Strategies in the Polish Gasoline Distribution Market

Entry and Pricing Strategies
in the Polish Gasoline
Distribution Market
• Wojtek Dorabialski, WSHiFM, WISER
• [email protected]
• January 2007
Goals
• To study entry and pricing strategies in an
imperfectly competitive markets
• A transition economy offers opportunities for this
type of study
– growing markets with a lot of entry and not much exit
– firms have market power and exercise it
• I selected the local retail gasoline markets
– homogenous good (although some advertising and
quality differences and loyalty programs exist)
– existence of station chains, operators of such chains
make entry decisions repeatedly
– evidence of imperfect competition (tacit collusion) in
the US market
Literature on
pricing-entry games
• Pre-entry limit pricing: difficult to justify (Spence
critique, chain-store paradox )
• Milgrom-Roberts limit pricing model does not fit
the market in question
• Reputation formation (Kreps-Wilson) hypothesis
could be tested
• „Sixth Avenue effect” (Caplin-Leahy) hypothesis
could be tested (whether entry may attract
competitors due to uncertainty about market
conditions)
Empirical studies on entry
• Bresnahan and Reiss (1991)
– effects of entry on competitiveness in concentrated markets
– significant differences between industries
– the entry of a second firm had a big impact on the level of prices,
but new entry did not have much pro-competitive effect in markets
with 3 or more incumbents
• Mazzeo (2002)
– entry decisions and product (quality level) choice
– entrants have a strong incentive to differentiate their product from
the incumbents’
• Toivanen and Waterson (2001)
– entry strategies of fast-food chain operators who set uniform prices
across all markets
– find evidence for learning; the probability of entry is positively
affected by competitor’s presence in a given market
Empirical studies on gasoline
distribution markets
• Borenstein and Shephard (1996)
– find evidence of (tacit) collusive pricing, that despite the fact that
the market concentration is low and the margin levels are far from
monopolistic
– margins behave in a manner consistent with the RottembergSaloner model of collusion
• Karrenbrock (1991), Borenstein, Cameron and Gilbert
(1997)
– asymmetries in downstream price response to changes in upstream
prices
– retail gasoline prices react more quickly to increases than to
decreases in the wholesale price. This asymmetry is evidence of
existence of retailers’ market power
• Slade (1987)
– test for tacit collusion is through direct estimation of demand, cost
and reaction functions, finds some level of collusion
Gasoline market in Poland
• Production: PKN Orlen (70%), Grupa Lotos
(30%)
• Imports  Exports (10% of consumption)
• Some competition in the wholesale, but Orlen and
Lotos are major players
• Retail (filling stations):
–
–
–
–
over 3000 independent stations
Orlen: 1900 stations (former CPN)
Lotos: 360 stations
Major iternationals: BP – 300, Shell – 250, Statoil –
230
– Small internationals and supermarkets
Gasoline market in Poland
• Orlen is reducing the number of stations, as
they (and the independents) have the lowest
average sales volume
• Internationals have the highest volume of
sales (When Lotos bougth Esso stations in
2005, they claimed that the acquired
stations sold on average 40% more fuel than
their old stations)
Liczba stacji
250
200
150
100
50
0
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
stacje Statoil
stacje Shell
rok
Market definition
• We assume that the markets in the geographical
sense are counties (powiaty). This is a result of
compromise:
– we are unable to precisely determine the proper
geographical market for each station
– gmina would be too small
– voivodship is too large
– powiat is probably too large (variation within marketfirm exists but is smaller than between market-firms
std. dev. 0.027 versus 0.078)
– problem with stations located in small local markets but
on major roads. To mitigate the problem (at least with
respect to pricing) we focus on the price of E95
unleaded gasoline
What we need
• We want to find the determinant of entry and
pricing decisions of the international chain
operators. The several things we should know:
– unobservable costs:
• entry cost: we assume that the entry costs are fixed and
uniform (at least within the chain)
• marginal costs: we assume that the entry costs are the uniform
across stations
– demand, demographic variables
• purchasing power
• number of vehicles
• number of inhabitants
– date of entry for each location
– price at every location
What we have
• The data come from 3 sources:
– station location and opening date: from Shell and
Statoil websites (we also have BP locations)
– E95 gasoline prices at Shell and Statoil stations have
been collected on a weekend in February 2005 by
telephone interviews
– demographic and infrastructural data come from GUS
Bank Danych Regionalnych for 2003
Summary stats
Powiaty with Shell and/or Statoil
All powiaty
Variable Obs
Mean
Std. Dev. Obs
Mean
voivod
153 0.104575 0.30701
373 0.042895
city
153 0.385621 0.48834
373 0.174263
cena
137 3.744805 0.079969
373 .
drogi_ulep~m 153 227.4941 164.0218
373 284.7517
wyd_biez
153 1.40E+08 3.24E+08
373 73222966
vehicles
153 57012.33 86669.22
373 41887.06
st_bezrob
153 19.17974 6.503009
373 21.97507
ludn
153
142044 170853.7
373
100748
powierzchnia 153 596.9346 525.3244
373 825.6595
west
153 0.039216 0.194745
373
east
153 0.035948 0.182315
373
overall
153 4.084967 5.999943
373
own
153 1.856873 1.803372
373
agem
153 55.9982 26.94076
373
Std. Dev.
0.202894
0.379845
153.9999
2.15E+08
58110.52
7.477223
117688.9
525.6141
Pricing strategies –
Conjectural variation
• The profit of an oligopolistic firm i in is:
 i  P(Q)qi  Ci (qi )  P(qi  Qi )qi  Ci (qi )
• The first order condition can be transformed into „supply
function” of an oligopolistic firm:
ci
P
1
1  i

• where i  [0, 1] is a parameter describing competitiveness
of the market (1 = monopoly, 0 = p.c.). 1/ i is the
equivalent of the number of firms in a symmetric Cournot
model
• With proper data (quantities, cost shocks, demand shocks)
the above can be estimated jointly with a demand function
as a stuctural model
• Our data only allows us to estimate the above as a reducedform model of the market
CV Regressions
• We will try to determine whether how the
demand conditions (elasticity) and the
competitiveness of the market affect prices
• We will regress the variables correlated with
demand elasticity and with i on price
• We use OLS to estimate a log-linear model
Results
Number of obs =
320
F( 11, 308) = 6.04
R-squared
= 0.1775
Adj R-squared = 0.1482
Root MSE
= .08428
cena
lnpopdens
lnroaddens1
voivod
lnwydatki
lnunemplrate
east
west
lnshare
new1
newcomp1
kodfirmy1
_cons
|
|
|
|
|
|
|
|
|
|
|
|
|
Coef.
-0.00687
-0.04582
-0.03858
0.009087
-0.06255
0.038134
0.098054
0.01117
-0.03223
-0.01345
0.030743
3.756607
Std. Err. t
0.008076
0.015232
0.017574
0.008905
0.015821
0.026561
0.022764
0.012808
0.024593
0.019538
0.009988
0.169459
-0.85
-3.01
-2.2
1.02
-3.95
1.44
4.31
0.87
-1.31
-0.69
3.08
22.17
P>|t|
0.396
0.003
0.029
0.308
0
0.152
0
0.384
0.191
0.492
0.002
0
Results
Number of obs =
170
F( 10, 159) =
3.59
Prob > F
= 0.0003
R-squared
= 0.1842
Adj R-squared = 0.1329
Root MSE
= .09283
cena
|
lnpopdens
|
lnroaddens1 |
voivod
|
lnwydatki
|
lnunemplrate|
east
|
west
|
lnshare
|
new1
|
newcomp1
|
kodfirmy1
|
_cons
|
Coef.
-0.01048
-0.03344
-0.03076
0.005402
-0.066
0.058194
0.101693
0.004172
0.047715
-0.0334
(dropped)
3.858078
Std. Err.
t
P>|t|
0.013984
0.028223
0.027336
0.013412
0.025399
0.048495
0.029426
0.024552
0.054604
0.027473
-0.75
-1.18
-1.13
0.4
-2.6
1.2
3.46
0.17
0.87
-1.22
0.455
0.238
0.262
0.688
0.01
0.232
0.001
0.865
0.384
0.226
0.257527
14.98
0
Results
Number of obs =
150
F( 10, 139) =
3.51
Prob > F
= 0.0004
R-squared
= 0.2016
Adj R-squared = 0.1442
Root MSE
= .07411
cena
|
Coef.
lnpopdens
lnroaddens1
voivod
lnwydatki
lnunemplrate
east
west
lnshare
new1
newcomp1
kodfirmy1
_cons
|
|
|
|
|
|
|
|
|
|
|
|
-0.00643
-0.04981
-0.05155
0.016565
-0.05039
0.023814
0.104974
0.024104
-0.05701
0.032072
(dropped)
3.621155
Std. Err.
t
P>|t|
0.009759
0.017219
0.022913
0.011685
0.020433
0.029477
0.057071
0.015658
0.0253
0.029739
-0.66
-2.89
-2.25
1.42
-2.47
0.81
1.84
1.54
-2.25
1.08
0.511
0.004
0.026
0.159
0.015
0.421
0.068
0.126
0.026
0.283
0.222127
16.3
0
Entry
• The proper way to estimate entry is to estimate
entry probability using a probit model (or an
ordered probit model)
• Entry and pricing strategies could be linked, as
entry is determined by the expected price-cost
margin
• Work in progress, but the results should be similar
to the ones in the pricing regressions, as the entry
variables are highly correlated with market
characteristics, but weakly correlated with price
and „strategy” indicators
OLS Results
Number of obs =
320
F( 10, 309) = 134.44
Prob > F
= 0.0000
R-squared
= 0.8131
Adj R-squared = 0.8071
Root MSE
= 1.7652
own
|
lnpopdens
|
voivod
|
lnroaddens1 |
lnwydatki
|
lnunemplrate |
east
|
west
|
kodfirmy1
|
lnprice
|
leader
|
_cons
|
Coef.
-2.0516
-0.04931
0.808579
3.319809
-1.30784
-0.63643
0.73686
-0.73023
10.35352
0.250719
-56.1456
Std. Err.
0.167491
0.367066
0.322109
0.181755
0.341847
0.555441
0.491568
0.21049
4.368462
0.242581
6.798719
t
P>|t|
-12.25
-0.13
2.51
18.27
-3.83
-1.15
1.5
-3.47
2.37
1.03
-8.26
0
0.893
0.013
0
0
0.253
0.135
0.001
0.018
0.302
0
Conclusions
• Our data does not allow us to uncover the pricing
strategies of the two firms
• Location near German border, the unemployment
rate and firm-specific fixed effect have a much
stronger impact on prices than „strategic”
variables; the level of competitiveness is similar
across local markets
• Firms’ pricing strategies differ: market share and
new entry variables affect Statoils’ prices, but not
Shell prices
• We find no evidence of dynamic entry deterrence
(price decreases in reaction to competitor’s entry)