=Paper=
{{Paper
|id=Vol-223/paper-24
|storemode=property
|title=Evolutionary Stability of Behavioural Types in the Continuous Double Auction
|pdfUrl=https://ceur-ws.org/Vol-223/38.pdf
|volume=Vol-223
|authors=Perukrishnen Vytelingum (University of Southampton),Dave Cliff (University of Southampton),Nicholas R. Jennings (University of Southampton)
|dblpUrl=https://dblp.org/rec/conf/eumas/VytelingumCJ06
}}
==Evolutionary Stability of Behavioural Types in the Continuous Double Auction==
https://ceur-ws.org/Vol-223/38.pdf
Evolutionary Stability of Behavioural Types in the
Continuous Double Auction∗
Perukrishnen Vytelingum, Dave Cliff and Nicholas R. Jennings
School of Electronics and Computer Science,
University of Southampton, Southampton, SO17 1BJ, UK
{pv03r,dc,nrj}@ecs.soton.ac.uk
November 17, 2006
Abstract
In this paper, we investigate the effectiveness of different types of bidding behaviour
for trading agents in the Continuous Double Auction (CDA). Specifically, we consider be-
havioural types that are neutral (expected profit maximising), passive (targeting a higher
profit than neutral) and aggressive (trading off profit for a better chance of transacting).
For these types, we employ an evolutionary game-theoretic analysis to determine the pop-
ulation dynamics of agents that use them in different types of environments, including
dynamic ones with market shocks. From this analysis, we find that given a symmetric de-
mand and supply, agents are most likely to adopt neutral behaviour in static environments,
while there tends to be more passive than neutral agents in dynamic ones. Furthermore,
when we have asymmetric demand and supply, agents invariably adopt passive behaviour
in both static and dynamic environments, though the gain in so doing is considerably
smaller than in the symmetric case.
Summary
The last decade has seen a significant change in the nature of electronic commerce with the
emergence of trading agents, software that is capable of autonomous and flexible action to
achieve its objectives and that is endowed with sophisticated strategies for maximising profit
in different types of market mechanisms. Now, one of the most important such mechanisms is
the Continuous Double Auction (CDA), a symmetric auction with multiple buyers and sellers.
CDAs are so important because they are the principal financial institution for trading securities
and financial instruments (e.g. the NYSE and the NASDAQ both run variants of the CDA
institution). However, developing agents that can participate in the CDA is difficult because it
is not amenable to a game-theoretic analysis and there is no known optimal strategy. Therefore,
a number of heuristic strategies have been proposed, each of which has a particular behaviour in
the market. Given this, in this paper, we are not concerned with developing yet another strategy,
but rather we are interested in how a particular characteristic of such strategies impacts upon
their behaviour and their effectiveness. The characteristic in question is the aggressiveness
of the bidding behaviour, here defined as how eager an agent is to transact. We focus on
aggressiveness in particular because we believe it is a key determinant of success in the market.
In particular, we consider three behavioural types:
1. The neutral agent always submits the quote (a bid or an ask) that maximises its expected
profit. This is the most common type of behaviour and is one that is often hardwired into
various strategies.
2. The passive buyer (seller) submits a lower (higher) quote than its neutral counterpart in
order to try and obtain a more profitable transaction. Thus if it does transact, it makes
more money because it pays less (if it is a buyer) or receives more (if it is a seller).
3. The aggressive buyer (seller) submits a higher (lower) quote than its neutral counterpart
in order to try and improve its chances of transacting as it offers more (if it is a buyer)
or asks for less (if it is a seller) with the aim of making sure that it can trade.
∗ This paper was published in the proceedings of the TADA/AMEC Workshop held at AAMAS 2006, Hako-
date, Japan.
At this time, comparatively few researchers have considered the issue of varying aggres-
siveness in a market context. Abreu et al. describe the evolutionary stability of behaviours
in a reputational model of bargaining, and how players can be more profitable when adopting
passive or aggressive behaviour. Thus, while they show that this attitude can have a significant
effect on the outcomes experienced, they do not consider the CDA. Walsh et al. do consider the
CDA and they use an evolutionary game-theoretic (EGT) analysis to examine the interaction
of a number of common strategies. Thus, their analysis provides an insight on the population
proportion that will adopt each strategy. But, while they restrict their analysis to a particu-
lar set of well-known strategies, we are more interested in whether a particular strategy can
perform better if it is passive, neutral or aggressive. Phelps et al. also use an EGT analysis,
but they compare two double auction mechanisms, the call and continuous, given that similar
strategies are available for both mechanisms. Thus, they are interested in the performance of
double auction mechanisms given particular strategies (including one that is evolved using a
GA).
Against this background, we believe this study of behavioural types is important because it
provides an insight into how this fundamental aspect of bidding behaviour affects the system’s
performance. Thus, these results apply to any CDA strategy that is capable of adjusting its
behaviour along this dimension (e.g. GD, ZIP or RB). Such insights are important because if
an agent can be more profitable by deviating to another behaviour, then it will do so. However,
with every agent in the market doing this, the population distribution of types can change
significantly. Now, an effective trading agent can use knowledge of such dynamics to decide on
what behavioural type to adopt given the particular population distribution of types. Further-
more, because behavioural aggressiveness is usually an endogeneous aspect of a strategy, such
an analysis can assist the strategy designer when assessing the effectiveness of various strate-
gies. To perform our analysis, we adopt a similar EGT approach to that of Walsh et al and we
choose EGT because it allows us to study the dynamics of the CDA when agents are allowed
to evolve in terms of the behaviour they adopt. To do this, however, we need to describe the
model in a form that abstracts the complex bidding that the CDA mechanism entails, into a
simple normal-form game. By so doing, the CDA then becomes amenable to such an analysis.
In particular, we develop a set of strategies that vary only in terms of their different behavioural
types.
As electronic marketplaces become ever more common, we believe software agents will in-
creasingly come to dominate the trading landscape. Their ability to quickly make informed
decisions, based on the available data, make them ideal candidates for automated trading. To
this end, analysing the impact of varying one of the fundamental characteristics of their bidding
behaviour in a range of market situations is an important step. In particular, in this paper,
we show that in a symmetric market, an agent is more likely to adopt an evolutionary stable
neutral behaviour. However, when there are market shocks that increase the equilibrium price
but maintain the symmetry of the market (meaning agents have to update their beliefs of the
market) neutral is no longer the behaviour agents are most likely to adopt. In this case, more
agents change to being passive. We also observe that changing behaviour is not particularly
profitable for an agent in an asymmetric market.
For future work, we intend to look at other types of symmetric and asymmetric demand and
supply, and other types of market shock in order to obtain further insights into how a trader’s
behaviour changes in yet other types of market. For completeness, we also aim to address the
limitation of our model where an agent has the same behavioural type when it is both a buyer
and a seller. In particular, we believe that separately analysing these two roles can be more
insightful, particular in asymmetric markets, where we do not expect the same behaviour from
them.
Acknowledgments
We kindly thank anonymous reviewers for their helpful comments on the paper. The first author
is funded by the DIF-DTC 8.6 project (www.difdtc.com).