Agent-Based Methods in Economics and Finance

MAML (Multi-Agent Modeling Language) is a macro-language for Swarm. Its aim is to ease the creation of the most common set of agent-based models by providing a couple of high level constructs and structures in the form of specialized keywords. In this paper we introduce the concepts of MAML through an extension of Chris Preist’s auction model on automated trading.

The original model proposes a persistent shout double auction setup for automated business-to-business electronic trading in which seller and buyer agents trade with an abstract good on a daily basis, one unit a day. Our version of the model extends the original setup in three ways. First, it relaxes the one-unit-a-day constraint. Secondly, it allows for the fluctuation of the supply and the demand by letting agents to enter and leave the market. Finally, it introduces time pressure on agents by setting a limit by which the agents must buy or sell the intended amount of good.

In some sciences, especially in the study of complex systems, computerprograms play an important role as scientific equipment. In the case of computer simulations, the programs under use, can be seen as experimental devices built in software.We don’t, however, forget that Social scientists are not computer scientists, but their skills in the field will have to improve to cope with the growing subject of social simulation and agent based modelling techniques.The Swarm project can be viewed as an aid to researchers to create a standard tool to build simulations.But there’s a problem. Programming with Swarm is not easy, the scientist have to spend a few hours (50 hours, in Pietro Terna’s Opinion) to learn the philosophy of the package, how to use the libraries, how to call the graphic routines, how to call data to plot, etc. etc, and how to program in objective-c. Actually, there’s also a java version of Swarm and the learning time may be decreased a bit, because it is very easy to find a java programmer, but the problem of learning the use and the philosophy of Swarm still remains.For this reason we have thought and programmed a package named VSB (Visual Swarm Builder). It can be viewed as the first visual Rapid Application Development (RAD) tool designed exclusively for

The goal of this work is to show the advantages of using the Swarm simulation platform as a tool for making Experimental Economics. A complex project, called SWIEE, will be presented. Furthermore, some useful examples will be proposed to spread this approach in the experimenters’ community, nowadays accustomed to use closed and proprietary programs. The idea of using Swarm for Experimental Economics derives from the possibility it gives to write simulations in the Java programming language, known to be object oriented and very useful for networked applications. Java allows the design of programs in which an object method can be called by other applications running on computers connected to the Internet, so the RMI (i.e. Remote Method Invocation) standard Java classes are the core of this new usage of Swarm. Furthermore, we illustrate an example (the prisoner’s dilemma) and we report some methodological considerations on the possibility of making experiments considering the Internet like a sort of virtual open laboratory. Finally, we display some future developments of the project, in particular those, very interesting, related to the construction of experiments (or better “simulative” experiments) with both human and artificial agents. More information about SWIEE can be obtained from its web pageshttp://​swiee.​econ.​unito.​it

As the world economy becomes increasingly global, will the financial sector become more stable or fragile? In this paper we study how the pattern of relations linking financial institutions - the network - affects the diffusion of a financial crisis. We analyze two such networks with a computational model: the local network, in which each bank is allowed to interact only with the most immediate neighbors, and the global network, in which each bank is allowed to interact with banks located anywhere in the system.

We find that the network matters both for the amount of illiquidity in the system and for the spread of bankruptcy. When interactions are local, bankruptcy spreads slower but illiquidity hits harder. When interactions are global, bankruptcy spreads faster, but illiquidity presents fewer problems. We conclude that a global system, in which financial institutions are not restricted to interact only with close neighbors, is more efficient in collecting and allocating funds, but is more vulnerable to contagion of bankruptcy crises.

In general, the absolute majority of financial market models is based on the stochastic properties of the asset returns, while the properties of the related asset quantities play a minor role. Starting from these remarks, in this paper we propose a system of nonlinear and stochastic difference equations in which the asset price behaviour and the corresponding asset quantity one are jointly taken into account. More precisely, in order effectively to represent the properties of the real asset price variations, we assume that (also on the basis of well known empirical evidences) their dynamics is distinguished by different stochastic processes alternating each other: the “classical” standard Brownian one, the fractional Brownian motion (which is able to represent the dependence among the returns), and the Pareto-Lévy stable one (which is able to represent the the non-Gaussian distributional features). All these processes are characterized by the same “fractal” quantity, the exponent of Hurst, which is properly utilized in the proposed dynamical model in order to represent the different stochastic properties of the asset price changes. Finally, because of the possible “bad” analytical peculiarities of the system itself, we investigate its dynamics by means of an agent-based approach developed in the Swarm software environment.

This paper examines some complex dynamic behaviour emerging in a contestable market. We use nearly zero-intelligence firms: they all sell at a fixed price an equal quantity of the good; they enter and exit according to the lagged profitability of the market. The financial requirements of production are considered, and a spread between creditor and debtor interest rates is introduced. A deterministic case is studied, whose analytical solution shows the presence of infinite equilibria with respect to the number and average capitalisation of the firms. We then proceed in simulating an extended model with the ACE approach, allowing for heterogeneity of firms with respect to their equity, and idiosyncratic shocks on prices. Some other macrobehaviours emerge, among which an asymmetric distribution of firms’ equity base, phase transition in the volatility of the equity base time series, out of equilibrium persistence.

We introduce here the SUM model-the Surprising (Un)realistic Market model-an agent based framework that allows us to deal with the micro-foundations of a stock market. We avoid any artificially simplified solution about price formation, such as to employ an auctioneer to clear the market; on the contrary, our model produces time series of prices continuously evolving, transaction by transaction.

This paper analyzes the organization of an economy where agents with heterogeneous endowments of (human and/or physical) capital and labor can join and form partnerships in order to produce an output^s. We suppose that the formation of a partnership is the result of a common willingness of a (sub)set of agents to pool and contribute their resources to a production process that takes place within the partnership that has been formed, without any externality onto or from other partnerships. The members of a partnership agree to divide the output among themselves according to an exogenously given distributive rule.

One of the central presumptions in general equilibrium models widely used in neoclassical macroeconomics today is that people are pretty much alike, or homogeneous in their preferences. A second presumption, one of mathematical convenience, is that people’s preferences are exogenous and fairly fixed over time and thus do not vary or adapt readily to changing circumstances. Lastly, rational actor behavior implies the availability of requisite information and the ability of individuals to process that information accurately. There is a vast class of theories that are consistent with these presumptions but such theories have difficulty in explaining persistent skewed distributional outcomes such as income and wealth inequalities within groups and across societies. Because of these limiting presumptions, the proffered explanations for such skewed distributions necessarily refer back to differences in initial resource endowments. We would restate the inequality question more provocatively by asking: Is poverty deliberate?

The CasinoWorld model presented in this paper uses agent-based simulation modeling in an attempt to overcome the limitations of standard equilibrium models. We use the SWARM programming environment to model agents with heterogeneous risk preferences in an environment of uncertainty and allow them to adapt their gaming strategies to maximize survival by preserving their wealth stakes. In this way SWARM allows us to endogenize heterogeneity and adaptability with regard to agent behavior. The results show that players with initialhigh risk preferences who are lucky can adapt their gambling strategies to increase their relative wealth. This model provides a rudimentary foundation for demonstrating that natural, rational behavior and luck can form the basis of inequality. We support the assumptions of the model with a discussion of experimental research in economic behavior, decision-making behavior and evolutionary psychology.

Job search theory involves perfectly rational, maximising actors that make decisions under imperfect information. In this paper we simulate a labour market which works differently: heterogeneous firms and workers interact following rules that place only bounded demand on their computational capacities. We stress the relevance of heterogeneity, formal education, and training on the job in determining employment and skill mismatch phenomena by comparing the ACE model with a more traditional competitive model. Results are interesting from both the theoretical and the empirical point of view.