In this paper, we propose a real world application which is an automated foreign exchange trading system 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.

In this research, we propose a method of construction of trading rules based on evolutionary computation. In our proposed method, trading rules are expressed by binary trees(so-called strategy trees). They include both technical indicators and their parameters. We tried to optimize the strategy trees using Genetic Programming. Then, we simulated using real foreign exchange market data. As a result, it showed high win probability.

本論文では，国債の銘柄選択問題を金利とデフォルト確率に基づいてN 本腕バンディット問題としてとらえ，これを複利型強化学習を用いて投資戦略を学習する方法を提案する．提案手法を用いて2010 年第2 四半期の日米欧各国の国債を対象にした強化学習タスクを作成し，複利型Q 学習を用いて学習を行った．また，学習した行動価値に基づいてポートフォリオを構成し，モンテ・カルロ・シミュレーションによってパフォーマンスを評価した結果を示す．

In financial markets, sudden unexpected changes occur frequently. We propose a new forecast method based on paired evaluators consisting of the stable evaluator and the reactive evaluator that is good at detecting and adapting to the consecutive market changes. We conduct a back-testing using financial data in US. The experimental results show that our method is effective and robust even against late-2000s recessions.

In this study, we proposed a new text-mining method for stock price indexes using newspaper articles. Using this method, we conducted extrapolation tests to evaluate the prediction accuracy for the year 2009. As a result, 11 sectors in 19 sectors (57.8 percent) showed over 52% accuracy. The prediction accuracy showed seasonality in some sectors. This is expected to be a measure of prediction confidence of text mining.

In recent years, a number of researches are conducted to analyze and forecast stock values in the area of artificial intelligence. The most researches use monthly news and focus on the stock value fluctuation forecast on a monthly basis, but financial traders usually require the daily forecast of stock market prices. To forecast daily change of stock market, sector news should be taken into consideration because we think that they impact on stock market. The primary objective of our research is to investigate whether or not next day's stock values can be forecasted using text mining of the daily sector news. In our method, sentences in the sector news are resolved into the morphemes by morpheme analysis and the co-occurrence frequency are counted. Then, the derived frequency data are transformed to principal components. Finally, the relationships between stock value fluctuations and the principal components in the multiple linear regression are examined. To clarify the effectiveness of our method, we compare the forecating obtained by the daily sector news with those done by daily whole news. The experimental results shows that the accuracy of the forecasts with daily sector news is higher than that with daily whole news in the period of the big fluctuation in the sector.

In this study, we applied the newly developed text-mining methods to English texts for the long-term market analyses. We analyzed monthly price data of foreign financial markets, in particular, the interest swap markets. Several extensions of the original method were suggested in order to extract English feature vectors from minutes of the monetary policy committee of The Bank of England. Trends of interest rates were estimated by using the regression analysis with the feature vectors. As a result, determination coefficients were found around 75%, and market trends were explained well. Using the predicted interest rates, we also simulated several implementation tests, which demonstrate the effectiveness of our extensions of the original method to English texts.

株価急落による市場不安定化を抑止するために，金融当局により空売りが規制されることがある．これまでに空売り規制が市場に与える影響を分析した研究は多々行われてきたが，その結果は賛否両論である．そこで本研究では，現実の金融市場を計算機上で模倣した市場（人工市場）を用いて，空売り規制が株式市場安定化にどのような効果をもたらすかを検証した．はじめに，空売り規制期間と市場の安定性の関係について分析し，規制期間が長くなればなるほど，バブル発生回数が増加し，市場が不安定になることを示した．次に，リスク資産率と市場効率性，安定性の関係について議論した．その結果，リスク資産率が小さいとき，規制なし市場は安定的となるのに対し，規制あり市場は不安定となることを示した．一方，リスク資産率が大きいとき，規制なし市場では市場が非効率となるのに対し，規制あり市場では市場が効率的となることを示した．

This article develops a methodology to compare market performances under different trading mechanisms for equities implemented in TokyoStock Exchange(TSE): countinuous double auction and call market. Our results show that patience on time-to-execution of traders has a significant impact on market performances. Hence fast moving equities tend to shrink their liquidity under call market having a long intervel on executions.

レバレッジの与える影響をマルチエージェントモデルを用いて分析する。そのために、まず、市場への入退場を考慮した新しいマルチエージェントモデルを開発した。