• Selected
• Entire List

Abstract:
We construct a computer simulation of a repeated double-auction market, designed to match those in experimental-market settings with human subjects, to model complex interactions among artificially-intelligent traders endowed with varying degrees of learning capabilities. In the course of six different experimental designs, we investigate a number of features of our agent-based model: the price efficiency of the market, the speed at which prices converge to the rational expectations equilibrium price, the dynamics of the distribution of wealth among the different types of AI-agents, trading volume, bid/ask spreads, and other aspects of market dynamics. We are able to replicate several endings of human-based experimental markets, however, we also and intriguing differences between agent-based and human-based experiments.

Agent-Based Models, Artificial Markets, Experimental Markets, Market Microstructure

Abstract:
We propose a theory of asset pricing based on heterogeneous agents who continually adapt their expectations to the market that these expectations aggregatively create. And we explore the implications of this theory computationally using our Santa Fe artificial stock market. Asset markets, we argue, have a recursive nature in that agents' expectations are formed on the basis of their anticipations of other agents' expectations, which precludes expectations being formed by deductive means. Instead traders continually hypothesize-continually explore-expectational models, buy or sell on the basis of those that perform best, and confirm or discard these according to their performance. Thus individual beliefs or expectations become endogenous to the market, and constantly compete within an ecology of others' beliefs or expectations. The ecology of beliefs co-evolves over time. Computer experiments with this endogenous-expectations market explain one of the more striking puzzles in finance: that market traders often believe in such concepts as technical trading, market psychology, and bandwagon effects, while academic theorists believe in market efficiency and a lack of speculative opportunities. Both views, we show, are correct, but within different regimes. Within a regime where investors explore alternative expectational models at a low rate, the market settles into the rational-expectations equilibrium of the efficient-market literature. Within a regime where the rate of exploration of alternative expectations is higher, the market self-organizes into a complex pattern. It acquires a rich psychology, technical trading emerges, temporary bubbles and crashes occur, and asset prices and trading volume show statistical features-in particular, GARCH behavior-characteristic of actual market data.

Abstract:
There is reliable evidence that simple rules used by traders have some predictive value over the future movement of foreign exchange prices. This paper will review some of this evidence and discuss the economic magnitude of this predictability. The profitability of these trading rules will then be analyzed in connection with central bank activity using intervention data from the Federal Reserve. The objective is to find out to what extent foreign exchange predictability can be confined to periods of central bank activity in the foreign exchange market. The results indicate that after removing periods in which the Federal Reserve is active, exchange rate predictability is dramatically reduced.

Abstract:
Recently, research has shown that simple technical trading rules have predictive power in foreign exchange markets. One feature that sets these markets apart from others is that certain large traders, central banks, may not be optimizing trading profits. This paper tests the performance of a few simple rules during intervention and nonintervention periods. The unusual performance of the rules is large while interventions are taking place, and not significantly different from zero otherwise. This is consistent with this dimension of market inefficiency being connected to exchange rate behavior influenced by the central bank.

Abstract:
There is reliable evidence that simple rules used by traders have some predictive value over the future movement of foreign exchange prices. This paper will review some of this evidence and discuss the economic magnitude of this predictability. The profitability of these trading rules will then be analyzed in connection with central bank activity using intervention data from the Federal Reserve. The objective is to find out to what extent foreign exchange predictability can be confined to periods of central bank activity in the foreign exchange market. The results indicate that after removing periods in which the Federal Reserve is active, exchange rate predictability is dramatically reduced.

Abstract:
This article exposes problems of the commonly used technique of splitting the available data into training, validation, and test sets that are held fixed, warns about drawing too strong conclusions from such static splits, and shows potential pitfalls of ignoring variability across splits. Using a bootstrap or resampling method, we compare the uncertainty in the solution stemming from the data splitting with neural network specific uncertainties (parameter initialization, choice of number of hidden units, etc.). We present two results on data from the New York Stock Exchange. First, the variation due to different resamplings is significantly larger than the variation due to different network conditions. This result implies that it is important to not over-interpret a model (or an ensemble of models) estimated on one specific split of the data. Second, on each split, the neural network solution with early stopping is very close to a linear model; no significant nonlinearities are extracted.

Model evaluation, Model uncertainty, Bootstrap, Resampling, Financial forecasting, Time series prediction, Linear bias of early stopping, Superposition of forecasts, Model merging

Abstract:
This article exposes problems of the commonly used technique of splitting the available data into training, validation, and test sets that are held fixed, warns about drawing too strong conclusions from such static splits, and shows potential pitfalls of ignoring variability across splits. Using a bootstrap or resampling method, we compare the uncertainty in the solution stemming from the data splitting with neural network specific uncertainties (parameter initialization, choice of number of hidden units, etc.). We present two results on data from the New York Stock Exchange. First, the variation due to different resamplings is significantly larger than the variation due to different network conditions. This result implies that it is important to not over-interpret a model, or an ensemble of models, estimated on one specific split of the data. Second, on each split, the neural network solution with early stopping is very close to a linear model; no significant nonlinearities are extracted.

Abstract:
Extreme Value Theory (EVT) offers a powerful framework to characterize financial market crashes and booms. This paper applies EVT to model the behavior of extreme events and compares tail thickness between emerging and developed market equity return distributions. We extend previous results by augmenting parametric Monte Carlo tests with nonparametric bootstrap tests. We construct Monte Carlo and Bootstrapping experiments to estimate the statistical significance of differences in tail behavior between markets and regions. Within each market we find little evidence for asymmetry between positive and negative tails. We find mixed evidence for uniformity inside each region, and strong evidence for differences in tail behavior between emerging and developed regions. Our regional results have important implications for the expected diversification benefits of international portfolio allocation decisions.

Power-laws, risk management, bootstrap

Abstract:
Recent research has shown the importance of time horizons in models of learning in finance. The dynamics of how agents adjust to believe that the world around them is stationary may be just as crucial in the convergence to a rational expectations equilibrium as getting parameters and model specifications correct in the learning process. This paper explores the process of this evolution in learning and time horizons in a simple agent based financial market. The results indicate that while the simple model structure used here replicates usual rational expectations results with long horizon agents, the route to evolving a population of both long and short horizon agents to long horizons alone may be difficult. Furthermore, populations with both short and long horizon agents increase return variability, and leave patterns in volatility and trading volume similar to actual financial markets.

Abstract:
We find a large positive correlation between daily trading volume in currency futures markets and foreign exchange intervention by the Federal Reserve over the period 1979-1996. Neither contemporaneous nor predicted volatility can fully account for the increases in trading activity. Whether or not the intervention operation is publicly reported appears to be an important determinant of trading volume.

Abstract:
There has been renewed interest in power-laws and various types of self-similarity in many financial time series. Most of these tests are visual in nature, and do not consider a wide range of possible candidate stochastic models capable of generating the observed results. This paper presents a relatively simple stochastic volatility model which is able to display power laws and scale invariance similar to actual financial data even though it is constructed to have none of these properties. The primary mechanism is that volatility is assumed to have a driving process with a half life that is long relative to the tested aggregation ranges. It is argued that this might be a reasonable feature for financial, and other macroeconomic time series.

Volatility, Power-laws, Long Memory

Abstract:
This paper introduces an order-driven market with heterogeneous investors, who submit limit or market orders according to their own trading rules. The trading rules are repeatedly updated via simple learning and adaptation of the investors. We analyze markets with and without learning and adaptation. The simulation results show that our model with learning and adaptation successfully replicates long-memories in trading volume, stock return volatility, and signs of market orders. We also discuss why evolutionary dynamics are important in generating these long memory features.

Microstructure, agent-based, long memory, order flow

Abstract:
This paper seeks to develop a structural model that lets data on asset returns and trading volume speak to whether volatility autocorrelation comes from the fundamental that the trading process is pricing or, is caused by the trading process itself. Returns and volume data argue, in the context of our model, that persistent volatility is caused by traders experimenting with different beliefs based upon past profit experience and their estimates of future profit experience. A major theme of our paper is to introduce adaptive agents in the spirit of Sargent (1993) but have them adapt their strategies on a time scale that is slower than the time scale on which the trading process takes place. This will lead to positive autocorrelation in volatility and volume on the time scale of the trading process which generates returns and volume data. Positive autocorrelation of volatility and volume is caused by persistence of strategy patterns that are associated with high volatility and high volume. Thee following features seen in the data: (i) The autocorrelation function of a measure of volatility such as squared returns or absolute value of returns is positive with a slowly decaying tail. (ii) The autocorrelation function of a measure of trading activity such as volume or turnover is positive with a slowly decaying tail. (iii) The cross correlation function of a measure of volatility such as squared returns is about zero for squared returns with past and future volumes and is positive for squared returns with current volumes. (iv) Abrupt changes in prices and returns occur which are hard to attach to 'news.' The last feature is obtained by a version of the model where the Law of Large Numbers fails in the large economy limit.

Abstract:
Evolutionary metaphors have been prominent in both economics and finance. They are often used as basic foundations for rational behavior and efficient markets. Theoretically, a mechanism which selects for rational investors actually requires many caveats, and is far from generic. This paper tests wealth based evolution in a simple, stylized agent-based financial market. The setup borrows extensively from current research in finance that considers optimal behavior with some amount of return predictability. The results confirm that with a homogeneous world of log utility investors wealth will converge onto optimal adaptive forecasting parameters. However, in the case of utility functions which differ from log, wealth selection alone converges to parameters which are economically far from the optimal forecast parameters. This serves as a strong reminder that wealth selection and utility maximization are not the same thing. Therefore, suboptimal financial forecasting strategies may be difficult to drive out of a market, and may even do quite well for some time.

financial time series, evolution, forecasting

Abstract:
Agent-based economics, and more generally agent-based social sciences, have been around in various forms for over 30 years. The advent of higher speed computing and new tools for the computational learning fields led to a major increase in activity in the early 1990s through today. Research activity continues to increase at the current time, but the field still remains somewhat of a "niche field" inside economics. Certain conferences and certain regions (such as Europe) are well populated with agent-based activity. However, at mainstream conferences inside the US one would have a hard time in finding agent-based researchers.

Abstract:
No abstract is available for this paper.

Abstract:
There has been renewed interest in power laws and various types of self-similarity in many financial time series. Most of these tests are visual in nature, and do not consider a wide range of possible candidate stochastic models capable of generating the observed results in small samples. This paper presents a relatively simple stochastic volatility model, which is able to produce visual power laws and long memory similar to those from actual return series suing comparable samples sizes. These are small-sample features for the stochastic volatility model, since asymptotically it possesses none of these properties. The primary mechanism for this result is that volatility is assumed to have a driving process with a half life that is long relative to the tested aggregation ranges. It is argued that this might be a reasonable feature for financial, and other macroeconomic time series.

Abstract:
This paper is intended to guide researchers interested in building their own agent-based financial markets. Key design questions are outlined, along with some of the major controversies about which directions to take.

Abstract:
This paper reafirms previous results that show a distinct change in performance of moving average trading rules on the Dow Jones Index over the past 10 years relative to the previous 90 years. The performance in forecasting conditional means has changed significantly. In contrast to this is the fact that technical rules continue to forecast conditional variances in a similar fashion to the previous 90 years. Several tests are performed to demonstrate the robustness of these results over alternative trading mechnasims. Finally, the results are discussed in the context of datasnooping versus structural changes in the Dow.

Abstract:
Computational models for financial markets with many interacting agents have recently appeared as a tool for examining learning and evolutionary issues in market dynamics. This paper surveys some of the early research in this area with emphasis on the many unsolved problems that researchers will need to confront. Several early papers are emphasized which focus on some of these problems, and references to many other papers are also given.

Abstract:
A market of artificially intelligent traders is constructed to buy and sell a risky asset along with a risk free bond. Prices of the risky asset are determined endogenously from the interactions of the strategies which make trades and gather data. Each trader tries to learn about the world around it while enhancing its trading strategies. The primary purpose of this paper is to demonstrate that such a market replicates some of the basic empirical features of many asset markets including the persistence of volatility and trading volume, weak trends in prices, and leptokurtosis in returns. Also, for certain parameter values agents converge to a well defined rational expectations equilibrium.

Abstract:
Recent evidence has shown possible scaling and self-similarity in high frequency financial time series. This paper demonstrates that many of these graphical scaling results could have been generated by a simple stochastic volatility model. This casts doubt on the power of these tests to discern between true scaling and simple highly dependent stochastic processes.