The Role of Heterogeneous Agents’ Past and Forward Time Horizons in Formulating Computational Models

The conditioning of strategies by market environment and the simultaneous emergence of market structure in the presence of evolving trading strategies are investigated with major international stock indexes. Models for price forecasting and trading strategies evolution are examined under different time horizons. The results demonstrate that trading strategies can become performative in thin markets, thereby shaping the price dynamics, which in turn feeds back into the strategy. The dominance in thin markets by some (short-memory) traders produces a better environment for learning profitable strategies with computational intelligence tools.The experiment conducted contradicts assertions that long-term fitness of traders is not a function of an accurate prediction, but only of an appropriate risk aversion through a stable saving rate. The stock traders’ economic performance is found to be best with a 1-year forward time horizon, and it deteriorates significantly for tests with horizons exceeding 2 years, identifying frequent structural breaks. To model the turmoil in an economic system with recurrent shocks, short-memory horizons are optimal, as older data is not informative about current or future states.     

A multi‐agent system for analyzing the effect of information on prediction markets

Prediction markets have been shown to be a useful tool for forecasting the outcome of future events by aggregating public opinion about the event's outcome. In this paper, we investigate an important aspect of prediction markets—the effect of different information‐related parameters on the behavior of the traders in the market. We have developed a multi‐agent based system that incorporates different information‐related aspects including the arrival rate of information, the reliability of information, the penetration or accessibility of information among the different traders, and the perception or impact of information by the traders. We have performed extensive simulations of our agent‐based prediction market for analyzing the effect of information‐related parameters on the traders' behaviors expressed through their trading prices, and compared our agents' strategies with another agent‐based pricing strategy used in prediction markets called the zero intelligence strategy. Our results show that information‐related parameters have a significant impact on traders' beliefs about event outcomes, and, frequent, reliable information about events improves the utilities that the traders receive. Overall, our work provides a better understanding of the effect of information on the operation of prediction markets and on the strategies used by the traders in the market. © 2011 Wiley Periodicals, Inc.     

Bubble Formation and Heterogeneity of Traders: A Multi-Agent Perspective

Financial bubble is an intensively discussed but quite controversial topic. In current literature, the researches usually focus on the (ir)rationality of traders and its impacts on the bubble. We thereby propose a completely different perspective, that is, of traders’ heterogeneity and its impacts on the formation of bubble in financial markets. As in the real financial markets, the agents are always heterogenous. For example, some of them are fundamentalists, some are chartists, some are noise traders, etc. To model the heterogeneity of agents in the real markets, we proposed a multi-agent model to control the constitution of traders. Based on four scenarios with different constitution of traders’ behaviors, we investigated three extreme situations where the market is occupied by homogeneous agents (no matter they are fundamentalists, chartists or noise traders), and one scenario where the market is made up of heterogeneous traders. By applying Log-Periodic Power-Law (LPPL) model, We studied the impacts of different investors’ behaviors on the bubble formation in the market and found that: (a) the public information has an important influence on the beginning of a bubble; (b) traders’ different expectations and their self-feedback is one of reasons for the existence of log-periodicity in bubble; (c) the existence of power–law growth and log-periodicity, which leads the probability of prediction for the bursting of bubble, is caused by the combined effects of public information, traders’ different expectations and their self-feedback.     

Imperfect Evolutionary Systems

In this paper, we propose a change from a perfect paradigm to an imperfect paradigm in evolving intelligent systems. An imperfect evolutionary system (IES) is introduced as a new approach in an attempt to solve the problem of an intelligent system adapting to new challenges from its imperfect environment, with an emphasis on the incompleteness and continuity of intelligence. We define an IES as a system where intelligent individuals optimize their own utility, with the available resources, while adapting themselves to the new challenges from an evolving and imperfect environment. An individual and social learning paradigm (ISP) is presented as a general framework for developing IESs. A practical implementation of the ISP framework, an imperfect evolutionary market, is described. Through experimentation, we demonstrate the absorption of new information from an imperfect environment by artificial stock traders and the dissemination of new knowledge within an imperfect evolutionary market. Parameter sensitivity of the ISP framework is also studied by employing different levels of individual and social learning     

An equilibrium analysis of market selection strategies and fee strategies in competing double auction marketplaces

In this paper, we propose a game-theoretic framework for analysing competing double auction marketplaces that vie for traders and make profits by charging fees. Firstly, we analyse the equilibrium strategies for the traders’ market selection decision for given market fees using evolutionary game theory. Using this approach, we investigate how traders dynamically change their strategies, and thus, which equilibrium, if any, can be reached. In so doing, we show that, when the same type of fees are charged by two marketplaces, it is unlikely that competing marketplaces will continue to co-exist when traders converge to their equilibrium market selection strategies. Eventually, all the traders will congregate in one marketplace. However, when different types of fees are allowed (registration fees and profit fees), competing marketplaces are more likely to co-exist in equilibrium. We also find that sometimes all the traders eventually migrate to the marketplace that charges higher fees. We then further analyse this phenomenon, and specifically analyse how bidding strategies and random exploration of traders affects this migration respectively. Secondly, we analyse the equilibrium strategies of the marketplaces when they have the ability to vary their fees in response to changes in the traders’ market selection strategies. In this case, we consider the competition of the marketplaces as a two-stage game, where the traders’ market selection strategies are conditional on the market fees. In particular, we use a co-evolutionary approach to analyse how competing marketplaces dynamically set fees while taking into account the dynamics of the traders’ market selection strategies. In so doing, we find that two identical marketplaces undercut each other, and they will eventually charge the minimal fee as we set that guarantees positive market profits for them. Furthermore, we extend the co-evolutionary analysis of the marketplaces’ fee strategies to more general cases. Specifically, we analyse how an initially disadvantaged marketplace with an adaptive fee strategy can outperform an initially advantaged one with a fixed fee strategy, or even one with an adaptive fee strategy, and how competing marketplaces evolve their fee strategies when different types of fees are allowed.     

Online double auction mechanism for perishable goods

One-sided auctions are used for market clearing in the spot markets for perishable goods because production cost in spot markets is already “sunk.” Moreover, the promptness and simplicity of one-sided auctions are beneficial for trading in perishable goods. However, sellers cannot participate in the price-making process in these auctions. A standard double auction market collects bids from traders and matches the higher bids of buyers and lower bids of sellers to find the most efficient allocation, assuming that the value of unsold items remains unchanged. Nevertheless, in the market for perishable goods, sellers suffer a loss when they fail to sell their goods, because their salvage values are lost when the goods perish. To solve this problem, we investigate the suitable design of an online double auction for perishable goods, where bids arrive dynamically with their time limits. Our market mechanism aims at improving the profitability of traders by reducing trade failures in the face of uncertainty of incoming/departing bids. We develop a heuristic market mechanism with an allocation policy that prioritizes bids of traders based on their time-criticality, and evaluate its performance experimentally using multi-agent simulation. We find out that our market mechanism realizes efficient and fair allocations among traders with approximately truthful behavior in different market situations.     

Adaptive stock trading with dynamic asset allocation using reinforcement learning

Stock trading is an important decision-making problem that involves both stock selection and asset management. Though many promising results have been reported for predicting prices, selecting stocks, and managing assets using machine-learning techniques, considering all of them is challenging because of their complexity. In this paper, we present a new stock trading method that incorporates dynamic asset allocation in a reinforcement-learning framework. The proposed asset allocation strategy, called meta policy (MP), is designed to utilize the temporal information from both stock recommendations and the ratio of the stock fund over the asset. Local traders are constructed with pattern-based multiple predictors, and used to decide the purchase money per recommendation. Formulating the MP in the reinforcement learning framework is achieved by a compact design of the environment and the learning agent. Experimental results using the Korean stock market show that the proposed MP method outperforms other fixed asset-allocation strategies, and reduces the risks inherent in local traders.     

Dynamic Testing of Wholesale Power Market Designs: An Open-Source Agent-Based Framework

In April 2003 the U.S. Federal Energy Regulatory Commission proposed a complicated market design—the Wholesale Power Market Platform (WPMP)—for common adoption by all US wholesale power markets. Versions of the WPMP have been implemented in New England, New York, the mid-Atlantic states, the Midwest, the Southwest, and California. Strong opposition to the WPMP persists among some industry stakeholders, however, due largely to a perceived lack of adequate performance testing. This study reports on the model development and open-source implementation (in Java) of a computational wholesale power market organized in accordance with core WPMP features and operating over a realistically rendered transmission grid. The traders within this market model are strategic profit-seeking agents whose learning behaviors are based on data from human-subject experiments. Our key experimental focus is the complex interplay among structural conditions, market protocols, and learning behaviors in relation to short-term and longer-term market performance. Findings for a dynamic 5-node transmission grid test case are presented for concrete illustration. This article is an abridged version of Sun and Tesfatsion ST07a     

An expansion matching method to improve transaction effectiveness in the double auction market

The double auction is an important transaction mechanism in electronic commerce. Buyers and sellers can interact and be matched with each other in a double auction e-market. Consequently, enhancing the effectiveness of the double auction market to help traders successfully complete their transactions is an important issue. In this research study, Trading Agent Competition (TAC) data were collected to examine double auction market mechanisms. The TAC is a worldwide, renowned competition in which intelligent agents are employed to simulate business/market operations, and the TAC Market Design (CAT) tournament is an individual TAC competition that focuses on the double auction market. Thus, we conducted simulation experiments on the CAT competition platform, and the transaction data were analyzed to identify the impact of market design strategies on market performance, such as market share, market profit and transaction success rate. Based on these results, we developed an expansion matching method to enhance market performance, and we conducted verification experiments to evaluate our method. The results show that our expansion matching method promotes improved performance of market policies in the double auction market.     

Traders' Long-Run Wealth in an Artificial Financial Market

In this paper, we study the long-run wealth distribution of agents with different trading strategies in the framework of the Genoa Artificial Stock Market.The Genoa market is an agent-based simulated market able to reproduce the main stylised facts observed in financial markets, i.e., fat-tailed distribution of returns and volatility clustering. Various populations of traders have been introduced in a`thermal bath' made by random traders who make random buy and sell decisions constrained by the available limited resources and depending on past price volatility. We study both trend following and trend contrarian behaviour; fundamentalist traders (i.e., traders believing that stocks have a fundamental price depending on factors external to the market) are also investigated. Results show that the strategy alone does not allow forecasting which population will prevail. Trading strategies yield different results in different market conditions. Generally, in a closed market (a market with no money creation process), we find that trend followers lose relevance and money to other populations of traders and eventually disappear, whereas in an open market (a market with money inflows), trend followers can survive, but their strategy is less profitable than the strategy of other populations.     

A case study on using prediction markets as a rich environment for active learning

In this paper, prediction markets are presented as an innovative pedagogical tool which can be used to create a Rich Environment for Active Learning (REAL). Prediction markets are designed to make forecasts about specific future events by using a market mechanism to aggregate the information held by a large group of traders about that event into a single value. Prediction markets can be used to create decision scenarios which are linked to real-world events. The advantages of this approach in the cognitive and affective domains of learning are examined. The unique ability of prediction markets to enable active learning in large group teaching environments is explored. Building on this theoretical work, a detailed case study is presented describing how a prediction market can be deployed as a pedagogical tool in practice. Empirical evidence is presented exploring the effect prediction market participation has on learners in the cognitive domain.     

An Heterogeneous, Endogenous and Coevolutionary GP-Based Financial Market

Stock markets are very important in modern societies and their behavior has serious implications for a wide spectrum of the world's population. Investors, governing bodies, and society as a whole could benefit from better understanding of the behavior of stock markets. The traditional approach to analyzing such systems is the use of analytical models. However, the complexity of financial markets represents a big challenge to the analytical approach. Most analytical models make simplifying assumptions, such as perfect rationality and homogeneous investors, which threaten the validity of their results. This motivates alternative methods.In this paper, we report an artificial financial market and its use in studying the behavior of stock markets. This is an endogenous market, with which we model technical, fundamental, and noise traders. Nevertheless, our primary focus is on the technical traders, which are sophisticated genetic programming based agents that co- evolve (by learning based on their fitness function) by predicting investment opportunities in the market using technical analysis as the main tool. With this endogenous artificial market, we identify the conditions under which the statistical properties of price series in the artificial market resemble some of the properties of real financial markets. By performing a careful exploration of the most important aspects of our simulation model, we determine the way in which the factors of such a model affect the endogenously generated price. Additionally, we model the pressure to beat the market by a behavioral constraint imposed on the agents reflecting the Red Queen principle in evolution. We have demonstrated how evolutionary computation could play a key role in studying stock markets, mainly as a suitable model for economic learning on an agent- based simulation.     

Evolutionary Frequency and Forecasting Accuracy: Simulations Based on an Agent-Based Artificial Stock Market

Learning how to forecast is always important for traders, and divergent learning frequencies prevail among traders. The influence of the evolutionary frequency on learning performance has occasioned many studies of agent-based computational finance (e.g., Lettau in J Econ Dyn Control 21:1117–1147, 1997. doi: 10.1016/S0165-1889(97)00046-8; Szpiro in Complexity 2(4):31–39, 1997. doi: 10.1002/(SICI)1099-0526(199703/04)2:4<31::AID-CPLX8>3.0.CO;2-3; Cacho and Simmons in Aust J Agric Resour Econ 43(3):305–322, 1999. doi: 10.1111/1467-8489.00081). Although these studies all suggest that evolving less frequently and, hence, experiencing more realizations help learning, this implication may result from their common stationary assumption. Therefore, we first attempt to approach this issue in a ‘dynamically’ evolving market in which agents learn to forecast endogenously generated asset prices. Moreover, in these studies’ market settings, evolving less frequently also meant having a longer time horizon. However, it is not true in many market settings that are even closer to the real financial markets. The clarification that the evolutionary frequency and the time horizon are two separate notions leaves the effect of the evolutionary frequency on learning even more elusive and worthy of exploration independently. We find that the influence of a trader’s evolutionary frequency on his forecasting accuracy depends on all market participants and the resulting price dynamics. In addition, prior studies also commonly assume that traders have identical preferences, which is too strong an assumption to apply to a real market. Considering the heterogeneity of preferences, we find that converging to the rational expectations equilibrium is hardly possible, and we even suggest that agents in a slow-learning market learn frequently. We also apply a series of econometric tests to explain the simulation results.     

基于深度强化学习的股市操盘手模型研究

股票市场具有变化快、干扰因素多、周期数据不足等特点,股票交易是一种不完全信息下的博弈过程,单目标的监督学习模型很难处理这类序列化决策问题。强化学习是解决该类问题的有效途径之一。提出了基于深度强化学习的智能股市操盘手模型ISTG（Intelligent Stock Trader and Gym）,融合历史行情数据、技术指标、宏观经济指标等多数据类型,分析评判标准和优秀控制策略,加工长周期数据,实现可增量扩展不同类型数据的复盘模型,自动计算回报标签,训练智能操盘手,并提出直接利用行情数据计算单步确定性动作值的方法。采用中国股市1400多支的有10年以上数据的股票进行多种对比实验,ISTG的总体收益达到13%,优于买入持有总体−7%的表现。     

Gaming Prediction Markets: Equilibrium Strategies with a Market Maker

We study the equilibrium behavior of informed traders interacting with market scoring rule (MSR) market makers. One attractive feature of MSR is that it is myopically incentive compatible: it is optimal for traders to report their true beliefs about the likelihood of an event outcome provided that they ignore the impact of their reports on the profit they might garner from future trades. In this paper, we analyze non-myopic strategies and examine what information structures lead to truthful betting by traders. Specifically, we analyze the behavior of risk-neutral traders with incomplete information playing in a dynamic game. We consider finite-stage and infinite-stage game models. For each model, we study the logarithmic market scoring rule (LMSR) with two different information structures: conditionally independent signals and (unconditionally) independent signals. In the finite-stage model, when signals of traders are independent conditional on the state of the world, truthful betting is a Perfect Bayesian Equilibrium (PBE). Moreover, it is the unique Weak Perfect Bayesian Equilibrium (WPBE) of the game. In contrast, when signals of traders are unconditionally independent, truthful betting is not a WPBE. In the infinite-stage model with unconditionally independent signals, there does not exist an equilibrium in which all information is revealed in a finite amount of time. We propose a simple discounted market scoring rule that reduces the opportunity for bluffing strategies. We show that in any WPBE for the infinite-stage market with discounting, the market price converges to the fully-revealing price, and the rate of convergence can be bounded in terms of the discounting parameter. When signals are conditionally independent, truthful betting is the unique WPBE for the infinite-stage market with and without discounting.     

On the effects of competition between agent-based double auction markets

Real market institutions, stock and commodity exchanges for example, do not occur in isolation. The same stocks and commodities may be listed on multiple exchanges, and traders who want to deal in those goods have a choice of markets in which to trade. While there has been extensive research into agent-based trading in individual markets, there is little work on this kind of multiple market scenario. Our work seeks to address this imbalance in the context of double auction markets. This paper examines how standard economic measurements, like allocative efficiency, are affected by the presence of multiple markets for the same goods, especially when the markets are competing for traders. We find that while dividing traders between several small markets typically leads to lower efficiency and worse convergence than grouping them into one large market, competition between markets for traders, can reduce these losses.     

A hybrid evolutionary dynamic neural network for stock market trend analysis and prediction using unscented Kalman filter

Stock market prediction is of great interest to stock traders and investors due to high profit in trading the stocks. A successful stock buying/selling generally occurs near price trend turning point. Thus the prediction of stock market indices and its analysis are important to ascertain whether the next day's closing price would increase or decrease. This paper, therefore, presents a simple IIR filter based dynamic neural network (DNN) and an innovative optimized adaptive unscented Kalman filter for forecasting stock price indices of four different Indian stocks, namely the Bombay stock exchange (BSE), the IBM stock market, RIL stock market, and Oracle stock market. The weights of the dynamic neural information system are adjusted by four different learning strategies that include gradient calculation, unscented Kalman filter (UKF), differential evolution (DE), and a hybrid technique (DEUKF) by alternately executing the DE and UKF for a few generations. To improve the performance of both the UKF and DE algorithms, adaptation of certain parameters in both these algorithms has been presented in this paper. After predicting the stock price indices one day to one week ahead time horizon, the stock market trend has been analyzed using several important technical indicators like the moving average (MA), stochastic oscillators like K and D parameters, WMS%R (William indicator), etc. Extensive computer simulations are carried out with the four learning strategies for prediction of stock indices and the up or down trends of the indices. From the results it is observed that significant accuracy is achieved using the hybrid DEUKF algorithm in comparison to others that include only DE, UKF, and gradient descent technique in chronological order. Comparisons with some of the widely used neural networks (NNs) are also presented in the paper.     

A note on the relationship between market efficiency and adaptability – New evidence from artificial stock markets

We developed various artificial stock markets populated with different numbers of traders using a special adaptive form of the Strongly Typed Genetic Programming (STGP)-based learning algorithm. We then applied the STGP technique to historical data from three indices – the FTSE 100, S&P 500, and Russell 3000 – to investigate the formation of stock market dynamics and market efficiency. We used several econometric techniques to investigate the emergent properties of the stock markets. We have found that the introduction of increased heterogeneity and greater genetic diversity leads to higher market efficiency in terms of the Efficient Market Hypothesis (EMH), demonstrating that market efficiency does not necessarily correlate with rationality assumptions. We have also found that stock market dynamics and nonlinearity are better explained by the evolutionary process associated with the Adaptive Market Hypothesis (AMH), because different trader populations behave as an efficient adaptive system evolving over time. Hence, market efficiency exists simultaneously with the need for adaptive flexibility. Our empirical results, generated by a reduced number of boundedly rational traders in six of the stock markets, for each of the three financial instruments do not support the allocational efficiency of markets, indicating the possible need for governmental or regulatory intervention in stock markets in some circumstances.     

A comparison of neural networks and adaptive neuro-fuzzy inference systems for the prediction of water diffusion through carbon nanotubes

Given the fact that artificial intelligence tools such as neural network and fuzzy logic are capable of learning and inferencing from the past to capture the patterns that exist in the data, this study presents an intelligent method for the forecasting of water diffusion through carbon nanotubes where predictions are generated from neuro-fuzzy structures using molecular dynamics data. Therefore, this research was mainly focused on combining molecular dynamics with artificial intelligence methods in order to reduce the computational time of biomolecular and nanofluidic simulations. Two different artificial intelligence methods are applied for the time-dependent water diffusion forecasting: artificial neural network (ANN) and adaptive neuro-fuzzy inference systems (ANFISs). The effects of different sizes of training sample sets on forecasting performance of ANN and ANFIS are investigated as well. Four different evaluation methods are used to measure the performance and forecasting accuracy of these two methods. As a result, ANFIS presents the higher accuracy than neural network method based on the comparison of these different evaluation methods adopted in this research. The results reported in this research demonstrate that combining of molecular dynamics with artificial intelligence methods can be one of the most powerful and beneficial tools for prediction of important nanofluidic parameters.     

Coarse and fine identification of collusive clique in financial market

Collusive transactions refer to the activity whereby traders use carefully-designed trade to illegally manipulate the market. They do this by increasing specific trading volumes, thus creating a false impression that a market is more active than it actually is. The traders involved in the collusive transactions are termed as collusive clique. The collusive clique and its activities can cause substantial damage to the market's integrity and attract much attention of the regulators around the world in recent years. Much of the current research focused on the detection based on a number of assumptions of how a normal market behaves. There is, clearly, a lack of effective decision-support tools with which to identify potential collusive clique in a real-life setting. The study in this paper examined the structures of the traders in all transactions, and proposed two approaches to detect potential collusive clique with their activities. The first approach targeted on the overall collusive trend of the traders. This is particularly useful when regulators seek a general overview of how traders gather together for their transactions. The second approach accurately detected the parcel-passing style collusive transactions on the market through analysing the relations of the traders and transacted volumes. The proposed two approaches, on one hand, provided a complete cover for collusive transaction identifications, which can fulfil the different types of requirements of the regulation, i.e. MiFID II, on the other hand, showed a novel application of well-known computational algorithms on solving real and complex financial problem. The proposed two approaches are evaluated using real financial data drawn from the NYSE and CME group. Experimental results suggested that those approaches successfully identified all primary collusive clique scenarios in all selected datasets and thus showed the effectiveness and stableness of the novel application.