Retail Gasoline Price Cycles: Evidence from Guelph, Ontario Using Bi-Hourly,... Retail Price Data

advertisement
Retail Gasoline Price Cycles: Evidence from Guelph, Ontario Using Bi-Hourly, Station-Specific
Retail Price Data*
Benjamin Atkinson (bja@ualberta.ca)
Department of Economics, University of Alberta
May 23, 2006
Abstract
In many Canadian cities, retail gasoline prices appear to move in cycles resembling
those predicted in the Edgeworth cycle literature. In the alternating-moves pricesetting oligopoly model developed by Maskin and Tirole (1988) and extended by
others, an equilibrium exists in which each firm slowly undercuts its rival’s price until
prices fall near marginal costs. At this point, one of the firms will raise its price,
initiating a new cycle. Unfortunately, tests of the basic theoretical predictions ideally
require high-frequency price data for every station in a market, and these data are not
publicly available.
To fill this gap in the literature, retail price data for 27 out of 28 stations in Guelph,
Ontario were collected bi-hourly from 8:00AM-10:00PM, every day from August 14 November 24, 2005. It has been found that while price movements in Guelph are
consistent with the alternating-moves theory in many respects, certain patterns in the
data suggest that the theory is incomplete. This paper also describes the retailers’
reactions to Hurricanes Katrina, Rita and Wilma and considers “gouging” accusations.
*
This paper will be included in my PhD dissertation, currently in progress. I wish to thank
Andrew Eckert and Douglas S. West for their guidance and supervision; Richard Gilbert, Stuart
Landon, Robin Lindsey, Henry van Egteren and Moin Yahya for helpful comments and
suggestions; my parents, Raymond and Lila Atkinson for providing food, shelter, emergency loans
for gas, and the exclusive use of a vehicle during the data collection process; and my wife,
Michaelle Tuz-Atkinson for collecting price data from the Internet while I was on the road. I
remain responsible for any errors or omissions.
1. Introduction
The timing and structure of retail gasoline price movements have been a source of controversy, both
in Canada and other countries for several decades. In some cities, prices appear to move in cycles
where they fall in small amounts over several days, until all stations appear to simultaneously raise
their prices by large amounts. These cycles interest economists because they seem to be independent
of cost movements, and also because it appears that maintaining a high price level is not a profitmaximizing strategy for some retailers. This suggests that cycling markets might be more competitive
than markets where prices are relatively stable. Thus, the competition authority should be interested
in understanding these cycles, because such knowledge can be very useful when evaluating mergers,
or when investigating complaints of price fixing and abuse of dominance in the industry.
A full understanding of these cycles requires empirical testing of the theoretical predictions,
which would ideally use high-frequency, station-specific data for an entire market. Since such data
are not publicly available for an unregulated market, station-specific price data were collected every
two hours from 8:00AM-10:00PM for 27 stations in Guelph, Ontario, every day for 103 days in late2005. Prices were also collected from the Internet for stations in the neighbouring cities of Kitchener,
Cambridge and Waterloo. Using data descriptive techniques, the basic predictions of the Edgeworth
cycle theory will be considered, and retailers’ reactions to Hurricanes Katrina, Rita and Wilma, which
caused retail and wholesale prices to rise significantly during the fall of 2005, will be described.
To anticipate results, preliminary findings support the theoretical predictions in many respects.
However, an interesting sub-cycle is observed after each price increase (i.e., “restoration”), where
some increases are completely reversed within 2-4 hours, and remain low until the end of the night
when they rise again. This contradicts the theoretical prediction that prices rise to the top of the
cycle, and then fall gradually to the bottom of the cycle. Second, while prices fall at different rates
across all four cities in the region, restorations appear to be synchronized both in terms of timing and
peaks. This suggests that while local market conditions generally dictate how prices move in each
city, retailers maximize profits by initiating one large regional cycle instead of four local ones. The
timing of cycle restorations also appears to depend on the time of day and day of the week, which is
contrary to the theoretical prediction that they are triggered by proximity to marginal costs. Finally,
the data shows that while margins did rise following the impacts of Hurricanes Katrina and Rita, they
were due to rapidly falling wholesale costs rather than increasing prices, and were short-lived. On
the other hand, refiner margins increased much more significantly and for longer durations, but they
also faced less competition during these weeks as a result of significant temporary reductions in North
American refinery capacity.
This paper is organized as follows. The theoretical Edgeworth cycle literature and its testable
implications are reviewed in section 2. The relevant empirical literature and alternative explanations
for asymmetric price movements are also critiqued. Section 3 summarizes the methodology that will
be used to test these predictions, while the data are reviewed in section 4, and the tests are conducted
in section 5. Section 6 briefly describes the reactions of retailers and refiners to the hurricanes, in
terms of changes in retail-rack and rack-crude margins, and section 7 concludes.
2. Literature Review
2.1. Testable Implications of the Edgeworth Cycle Theory
In the standard Edgeworth cycle model, Maskin and Tirole (1988) assume that two identical firms
sell homogeneous products in an alternating-moves game, so firm reactions are based on short-run
commitment. Marginal costs are constant, there are no fixed costs or capacity constraints, and the
Page 2 of 33
lowest-priced firm serves the entire market. If both charge the same price, then they split market
demand evenly. The authors show that for a sufficiently fine pricing grid and a discount factor near
one, prices will move cyclically as shown in Figure 1.1
There are five structural predictions which describe the shape and size of a cycle.2 Each cycle
begins with a large price increase in a single period, followed by several small price decreases over
consecutive periods (Structural Prediction #1). Cycles are not caused by cost movements, because
marginal costs do not cycle (Structural Prediction #2).3 Instead, a firm undercuts its rival’s price to
serve the entire market for one period. Since the firm will only lower its price by enough to undercut
its rival, cost movements do not influence the size of these price decreases (Structural Prediction #3).
Prices continue to fall until they approach marginal costs, after which there is a war of attrition
of indeterminate length while each firm waits for its rival to raise its price. When one of them does
“relent” by increasing its price to some function of the monopoly price, the other will undercut this
price by a small amount and the cycle is repeated. Since price increases are triggered by proximity
to marginal costs (Structural Prediction #4), the margins at the bottom of each cycle are relatively
constant. However, cycle peaks also depend on demand and therefore should be less tied to marginal
costs (Structural Prediction #5).
1
Noel (2004a) extends the basic model to allow for randomly fluctuating marginal costs,
and shows computationally that if there is product differentiation (such as differences in station
characteristics), capacity constraints or three firms, then cycles can still exist in equilibrium.
2
Identifying predictions as structural or behavioural is adopted from Noel (2005).
3
Noel (2004a) demonstrates that if the model is extended to include three firms, then the
success of a restoration attempt can depend on marginal cost movements: if marginal costs fall
after the first firm raises its price, and its rivals therefore do not follow the lead then the leader
might abandon the restoration attempt. Noel (2004a) calls this a “false start”.
Page 3 of 33
There are also three behavioural predictions which describe how firms interact with one
another along the cycle. Eckert (2003) extends the basic model to allow firms to differ in size, which
is measured by the number of stations each one operates in the market; if both charge the same price,
their shares of market demand are proportional to their relative sizes. He finds that large firms tend
to lead price increases (Behavioural Prediction #1) and small firms tend to lead price decreases
(Behavioural Prediction #2). Intuitively, large firms can initiate market-wide restorations more easily
than small ones since they control prices at more stations across the market. On the other hand, small
firms undercut rival prices since the gain from matching is relatively small (due to the market sharing
rule).4 Finally, Maskin and Tirole (1988) demonstrate that if the timing of price movements is
determined endogenously, then firms still move sequentially (Behavioural Prediction #3).
2.2. Empirical Studies
The data used in most studies of retail gasoline price cycles, such as Eckert (2002) and Noel (2004b)
are insufficient to test some of the basic theoretical predictions, because they are aggregated across
stations and/or sampled once every one or two weeks. Noel (2005), however collected stationspecific prices for 22 stations in Toronto, every 12 hours for 131 days in 2001. He tests the above
three behavioural predictions as well as the first structural prediction, and finds that all are supported
by his data. Unfortunately, all of these stations are distributed along major city routes in a relatively
small section of Toronto, so it is reasonable to expect that these retailers react to one another’s price
4
Also, Noel (2004a) shows that if one firm is a recognized leader of price increases, then
followers will have a “stepping-up” strategy where they raise their prices slightly at the bottom of
the cycle to remain profitable. Given this strategy, the recognized leader will always lead price
increases because it knows its rival will never relent, so it will not be observed in equilibrium.
Page 4 of 33
changes much more quickly than 12 hours. These stations also appear to have been selected nonrandomly based on the data collector’s commuting patterns, and many of the excluded stations are
located within 1-2 blocks of the sampled stations. Thus, true leaders of price changes might have
been excluded from the sample. Finally, all observations are for self-serve gas and each major-brand
station is company-operated, suggesting that other than the major-independent delineation, tests
based on differences in station characteristics and source of price-control could not be conducted.
Eckert and West (2004a) collected station-specific price data from Internet gas price sites for
Ottawa and Vancouver, and find that the existence of cycles appears to be affected by the presence
of aggressive “maverick” retailers that prevent tacit collusion, such as Sunoco and Pioneer in Ottawa,
and ARCO and Tempo in Vancouver. Eckert and West (2004b) also find that price decreases in
Vancouver appear to originate in regions where ARCO and Tempo are most concentrated, and that
cycle restorations are usually initiated on Tuesdays and Wednesdays. This is consistent with an
observation made by the Conference Board of Canada (2001) that prices tend to rise early in the
week (after the morning rush) when demand is relatively low.5 Unfortunately, while Eckert and West
(2004b) are able to describe the structural characteristics of these cycles, the behavioural predictions
cannot be tested because prices were voluntarily (and non-randomly) reported by consumers. Since
consumers report several prices together, the timing and order of price changes are unobserved and
true price leaders might have been omitted from the sample. Thus, Internet data cannot be used to
reliably test the behavioural predictions of the Edgeworth cycle theory.
Since price movements in Guelph are unregulated, the data reveal some interesting cyclical
5
The Conference Board of Canada (2001) also observes that the restoration will be
abandoned or modified before the evening rush, if it is not followed either quickly or completely.
Page 5 of 33
patterns that likely would not be observed in a price-regulated market. However, it should be noted
that due to the highly regulated nature of the Western Australian retail gas market, Wang (2005a) was
able to compile an excellent data set that includes every price change by every station in Perth from
January 3, 2001 to October 31, 2003. 6 He finds that even though prices change simultaneously and
once per day, they still cycle and firms still raise them sequentially.7
Finally, LECG Canada (2006) conducted a study for the Canadian Competition Bureau on
retail gasoline profitability in Canada, initiated in response to separate complaints that retailers were
both predating and gouging. Although a stated objective is to empirically evaluate the impact of local
competition on retail prices, only five stations located between Oshawa and Kitchener-Waterloo were
sampled, covering a road distance of 150 km. Also, the firms chose the stations as “representative
of their retail operations in the Greater Toronto Area and surrounding areas” (page 34), but no
evidence is given that the sample is representative, so the results cannot be extended to retailers in
general. The timing of price movements is unobserved since the data range from average daily to
monthly revenues per litre sold, and the impact of independent-brand competition is also unclear
because the operations of the two independent retailers are not primarily focussed on gasoline sales.
Finally, this study provides no insight into the validity of the theoretical structural predictions since
it is unknown if the prices of any of the five stations cycled.
6
By 2:00PM each day, retailers are required to report their prices for the following day to
the Western Australian government, which by 2:30PM are all published simultaneously on the
Internet (http://www.fuelwatch.wa.gov.au). Every retailer is then required to change its price at
exactly 6:00AM and keep it unchanged for 24 hours. This is known as the “24 Hour Rule”. Oil
companies are also legally prohibited from operating more than 5% of stations in Australia.
7
Adding quantities to his data set, Wang (2005c) estimates station-level price elasticities
of demand and finds that gasoline is likely used as a loss-leader.
Page 6 of 33
2.3. Competing Explanations of Asymmetric Price Movements
If the predictions of the Edgeworth cycle theory are confirmed in this paper, then other explanations
of asymmetric price movements will be indirectly disputed since they make contrary predictions. For
example, the oligopolistic “sticky” pricing theories cited by Borenstein, Cameron, and Gilbert (1997)
predict that prices move in response to cost shocks, contrary to Structural Prediction #2. Another
potential explanation for retail price cycles is that they are caused by demand cycles: prices fall with
demand until a positive demand shock forces them to rise again, such as before a long weekend.
However, this contradicts the prediction made by the Edgeworth cycle theory that prices cycle even
when demand is constant. It also contradicts the empirical finding that price restorations tend to
occur early in the week when demand is relatively low, as referenced above.
It has also been suggested that inventory fluctuations might cause prices to cycle. Using a
monopolistically competitive, (s, S) threshold model where firms choose both price and inventory
levels, Aguirregabiria (1999) does predict price cycles. However, prices rise slowly until inventories
are replenished and then fall rapidly. Empirically, inventory deliveries occur much more frequently
than cycle restorations and on no particular days. According to brand representatives, inventory
deliveries can occur 3-4 times per week, and deliveries have been observed on Saturdays.
Other explanations are based on different assumptions. First, Conlisk, Gerstner, and Sobel
(1984) predict cycles in equilibrium due to intertemporal price discrimination. However, they assume
that the firm is a durable-goods monopolist, and Sobel (1984) shows that the cycles disappear when
the model is extended to an oligopoly setting. Second, the Conference Board of Canada (2001, page
50) observes that “most consumers view the volatility in gasoline prices and the uniform lowering or
raising of prices as a sign of collusion.” This not only contradicts the alternating-moves assumption,
Page 7 of 33
but also intuition. Noel (2005) argues that although such cycles can occur in equilibrium by the Folk
Theorem, coordination and enforcement costs would likely be prohibitive; strategies of more subtle
price movements that would attract less public attention and scrutiny seem more reasonable.8
3. Methodology
3.1. Definitions
Tests of the above structural and behavioural predictions rely on accurately identifying the beginning
of a cycle, as well as when a restoration has been completed. Therefore, modifying a definition used
by Eckert and West (2004b) to include unsuccessful restoration attempts, an attempted restoration
is defined as an increase in the statistical mode price that results from a retail price increase.9 Also,
the first day of an attempted initiation is labelled Day 0, the next Day 1, and so on, which is similarly
modified from Eckert and West (2004b). Finally, a relenting phase is defined to begin with the first
price increase on Day 0, and end after all stations have raised their prices. The remaining days
comprise the undercutting phase. If a retailer does not raise its price but instead continues to lower
it over the course of the new cycle, then it is considered to have not participated in the restoration.
3.2. Structural Predictions
The first structural prediction is that prices rise by large amounts in one period and fall by small
8
Gas retailers in Ballarat, Australia were convicted of trying to coordinate price increases
(but not decreases) by phone. Even so, Wang (2005b) finds that these attempts sometimes failed.
9
Mode price increases have occasionally been observed after a retailer lowered its price
from the previous mode price, thus causing a higher price to be more frequently observed. These
mode price increases are not considered restoration attempts.
Page 8 of 33
amounts over several consecutive periods, and will be tested in four parts. First, the magnitude of
price increases should be much larger than the magnitude of price decreases, on average. Second,
the signs of two consecutive price changes should rarely, if ever be positive; price decreases should
almost always follow price changes in either direction. Third, a very small proportion of price
changes should be positive during each cycle, with the exception of the first price increase on Day
0 and the first increase on the next Day 0. Finally, when a retailer raises its price to the top of the
cycle, the top should be reached in a single period.
The second structural prediction is that wholesale prices do not cycle. This prediction will
be tested by creating a transition matrix which includes the proportions of price increases and
decreases that are followed by other price increases or decreases. While the first structural prediction
is that price changes are usually followed by decreases, this prediction is that approximately 50% of
price changes should be followed by price changes in either direction. This prediction will also be
considered visually by plotting the time series for each price.
The third structural prediction is that the size of rack price changes should have little effect
on the magnitude of retail price decreases. To test this prediction, the daily decrease in the mode
retail price will be calculated each day, as will the daily change in the wholesale price (except for Days
0). The correlation between these two changes will then be calculated. A second test will calculate
all price decreases over the same sub-sample, and the distribution of price changes will be described.
Price changes should tend to be distributed within a small band that is near zero.
Fourth, price increases are triggered by their proximity to wholesale prices, so the mode retailrack margins immediately preceding each restoration should be relatively constant, as should the
Page 9 of 33
margins for each individual station. 10 Finally, the fifth prediction is that cycle peaks will vary over the
sample based on demand conditions. Therefore, immediately following the initiation of each cycle
on a Day 0, the mode margins should not be tightly distributed around any particular mass point.
3.3. Behavioural Predictions
The first prediction is that vertically-integrated (“major brand”) firms tend to lead price increases,
which can be defined as the first station to raise its price on Day 0. However, a problem with this
definition is that a leader might have actually been following a stepping-up strategy. Thus, in order
to test the robustness of the results, leaders will also be identified based on the definition that a leader
is the first station to raise its price to the top of its cycle.11
The second behavioural prediction is that relatively small firms typically lead price decreases.
However, since prices tend to fall very frequently along a cycle, this prediction should be refined as
follows: independents tend to undercut the mode major brand price, while major brand stations tend
to match independent station prices. This prediction will be tested by dividing stations into four
groups: vertically-integrated stations that are known or reasonably believed to be company-operated
(Group 1); major brand stations that are known or reasonably believed to be independently-operated
(Group 2); “large” independent stations which resemble major brand stations in terms of capacities
and ancillary services offered (Group 3); and “small” independents with operations that appear to be
less focussed on gasoline sales, and more on automotive repair services (Group 4) . It is predicted
10
All margins will exclude the federal gas tax (10 cpl), the Ontario provincial gas tax (14.7
cpl), and the federal Goods and Services Tax (7%) which is also levied on the two excise taxes.
11
These tests will also control for the possibility that a follower’s price increase was
observed before the leader’s price increase, due to the order in which prices were collected.
Page 10 of 33
that Group 1 stations usually set the same price during the relenting phase and match independent
station prices during the undercutting phase; Group 2 stations might either match or undercut prices
in both phases, since they are expected to be privy to head-office strategies but have some latitude
to set their own prices; Group 3 stations likely undercut the mode major brand price, since they have
relatively more to gain than if they simply matched these prices (because major brands are expected
to serve a larger share of the market if they match prices); and Group 4 stations are less likely to
compete aggressively because gasoline sales are relatively less important.
Finally, according to the third behavioural prediction retailers play an alternating-moves game.
Therefore, the average and mode number of stations that change their prices in each period will be
calculated, for both increases and decreases. Since some retailers can react to rival price movements
in minutes, one observed price change per period is unlikely. However, if the theoretical prediction
is correct, then the number of price changes observed in each period should be near one.
4. Data
Regular-grade fuel prices in cents per litre (cpl) were collected every two hours (8:00AM - 10:00PM)
from August 14 - November 24, 2005 for 27 stations in Guelph, a city in southern Ontario with an
approximate population of 106,000. Station characteristics were also collected,12 and daily rack price
data for London, Ontario were obtained from MJ Ervin & Associates. Finally, two dummy variables
are defined to equal one when a queue is observed and when a station is closed, respectively.
12
Characteristics include brand type (vertically-integrated or independent), operating
hours (24 hours or not), price sign (electronic or manually-adjusted), pay-at-the-pump, service
levels (full, self or both), number of regular-grade nozzles, number of grades (up to 4 grades
distinguished by octane level), other fuels (diesel, auto propane), and amenities (repair bay,
variety store, car wash).
Page 11 of 33
Station locations are plotted in Figure 2, and numbered in the order that they were sampled.
Esso (8 locations), Petro Canada (4), Shell/Beaver (3) and Sunoco (3) are the vertically-integrated
brands, and Canadian Tire (2), 7-Eleven (1) and Pioneer (1) are also in the market.13 Each of these
seven brands were contacted by e-mail and asked for the source of price control at each station. Only
Esso refused to cooperate. The rest replied that each station is company-controlled, but the manager
of Petro Canada station #14 is permitted limited price-control based on competitive conditions.
Fortunately, the source of price-control for each Esso station can be reasonably predicted.
Using price, sales and contract data for Vancouver gas retailers, Slade (1998) finds that brands are
more likely to delegate price-setting authority to stations where delegation will potentially increase
profits. Specifically, stations are expected to be company-controlled if they are open longer, sell
higher volumes of gas or have convenience stores. Similarly, firms are more likely to delegate pricesetting authority to full-service stations and stations with repair bays. Thus, it is reasonable to expect
that Esso controls prices at Stations #19, 21 and 26 because they are self-serve, open 24 hours,
relatively less capacity-constrained (8-12 nozzles each), and operate Esso-branded variety stores and
car washes, but no repair bays. Stations #6 and 13, on the other hand seem to be privately operated,
since both close by 9:00PM, are relatively capacity-constrained (4 nozzles each), operate repair bays
and no stores, and Station #6 is a full-serve station.
Finally, there is evidence that Stations #7 and 17 are not operated by Esso. Station #7 has
a 7-Eleven convenience store, and according to MJ Ervin & Associates Inc. (2004) 7-Eleven Canada
13
Station #28 was excluded from the sample because its price was never posted, but
prices collected once per night from the pump from August 14 - September 29, 2005 suggest that
prices at the other 27 stations move independently of this station’s price. Also, Beaver is a
regional brand that is controlled by Shell Canada, and is therefore included as a Shell station.
Page 12 of 33
typically buys gas at the wholesale price from Imperial Oil (Esso’s parent company), along with the
right to use the Esso logo and participate in Esso marketing programs, but retains price control.
Also, the correlation between the retail-rack margins of Stations #7 and 10 (controlled by 7-Eleven)
is 0.9722. As for Station #17, until August 24, 2005 it operated as Rainbow Car Wash & Gas Bar
(the car wash and variety store remained under the Rainbow brand name), and is described on an
Esso-affiliated website as a “dealer” that is family-operated by Sam and Al Destro.14
In an attempt to ensure that no relevant competitors are excluded from the data, prices were
collected once-daily from the 10 closest stations to Guelph (plotted in Figure 3); nine are at least 7.7
km from their nearest Guelph neighbours. Only one station’s price movements resemble those of its
nearest neighbours (Pioneer station #29), perhaps because three of these neighbours are Sunoco and
Pioneer stations (Pioneer is 50% owned by Sunoco’s parent company, Suncor). Prices were also
collected each day at noon and midnight from the Internet15 for stations in Kitchener, Waterloo and
Cambridge, and are observed to fall at different rates than prices in Guelph. 16
An initial overview of the data reveals that out of 22,140 observations where a price change
can be calculated, there are 3,695 (16.7%) changes (up to six per day for each station), of which only
14
Esso’s Rebecca Run for SMA (http://www.rebeccarun.com/esso_dealers/rainbow.html).
15
Consumers voluntarily post the brands, locations and prices of gas retailers on this site
(http://www.ontariogasprices.com); the time of each post and the member’s nickname (“Visitor”
for non-members) are also listed. Membership is free and anonymous, and members earn 150
points for every price posted (up to 750 points per day) which can be used to participate in raffles
for prizes such as U.S.$250 gas cards. Additional points can be earned for participating in other
features of the site, such as voting in opinion polls and posting messages on the message forum.
16
Paradigm Transportation Solutions Limited, Tottem Sims Hubicki Associates, and GSP
Group Incorporated (2005) report that the number of commuters to other regions in 2001 were
low in proportion to Guelph’s population: Peel (2,255), Toronto (1,570), Halton (1,525), and
Hamilton (375). This suggests that Guelph prices are set independently of these outside prices.
Page 13 of 33
575 (15.6%) are price increases.17 Also, 16 attempted cycle restorations are identified. The mode
retail price is plotted in Figure 4, along with the London, Ontario rack price and the par Edmonton
price of crude oil.18 Selected station characteristics are also listed in Table 1, where it can be seen
that Group 1 stations tend to be self-serve with relatively high capacities, are open 24 hours, operate
variety stores and car washes, and do not operate repair bays. The only exceptions are Stations #9,
11 and 20 which are company-operated, but more closely resemble Group 4 stations: low capacities,
no amenities, limited hours and repair bays. Group 3 stations are independents that resemble Group
1 stations (high capacities and more amenities), and thus might price aggressively in order to attract
consumers to both their pumps and stores. Finally, Group 2 stations #7, 14 and 17 seem to differ
from Group 3 stations only in terms of the brand names on their pumps, while Stations #6 and 13
more closely resemble Group 4 stations. In summary, tests of the leadership predictions might reveal
interesting behavioural differences not only between groups, but also within groups.
5. Empirical Tests of the Basic Theoretical Predictions
5.1. Structural Test Results
5.1.1. Magnitudes and Frequencies of Price Changes
As predicted by the theory, prices in Guelph tend to rise by large amounts in single periods, and fall
17
Data for September 22-23, which account for 22 of these increases are excluded from
the tests in section 5, because there is evidence that a demand shock on September 22 affected
price movements on these two days. For example, 109 queues are observed in Guelph on
September 22, while less than 4 queues are observed during each of the other 102 days, on
average. Price movements on these two days will be described in section 6.
18
Daily crude oil price data were downloaded from Natural Resources Canada’s website
(http://www2.nrcan.gc.ca/es/erb/prb/english/View.asp?x=476).
Page 14 of 33
by small amounts over several consecutive periods: the average mode price increase is 8.4 cpl, or
approximately 5 times higher than the average daily mode price decrease of 1.8 cpl. Restoration
phases are also relatively brief, with an average duration of approximately 24 hours, which can be
contrasted with the average duration of an undercutting phase of approximately 5 days. Relenting
phases end by 8:00AM on Day 2, and an average of 96.0% of all retailers participate in each one.
Furthermore, Table 2 shows that 93.5% of all price increases are followed by decreases, and
86.7% of all price decreases are also followed by price decreases;19 only 25 increases are observed
during undercutting phases, which are spread over the entire sample and involve 17 retailers, so prices
rarely increase between Days 0. The data also show that retailers tend to reach the tops of their
cycles in one period, except the six Group 4 stations: 74.1% of the 27 consecutive increases reflected
in Table 2 involve these six stations, 15 of which specifically involve Stations #3 and 4. Looking once
again at all 14 restorations in which they participated, these two stations completed 50-57% of them
after 2-3 consecutive increases. Therefore, it appears that this prediction is less likely to be satisfied
for small independent stations. However, it supports the intuition that small firms tend to undercut
large firms, because they have relatively less to gain by matching prices.
The most interesting result to arise under this prediction is that an unexpected sub-cycle is
consistently observed during all 16 restorations, where some stations lower their prices back to the
bottom of the cycle, and then several hours later raise them again to the top of the cycle. Only then
does the cycle progress into the undercutting phase as predicted by the theory. In total, 67 of these
sub-cycles are observed in the data, and while the identities of the stations are not always the same,
19
Observations for August 30 - September 2 are excluded from Table 1 because three
restorations were identified on these four days, and therefore 51 consecutive price increases that
would skew the results.
Page 15 of 33
every participant initially raises its price by 4:00PM on Day 0. This implies that the prices of the
leaders are more likely to sub-cycle. Furthermore, 91% of the participants reduce their prices back
to the bottom of the cycle within 2-4 hours, and are not observed to rise back to their peaks until after
8:00PM. This suggests that the sub-cycle allows retailers to avoid the war of attrition at the trough
of the cycle, because the leaders can initiate a restoration without sacrificing rush-hour demand. It
thus appears that when the Conference Board of Canada (2001) observed stations abandoning
restorations before the evening rush hour, it might have actually been observing part of this sub-cycle.
5.1.2. Do Wholesale Prices Cycle?
As seen in Table 3, wholesale prices do not appear to move cyclically: over 40% of all price changes
are followed by price increases, despite the direction of the original change. Therefore, it seems clear
that wholesale prices do not cycle, which is confirmed graphically in Figure 4.
5.1.3. Does the Magnitude of Retail Price Changes Vary with Rack Price Movements?
A visual inspection of Figure 4 reveals that prices tend to fall by small amounts despite movements
in the wholesale price. Excluding the sub-cycles, 72.6% of all price decreases are less than or equal
to 1.0 cpl (although the Group 4 stations have much lower proportions), and the correlation between
daily decreases in the mode price and the corresponding changes in the London rack price is 0.08904.
However, there are a few exceptions, the most notable occurring on September 9-10 when the mode
price decreased by 16.8 cpl, from 125.3 to 108.5. This can be reconciled with the theory by noting
that the mode margin fell from 21.2 cpl to 5.4 cpl, which suggests that these price decreases were less
a reaction to wholesale price movements, and more a reaction to the fact that margins were
Page 16 of 33
substantially above the mean peak price of 7.1 cpl (see Table 4). Similarly, the negative mode price
change on September 30 from 115.5 to 109.0 corresponds with a 1.6 cpl decrease in the wholesale
price, and a resulting increase in the mode margin to 8.9 cpl. Finally, the large decreases on October
13 and 19 correspond with false starts.20
5.1.4. Does Proximity to the Wholesale Price Trigger a Restoration?
The fourth structural prediction is that restorations are triggered when retail-rack margins fall within
a certain band, but Table 4 shows the mode pre-restoration margins range from -8.5 to 6.7 cpl. It
instead appears that restorations are tied to the times of day and days of the week when demand is
expected to be lowest: 13 of the 16 restorations are initiated between Monday and Wednesday, which
is consistent with findings in the empirical literature. Furthermore, there are 10 restorations where
the mode trough margin is negative. When the mode margin falls below zero before 10:00AM on
Day 0 (6 cases) then at least one station raises its price to the mode restoration price between noon
and 2:00PM. However, in the four cases where the margin becomes negative after noon or late in
the week, restorations are not initiated for another 1-3 days, depending on which day it becomes
negative. Thus, when margins become negative before a weekend, they sometimes remain negative
until the following Monday or Tuesday.
It is also found that six restorations are attempted when margins were still non-negative.
After further analysis, it seems that the timing of restoration attempts is also influenced by proximity
20
The reasons for the relatively large mode price decreases on September 17 and
November 11-12 are less clear, because margins did not rise above 7.1 cpl. However, the
decreases on November 11-12 correspond with the Remembrance Day weekend, and therefore
might reflect more aggressive competition for drivers participating in weekend services.
Page 17 of 33
to wholesale costs in the region. Internet data show that all 16 restorations are initiated in each of
Kitchener, Waterloo, Cambridge and Guelph on the same afternoons, and that the mode restoration
prices are also consistently identical across cities. Furthermore, mode margins are observed to be less
than 0.1 cpl in at least one of these cities before each restoration. It therefore appears that while price
movements are usually determined by local market conditions, the timing of restorations and peak
prices are determined for the region as a whole. Intuitively, initiating region-wide restorations can
be advantageous, because it might be easier to determine the optimal timing of one large restoration
than four individual ones.
5.1.5. Do Cycle Peaks Vary Over Time?
It is unknown if demand fluctuated because quantity data is not available. However, it has been found
that cycle peaks do not vary much over time. Table 4 shows that the mode margin range from 6.4
to 7.3 cpl for 15 restorations, and the most frequently observed margins are 7.0 and 7.2 cpl. Thus,
some preliminary tests were conducted to determine if the mode peak margins could be predicted,
and the following rule has been remarkably successful: take the current London wholesale price and
add 7.0 cpl; then add all taxes. If the resulting price ends in a “5” or “9”, charge that price; otherwise,
round to the nearest “9”. This rule successfully predicts 12 of the mode peak margins. Three of the
other four days follow large post-hurricane cost shocks, and thus might have differed to reflect these
circumstances. Finally, on November 4 the price is reduced even further to the nearest “5”, possibly
because it is a Friday. The significance of this finding is that retailers might set specific margins at
the beginning of the cycle, with the intent that the next restoration will occur early in the following
week when demand is relatively low.
Page 18 of 33
5.2. Behavioural Test Results
5.2.1. Leaders of Price Increases
Consistent with the first behavioural prediction, it is observed that company-controlled, verticallyintegrated stations tend to lead restorations. If a leader is defined as the first station to raise its price
on Day 0, then Group 1 stations are observed to lead 14 restorations (87.5%), while Group 4 stations
#3, 4 and 16 are observed to lead 6 restorations.21 Stations from the other two groups are never
leaders. However, during four of the restorations where a Group 4 station is observed to be a leader,
its increase is only slightly higher than necessary to raise its margin above zero, suggesting it is a
stepping-up strategy. Therefore, if a leader is defined as the first station to raise its price to its cycle
peak, then these same Group 1 stations are observed to be leaders during all 16 restorations, while
Group 4 stations #3 and 4 are leaders on only 2 dates.
Under both definitions, these Group 1 stations operate under the Petro Canada (#5, 18 and
25), Esso (#19 and 26) and Sunoco (#9, 20 and 22) brands. It is reasonable to expect that these
brands lead restorations, because they are located in all four cities and can thus lead restorations in
each one simultaneously. There is also evidence that the five Petro Canada and Esso stations are the
recognized price leaders in Guelph: out of 80 initial increases for these five stations, 90% of them are
made between noon and 2:00PM on Day 0 and they always charge the mode peak price, implying that
they can be used by other retailers as reliable information sources. On the other hand, no Sunoco
station is observed to increase its price until after 8:00PM during 10 of the relenting phases.
21
Since multiple stations of different types can be observed to be leaders of the same
restoration, these numbers do not add to 16.
Page 19 of 33
5.2.2. Leaders of Price Decreases
The restoration of September 12 will be excluded from these tests, because only 7 (Group 1) stations
raised their prices that day, and most of the remaining stations were pricing higher than the mode
peak price before the relenting phase began. Also, the two full-serve Sunoco stations #9 and 20 are
excluded from this analysis, since their prices appear to be based on the price of Station #22.22
The data suggest that competition in Guelph is centred on the price movements of Pioneer
station #23. With the exceptions of Stations #11 and 27, all Group 1 stations set their prices equal
to the mode peak price, while Pioneer always undercuts that price by at least 0.3 cpl. Table 5 shows
that there is also a tendency for these major brand stations to set their prices up to 0.3 cpl higher than
Pioneer, while Pioneer tends to undercut the mode Group 1 price by either 0.3 or 0.7 cpl.
The only exceptions are Stations #11 and 27, which often undercut the mode peak price, as
seen in Table 6. This might reflect a policy of Shell Canada to set different strategies for each of the
stations that it controls, based on station characteristics and spatial considerations. In particular,
Station #11 is a Beaver station which resembles Group 4 stations, and is also on the same corner as
7-Eleven station #10. Therefore, Beaver might price aggressively to compete with 7-Eleven; in fact,
over 80% of its price decreases result in identical prices with this 7-Eleven station.
Table 6 also lists other stations that appear to be aggressive price-setters. Specifically, it can
be seen that the two relatively small Group 2 Esso stations #6 and 13 are not included in the table,
because they rarely undercut the mode Group 1 price. Canadian Tire station #24 also appears to not
be an aggressive competitor, because it matches 80% of the mode peak prices. Furthermore, while
22
The margin correlations between Stations #9 and 20 and Station #22 are 0.98478 and
0.96547, respectively, and their prices tend to be set 0.4 cpl higher than prices at Station #22.
Page 20 of 33
85.7% of its price decreases undercut the mode Group 1 price, it appears that it is an indirect result
of the fact that 46.2% of all observed price decreases matched Pioneer’s previous price. Therefore,
the identification of a firm as a leader of price cuts depends not only on whether it is a major brand
or an independent, but also on source of price control and station characteristics.
Finally, the only Group 4 stations that appear aggressive are Stations #3 and 4, since less than
22% of price decreases for the other four stations undercut the mode major brand price. It seems that
the only strong difference between these two stations and the other Group 4 stations, is that Stations
#3 and 4 are on Highway 7 while the others are on arterial and collector roads. Therefore, these two
stations might price more aggressively to compete with Petro Canada station #5 for the traffic on that
highway, since Petro Canada has more brand recognition and is therefore expected to serve more
consumers than these independents, even if they all have the same prices.
5.2.3. The Alternating-Moves Assumption
On average, 64.8% of stations raise their prices to the peaks of their individual cycles by the end of
Day 0, while the others wait until Day 1 to raise their prices. Also, there are 3.7 price increases in
each period of a relenting phase, on average (mode = 1, range = 1-13), while there are an average of
4.5 price decreases in each period of the entire sample, excluding Days 0 and September 22-23 (mode
= 2, range = 1-15). Thus, the data support the expectation that firms do not move in unison.
6. Price “Gouging” Complaints
When retail gasoline prices rose significantly in response to Hurricanes Katrina and Rita, there were
angry complaints from both consumers and politicians that they were being “gouged” by the retailers.
Page 21 of 33
However, the actual definition of “gouging” is unclear. The Conference Board of Canada (2001) uses
the term four times but never defines it; the Competition Bureau defines it as “charging high prices
at times of actual or anticipated excess demand”23 and argues that firms can legally charge prices at
levels markets will bear, which is vague. However, it seems that the implied definition is that retailers
are pricing in such a way that their profit margins rise substantially. Profits cannot be calculated
without quantity data, but it is likely that if any excess profits were earned it would be due to an
increase in margins. Therefore, retail-rack margins will be described to see if they rose significantly
in response to the hurricanes. A caveat is that these margins will not provide definitive proof of
whether profits rose, since higher margins might have coincided with reduced demand.
As shown in Table 7, the average retail-rack margin did rise considerably from 4.6 to 7.1 cpl
(54.3%) after the impact of Hurricane Katrina, and again to a lesser extent after Hurricane Rita.
However, these increased margins are a direct result of rapidly falling wholesale costs and not rising
retail prices, such as on September 8-9 when the rack price fell by 14.4 cpl. When costs became
relatively stable at the end of October, margins fell once again to pre-hurricane levels and the mode
restoration margins were quite stable. These higher margins can be expected by the third structural
prediction, i.e., the size of retail price decreases are independent of wholesale price movements.
A second concern raised by the public is that many retailers in southern Ontario were reported
to have raised their prices to $2-3 per litre on September 22, before the impact of Hurricane Rita.
However, the data for Guelph show that while 10 stations did raise their prices as high as 130.9, it
appears that it was in response to a significant demand shock on that day, as referenced above. All
23
“Competition Bureau Concludes Examination into Gasoline Price Spike Following
Hurricane Katrina (http://www.competitionbureau.gc.ca/internet/index.cfm?itemID=2047&lg=e).
Page 22 of 33
but one of these price increases were completely reversed by the following morning, after queues
ceased to accumulate. In fact, most stations did not raise their prices on September 22, and the mode
price fell by 2.0 cpl from the previous evening. Thus, it does not appear that Guelph retailers were
“gouging” consumers on September 22.
In summary, Figure 5 shows that margins tended to fluctuate between 0 and 10 cpl during the
entire sample, with a notable exception of the spike in early September. This can be contrasted with
Figure 4, where it can be seen that rack prices rose significantly despite relatively stable crude prices.
However, a caveat is that there were significant shortages in North American refining capacity during
these weeks, so higher margins could be reasonably expected.
7. Summary Remarks
The purpose of this paper has been to test the basic predictions of the Edgeworth cycle theory, using
high-frequency, station-specific price data collected for 27 retail gasoline stations in Guelph, Ontario
in 2005. While a preliminary analysis shows that some of the basic predictions are confirmed by the
data, there are some important characteristics of these cycles which have not been found in any of the
literature, and likely could not even be found with data collected as frequently as every 12 hours.
First, stations that raise their prices early in the afternoons of restoration days often follow an
interesting sub-cycle, which appears to result in the avoidance of the predicted war of attrition at the
bottom of the cycle. Second, Internet data suggest that while price movements in Guelph are
typically determined by local market conditions, the timing of restorations and the new mode prices
are determined regionally. Intuitively, it is easier for retailers to determine the optimal timing of one
large restoration, than of four smaller ones. Thus, cycle restorations are occasionally initiated in all
Page 23 of 33
four cities even when some of them are still setting positive margins, which have been found to result
in local false starts. Finally, the initiation of price cycles appear to depend not only on proximity to
wholesale costs across the region, but also on the days of the week and times of day when demand
is presumably lowest. The cycle peaks also appear to be determined based a specific markup over
wholesale prices and preferred price endings.
The hypothesis that retailers were “gouging” consumers after the impacts of Hurricanes
Katrina and Rita was also examined, and it was found that while margins did rise significantly
following these two hurricanes, these higher margins resulted from rapidly falling costs, not rising
prices. With respect to upstream prices, margin increases were larger and more prolonged but such
patterns can be expected due to the increased market power experienced by Canadian refiners.
References
1.
Aguirregabiria, Victor (1999), "The Dynamics of Markups and Inventories in Retailing Firms,"
The Review of Economic Studies 66 (2), pp. 275-308.
2. Borenstein, Severin, A. Colin Cameron, and Richard Gilbert (1997), "Do Gasoline Prices
Respond Asymmetrically to Crude Oil Price Changes?," The Quarterly Journal of
Economics 112 (1), pp. 305-39.
3.
Conference Board of Canada (2001), "The Final Fifteen Feet of Hose: The Canadian Gasoline
Industry in the Year 2000," Empirical examination of Canadian petroleum market.
4. Conlisk, John, Eitan Gerstner, and Joel Sobel (1984), "Cyclic Pricing by a Durable Goods
Page 24 of 33
Monopolist," The Quarterly Journal of Economics 99 (3), pp. 489-505.
5.
Eckert, Andrew (2002), "Retail Price Cycles and the Response Asymmetry," Canadian Journal
of Economics 35 (1), pp. 52-77.
6. ----- (2003), "Retail Price Cycles and the Presence of Small Firms," International Journal of
Industrial Organization 21 (2), pp. 151-70.
7. Eckert, Andrew and Douglas S. West (2004a), "A Tale of Two Cities: Price Uniformity and
Price Volatility in Gasoline Retailing," Annals of Regional Science 38 (1), pp. 25-46.
8. ----- (2004b), "Retail Gasoline Prices Cycles Across Spatially Dispersed Gasoline Stations,"
The Journal of Law & Economics 47 (1), pp. 245-73.
9.
LECG Canada (2006), "What Determines the Profitability of a Retail Gasoline Outlet? A Study
for the Competition Bureau of Canada," Independent Expert Report.
10.
Maskin, Eric and Jean Tirole (1988), "A Theory of Dynamic Oligopoly, II: Price Competition,
Kinked Demand Curves, and Edgeworth Cycles," Econometrica 56 (3), pp. 571-99.
11. MJ Ervin & Associates Inc. (2004), "National Retail Gasoline Site Census 2004," Annual
survey of retail gasoline representation in Canada.
12.
Noel, Michael (2004a), "Edgeworth Cycles and Focal Prices: Computational Dynamic Markov
Equilibria," UCSD Working Paper (September 28, 2004).
13. ----- (2004b), "Edgeworth Price Cycles, Cost-based Pricing and Sticky Pricing in Retail
Page 25 of 33
Gasoline Markets," UCSD Working Paper (October 23, 2004).
14. ----- (2005), "Edgeworth Price Cycles: Evidence from the Toronto Retail Gasoline Market,"
UCSD Working Paper (January 27, 2005).
15.
Paradigm Transportation Solutions Limited, Tottem Sims Hubicki Associates, and GSP Group
Incorporated (2005), "Guelph-Wellington Transportation Study: July 2005 Final
Report," Transportation study jointly undertaken by the City of Guelph and the County
of Wellington (http://guelph.ca/living.cfm?itemid=68351&smocID=1725).
16. Slade, Margaret E. (1998), "Strategic Motives for Vertical Separation: Evidence from Retail
Gasoline Markets," The Journal of Law, Economics & Organization 14 (1), pp. 84113.
17.
Sobel, Joel (1984), "The Timing of Sales," The Review of Economic Studies 51 (3), pp. 353-68.
18.
Wang, Zhongmin (2005a), "Strategy, Timing and Oligopoly Pricing: Evidence from a Repeated
Game in a Timing-Controlled Gasoline Market," Northeastern University Working
Paper (September 2005).
19. ----- (2005b), "Edgeworth Price Cycle and Oligopoly Coordination: Trial Evidence from
Australia," Northeastern University Working Paper (October 2005).
20. ----- (2005c), "Station-Level Gasoline Demand in a Market with Price Cycles," Northeastern
University Working Paper (November 5, 2005).
Page 26 of 33
Page 27 of 33
Page 28 of 33
Page 29 of 33
Table 1: Selected Station Characteristics24
Brand
ID
Nozzles
24 Hrs
Self-Serve
Store
Wash
Repair
Group 1: Vertically-Integrated, Company-Operated Stations
Esso
Esso
Esso
Petro Canada
Petro Canada
Petro Canada
Shell
Shell (Mac’s)
Shell (Beaver)
Sunoco
Sunoco
Sunoco
19
21
26
5
18
25
12
27
11
9
20
22
12
10
8
8
8
8
8
8
6
6
4
8
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
Group 2: Vertically-Integrated, Independently-Operated Stations
Esso (Norm’s Garage)
Esso (7-Eleven)
Esso (Gas-Up Carwash)
Esso (Rainbow)
Petro Canada
6
7
13
17
14
4
8
4
6
8
T
T
T
T
T
T
T
T
T
T
T
T
T
Group 3: “Large” Independent, Company-Operated Stations
7-Eleven
Canadian Tire
Canadian Tire
Pioneer
10
15
24
23
8
8
6
8
T
T
T
T
T
T
T
T
T
T
T
T
Group 4: “Small” Independent Stations
Amco
Cango
Hilton Group Gas (“Sam’s”)
Maple Leaf Gas and Fuels
CAN-OP
Quik-N E-Zee Gas and Snacks
1
2
3
4
8
16
4
3
4
4
2
2
24
T
T
T
T
T
T
Petro Canada station #18 is the only station with both full- and self-serve pumps. Since the selfserve price is observed, only those pumps are counted. Also, a “store” means a convenience store, except
for Canadian Tire station #15 where it is a Canadian Tire department store.
Page 30 of 33
Table 2: Transition Matrix of Guelph Retail Price Movements
)pt > 0
)pt < 0
)pt-1 > 0 (N = 418)
6.5%
93.5%
)pt-1 < 0 (N = 2958)
13.3%
86.7%
Table 3: Transition Matrix of London Rack Price Movements
)rt > 0
)rt < 0
)rt-1 > 0 (N = 30)
43.3%
56.7%
)rt-1 < 0 (N = 38)
42.1%
57.9%
Table 4: Comparison of Mode Price Changes and Margins Across Cycles
Period
Day 0 of Attempted
Restoration
Mode Price
Change (cpl)
Pre-Katrina
Monday August 15
Tuesday August 23
15.0
6.9
-7.1
0.1
6.9
6.5
Hurricane Katrina
N/A
N/A
N/A
N/A
Post-Katrina
Tuesday August 30
Wednesday August 31
Friday September 2
Monday September 12
6.9
18.9
10.0
0.7
-0.2
-8.5
-2.8
6.7
7.2
9.2
6.6
7.3
Hurricane Rita
Tuesday September 20
8.7
-1.4
6.7
Post-Rita
Monday September 26
Thursday September 29
Wednesday October 12
6.0
11.6
2.7
1.5
-3.5
4.7
7.1
7.3
7.2
Hurricane Wilma
Tuesday October 18
6.6
1.0
7.2
Post-Wilma
Monday October 31
Friday November 4
Wednesday November 9
Tuesday November 15
Tuesday November 22
8.4
6.8
7.7
9.3
7.6
-1.1
0.1
-0.2
-1.7
-0.5
6.7
6.4
7.0
7.0
7.0
8.4
-0.8
7.1
Average
Page 31 of 33
Mode Trough
Margin (cpl)
Mode Peak
Margin (cpl)
Table 5: Relationship Between Selected Group 1 Stations and Pioneer (#23)
Brand
ID
Esso
Esso
Esso
Petro Canada
Petro Canada
Petro Canada
Shell
Sunoco
Major Mode Price - lagged Pioneer Price
19
21
26
5
18
25
12
22
Pioneer Peak Price - Mode Peak Price
0.3 cpl
0.2 cpl
0.0 cpl
Total
9.7%
13.0%
8.5%
27.5%
16.4%
11.3%
24.1%
24.2%
17.7%
10.7%
10.5%
8.5%
11.1%
7.3%
10.7%
7.9%
25.7%
24.2%
30.4%
18.5%
25.4%
21.3%
16.1%
18.5%
53.1%
47.9%
49.4%
54.5%
52.9%
39.9%
50.9%
50.6%
Pioneer Price - lagged Major Mode Price
-0.3 cpl
-0.7 cpl
-1.0 cpl
Total
-0.3 cpl
-0.7 cpl
-1.0 cpl
Total
60.0%
33.3%
6.7%
100.0%
25.9%
15.7%
7.5%
49.1%
Table 6: Other Potentially Price-Aggressive Stations
Group
Brand
ID
Proportion of Price
Increases Where
Undercut Mode Peak25
Proportion of Price
Decreases Where Undercut
Mode Group 1 Price
1
Shell (Beaver)
Shell (Mac’s)
11
27
93.3%
46.7%
N/A
N/A
2
Esso (7-Eleven)
Esso (Rainbow)
Petro Canada
7
17
14
73.3%
60.0%
86.7%
34.8%
63.3%
45.2%
3
7-Eleven
Canadian Tire
10
15
73.3%
60.0%
56.3%
30.3%
4
Hilton Group Gas
Maple Leaf Gas and Fuels
3
4
100.0%
100.0%
40.4%
34.6%
25
If a station did not participate in a restoration, it was counted as undercutting the mode price.
Page 32 of 33
Table 7: Cross-Period Margin Comparisons
Period
Dates
Observations
Total Rack Change (cpl)
Sub-Period
Mean Retail Margin (cpl)
Overall
Sub-Period
Overall
Pre-Katrina
14-AUG – 15-AUG
16-AUG – 23-AUG
432
1,728
2.7
-1.5
1.2
-1.3
4.3
3.2
Hurricane Katrina
24-AUG – 29-AUG
1,296
-1.7
-1.7
4.6
4.6
Post-Katrina
30-AUG – 02-SEP
03-SEP – 06-SEP
07-SEP
08-SEP – 09-SEP
10-SEP
11-SEP – 12-SEP
13-SEP – 14-SEP
15-SEP – 17-SEP
864
864
216
432
216
432
432
648
30.6
-3.1
-6.0
-14.4
0.0
-2.5
-4.1
0.8
1.3
0.5
5.1
9.3
19.6
10.9
7.1
10.3
6.1
7.1
Hurricane Rita
18-SEP – 24-SEP26
1,080
1.1
1.1
4.2
4.2
Post-Rita
25-SEP – 15-OCT
4,536
-8.9
-8.9
5.0
5.0
Hurricane Wilma
16-OCT – 24-OCT
1,944
-4.4
-4.4
5.3
5.3
Post-Wilma
25-OCT – 30-OCT
31-OCT – 24-NOV
1,296
5,319
-1.6
-1.8
-3.4
2.3
3.9
3.6
21,735
-14.8
-14.8
4.8
4.8
Total
26
Observations for September 22-23 were excluded to avoid skewing the averages.
Page 33 of 33
Download