E-commerce and prices - theory and evidence

Richard Friberg, Mattias Ganslandt Mikael Sandstrmy ,

Abstract: We construct a model which captures some key features of competition between Internet retailers and conventional stores. The model derives demand from consumers maximization problems - a key feature is the trade-o between convenience costs of shopping in the regular stores vis--vis transport and navigation costs on the Internet. In particular we are interested in the impact of Internet as a retail channel on prices in dierent locations. Predicitions of the model are tested on the Swedish markets for books and CDs. The sample includes price information on a number of well specied goods from all Swedish e-commerce sites and a very large sample of conventional stores including background information on these stores. JEL codes: L11, L81, D12, D43 Keywords: retail pricing, consumer behavior, e-commerce, price discrimination

Correspondence: Mattias Ganslandt, University of Colorado at Boulder, Department of Economics, CB 256, Boulder CO 80309-0256, USA, e-mail: mattiasg@iui.se. Part of this research was conducted when Ganslandt was visiting the Department of Economics at University of Colorado at Boulder. Their hospitality is gratefully acknowledged. Ganslandt also wants to thank The Swedish Foundation for International Cooperation in Research and Higher Education (STINT) for nancial support. y Swedish Research Institute of Trade, SE-103 29 Stockholm, Sweden. E-mail: mikael.sandstrom@hui.se.

E-commerce and prices - theory and evidence

Introduction

While currently only a small share of total sales in almost any retail market is through Internet, the future potential may be great - many commentators see e-commerce as enabling a frictionless economy and implying the death of distance. This in turn impacts issues such as the strength of competition on markets, the constraints facing scal and monetary policy on a national level and economic geography (where transport costs and distance play a central role; see for instance Fujita, Krugman and Venables, 1999). Consequently, the rapid growth of e-commerce has sparked tremendous attention amongst policy makers, media and led to a large number of management books. 1 In contrast, there has been little formal modeling or comprehensive empirical analysis of e-commerce. This paper attempts to help ll this gap. We construct a model which captures some key features of competition between Internet retailers and conventional stores. In particular we are interested in the impact of Internet as a retail channel on prices in dierent locations. Predictions of the model are tested on the Swedish markets for books and CDs. The sample includes price information on a number of well specied goods from all Swedish e-commerce sites and a very large sample of conventional stores including background information on these stores. E-commerce is still in its infancy and many underlying factors can be expected to change in the future. Such factors are the ease of navigating shopping sites, the cost of transporting goods to the consumer (for instance downloadable books and music), and the share of the population having access to the internet. For these reasons it is important to consider not only current empirical evidence but also to try understanding the mechanisms at play. The model in this paper derives demand from consumers maximization problems - a key feature is the trade-o between convenience costs of shopping in the regular stores vis--vis transport and navigation costs on the Internet. The share of consumers with access to Internet is an important parameter in the model. Based on the model we should expect a discrete fall in prices in the conventional retail stores once Internet usage reaches a critical level. This prediction is generally premature to test but is an important caution not to dismiss future price eects of the Internet based on current evidence.2 The
See for instance OECD (1999), U.S. Government Working group on Eclectronic Commerce (1999), Shapiro and Varian (1999) or Schwartz (1999). 2 Stock trades - some circumstancial evidence?
1

E-commerce and prices - theory and evidence

model also stresses that equilibrium prices may be both higher or lower on the Internet than in conventional stores, it will depend on the perceived costs of shopping in conventional stores and on navigation and transport costs. Based on a combination of these predictions we get the somewhat surprising result that prices may initially be lower online than in conventional stores, while the relation is reversed in the long run. A nal prediction of the model is that retailers with both an internet and regular presence (where the two sales channels may serve as an instrument for second degree price discrimination) will have higher online prices than pure Internet retailers. This prediction is borne out by the empirical analysis. To date there exists little previous theoretical analysis of competition that pays attention to the institutional characteristics of the Internet. The perhaps closest in modeling terms is Anderson and Ginsburgh (1999) who examine deviations from LOP when there is both second and third discrimination. Even though our model could be used to discuss such (and many other) issues we choose to focus on the above questions. It should also be noted that our focus is on the interaction between traditional and online markets. The interaction between e-retailers, which is an important eld of study, is beyond the scope of this paper. The limited empirical evidence on e-commerce that exists so far has largely focused on the issue of whether it is cheaper to buy over the internet, Smith, Bailey and Brynjolfsson (1999) provide a survey. As regards price levels Lee (1997) nds that prices for used cars are higher on the net and Bailey (1998) nd the same for books, CDs and software. Brynjolfsson and Smith (1999) examine the prices of books and CDs in regular and internet stores concluding that prices are lower on the Internet. A number of issues leads one to question the generalizability of these results however - Lee does not control for quality dierences and Brynjolfsson and Smith only include data from four conventional stores dispersed across the US. As we shall see in our empirical analysis there is very substantial price dispersion across conventional stores, making it important to have a representative sample of stores to be able to make statements that generalize beyond the specic sample. Sweden oers a good laboratory for examining e-commerce since it has been in the forefront of Internet development. An estimated 0.7 percent of Swedish 1999 retail sales were over the Internet (only the US has a higher share, 1.2 percent). The homogeneity of sales taxes and costs across locations in Sweden facilitates analysis compared to the US - where dierences in local sales taxes appears to be an important factor driving internet

E-commerce and prices - theory and evidence

buying (Goolsbee, 1998). The next two sections presents the formal model and section 4 analyses the empirical evidence, while the last section concludes.

The Model

There are n identical, local markets denoted with subscripts A; B; C; :::. Without loss of generality we will focus attention on market A. In each market there is a continuum of consumers with valuations uniformly distributed on [0; 1]. Assume further that a uniformly distributed fraction v of these consumers has access to Internet. Such a consumer in market A has utility t pA u = > T pI : 0
8 > <

traditional shopping online shopping no cons.

(1)

where pA is the price in the local store and t the opportunity cost of shopping (high-income consumers have high valuations but also higher opportunity costs), pI is the price online and T the xed cost of buying the goods online (this includes access costs a as well as pure transportation costs ). For consumers without access to Internet utility is given by u=
(

t pA 0

traditional shopping no cons.

(2)

The supply-side is very simplied. The sum of the local markets (cities) is the total market (the country). Local markets are indexed with subscripts. In market A there is a single local retail store setting a price pA with marginal cost cA. A local rm can sell in the local market only. Online stores can sell in all local markets but has to charge the same price, pI , in all markets. They produce at marginal cost cI . We are rst analyzing the eect of internet competition on local prices under the assumption that there is free entry of online retail stores. In equilibrium the price online has to equal average cost pI = ACI . In the following section we treat the online price as parametric for simplicity. We are interested in an equilibrium in which both types of rms can be active. The average cost of online rms is, therefore, assumed to be suciently low to allow for entry in the segmented equilibrium, that is ACI + T < p + t. A

E-commerce and prices - theory and evidence

3 3.1

Prices on the Internet and in local markets Does Internet lead to lower local prices?

The rst issue that we want to examine is if e-commerce does indeed exert a competitive pressure on prices in local markets. Our aim is to illustrate the conditions for a pro-competitive eect as well as the magnitude of the eect. To do this we rst derive the local prices in the absence of Internet and under perfect segmentation before proceeding to a comparison of local prices when there are Internet retailers.

3.1.1

Perfect segmentation

Consider a perfectly segmented equilibrium without online stores. We can think of this benchmark as a situation in which consumers have high search costs or prohibitive transportation costs across markets, but perfect information and no transactions costs (except the opportunity cost of shopping) within the local market. A consumer with valuation would buy from the local store if t pA 0 which implies a cut-o value, A (pA ) or A for short, where p A = A 1t and the local monopolist solves max (1 A) (pA cA ) p
A

(3)

(4)

(5)
#

which has a rst order condition

d A (1 A) (pA cA ) = 0 dpA 1 + cA t 2

(6)

which gives an optimum at p = A (7)

where the superscript denotes the equilibrium and p is henceforth referred A to as the local monopoly price.

E-commerce and prices - theory and evidence

3.1.2

Internet competition

Next, consider the impact of Internet competition. In the population with access to Internet a consumer with valuation would buy from an online store if T pI t pA 0 (8)

which gives an indierent consumer with valuation e (pI ; pA) or e for short, where
e

1 = min (pI + T pA) ; 1 t

(9)

and the demand at the local store in this population is DA (pA ; pI ) =

if pA (pI + T ) (1 t) if pA > (pI + T ) (1 t)

(10)

3.1.3

Yes, and it says bang!

The local rm solves max (1 v) (1 A ) + v e A p

(pA cA )

(11)

which has one or two local optima; an interior and a corner solution. First, consider the interior local optimum. In the interior solution the local rm sets a price which yields revenues from both populations (consumers with access as well as without access to online stores). The interior rst order condition is
" # " # d A de (1 A) (pA cA ) v 1 e (pA cA) = 0 dpA dpA

(12)

which is partly identical to the rst order condition in the segmented equilibrium, eq. (6). The rst term on the left hand side is the monopolists incentive to maximize prots while the second term in is the pro-competitive eect of online-competition. The pro-competitive eect shows that the direct eect on prots of a price increase is less than in the segmented equilibrium as a share of consumers is now buying from the online stores but also result in a ow of consumers to the online retailers. The pro-competitive eect is increasing in the share of consumers with online access (v).

E-commerce and prices - theory and evidence

The interior local optimum, however, is not always the global optimum. For very low prices online, such that e A < 0 at p , and a small frac A tion with online access, the local store prefers to sell at a high price to the population without access rather than trying to steal some customers from the online stores. More precisely, for suciently low v (close to zero) the corner solution is the global optimum and the equilibrium price is p . HowA ever, as the fraction with access increases, the weight of the revenues from the online-population increases in the local rms prot function. At some critical fraction v the equilibrium shifts to the interior solution and the price falls discretely from p to pv . The critical value is (implicitly) given by A A
e (1 v ) v 1 A
v

(pv cA ) = (1 v) (1 ) (p cA) A A A

(13)

where the superscripts refer to the value in the local optimum. The following proposition states the result: Proposition 1 If only a small fraction of the population has access to online stores, v v; the equilibrium price in the local market is equal to the local monopoly price p ; but if the fraction is suciently large, v > v; the A equilibrium price falls to pv , where pv is falling in v and pv < p for all A A A A v 2 (0; 1]. Proof. (Sketch) For v = " and low pI + T the monopoly prot in population (1 v) is larger than the duopoly prot in the whole population. For v = 1 the opposite is true. From the theorem of intermediate values there exist a v, such that (13) holds with equality. The solution is unique and equals [cA (1 t) (pI + T )] 2 v= 1+ [tcA + (1 2 (pI + T )) (1 t) t] (1 t cA)
!1

(14)

evaluated at cA. The rest of the proposition follows from (12). From the rst order condition it follows that the optimal price, pv , in the interior solution A is falling in the fraction of the population with access to online stores (v), increasing in the price and cost of online shopping (pI ; T ), and increasing in the marginal cost (cA). In gure 1 we illustrate the equilibrium price in the local market as a function of the fraction with Internet access. Interestingly, the critical value v is falling in pI + T . We conclude that if the price in the online store is low, a pro-competitive eect of Internet competition in the local market is less likely, i.e. a larger

E-commerce and prices - theory and evidence

Figure 1: The pro-competitive eect of internet competition fraction of the population must have access to Internet, but stronger. In other words, if the price on the Internet is very low it is less attractive for the local store to try to compete for the customers with Internet access. If it is optimal to set price so that you are also able to attract online users, on the other hand, the pro-competitive eect will be stronger the lower the Internet price is. To summarize, the pro-competitive eect of Internet competition depends on the degree of access to online retailers. The result suggests that equilibrium prices can be radically dierent in a situation with relatively limited access to internet retailers than in a mature market with general access to internet and acceptance of e-commerce. It should, therefore, be stressed that we have to be cautious with the interpretation of the relation between local prices and online prices as long as the market is immature. More specically, if a relatively small fraction of the population has access to Internet it is likely that prices online are lower than in traditional stores since the local prices are at their initial, high, level. The traditional stores prefer to have high revenues in the population without access to e-commerce, rather than compete for consumers with access to lower prices at online stores. However,

E-commerce and prices - theory and evidence

as a larger proportion of the population gains access to online stores we can expect a drastic eect of Internet competition in the traditional market. At some point it is no longer protable for traditional rms to neglect the online market and, hence, the local price will fall discretely.

3.2

Are prices lower on the net?

Having focused on the pro-competitive eect of Internet competition in the local market we now characterize the relative prices in local stores and online stores. We do so under the assumption that everyone has access to the Internet, that is v = 1. The local store has to charge a price which is suciently low, i.e. pA < (1 t) (pI + T ), or else it would not have any positive sales. The local retail store solves the program max e A (pA cA) p
A

(15)

and the rst order condition is the limit of (12) as v ! 1, i.e.

de d + A (pA cA) = 0 dpA dpA

(16)

which yields a local price pA = (1 t) (pI + T ) + cA 2 (17)

and, hence, the optimal price is increasing in the online price including transportation costs and it is falling in the opportunity cost of traditional shopping. Moreover, the price online, pI can be higher or lower than pA . For relatively high access costs and relatively low opportunity costs in traditional stores, the price in traditional stores are higher than online prices. For relatively low access costs or relatively high opportunity costs, however, the price online is higher. More formally, we have the following result: Proposition 2 For a given price online, pI , the optimal price in the local store, pA; is increasing in the access cost to online stores (T ) and decreasing in the opportunity cost of shopping (t).

E-commerce and prices - theory and evidence

Proof. We dierentiate eq. (17) w.r.t. T and t to obtain @ pA 1t = >0 (18) @T 2 @ pA (pI + T ) = <0 (19) @t 2 and the rest of the proposition follows from the assumption about both stores being active in equilibrium, i.e. pA < (1 t) (pI + T ) ; and eq. (17). The opportunity cost of traditional shopping (t) as well as the cost of online shopping (T ) is going to vary between dierent markets and for different products. The opportunity cost of traditional shopping is determined by a number of factors; including the importance of pre-sale services as well as how much time the transaction is taking. Generally, the opportunity cost of traditional shopping tend to be lower, if the traditional stores provide pre-sale services which are important to the consumer and these services are not provided by the online counterparts. The opportunity cost of traditional shopping tend to be higher, if shopping is time-consuming or stressful relative to the value of the product. Hence, we expect local prices to be relatively lower compared to the online price when traditional shopping is time-consuming or pre-sale services are unimportant. The cost of online shopping is also determined by a number of factors; including direct access and transportation costs as well as the complexity of the transaction. The former costs are straightforward.. The magnitude of the latter, i.e. the complexity of the transaction, is aected by the design of the online shopping system as well as consumer behavior. Some products are recognized by name, brand or trademark, which can easily be coded and the complexity of online shopping can be relatively low. Other products are recognized by shape, color or physical location in the store, which makes digital coding harder and the complexity of the online shopping system potentially higher. Hence, we expect the local price to be relatively higher compared to the online price when the online transportation cost, the access cost or the complexity of online shopping is high. Now, using the previous result we can characterize the exact relationship between the local and online prices. To obtain the price for which both types of rms are active in equilibrium and the optimal price in the local market is equal to the price online we solve eq. (17) for p to obtain the critical price (1 t) T + cA e p (t; T; cA) = (20) (1 + t)

E-commerce and prices - theory and evidence

10

which is a function of the primitive parameters of the model. For suciently e low prices online, i.e. pI < p (t; T; cA), the local price is higher than the price charged by the e-retailers, pA > pI . The intuition for this is straightforward. If the xed cost of online transactions is high or the opportunity cost of conventional shopping is low for a given price online, the conventional retailer can charge a higher price without loosing too many customers. Hence, the price in local stores can be higher than the price in online stores. For a higher e price online, i.e. pI > p (t; T ; cA), the local price has to be lower than the price charged by the e-retailers, pA < pI . The local rm has to compensate the customers for high opportunity costs of conventional shopping and the low xed cost of online shopping makes e-retailers more attractive to the customer. Consequently, the price in the local market has to be lower than the price online. The last result combined with proposition 1 gives an interesting and somee what surprising result. For any online price such that pI 2 (p (t; T; cA) ; p ) A the local price is higher than the online price as long as a small fraction of consumers have access to internet, i.e. v < v, while the local price falls below the price online as soon as a larger fraction of consumers have access to internet. 3 The initial relationship with a lower price online than in the conventional store may, thus, be reversed to a lower price in the conventional store in the long run. This result is particularly important for the empirical analysis of prices in conventional and online stores. The observation that prices are lower online than in conventional stores can either depend on a low transaction cost online and high opportunity cost of conventional shopping or a small fraction of consumers with internet access. In the former case the relationship holds in the long run, while it may be reversed in the latter case for any online price e such that pI > p (t; T ; cA). From an empirical point of view, it should also be noted that the exact long-run relationship depends on various parameters which are not directly observable. As we have previously noted, the opportunity cost of conventional shopping as well as the cost of online transactions depends on a number of factors that are hard to measure. The analysis shows, however, that the relative prices in traditional and online stores depend on opportunity costs, transportation costs and access costs. People with high valuation and high opportunity costs of traditional
3

(e (t; T ; cA ) ; p ) is a non-empty set if 2T + c < 1 + t. p A

E-commerce and prices - theory and evidence

11

shopping prefer to shop online and, thus, accept a higher price. Local prices can be expected to be relatively low compared to the online price when traditional shopping is time-consuming or pre-sale services are unimportant. Moreover, local prices can be expected to be relatively high compared to the online price when the online transportation cost, the access cost or the complexity of online shopping is high.

3.3

Are pure Internet retailers cheaper than mixed retailers?

Having considered the price in the local store we can now turn to the prices in integrated as well as pure online stores. In this section we are, therefore, going to assume that the online price is determined by a prot maximizing rm. The issue of pricing by mixed and pure internet retailers is interesting since it has been noted that in many industries the structure on the internet is changing from being dominated by recent start-ups and pure Internet retailers to being dominated by the rms that have important market shares online but also considerable sales in regular stores. First, consider an integrated rm with a local store as well as a store online. This rm solves max e A (pA cA) + 1 e (pI cI ) p ;p
A I

(21)

and the rst order condition with respect to the local price, pA , is
" e
A

de d de + A (pA cA) (p cI ) = 0 dpA dpA dpA I

"

(22)

and the rst order condition with respect to the price online, pI ,
"

and combining these two conditions we nd that the optimal local price is equal to the local monopoly price as derived in eq. (7), i.e. p = p . A A The intuition for this result is that the integrated rm considers the ow of consumers from the local store to the online store as he sets the local price. A reduction of the local price would result in a ow from the online store to

1e

de de (pI cI ) + (p cA) = 0 dpI dpI A

(23)

E-commerce and prices - theory and evidence

12

the local store which would reduce the revenues in the Internet store. The direct eect on prots as well as the eect on sales (i.e. the eect on the total demand in the local and online store) is identical to the problem of the local monopolist in a segmented equilibrium. However, the optimization problem of the local monopolist (with local sales only) and the integrated rm dier in a very important respect. The integrated rm can use the online market for second degree price discrimination. The integrated rm can set a price in the online store such that high-valuation customers choose to buy online, while consumers with lower valuations choose to buy the goods in the local store. Solving (22) and (23) for pI we nd the optimal price online, which is p = I 1 + cI T 2 (24)

and a necessary condition for opening the Internet market is that the markup on this market is strictly larger than the markup in the traditional store. 4 If the marginal cost in the online store is larger than the marginal cost in the traditional store, the price online (including the charge for transportation) would be higher than in the traditional store or it would not be protable to open the online store. If the marginal cost online is very low relative to the traditional store, the latter would be closed. Next, consider the optimal decision by a pure internet retailer. In this case the online store solves max 1 e (pI cI ) p
I

(25)
#

and the rst order condition is

which has an optimum pe . The main dierence between the rst order conI dition for the independent online store and the rst order condition for the integrated rm (23) is that the integrated rm takes into consideration the ow from the traditional store to the online store when the online price is
which is true for cA + tA > cI > cA (1 tA ) (upper bound from mark-up condition and lower bound from e > A ). If the left hand inequality is violated the rm would not open the online market. If the right hand inequality is violated the rm would close the traditional store.
4 1

de 1e (p cI ) = 0: dpI I

(26)

E-commerce and prices - theory and evidence

13

reduced, reected in the second term in square brackets in (23). Lower demand in the local store would reduce the revenues in the local market. The optimal price online for the integrated rm is, therefore, higher than the optimal price online for the independent rm. Proposition 3 The online price in an integrated rm with local sales is strictly higher than the price online of an independent (pure) online store, i.e. p > pe . I I Proof. Follows from the rst order conditions (23) and (26). This proposition shows that for a given local price pA, the optimal online price of an integrated rms is strictly higher than the optimal price of an independent (pure) online retailer. The intuition for this result is, as the discussion above shows, that the integrated rm is charging a higher price online to avoid stealing customers from its local store. The pure internet retailer, on the other hand, is going to compete for local customers by charging a lower price, neglecting the negative eect on the total industry protability.

Empirical results

We now proceed to an empirical examination of the Swedish markets for books and CDs - two markets that are comparatively well suited for Internet sales: transport costs are relatively low, the goods are physically homogenous and the demand for services is small. They are also mature markets by Internet standards.5 We also collected prices for an industry where the cost of shopping in regular stores is high - grocery. The empirical section ends with a few reections on these prices. We collected the prices of four books and six CDs in a total of 64 conventional book stores and 98 conventional music stores located in 19 cities across Sweden. We also collected prices of these products at the seven and eight Swedish e-commerce sites selling books and CDs, respectively. For both books and CDs the conventional retailers are a sample of randomly picked
Evidenced by that these (US) markets harbor some of the most recognized Internet brand names, such as Amazon and CDnow.
5

E-commerce and prices - theory and evidence

14

stores, while the e-commerce sites are all that were active in Sweden at the time.6 All observations were made during two weeks in October 1999. It is well established that even though, as in this case, we examine physically homogenous goods the prices do not represent identical goods since service and location are also attributes of the good. Clearly a book bought over the Internet is in some ways a dierent good from one bought in a book store in the store in the next block. Similarly, a book in a department store with little assortment and no specialized clerks out in a suburb is not the same thing as a book in a specialized book store in the city center. To control for this we therefore also collected information on the stores: on location, type of store (such as book store, department store, supermarket), assortment (large/small) and whether it belonged to a retail chain. It is in many senses premature to test Propositions 1 and 2 and it is left for future work. Instead we focus on testing Proposition 3 and on providing an analysis of how the prices on Internet relate to prices in regular stores.

4.1

Books

First, we analyze the market for books. We study the prices of four books: a childrens book: Alfons, a recent bestseller: Carambole, a Swedish dictionary: SAOL and an English dictionary: Longman. Figure 2 plots the price of the four books in each location. 7

[FIGURE 2]

A standout feature is the very substantial price dispersion among conventional stores. For instance in the nine Stockholm stores that carried
6 The conventional book stores and music stores were selected at random from the Yellow pages in the 19 largest Swedish cities. 7 Throughout prices are in Swedish kronor (SEK), in January 1st, 2000 the SEK/USD exchange rate was 8.50.

E-commerce and prices - theory and evidence

15

Carambole, price ranged from 239 to 315 (prices were: 239, 265, 288, 288, 288, 295, 298, 302, 315). So the most expensive store had some 30 percent higher price than the cheapest. Secondly, we see from Figure 2 that prices in the e-commerce sites appear lower than prices in conventional stores. Thirdly, there is substantial price dispersion also among the Swedish e-commerce sites. For instance the price of Carambole ranges from SEK 183 to SEK 263.8 It is clear that there by no means is perfect competition on the net. A rst summary is thus that prices of these books are lower on the internet, but that there is considerable dispersion in prices both among conventional stores and e-retailers. Table 1 presents some summary statistics on the price of Carambole and of a (unweighted) basket comprising three books: SAOL, Carambole and Alfons. Longmans dictionary was only available in seven regular stores (with exceptional price dierences) and was therefore excluded from the basket.

[TABLE 1]

On average prices of these books are lower on the internet. The basket is on average close to 15 percent cheaper if bought on the Internet than in conventional stores. A two tailed t-test rejects that the mean price in conventional equals the mean price on the Internet at the 10 percent level. If transport costs are included in the price of the Internet basket is still cheaper on the Internet but the dierence is lower, now the basket is on average some 9 percent cheaper. The dierence is, however, not signicant at any usual levels of signicance using a two tailed t-test. Proposition 3 predicts that mixed stores have higher online prices than the pure internet retailers. This prediction is conrmed, all the pure internet retailers have lower prices than the mixed rms internet prices (the max for pure Internet is lower than the min for the mixed retailers). Given that this is the full population of Swedish internet stores the dierence in means is by denition signicant. The prices of the pure internet retailers are lower than in conventional stores also when transport are included, the hypothesis
8 Prices were: [price including transport within brackets]: 183[222], 202[241], 207 [236], 209[248], 215[254], 259[299], 263[263]

E-commerce and prices - theory and evidence

16

that the mean prices are equal in conventional stores and pure internet stores can be rejected at the 1 percent level both with and without transport costs. Notable is that this holds also for Carambole, even when buying just one book the gain is thus suciently large that the price dierential is not eradicated by transport costs. As noted previously, the price of a good should reect not only if it is bought on Internet, but also on other characteristics of the outlet. Table 2 reports the result of regressions where we examine if how prices depend on characteristics of the outlet. A number of dummy variables are used as explanatory variables - INTERN (1 if the price observation is from a internet retailer), INTPURE (1 if the price observation is from a pure internet retailer), INTMIX (1 if the price observation is from a mixed retailer), BOOKST (1 if the price observation is from a specialized book store), CENTER (1 if the price observation is from a store located in the city center), LARGE (1 if the price observation is from a conventional store with large assortment) and nally NOCHAIN (1 if the price observation is from a retailer that does not belong to a chain).

[TABLE 2]

The regressions conrm the observations from Table 1 - it is cheaper to buy this basket of books on the internet but only if transport costs are not included. As seen above the basket is cheaper also when transport costs are included if the prices of pure internet retailers are compared to the average price. The more aggressive pricing policy of the pure internet retailers is again conrmed by comparing the coecients on INTPU with those on INTDU. Prices of the basket in specialized bookstores (BOOKST) are higher, likely reecting better service but there is some variation across individual books, Carambole is cheaper in specialized book stores. Being located in the center of the town is associated with some 5 percent lower prices. This may be because of greater proximity to competitors is driving prices downwards and is largely due to the lower prices of Carambole in city centers. Explanatory power is the greatest for Carambole - this is a recent bestseller, volumes are likely to be high and the loss of having a price that is not optimally adjusted is higher than for Alfons and SAOL. Independent rms tend to have lower

E-commerce and prices - theory and evidence

17

prices but the eect is generally not statistically signicant. The explanatory value of regressions is reasonably high for cross-sectional data. Of some interest is also comparing prices among conventional stores. There exists a large literature that examines prices of the same good across local markets - see for instance the book edited by Weiss (1989) or Asplund and Sandin (1999) for a recent contribution. As evidenced by Figure 2 the price dispersion is great within locations and there is little obvious signs of price dispersion across locations. The regressions reported in Table 3 partly conrm that pattern - here prices across regular stores are examined. In addition to the above variables these regressions include one cost variable RENT (rental cost of oce space), a measure of the number of rms in the location CONC, and two variables capturing demand conditions, INC (income per capita at county level) and UNIV (a dummy variable for University town).

[TABLE 3]

4.2

Music

Next, we proceed to the market for popular music. The price of six music CDs is examined. Three are international best-sellers at the time of the study: Red Hot Chilli Peppers - Californication (RHCP), Back Street Boys - Millenium (BSB): Shania Twain - Come on Over (TWAIN), one is an international classic: U2 - The Joshua Tree (U2) and two are Swedish bestsellers: Tomas Di Leva - Fr Sverige i Rymden (DILEVA) and Eva Dahlgren - En Blekt Blondins Hjrta (DAHLGREN). The summary statistics show that the price dispersion in conventional stores is substantial. For instance, the six Stockholm retailers selling the BSB album charged prices ranging from 139 to 199 (the prices were 139, 159, 159, 189, 189, 199). So the most expensive store is over 40 percent more expensive than the cheapest. The prices online appear to be lower as well as less dispersed. Prices of BSB, excluding transportation costs, vary approximately 15 percent from 129 to 149. Adding transportation costs raises the price level as well as the dispersion. Prices including transport costs

E-commerce and prices - theory and evidence

18

ranges from 149 to 178 or some 20 percent for BSB. 9 It is also worth noting that prices charged by e-commerce sites including transportation costs are not always lower than prices in traditional stores. Indeed, the lowest price of BSB in one of the conventional stores in Stockholm is approximately 7 percent lower than the lowest price on the internet. Table 4 presents some summary statistics on prices of RHCP, U2 and a basket comprising of 4 current bestsellers - RHCP, BSB, TWAIN and DILEVA.

[TABLE 4]

Looking rst at the basket we see that prices are some 14 percent lower on the Internet than in conventional stores, using a two tailed t-test the hypotheses that the outlets have equal mean prices is rejected. When transport costs are included the dierence between average prices decreases to some 10 percent - it can still be rejected that the two outlets have equal means. If one chooses to buy just one record over the internet the picture changes - prices including transportation costs are close to prices charged by traditional retailers for single item purchases. The dierence between the two is not statistically signicant. On average shipping and handling for a single purchased CD adds 15 percent to the price charged online. It should be noted, however, that the price including transport costs for a single item is an upper bound for online prices. For all eight e-commerce sites the shipping and handling charges are xed and independent of the number of CDs bought and the average transportation cost falls when more than one item is bought at the same occasion. The prediction of more aggressive pricing of the pure internet retailers is conrmed (with the exception of prices for U2 where the very low price of one of the mixed retailers lowers the average so much that it is lower than for the pure internet retailers). For the basket the maximum price among pure internet retailers is lower than the minimum price among the mixed retailers when transport costs are excluded and they are very close also when transport costs are included. Since these are the full population of Swedish internet retailers the dierence is signicant per se.
9 The prices without [with] transportation costs were 129 [149], 137 [164], 139 [164], 139 [164], 139 [157], 149 [149], 149 [174], 149 [178]).

E-commerce and prices - theory and evidence

19

The price dispersion in both conventional and internet stores is substantial. Among conventional stores the maximum price of the basket is approximately 35 percent above the minimum and the standard deviation is close to 7 percent of the average price. Prices are less dispersed on the internet with a standard deviation of some 4.5 percent of the average price. These results are consistent with our theoretical model to some extent. Given a relatively early stage of e-commerce with limited access to internet for a majority of consumers, the local rms do not respond to internet competition. The effects of internet competition in the long-run might be quite dierent as our model suggests. Table 5 reports the result of the regressions when we control for dierent characteristics of the goods, such as location and services - the explanatory variables are analogous with the variables used for books.

[TABLE 5]

The regressions conrms that it is on average cheaper to buy these baskets of music CDs from e-commerce sites than from conventional retailers. The coecient of INTERN is negative and signicant both when transport costs are included and excluded. For single item purchases, however, the coecient is small and not signicantly dierent from zero. Also, the more aggressive pricing of pure internet retailers is conrmed. The coecients on INTPURE are always more negative than those on INTMIX. Pure internet retailers set prices which are strictly lower than prices set on the internet by mixed rms. Since this is the full sample of pure and mixed internet retailers, the dierence is signicant per se. The coecient of CDSTORE is positive in all the regressions (however only signicantly different from 0 for RHCP), probably reecting that specialized stores provide better service than supermarkets or general department stores. For the large basket the coecient on CENTER is negative (as was the case with books) this is not the case in the other regressions however. In contrast to the case for books larger stores tend to have higher prices. There is little evidence that the price of independent stores diered from the average.

E-commerce and prices - theory and evidence

20

4.3

Grocery

Finally, we analyze the grocery market. The characteristics of a typical grocery product is such that we expect transport costs to be important. One reection of this is that the main picture that emerges from the Internet sites in grocery is of an industry very much in its infancy in comparison with the markets for books and CDs. The sites were not easy to navigate and Java errors were common. Given our limited sample of prices from regular stores we will limit ourselves to a few observations for this market. In July 1999 a total of 696 price observations on 30 narrowly dened goods from 27 retailers were collected. Of these there were 22 mixed retailers that had sales both over Internet and in conventional stores. For 5 of these rms the data set includes prices from both channels: in each of these cases the price on the internet (excluding transport costs) was identical to the price in the traditional store. When transport costs where included it was clearly more expensive to buy over the internet. These rms apparently do not have a separate pricing strategy for the internet. Asplund and Friberg (1999) examine the prices of some 1000 stores across Sweden of a basket of well dened goods. The Asplund and Friberg data-set contain the 1997 prices of 8 of the 22 Swedish mixed retailers. The prices in these 8 stores were on the lower end of the spectrum in 1997 (two of the stores had prices slightly above average the rest had below average prices, with three having prices in the lowest decile). The low prices is likely to be a reection of that most of the mixed stores are large retailers (all else equal larger stores have lower prices in the dataset). Of the 27 rms ve are pure internet retailers - three German and two Swedish. In the two Swedish retailers prices were roughly equal to the average for all goods when transport costs were excluded . Both featured next day delivery with one oering free delivery for sales of more than 500 kronor (in a major metropolitan area of Sweden) and the other requiring a monthly fee of 100 which included free delivery (in the Stockholm area). The 30 goods were well represented on these sites. The three Internet retailers based in Germany had their web sites in Swedish and were in this sense geared to Sweden. However the assortment featured mainly German and generic brands - 15 of the 30 goods were not available in any of the stores. Prices of most of these goods were substantially lower when transport costs were not included. Delivery took place from Germany and took 5-7 days for one, a maximum of 10 days for another and

E-commerce and prices - theory and evidence

21

the third did not clearly specify how long the delivery would take. In conclusion: buying from the Swedish Internet sites does not pay o in lower prices - having the goods delivered home is the main motive to buy over the Internet - the sites are convenient only to the frequent shopper however and not very easy to navigate. Buying from the German sites oers savings if large bundles are bought - these savings come at a considerable cost of delayed delivery and unfamiliar brands however.

Conclusions

The theoretical analysis in this paper shows that the prices in traditional and online stores depend on opportunity costs, transportation costs and access costs. Although it is to early to formally test this result, it nds some support in the data. The empirical analysis of the markets for books and music CDs shows that prices, excluding transportation costs, are lower online than in conventional stores. Prices of a basket of books is close to 15 percent cheaper if bought on the Internet than in conventional stores. For a basket of music CDs the prices is some 14 percent lower on the Internet than in conventional stores. The analysis also shows that transportation costs are important for determining the relationship between prices in conventional and online stores. Our data suggests that when transportation costs are included in the prices for single item purchases the pattern changes; books are on average cheaper, CDs equally expensive and groceries more expensive online than in conventional stores. This supports to some extent our theoretical nding that the price relation between online and conventional stores is determined by the opportunity cost of conventional shopping as well as the costs access and transportation for online shopping. It also highlights that a reduction of transportation costs - e.g. downloadable music and books - can have a considerable impact on consumer prices. A second result from the theoretical analysis is that an integrated rm charges an optimal price online which is strictly higher than the optimal price of an independent (pure) online retailer. The more aggressive pricing policy of the pure internet retailers is conrmed for both books and music CDs. This result suggests that the long run impact of e-commerce is going to be determined by the relative dominance of mixed and pure internet retailers. The pro-competitive eect can be expected to be less substantial when the

E-commerce and prices - theory and evidence

22

market is dominated by mixed retailers. Finally, a word of caution. Our analysis shows that the pro-competitive eect of Internet competition depends on the degree of access to online retailers. This result suggests that equilibrium prices can be radically dierent in a market with general access to internet and acceptance of e-commerce than in an initial situation with relatively limited access to internet retailers. It should, therefore, be stressed that we have to be cautious with the interpretation of the relation between local prices and online prices as long as the market is immature. Future price eects of internet competition should, therefore, not be dismissed based on current evidence.

E-commerce and prices - theory and evidence

23

References
Anderson, Simon P. and Victor A. Ginsburgh, 1999, International pricing with costly consumer arbitrage, Review of International Economics 7, 126-139. Brynjolfsson, Erik and Michael D. Smith, 1999, Frictionless commerce? A comparison of Internet and conventional retailers, manuscript, MIT Sloan School of Management, available at http://ecommerce.mit.edu/papers/frictions. OECD, 1999, The economic and social impact of electronic commerce, Paris: OECD. Fujita, Masahisa, Paul Krugman and Anthony J Venables, 1999, The spatial economy: Cities, regions, and international trade, Cambridge and London: MIT Press. Goolsbee, Austan, 1998, In a world without borders: The impact of sales taxes on internet commerce, NBER Working paper No 6863. Kraemer, Kenneth L., Jason Dedrick and Sandra Yamashiro, 1999 Rening and extending the business model with information technology: Dell computer corporation, Center for Research on Information Technology and Organizations (CRITO), University of California, Irvine. Schwartz. Evan I., 1999, Digital Darwinism: Seven Breakthrough Business Strategies for Surviving in the Cutthroat Web Economy, Broadway Books. Shapiro, Carl and Varian, Hal, 1999, Information Rules: Competitive Strategy for the Information Economy Harvard Business School Press. U.S. Government working group on electronic commerce, 1999, Towards digital equality, Second annual report.

Caram- All bole (SEK) B&M All internet Mixed internet Pure internet Basket All ( Dev. from B&M sample m e a n , All %) internet Mixed internet Pure internet

Mean Min Median M a x Mean Min Median M a x Std.dev 2 5 th 7 5 th S td.dev 2 5 th 7 5 th Exclucing transport Including transport 263.7 119 259 332 267.4 119 259 332 37.91 249 288 34.35 249 288 269.5 119 259 332 269.5 119 259 332 35.1 259 288 35.1 259 288 219.7 183 209 263 251.8 222 248 299 29.94 24.59 261 259 263 281 263 299 2.828 25.46 203.2 183 207 215 240.2 222 238 254 12.21 12.26 0 -30.04 1.47 18.24 0 -23.96 0.77 18.60 10.88 -7.34 6.05 9.97 -7.98 5.31 1.94 -14.11 2.49 18.24 1.23 -14.71 1.78 17.42 8.57 -4.12 8.25 8.51 -4.79 7.50 -12.59 -30.04 -18.18 12.65 -7.98 -23.96 -13.03 18.60 16.22 15.34 7.06 1.47 12.65 9.68 0.77 18.60 7.91 12.61 -22.41 -30.04 -20.80 -18.01 -16.81 -23.96 -15.64 -12.02 5.62 5.23 -

60 53 7 2 5 45 39 6 2 4

Footnote to table1: Ordinary 2-tailed t-tests were used to test if the differences between the means of the different groups of stores were significant. The variances of the prices in the groups was not assumed to be the same. The difference in the mean prices in B&M stores and internet stores are significant at least on a 10 percent level when transport costs are not included in the price. (1 percent for Carambole.) When transport costs are included, the differences are no longer significant at any usual level of significance. However, in tests of the equality of the mean prices for pure internet stores and B&M stores, the null (i.e. that the means are equal) was rejected at a 1 percent level of significance for both Carambole and the basket whether or not transport costs were included.

Alfons N Std.d. R2 Adj. R2 Const. Intern Intpure Intmix Bookst Center Large Noch. 0.108 0.071 0.033 0.073 0.007 0.062 -0.050 0.049 49 0.185 0.107 0.003 -0.102 0.063 -0.117 0.072 -0.232 0.062 0.057 0.063 0.117 0.067 0.038 0.067 0.012 0.062 -0.010 0.053
* ***

Carambole Excluding transport 49 60 0.120 0.357

SAOL 67 0.160 0.041 67 0.153 0.142 0.056 -0.004 0.087

Basket Excluding transport Including transport 45 0.097 0.290 0.199 0.028 0.053 -0.119 * 0.065 45 0.084 0.483 0.402 -0.033 0.046 45 0.095 0.187 0.083 0.015 0.052 -0.068 0.063 -0.243 *** 0.044 0.058 0.043 0.064 ** 0.031 -0.066 * 0.036 -0.004 0.031 -0.044 0.038 0.097 *** 0.066 ** 0.032 0.031 * -0.055 -0.064 * 0.031 0.003 0.030 0.005 0.034 0.035 -0.003 0.030 -0.039 0.037 -0.179 *** 0.043 0.091 0.065 0.097 *** 0.031 -0.054 * 0.031 0.003 0.030 0.005 0.034 45 0.085 0.374 0.275 -0.039 0.046

60 0.119 0.381

0.180 0.165 0.046 -0.130 0.058

0.298 0.227 *** 0.057 -0.137 *** 0.042

0.311 -0.038 0.211 *** 0.031 0.060 0.095 -0.092 0.075 -0.188 *** 0.050 -0.037 * 0.021

-0.247 *** 0.046 0.084 0.088 0.013 0.072 -0.026 0.052 0.022 0.045 0.026 0.040

-0.106 *** -0.097 ** -0.003 0.039 0.041 0.079 *** *** -0.136 -0.132 -0.026 0.048 0.046 0.033 -0.077 * 0.040 0.048 0.045 0.033 -0.059 0.044 0.053 0.013 0.045 -0.013 0.040

Standard errors in italics. One, two and three asterisks denote that the coefficient estimate is significant at, respectively, the 10, 5 and 1 percent level of significance.

Basket N Std.d. R2 Adj. R 2 Const. Bookst Center Large Noch. Rent Conc Univ Inc

Alfons Carambole SAOL 39 43 53 60 0.0773 0.1909 0.1171 0.1555 0.3586 0.1238 0.3454 0.0851 0.1876 -0.0824 0.2264 -0.0584 0.3018 0.4474 -0.3009 -0.3829 0.5533 1.2506 0.6373 0.8280 ** ** * 0.1282 0.1331 -0.0732 0.0118 0.0505 0.0622 0.0401 0.0817 -0.0291 0.0670 -0.1082 ** -0.0177 0.0247 0.0631 0.0450 0.0516 -0.0258 -0.0119 0.0274 0.0138 0.0245 0.0591 0.0279 0.0454 -0.0023 -0.0585 -0.0444 0.0120 0.0363 0.0607 0.0351 0.0436 -1.162E-06 -1.314E-05 1.006E-05 -7.720E-06 6.768E-06 1.674E-05 1.066E-05 1.032E-05 0.0036 0.0073 -0.0018 0.0004 0.0028 0.0070 0.0037 0.0049 -0.0835 *** 0.0038 -0.1171 ** -0.0550 0.0278 0.0763 0.0489 0.0420 -0.0026 -0.0039 0.0030 0.0030 0.0042 0.0087 0.0046 0.0058

Standard errors in italics. One, two and three asterisks denotes that the coefficient estimate is significant at, respectively, the 10, 5 and 1 percent level of significance.

Red Hot All Chilli Peppers B&M (SEK) All internet Mixed internet Pure internet U2 (SEK) All B&M All internet Mixed internet Pure internet Basket All ( Dev. from B&M sample m e a n , All %) internet Mixed internet Pure internet

Mean Min Median M a x Mean Min Median M a x N Std.dev 2 5 th 7 5 th Std.dev 2 5 th 7 5 th 1 Exclucing transport Including transport 158.5 127 159 189 160.4 139 159 189 96 13.41 149 169 11.85 149 169 160.5 139 159 189 160.4 139 159 189 88 12.05 149 169 12.05 149 169 137.4 127 139 149 159.9 149 157.4 178 8 8.815 9.736 145.7 139 149 161.5 149 178 3 5.774 14.92 132 127 129 139 158.8 153 164 5 6.066 6.076 166.2 109 169 189 168.2 126 196 189 79 18.3 149 179 16.42 159 179 169.0 126 169 189 169.0 126 169 189 72 16.1 159 179 16.15 159 179 137.3 109 139 149 159.4 127.4 164 178 7 14.4 18.62 135.7 109 149 151.5 127.4 178 3 23.09 25.4 138.5 129 139 147 167.3 164 174 4 7.371 5.774 0 -18.85 0.02 22.04 0 -13.06 -0.55 21.34 74 7.84 -6.27 4.74 7.046 -6.81 4.14 1.52 -12.56 0.02 22.04 0.94 -13.06 -0.55 21.34 66 6.71 -3.12 6.31 6.67 -3.68 5.70 -12.58 -18.85 -12.56 -6.27 -8.85 -12.75 -2.27 8 4.47 3.48 -11.66 -6.81 -8.36 -12.56 -6.27 -6.42 -10.18 -2.27 3 3.63 3.97 -15.1 -18.85 -15.39 -12.56 -10.68 -12.75 -9.15 5 2.66 1.82 -

Footnote to table 4: Ordinary 2-tailed t-tests were used to test if the differences between the means of the different groups of stores were significant. The variances of the prices in the groups was not assumed to be the same. The difference in the mean prices in B&M stores and internet stores are significant at least on a 1 percent level when transport costs are not included in the price, both for the two records and for the basket.

For one pure on-line store, the transport cost is missing, thus, the number of observations

for the statistics including transport are one less than the number presented in the rightmost column of the table for the all rows except B&M and mixed internet.

When transport costs are included, the differences between the prices for the individual CDs are no longer significant at any usual level of significance. However, the difference between the mean prices of the basket is still significant at any usual level of significance.

Californication Excluding transport N 95 95 Std.d. 0.070 0.069 R2 0.351 0.373 Adj. R2 . 0.315 0.331 Const. -0.023 -0.026 0.020 0.020 Intern -0.160 *** 0.024 Intpure -0.195 *** 0.022 Intmix -0.105 *** 0.018 *** CDst 0.051 0.053 *** 0.017 0.017 Center 0.000 0.000 0.022 0.022 Large 0.036 ** 0.036 ** 0.016 0.016 Noch -0.035 * -0.033 * 0.019 0.019

Basket 1: All 6 CDs Excluding transport Including transport 34 34 33 33 0.060 0.059 0.058 0.059 0.678 0.692 0.599 0.603 0.620 0.623 0.525 0.512 0.083 0.068 0.070 0.063 0.055 0.056 0.055 0.055 -0.213 *** -0.181 *** 0.019 0.015 *** -0.234 -0.192 *** 0.018 0.016 -0.173 *** -0.163 *** 0.023 0.018 0.011 0.025 0.015 0.021 0.025 0.027 0.024 0.026 -0.105 * -0.105 * -0.104 * -0.104 * 0.055 0.055 0.055 0.054 ** ** ** 0.046 0.049 0.047 0.048 ** 0.023 0.023 0.022 0.023 0.027 0.029 0.027 0.028 0.023 0.023 0.023 0.023

Basket 2: Top 4 Excluding transport Including transport 73 73 72 72 0.065 0.064 0.064 0.064 0.369 0.394 0.238 0.251 0.322 0.339 0.180 0.182 -0.026 -0.029 -0.033 -0.034 0.021 0.021 0.021 0.021 -0.161 *** -0.118 *** 0.022 0.019 *** -0.192 -0.142 *** 0.020 0.018 -0.113 *** -0.088 *** 0.016 0.019 0.025 0.028 0.026 0.027 0.017 0.017 0.017 0.017 0.020 0.020 0.020 0.020 0.023 0.023 0.023 0.023 0.017 0.016 0.017 0.016 0.017 0.017 0.017 0.017 0.004 0.006 0.005 0.006 0.020 0.020 0.020 0.020

Standard errors in italics. One, two and three asterisks denote that the coefficient estimate is significant at, respectively, the 10, 5 and 1 percent level of significance.

N Std.d. R2 Adj. R 2 Const. Bookst Center Large Noch. Rent Conc Univ Inc

Basket 65 0.061 0.280 0.177 -0.862 0.348 0.025 0.016 0.014 0.018 0.018 0.015 0.002 0.018 -6.57E-06 3.72E-06 -0.002 0.001 0.025 0.018 0.006 0.002

**

**

Californication U2 87 71 0.067 0.094 0.292 0.169 0.219 0.062 -0.624 ** -0.147 0.269 0.415 *** 0.051 -0.016 0.016 0.032 0.009 0.038 0.019 0.029 0.039 *** 0.031 0.015 0.019 -0.031 * -0.033 0.017 0.025 -9.15E-06 ** -1.04E-05 * 3.79E-06 5.77E-06 -0.002 0.000 0.001 0.002 0.000 -0.016 0.021 0.038 ** 0.005 0.002 0.002 0.003

Standard errors in italics. One, two and three asterisks denote that the coefficient estimate is significant at, respectively, the 10, 5 and 1 percent level of significance.