An agent-based model of price flexing by chain-store retailers
Ondˇrej Krˇc´al 1
Abstract. This paper investigates the effects of local price flexing on market structure
and welfare in the supermarket sector. It presents an agent-based model of a sector
with two chain-store retailers in which the number and location of stores and prices
charged by the stores are determined endogenously. The outcome in which all the
stores within each chain charge the same price is compared to the situation in which
each store sets prices according to local market conditions. The paper finds that local
pricing reduces the number of stores and total welfare in the market. Furthermore,
local pricing is more likely to increase the average prices and total revenue in the
market, if the reservation price of consumers is high relative to the equilibrium price
under uniform pricing. In this situation, local pricing is likely to reduce not only total
welfare but also aggregate consumer surplus in the market.
Keywords: retail, supermarket, price flexing, agent-based, local, uniform.
JEL classification: L10, L40, L81, R30
AMS classification: 37M05, 91B26
1 Introduction
Local price flexing (or local pricing) by chain-store retailers is a form of third-degree price discrimination in which
individual stores set their prices according to their local market power. Prices are typically higher in areas with
higher geographical distance between competing stores. In its investigation of the UK supermarket sector in 2000,
Competition Commission found evidence of local price flexing. Moreover, it concluded that this practice distorted
competition and adversely affected public interest. One of the remedies considered by Competition Commission
was the imposition of uniform pricing, so that all stores within one chain would have to charge the same prices for
the same products (for a detailed account of the use of local price flexing in retail markets, see [2], [3], and [4]).
This paper investigates the effects of local pricing on market outcome in the retail sector. This problem is
closely related to the literature on third-degree price discrimination in oligopolistic markets ([1] and [5]). A more
specific theoretical approach to the problem of local price flexing was proposed by Dobson & Waterson [2] and [3].
In [3], they present a stylized model of a supermarket sector with two retailers. Each of the retailers operates in two
separate local markets: they compete against each other in one of the markets (there is a differentiated Bertrand
competition) and have local monopoly in the other. Using a linear demand specification, they find that the results
depend on two parameters: a measure of substitutability between the products sold in the common market and the
relative size of the demand in the local and common market. They show that if the demand functions are similar
in both markets, local pricing increases total industry profit for highly substitutable products and reduces it if the
substitutability is low. They also show that for similar demand functions and high and intermediate subsitutability,
local pricing reduces aggregate consumer surplus in the market.
One of the problems of the approach by Dobson & Waterson [2] and [3] is that pricing strategy affects only
prices while leaving the market structure unchanged. This paper proposes a solution to this problem. It presents
a version of the agent-based model of a monopolisticaly competitive market with endogenous number and location
of firms introduced by Krˇc´al [6]. In the model presented here, local pricing affects not only prices charged by the
stores but also the number and location of stores in the market. Therefore, the model provides a more realistic
framework for analyzing the effects of local pricing in retail sector. The structure of the paper is straightforward.
Section 2 introduces the model, Section 3 describes the data and discusses the results of the model, and Section 4
concludes.
1Masaryk University/FEA, Economics department, Lipov´a 41a, 602 00 Brno, krcalo@mail.muni.cz
2 Model
In agent-based models, agents act or interact according to specific rules that are followed in a given order. In this
section, I introduce the agents and explain the rules of the model in a step-by-step way similar to the order in
which the simulation of the model proceeds (the model was implemented in the multiagent modeling environment
Netlogo 4.1.3). In each run, the model is first initialized and then it runs for a certain number of periods.
In the initialization phase, the model creates a landscape, a population of consumers and an initial population
of stores. Landscape: The retail market is set in a square landscape of 40 × 40 patches (a patch is a square field
in Netlogo). In this model, a patch can be interpreted as any unit of distance. So, for instance, the landscape can
be seen as a square island with a side of 40 kilometers. Population of consumers: The landscape is populated by
1,000 identical consumers who differ only in their locations. Consumers live in settlements. Each inhabitant of a
settlement gets a location with a random direction from the center of the settlement and with a random distance
from the center of between 0 and
√
h/(πu), where h is the number of inhabitants of the settlement, u > 0 is the
population-density parameter and π 3.14. If u = 1, the average population density of a round settlement with a
perimeter of
√
h/(πu) is 1 consumer per square patch. Only if a settlement is too close to the edge of the landscape,
consumers who would be located outside of the landscape live on the edge of the square (or on the coastline of
the square island), keeping their original direction from the center of the settlement. Then, the average population
density in such a settlement is higher. Initial population of stores: The market is served by two chain-stores (chain
1 and chain 2) who sell identical product. Before the first period, ten stores of each chain locate randomly in the
market. Each store sets its price equal to pR/2, where pR > 0 is the reservation price of consumers.
Once the model is initialized, the simulation proceeds in periods. In each period, the agents do their actions
in four sequential steps: 1) Both chains consider opening new stores. 2) The existing stores adjust their prices.
3) Consumers shop in the stores. And 4) both chains consider closing some of their stores.
1) Opening stores At the beginning of period t, each chain considers building v new stores. They do it in
turns. First, chain 1 and then chain 2 considers opening a store. This process is repeated v times. For each new
store, each chain is assigned a random location. It would open the store in this location only if for given prices the
new store increased the profit of the chain [the profit is determined in the same way as described in points 3) and
4)]. Whether a new store will enter the market therefore depends not only on its assigned location but also on the
initial price it charges. The price depends on the pricing strategy of the chain. Under uniform pricing (strategy U),
the new store charges the same price as any store in its chain. Under local pricing, the chain can assign any price to
the new store. In order to test the sensitivity of the result to different pricing strategies of the new stores, I introduce
three different local-pricing strategies: 1) Under local pricing with minimal entry price (strategy L), the new store
sets its price equal to the lowest price charged by an incumbent store of its chain. 2) Under local pricing with
average entry price ( ˆL), the new store sets its price equal to the average price charged by the stores of its chain.
And 3) under local pricing with local entry price (LL), the new store sets its price equal to the price of the store (of
any chain) with the lowest distance to the new store.
2) Adjusting prices After the new stores have been opened, each store can increase or decrease its price by a
constant > 0 or keep it at the same level as before. The adjustment process depends on the pricing strategy of
the chain. Under uniform pricing (U), each store in a given chain charges the same price. Therefore, each chain
chooses the price (out of the three options) that maximizes its profit given the price charged by the other chain.
Under local pricing (L, ˆL, or LL), each store may charge a different price. Therefore, each store chooses the price
that maximizes its chain’s profit given the prices charged by all the other stores. Again the profit is determined in
the way described in points 3) and 4).
3) Shopping Each consumer makes a decision whether to shop in store i based on the price of the product pit,
on the per-patch transportation cost c > 0 and on the distance to the store i dit. Each consumer buys one unit of
the product from the store with the lowest pit + cd2
it if pR > p∗
jt + cd∗
jt
2
, and zero units if pR ≤ p∗
jt + cd∗
jt
2
, where
p∗
jt and d∗
jt are the price and the distance of the optimal store for consumer j in period t. The intuition behind the
zero purchase is that each consumer can buy the product in a near-by mom-and-pop store for the price pR. After
the shopping and before the exit, all the variables used for analyzing the situation in the market are calculated and
recorded.
4) Closing stores The closing decision is based on profit of stores. Assuming zero marginal cost, the profit of
store i in period t is πit = qit pit − F, where qit are units of product sold, and F is the quasi-fixed cost that is the
same for all the stores. In period t, the chain closes store i with a probability −πit/F.
3 Results
This section shows how pricing strategy affects the situation in the retail market in the model. Subsection 3.1
provides details about the data and Subsection 3.2 presents the main results of the model. The data presented in
this section was generated using the function Behavior Space in Netlogo and analyzed using the econometric
software Gretl 1.9.5cvs.
3.1 Data
The data analyzed in this section come from the total number of 1,024 runs corresponding to all possible combinations
of the following settings and parameter values: 2 types of landscape: 1) urban landscape containing
1 settlement with the center in the central point of the landscape and the number of inhabitants h = 400, and 20 settlements
with randomly located centers and h = 30, and 2) rural landscape containing 1 settlement with the center
in the central point of the landscape and h = 100, and 30 settlements with randomly located centers and h = 30;
2 population-density parameters u = 0.5 and u = 1; 2 reservation prices pR = 0.5 and pR = 1; 2 numbers of new
stores per chain and period v = 2 and v = 4; 4 strategy profiles (U, U), (L, L), ( ˆL, ˆL) and (LL, LL); 2 transportationcost
parameters c = 0.01 and c = 0.02; 2 price-change parameters = 0.02 and = 0.03; 1 quasi-fixed cost F = 5;
and 4 random initializations with random seed of 1, 2, 3, and 4 (using the random-seed function in Netlogo).
Each run of the simulation generates the following variables:
• Quantity Q = 1
100
200
t=101 ¯nt, where ¯nt is the number of consumers who bought the product in one of the
chains in period t (this consumers are called customers).
• Price P = 1
100
200
t=101( 1
¯nt
¯nt
j=1 pjt), where pjt is the price paid by customer j in period t.
• Number of stores of chain k Mk = 1
100
200
t=101 mkt, where mkt is the number of stores in chain k in period t.
• Revenue of chain k Rk = 1
100
200
t=101
mkt
l=1 qlkt plkt, where qlkt and plkt are the quantity and price of store l of
chain k in period t.
• Distance D = 1
100
200
t=101
¯nt
j=1 d∗
jt, where d∗
jt is the distance to the store of customer j in period t.
• Consumers’ surplus CS = 1
100
200
t=101
¯n
j=1(pR − pjt − cd∗
jt
2
) = QpR − R − cD2
where R = R1 + R2.
• Profit of chain k Πk = 1
100
200
t=101
mkt
l=1(plktqlkt − F) = Rk − MkF
• Total profit Π = Π1 + Π2 = R − MF, where M = M1 + M2.
• Welfare W = CS + Π = QpR − cD2
− MF.
3.2 Comparing the market outcome for uniform and local pricing
This subsection compares the outcomes of the model with both chains pricing uniformly to the outcome of the
model with both chains using local pricing. First, it explains how the outcomes will be compared. Then, it shows
how local pricing affects pattern of prices and what are the effects of this change on quantity Q, number of stores
M, and distance D. Finally, it investigates the effect of local pricing on welfare W, consumers’ surplus CS and
profit Π.
The market outcomes are compared as follows: There are three pairs of strategy profiles to be compared [(U, U)
to (L, L), (U, U) to ( ˆL, ˆL), and (U, U) to (LL, LL)]. For each of the pairs, I investigate the effect of local pricing on
a given variables using linear OLS regressions. The dependent variables are regressed against all the exogenous
variables of the model changed in the simulation (except for the random-seed variable). The independent variables
are denoted as follows: transportation-cost parameter c is TRANSP COST, population-density parameter
u is POP DENSITY, reservation price pR is RES PRICE, price-change parameter is EPSILON, the number of
entrants per chain and period v is ENTRANTS, the variable that indicates the type of landscape is URBAN – it
takes the value of 1 for urban landscape and 0 for rural landscape, and the variable that indicates pricing strategy is
LOCAL – it takes the value of 1 for local pricing, and 0 for uniform pricing. For instance, the parameter of LOCAL
in the following regression shows the effect of local pricing [strategy profile (L, L)] on number of stores M:
STORES NO = 19.307
(0.958)
+ 507.176
(23.652)
TRANSP COST − 3.454
(0.473)
POP DENSITY + 5.935
(0.473)
RES PRICE
+19.293
(23.652)
EPSILON + 0.325
(0.118)
ENTRANTS − 1.336
(0.237)
URBAN − 4.523
(0.237)
LOCAL (1)
T = 512 ¯R2
= 0.677 F(7, 504) = 153.64 ˆσ = 2.676
(standard errors in parentheses)
For all other regressions the paper will report only the mean value of the dependent variable under uniform
pricing, and the parameter and standard error of LOCAL. For instance, the paper will report 29.6 and −4.52 (0.24)
for the equation 1 (see Table 1, column M). The meaning of the numbers is as follows: The change from uniform
pricing (U, U) to local pricing with minimal entry price (L, L) reduces the mean value of number of stores M from
29.6 by 4.52 to roughly 25.1. The standard error is 0.24, which indicates a highly significant effect of pricing
strategy on number of stores. Table 1 (section all data) summarizes the effect of local pricing on selected variables
for the entire dataset. Each field in the rows (L, L), ( ˆL, ˆL) and (LL, LL) corresponds to one regression. The total
number of regressions run for the entire dataset is 24. Furthermore in order to test sensitivity of the results, I run
the same 24 regressions for each of the 12 different partitions of the data (with 256 observations each) restricted
to TRANSP COST = 0.01 or 0.02, POP DENSITY = 0.5 or 1, RES PRICE = 0.5 or 1, EPSILON = 0.02 or
0.03, ENTRANTS = 2 or 4, and URBAN = 0 or 1. The parameter that affects most the outcome of the model is
reservation price. For this reason, Table 1 (sections pR = 0.5 and pR = 1) also presents the effects of local pricing
for the partitions restricted to RES PRICE = 0.5 and 1. In most of the regressions, White test or Beusch-Pagan
test indicate heteroskedasticity. All the regressions in this paper therefore report heteroskedasticity-robust standard
errors.
Dataset Pricing Q M cD2
W P R CS Π
all data
(U, U) 965.4 29.6 64.7 519.9 0.285 275.2 392.5 127.4
(L, L)
−27.8 −4.52 35.3 −29.1 −0.004 −10.1 −41.6 12.5
(2.94) (0.24) (1.23) (1.38) (0.003) (3.10) (3.21) (2.47)
(T = 512)
( ˆL, ˆL)
−9.23 −1.59 20.2 −17.6 −0.013 −15.1 −10.6 −7.09
(2.63) (0.24) (0.84) (1.26) (0.003) (2.92) (2.78) (2.11)
(LL, LL)
−10.8 −1.39 14.09 −13.2 0.002 −0.57 −19.6 6.38
(2.69) (0.24) (0.76) (1.21) (0.003) (3.12) (2.89) (2.24)
pR = 0.5
(U, U) 932.0 28.1 56.7 269.0 0.262 243.6 165.8 103.2
(L, L)
−45.6 −4.55 17.8 −17.8 −0.028 −37.0 −3.51 −14.3
(3.27) (0.29) (0.71) (1.64) (0.003) (2.46) (2.32) (1.63)
(T = 256)
( ˆL, ˆL)
−15.5 −2.12 12.2 −9.33 −0.023 −25.6 5.71 −15.1
(2.77) (0.26) (0.57) (1.63) (0.003) (2.36) (2.38) (1.67)
(LL, LL)
−18.6 −2.26 7.63 −5.63 −0.013 −17.3 0.34 −5.98
(2.79) (0.27) (0.57) (1.58) (0.003) (2.54) (2.53) (1.69)
pR = 1
(U, U) 998.8 31.0 72.8 770.8 0.307 306.7 619.3 151.5
(L, L)
−9.99 −4.49 52.9 −40.4 0.02 16.8 −79.7 39.3
(0.71) (0.32) (1.31) (1.99) (0.004) (3.81) (4.05) (2.98)
(T = 256)
( ˆL, ˆL)
−3.01 −1.07 28.3 −26.0 −0.003 −4.47 −26.8 0.86
(0.33) (0.33) (0.88) (1.79) (0.004) (4.06) (4.06) (2.91)
(LL, LL)
−2.88 −0.51 20.5 −20.8 0.017 16.2 −39.6 18.7
(0.31) (0.34) (0.81) (1.72) (0.004) (4.28) (4.19) (3.07)
Table 1 The effect of local pricing different dependent variables
The table shows the means of different dependent variables [see the lines denoted by (U, U)], and the parameters
and standard errors (in parentheses) of the dummy variable LOCAL for three different local pricing strategies [see
the lines denoted by (L, L), ( ˆL, ˆL), and (LL, LL)]. This information is reported for the entire dataset (all data) and
two different partitions of the dataset (pR = 0.5 and pR = 1).
Pricing strategy affects the prices charged by stores within each chain. While under uniform pricing all stores in
a given chain charge the same price, the prices charged by stores under local pricing differ. The average difference
between the highest and lowest price charged within a chain is approximately 0.24 for the reservation price pR =
0.5, and 0.54 for the reservation price pR = 1. Moreover, local pricing affects the geographical pattern of prices.
The local level of prices depends on the density of population and other characteristics of the area. Typically,
stores in larger and more densely populated areas charge lower prices. Figure 1 shows a typical spatial pattern
of prices for pR = 0.5 (left panel) and pR = 1 (right panel) for the pricing strategy (LL, LL). Because of lower
reservation price, lower proportion of customers in the left panel have prices higher than 0.3. And also, many of
consumers living in relatively remote areas in this panel do not shop in the stores at all (see the dots in the left
panel of Figure 1).
Figure 1 Examples of pattern of prices for pR = 0.5 (left panel) and pR = 1 and the strategy (LL, LL)
The panels show the market outcome in a market with the size of 40 × 40 patches in period t = 200. The crosses
and dots are consumers located in the landscape. Dots show location of consumers who do not buy the product
from any of the stores. Black crosses show location of customers with pjt ≤ 0.2, dark gray crosses customers with
0.2 < pjt ≤ 0.3, and light gray crosses customers with pjt > 0.3.
The pattern of prices reduces the total quantity of product bought Q (see Table 1). This effect is due to the
fact that stores in less densely populated areas prefer charging higher prices even at the cost of losing some of the
potential customers (see location of the dotted consumers in Figure 1). The effect is negative and significant for all
partitions of the data. The pattern of prices also affects the number and location of stores in the market. Competition
reduces prices and profit margins of stores in larger and more densely populated settlements. Therefore, each
individual store needs to serve larger market in order to cover its quasi-fixed cost. On the other hand, the number
of stores in less densely populated areas might increase because the prices charged here are higher than under
uniform pricing. However, markets in smaller settlements often cannot support more stores. This seems to be the
reason why local pricing reduces the number of stores M for all the local pricing strategies for the entire dataset and
for pR = 0.5 and pR = 1 (see Table 1). Furthermore, the effect of local pricing on the number of stores is negative
and highly significant for all the remaining partitions of the data. Lower number of stores implies higher distance
D, which increases the total shopping cost cD2
(see Table 1). Again, this effect of local pricing on shopping cost
is positive and highly significant for all the remaining partitions of the data.
The effect of local pricing on welfare W is negative and highly significant for the entire dataset as well as
for all the partitions of the data (see Table 1). The change in welfare is ∆W = pR∆Q − c∆D2
− ∆MF. Welfare
decreases because the negative effect of lower welfare due to lower quantity traded (pR∆Q < 0) and higher distance
(−c∆D2
< 0) outweighs the positive effect of lower number of stores and therefore lower total quasi-fixed cost paid
by all the stores (−∆MF > 0). Table 2 shows the contributions of the individual effects to the change in welfare
for pR = 0.5 and pR = 1. Interestingly, the absolute value of ∆W is lower for low reservation price. It is because
the consumer surplus from buying in supermarkets is low to start with, so the increase of their distance d∗
jt due to
local pricing forces some of the consumers out of the market (see Figure 1). Which means that they do not waste
as much welfare on shopping cost as consumers with high reservation prices.
reservation price pricing strategy ∆W ∆QpR −c∆D2
−∆MF
pR = 0.5
(L, L) −17.8 −22.8 −17.8 22.8
( ˆL, ˆL) −9.3 −7.7 −12.2 10.6
(LL, LL) −5.6 −9.3 −7.6 11.3
pR = 1
(L, L) −40.4 −10 −52.9 22.5
( ˆL, ˆL) −26.0 −3.0 −28.3 5.3
(LL, LL) −20.8 −2.9 −20.5 2.6
Table 2 The welfare effects of local pricing
And finally, Table 1 shows that the effect of local pricing on profits Π and consumers’ surplus CS is ambiguous.
It is due to the fact, that local pricing may lead to lower price P and revenue R, typically if the reservation price
is low, or to higher price P and higher revenue R, if the reservation price is high. The intuition behind this
result is as straightforward. If the reservation price is high relative to the equilibrium price in the market with
uniform prices (e.g. pR = 1 and P = 0.307), stores in areas with less competition are able to increase their prices
substantially, which may increase also the average price P and revenue R. On the other hand, stores with relatively
low reservation prices (pR = 0.5) have less market power, therefore the average price P and the revenue R in the
market are more likely to decline.
4 Conclusion
The aim of this paper was to investigate the effect of local pricing on market outcomes in the supermarket sector.
For this purpose, the paper introduces an agent-based model with endogenous entry and location of stores. It finds
that local pricing reduces the number of stores and the number of products sold in the market and increases the total
shopping cost incurred by consumers. And since the welfare loss due to lower quantity sold and higher shopping
cost outweighs the gain due to lower aggregate quasi-fixed costs, local pricing reduces total welfare. Similarly
to [3], the effect on profit and consumers’ surplus is ambiguous, but it is more likely that local pricing reduces the
consumers’ surplus. But differently to their model, the main factor that influences the direction of this effect in this
model is the size of reservation price relative to the equilibrium price under uniform pricing.
Acknowledgements
This paper was created as a part of the specific research project no. MUNI/A/0796/2011 at Masaryk University.
References
[1] Borenstein, S.: Price discrimination in free-entry markets. RAND Journal of Economics, 16 (1985), 380–397.
[2] Dobson, P. W., and Waterson, M.: Chain-Store Pricing across Local Markets. Journal of Economics & Management
Strategy, 14 (2005), 93–119.
[3] Dobson, P. W., and Waterson, M.: Price Flexing and Chain-Store Competition. Proceedings of the 32nd
EARIE Annual Conference, (2005) available at
http://www.fep.up.pt/conferences/earie2005/cd rom/Session%20VI/VI.J/Dobson.pdf
[4] Dobson, P. W.: Relationship between Buyer and Seller Power in Retailing: UK Supermarkets, In: Cases in
European Competition Policy (B. Lyons, ed.). Cambridge University Press, Cambridge, (2009), 100–128.
[5] Holmes, T. J.: The effects of third degree price discrimination in oligopoly. American Economic Review, 79
(1989), 244–250.
[6] Krˇc´al, O.: Agent-Based Model of an Urban Retail Market. In Proceedings of the 29th International Conference
Mathematical Methods in Economics 2011, (M. Dlouh´y, and V. Skoˇcdopolov´a, eds.), University of
Economics, Faculty of Informatics and Statistics, Prague, 2011, 369–374.