We develop a behavioral model as a a computer testbed we can use to study the probable performance of a wide range of mechanisms prior to testing them in a laboratory or using them in practice. In this paper, we describe an implementation of our model and the computer testbed methodology to Groves-Ledyard (1977) mechanisms for provision of public goods. Previous experimental evidence, and some theory, strongly suggest that the value of a free mechanism parameter is important for the dynamic performance of the mechanism (when it is simulated as a repeated game). In our model messages converge to the Nash Equilibrium for all of the values of the mechanism parameter that we studied. However, the convergence times depend on the value of the parameter. Our analysis suggests there are values of the free parameter that result in the fastest convergence. The range of values is robust with respect to the changes in the behavioral model's parameter values and details of the updating procedures. This prediction is validated with data from experiments with human subjects.

We develop a model of individual learning that should allow us to study the probable performance of a wide range of mechanisms. Such a model could be a valuable design tool, allowing mechanism designers to lter out inappropriate designs prior to testing them in a laboratory or using them in practice. In this paper, we study our model in the context of Groves-Ledyard (1977) mechanisms for provision of public goods. Theoretical model does not address the issue of mechanisms implementation in practice in terms of the values of the free parameter of the model. Experimental evidence strongly suggests that this value is important for the implementation of the mechanism (when it is simulated as a repeated stage game) in terms of the speed of convergence to the Nash equilibrium. We investigate the dynamics of our model for a wide range of values of the mechanism's free parameter. We nd that our model converges to Nash Equilibrium for all of the values that we studied. However, the convergence time does depend on the value of the parameter. Our analysis suggests there is a range of values of the free parameter that results in the fastest convergence. This result (and the range of values) is robust with respect to the changes in the learning model's parameter values and details of the updating procedures. Thus our model yields a prediction that can be tested in the experiments with human subjects. The same range of values results in the least amount of variation across dierent runs in terms of the convergence times. This is another testable prediction that could be very important for mechanism's experimental and actual implementation. Finally, we also compare the performance of our model to the results of the experiments with human subjects (Chen and ...

We introduce learning in a principal-agent model of stochastic output sharing under moral hazard. Without knowing the agents ’ preferences and technology the principal tries to learn the optimal agency contract. We implement two learning paradigms- social (learning from others) and individual (learning by doing). We use a social evolutionary learning algorithm (SEL) to represent social learning. Within the individual learning paradigm, we investigate the performance of reinforcement learning (RL), experience-weighted attraction learning (EWA), and individual evolutionary learning (IEL). Overall, our results show that learning in the principal-agent environment is very difficult. This is due to three main reasons: (1) the stochastic environment, (2) a discontinuity in the payoff space in a neighborhood of the optimal contract due to the participation constraint and (3) incorrect evaluation of foregone payoffs in the sequential game principal-agent setting. The first two factors apply to all learning algorithms we study while the third is the main contributor for the failure of the EWA and IEL models. Social learning (SEL), especially combined with selective replication, is much more successful in achieving convergence to the optimal contract than the canonical versions of individual learning from the literature. A modified version of the IEL algorithm using realized

The views expressed are those of the individual authors and do not necessarily reflect official positions of the Federal Reserve Bank of St. Louis, the Federal Reserve System, or the Board of Governors. Federal Reserve Bank of St. Louis Working Papers are preliminary materials circulated to stimulate discussion and critical comment. References in publications to Federal Reserve Bank of St. Louis Working Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors. Social Learning and Monetary Policy Rules

This paper gives an overview of the current research program and some recent results in modelling of modern telecommunication environment obtained in the Department of Industrial Economics and Technology Management, NTNU. We concentrate on different strategic decision problems when it is necessary to take into account uncertainties in demand, technology and other important variables which characterize this rapidly changing environment. Besides, we look at cases that are characterized by the interaction of different agents engaged in relations of competition and collaboration. One such example deals with a quantitative evaluation of business models for collaborative provision of advanced mobile data services while another looks at the relations between network operators and service providers. Besides, we put our research focus in a wider perspective by presenting a survey of several promising modelling approaches of quantitative network analysis and formation. These include stochastic optimization, statistical mechanics of networks, network formation games and agent based computational economics. All of these research domains study complex interactive systems that are either explicit networks or can be described as such. The paper explores similarities, dissimilarities and linkages of these concepts including approaches on the border between these methodologies. Finally, implications for the research in telecommunications will be given.

I develop a two market agent-based model to study how global portfolio managers affect global financial crises and stability. The Markowitz model is extended by incorporating several insights from behavioral finance. Simulation results of an agent-based version of the Markowitz model reveal that global financial crises do not occur when global managers are added to the model. However, when risk is determined based on investors ’ historical losses and exponential averaging, slight global manager losses can trigger a widening of both markets ’ risk premium, accelerating the decline in asset prices worldwide. Statistical analysis reveals that global managers are a stabilizing force in smaller numbers; however, they become destabilizing in larger numbers. The ability to reduce risk by diversifying across markets results in excessive risk taking. If global managers exist in larger numbers, systematic over leverage may result such that a deleveraging process can lead to the spreading of financial crises. Other results indicate the existence of contagion and the importance of the real linkage versus the global manager linkage depends on the amount of total assets global managers control in each market. ∗I am grateful Dan Friedman for invaluable advice. I retain sole responsibility for remaining idiosyncrasies and errors.

We analyze the e¤ects of social learning in a widely-studied monetary policy context. Social learning might be viewed as more descriptive of actual learning behavior in complex market economies. Ideas about how best to forecast the economy’s state vector are initially heterogeneous. Agents can copy better forecasting techniques and discard those techniques which are less successful. We seek to understand whether the economy will converge to a rational expectations equilibrium under this more realistic learning dynamic. A key result from the literature in the version of the model we study is that the Taylor Principle governs both the uniqueness and the expectational stability of the rational expectations equilibrium when all agents learn homogeneously using recursive algorithms. We …nd that the Taylor Principle is not necessary for convergence to MSV equilibrium in a social learning context. We also study whether social learning can coordinate on sunspot equilibria which exist in the region of the parameter space where the Taylor Principle does not hold, MSV rational expectations equilibrium is not unique, and MSV solutions are expectationally unstable. We …nd that our agents cannot coordinate on a sunspot equilibrium in a general form speci…cation, however, they are able to coordinate on sunspot solutions in a common factor speci…cation. We also contribute to the use of genetic algorithm learning in stochastic environments.

This paper aims at reassessing some central issues of monetary policy by offering a model in which a central bank tries to stabilize fluctuations in aggregate output and inflation in an adaptive complex economy. We resort to evolutionary algorithms to model the central bank behaviour under discretion, and confront the efficiency of discretion with the choice of full commitment to a fixed rule.