have had Wal-Mart discount stores for years in advance of Supercenters' entry into the state. (7) Of the forty counties hosting a Supercenter by October, 2005, thirty-five had an existing discount store before the Supercenter's entry. We predict the timing of Supercenter entry using the number of years since a Wal-Mart discount store opened in the county. (8) In the first stage of the estimation, [W.sub.it] is regressed on the number of years a discount store has been open in county i at time t along with county and year effects. The predicted values from this regression then replace [W.sub.it] in the estimation of equation (3).

Results and Discussion

Table 2 presents the results from both ordinary least squares (OLS) and instrumental variables (IV) estimation of equation (3). In both specifications, the coefficient on the Wal-Mart variable is negative and statistically significant for the year of entry and one year after. The OLS estimates suggest that the entry of a Wal-Mart Supercenter in a nonmetropolitan county reduces growth in food and beverage sales by existing grocery stores by 2.8 percentage points in the opening year and an additional 6.3 percentage points in year 1 relative to counties without Supercenters. The effect dissipates by the second year after opening. (9) The IV estimates are even larger, implying that a Supercenter's entry reduces food and beverage growth by 6.3 percentage points in its opening year and another 10.5 percentage points in the succeeding year. Figure 3 plots the estimated evolution of growth rates for food and beverage sales for host nonmetropolitan and metropolitan counties obtained from the IV estimation. (10) The effect of a Supercenter opening in a metropolitan county is smaller and is statistically significant only in the opening year. The IV estimates imply that Supercenter entry reduces growth in food and beverage stores by 4 percentage points relative to counties without a Supercenter.

[FIGURE 3 OMITTED]

Average food and beverage sales for a nonmetropolitan Mississippi county over the study time period were approximately $46.8 million. The estimated two-year reduction in sales growth estimated to occur with Supercenter entry translates to a sales decline of roughly $7.5 million in the host county. Data from the 2002 Economic Census report that statewide, average grocery sales per establishment were about $3.4 million; thus we might expect that Supercenter entry would replace the equivalent of two existing average supermarkets, or more, if per establishment sales for nonmetropolitan stores were on average lower than the state mean. In metropolitan counties, grocery store sales averaged $314.7 million over the study period. Our estimates suggest a one-year decline roughly equivalent to $12.6 million, or the equivalent of 3.7 average-sized host county supermarkets.

What does this decline in food and beverage sales mean? Since grocery sales at Supercenters are classified in the general merchandise category and not in the food and beverage category, these estimates represent the effect of the Supercenter on grocery sales, net of the Supercenter's own sales. Certainly, the findings seem to suggest that Mississippi residents are shifting their grocery purchases from supermarkets to Supercenters. Yet, there may be another effect of the Supercenter reflected in these results. Sales in competing grocery stores may be declining partially because these stores have lowered prices in response to increased competition from Supercenters. (11) Hausmann and Leibtag (2005) estimate this indirect price effect to be approximately 3% to 5% overall, which might account for the decline in metropolitan markets, but is much smaller than the decline estimated for nonmetropolitan markets.

Conclusions

This study has evaluated the impact of the entry of a Wal-Mart Supercenter on the sales growth of existing retail grocery stores in the local market. We find that Wal-Mart's entry into nonmetropolitan markets reduces growth of grocery store sales by nearly 17 percentage points within two years of entry. In metropolitan counties, the proportionate reduction in sales growth is much smaller, 4 percentage points within one year of Supercenter entry, but the associated dollar amount of sales captured by the Supercenter is nearly twice as large. Some of the decline may be attributable to lower prices induced by Wal-Mart competition, but the magnitude of the effects suggests that, particularly in more rural areas, the new store captures a significant amount of business from existing grocery retailers in the host county.

This analysis has implications for economic development policy. Retail recruitment, particularly of "big-box" or discount department stores, has become an increasingly popular economic development strategy in more rural areas. Many local officials anticipate increased property and sales taxes from a new Supercenter, but fail to recognize other potentially offsetting effects. The findings in this study will help local policymakers better understand some of the costs and benefits of recruiting and providing incentives for retail development.

References

Angrist, J.D., and A.B. Krueger. 1999. "Empirical Strategies in Labor Economics." In O. Ashenfelter and D. Card, eds. The Handbook of Labor Economics, Vol 3. Oxford, UK: Elsevier Ltd, pp. 1277-1366.

Artz, G.M. 1999. "The Impact of Wal-Mart on Retail Market Structure in Maine." MS thesis, Department of Resource Economics and Policy, University of Maine.

Ayres, I., and S.D. Levitt. 1998. "Measuring Positive Externalities From Unobservable Victim Precaution: An Empirical Analysis of LoJack." Quarterly Journal of Economics 113:43-77.

Basker, E. 2005a. "Job Creation or Destruction? Labor Market Effects of Wal-Mart Expansion." Review of Economics and Statistics 87:174-183.

--. 2005b. "Selling a Cheaper Mousetrap: Wal-Mart's Effect on Retail Prices." Journal of Urban Economics 58:203-229.

Blair, J.P., and R. Kumar. 1997. "Is Local Economic Development a Zero-Sum Game?" In R.D. Bingham and R. Mier, eds. Dilemmas of Urban Economic Development. Thousand Oaks, CA: Sage Publications, pp. 1-20.

Dube, A., B. Eidlin, and B. Lester. 2005. "Impact of Wal-Mart Growth on Earnings throughout the Retail Sector in Urban and Rural Counties." Working Paper, Institute of Industrial Relations, University of California, Berkeley.

Franklin, A.W. 2001. "The Impact of Wal-Mart Supercenter Food Store Sales on Supermarket Concentration in U.S. Metropolitan Areas." Agribusiness 17:105-114.

Gruidl, J., and S. Kline. 1992. "The Impact of Discount Stores on Retail Sales in Illinois Communities." Rural Research Report, Western Illinois University.

Hausmann, J., and E. Leibtag. 2005. "Consumer Benefits from Increased Competition in Shopping Outlets: Measuring the Effect of Wal-Mart." Working Paper, National Bureau of Economic Research.

Irwin, E., and J. Clark. 2006. "The Local Costs and Benefits of Wal-Mart." Department of Agricultural, Environmental and Development Economics, The Ohio State University. Available at: http ://www-agecon.ag.ohio- state.edu/programs/ComRegEcon/walmart/Irwin_ClarkWalMart_Final_03_06.pdf

Mississippi State Tax Commission. Annual Report, 1990-2005. Data from fiscal year 1998 through 2005 are available online at www.state.ms.us.

Neumark, D., J. Zhang, and S. Ciccarella. 2005. "The Effects of Wal-Mart on Local Labor Markets." Working Paper, National Bureau of Economic Research.

Stone, K.E. 1995. "Impact of Wal-Mart Stores on Iowa Communities: 1983-93." Economic Development Review 13 (2):60-69.

U.S. Department of Commerce, Bureau of the Census. County Business Patterns (various years). Washington DC. Available at http://www.census.gov/epcd/cbp/view/cbpview.html.

--. Population Estimates (various years). Washington DC. Available at http://www.census.gov/popest/estimates.php.

U.S. Department of Commerce, Bureau of Economic Analysis. Regional Economic Accounts. Washington DC. Available at http://www.bea.gov/.

(1) In fiscal year 2005, 159 of 242, or 66%, of new Supercenters resulted from conversions of existing stores. See Wal-Mart SEC form 10K, January 31, 2005.

(2) Recent store openings (2004 to current) are posted on the following Wal-Mart websites: http://walmartstores.com and http://www.walmartfacts. For earlier openings, we used local newspaper archives to find press releases announcing the opening dates or contacted the stores directly. Three counties gained a Super-center in fiscal year 2005. These counties do not enter as host counties since no post-entry data are available.

(3) The Mississippi State Tax Commission reports county sales by merchandise category annually for a fiscal year spanning July 1 to June 30.

(4) On average, host county population in 2005 was 16% higher than in 1990: population growth in non-host counties was only 4% higher. Income growth in host counties rose 116% while non-host counties' income grew 88%.

(5) [W.sub.it] is the sum of the number of years open over all Wal-Mart stores in county i at time t. For example, if the county had one Supercenter open for three years and one Supercenter open for one year, [W.sub.it] would equal (4). This provides a measure of exposure, allowing Wal-Mart's effect to evolve over time and to increase with the number of stores in the county.

(6) This measure is analogous to the exposure measure for Supercenters. It is the sum of years open over all casinos in the counties.

(7) Of the 82 counties in Mississippi, 31 had neither a discount store nor Supercenter; 11 had a discount store only; 35 had a discount store prior to adding (or converting to) a Supercenter; and (5) counties had only a Supercenter. This strategy is similar to one used by Ayres and Levitt (1998) in their study of the impact of LoJack on crime rates.

(8) There may be a spatial correlation in retail sales growth across counties due to the pull of regional trade centers. Any spatial patterns that are fixed over time are differenced away in equation (3). Since Supercenters tend to locate in regional trade centers, we might expect little change in the trade patterns resulting from the Supercenter's entry. Yet, there is evidence that Supercenters attract shoppers from neighboring counties, so entry of a Wal-Mart Supercenter could alter the spatial patterns in retail trade. If so, our estimates of the Supercenter's total effect are understated, capturing the relative difference between host counties and non-host counties. but not accounting for any negative impact the Supercenter may have on grocery sales in adjacent counties. We experimented with interacting [W.sub.it] and its lagged values with a variable indicating host county adjacency. These coefficients were insignificant, suggesting that Supercenter entry did not significantly alter existing trade patterns.

(9) This finding may be due in part to sample attrition. The average host county in the sample had a Supercenter on average only 2 years.

(10) A 95% confidence interval is shown. It Estimates of the indirect price effect provide an upper bound on the amount of the change in sales that may be due to lowered prices. Sales (S) is the product of price (p) and quantity (q). Taking the total differential, we have [partial derivative]S = [partial derivative]p * q [partial derivative]q * p. The price effect [partial derivative]p is negative. If there were a quantity effect. [partial derivative]q, it would be positive; lower prices would induce shoppers to purchase more. This would provide an offsetting effect, reducing any revenue loss due to lowered prices. Therefore, [partial derivative]p is the most that sales might fall due to lowered prices and not to redistribution of sales to the Supercenter.

Georgeanne M. Artz is assistant professor, Department of Agricultural Economics and Truman School for Public Affairs, University of Missouri. Kenneth E. Stone is Professor Emeritus, Department of Economics, Iowa State University. We are grateful for helpful comments provided by Judy Stallman, Laura McCann, Jeffrey Milyo, and participants in a Truman School research seminar.

This article was presented in a principal paper session at the AAEA annual meeting (Long Beach, CA, July 2006). The articles in these sessions are not subjected to the journal's standard refereeing process. Table 1. Summary Statistics for Mississippi Counties With and Without Wal-Mart Super-centers

Sample Means Item Category Host Non-host All Annual Food and beverage sales growth 0.023 0.025 0.024 Annual Population growth 0.009 0.003 0.006 Annual Income growth 0.055 0.046 0.050 WM supercenters weighted by years open 2.057 0.000 1.029 1990 Food and beverage sales

($millions) $87.43 $19.29 $53.36 1990 Population 46,848 15,969 31,408 1990 Income ($millions) $647.96 $175.31 $411.64 Table 2. Regression Results of Food and Beverage Sales Growth

OLS

Parameter Standard Variable Estimate Error In [Population.sub.t]--In

[Population.sub.t-1] 0.175 * 0.095 In [Income.sub.t]--In

[Income.sub.t-1] 0.290 *** 0.090 In [Restaurants.sub.t]--In

[Restaurants.sub.t-1] -0.002 0.002 [Casinos.sub.t]--

[Casinos.sub.t-1] 0.006 0.005 [Year.sub.t-3] -0.031 0.024 [Year.sub.-3] * Metro 0.058 * 0.032 [Year.sub.-2] 0.006 0.007 [Year.sub.-2] * Metro -0.031 ** 0.015 [Year.sub.-1] 0.006 0.054 [Year.sub.-1] * Metro -0.027 0.056 [Year.sub.0] (entry) -0.028 *** 0.008 [Year.sub.0] * Metro 0.018 * 0.010 [Year.sub.1] -0.063 *** 0.012 [Year.sub.1] * Metro 0.057 *** 0.013 [Year.sub.2] 0.002 0.005 [Year.sub.2] * Metro 0.002 0.007 [Year.sub.3] 0.012 * 0.007 [Year.sub.3] * Metro -0.003 0.008

IV

Parameter Standard Variable Estimate Error In [Population.sub.t]--In

[Population.sub.t-1] 0.174 * 0.102 In [Income.sub.t]--In

[Income.sub.t-1] 0.287 *** 0.090 In [Restaurants.sub.t]--In

[Restaurants.sub.t-1] -0.002 0.002 [Casinos.sub.t]--

[Casinos.sub.t-1] 0.005 ** 0.002 [Year.sub.t-3] -0.008 0.014 [Year.sub.-3] * Metro 0.066 *** 0.023 [Year.sub.-2] 0.026 ** 0.011 [Year.sub.-2] * Metro -0.031 * 0.019 [Year.sub.-1] 0.006 0.016 [Year.sub.-1] * Metro -0.001 0.023 [Year.sub.0] (entry) -0.063 *** 0.019 [Year.sub.0] * Metro 0.023 0.022 [Year.sub.1] -0.105 *** 0.021 [Year.sub.1] * Metro 0.087 *** 0.023 [Year.sub.2] -0.001 0.009 [Year.sub.2] * Metro -0.003 0.017 [Year.sub.3] 0.015 0.012 [Year.sub.3] * Metro -0.001 0.021 Notes: Asterisks denote significance: * significant at the 10-percent level; ** significant at the 5-percent level: *** significant at the 1-percent level. Time dummies arc included in the estimat4ion, but not reported above. The sample size is 1.148. The [R.sup.2]1 for the OLS estimation is 0.242: for IV estimation it is 0.240. Robust standard errors, adjusted for serial correlation, are reported.