Agent-Based Simulation Models, Computational Policy and Market Design



Marginal contribution, reciprocity and equity in segregated groups: Bounded rationality and self-organization in social networks (2007)



We study the formation of social networks that are based on local interaction and simple rule following. Agents evaluate the profitability of link formation on the basis of the Myerson–Shapley principle that payoffs come from the marginal contribution they make to coalitions. The NP-hard problem associated with the Myerson–Shapley value is replaced by a boundedly rational ‘spatially’ myopic process. Agents consider payoffs from direct links with their neighbours (level 1), which can include indirect payoffs from neighbours’ neighbours (level 2) and up to M-levels that are far from global. Agents dynamically break away from the neighbour to whom they make the least marginal contribution. Computational experiments show that when this self-interested process of link formation operates at level 2 neighbourhoods, agents self-organize into stable and efficient network structures that manifest reciprocity, equity and segregation, reminiscent of hunter gather groups. A large literature alleges that this is incompatible with self-interested behaviour and market oriented marginality principle in the allocation of value. We conclude that it is not this valuation principle that needs to be altered to obtain segregated social networks as opposed to global components, but whether it operates at level 1 or 2 of social neighbourhoods. Remarkably, all M>2 neighbourhood calculations for payoffs leave the efficient network structures identical to the case when M=2.

Kirman, A., Markose, S., Giansante, S. and Pin, P. (2007). Marginal contribution, reciprocity and equity in segregated groups: Bounded rationality and self-organization in social networksJournal of Economic Dynamics and Control, 31(6), pp.2085-2107.


Designing large value payment systems: An agent-based approach (2011)



The purpose of this paper is to show how agent-based simulations of payment systems can be used to aid central bankers and payment system operators in thinking about the appropriate design of payment settlement systems to minimise risk and increase their efficiency. Banks, which we model as the ‘agents’, are capable of a degree of autonomy with which to respond to payment system rules and adopt a strategy that determines how much collateral to post with the central bank at the start of the day (equivalently how much liquidity to borrow intraday from the central bank) and when to send payment orders to the central processor. An interbank payment system with costly liquidity requires banks to solve an intraday cash management problem, minimising their liquidity and delay costs subject to their beliefs about what the other banks are doing. Some preliminary results are given on how banks learn to endogenously determine how much liquidity to post in the interbank liquidity management game.

Markose, Sheri M and Alentorn, Amadeo and Millard, Stephen and Yang, Jing (2011) Designing large value payment systems: An agent-based approachEconomics Department Discussion Paper No. 700, University of Essex.


  • Interbank Payment System Simulator – Project with the Bank of England to study liquidity and risk in large payment systems with agents.

Presentation at the CCBS (Bank of England) at the Expert Forum: Payment System Architecture and Oversight: 31-Jan-2005 to 02-Feb-2005



Dynamic Learning, Herding and Guru Effects in Networks (2004)



It has been widely accepted that herding is the consequence of mimetic responses by agents interacting locally on a communication network. In extant models, this communication network linking agents, by and large, has been assumed to be fixed. In this paper we allow it to evolve endogenously by enabling agents to adaptively modify the weights of their links to their neighbours by reinforcing ‘good’ advisors and breaking away from ‘bad’ advisors with the latter being replaced randomly from the remaining agents. The resulting network not only allows for herding of agents, but crucially exhibits realistic properties of socio-economic networks that are otherwise difficult to replicate: high clustering, short average path length and a small number of highly connected agents, called ‘gurus’. These properties are now well understood to characterize ‘small world networks’ of Watts and Strogatz (1998).

Markose, Sheri M and Alentorn, Amadeo and Krause, Andreas (2004) Dynamic Learning, Herding and Guru Effects in Networks. Economics Department Discussion Paper No. 582, University of Essex.


The Herding and Networks Simulator  Presentation (2007)
The Herding and Networks Simulator was developed at CCFEA by Sheri Markose and Amadeo Alentorn.

  • Herding Simulator – an application to study herding, guru effects and star formations with dynamic learning on networks.

For any further information, please contact