This study presents an ANSeR model (asset network systemic risk model) to compute a bankruptcy amplification ratio in analyzing the risk of bank failures in a financial crisis. With the model we compare the instability of financial systems in Japan, the United States, and the United Kingdom.

There are substantial discussions on the pros and cons, from regulators, market participants and academics, regarding the introduction of regulated short selling. In this research, we focus on two of the four regulatory measures on short- selling in Japan: "uptick rule requirement" and "naked short-selling ban." We analyze the effectiveness of each measure by comparing the simulations for markets with and without the regulation. The unique feature of our model is the stock lending and borrowing transactions between traders. The lending/borrowing fee is modeled to influence the investment decisions of the traders. We calculate the price fluctuation, divergence of the price from the fundamental value and price volatility for each simulated market. We evaluate robustness of the markets by giving an exogenous shock to the markets causing drastic declines in the fundamental price. As a result, we find that both regulatory measures make the market more stable and robust against abrupt shocks to the market.

バブル崩壊時や金融危機時に重要である学習プロセスを実装した人工市場を用いてシミュレーションを行い，値幅制限制度と完全空売り規制，およびアップティックルールの効果を比較した．その結果，規制がない場合にバブル崩壊がおこるとファンダメンタル価格よりもさらに価格が下落するというオーバーシュートが発生することが分かった．一方，規制がある場合はオーバーシュートが発生せず効率的な市場となることが分かった．しかし，完全空売り規制とアップティックルールは平常時に，割高な価格でしか取引されないという副作用をもっていることが分かった．これらを総合すると，値幅制限制度が平常時の副作用も無く，もっとも効率的な市場をもたらす可能性があることを示した．

In recent years, several stylized facts have been uncovered in econophysics. Here, we perform an extensive analysis of forex data that leads to unveil a statistical ?nancial law. First, our ?ndings show that, in average, volatility increases more when the price exceeds the highest (lowest) value (i.e. breaks resistance line). We call it (breaking-acceleration e?ect). Secondly, our results show that the probability P(T) to break the resistance line in the past T time follows power-law in both real data and theoretically simulated data. However, the probability calculated using real data is rather lower than the one obtained using a traditional Black-Scholes (BS) model. Taken together, the present analysis characterizes a new stylized fact of ?nancial markets and shows that the market exceeds a past (historical) extreme price fewer times than expected by the BS model (resistance e?ect). However, when the market does it, we predict that the average volatility at that time point will be much higher. These ?ndings indicate that any markovian model does not faithfully capture the market dynamics.

The purpose of this research is to discuss unexpected fluctuations such as "unintended consequences" in markets. "Unintended consequences" are illustrated as phenomena based on mutual-interaction between the macro and micro level, known as macro-micro linkage. In general, macro and micro views are analyzed separately in social science. I am interested in the results of feedback loops in social phenomena. I am focusing on "emergence" especially transilient types from mutual-interaction between traders (agent) and markets (system) through actions.

本論文では，複利型強化学習において投資比率を最適化する方法について検討する．これまでに提案した方法ではオンライン勾配法を用いて投資比率を最適化するが，強化学習によって有効な行動規則を学習できない場合には投資比率が0 に収束してしまう場合がある．そこで，本論文では，複利型共学習における投資比率の最適化方法について検討し，新たな方法を提案する．

We introduce a method of extracting causal information (e.g., Demand for semi conductor manufacturing equipments is good ) from Japanese financial articles concerning business performance of companies. Causal information is useful for investors in selecting companies to invest. Our method automatically extracts causal information as a form of causal expression by using statistical information and initial clue expressions. We extract keywords from Web sites of companies and improve the previous method by using them. Moreover, our method automatically determines the most important causal expression by using these keywords extracted from Web sites of companies.

ツイッター情報からテキスト特徴量とグラフ特徴量を抽出し、得られた情報と業種別平均株価との関連性を明確にすることで、経済動向の分析に有用な情報を得る。

In this study, we propose a method that acquires a topic dictionary from financial news articles. The topic dictionary is the knowledge related to the stock brands, and is composed of topics (e.g. influenza), groups of stock brands (e.g. pharmaceutical companies and / or spinning companies), and the strength of relations between them. The strength of a relation is updated by referring to the number of news headlines of the topic and the volume of transaction of the stock, and is used in order to strengthen the relation by calculating the weighted sum of news headlines. This study shows the correlation coefficient between the weighted sum of transactions and the volume is higher than the one between the non-weighted sum and the volume of transactions. The topic dictionary is expected to help catching the influence which newest topics give to the volume of transactions.