ORGANIZATIONAL FORM AND PRODUCT MARKET ... - CiteSeerX

ORGANIZATIONAL FORM AND PRODUCT MARKET COMPETITION ARE FOCUSED FIRMS WEAK COMPETITORS?

NAVEEN KHANNA and SHERI TICE* September 2002 First Draft: Feb 2002

*Khanna is from the Eli Broad College of Business, Michigan State University and Tice is from the A.B. Freeman School of Business, Tulane University. We thank Rob Hansen, Paul Spindt, Tom Noe, Suman Banerjee, Ted Fee, Charlie Hadlock, Long Chen and Zsuzsanna Fluck for their comments. We thank David Scharfstein for suggesting the source of the pricing data. Tice gratefully acknowledges research support from a Freeman Faculty Research Fellowship at the A.B. Freeman School of Business. Address correspondence to Sheri Tice, A.B. Freeman School of Business, Tulane University, New Orleans, LA 70118, or email: [email protected].

ORGANIZATIONAL FORM AND PRODUCT MARKET COMPETITION ARE FOCUSED FIRMS WEAK COMPETITORS?

ABSTRACT

We examine product market behavior of retailers to determine if diversified or focused firms behave as weak competitors and are perceived as such. We do so in three distinct ways. We first test pricing predictions of a simple switching cost model of market competition, that weaker firms charge higher prices in equilibrium (and sacrifice market share). Not surprisingly, we find higher average prices in cities with a larger proportion of high debt and low efficiency firms. Surprisingly, the same result holds for cities with a larger proportion of focused firms. Our second test documents that firms with these characteristics are more likely to exit after facing a negative shock to their city, again suggesting that like high debt, low efficiency, and focused firms are weak competitors. Given that certain characteristics are associated with weakness, and these characteristics are observable, stronger competitors should be able to take advantage of firms with such characteristics. We model the optimal location decision of a new entrant into a city with existing incumbent stores. The prediction is the entrant locates closer to a weaker incumbent. We document a new entrant locates closer to high debt as well as focused firms suggesting that at least in the retail industry, focused firms are perceived as weak competitors.

JEL Classification: D43, G31, G32

2

The presence of a diversification discount has been a troublesome puzzle since documented by Lang and Stulz (1994) and Berger and Ofek (1995). Given the compelling advantages of diversification, and that much of the economic activity in the U.S. is conducted by diversified firms, there is a natural resistance to the notion that diversified firms destroy value. Not surprisingly, a number of recent papers have issued a serious challenge to the existence of the discount and the arguments in support of it. The challenges have come on two fronts. Researchers like Fluck and Lynch (1999), Maksimovic and Phillips (2002), Campa and Kedia (2002) and Graham Lemmon and Wolf (2002) argue that for endogeneity and self-selection reasons, comparing diversified firms to stand-alones is inappropriate. If poorly performing firms are more likely to diversify or if acquired firms are on average poor performers, this benchmark consisting of stand-alones is too stringent. Researchers like Whited (2001), Mansi and Reeb (2002), and Villalonga (2002) argue that the documented inefficient cross-subsidization as well as the documented diversification discount result from measurement problems. When corrected for, either organizational form is irrelevant, or arguably there is a diversification premium.

We contribute to the debate from a different angle. Instead of focusing on differences in valuation of diversified and stand-alone firms, we focus on differences in their equilibrium decisions in the product market. This not only permits us to circumvent some of the perceived shortcomings of previous tests, but also provides new insights about how firms compete at the market level. If organizational form is irrelevant, not only should both diversified and standalone firms make similar decisions, but their competitors should perceive the firms to be similar and respond similarly. Since these responses are conditioned also on competitors’ own organizational forms, equilibria across markets with different mixes of diversified and focused firms should be similar. However, if either decisions or responses depend on own and/or competitors’ organizational form, organizational form would appear to play an important role in product market competition.1

We are not the first to compare decisions of focused and diversified firms. Papers like Lamont (1997), Shin and Stulz (1998), Scharfstein (1998), Rajan, Sarvaes and Zingales (2000), Khanna and Tice (2001) and Maksimovic and Phillips (2002) examine capital expenditure decisions of diversified versus focused firms and find that they invest differently. The contribution of this

1

The literature is appropriately concerned whether such results can be attributed to organizational form, or if there is an unobservable variable driving both organizational form and differences in decisions and responses. We discuss this issue later.

3

paper is two fold. One, we look at a different set of decisions to establish whether they too are affected by organizational form. Two, we are able to document that focused firms act as weak firms without having to measure their prospects directly. Instead, we infer prospects from the equilibrium behavior of a firm and its competitors. The switching cost model of market structure provides predictions about how weaker firms behave, and how competitors take that into account while making their own decisions. By observing actions of firms in the product market, we can determine whether a firm views itself as weak and whether that view is shared by its competitors. This allows us to avoid measures like profitability, productivity or industry Q as proxies for prospects.2

We use city level observations for the discount department industry in our study. This industry is particularly suited to our tests because of the heterogeneity of incumbent characteristics both within and across cities. Cities differ not only in the total number of competing stores, but also in the mix of firms owning these stores. While some cities are served by only one firm, others have six or seven firms competing. The firms themselves are quite heterogeneous, differing on variables like size, profitability, leverage, and organizational form. Another benefit of this industry is Wal-Mart. Wal-Mart, expanded dramatically across the United States during this period fanning out from its base in Arkansas. Since the merchandise sold is relatively homogeneous, entry by a technologically superior and low cost firm puts pressure on all incumbents, particularly weaker ones. As in Chevalier (1995), by using city level observations, the potential endogeneity of firm characteristics as a response to a negative shock to a city is less of a concern than in other settings.

We examine the link between organizational form and firm weakness in three distinctly different ways. We first test pricing predictions using a simple switching cost model of market competition. The main insight from this model is that weaker firms charge higher prices in equilibrium and consequently sacrifice market share. Thus, markets with one or more weak competitors have softer competition, allowing other incumbents to also charge higher prices.3

2

At first glance, focused firms acting weak may appear at variance with the presence of a diversification discount. However, it is possible that for the period of our study, acting weak may have been the more efficient action. The issue though is complicated. We also document that acting weak results in competitor actions that are detrimental to a firm. Thus, there are benefits to acting strong. However, that entails further investing in market share, which is costly. The fundamental question then is whether focused firms are solving this tradeoff more efficiently. Investigation of this issue is left to future research. 3 See Appendix I for our simple model demonstrating this result.

4

This results in higher average prices for such markets.4 As predicted, variables traditionally used as proxies for weakness are positively correlated with equilibrium prices. Cities with a larger proportion of high debt and low efficiency firms have higher prices. However, prices are also positively correlated with the fraction of focused firms in the city. The focus variable is persistently significant in all the specifications tested and remains significant when we include low efficiency, high debt, small division size, and small parent firm size in our specifications. The results are consistent with focused firms being weak competitors.

If a firm is weak, it should be especially vulnerable when there is new entry as that increases aggregate supply and pushes prices lower. For our second test, we examine city level exit decisions of incumbent stores within four years after Wal-Mart’s entry into the city. 5 We find that incumbents that have low efficiency, high debt or the focused organizational form have a higher probability of exit from a city within four years after Wal-Mart’s entry. The focus variable remains a strong predictor of exit after controlling for other traditional measures of weakness. Unlike Chevalier (1995), we do not find evidence of predatory pricing, as prices are the highest in cities with the largest amount of subsequent exit.

For our third test we look at competitor decisions. Given that certain characteristics are associated with a firm being weak, and these characteristics are observable, stronger competitors should be able to take advantage of firms with such characteristics. Our pricing data consists of average prices at the city level so firm level prices cannot be used to directly examine this. However, there is a test we can perform which explicitly documents that strong firms do condition on characteristics associated with weak incumbents. When Wal-Mart enters a city it knows the location and characteristics of all incumbents. An important decision it makes is where to locate its store relative to the existing stores in the city. Its location affects not only the prices and market share of its immediate competitors but those of all other incumbents. Thus, it chooses that location which maximizes its own profits after accounting for the equilibrium responses of all incumbents. 4

This may not hold if there is an attempt to force exit of weaker incumbents through aggressive, even predatory pricing. Since this forces the rival incumbent to also lower its price, the average price in a market with a weak incumbent could be lower. However, in markets with switching costs, firms are localized monopolies and make positive profits. Thus, the option to run out a rival by charging a very low price is quite expensive. It requires either that the rival can be ousted quickly or that the resulting market share after ousting the rival is very valuable. We later provide evidence that predatory pricing does not appear to occur in our data set.

5

To examine this, we build a simple extension of the Klemperer (1987, 1995) model of pricing and spatial competition. Our model predicts that in equilibrium the new entrant should locate closer, but not too close to the weaker incumbents.6 It locates closer to a weak incumbent, because such incumbents charge higher prices, allowing the new entrant to also charge higher prices. It does not locate too close though, because locating closer decreases the switching costs for consumers, forcing both firms to lower prices. Thus, where Wal-Mart locates should provide information about its perception of incumbent weakness. Using a sub-sample, we find that when Wal-Mart enters a city it places its first store closest to stores owned by high debt and focused firms suggesting that Wal-Mart perceives these firms to be the weaker incumbents. On average it locates 1.5 miles away from the closest incumbent consistent with the model predictions of not locating too close.

If a particular organizational form reflects weakness and competitors base their decisions on it, a natural question is whether the firm should change its organizational form. This is not unlike the endogeneity issue driving the current debate. Since organizational form is a decision variable, it is endogenously and optimally determined. Thus, identifying a benefit associated with it does not imply the firm would gain from changing it. Other costs may outweigh the benefits of changing organizational form. This is especially relevant for our data set where the same firm operates in numerous local markets. If it decides to change organizational form in response to what happens in one or few of its markets, it must consider the effect of the change it all its other markets and, thus, on total efficiency. Another question is whether it is organizational form that is important or some unobservable factor is driving both organizational form and firm weakness. As in the literature, we try to handle this concern by controlling for a number of potentially relevant variables in our tests. Two of these are financial leverage and operating efficiency. If the unobservable factor is important, it is likely to impact at least one of these variables. Thus, we can suggest the following policy: firms should switch to the diversified organizational form to gain the benefits of being tough only if they can do so without increasing debt or lowering overall efficiency, as these too are attributes of weak competitors.

5

This test is similar to one in Zingales (1998) to test the impact of debt on firm survival when trucking firms faced a deregulation shock. 6 Our model demonstrating this result is located in Part B of Appendix I.

6

The rest of the paper proceeds as follows. Section I describes our data and sample selection. Section II contains pricing and exit tests. Section III contains the new entrant location tests. Section IV concludes.

I. City Level Data and Sample

A. Pricing Data The quarterly price data used in this study comes from the American Chamber of Commerce Researchers Association (ACCRA) Cost of Living Index. Under the guidelines of the ACCRA, local chambers of commerce offices collect quarterly prices on a variety of items. The cities included in the survey are those where chambers of commerce or similar organizations have volunteered to participate. The number of respondents providing prices varies from quarter to quarter. ACCRA stringently reviews all prices reported each quarter and attempts to eliminate errors and noncompliance with specifications. The Cost of Living Index generally includes only cities with populations over 40,000.7 The composition of the items in the Cost of Living Index changed substantially between 1981 and 1982. To standardize results, we only use prices in 1982 and beyond in our tests.

The Miscellaneous Goods & Services Index is one of the indexes in the Cost of Living Index. It contains several items that are sold in discount department stores. We divide the items contained in the Miscellaneous Goods & Services Index into three groups: (1) Discount Items Group: Seven items sold by discount department stores; (2) Non-Discount Items Group: Ten items not sold by discount department stores; (3) Alcoholic Beverage Group.

Another index available in the Cost of Living Index is a Grocery Items Index. We add three nonfood items from the Grocery Items Index to the discount items group, as they are items likely to be sold in discount department stores over the period of the study. The items are “facial tissue” (Kleenex), “washing powder” (Tide, Bold or Cheer) and “soft drink” (2 liter Coca Cola). The Grocery Items Index also consists of many food items. In general, discount department stores chains have gradually added food items in their stores during the 1980’s and 1990’s. However, this has been done at different paces and times with various discount chains. To keep the

7

According to ACCRA, there are, however, a small number of special case exceptions where communities have proven their ability to provide data coverage.

7

introduction of noise to a minimum, food items from the Grocery Items Index are not added to the discount item group or the non-discount item group.8

We use the discount items group to measure discount price levels and the non-discount items group as a control for overall price levels in a city caused by unobservables. A detailed listing of the composition of the discount and non-discount item groups is displayed in Table I. 9

B. Store Location Data Two trade journals are used to identify industry participants. Using trade journals to define an industry is a cleaner way to identify industry participants than using the Compustat Industrial Segment Data where firm managers have discretion with respect to how they aggregate their activities into segments. We first make a list of all discount department store chains that are on the “Leading Discounters” list in Discount Merchandiser for at least one year during the 1975 – 1996 time period. This step eliminates very small chains of stores. The Directory of Discount Department Stores is then used to determine store locations for each chain for each year in which they operate in the industry during the 1982 – 1996 time period.10 A discount department store chain must be in both of these trade journals to be included in our tests. The trade journals show the name of the firm that owns each chain. Sometimes a firm may own more than one discount department store chain. If this is the case, the details with respect to the chains are combined within the firm for our tests.

C. The Sample The Directory of Discount Departments Stores lists store locations at the beginning of each year. Therefore, there is ambiguity regarding the quarter in which the new store actually opens. For example, according to the Directory of Discount Department Stores a Wal-Mart store is first located in Covington, Louisiana at the beginning of 1984. This store could have opened anytime between the beginning of 1983 and the beginning of 1984. We define the beginning of the

8

The prices of alcoholic beverage prices are not used, as several reporting cities are apparently located in dry counties and do not report prices of alcoholic beverages. 9 During our sample period ACCRA switched Johnson’s Baby Shampoo with Alberto VO5 in the Miscellaneous Goods & Services Index. We adjust the price of the item after the switch to reflect this change. The details of the adjustment methodology used are provided in Appendix II. 10 After 1995 discounters are pooled with other general merchandise retailers, and firm specific details are no longer provided. Thus, our sample is constrained on the lower end at 1982 due to pricing data limitations and at the upper end at 1996 due to firm characteristic limitations.

8

calendar year in which Wal-Mart enters a city as quarter zero, with the understanding that WalMart enters the city sometime between quarter zero and quarter plus four.

Wal-Mart enters 1,588 cities during the 1984 – 1996 time period. For each of these cities we identify whether prices are available at quarter minus one. If prices are not available for a city at quarter minus one we go back to quarter minus two. Cities are dropped from the sample if there are no incumbent stores in the city when Wal-Mart enters, if there are any stores owned by a foreign firm, or if there are any stores owned by a franchise operation. This results in 180 cities for our first set of tests. There are 43 different incumbent firms in the 180 sample cities, and stores in any one city are owned by between one to seven incumbent firms.11 Kmart is an incumbent in all but seven of the 180 cities and is the largest incumbent firm based on the number of stores or sales over the sample period.

To minimize potential endogeneity arising directly from the effect of Wal-Mart’s entry on firm characteristics, we measure firm characteristics for the fiscal year that ends around quarter zero. However, we do not feel that this type of endogeneity is likely given the multi-market structure of this industry. A time line illustrating the basic empirical set-up is shown below:

Incumbent Firm Characteristics Measured

Wal-Mart’s 1st Store Observed in City j

Incumbent Exit Measured City j

--------------------------------------------------------------------------------------------- Qtr – 1 Qtr 0 Qtr +4 Qtr +20 Wal-Mart Enters City j Between Qtr 0 and Qtr +4

Prices Measured In City j

11

If a firm spins-off a division, the spin-off company is treated as a new firm.

9

II. Pricing and Exit Tests

A. Determinants of Market Prices A standard finding in the industrial organization literature and a result that emerges in a simple switching cost model (see Appendix I), is that in equilibrium, high cost firms charge higher prices. In their extension of the Klemperer (1987, 1995) switching cost model, Chevalier and Scharfstein (1996) show that equilibrium prices will also be higher in markets with a higher fraction of high debt firms. High debt firms have a higher probability of exit making future market share less important. Thus, these firms are more likely to charge higher current prices at the expense of market share. Since both high costs and high leverage have traditionally been associated with weakness, our hypothesis is that cities with a higher fraction of weak incumbents will charge higher prices.

Higher average prices in cities with a larger fraction of weak incumbents, is a general result that should hold independent of when or whether Wal-Mart enters a particular city. However, this finding is likely to be strongest around Wal-Mart’s entry. Given that Wal-Mart is a cost effective and technologically superior firm, its entry is likely to hurt the prospects of all incumbents in general and the weaker ones in particular. With expected deterioration in prospects, these firms are likely to care less about future market share and charge higher pre-entry prices. 12 The same holds for post-entry prices. However, Wal-Mart’s entry initially increases the total number of stores in a city and later results in exit of a number of incumbents. The large changes in spatial competition after Wal-Mart’s entry make post-entry prices transitional for a time, making test results unreliable. For these reasons, we measure price levels one quarter before Wal-Mart’s entry.13 14 The dependent variable is defined as the sum of ten discount item prices divided by the sum of ten non-discount item prices:

12

There is a potential endogeneity problem if entry is anticipated, as incumbents may choose to change their characteristics (right hand side variables). However, as shown in Khanna and Tice (2000), firms do not alter their characteristics around their first interaction with Wal-Mart. This is probably due to their presence in multiple heterogeneous markets. What happens in one market or in a few of their markets is unlikely to cause them to change firm characteristics, as this would impact all of their other markets as well. Given heterogeneity of markets, the total impact of such a change will be hard to gauge. 13 We do not examine price changes due to entry since our model’s predictions relate only to price levels. Post-entry prices, though, should be lower both because of the increase in number of stores and because the new incumbent has lower costs. This is borne out by the data. 14 Sometimes prices are not available in a particular quarter as the city failed to report them. If pricing data is not available at quarter minus one, prices are measured at quarter minus two if they are available.

10

10

Relative Discount Prices j, t

∑ P d, j ,t d=1 = ----------------------10 ∑ P nd j t nd = 1

where P is defined as an item’s price, j indexes cities, d indexes discount items, nd indexes nondiscount items and t indexes time.

The use of relative discount prices is a way to control for city level unobservables as it adjusts for cost of living differences and other unobservables that may impact prices at the city level. For example, if residents of a particular city are willing to pay higher prices for better service we would expect that to show up in discount prices as well as in non-discount item prices. Similarly if inflation is higher in a city than in other cities it should affect discount item prices as well as non-discount item prices.

The independent variables used in our tests are defined below:

Discount Sales Per Square Foot Sales-per-square foot is frequently used as a measure of operating costs/efficiency in the discount department stores industry. It measures firm performance using what is known as the retailing productivity loop. If a retailer has low cost of goods sold or low SG&A expenses, they are able to charge lower prices and remain profitable. When prices are lowered, they pick up additional volume leveraging fixed costs further. The leveraging of fixed costs provides the ability to further lower prices and capture higher volume leading to an even lower expense ratio. How good a firm is at exploiting the retailing productivity loop is typically measured in the industry through sales per square foot. We follow the industry standard of measuring firm costs/productivity with discount sales per square foot. Firm level sales-per-square foot is available for the discount divisions of both public and private firms in our sample from The Directory of Discount Department Store.15

15

In a few cases, discount sales per square foot for a specific fiscal year is missing for some firms in the Directory of Discount Department Stores. In these instances, the value for the prior year for the firm is used instead. In the case of one firm, sales per square foot is not shown for several years, but firm sales and the number of discount stores is shown for each fiscal year. For this firm, the average square footage of the

11

Parent Firm Size Parent firm size could matter for many reasons. It may be a measure of efficiency, as efficient firms may be more likely to grow large. Larger firms may also have better access to external financing.

Discount Size Like parent firm size, the size of a firm’s discount department store operations is used in the specification for many reasons. The size of a firm in the industry may be a proxy for operating costs due to volume discounts when purchasing merchandise. It may also proxy for efficiency, as those firms that are efficient in this business are more likely to grow large.

Firm Debt Firm indebtedness is a commonly used proxy for financial constraints. The level of firm debt has been found to be relevant for investment and pricing decisions in the product market literature. We measure this using a firm’s total debt ratio defined as: Total Assets (Compustat item A6) minus stockholder equity (Compustat item A216) all divided by total assets (Compustat item A6).16 Unfortunately, this information is not available for the privately held firms in our sample.

Focused Organizational Form Firms with the diversified organizational form may be tougher competitors for several reasons. They may have lower costs due to synergies, more access to capital due to a higher borrowing capacity, financing flexibility due to an internal capital market, and a lower probability of bankruptcy. However, firms with the diversified organizational form may be weaker competitors. This can happen if agency conflicts due to the more complex organizational form create high agency costs paid out as extra managerial compensation or wasted in rent-seeking behavior by division managers, if there is less access to external funds due to lower transparency, or if there is

firms stores is calculated for the last year it is shown, and implied sales per square foot is calculated assuming average store size remains unchanged over the subsequent four years. 16 In a few cases, this information is not available in Compustat for a specific fiscal year. In these instances, the value for the prior year for the firm is used instead. In one case, this was not available, so the value for the following fiscal year is used. One firm is not in Compustat. Information available in Moody’s Industrial Manuals is used to calculate the debt ratio for this firm.

12

less managerial effort due to an inability to offer division managers incentive contracts linked directly to division stock price performance. 17,18

Following the capital expenditure literature, focused firms are defined as those with more than 90 percent of their total firm sales originating from the discount department store industry. Discount department store sales come from the Directory of Discount Department Stores. Occasionally this data is missing. In these instances, discount department store sales come from Discount Merchandiser. For most of the sample, total firm net sales come from Compustat. If not available there, the data comes from the Million Dollar Directory, or Wards Business Directory.19

Privately Held Equity It is possible that private firms have less access to capital and consequently may be weaker than firms with access to public equity markets.

The prices in our data set are available at the city level. Due to this, measures of weakness have to be measured at the city level for our tests. The explicit definitions of the independent variables used in the first set of tests are:

Frlowssq: The fraction of firms in city j with CPI inflation adjusted discount sales per square foot of less than $177 (in 1996 dollars). This is approximately the mean/median industry sales per square foot documented for the discount department store industry in Khanna and Tice (2001).

FrSmFirmSize: The fraction of stores in a city j owned by firms with CPI adjusted total firms sales of less than 1.475 billion (in 1996 dollars). This results in half of the firms in the sample being classified as small.

17

For papers discussing the theoretical benefits of diversification see Lewellen (1971), and Stein (1997). For papers discussing the theoretical disadvantages see Meyer, Milgram and Roberts (1992), Scharfstein and Stein (1998), Rajan, Servaes and Zingales (2000). 18 It is unclear if inefficient cross-subsidization via an internal capital market would make firms tougher or weaker competitors. In these models good prospect divisions receive too little funding and bad prospect divisions receive too much funding vis-à-vis stand-alone firms in the same industries. 19 In a few cases, this ratio was not available for a particular firm in a particular fiscal year. In these cases the focus variable is measured one year earlier.

13

FrSmSegmentSize: The fraction of stores in a city j owned by firms with CPI adjusted discount department store sales of less than 1.475 billion (in 1996 dollars). This is the same size used to identify small firm size.

Frhighdebt: The fraction of stores in city j owned by firms with a total debt to total assets ratio of at least 70%. A debt ratio cut-off of 70% causes around 50% of the cities to be classified as having at least one store owned by a high debt incumbent firm.

FrFocused: The fraction of firms in city j with more than 90% of firm sales attributed to the discount department store industry.

FrPrivate: The fraction of stores in city j owned by firms without publicly traded stock.

Herf: This is the city Herfindahl measure. It is calculated as the sum of the squares of the market share held by each incumbent firm i in city j at the beginning of the event window. It is used as a control variable for market concentration.

The summary statistics for the variables are shown in Table II, while regression results are shown in Table III. We find that prices are higher the larger the fraction of low sales per square foot incumbents. This result is not surprising, given firms with higher costs would be expected to charge higher prices. However, we also find that prices are higher the larger the fraction of focused incumbents, a finding that is robust to the inclusion of other variables that measure firm weakness.20 The results on the focus variable are also economically significant. Using the coefficient estimates shown in column 1, if the fraction of focused firms in a city increases by 20%, the coefficient implies an increase of .0135 in relative discount prices. At the sample average relative discount price ratio of .8680, this translates to a 1.55% increase in discount item prices holding non-discount item prices fixed. Total parent size, discount operation size and whether a firm is privately held are not determinants of relative discount prices before the shock.

20

Using the same industry over the longer period of 1975 - 1996, Khanna and Tice (2001) document that diversified firms are more sensitive to their relative productivity/efficiency when making capital expenditure decisions when competing against Wal-Mart. They document that high productivity diversified firms invest more in stores than high productivity focused firms and low productivity diversified firms invest less in stores and exit earlier than low productivity focused firms. Due to pricing data limitations, the current study examines the 1982 – 1996 interval. By 1982 several of the low productivity diversified firms have exited the industry making it more likely that the surviving diversified firms have high relative productivity.

14

The coefficient on the Herfindahl measure is positive and significant in some specifications providing some evidence that prices are higher in cities with less competition.

Previous studies have documented that firm financial leverage is an important determinant of prices suggesting there may be an omitted variable bias in our estimates.21 We next test whether the documented results for the focus variable continue to hold after including firm financial leverage in the specifications. We drop all cities in the sample that have any incumbent firms lacking debt information. This results in a sub-sample of cities that is 75 percent the size of the original sample. The relative price tests using this sub-sample are shown in Table IV. The fraction of stores owned by high debt firms in a city is positive and statistically significant indicating that the higher the fraction of high debt stores in a city, the higher the discount prices. The coefficient on the financial leverage variable is also economically significant and of a similar magnitude to the focus coefficient. For example, using the coefficient estimates shown in column 1, if the fraction of high debt firms in a city increases by 20%, the coefficient implies an increase of .0136 in relative discount prices. At the sample average relative discount price ratio of .8680, this translates to a 1.56% increase in discount item prices holding non-discount item prices fixed. These results are consistent with those obtained by Chevalier (1995) who showed that supermarkets that had recently undergone a LBO charge higher prices than their non-LBO rivals.

It is also noteworthy that the fraction of stores in a city owned by focused firms continues to be positive and significant indicating that cities with a larger fraction of focused firms in them have higher prices. Using this sub-sample, the fraction of low sales per square foot incumbents is not a significant determinant of relative prices. This is probably due to a high correlation between leverage and sales per square foot. As with the previous tests using all cities, firm size, and discount size are not significant independent variables.

B. Firm Characteristics That Predict Exit We next examine which incumbent firm characteristics predict exit from a city after Wal-Mart’s entry. Weak firms are more likely to exit after the negative shock of Wal-Mart’s entry. In fact, 34% of incumbent stores in our sample exit a city by quarter +20 (Wal-Mart enters sometime between quarter 0 and quarter +4) suggesting that Wal-Mart’s entry is a severe, negative shock to incumbents. The design of our tests allows us to examine the link between the level of prices 21

See Chevalier (1995), Phillips (1995), Chevalier and Scharfstein (1996), Zingales (1998), and Campello (2002).

15

immediately prior to Wal-Mart’s entry and the level of exit after Wal-Mart’s entry. The Chevalier and Scharfstein (1996) model predicts that prices should be higher in cities containing firms with a higher probability of exit. According to their model and our pricing results cities with a higher fraction of low efficiency firms, high debt firms or firms with the focused organizational form should have a higher probability of exit. However, this prediction may not hold in cities with a mix of strong and weak incumbents if there is an attempt to force exit of weaker incumbents through aggressive, even predatory pricing.

We start with the 180 city observations used in Table III. We drop all observations for cities with quarter zero occurring after 1991, as store location data is needed for twenty quarters after quarter zero and our store location data ends in 1996. This results in a sub-sample of 115 city observations for the exit tests. The dependent variable is the fraction of incumbent stores at quarter zero that exit city j by quarter plus twenty. We estimate this as a latent variable or using a Tobit Model: y* = βo + xβ β + µ where µ/x ∼ Normal (0, σ2) and Fraction Exit j,

0 to +20

= max (0,

y*). Formally: if #Strs n,j, +20 ≥ #Strs n, j, 0 then y = Fraction Exit j , 0

to + 20

= 0

if #Strs n,j,+ 20 < #Strs n, j, 0

then y = y* = Fraction Exit j, 0

to +20

N ∑ [Net Reduction #Strs n ,j] 0 to +20 n=1 = ---------------------------------------------------N ∑ #Strs n ,j 0 n=1

where [Net Reduction #Strs n ,j] 0 to +20 = [#Strs n, j, 0 − #Strs n, j, +20], #Strs is the number of stores, j indexes cities, n indexes incumbent firms in city j and Wal-Mart enters city j between quarter 0 and quarter +4. The estimation results are shown in Table V.22 The fraction of incumbent stores that exit a city after Wal-Mart’s entry is larger the higher the fraction of stores owned by low sales per square 22

If a diversified firm spins off its discount division and continues to operate a store in a market this is not treated as an exit. Similarly, if a firm goes private and continues to operate a store in a city this is not

16

foot firms, and the higher the fraction of stores owned by focused firms. Consistent with the pricing results, total firm size, firm discount operations size, and access to public equity markets are not important predictor’s of firm exit. Therefore, incumbent characteristics that lead to higher equilibrium prices in cities, also predict more exit after new entry. We then repeat these tests using the sub-sample of cities where debt is known for all incumbents and then include debt levels as an explanatory variable. We find that the fraction of incumbent stores that exit the city is greater the higher the fraction of stores owned by low sales per square foot firms, the higher the fraction of stores owned by focused firms, and the higher the fraction of stores owned by high debt firms.23 These results are consistent with low efficiency (high cost) firms, high debt firms, and focused firms being weaker competitors as they have a higher exit rate after the negative shock.

As in Chevalier (1995), by using city level observations, the potential endogeneity of

firm characteristics as a response to a negative shock to a city is less of a concern than in other settings. This suggests that not only are focused firms weak, but that the focused organizational form is likely the cause of the observed weakness in the product market rather than the reverse.

As is shown in Table VI, we do not find evidence that stronger incumbents are pricing low to prey on weaker incumbents as the fraction of firms that exit after the negative shock is higher in markets that have the highest prices immediately before the shock. If predation exists, one would expect a higher incidence of exit in markets with lower prices. Furthermore, we find prices are the lowest in markets that have all strong competitors and these markets have the least amount of exit subsequently. In our sample, twenty cities have only diversified, low debt, high efficiency competitors. The average relative discount item prices in these cities is .82 versus .87 for the other one hundred and sixty cities. This difference is statistically significant at the 2% level and suggests that prices are lower because of stronger competition rather than the existence of predatory pricing. In models with switching costs this result is not surprising as incumbents are localized monopolists and make positive profits. Thus, attempts to drive out a rival by charging low prices are probably justified if either the rival can be driven out quickly, or gaining market share is very valuable. Neither of these is likely in our sample. Since competition takes place at city level and the same firms do not compete in every city, the effect of lowering prices in only some of the rival’s markets may not result in quick exit. Given low barriers to entry, exit by one rival may make entry by another more likely, reducing the importance of gaining market share.

treated as an exit. However, if an incumbent firm is acquired by another firm, or an incumbent store is sold to a competitor, this is treated as exit by the incumbent firm. 23 The exit results for financial leverage are consistent with those found by Zingales (1998).

17

The nature of the business also suggests that weak firms are unlikely to lower prices to compensate customers for increased distress risk.

We next examine the firm specific characteristics of the firms that retrench to verify that it is the “weak” firms that are exiting cities after Wal-Mart’s entry. The 115 cities in Table V are used for this robustness test. For each of the 115 cities, we record if each firm i has less stores in city j at quarter plus twenty than it has at quarter zero (recall that Wal-Mart enters the city sometime between quarter 0 and quarter plus four). This results in 307 firm i, city j pairs of observations. A difference in the mean test is done to see if there are statistically significant differences in the average firm characteristics of the firms that retrench in a city versus those that do not retrench after Wal-Mart’s entry. The results are shown in Table VII. Retrenching firms tend to have the focused organizational form, high debt levels, low operating efficiency (or high operating costs), be smaller industry players, and be smaller firms. It appears firms with “weak” characteristics are the ones exiting after Wal-Mart’s entry, and focus is one of these characteristics.

III. New Entrant Location

We next test for firm weakness from a new and different angle. We examine whether incumbent firms’ characteristics affect a new competitor’s decisions. If certain observable characteristics of incumbents reveal weakness, an entrant should condition on it when making its own decisions. One such decision is to determine where to locate its new store in relation to existing incumbents’ stores. To fix our hypotheses, we build a simple model where an entrant observes heterogeneous incumbents’ characteristics and locations and determines a location that maximizes its own profits after accounting for the equilibrium responses of both incumbents. The model, provided in Section B of Appendix I, suggests that the new entrant (Wal-Mart) will locate closer (but not too close) to the weaker incumbent. The reason is that a weaker incumbent charges higher prices in equilibrium. By locating closer to the weaker incumbent, the entrant can charge higher prices than it could if it located closer to the stronger incumbent. The new entrant would not locate too close, though, because switching costs become smaller the closer the new entrant locates to the weaker incumbent, putting downward pressure on prices.

The 1987, 1991 & 1995 Directory of Discount Department Stores contain detailed store location data and were available to us. Using these three volumes of store location data, we identify all cities entered by Wal-Mart for the first time during each of these years. We drop all cities with

18

only one incumbent firm with stores in a city at the time of Wal-Mart’s entry. We also drop cities with any stores owned by a foreign firm, or a franchised firm, as no firm level data is available for these firms. Cities with multiple Wal-Mart stores entering at the same time are dropped, as measuring the distance from each incumbent store to the Wal-Mart stores is complex.

We next determine the driving distance in miles between the new entrant (Wal-Mart) and each of the incumbent stores in the city using the point-to-point driving directions on Yahoo.com. If any incumbent store address cannot be found on Yahoo.com or Mapquest.com, the city in which the store is located is dropped from the sample.24 Sometimes instead of listing the address of the new Wal-Mart store, the 1987 store location book identifies the new Wal-Mart as a “projected new store” in the city. In these cases, the Wal-Mart address shown in the 1995 volume is used as long as there is only one Wal-Mart store in the city in 1995, and the square footage of the store matches.

This process results in 205 cities and 665 firm i - city j pairs of observations for our tests. Consistent with model predictions, we find that Wal-Mart locates close, but not too close to the closest incumbent when it enters a city. The average driving distance between the new Wal-Mart store and the closest incumbent store is 1.53 miles. The average distance between the new WalMart and the other incumbents in each city is 4.95 miles.

We next test to see if Wal-Mart locates its stores closest to stores owned by “weak” incumbents when it enters new cities. The dependent variable is the distance (in miles) from the new entrant’s store to incumbent store i. All of the following independent variables are measured for firm i for the fiscal year ending when Wal-Mart enters: (1) Focus Dummy: dummy = 1 if discount department store sales are ≥ 90% of firm sales (2) High Debt Dummy = 1 if Total Debt over Total Assets ratio ≥ 70% (only available for firms that have publicly traded stock) (3) Natural log of Inflation Adjusted Discount Sales (’96 dollars) (4) Inflation Adjusted Discount Sales per Square Foot (’96 dollars) (5) Natural log of Inflation Adjusted Parent Sales (’96 dollars) (6) The number of incumbent stores in the city when Wal-Mart enters is included as a control variable to adjust for city size.

24

For example, sometimes a shopping mall or shopping center will be named rather than a street address. If an internet search doesn’t lead to a street address, the city was dropped from the sample.

19

The regression results are shown in Table VIII. Columns 2 & 4 consist of only those firm i-city j pairs where debt is known for the firm while Columns 1 & 3 consist of all firm - city pairs of observations. It appears that Wal-Mart locates its stores closer to focused firms and high debt firms when it enters a city, controlling for firm efficiency, discount size, total firm size and the size of the city. This is consistent with Wal-Mart perceiving high debt and focused firms as weaker competitors.

IV. Conclusion This paper contributes to the debate of whether organizational form matters from a different angle. Instead of focusing on differences in valuation of diversified and stand-alone firms, we focus on differences in their equilibrium decisions in the product market. This not only permits us to circumvent many of the perceived shortcomings of previous tests, but also provides new insights about how firms compete in these markets. If organizational form is irrelevant, not only should both diversified and stand-alone firms make similar decisions, but their competitors should perceive the firms to be similar and respond similarly. Since these responses are conditioned also on competitors’ own organizational forms, equilibria across markets with different mixes of diversified and focused firms should be similar. However, if either decisions or responses depend on own or competitors’ organizational form, this would suggest that either organizational form or an unobservable driving organizational form plays a role in product market competition.

Using data from the discount department store industry, we examine the link between organizational form and firm weakness in the product market in three distinctly different ways. We first test pricing predictions of a simple switching cost model of market competition, demonstrating that weaker firms charge higher prices in equilibrium (and sacrifice market share). Not surprisingly, we find higher average prices in cities with a larger proportion of high debt and low efficiency firms. Surprisingly, the same result holds also for cities with larger proportion of focused firms. Our second test documents that firms with these characteristics are also more likely to exit after facing a negative shock to their city, again suggesting that like high debt and low efficiency firms, focused firms are weak competitors. Given that certain characteristics are associated with weakness, and these characteristics are observable, stronger competitors should be able to take advantage of firms with such characteristics. We model the optimal location decision of a new entrant into a city with existing incumbent stores. The prediction is the entrant locates closer to a weaker incumbent. We document a new entrant locates closer to high debt as

20

well as focused firms suggesting that in the retail industry, focused firms are perceived as weak competitors.

If the focused organizational form reflects weakness and competitors base their decisions on it, a natural question is whether the firm should change its organizational form. This is a difficult question and one we address only indirectly. The reason is organizational form is endogenously determined. Thus, identifying a benefit associated with the diversified form does not imply the firm would gain from changing to it. Other costs may outweigh the benefit. So it is important to look at the net effect through the expected impact on profitability or efficiency. If this is positive, changing organizational form may be reasonable. Another unresolved issue is whether some unobservable factor is driving both organizational form and firm weakness. We try to address this by controlling for a number of potentially relevant variables. Two of these are leverage and efficiency. If the unobservable factor is important, it is likely to impact at least one of these variables. Thus, we suggest that firms should switch to the diversified organizational form only if they can do so without increasing debt or lowering operating efficiency, as these too are attributes of weak competitors.

This study uses data for the discount department store industry, an industry for which extensive detailed data is available. It would be interesting to see whether the results hold more generally. Given that Maksimovic and Phillips (2002) use manufacturers and find similar results to Khanna and Tice (2001) who use the discount department store industry, the results of this study could well extend to manufacturers too. If not, an investigation of potential differences would increase our understanding of why organizational form matters for some industries and not others. We leave these and other issues to future research.

21

Table I Discount and Non-Discount Item Descriptions ACCRA descriptions of the items from the Cost of Living Index used in this study are shown below. All items are from the ACCRA Miscellaneous Goods & Services list except, facial tissue, washing powder and soft drink, which are from the ACCRA Grocery Items list. Discount Department Store Items Group

Non-Discount Items Group

Toothpaste:

Hamburger Sandwich:

6 oz – 7 oz tube, Crest or Colgate

¼ lb patty with cheese, McDonalds, if available

Shampoo:

Pizza:

11 oz bottle Johnson’s Baby Shampoo

12” – 13” thin crust, cheese, Pizza Hut or Pizza Inn, if

(Switches to 11 oz bottle of Alberto VO5 in 1991 Qtr 4 )

available. (11” – 12” size starting in 1994 Qtr 4)

Man’s Dress Shirt:

Fried Chicken:

White, cotton/poly blend, long sleeves

Thigh and drumstick, with or without extras, whichever

(Various combinations of brands used over time

is less expensive, Church’s or Kentucky Fried Chicken,

including a change to 100% cotton pinpoint Oxford, long

if available.

sleeves in 1994 Qtr 4) Man’s Denim Jeans:

Major Appliance Repair:

Levi brand, size 28/30 – 34/36

Home service call, clothes washing machine; minimum

(Specific Levis jean changes: Levi’s straight leg, Levi’s

labor, excluding parts.

500 series, Levi’s 501s or 505s) Boy’s Underwear:

Beauty Salon:

Package of 3 cotton briefs, lowest price

Woman’s shampoo, trim and blow dry.

Tennis Balls:

Dry Cleaning:

Wilson or Penn brand, can of 3 extra-duty yellow.

Man’s two-piece suit.

Board Game:

Haircut:

Parker Brothers “Monopoly” No. 9 standard edition

Man’s barbershop haircut, no styling.

Facial Tissue:

Newspaper Subscription:

175 count box Kleenex brand

Monthly cost of daily and Sunday home delivery.

Washing Powder:

Movie:

42 oz or 49 oz Tide, Bold or Cheer

First-run, indoor, evening rate, no discount.

Soft Drink:

Bowling:

2 liter Coca Cola excluding any deposit

Price per game, evening rate.

22

Table II Summary Statistics for Variables This table contains means and standard deviations for the variables used in the subsequent specifications. The 180 cities used have prices available in qtr −1 (or qtr –2 if prices are not available in qtr –1). Wal-Mart enters city j between qtr 0 and qtr +4. Cities have at least one incumbent firm at the beginning of the event window, and no foreign owned or franchised stores. The variables measure the characteristics of the incumbent firms in city j for the fiscal year that ends at the beginning of qtr 0. FrFocused is the fraction of incumbent stores with more than 90% of sales coming from the discount department store industry, FrLowssq is the fraction of incumbent stores owned by firms with inflation-adjusted sales per square foot of less than $177 in 1996 dollars, FrSmFirmSize is the fraction of incumbent stores owned by firms with total firm inflation adjusted sales of less than 1.475 billion in 1996 dollars, FrSmSegmentSize is the fraction of incumbent stores owned by firms with discount department store inflation-adjusted sales of less than $1.475 billion in 1996 dollars, FrPrivate is the fraction of incumbent stores owned by firms that do not have publicly traded stock, Herf is the Herfindahl measure for the city and is defined as the sum of the squares of the market share held by each incumbent firm i in city j at the beginning of the event window, Relative Disc Prices is the sum of the prices of the 10 discount items divided by the sum of the prices of the 10 non-discount items measured at quarter –1 (or qtr –2 if qtr –1 prices are not available). 43 firms are represented in the sample.

Variable

Mean

Std Dev.

FrFocused

.3442

.3250

FrLowssq

.3145

.3314

FrSmFirmSize

.1214

.1924

FrSmSegmentSize

.1545

.2274

FrPrivate

.0736

.1513

Herf

.4858

.2413

Relative Disc Prices

.8680

.0951

23

Table III Determinants of Prices The dependent variable is the relative prices of discount items in city j defined formally as: 10 ∑ Pd ,j ,-1 d=1

Relative Discount Prices j, -1 =

÷

10 ∑ Pnd ,j ,-1 nd = 1

where P is defined as an item’s price, j indexes cities, d indexes discount items, nd indexes non-discount items. The 180 cities used have prices available in qtr −1 (or qtr –2 if prices are not available in qtr –1). Wal-Mart enters city j between qtr 0 and qtr +4. Cities have at least one incumbent firm at the beginning of the event window, and no foreign owned or franchised stores. The variables measure the characteristics of the incumbent firms in city j for the fiscal year that ends at the beginning of qtr 0. FrFocused is the fraction of incumbent stores with more than 90% of sales coming from the discount department store industry, FrLowssq is the fraction of incumbent stores owned by firms with inflation-adjusted sales per square foot of less than $177 in 1996 dollars, FrSmFirmSize is the fraction of incumbent stores owned by firms with total firm inflation adjusted sales of less than 1.475 billion in 1996 dollars, FrSmSegmentSize is the fraction of incumbent stores owned by firms with discount department store inflation-adjusted sales of less than $1.475 billion in 1996 dollars, FrPrivate is the fraction of incumbent stores owned by firms that do not have publicly traded stock, Herf is the Herfindahl measure for the city and is defined as the sum of the squares of the market share held by each incumbent firm i in city j at the beginning of the event window. 43 different firms are represented in the sample. A time dummy representing the 1980’s is included, but the estimate of the coefficient is not shown. The estimations shown use White’s adjustment for heteroskedasticity. P-values are in parentheses.

1

2

3

4

FrLowssq

.0441** (.044)

.0476* (.053)

.0458* (.053)

.0405* (.070)

FrFocused

.0676*** (.001)

.0709*** (.001)

.0689*** (.001)

.0624*** (.002)

−.0175 (.671)

FrSmFirmSize

−.0074 (.819)

FrSmSegment Size

FrPrivate Herf.

# Observations R-Squared

.0504 (.273) .0428 (.124)

.0405 (.141)

.0414 (.133)

.0481* (.079)

180

180

180

180

.1008

.1017

.1011

.1062

***, **, or * denote t-statistics significant at the one, five and ten percent respectively

24

Table IV Determinants of Prices: Incumbent Debt Known The dependent variable is the relative prices of discount items in city j defined formally as: 10 10 Relative Discount Prices j,-1 = ∑ Pd ,j ,-1 ÷ ∑ Pnd ,j ,-1 d=1 nd = 1 where P is defined as an item’s price, j indexes cities, d indexes discount items, nd indexes non-discount items, and q indexes the specific quarter. The 134 cities have prices available in quarter −1 (or quarter –2). Wal-Mart enters city j between qtr 0 and qtr +4. All cities used in each panel have at least one incumbent firm at the beginning of the event window, and no foreign owned or franchised stores. The variables measure the characteristics of the incumbent firms in city j for the fiscal year that ends at the beginning of qtr 0. FrFocused is the fraction of incumbent stores with more than 90% of sales coming from the discount department store industry, FrLowssq is the fraction of incumbent stores owned by firms with inflationadjusted sales per square foot of less than $177 in 1996 dollars, FrHighDebt is the fraction of incumbent stores owned by firms with a total debt to total assets ratio of at least 70%, FrSmFirmSize is the fraction of incumbent stores owned by firms with total firm inflation adjusted sales of less than 1.475 billion in 1996 dollars, FrSmSegmentSize is the fraction of incumbent stores owned by firms with discount department store inflation-adjusted sales of less than $1.475 billion in 1996 dollars, Herf is the Herfindahl measure for the city and is defined as the sum of the squares of the market share held by each incumbent firm i in city j at the beginning of the event window. A time dummy representing the 1980’s is included, but the estimate of the coefficient is not shown. P-values are in parentheses. The estimations shown use White’s adjustment for heteroskedasticity.

1

2

3

FrLowssq

.0210 (.447)

.0193 (.524)

.0204 (.501)

FrFocused

.0680*** (.002)

.0663*** (.006)

.0676*** (.003)

.0632* (.070)

.0634* (.076)

.0633* (.076)

FrHighDebt FrSmFirmSize

.0128 (.823)

FrSmSegment Size Herf.

# Observations R-Squared

.0034 (.933) .0289 (.370)

.0300 (.369)

.0294 (.374)

134

134

134

.1121

.1124

.1121

***, **, or * denote t-statistics significant at the one, five and ten percent respectively

25

Table V Determinants of Exit The dependent variable is the fraction of incumbent stores that exit city j by qtr +20. Wal-Mart enters city j between qtr 0 and qtr +4. We estimate this as a Tobit Model: y* = βo + xβ β + µ where µ/x ∼ Normal (0, σ2) and Fraction Exit j, 0 to +20 = max (0, y*). If #Strs n,j, +20 ≥ #Strs n, j, 0 then y = Fraction Exit j ,0 to +20 = 0. If #Strs n,j,+20 < #Strs n, j, 0 then y = y* = Fraction Exit j, 0 to +20 where N ∑ [Net Reduction #Strs n ,j] 0 to +20 n=1 Fraction Exit j, 0 to +20) = ---------------------------------------------------N ∑ #Strs n ,j ,0 n=1 and [Net Reduction #Strs n ,j] 0 to +20 = [#Strs n, j, 0 − #Strs n, j, +20], #Strs is the number of stores, , j indexes cities, n indexes incumbent firms in city j. Only observations with quarter zero equal to ’91 or earlier are used, as data through quarter +20 is needed. The 115 city observations used in Panel A have prices available in quarter −1 (or qtr –2 if qtr –1 is not available), at least one incumbent firm at the beginning of the event window, no foreign owned or franchised stores. The dependent variable is in decimal form. The independent variables are measured for each city j using data for the year ending at qtr 0. A time dummy representing the 1980’s is included, but the estimate of its coefficient is not shown. The 86 city observations used in Panel B are the observations where debt is known for all incumbents in the city. NOTE: The Tobit coefficient estimates are not the partial effects of the conditional expectations. P-values are in parentheses. Panel A: Fraction of Incumbent Stores Exiting by Quarter +20: All Cities 1 2 3

4

FrLowssq

.7197*** (.000)

.7371*** (.000)

.7255*** (.000)

.7363*** (.000)

FrFocused

.2653*** (.010)

.2854*** (.000)

.2705** (.014)

.2900*** (.007)

− .0834 (.668)

FrSmFirmSize

− .0220 (.890)

FrSmSegmentSize

−.1657 (.460)

FrPrivate

Herf.

Pseudo R-Squared # Observations

−.4914*** (.002)

−.5077*** (.002)

−.4967*** (.003)

−.5015*** (.002)

.4061

.4073

.4062

.4096

115

115

115

115

***, **, or * denote t-statistics significant at the one, five and ten percent respectively

26

Table V (continued)

Panel B: Fraction of Incumbent Stores Exiting by Quarter +20: Debt Known for All Incumbents 1

2

3

FrLowssq

.6691*** (.000)

.6214*** (.000)

.6315*** (.000)

FrFocused

.3029** (.022)

.2656** (.048)

.2802** (.035)

FrHighDebt

.5770*** (.002)

.6344*** (.001)

.6151*** (.002)

FrSmFirmSize

.3118 (.298)

FrSmSegmentSize

Herf.

Pseudo R-Squared # Observations

.1912 (.365) −.5140** (.016)

−.4401** (.044)

−.4517** (.040)

.4328

.4413

.4392

86

86

86

27

Table VI Price Levels and Subsequent Exit The dependent variable is the fraction of incumbent stores that exit city j by qtr +20. Wal-Mart enters city j between qtr 0 and qtr +4. We estimate this as a Tobit Model: y* = βo + xβ β + µ where µ/x ∼ Normal (0, σ2) and Fraction Exit j, 0 to +20 = max (0, y*). If #Strs n,j,+20 ≥ #Strs n, j, 0 then y = Fraction Exit j , 0 to +20 = 0. If #Strs n,j,+20 < #Strs n, j, 0 then y = y* = Fraction Exit j, 0 to +20 where N ∑ [Net Reduction #Strs n ,j] 0 to +20 n=1 Fraction Exit j, 0 to +20 = ---------------------------------------------------N ∑ #Strs n ,j ,0 n=1 and [Net Reduction #Strs n ,j] 0 to +20 = [#Strs n, j, 0 − #Strs n, j, + 20], #Strs is the number of stores, , j indexes cities, n indexes incumbent firms in city j. Only observations with quarter zero equal to ’91 or earlier are used, as data through quarter +20 is needed. The 115 city observations used have prices available in quarter −1 (or qtr –2 if qtr –1 is not available), at least one incumbent firm at the beginning of the event window, no foreign owned or franchised stores. The dependent variable is in decimal form. Relative Prices for city j are measured at quarter –1; The Herfindahl measure for city j is measured at quarter 0. A time dummy representing the 1980’s is included, but the estimate of its coefficient is not shown. NOTE: The Tobit coefficient estimates are not the partial effects of the conditional expectations. P-values are in parentheses.

Relative Prices City j Herfindahl City j

#Observations Pseudo R-Squared

1.0205** (.018) − .7568*** (.000) 115 .1246

***, **, or * denote t-statistics significant at the one, five and ten percent respectively

28

Table VII Difference in Mean Characteristics of Retrenching vs Non-Retrenching Firms The 180 cities used have prices available in quarter −1 (or quarter –2), have at least one incumbent firm at the beginning of the event window, and no foreign owned or franchised stores. Wal-Mart enters city j between qtr 0 and qtr +4. If prices are not available in quarter −1, the city is included if prices are available in quarter −2. The sample consists of 307 firm i - city j observations. In 104 cases firm i has less stores in city j at the end of the event window. In the remaining 203 observations, no reduction in stores occurred for firm i, in city j over the event window. Two-sample t-tests with equal variances assumed are shown below to test if the characteristics of retrenching firms differ from those that do not retrench. Each firm characteristic is measured at quarter 0. These are defined as: a Focus dummy = 1 if discount sales ≥ 90% of firm sales; total debt to total assets for firm i (not available for the private firms); inflation adjusted discount sales per square foot for firm i (1996 dollars); inflation adjusted discount sales for firm i in millions (1996 dollars); inflation adjusted parent sales for firm i in millions (1996 dollars); a dummy =1 if the firm does not have publicly traded stock (private dummy).

Variable

#Firm i - City j Pairs

Mean for Firm i - City j Pairs

#Firm i - City j Pairs

Mean for Firm i - City j Pairs

Focus Dummy

Firm Retrenches 104

Firm Retrenches .404

Firm Does Not Retrench 203

Firm Does Not Retrench .246

t-stat Ho: Mean (diff) = 0 2.88***

Debt Ratio

89

.723

185

.611

6.09***

Inf. Adj. S/Sq ft (’96 dollars) Inf. Adj.Disc. Sales (000’s) (’96 dollars) Inf. Adjusted Parent Sales (000’s) (’96 dollars) Private Dummy

104

$154.26

203

$203.86

8.49***

104

$4518.60

203

$15876.85

8.33***

104

$7824.89

203

$20414.68

8.65***

104

.144

203

.123

.605

***, **, or * denote t-statistics significant at the one, five and ten percent respectively

29

Table VIII New Entrant Distance to Incumbents The dependent variable is the distance in miles between the new entrant’s store and the incumbent store in city j. This sample contains cities Wal-Mart entered in 1987, 1991 and 1995. 205 cities and 665 firm i – city j pairs are represented in the tests below. The independent variables are measured for firm i for the fiscal year ending when Wal-Mart enters: (1) Focus Dummy: Equals 1 if discount sales ≥ 90% of firm sales for firm i , otherwise = 0; (2) High Debt Dummy: Equals 1 if the total debt to total assets ratio for firm i ≥ 70% (not available for some firms as they are privately held); (3) LnDiscSales: natural log of inflation adjusted discount department store sales for firm i (’96 dollars); (4) Inf. Adj. Sales/SqFt: Inflation Adjusted Sales per Square Foot for firm i (’96 dollars); (5) LnFirmSales: natural log of inflation adjusted firm sales for firm i (’96 dollars); (6) #Incumbent Stores: The number of stores in city j when Wal-Mart enters (a proxy for the size of city j) All estimations include 1991 and 1995 year dummies, however, the coefficient estimates are not shown. The estimations shown use White’s adjustment for heteroskedasticity. Columns 2 & 4 consist of the sub-sample of incumbent firms where the debt ratio is known (ie: public).

Focus Dummy

Public & Private

Public Only

Public & Private

Public Only

1

2

3

4

− 1.018** (.020)

− .7574* (.100)

− 1.1063** (.041)

− 1.1216* (.071)

− 0.6314* (.076)

High Debt Dummy LnDiscSales

Inf Adj. Sales/SqFt

− 0.0848 (.493)

− 0.2146 (.167)

0.0002 (.964)

0.0035 (.409)

LnFirmSales

#Incumbent Stores #Observations R-Squared

− 0.6111* (.083)

− 0.0001 (.969)

0.0027 (.505)

− 0.0899 (.564)

− 0.2830 (.200)

0.6049*** (0.000)

.6193*** (0.000)

0.6059*** (0.000)

0.6200*** (0.000)

665

597

665

597

.3932

.3985

.3931

.3980

***, **, or * denote t-statistics significant at the one, five and ten percent respectively

30

Appendix I

A. Model Base Case We start with the original model developed by Hotelling (1929). Assume a “linear city” of length 1, with consumers uniformly distributed over its length. There are two stores belonging to different chains located at the extremes of the city; store A at x = 0, and store B at x = 1.25 The stores sell homogeneous goods, but one can have lower unit costs than the other. Let CA and CB represent their respective one unit costs. Consumers incur transportation cost t per unit of length. Since we use a variant of Hotelling’s model by assuming quadratic instead of linear costs, a consumer living at x incurs a cost tx2 to go to store A and a cost of t(1-x)2, to go to B. Thus, if prices charged by the two stores are PA and PB, total price to this customer is PA + tx2 to buy from A and PB + t(1-x)2 to buy from B. Each consumer demands one unit of the good per period and has a reservation price of R. As a starting point, we use the two period model of Klemperer (1987). Since these models assume switching costs in the second period, firms retain the customers they attract in the first period and can charge them up to their reservation price, R, in the second period.

Assuming quadratic transportation/switching costs, and that firms charge prices PA and PB simultaneously in period one and R in the second period, we look for the Nash equilibrium in prices. We also assume prices are close together so that both firms capture positive market share and they are low enough so all customers buy.

Proposition 1: The equilibrium period-1 prices are : PA = t + 4/3 CA + 2/3 CB – R and PB = t + 4/3 CB + 2/3 CA – R. Proof: A consumer located at x is indifferent between shopping at A or B if his total price including transportation costs is equal across stores. That is: PA + tx2 = PB + t(1-x)2 or x = 1/2 + (PB – PA)/2t.

25

With the cost function we use in this model, locating at the extremes is an equilibrium outcome. See d’Aspremont, Gabszewicz, and Thisse (1979) and Economides (1986) among others.

31

Given this formulation of the problem, x represents the portion of the market captured by store A. Thus A captures a larger portion of the market if PA < PB. If PA = PB, both split the market equally in the first period, and because of switching costs in the second period retain that share. Thus A’s two period profit, ΠA, is: ΠA = (PA – CA)(1/2 + (PB – PA)/2t) + (R – CA)(1/2 + (PB – PA)/2t)

(1)

Equating ∂ΠA /∂PA to zero, gives the reaction curve of PA to a given PB, results in:

2PA = t + PB +2CA –R

(2)

Similarly, the reaction curve of PB to a given PA is: 2PB = t + PA +2CB –R

(3)

Substituting PB from (3) into (2) gives the desired expression for PA. The desired expression for PB is by symmetry.

QED

The proposition shows that an incumbent’s equilibrium price is positively related to its own cost and that of its competitor and is negatively related to a customer’s reservation price in the second period. Thus, markets with an inefficient (high cost) competitor display higher prices in equilibrium. A firm with higher costs charges a higher price. This permits its more efficient competitor to raise prices and increase its own profits. Thus weaker competition results in higher average prices in such markets.26 The reason equilibrium prices are negatively related to R is a result of the following trade off in a two period model. Since this model assumes stores retain their first period customers for the second period and charge them their reservation price, higher R makes the second period profit more attractive. To maximize this profit opportunity, firms try to capture a larger share of the market in the first period. However, this necessitates a lower price. Thus, as R increases, prices in the first period decrease.

26

Similar results occur if we also introduce a probability of exit for incumbents as in Chevalier and Scharfstein’s (1996). Average prices are higher in markets with higher probability of exit. Reasons are similar to markets with a high cost or low efficiency incumbent. A firm with a higher probability of exit is more inclined to sacrifice future market share and charges a higher current price. This allows its competitors to also raise price, making the average price higher. Since high leverage is related to higher probability of exit, markets with high debt firms should see higher average prices.

32

B. New Entry Next, we allow for new entry. A store belonging to firm C, with cost CC per unit enters the market with the two firms A and B located as above.27 C has to make two decisions, where to locate, and given the location what price to charge assuming that A and B will respond optimally by changing their prices to maximize their profits given C’s location.28 In effect we are assuming that prices are again determined simultaneously as a Nash equilibrium, after C’s location is observed. We solve for the equilibrium by identifying, for every possible location for C, say y, the PA, PB, and PC which maximize each store’s profits given the other stores’ reaction function. The y we are interested in gives C the highest profits over all other y.

It is instructive to draw a diagram of the linear city. 0______ xCA__ y______xCB__________1 A

C

B

A and B are at 0 and 1 as before. C locates at some y between them. Now consumers lying between 0 and y, choose between A and C, while those lying between y and 1 choose between C and B. Given PA, PB, PC and y, we refer to xCA as the consumer who is indifferent between shopping at C and A, and xCB as the consumer indifferent between C and B. Thus, xCA will represent portion of the market between 0 and y captured by C, and xCB as the portion between C and B captured by C. Thus the total market share captured by C will be xCA + xCB for the given prices and location. Since xCA is indifferent between C and A, as before: 2 PC + txCA = PA + t(y - xCA)2 . This gives:

xCA = y/2 + (PA – PC)/2ty, which is similar to the expression for market share without entry except for y. By symmetry, A’s market share now is: xA = y/2 + (PC – PA)/2ty. It can similarly be shown, that xCB = (1 – y)/2 + (PB – PC)/2t(1 – y), xB = (1 – y)/2 + (PC – PB)/2t(1 – y). 27

We do not study the entry decision of C, only its location decision after the entry decision has been made. However, the location decision should be the same whether or not the entry decision is considered. 28 Here we are restricting A and B to react only through prices and not through spatial competition either through a change in location or by adding more stores. For quadratic cost functions, it remains optimal to maintain maximal dispersion. Thus neither A nor B are better off by leaving the extremes. Also, opening another store reduces profits through both cannibalization and lowering of prices because of reduced distance between stores. It can be shown that under some weak assumptions, it is also not optimal to respond by opening another store. Proofs are available from authors on request.

33

Proposition 2: Given PA, PB and y, the price that maximizes C’s profits is given by: 2PC = ty(1 – y) + PA(1 – y) + PBy + 2CC – R. Proof: ΠC = (PC - CC)( xCA + xCB) + (R - CC)( xCA + xCB), and (xCA + xCB) = ½ + (PA – PC)/2ty + (PB – PC)/2t(1 – y) = (1/2ty(1-y))(ty(1 – y) + PA(1 – y) – PC + PBy). Thus, ΠC

= (1/2ty(1-y))(ty(1 – y) + PA(1 – y) – PC + PBy) (PC + R – 2CC).

Taking f.o.c. of ΠC with respect to PC and simplifying, results in the required expression for optimal PC. This is C’s reaction function for different PB.

QED

Like in Proposition 1, the equilibrium PC is positively related to its own cost and competitor prices, which are functions of their own cost as is made explicit in the next proposition. It is also dependent on y in an intuitive way. As y increases, C locates closer to B and B’s price becomes more important in determining PC. Similarly PA becomes more important when C locates closer to A. Proposition-1 shows that equilibrium prices are more sensitive to own costs than other’s prices/costs. Thus, the store with higher costs will charge higher prices. Let this be incumbent B. This suggests that C would benefit by locating closer to B, as that would enable it to charge a higher price to capture the same portion of the market between them, than it would need against A. However, locating closer to B has a downside too. As C locates closer to B, switching costs between B and C decrease, putting downward pressure on prices. Equilibrium prices and location result from balancing these tradeoffs. Propositions 3 and 4 formalize this intuition.

Proposition 3: Given y and PC as above, the equilibrium prices, PC, PA and PB as functions of only exogenous variables are: PC = ty(1 – y) + 2/3 CA(1 – y) + 2/3 CBy + 4/3 CC – R, PA = ty/2 + 4/3 CA + 1/3 y(CB - CA) + 2/3 CC – R, and PB = t(1 – y)/2 + 1/3 CA + CB + 1/3 y(CB – CA) + 2/3 CC – R. Proof : A’s profit in both periods comes from the portion of the market it captures from C in the first period. Like B and C it charges R to its customers in the 2nd period. Therefore: ΠA = (PA – CA)(y/2 + (PC – PA)/2ty) + (R – CA)(y/2 + (PC – PA)/2ty). Taking the f.o.c. with respect to PA and simplifying gives,

34

2PA = ty2 + PC + 2CA –R. This is the reaction function for A for different PC. To get the equilibrium PA, we need only substitute for PC. Before that, though, we need to express PC as a function of only the exogenous variables. We achieve that by substituting the above expression for PA and the following expression for PB into the expression for PC in Proposition-2. The reaction curve for B, given PC and y can be shown to be: 2PB = t(1-y)2 + PC + 2CB – R. Substituting for PA and PB gives us the necessary expression for PC: PC = ty(1 – y) + 2/3 CA(1 – y) + 2/3 CBy + 4/3 CC – R. Substituting this expression for PC into above expressions for PA and PB gives us the equilibrium PA and PB to be: PA = ty/2 + 4/3 CA + 1/3 y(CB - CA) + 2/3 CC – R, and PB = t(1 – y)/2 + 1/3 CA + CB + 1/3 y(CB – CA) + 2/3 CC – R.

QED.

As before, the equilibrium prices are positively related to own and competitor costs, are more sensitive to own costs, and are negatively related to R. They are also positively related to switching costs and to how far C locates in relation to the incumbents.

The next proposition determines the optimal location of the entrant as a function of the exogenous variables, CA, CB, CC, t, and R.

Proposition 4: The optimal location, y, is a solution to: 2ty3 – 3ty2 + ty – 2/3 CA + 2/3 yCA + 2/3 yCB + 2/3 CC – 4/3 yCC = 0. Proof: The optimal location is one that maximizes C’s profits given optimal price responses by incumbents A and B to the choice of y and PC by C. From Proposition 2, ΠC

= (1/2ty(1-y))(ty(1 – y) + PA(1 – y) – PC + PBy) (PC + R – 2CC), and from

Proposition 3, the optimal PA and PB, given y and PC are given by: 2PA = ty2 + PC + 2CA –R, and 2PB = t(1-y)2 + PC + 2CB – R . Substituting for PA and PB in equation for ΠC gives us: ΠC = (1/4ty(1-y))[ty(1 – y) + (ty2 + PC + 2CA –R )(1 – y) – PC + (t(1-y)2 + PC + 2CB – R)y](PC + R – 2CC), or ΠC = (1/4ty(1-y))[3ty – 3ty2 + 2CA(1-y) + 2CBy – PC – R](PC + R – 2CC).

35

Substituting for PC from Proposition 3 reduces this to: ΠC = (1/2ty(1-y))[ty(1 – y) + 2/3 CA(1 – y) + 2/3 CBy – 2/3 CC]2 . Taking f.o.c. of ΠC with respect to y gives: 4ty(1 – y)(t – 2ty – 2/3 CA + 2/3 CB) – (2t – 4ty)[ty(1-y) + 2/3 CA(1 – y) + 2/3 CBy – 2/3 CC] = 0, which simplifies to the cubic equation 2ty3 – 3ty2 + ty – 2/3 CA + 2/3 yCA + 2/3 yCB + 2/3 CC – 4/3 yCC = 0.

QED

The solution to this equation is messy. However, we can find the first derivative of y with respect to CB to see where the new entrant locates as CB increases. To do so we find the total derivative of the cubic, and set dCA and dCC = 0. Then, we get: dy/dCB = (- 1/3 y)/[3ty(y – 1) + t/2 + 1/3 CA + 1/3 CB – 2/3 CC]. This derivative is positive as long as the firms’ costs do not differ by too much, an assumption we have needed all along, and y is not too close to the extremes. Under these conditions, y is increasing in CB. That implies, C chooses to locate closer to B (the weaker incumbent) as CB increases.

A number of interesting observations can be made with the help of a numerical example. For instance, assume the following: t = .20; CA = .85; CB = .86; CC = .83; R = 1.3., making B the weaker incumbent. As can be seen from the table below, C’s profit, ΠC, is maximized at y = .55, closer (though not too close) to B. PC also has an interior maximum, though not necessarily at the same y. PA is monotonically increasing and PB monotonically decreasing in y. This is as expected, since who ever C locates closer to has to charge a lower price because of lower switching costs. The weaker incumbent, B, charges a similar price to A at the equilibrium location for C, i.e., y = .55. This occurs even though B has higher costs and would prefer to charge higher. Because C has located closer to it, B’s ability to charge higher is reduced. It is also trivial to show that prices are lower after C enters than before. This is because with three competitors, switching costs are lower for consumers, reducing the ability of firms to charge higher prices. Since C is the most efficient, it makes the highest profits. Though not shown, C also captures the largest market share. The results hold over a reasonable parameter space and as long as costs are close enough so all firms capture positive market shares.

36

Two Period Model with Quadratic Costs: Entry Decision of C at Optimal y t = .20; CA = .85; CB = .86; CC = .83; R = 1.3.

y 0.1 0.20 0.3 0.4 0.5 0.53 0.55 0.57 0.6 0.7 0.8 0.9

P*c 0.392 0.407 0.417 0.424 0.427 0.427 0.427 0.426 0.425 0.420 0.411 0.397

P*a 0.397 0.407 0.418 0.428 0.438 0.441 0.444 0.446 0.449 0.459 0.469 0.480

P*b 0.487 0.477 0.468 0.458 0.448 0.445 0.444 0.442 0.439 0.429 0.419 0.410

Πc 0.02844 0.03403 0.03913 0.04267 0.04444 0.04463 0.04467 0.04464 0.04446 0.04286 0.04011 0.03872

Πa 0.0002 0.0007 0.0026 0.0049 0.0073 0.0081 0.0086 0.0091 0.0099 0.0124 0.0150 0.0176

Πb 0.0125 0.0103 0.0081 0.0060 0.0040 0.0034 0.0031 0.0027 0.0022 0.0007 0.0000 0.0027

37

Appendix II Adjustments for ACCRA Discount Items Replacement

In the fourth quarter of 1991 the ACCRA replaced Johnson’s Baby Shampoo with Alberto VO5 shampoo. We adjust the prices of the items after the switch to reflect these changes by replacing the new price with the price of the old item had no switch occurred. Formally, 9 ∑ [(P j,i ,q − P j,i,q-1 )/ P j,i,q-1] i=1 DPC j,q ≡ Average Discount Item Price Change j,q = ------------------------------------9 where P represents item price, j indexes the city, and i indexes the nine discount items in city j not switched during quarter q.

Price Old Item j, q– 1 (1+ DPC j,q ) Adj. Price New Item j, q + n = ----------------------------------------- X Price New Item j, q + n Price New Item j, q where q is the quarter that the item is switched, and n indexes the number of quarters after the switch.

NOTE: In some cases we had to go forward to quarter q+2, or backwards to quarter q-3 to get prices before and after the switch. For a couple of cities, prices before and after the switch are not available. In these cases, we use the average multiplier for the other cities defined as: [Price Old Item j, q– 1 (1+ DPC j,q )]/ Price New Item j, q .

38

REFERENCES American Chamber of Commerce Researchers Association, Cost of Living Index, 1982 – 1997. Berger, Phillip G., and Eli Ofek, 1995, Diversification’s effect on firm value, Journal of Financial Economics 37, 39-65. Campa, Jose Manuel and Simi Kedia, 2002, Explaining the diversification discount, Journal of Finance, forthcoming. Campello, Murillo, 2002, Capital Structure and product market interactions: Evidence from business cycles, Journal of Financial Economics, forthcoming. Chevalier, Judith A., 1995, Do LBO supermarkets charge more? An empirical analysis of the effects of LBOs on supermarket pricing, Journal of Finance 50, 1095 – 1112. Chevalier, Judith A., and David S. Scharfstein, 1996, Capital market imperfections and countercyclical markups: Theory and evidence, American Economic Review 86, 703 – 725. d’Aspremont, Claude, Jean J. Gabszewicz, and Jacques F. Thisse, 1979, On Hotelling’s stability in competition, Econometrica 17, 1145-1151. Directory of Discount Department Stores, Directory of Discount Stores and Directory of Discount and General Merchandise Stores, Business Guides Inc., New York, NY, 1983 – 1996. Discount Merchandiser, Schwartz Publications, New York, NY, 1983 – 1996. Economides, Nicholas, 1986, Minimal and maximal product differentiation in Hotelling’s duopoly, Economic Letters 21, 67-71. Fluck, Zsuzsanna, and Anthony W. Lynch, 1999, Why do firms merge and then divest? A theory of financial synergy, Journal of Business 72, 319-346. Graham, John R., Michael L. Lemmon, and Jack Wolf, 2002, Does corporate diversification destroy value? Journal of Finance 57, 695-720. Hotelling, Harold, 1929, Stability in competition, Economic Journal 39, 41 – 57. Khanna and Tice, 2000, Strategic responses of incumbents to new entry: The effect of ownership structure, capital structure and focus, Review of Financial Studies 13, 749 – 779. Khanna and Tice, 2001, The bright side of internal capital markets, Journal of Finance 56, 1489 – 1528. Klemperer, Paul, 1987, Markets with consumer switching costs, Quarterly Journal of Economics 102, 375 – 394.

39

Klemperer, Paul, 1995, Competition when consumers have switching costs: An overview with applications to industrial organization, macroeconomics and international trade, Review of Economic Studies 62, 515-539. Lamont, Owen, 1997, Cash flow and investment: Evidence from internal capital markets, Journal of Finance 52, 83-109. Lang, H.P., and Rene M. Stulz, 1994, Tobin’s q, corporate diversification, and firm performance, Journal of Political Economy 102, 1248-1280. Lewellen, Wilbur G., 1971, A pure financial rationale for the conglomerate merger, Journal of Finance 26, 521 – 537. Mansi, Sattar A., and David M. Reeb, 2002, Corporate diversification: What gets discounted? Journal of Finance, forthcoming. Maksimovic and Phillips, 2002, Do conglomerate firms allocate resources inefficiently? forthcoming Journal of Finance. Meyer, Margaret, Paul Milgrom and John Roberts, 1992, Organizational prospects, influence costs, and ownership changes, Journal of Economics and Management Strategy 1, 9-35. Phillips, Gordon M., 1995, Increased debt and industry product markets: An empirical analysis, Journal of Financial Economics 37, 189 – 238. Rajan, Raghuram, Henri Servaes, and Luigi Zingales, 2000, The cost of diversity: The diversification discount and inefficient investment, Journal of Finance 55, 35 – 80. Scharfstein, David S., 1998, The dark side of internal capital markets II: Evidence from diversified conglomerates, Working paper, MIT and NBER. Scharfstein, David S., and Jeremy C. Stein, 2000, The dark side of internal capital markets: Divisional rent seeking and inefficient investment, Journal of Finance 56, 2537 – 2564. Shin, Hyun-Han, and Rene M. Stulz, 1998, Are internal capital markets efficient?, Quarterly Journal of Economics 113, 531-552. Stein, Jeremy C., 1997, Internal capital markets and the competition for corporate resources, Journal of Finance 52, 111-133. Villalonga, Belen, 2002, Diversification discount or premium? New evidence from BITS establishment-level data, Harvard Business School working paper. Whited, Toni M., 2001, Is it inefficient investment that causes the diversification discount? Journal of Finance 56, 1667-1691. Zingales, Luigi, 1998, Survival of the fittest or the fattest? Exit and financing in the trucking industry, Journal of Finance 53, 905 – 938.

40