MEASUREMENT OF CONJECTURAL VARIATIONS IN OLIGOPOLY

ELECTRICITY MARKETS

J.D. Liu and T.T. Lie

Power Market Research Group

School of Electrical and Electronic Engineering
Nanyang Technological University
Singapore 639798

Abstract

The concept of conjectural variation and its application to strategic bidding have been
introduced in oligopoly electricity market. Various values have been assumed for the
value of conjectural variation, but few attempts to measure it empirically have been made.
In this paper, the theory and method for estimation of the conjectural variations of firms
are proposed. This measurement method is applied to the Australia National Electricity
Market (NEM). An empirically investigation about market behavior of firms in the NEM
is presented.

1. INTRODUCTION This proposed method is applied to the Australia

In recent years, the traditional vertically integrated
electric utility structure has been deregulated and
replaced by a competitive market scheme in many
countries worldwide. This activity is called
deregulation, restructuring or unbundling in different
applications. The objective of this activity is to create
a competitive environment for trade in electricity in
order to improve social welfare and market efficiency.
However, due to high barriers for new entrants to
enter electricity market, the new electricity markets
are more akin to oligopoly than perfectly competitive
market. As a result, suppliers can increase profits
through strategic bidding, or in other words, through
exercising market power. The exercise of market
power by market participants may raise price above
marginal cost and lead to productive and allocative
inefficiencies and inefficient signals for new
investment. Therefore, it is very important for system
regulator to analyze the competitiveness of a set of
oligopolists and to recognize market power.
The concept of conjectural variation was brought
forward by Bowley in 1924, but named as ‘conjectural
variation’ by Frisch in 1933 [1]. In this paper, a theory
and method for estimation of the conjectural
variations of firms are developed. Then the conjectural
variation of firms is used to analyze the market
behavior of oligopolists. That is to draw inferences
about whether Bertrand, Cournot, or Monopoly, which
are classical oligopoly models, is supported by the
estimated conjectural variations. Then the market
structure and potential market power can be analyzed.
National Electricity Market (NEM).
This paper is arranged as follows. Section 2 presents
the theoretical framework of measurement method.
The proposed method is used to analyze the NEM.
Section 3 describes the actual historical data of NEM
and estimates corresponding parameters for
calculating the conjectural variation. The detail
method for estimating the marginal cost of firms will
be described in Section 4. Section 5 presents the
estimate of conjectural variations and analyzes the
market behavior of firms in the NEM. Finally,
conclusions are drawn in Section 6.
2. A THEORETICAL FRAMEWORK
OF ANALYSIS
Suppose there is an oligopolistic market of a few
supply firms with one homogeneous product. Let the
number of firms be N and the supply of the ith firm be
qi. Then the total supply Q = q1 + q 2 + K + q N must
be equal to the total demand D:
D = Q = q1 + q 2 + K + q N (1)
Let the price be p, then the inverse market demand
function will be as follows:
p = f ( D) = f (Q) (2)
The derivative of this inverse market demand function
is to be negative for any positive D.
Assume that each firm is rationally aiming at Specifically, consider the N-firm electricity market
maximizing its profit. Define profit as with a linear inverse demand curve and quadratic cost
π i (qi ) = Ri (qi ) − C i (qi ) = pqi − Ci (qi ) , where functions: Ci (qi ) = ai + bi qi + 0.5ci qi2 . Then if the
Ri (qi ) and Ci (qi ) are total revenue and cost firms can be described by Bertrand, Cournot, and
function respectively. The marginal revenue can be Monopoly model, then the conjectural variation of
expressed as follows: N
1
dR
MRi (qi ) = i =
d ( pqi ) d p dD
= qi + p
firms are –1, 0, ci ∑ c
j =1, j ≠i j
, respectively. These
dqi dqi dD dqi
results are independent of the estimation of the
N
dp dq j conjectural variation. Therefore, these conjectural
=
dD
(1 + ∑
j =1, j ≠i
dqi
)qi + p
variations can be considered as indicators for
analyzing the market behavior of firms in actual
dp electricity market. That is to draw inferences about
= (1 + CVi )qi + p (3)
dD whether Bertrand, Cournot, or Monopoly are
where supported by the estimated conjectural variations.
N
dq j
CVi = ∑
j =1, j ≠i
dqi
(4) 3. THE DATA

The proposed method is used to analyze the Australia

It is known from industrial organization theory [2] that National Electricity Market (NEM). The NEM was
CVi is the conjectural variation. The conjectural initiated in December 1998 to represent the wholesale
variation of Firm i is its belief or expectation on its market for the supply and purchase of electricity in
rivals’ reaction to its output changes. several Australian states and territories in addition to
an administration of open access to transmission and
The first order condition for profit maximization is distribution networks in those states and territories.
dπ i dp
= (1 + CVi )q i + p − MC i = 0 (i = 1, L N ) (5)
dqi dD The National Electricity Market Management
Company Limited (NEMMCO) manages and
where the marginal cost MCi = dCi dqi .
facilitates the wholesale electricity market. The main
functions of NEMMCO include managing the power
The price elasticity of electricity demand is defined as system to keep supply and demand in balance based
the percentage change in the quantity of electricity on the generating capacity available to the wholesale
demand in response to a percentage change in the market, keeping security of power system,
price of electricity. If the price elasticity of demand is administering the spot market including calculation of
dD D spot prices and so on.
denoted by α(<0), α = , then eqn. (5) can be
dp p
expressed as follows: In the wholesale electricity market, generators decide
1 p their own unit commitment respectively and provide
(1 + CVi )qi + p − MC i = 0 (i = 1, L N ) (6) dispatch offers (prices for different levels of
α D generation) to NEMMCO. The electricity output from
all generators is centrally pooled and scheduled to
Define the market share of Firm i as follows: meet the electricity demand. A spot price for
q wholesale electricity is calculated for each half-hour
si = i (7)
D period during the day and is the clearing price to
match supply and demand. The price is determined as
Then the conjectural variation CVi is expressed as the highest price of the power producers that satisfies
follows: the load requirement.
MCi − p 1
CVi = α − 1 (i = 1, L N ) (8) The data used in this paper are obtained from [3]. All
p si
these data including bidding data of all generation
companies (Gencos), market clearing price and
Thus the conjectural variation CVi can be calculated if demand are the actual historical market data of NEM.
the values of the price elasticity α, the marginal cost The period of analysis is confined from April 2002 to
MCi, the market clearing price p and the market share December 2002, because the bidding data of Gencos
si are known. These parameters can be estimated for are only available in this period. Normally, every
each firm from the historical data of actual electricity Genco has several power stations, and each of them
market. submits its own bidding curve everyday to the NEM.
Assume the objective of these units’ bidding strategies of demand of NEM is obtained from a report for the
is maximizing profit for the whole Genco owning National Electricity Market Management Company
these units. It means that the market behavior of [4], which is about the price elasticity of demand for
Genco is considered as a whole. In general, there are electricity in NEM regions.
more than 160 units, which are owned by about 58
Gencos, offering their bidding data everyday in the The estimated long run own price elasticities of
NEM. But 11 of all Gencos supply almost more than electricity demand by the states are shown in Table 2.
70 per cent of total demand. The minimal share of Both a range and mean value are provided for each
these 11 firms is 4.13 per cent and the maximal share region.
is 8.40 per cent. For simplicity, only the conjectural
variations of these 11 firms are estimated. Their In this paper, in order to analyze the sensitivity of
market behavior is analyzed while the behavior of price elasticity of demand, -0.20, -0.35 and -0.50 are
fringe Gencos is ignored. used as elasticities of demand to calculate the
conjectural variation. As a result, the extreme bounds
The pool operation of NEM, managed by NEMMCO, of conjectural variation can be obtained.
mainly covers an interconnected power system
including New South Wales, Queensland, South Table 2 Estimated long run price elasticity of
Australia, SNOWY, and Victoria. There is some little electricity demand of NEM
difference in clearing prices among these different
areas for the transmission loss and transmission Range Mean
constraint. Every area has its own demand New South Wales -0.22 to -0.52 -0.37
respectively. For simplicity, the mean of prices and Victoria -0.23 to -0.53 -0.38
the aggregation of demands are used as the final Queensland -0.14 to -0.44 -0.29
clearing price and total demand of every time period. South Australia -0.17 to -0.47 -0.32
NEM -0.20 to -0.50 -0.35
From the data [3], the mean market clearing price of
NEM from April 2002 to December 2002 can be
calculated, which is 28.55 ($/MWh). The bidding 4. ESTIMATION OF MARGINAL
curve of every firm in every half-hour period can also COST
be obtained. Then the market share of firm’s supply
can be calculated. The mean estimated market share of
Let Firm i denotes the Genco whose bidding strategy
11 firms from April 2002 to December 2002 are
is being computed.
shown in Table 1. Assume each of 11 Gencos
considered only takes the behavior of these 11 Gencos
Define:
into account. Then the values of market share used for
D - Total market demand
calculating the conjectural variation are proportionally
SOi(p) - Amount of capacity bid by all other firms
more than the values in Table 1.
except Firm i as a function of the market
price p
Table 1 Estimated market shares of firms
DRi ( p) = D − SOi ( p) - Residual demand faced
No. GencoID Market share by Firm i, specifying the demand faced by
Firm i as a function of the market price p
1 BW 8.40% π i ( p ) - Profit of Firm i at price p
2 ER 6.36%
Ci(p) - Total cost of Firm i
3 GSTONE 5.36% MCi - Marginal cost of Firm i
4 HWPS 6.44% Si(p) - Bid function of Firm i giving the amount it
5 LD 6.81% is willing to supply as a function of the
6 LOYY 4.13% market price p
7 LY 8.15%
Assume that Firm i is able to observe the market
8 MP 7.76%
demand and bids submitted by all other market
9 STAN 4.96% participants from the historical data. It can construct
10 TARONG 5.63% its residual demand function implied by the market
11 YWPS 5.93% demand and these bids. Then Firm i selects the profit
maximizing price associated with this residual demand
Estimation of conjectural variation calls for the via assuming the market clearing price p is determined
elasticity of demand, which requires that some by solving the smallest price such that the equation
elasticity estimation should be obtained. The elasticity S i ( p) = DRi ( p) holds for Firm i.
Then, the profit function of Firm i can be expressed as Table 3 Mean marginal costs of 11 firms
follows: in the NEM
π i ( p) = S i ( p) p − Ci ( S i ( p))
= DRi ( p) p − Ci ( DRi ( p)) (i = 1, L , N ) (9) No. GencoID Marginal cost($/MWh)
1 BW 15.22
To compute the optimal price associated with the 2 ER 16.76
residual demand function, DRi(p), differentiate the 3 GSTONE 17.65
profit function with respect to p and set the result 4 HWPS 17.31
equal to zero:
5 LD 16.99
dπ i ( p) dDRi ( p)
= p + DRi ( p) 6 LOYY 19.75
dp dp
7 LY 15.80
dC ( DRi ( p)) dDRi ( p) 8 MP 15.61
− i = 0 (i = 1, L , N ) (10)
dDRi ( p ) dp 9 STAN 18.45
10 TARONG 18.06
This first-order condition as shown in eqn. (10) can be 11 YWPS 17.74
used to estimate of the marginal cost at the observed
market clearing price, p*, as follows:
dCi ( DRi ( p*)) Table 4 The parameters of the implied cost function
MCi ( DRi ( p*)) =
dDRi ( p) regression of each firm
= p * + DRi ( p*) / DRi' ( p*) (i = 1, L , N ) (11)
No. GencoID b c
DRi(p*) and DRi ' ( p*) can be computed using the 1 BW -2.435453 0.005952
actual market outcome and bid functions submitted by 2 ER 12.42053 0.006533
all other market participants except Firm i. The market 3 GSTONE 4.882643 0.007356
clearing price, p*, can be directly observed. 4 HWPS -48.426416 0.032943
5 LD -10.883333 0.013645
Now the proposed technique mentioned above will be 6 LOYY -69.044431 0.023335
applied to the NEM. This requires collecting data on 7 LY -30.077832 0.019167
Gencos bids and market outcomes for a half-hourly
8 MP -4.64969 0.008039
time period. This information for a market participant
in the NEM from the period April 2002 to December 9 STAN -9.024203 0.019785
2002 can be obtained from [3]. 10 TARONG -12.664782 0.02076
11 YWPS -71.460874 0.033169
Firstly, compute implied marginal cost estimations
using bid data submitted by all Gencos, actual market
prices and total market demand. The mean marginal 5. ESTIMATION RESULTS OF
costs of all firms from April 2002 to December 2002 CONJECTURAL VARIATION
are shown in Table 3.
When all the data needed to calculate the conjectural
From the historical data, one can get a series of the variation, the values of the marginal cost MCi, the
marginal cost and associated output pairs,
price elasticity α, the market clearing price p and the
( Ci ' ( DRi ( p*)) , DRi ( p*) ), for all values of the half- market share si, are estimated, the value of conjectural
hourly market clearing price p*. Assume firms have variation of each firm is calculated using eqn. (8).
quadratic cost functions: Ci (qi ) = ai + bi qi + 0.5ci qi2 .
Using regression analysis method, one can get the Table 5 shows the results of conjectural variations of
linear regression of the implied marginal cost, Ci ' (qi ) , all firms with different price elasticity α = -0.20, -0.35
and -0.50.
on q, the associated implied output of Firm i:
Ci ' (qi ) = bi + ci qi (12) As it was mentioned above, if the firms can be
described by Bertrand, Cournot, and Monopoly model,
The value of ci can be used in calculating the then the conjectural variations of firms are –1, 0,
conjectural variations of firms in Monopoly case. N
1
Table 4 gives the results of the implied marginal cost
function regression of every firm.
ci ∑ c
j =1, j ≠i j
, respectively (ci can be obtained from
Table 4). The CVs of all firms in Bertrand, Cournot, variations are close to Cournot value 0 for all firms.
and Monopoly model are shown in Table 6. Both the monopoly hypothesis (CV>>1) and the
Bertrand hypothesis (CV = -1) will be strongly
Table 5 Conjectural variations of all firms with rejected. However, it can be seen that when the price
different price elasticity elasticity α = -0.35 and -0.50, the conjectural variation
increases observably. This result is implied by eqn. (8)
CV CV CV that the conjectural variation is increasing with the
No. GencoID
(α=-0.2) (α=-0.35) (α=-0.50) price elasticity α. If the price elasticity is higher than
1 BW -0.22 0.36 0.95 it is expected, a given price-cost margin would be
2 ER -0.09 0.59 1.27 “explained” by less competitive, i.e. larger conjectural
variation. A natural question to ask is how sensitive
3 GSTONE 0.00 0.74 1.49 these results are to various modifications in the
4 HWPS -0.14 0.50 1.14 underlying price elasticity and cost parameter
5 LD -0.17 0.46 1.08 estimation.
6 LOYY 0.04 0.83 1.61
7 LY -0.23 0.34 0.92 In order to check on the fragility or robustness of the
8 MP -0.18 0.43 1.04 basic conclusions, a sensitivity analysis over the range
of plausible parameters is investigated. The values of
9 STAN 0.00 0.75 1.50
price elasticities α are -0.2, -0.35, -0.50, while the
10 TARONG -0.09 0.60 1.29 values of marginal cost are 95%MC, 100%MC,
11 YWPS -0.11 0.56 1.23 105%MC, where MC is the mean marginal cost of any
firm in Table 3. There are therefore nine sets of results,
which are shown in Table 7 (only results of Firm BW
Table 6 Conjectural variations of all firms in different are reported). It shows that increase in price elasticity
oligopoly models causes the estimated conjectural variation to rise
substantially. It is clear that an error in price elasticity
No. GencoID Bertrand Cournot Monopoly could undermine the results substantially. As for 5%
1 BW -1 0 4.41 variation in marginal cost, the estimated conjectural
variations are decreasing as marginal cost becomes
2 ER -1 0 4.94
larger, but this effect is relatively modest. It means
3 GSTONE -1 0 5.69 that the sensitivity of results to the price elasticity is
4 HWPS -1 0 28.94 probably of more concern than sensitivity to marginal
5 LD -1 0 11.4 cost.
6 LOYY -1 0 20.21
7 LY -1 0 16.42 The final observation of this sensitivity analysis is to
8 MP -1 0 6.31 take note of the “extreme bounds” of the conjectural
variation. The highest estimated conjectural variation
9 STAN -1 0 16.98 is 1.06, while the lowest is -0.27. Even these extreme
10 TARONG -1 0 17.87 bounds don’t approach either the Bertrand or
11 YWPS -1 0 29.15 Monopoly. Thus it shows that strong evidence against
the monopoly hypothesis and against the highly
Now, the behavior of firms in oligopoly market can be competitive Bertrand hypothesis has been found. The
analyzed. Comparing Table 5 with Table 6, one can Cournot behavior can be plausible for the NEM
see that all the calculated conjectural variations in market, taking into account the various errors and
Table 5 are in the “reasonable range”. It means that approximations. For the Bertrand model to be
the estimated values are between the extreme bounds, supported in the market, prices have to be at or close
which are the conjectural variations of Bertrand case to marginal costs. The main reason for the poor
and monopoly case as shown in the columns 3 and 5 performance of the Bertrand model here is simply that
of Table 6. Note that there is nothing that forces the the mean price is more than the marginal costs. As for
estimated conjectural variations to be in this range. monopoly model, the mean price was simply not large
These values can, from a purely computational point enough in the year 2002 to be consistent with
of view, quite easily be large negative or positive monopoly behavior. Both the Bertrand and monopoly
numbers. The apparent reasonableness of the might be regarded by many economists as behavioral
calculated conjectural variations shows the extremes which will not be observed very often in the
effectiveness of the proposed method. actual data. Thus, a rejection of these models would
not be surprising. The relatively strong performance of
From Tables 5 and 6, it shows that when the price the Cournot model is perhaps more reasonable in
elasticity α = -0.20, the estimated conjectural NEM.
 Table 7 Sensitivity of conjectural variations
(Firm BW)

α=-0.2 α=-0.35 α=-0.5

95%MC -0.18 0.44 1.06
100%MC -0.22 0.36 0.95
105%MC -0.27 0.28 0.83

6. CONCLUSION

This paper presents a theory and method for the

statistical estimation of the conjectural variation. This
proposed method is applied to the Australia National
Electricity Market (NEM). All the parameters needed
to calculate the conjectural variation, the values of the
marginal cost MCi, the price elasticity α, the market
clearing price p and the market share si, are estimated
based on the actual data of the NEM. The values of
conjectural variation of firms are calculated based on
these estimated parameters. From the analysis of
conjectural variations in the NEM, the Cournot
behavior will be accepted more reasonably, while the
Bertrand and monopoly behavior are strongly rejected.

This paper empirically analyzes the static behavior of

firms in the NEM in a period by the empirical
knowledge of the conjectural variation of firms. In real
life, firms play a repeated game. It can be imagined
that a firm can develop its behaviors by learning the
behaviors of its rivals. Dynamic strategic interaction
could be a more important aspect of strategic bidding
study in the oligopolistic electricity market. The
dynamic interaction among firms in the actual
electricity market will be studied in the future.

7. REFERENCES

[1] Tirole, J., “The theory of industrial

organization”, The MIT Press, Cambridge,
Massachusetts, London, England, 1989

[2] Church, J., and Ware, R., “Industrial

Organization: A Strategic Approach”, Irwin
McGraw-Hill, Boston, 2000.

[3] Website: http://www.nemmco.com.au/

[4] National Institute of Economic and Industry

Research, “The price elasticity of demand for
electricity in NEM regions”, a report for the
National Electricity Market Management
Company, 2002