COSMOS description

CWE Market Coupling algorithm

Version 1.1
Date 14 January 2011
Status Draft Final

Document creation and distribution

Document Owner Rouquia Djabali, Joel Hoeksema, Yves Langer,
Function Cosmos development group
File location
Distribution Publicly available

Approval
Version Date Name Function Signature

Previous versions
Version Date Author Summary of changes

Related documents

Attachments
1 Summary ................................................................................................ 3

2 General principles of market coupling .................................................... 3

2.1. General principle of market coupling ................................................... 3
2.2. ATC market coupling ......................................................................... 3

3 COSMOS in a nutshell ............................................................................. 5

4 Market constraints ................................................................................. 5

4.1. Hourly orders ................................................................................... 5
4.2. Profile block orders ........................................................................... 6

5 Network Constraints .............................................................................. 6

5.1. ATC-Based constraints ...................................................................... 6
5.2. Flow-Based constraints (used for FB parallel runs) ................................ 7

6 Functioning of COSMOS .......................................................................... 7

6.1. Algorithm ........................................................................................ 7
6.2. Precision and rounding ...................................................................... 9
6.3. Price boundaries ............................................................................... 9
6.3.1. Price boundaries and network constraints ................................. 9
6.3.2. Extreme prices and curtailment .............................................. 10
6.4. Optimality and quality of the solution ................................................. 10
6.5. Time control ................................................................................... 10
6.6. Transparency .................................................................................. 10

7 Further geographic and product extensions ......................................... 11

8 APPENDIX 1: Mathematical formulation ............................................... 12

8.1. Sets ............................................................................................... 12
8.2. Data .............................................................................................. 12
8.3. Variables ........................................................................................ 12
8.4. Market Constraints .......................................................................... 12
8.5. Network Constraints ........................................................................ 13
8.6. Objective ........................................................................................ 13
8.7. Summary ....................................................................................... 13

9 APPENDIX 2: Price indeterminacy rules ............................................... 14

9.1. Notations........................................................................................ 14
9.2. Mid point rule formulation for price determination................................ 14
9.3. Prices to be published ...................................................................... 15
9.4. Summary ....................................................................................... 15

10 APPENDIX 3: Volume indeterminacies and curtailment rules ............... 16

10.1. Definitions and objective .................................................................. 16
10.2. Avoiding curtailment ........................................................................ 17
10.3. Minimizing and sharing curtailment .................................................... 18
10.4. Maximizing traded volume ................................................................ 19
10.5. Summary ....................................................................................... 19
COSMOS description CWE Market Coupling algorithm

1 Summary

The CWE project parties have selected COSMOS as the algorithm to calculate daily market
coupling results. COSMOS is a branch-and-bound algorithm designed, in collaboration with
N-SIDE, to solve the problem of coupling spot markets including block orders. It naturally
treats all technical and product requirements set by the CWE project, including step and
interpolated orders, flow-based network under PTDF representation, ATC links and DC
cables (possible with ramping, tariffs and losses), profiles block orders, flexible blocks
orders and linked block orders.

COSMOS outputs net export positions and prices on each market and each hour, the set of
executed orders, and the congestion prices on each tight network element. These outputs
satisfy all requirements of a feasible solution, including congestion price properties and the
absence of Paradoxically Accepted Blocks.

This document only describes the features that are currently in use in the CWE context,
though COSMOS already integrates many additional features such as those to be expected
in a context of product and geographic extensions.

As such, COSMOS algorithm is also used for local and coupling auctions such as:
- EPEX SPOT for CH;
- HUPX, OTE and OKTE for CZ-HU-SK coupling.

In those market areas, the COSMOS market constraints are defined in local market rules.

2 General principles of market coupling

2.1. General principle of market coupling

Market coupling is both a mechanism for matching orders on power exchanges (PXs) and
an implicit cross-border capacity allocation mechanism. Market coupling optimizes the
economic efficiency of the coupled markets: all profitable deals resulting from the
matching of bids and offers in the coupled markets of the PXs are executed; matching
results are however subject to capacity constraints calculated by Transmission System
Operators (TSOs) which may limit the flows between the coupled markets.

Market prices and schedules of the connected markets are simultaneously determined with
the use of the available capacity defined by the TSOs. The transmission capacity is thereby
implicitly auctioned and the implicit cost of the transmission capacity is settled by the price
differences between the markets. In particular, if no transmission capacity constraint is
active, then there is no price difference between the markets and the implicit cost of the
transmission capacity is null.

2.2. ATC market coupling

Under ATC, Market coupling relies on the principle that the markets with the lowest prices
export electricity to the markets with the highest prices. Between two markets, two
situations are possible: either the ATC is large enough and the prices of both markets are
equalized (price convergence), or the ATC is not sufficient and the prices cannot be
equalized. These two cases are described in the following examples.

Suppose that, initially, the price of market A is lower than the price of market B. Market A
will therefore export to market B, the price of market A will increase whereas the price of
market B decreases. If the ATC from market A to market B is sufficiently large, a common
price in the market may be reached (PA* = PB*). This first case is illustrated in Figure 1.

Version 1.1 – Date 14 January 2011 Page 3 of 19

COSMOS description CWE Market Coupling algorithm

Figure 1: Market coupling of two markets with no congestion

The other case, illustrated in Figure 2, happens when the ATC is not sufficient to ensure
price harmonization between the two markets. The amount of electricity exchanged
between the two countries is then equal to the ATC and the prices PA* and PB* are given
by the intersection of the purchase and sale curves. Exported electricity is bought in the
export area at a price of PA* and is sold in the import area at a price of PB*. The
difference between the two prices multiplied by the exchanged volume – i.e. the ATC – is
called “congestion revenue”, and is collected and used pursuant to article 6.6 of the
Regulation (EC) N° 1228/2003 of the European Parliament and of the Council of 26 June
2003 on condition for access to the network for cross-border exchanges in electricity.

Figure 2: Market coupling of two markets with congestion

Version 1.1 – Date 14 January 2011 Page 4 of 19

COSMOS description CWE Market Coupling algorithm

3 COSMOS in a nutshell
This section describes the model and the algorithm that have been chosen to solve the
problem associated with the coupling of the day-ahead power markets in the CWE region.

Market participants submit orders on their respective power exchange. The goal is to
decide which orders to execute and which to reject and publish prices such that:

1
• The social welfare generated by the executed orders is maximal.
• Orders and prices are coherent.
• The power flows induced by the executed orders, resulting in the net positions do
not exceed the capacity of the relevant network elements.

COSMOS has been initially developed by BELPEX, in collaboration with N-SIDE, a company
specialized in optimization solutions based on operations research & modeling. The
purpose of this algorithm is to deal with the CWE coupling problem in way that allows
considering more general aspects of market coupling such as constraints that would arise if
coupling with neighboring markets. COSMOS is currently co-owned by APX-ENDEX, BELPEX
and EPEX SPOT.

In summary, the COSMOS algorithm:

• naturally treats standard and “new” order types with all their requirements,
• naturally handles both Available Transmission Capacity (ATC) and Flow-Based (FB)
network representations as well as possible alternative models and HVDC cable
features,
• implements specific curtailment rules for those cases where price boundaries are
not harmonized
• is not limited by the number of markets, orders or network constraints,
• finds quickly (within seconds) a very good solution in all cases (even with problems
with 350000 hourly orders and 1800 block orders in more than 10 markets),
• continues improving this initial solution until the time limit (e.g. 10 min) is
reached,
• generating several feasible solutions during the course of its execution,
• unless it can show that the mathematically optimal solution has been found (which
is most often the case), in which case it stops before the time limit.

In the two following sections, we detail which products and network models can be
handled by COSMOS. Section 6 gives a high-level description of how COSMOS works, and
section 7 provides additional information related to the functionalities and behaviors of the
algorithm.

4 Market constraints
Market constraints2 are those applying to the orders submitted to the exchanges. The list
presented hereunder proposes a set of all products available in at least one CWE
Exchange.

4.1. Hourly orders

Depending on markets needs and on already existing systems, hourly orders can be either
stepwise (BELPEX, APX-ENDEX,) or linearly interpolated (EPEX SPOT).

The fixing of hourly orders satisfies the following constraints:

1
Social welfare is defined as: consumer surplus + producer surplus + congestion revenue
across the region. It is the objective function of COSMOS (see “objective” in technical
appendix 1).
2
See also “Market constraints” section in appendix 1 for a mathematical formulation of
these constraints.

Version 1.1 – Date 14 January 2011 Page 5 of 19

COSMOS description CWE Market Coupling algorithm

• An Hourly Offer is rejected when the Market Clearing Price is lower than the offer
(lowest) price limit.
• An Hourly Bid is rejected when the Market Clearing Price is higher than the bid
(highest) price limit.
• An Hourly Offer is executed when the Market Clearing Price is higher than the offer
(highest) price limit.
• An Hourly Bid is executed when the Market Clearing Price is lower than the bid
(lowest) price limit.
• An Hourly Order may be partially executed if and only if the Market Price is equal
to the price limit of that order / is between the two price limits of that order.
• An Hourly Order is not executed for a quantity in excess of the volume limit
specified in the Order.

4.2. Profile block orders

Compared to block orders that were available prior to the CWE market coupling go-live,
block orders in COSMOS can represent profiles, i.e. are defined by distinct volume limits at
each hour.

Block orders are neither partially nor paradoxically executed. Therefore, all orders can only
be either executed fully, or rejected fully. Because of this constraint – called the “fill or kill
3
constraint” - some block orders can be rejected even if they are in the money , in which
case they are called Paradoxically Rejected Blocks (PRB). On the contrary, no block orders
can be executed paradoxically (i.e. executed even if out of the money).

The fixing of block orders satisfies the following constraints:

• A Block Offer is not executed when the average of the rounded Market Clearing
Prices over the relevant hours and weighted by the corresponding volume limits is
lower than the price limit of this order.
• A Block Bid is not executed when the average of the rounded Market Clearing
Prices over the relevant hours and weighted by the corresponding volume limits is
higher than the price limit of this order.
• A Block Order can only be executed at all hours simultaneously, for a quantity
equal to the hourly volume limits specified in the order.

5 Network Constraints
COSMOS is able to tackle the network constraints associated with several network
configurations4 (ATC-Based and Flow-Based – as well as with HVDC cables and ramping
constraints in case of further extensions).

5.1. ATC-Based constraints

With an ATC-Based representation of the network, the cross border bilateral exchanges are
only limited by the ATCs as provided for each hour and each interconnection in both
directions. The algorithm will thus compute the cross border bilateral exchanges that are
optimal in terms of overall social welfare.

The fixing with ATC-Based constraints satisfies the following constraints:

• Cross-border bilateral exchanges are smaller than or equal to the relevant ATC
value.
• Whenever the cross-border exchange is strictly smaller than the relevant ATC
value, then the clearing prices on both sides of the border are equal.

3
A supply (respectively demand) order is said to be in the money if the submission price
of the order is below (resp. above) the average market price.
4
See also “Network constraints” section in appendix 1 for a mathematical formulation of
these constraints.

Version 1.1 – Date 14 January 2011 Page 6 of 19

COSMOS description CWE Market Coupling algorithm

• Whenever there is a price difference between two areas, the ATC between these, if
any, is congested in the direction of the high price area.

5.2. Flow-Based constraints (used for FB parallel runs)

Flow-based network representations are set to model more precisely physical electricity
laws.

In a flow-based representation of the network, the flows on a set of given critical network
elements are equal to the product of a PTDF (Power Transfer Distribution Factor) matrix
with the vector of the areas’ net positions. These (unidirectional) flows are limited by the
corresponding transmission capacities provided for each hour. Such PTDF constraints allow
representing explicitly all critical elements and security constraints, but would also support
more simplified network models.

The fixing with Flow-Based constraints satisfies the following constraints:

• For each flow-based constraint, the sum of the area’s net positions of all markets
weighted by the PTDF value is smaller than or equal to the corresponding
transmission capacity.
• The sum of the areas’ net position is equal to zero
• For each flow-based constraint, whenever the sum of the area’s net positions of all
markets weighted by the PTDF value is strictly smaller than the corresponding
transmission capacity, then the congestion price of this constraint is null.
• The price difference between two areas is equal to the sum of the congestion
prices of all capacity constraints weighted by the difference of the corresponding
PTDF values.

6 Functioning of COSMOS
In this section we describe how COSMOS selects orders to be executed or rejected, under
the Market and Networks Constraints. The objective of COSMOS is to maximize the social
welfare, i.e. the total market value of the day-ahead auction.

The main difficulty in determining which orders to execute or reject comes from the fact
that block orders must satisfy the “fill or kill” property.

Without those block orders, the problem is much simpler to solve. Indeed, the problem can
then naturally be modeled as a Quadratic Program (QP)5, which can be routinely solved by
off-the-shelf commercial solvers6. The use of a commercial solver to directly solve this
Quadratic Program would then be the most efficient solution.

The presence of block orders in the order book however makes the problem substantially
more difficult. The problem with block orders can be formulated as a Mixed Integer
Quadratic Program (MIQP) allowing modeling the fill or kill condition of block orders. The
state-of-the-art method used to solve MIQP is called branch-and-bound7. COSMOS has
been designed as a dedicated branch-and-bound algorithm for solving the CWE Market
Coupling problem (the mathematical model of this problem is proposed in appendix 1).

6.1. Algorithm
COSMOS proceeds step by step.
At the first step, COSMOS solves a market coupling QP without fill or kill constraints, hence
allowing all block orders to be partially executed. By chance, the solution of this problem
might satisfy the fill or kill condition for all block orders and is therefore a feasible solution

5
A Quadratic Program (QP) is an optimization problem where an objective (function) of
the second order is to be optimized under linear constraints.
6
COSMOS uses the CPLEX solver.
7
For a more extensive discussion on the branch and bound technique, see for instance:
Integer Programming, Wolsey, 1998

Version 1.1 – Date 14 January 2011 Page 7 of 19

COSMOS description CWE Market Coupling algorithm

of the CWE market coupling problem. In this case, the solution that has been found is the
optimal solution.

Otherwise, COSMOS gradually forces the partially executed block orders to be either fully
rejected or fully executed in subsequent steps, in order to obtain a solution to the CWE
market coupling problem which respects all fill or kill constraints.

At a given step, two situations can occur:

1. COSMOS has produced a solution in which some block orders are either fully
executed or rejected and some block orders are partially executed. This solution
has been computed by solving the initial QP, but in which some block orders have
been forced to be executed or rejected (as the result of some previous steps).
Since it contains partially executed orders, it is called a partial solution. The
property of this solution is that its objective value is an upper bound of the welfare
of any solution that could be produced by extending this partial solution into a
feasible solution by adding further constraints. Two sub-cases can occur:
o Sub-case 1a: If the upper bound associated to this partial solution is
smaller than the welfare of the best feasible solution found so far, COSMOS
will discard this partial solution and won’t consider it anymore.
o Sub-case 1b: Otherwise, COSMOS will select a block order partially
executed and create two new steps to be analyzed: in the first of these
new steps, the selected block is forced to be executed, and in the second
one it is forced to be rejected.

2. COSMOS has produced a solution in which all block orders are either fully executed
or fully rejected (even those that were not forced to). In this case, COSMOS must
still check whether there exist prices that are compatible with this solution and the
constraints (which is done by verifying that all market and network constraints are
satisfied). Two cases can occur:
- Sub-case 2a: If such prices exist, COSMOS has found a feasible
solution. If this solution is better than the best one found so far8, it
is marked as such.
- Sub-case 2b: If no such prices exist, then a new step is created
with a transformed problem containing additional constraints to
exclude this non feasible solution.
During the course of its execution, COSMOS might sometimes increase the number of
steps that it has yet to consider (e.g. sub-cases b) or reduce it (sub-cases a). When there
remains none, this means that COSMOS has finished and has found the best possible
solution. Possibly, COSMOS will reach the time limit although there remain some partial
solutions that were not analyzed. In this case, COSMOS will output the best solution found
so far without being able to prove whether it is the very best possible one.

8
Or if it is the first feasible solution found.

Version 1.1 – Date 14 January 2011 Page 8 of 19

COSMOS description CWE Market Coupling algorithm

Here is a small example of the execution of COSMOS:

- First node
- Solution value 3500
- Blocks 23 and 54 fractional
Case1a

Case2a
- third node, block 23 executed
- Solution value 3440
- all blocks integral, no prices

constraints added
Case2b
- second node, block 23 rejected
- Solution value 3050
- fourth node, block 23 executed + constraints
- all block integral, there exist prices
Case1b - Solution value 3300

feasible solution found
- block 30 fractional

- fifth node, block 23 executed, block 30 rejected + constraints - sixth node, block 23 executed, block 30 executed + constraints
- Solution value 3100 - Solution value 3000
- all block integral, there exist prices
there cannot exist better solutions here!

better solution found!

6.2. Precision and rounding

COSMOS provides exact results which satisfy all constraints with a target tolerance of 10-5
(and in any case 10-3). Those exact prices and volumes (net positions) are rounded by
applying the commercial rounding (round-half-up) convention before being published. The
size of the tick varies depending on the data considered. For instance:
• Prices at APX-ENDEX BELPEXSPOT are rounded with two digits (e.g. 0.01 €/MWh)
• Prices at EPEX SPOT are rounded with three digits on the FR market (e.g. 0.001
€/MWh) and with two digits on the DE/AU market
• Net positions in BE, DE and NL are rounded with 1 digit (e.g. 0.1 MWh)
• Net positions in FR are rounded with no digit (e.g. 1 MWh)

NB: rounding the results imposes to accept some tolerances on constraints. Typically, this
tolerance is equal to the sum the precision ticks of all rounded values divided by two. For
instance, the sum of the net positions of all bidding areas must be zero, with a tolerance of
0.65 MWh (the sum of the net position ticks of all markets divided by two) for the CWE
MC.

6.3. Price boundaries

Published prices must be within predefined boundaries. It is intended that all price
boundaries will be harmonized so that prices are in the [-3000 €/MWh,+3000 €/MWh]
interval, though the algorithm is designed to also support different price boundaries

6.3.1. Price boundaries and network constraints

Generally speaking, different price boundaries can be implemented in COSMOS, but not
together with the network price properties as commonly defined. In particular, flow-based
models in general hinder the possibility to impose boundaries on prices at all in some
particular cases.

In order to accommodate technical price boundaries and to compute coherent prices (in
the sense that they respect market and network constraints), COSMOS guarantees on the
one hand that market and network constraints are satisfied with respect to unrounded
prices. On the other hand, COSMOS also ensures that market constraints are satisfied

Version 1.1 – Date 14 January 2011 Page 9 of 19

COSMOS description CWE Market Coupling algorithm

using rounded and within bound prices. Hence some network constraints are not checked
against rounded and within bound prices, but only against unrounded and possibly out of
bounds prices. This allows computing coherent prices while respecting the local price
boundaries and is currently only relevant in flow-based simulations.

See technical appendix 2 for more information.

6.3.2. Extreme prices and curtailment

Generally speaking, Cosmos is designed to avoid curtailment situations – i.e. situations
when price taking orders are not fully satisfied. More precisely, COSMOS enforces local
matching of price taking hourly orders with hourly orders in the opposite direction and in
the same market as counterpart. Hence, whenever curtailment of price taking orders can
be avoided locally on an hourly basis – i.e. the curves cross each other - then it is also
avoided in the final results. In case the local matching does not allow to fulfill all the price
taking orders, then this curtailment is equally apportioned between all markets (subject to
network constraints).

See technical appendix 3 for more information.

6.4. Optimality and quality of the solution

During the course of its execution, COSMOS will typically generate several feasible
solutions. The best one in terms of welfare is selected among these solutions at
termination of the algorithm.
By optimizing welfare, COSMOS also avoids – whenever possible - generating solutions
with paradoxically rejected orders (PRBs), and especially the ones with large volume
and/or largely in the money.

COSMOS algorithm selects the solutions with the largest welfare, but discards during its
computation the solutions with paradoxically rejected blocks that are very deep in the
money. This is implemented to guarantee fairness, as this could only happen with blocks of
small volume.

6.5. Time control

COSMOS is tuned to provide very quickly a first feasible solution. It can be shown that the
upper bound in terms of computing time to obtain a first feasible solution is linear in terms
of number of block orders. In practical cases, the first feasible solution has been found
within less than 30 seconds on all our CWE instances.

Due to the combinatorial complexity of the problem, this is obviously not true for the
computing time to obtain the optimal solutions. Nevertheless, most of the instances were
solved at optimality in less than 10 minutes (which is the maximal time allowed in the CWE
coupling process), the remaining showing quite small distances to optimality after this time
limit.

6.6. Transparency
Generally speaking, COSMOS is based on sound and robust concepts and has a good
degree of transparency. In particular, COSMOS is perfectly clear and transparent as to
what are feasibility and optimality. More precisely, COSMOS will typically consider all
feasible solutions and choose the best one according to the agreed criterion (welfare-
maximization).

Also, COSMOS optimizes the total welfare, so that the chosen results are well explainable
to the market participants: published solutions are the ones for which the market value is
the largest. In addition, in order to avoid undesirable solutions, COSMOS will not output
solutions in which blocks that are unduly deep in the money are rejected paradoxically.

Version 1.1 – Date 14 January 2011 Page 10 of 19

COSMOS description CWE Market Coupling algorithm

7 Further geographic and product extensions

During the design and implementation of COSMOS, great care has been taken to ensure
that the additional requirements aiming at supporting potential extensions in the product
range or the geographical scope of the coupling (or possibly both of them) are also met. In
particular, all the foreseeable requirements necessary to facilitate the mid-term
extendibility of the COSMOS solution (linked and flexible orders of NPS, ramping constraint
of NorNed, charges and losses of BritNed and IFA) are already supported by COSMOS.

Furthermore, COSMOS uses a very general method for solving the market coupling
problems with “fill or kill” constraints. The ability of the algorithm to handle new products
or new requirements is thus excellent as long as the constraints remain of the same type
(linear constraints, with possible fill or kill conditions).

Version 1.1 – Date 14 January 2011 Page 11 of 19

COSMOS description CWE Market Coupling algorithm

8 APPENDIX 1: Mathematical formulation

In this appendix, we describe the mathematical formulation of the Market Coupling
Problem that is solved by COSMOS. As mentioned earlier, it is a Mixed-Integer Quadratic
Program (MIQP).

8.1. Sets
Let us first introduce some sets, and their indices:
- Bidding areas : m
- Hours: h
- Hourly orders: o
- Profile block orders: b
- Unidirectional ATC lines: l

8.2. Data
Here are the input data of the Coupling problem:
- qo is the quantity of hourly order o; it is considered positive for
supply orders and negative for demand orders
- p0o is the price at which an hourly order starts to be accepted
- p1o is the price at which an hourly order is fully accepted
(in the case of step orders, p0o= p1o; in the case of interpolated
supply orders, p0o ≤ p1o; in the case of interpolated demand order
p0o ≥ p1o)
- qb,h is the quantity of profile block b on period h; it is considered
positive for supply orders and negative for demand orders
- bidding area (b) is the bidding area in which profiled block order
b originates
- hours(b) is the set of hours on which profile block order b spans
- bidding area(o) is the bidding area in which hourly order o b
originates
- hour(o) is the hour on which hourly order o spans
- from(l) is the bidding area from which line l is originating
- to(l) is the bidding area to which line l is leading
- capacityl,h is the capacity on the ATC line l on period h

8.3. Variables
- 0 ≤ ACCEPTo ≤ 1 Acceptance of the hourly order
- ACCEPTb∈{0,1} Acceptance of the block order
- 0 ≤ FLOWl,h Flow on the line l at period h
- MCPm,h Market clearing price at market m and period h
- 0 ≤ ATC_PRICEl,h Congestion price of the capacity constraint of line l at period h

8.4. Market Constraints

These are the constraints linking the prices to the orders selection.

- An hourly order o may be accepted only if it is at- or in-the-money:

( )
ACCEPTo > 0 ⇒ qo ⋅ MCPbidding area( o ),hour( o ) − po0 ≥ 0
- An hourly order o must be refused if it is out-of-the-money:
( )
qo ⋅ po0 − MCPbidding area( o ),hour( o ) > 0 ⇒ ACCEPTo = 0

Version 1.1 – Date 14 January 2011 Page 12 of 19

COSMOS description CWE Market Coupling algorithm

- An hourly order o may be partially rejected only if it is at-the-money:

0 < ACCEPTo < 1 ⇒ MCPbidding area( o ),hour( o ) = po0 + p1o − po0 ⋅ ACCEPTo ( )
- An hourly order o must be fully accepted if it is in-the-money:
( )
qo ⋅ p1o − MCPbidding area( o ),hour( o ) < 0 ⇒ ACCEPTo = 1
- An accepted block b must be in-the-money:
ACCEPTb = 1 ⇒ ∑ q ⋅ (MCP
b,h
h∈hours ( b )
bidding area ( b ), h − pb ) ≥ 0

8.5. Network Constraints

These are the constraints on the physical exchanges and the prices arising
from the ATC network model of the CWE region.

- The bidding area m must be in balance on hour h, meaning that the sell and the
import volumes must match the purchase and the export volumes:

∑ ACCEPT
o ,bidding area ( o ) = m ,
o ⋅ qo + ∑
b ,bidding area ( b ) = m ,
q ACCEPT = ∑ FLOWl ,h − ∑ FLOWl ,h
b, h b l , from (l ) =m l ,to ( l ) = m
hour ( o ) = h h∈hours ( b )

- The maximum flow on an ATC line l on hour h must not exceed the available
capacity.
FLOWl ,h ≤ Capacity l ,h
- The congestion price of an ATC line l on hour h can be positive only if the line is
congested.
ATC _ PRICE l , h > 0 ⇒ FLOWl , h = Capacity l , h

- The (positive) congestion prices of the capacity constraints must be equal to the
price difference between the destination and the source bidding areas of the ATC
line l on hour h:
- ATC _ PRICE l , h = MCPto(l ), h − MCPfrom(l ), h

8.6. Objective
The objective of the Market Coupling Problem is to maximize the total welfare:
 p 0 + po1 po0 − po1 
max ∑ − qo ACCEPTo  o + (1 − ACCEPTo ) − ∑ pb ACCEPTb ∑ qb ,h
o  2 2  b h∈hours ( b )
Here the welfare is defined as the difference between the cumulative amount that the
buyers are ready to pay and the cumulative amount that the sellers want to be paid
(quantities are signed) over all markets and hours. This difference corresponds to the sum
of the surplus of the producers and the consumers, plus the congestion revenue.
Note that the objective is quadratic in the ACCEPTo variables (for interpolated orders).

8.7. Summary
COSMOS is an algorithm that generates several (i.e. virtually all) solutions which
satisfy the constraints listed under “Market Constraints”, and “Network
Constraints” and chooses the best one according to the criteria formulated under
“Objective”.

Version 1.1 – Date 14 January 2011 Page 13 of 19

COSMOS description CWE Market Coupling algorithm

9 APPENDIX 2: Price indeterminacy rules

There are cases where multiple solutions of the Market Coupling Problem have identical
block selections and identical welfares but different clearing prices. In the most frequent
cases, this arises when offer and demand curves overlap on a vertical segment. These
cases are called price indeterminacies. The aim of this appendix is to describe the rules
implemented in COSMOS to choose amongst these clearing prices.

UB

MCP*
LB

Figure 3: Simple case of price indeterminacy

In a nutshell, COSMOS will pick the mid-point of the intersection, that is MCP* in the figure
above. However, in a realistic problem with several markets coupled over several hours,
this solution might be non-feasible. This would be the case for example if MCP* invalidates
the selection of block (i.e. the price acceptance of block orders are not compatible with
MCP*). In such a case, COSMOS will pick up the price which is closed to the mid point and
which respects all other price constraints.

This section explains in details how COSMOS selects prices in indeterminate cases9.

9.1. Notations
Let’s note:
LBm,h: the lowest possible price in the market m on hour h for a defined net position Q*.
UBm,h: the highest possible price in the market m on hour h for a defined net position Q*.
MCPm,h: the (unrounded) Market Clearing Price of the market m on hour h.

The mid point of the intersection defined by [LBm,h, UBm,h] is equal to (UBm,h+LBm,h)/2.

9.2. Mid point rule formulation for price determination

A potential feasible solution produced by the algorithm is definitively a feasible solution
once unique rounded prices satisfying all constraints can be defined for this solution.
In case of non-unique price solution, for a given hour h, the mid-point rule is defined
as the minimal square distance between MCPm,h and the distance to its “mid point”. In
case of several indeterminacies, the sum of these distances is considered. The mid-
point price expression can be formulated with the following function

9
Note that these rules are independent of the ATC or FB network model

Version 1.1 – Date 14 January 2011 Page 14 of 19

COSMOS description CWE Market Coupling algorithm

2
 UBm, h + LBm,h 
min ∑  MCPm, h − 

m, h  2 
Subject to the following constraints:
• All market and network constraints
• B_ACCEPTo = 1 if the block order o is executed in the feasible solution
= 0 otherwise

9.3. Prices to be published

These prices must be rounded and bounded before being published. This is done as follows

Published _ price m,h = min(max( round (MCPm, h , precision m ), P minm ), P max m )

With:
- Published_Pricem,h: the price published by the Exchange in market m on
hour h
- Precisionm: the precision of prices in market m
- Pminm: the lower price boundary of market m
- Pmaxm: the upper price boundary of market m

Note that market constraints are checked against these prices as well before being
published

9.4. Summary
COSMOS lifts potential price indeterminacies by selecting the prices at the middle
of the feasible interval, taking into account the price constraints arising from
block selection and network configuration. Then Cosmos rounds and enforces
bounds on these prices before publication.

Version 1.1 – Date 14 January 2011 Page 15 of 19

COSMOS description CWE Market Coupling algorithm

10 APPENDIX 3: Volume indeterminacies and

curtailment rules
There are cases where multiple solutions of the Market Coupling Problem have identical
prices, identical block selections and identical welfares but different volumes sold or
bought on some markets. The cases are called volume indeterminacies. The aim of this
appendix is to describe the rules implemented in COSMOS to choose amongst these
solutions.

MCV*

Figure 4: Simple case of volume indeterminacy

In a nutshell, COSMOS will choose the solutions for which the traded volumes are
maximal, that is MCV* in the figure above.

In addition, these volume indeterminacy rules encompass curtailment rules which are rules
for special volume indeterminacy situations, namely at minimum/maximum prices. Indeed,
a secondary objective of COSMOS is to avoid that price taking orders - i.e. orders at
extreme prices - are not fully executed, this situation being called curtailment10.

10.1. Definitions and objective

Volume indeterminacy occurs when, for
• a given set of market clearing prices (and congestion prices) satisfying the
constraints of the COSMOS model
• a given feasible block selection
There exists many solutions for the hourly orders such that
• they are compatible with the market clearing prices (i.e. out-of-the-money orders
are entirely rejected and in-the-money orders are entirely executed)
• they are compatible with congestion prices (i.e., elements with non-zero
congestion prices are congested).
In this case, all these solutions have an identical welfare. So welfare cannot be a criterion
to select one of them. Thus additional criteria are required for such cases.

In general (not at extreme prices), in case of volume indeterminacy, the criterion is to

maximize traded volume and to share volumes of possible further indeterminacies equally
amongst all the markets. Section 10.4 describes how to achieve this goal.

However, because of volume indeterminacy at extreme prices, hierarchical rules have been
implemented:

10
These rules are independent of the ATC or FB network model

Version 1.1 – Date 14 January 2011 Page 16 of 19

COSMOS description CWE Market Coupling algorithm

• Firstly: avoid curtailment by temporarily matching price taking order locally (see
10.2)
• Secondly: share curtailment (see 10.3).In order to provide fairness, the algorithm
will try to equally share the “curtailed volumes” between all markets.
• Thirdly: maximize traded volume (see 10.4)

10.2. Avoiding curtailment

Avoiding curtailment is desirable in general, as price taking orders allow to close positions
prior to day-ahead nominations. This is done by be the local matching constraint which
enforces the local matching of price taking orders prior to the welfare maximization
procedure (see Figure 5).

Figure 5 Removing price taking volume from Demand and Supply curves

However, it is possible that a market is initially in curtailment, that is, it is a priori

impossible to match locally all price taking orders. In this case only the price taking order
that can be met locally can be discarded from the curves, but the remaining price taking
volume should be submitted, since not providing this volume could in extreme cases lead
to infeasible constraints (e.g. if all markets are initially in curtailment and none would
submit any price taking volume then all impose a constraint that they should export. Since
no market is willing to import, the balance constraint will be violated). The initial
curtailment case is illustrated in Figure 6.

Version 1.1 – Date 14 January 2011 Page 17 of 19

COSMOS description CWE Market Coupling algorithm

Figure 6 Removing price taking volume from Demand and Supply from a market
initially in curtailment

In practice, COSMOS will force the acceptance of price taking hourly supply (respectively
demand) orders by matching them with hourly demand (resp. supply) orders. Let us call
such a constraint “LocalMatchingpmin” (resp. “LocalMatchingpmax”).

The local matching constraints can be written as follows:

• For price taking supply orders:

 
∑ qd
 d ∈Demand 
LocalMatching p min : ACCEPTSupply p min ≥ min 1; 
 ∑ qs 
 s∈Supply p min 
• For price taking demand orders:
 ∑ qs 
 s∈Supply 
LocalMatching p max : ACCEPTDemand p max ≥ min 1; 
 ∑ qd 
 d ∈Demand p max 
Where
• Supplypmin is the set of all supply orders at minimum price (i.e. price taking),
• Demandpmax is the set of all demand orders at maximum price (i.e. price taking).

The local matching constraints are enforced for each market individually prior to the
welfare maximization process. This rule helps finding block selections where curtailment is
reduced. Upon lifting volume indeterminacies, the local matching constraint is dropped in
order to allow a fair sharing of curtailment.

10.3. Minimizing and sharing curtailment

When one or more markets are initially curtailed (see above), the curtailment might not be
avoided through local matching. In this case, the secondary goal is to minimize the volume
curtailed. However two additional issues must be dealt with:
1. The solution must be uniquely defined

Version 1.1 – Date 14 January 2011 Page 18 of 19

COSMOS description CWE Market Coupling algorithm

2. The solution must be fair in apportioning the curtailment between two (or more)
markets initially in curtailment.
To this end, the objective is defined as a quadratic function of the acceptance of the
curtailed orders:

min ∑ qo (1 − ACCEPTo )
2

o∈C

Subject to the following constraints:

• All market and network constraints
• B_ACCEPTo = 1 if the block order o is executed in the feasible solution
= 0 otherwise

where C is the set of price taking orders of markets of whom the curtailment has to be
apportioned and qo is the volume of these price taking orders. In plain English, we
minimize the sum of the non-acceptance of price taking orders, measuring non-acceptance
by the square of the rejected proportion of the price taking order times its volume. This
objective function has several desirable properties:
• it is strictly convex in the space of price taking orders of markets in curtailment,
thus guaranteeing uniqueness of solution for these orders
• It will apportion curtailment according to the size of initial curtailment. For
example suppose two markets initially in curtailment by X and Y MW are coupled
(with infinite capacities) with a third market that can offer Z MW before being itself
in curtailment. Then X/(X+Y)*Z will be given to reduce curtailment in market X
and Y/(X+Y)*Z will be used to reduce curtailment in market Y. Therefore, each
market will see its curtailment decrease by the same proportion Z/(X+Y). Thus
price taking orders in each curtailed market are treated equally.

If two markets are curtailed to a different degree, the market with the least severe
curtailment (in proportion) would help the other reducing its curtailment, so that both
markets end up with identical curtailment ratios.

10.4. Maximizing traded volume

We assume now that for market at Pmin and Pmax the volume indeterminacy has been
settled according to previous section, and volume in these markets in curtailment is fixed.
We turn now our attention to dealing with volume indeterminacy for markets and/or hours
not in curtailment. Here our goal will be to maximize traded volume.

We therefore solve the following optimization problem:

min ∑ qo (1 − ACCEPTo )
2

o∈H
Subject to the following constraints:
• All market and network constraints
• B_ACCEPTo = 1 if the block order o is executed in the feasible solution
= 0 otherwise

where H is the set of supply and demand step curves that are exactly at the market
clearing price of their respective markets and are not at Pmin or Pmax. Again, we want to
minimize the rejection of all (demand & supply) orders and apportion this equally across
markets.

10.5. Summary
COSMOS lifts potential volume indeterminacies with the following priorities:
1. Avoiding curtailment of price taking orders whenever possible
2. Minimizing and sharing the curtailment when unavoidable
3. Maximizing the volume traded in other cases.

Version 1.1 – Date 14 January 2011 Page 19 of 19