In this study, we propose the use of exchange value as the stable numeraire for pricing financial assets. The price of currency is used to illustrate our method. In the currency market, the price of a currency is usually measured by using its exchange rate to another currency, for example, the dollar-yen rate. We point out that the price of currency can also be measured by using the proposed numeraire, namely the exchange value. In particular, we calculated the exchange values of different currencies from their exchange rates under a statistical approximation. Finally, probability distributions of differently measured price returns are compared with each other to evaluate the stability with the use of the proposed numeaire.

In recent years, portfolio optimization with the use of a vector autoregressive (VAR) model to estimate future stock returns has been subject of research. In this paper, we propose an optimization model that uses time series models to predict not only the future returns but also the conditional variance-covariance matrix of returns. More speci?cally, the future returns are predicted by using the VAR model, and the conditional variance-covariance matrix is estimated by using a dynamic conditional correlation (DCC) multivariate GARCH model. We evaluate the out-of-sample investment performance of our model using historical data of U.S. stock market.

本論文では，新聞記事で報道される情報をもとに，深層学習による株価動向推定の手法を提案する．一日の新聞記事中には，様々な銘柄に直接的・間接的に関係するニュース記事が混在しており，それらに出現する語彙を全て同等に扱うのは適切でない．そこで，記事ごとに作成した特徴ベクトルを深層学習によって統合することで，一日の出来事に関する圧縮表現を獲得する手法を提案する．さらに，このように獲得した特徴ベクトルを用いて，数種の銘柄の株価動向予測を行った結果について報告する．

We detect turning points of the non-stationary time series data of Nikkei 225 index for the period between 1993 and 2010 using 'change finder'. We also calculate the market sentiment using news data prior to the turning point. Our findings are in two-fold. Firstly, the 'change finder' signals the bullish turning points following the rise of the optimistic sentiment and vice versa. Secondly, bullish change occurs significantly more in the first half of the year than the latter half. Our findings are consistent with the view that the reported 'Dekansho-bushi' effect in the Japanese Stock Market is driven by the market psychology.

We put forward an optimized method of stock factors prediction model which can be easily extended to realtime prediction system., based on the new version of Yahoo Financial text board of 2012.11~2013.6 with about 4000 companies. Preceding studies have verified that BBS text can be used to forecast trade volume and return. On this basis, LSA (Latent Semantic Analysis) and multi-SVM model are put forward in our framework to improve the accuracy of natural language processing and the prediction.

Recent research has explored the proper method to analyse the relationships in financial markets for risk management, In this paper, we apply transfer entropy to construct a stationary network which represents the information propagation between stocks. This network can differ significantly from other static networks, such as correlations network and minimal spanning tree network, because it can include the direction information. We demonstrate that this method reveals meaningful hidden relations of cause and effect between stocks.

It had been believed that the risk of a bank going bankrupt is lessened in a straightfor-ward manner by transferring the risk of loan defaults. But the failure of American International Group in 2008 posed a more complex aspect of nancial contagion. This study presents an ex-tension of the asset network systemic risk model (ANWSER) to investigate whether credit default swaps mitigate or intensify the severity of nancial contagion. The empirical distribution of the number of bank bankruptcies is obtained with the extended model. Systemic capital bu er ratio is calculated from the distribution. The ratio quanti es the e ective loss absorbency capability of the entire nancial system to force back nancial contagion. The key nding is that the the risk transfer in an interbank network does not mitigate the severity of nancial contagion.

In this study, we propose a new forecasting method of a financial time series based on Deep Belief Network (DBN) by enhancing the approach of the Chao et al. First, a new topology for a regression training is proposed. Second, we forecast a Nikkei Stock Average renewing a training term. Third, Self-Organized-Map (SOM) is introduced for reducing the computational time in DBN. It is shown by some experiments that some improved performance indexes can be obtained, and reduction of the computation time is achieved.

We propose a method to predict stock prices by SVMs using news on foreign exchange rates on the Web. Our method targets Japanese news and stocks. We compare several parameters for predicting the span, and fixed span to 50 minutes. We then apply the proposed method to 15 different stock issues from Nikkei 225. Although our preliminary results are encouraging, we plan to further improve the accuracy of our approach in future.

To clarify connections between social media data and ?nancial market data, we studied the quantitative evaluations of relations between time series of word appearances on Japanese blogs and those of stock prices. In particular, we proposed the method of comparison of some correlation indices such as the Spearman's rank correlation coe?cient, based on the man-made related stock information. We found that the Spearman's rank correlation coe?cients over time series of 562 keywords can hardly pick the correct combinations of related stocks out the pool of more than 3,000 stocks on the Tokyo Stock Exchange. However, we show that the composite correlation indicators, which re?ect multiple features of the time series, can pick the correct stocks up to a certain level of statically signi?cant.

In this paper, we analyze how the relationship among Nikkei average futures and individual stocks changes when a big event occurs, intending to give investors useful information for the risk management. We showed that the strong relationships had appeared after the Great East Japan Earthquake and we could detect them by using order books properly.

We introduce analysis of arbitrage opportunities in the foreign exchange market by using high-frequency data. We showed two kinds of arbitrage opportunities, negative spread arbitrages and triangle arbitrages, and we modeled the occurrence of the arbitrage with volatility, the number of deals and the number of computer traders. The market has changed over the last ten years. In particular an emergence of computer traders, which have trading algorithms in computers, is one of the biggest news in financial markets, and the computer traders can detect triangular arbitrages much faster than human traders. We also modeled the disappearance probability of triangular arbitrages within one second that is the minimum observation interval of our data by using volatility, the number of deals and the number of computer traders.

本論文では，株を実際に取引できるTradeStation において，複利型強化学習を用いて取引戦略を獲得する方法について述べる．従来のカブロボでは1 日に3 回プログラムが実行されるだけだったが，TradeStation では分足の価格情報が更新されるごとにプログラムが実行されるため，デイトレードを行うことができる．そこで，本論文では，これまで日次取引を対象に開発してきた手法をデイトレードに適用する方法を提案する．また，SPDR S&P 500 ETF Trust を対象とした実験結果を示す．

株式市場において，注文を公開せずに注文を付き合わせる，ダーク・プールという取引市場が普及してきている．ダー ク・プールは市場の安定化につながると言われている一方，市場の価格発見機能が低下し，市場が不安定になる恐れが あるという批判もある．本研究では１つのリット市場（注文情報が公開されている通常の市場）と１つのダーク・プー ルが存在する人工市場モデル（マルチ・エージェント・シミュレーション）を構築し，ダーク・プールの普及が市場を安 定化させるのかどうかシミュレーションを行い分析をした．その結果，ダーク・プールの普及によりボラティリティが 低下し市場が安定化することが分かった．また，アルゴリズム・トレードでは，ダーク・プールへの発注を増やすほど マーケット・インパクトをおさえることが分かった．しかし，他の投資家がダーク・プールをとても多く使用している状 況だと，アルゴリズム・トレードはそれ以上にダーク・プールを使用しないとマーケット・インパクトを押さえられない ことが分かった．さらに，リット市場のティックサイズが大きいときはよりダーク・プールの有用性が高い可能性を示せ た．普及しすぎた場合にはさまざまな悪影響が示唆されたが，その悪影響がではじめる普及率は実際の金融市場での現 在のダーク・プールの普及率よりかなり高い可能性があることが示唆された．