Auction Parameters
This page contains explanations of many of the auction parameters which can be manipulated to create different types of auctions.
Contents

I. Auctions

II. Auctioneers

III. Pricing Policies

IV. Valuers

V. Strategies

VI. Learners



I. Auctions

An auction is a mechanism which traders of goods use to conduct transactions. There are many different types of auctions, each of which may differ from the others in a number of parameters.

1.Round Robin Auction

2.Random Robin Auction



1. Round Robin Auction

An auction in which RoundRobinTraders can trade by placing shouts in a synchronous round-robin shedule. This means that the order of bidding in each round will be the same, and bids will be recieved in a specific order, without any bids occuring at the same time.

2. Random Robin Auction

A round-robin auction in which the ordering of traders in randomized for each round.



II. Auctioneers

The Auctioneer is the 'person' who runs the auction. Changes in the type of auctioneer will change the way in which the auction is run. This will significantly change the outcome of the auction and must be tailored to the specific requirements of the auction.

1. Ascending Auctioneer

2. Double Auctioneer



1. Ascending Auctioneer

Auctioneer for standard multi-unit english ascending auction. This is what we commonly think of as an 'auction'. There is one seller who is selling multiple units. A number of buyers will continuously bid higher than the previous highest bid. When none of the buyers are willing the bid any higher, the good will be sold to the highest bidder.

2. Double Auctioneer

In a Double Auction, there are numerous buyers and sellers. The auctioneer will arrange the transactions based on the mechanism and pricing policy.

Clearinghouse

In a Clearninghouse Auction the auctioneer will arrange the transactions at specified intervals (e.g.- At the end of each round or day) based on all outstanding bids and asks.

Continuous

An auctioneer for a double-auction with continuous clearing. The clearing operation is performed every time a shout arrives. Shouts must beat the current quote in order to be accepted.

No Queue

An auctioneer for a Double Auction with continuous clearing and no order queuing. Every time an offer is cleared any pending offers are discarded.



III. Pricing Policies

Pricing policies are the mechanisms by which the auctioneer decides the transaction price for a specific transaction in a Double Auction.

1. Uniform

2. Discriminatory



1. Uniform

A pricing policy in which we set the transaction price in the interval between the ask quote and the bid quote as determined by the parameter k. The pricing policy is uniform in the sense that individual bid and ask prices are ignored, thus all agents performing transactions in the clearing operation will pay the same price.

2. Discriminatory

A pricing policy in which we set the transaction price in the interval between the matched prices as determined by the parameter k. The price for each transaction will be different, and based upon the individual bid and ask prices.



IV. Valuers

Valuers are the policies used by agents in an auction to determine how much a commodity is worth to them. Valuers can only assign a value that falls between the minimum and maximum value for that particular agent. Click on each valuer below to learn how each one determines the value for the agent.

1. Random Valuer

2. Fixed Valuer

3. Interval Valuer

4. Random Schedule Valuer



1. Random Valuer

A valuation policy in which the value is randomly determined across all auctions and all units at agent-initialisation time. Valuations are drawn from a uniform distribution with the specified range.

2. Fixed Valuer

A valuation policy in which we maintain a fixed private valuation independent of time or auction.

3. Interval Valuer

Agents configured with this valuation policy will receive a unique private value from a common set of values starting at the specified minimum value and incrementing by a specified 'step' amoung as each agent is assigned a valuation at agent setup time. This is useful for quickly specifying supply or demand curves with a constant "slope" .

4. Random Schedule Valuer

A valuation policy which specifies a randomly-generated series of valuations for each unit of commodity.



IV. Strategies

Strategies are used by trading agents to determine how they should proceed to bid or ask in an auction. Click on each strategy below to learn about how each one determines how an agent should bid or ask.

1. Equilibrium Price Strategy

2. Fixed Price Strategy

3. GD Strategy

4. Kaplan Strategy

5. MDP Strategy

6. Mixed Strategy

7. Priest Van Tol Strategy

8. Proportional Response Strategy

9. Pure Simple Strategy

10. Random Constrained Strategy

11. Random Unconstrained Strategy

12. Simple Momentum Strategy

13. Stimuli Response Strategy

14. Truth Telling Strategy



1. Equilibrium Price Strategy

A strategy which will bid at the true equilibrium price, if profitable, or bid truthfully otherwise. Although this is not a realistic strategy, it can be useful for testing and control experiments.

2. Fixed Price Strategy

A strategy which maintains a fixed price for bidding.

3. GD Strategy

An implementation of the Gjerstad Dickhaut strategy. Agents using this strategy calculate the probability of any bid being accepted and bid to maximize expected profit. The agents use previous bids and asks to form a belief that their individual bid or ask will be accepted. They then choose an action based on that belief which will maximize their own profits. See "Price Formation in Double Auctions" S. Gjerstad, J. Dickhaut and R. Palmer

4. Kaplan Strategy

An implementation of Todd Kaplan's sniping strategy. Agents using this strategy wait until the last minute before attempting to "steal the bid". See "Behaviour of trading automata in a computerized double auction market" J. Rust, J. Miller and R. Palmer in "The Double Auction Market Institutions, Theories and Evidence" 1992, Addison-Wesley.

5. MDP Strategy

A trading strategy that uses an Markov Decision Process learning algorithm, such as the Q-learning algorithm, to adapt its trading behaviour in successive auction rounds. The current market-quote is hashed to produce an integer state value.

6. Mixed Strategy

A mixed strategy is a strategy in which we play a number of pure strategies with different probabilities on each auction round.

7. Priest Van Tol Strategy

This strategy uses calculations based on profit margin as described in Adaptive Agents in a Persistent Shout Double Auction -Chris Preist, Maarten van Tol.

8. Proportional Markup Strategy

This strategy bids at the specified percentage markup over the agent's current valuation.

9. Pure Simple Strategy

A trading strategy in which we bid a constant mark-up on the agent's private value.

10. Random Constrained Strategy

A trading strategy that in which we bid a different random markup on our agent's private value in each auction round. This strategy is often referred to as Zero Intelligence Constrained (ZI-C) in the literature.

11. Random Unconstrained Strategy

A trading strategy in which an agent bids regardless of its private value. This strategy is often referred to as Zero Intelligence Unconstrained (ZI-U) in the literature.

12. Simple Momentum Strategy

13. Stimuli Response Strategy

A trading strategy that uses a stimuli-response learning algorithm, such as the Roth-Erev algorithm, to adapt its trading behaviour in successive auction rounds by using the agent's profits in the last round as a reward signal.

14. Truth Telling Strategy

A strategy that bids/asks based on the agent's truthful individual value.



VI. Learners

The learners are different algorithms which the agents use to change the way they interact over time. They learn from previous bidding rounds and use that knowledge to change their future behavior.

1. Dumb Learner

2. Meta Learner

3. Roth Erev Learner

4. Q Learner

5. Windrow Hoff Learner



1. Dumb Learner

A learner that chooses the same specified action on every iteration.

2. Meta Learner

3. Roth Erev Learner

A class implementing the Roth-Erev learning algorithm. This learning algorithm is designed to mimic human-like behaviour in extensive form games. See: A.E.Roth and I. Erev "Learning in extensive form games: experimental data and simple dynamic models in the intermediate term" Games and Economic Behiour, Volume 8

NPT Roth Erev Learner

A modification of RothErev to address parameter degeneracy, and modified learning with 0-reward. These modifications are made in the context of using the RE algorithm for trader agents in a double auction. See: "Market Power and Efficiency in a Computational Electricity Market with Discriminatory Double-Auction Pricing" Nicolaisen, Petrov & Tesfatsion in IEEE Transactions on Evolutionary Computation Vol. 5, No. 5, p 504.

4. Q Learner

An implementation of the Q-learning algorithm, with epsilon-greedy exploration.

5. Windrow Hoff Learner

An implementation of the Widrow-Hoff learning algorithm for 1-dimensional training sets.