We study unconditional distribution derived from the Fokker-Planck equation for an artificial market. The artificial market consists of two kind of participants; fundamentalists and chartist(noise traders). We compute parameter estimates for absolute-log-return time series of exchange rates by means of the maximum likelihood method. We compare the model-based tail index with Hill's estimators.

We propose a method to assess the risk of ?nancial time series with an unconditional distribution estimated from them. Because it is not easy to infer its tail shape due to a lack of data in a practical manner, we adapt a parametric method with a q-Gaussian distribution. We introduce Value-at-Risk (VaR) to measure risk and compare it with variance under the q-Gaussian assumption. We examine performance of the maximum likelihood estimator with the q-Gaussian log-likelihood function. By using the distribution estimates, we compute the errors, de?ned as the di?erence between estimation and the real value. Finally,we conduct an empirical analysis on log-returns of a stock traded in the Tokyo Stock Exchange by using the proposed method.

We introduce minimal agent based model of foreign exchange markets from the view point of econophysics. In order to reproduce major statistical properties of real market, we start from the simplest model and we add two important feedback effects to this simplest model; one is feedback effect of price change and the other is feedback effect of transaction intervals. As an application of this realistic model, we simulate the case of government intervention on exchange rate and discuss a relationship between a dealers' trend-follow effect and efficiency of the intervention.

In arti?cial market studies, agents must do the similar behavior with real market participants. In this paper, we estimate the composition of the real market participants using arti?cial markets. We use an inverse simulation to optimize agents who participate in arti?cial markets. We use stock price time series of the market to evaluate arti?cial markets. In the experiment for verifying the validity of the proposed method, we applies the proposed method to markets in which di?erent participants exist. The result indicates that the proposed method is useful to estimate market participants.

In this study, we propose a new paired evaluators method with consideration of turn over cost for sudden unexpected changes in financial markets. This consideration is necessary to evaluate the usability of forecasting models in realistic portfolio management. We conduct empirical analysis using Japanese stock market data from Jan.2001 to Sep.2010 to test how our proposing method switches the long term portfolio and short term portfolio in efficient ways. The results of the empirical analysis show that our method achieves higher return and reduces risks compared to the cases of fixed portfolios, either long-term portfolio or short-term portfolio.

In this paper, we propose a real world application which is an automated foreign exchange trading system which update the trading rule automatically, using Differential Evolution( DE). There are many studies which have focused on trading system. In recent years, such studies have attracted attention because system can catch movement of market price accurately and quickly. DE is a simple yet powerful evolutionary algorithm for global numerical optimization. To verify the effectiveness of the method, we performed simulations using real historical trading data. DE was found to be superior compared to other previous methods in terms of precision and reliability.

We report our on-going research to develop a system to predict volume of transactions of a brand using news articles related to the brand. We investigated the e?ect of keyword-based article selection and online learning algorithms on prediction accuracies.

Through intensive analysis of high frequency market data, empirical laws of the foreign. exchange markets have been discovered. For examples, the power law distributions of the absolute value of price changes, abnormal diffusion of prices in the short time scale and a normal diffusion in the large time scale. The dealer model is a model that directly simulates the dealer's behavior and satisfies these empirical laws. In this study, we introduce the new quantity of position to the basic dealer model defined by the ratio of yen property and dollar property. If a dealer 's position is leaned to one side he tries to balance the position to avoid potential risks even the action is less profitable. The self-interest pursuit is the effect that the dealer acts so that his position may work advantageously by changing his price. As a result we obtain a phase diagram categorizing the difference of market price dynamics. With this revised model we can simulate the effects of loss-cut and loss-limit which are actually applied to real dealers in financial institutes.

In this paper, we propose a new simple artificial market model based on an idea of the U-Mart system. We designed a reinforcement learning agent which runs on this artificial market with other simple computer agents. We simulated the process that the reinforcement learning agent adapts to the price series of the market. This simulation clearly showed the emergence of market.