近年，取引市場同士の競争や投資家の要望などにより，取引市場のシステムの高速化が進んだ．取引システムの高速化により，流動性を供給して利益を得るマーケットメーカー戦略の注文量が以前より増え流動性が向上したという評価がある一方，投資家同士の取引のスピード競争を招き，その競争のために費やされたシステムコストは他の投資家に転化されるという批判がある．高速な取引の優位性を無効にする取引方式としてバッチオークション方式が提案されているが，流動性を供給するマーケットメーカー戦略は損益のリスクが高くなるため継続することが難しくなり，流動性の供給が減ってしまい，むしろ投資家の取引コストが上昇する可能性があるという批判もある．本研究では，ザラバ方式とバッチオークション方式を比較可能な人工市場シミュレーションを用いて，マーケットメーカー戦略の損益のリスクを分析することによりその存続可能性を議論し，マーケットメーカー戦略がバッチオークション方式においても流動性を供給し続けることができるかどうかを調べた．その結果，板寄せ間隔が大きくなると，取引成立率が減少し流動性供給が減少する可能性が示された．さらに，板寄せ間隔が大きくなると，オーバーナイトのポジションをゼロにすることが難しい上にポジションも大きく価格変動リスクが大きくなり，リスクをおさえてマーケットメーカー戦略を継続することが困難になる可能性が示された．さらに，ザラバ方式のときのみ，リスクに見合った収益を得る可能性がある可能性が示され，バッチオークション方式のときは，少なくともザラバ方式では機能したマーケットメーカー戦略では，リスクに見合った収益をあげるのは難しくなる可能性が示された．これらの示唆は，ザラバ方式では流動性を供給していたマーケットメーカー戦略が，バッチオークション方式になるとその供給を維持できなくなり，それらが撤退することにより，流動性が低下する可能性を暗示しているとも考えることができる．

Financial markets occasionally become highly volatile, as a result of a financial crisis or other factors. Previously, index futures trading and program trading have been singled out as direct causes of market destabilization, but more recently it has been suggested that leveraged ETFs (funds aimed at amplifying several-fold the movement of a price index such as the Nikkei Stock Average or underlying assets) rebalancing trades may also be a factor. This study uses a financial market simulation (artificial market) constructed virtually on a computer to assess the impact of leveraged ETF rebalancing trades on the underlying assets market. Analysis results showed that a larger amount of the managed assets of leveraged ETFs corresponds to a higher volatility of the underlying securities market. They also demonstrated that leveraged ETF trading can destroy the underlying assets market, if the leveraged ETF trading impact on the market is greater than that of ordinary volatility of the underlying assets.

投資信託の取引において，投資家が想定外のリスクを負うのを未然に防ぐため，分散投資規制が設けられた．分散投資規制は，1 つの発行体への投資が過度に集中しないよう，1 つの発行体への投資額を投資信託純資産の一定割合以下にするものである．これまでに，2 資産市場の人工市場において分散投資規制が資産価格急落時の市場に与える影響について検証されており，資産価格が急落したとき全ての資産の保有額が純資産に対して非常に少なくなってしまうことが確認されている．一方，実際の投資信託運用者は，資産価格の変動などによって予め設定している保有額の下限を下回らないよう，保有額の少ない資産を買い増ししたりして純資産に対する全資産保有額を調整している．そこで本研究では，2 資産市場の人工市場にまだ分散投資していない資産（資産 3）を追加し，ある資産の価格が急落したときに分散投資規制が市場にどのような影響を与えるかを，純資産に対する全資産保有額の下限がある場合とない場合に分けて調査した．その結果，資産 1 のファンダメンタル価格が急落し，分散投資規制が導入されている市場において，純資産に対する資産の保有額に下限を設けた場合は，資産 1 と資産 2 の市場価格は下落する一方で，初期状態では分散投資されていなかった資産 3 が購入され，市場価格が上昇することが確認できた．

In recent years, many researchers pay attention to behavioral finance. Behavioral finance focuses on investor's irrationality. In the field of behavioral finance, researchers are trying to elucidate cause of investor's decision making by observation. Some researchers constructed mathematical model of investor's decision making considering investor's psychological factor using an artificial market. However, these researches didn't fully express the psychological factors and real investor's decision making. Therefore, we construct investor's decision making model depending on investor's loss and profit using an artificial market. And we tested efficiency of our model.

In the past few decades, researchers have realized that human psychology can affect traders in decision-marking process and finally affect the financial market based on behavior finance theory and cognitive psychology. Group behavior bias is one of them. Group. Some studies have been done on group behavior bias from behavior finance viewpoint to tell the differences of group behavior bias. However they only qualitatively analyzed macro phenomenon without quantitatively measure the detail micro thinking process in individuals. In this paper, we proposed and validated three types of group behavior bias models including majority following, winner following and hub following models based on different nationalities. We also compared these models and figured out that the majority following bias is the easiest to form in the market, however the market impact is the least. On the other hand, hub following bias is the hardest to emerge but the market impact is the most. Besides, we introduced short selling regulation as well as multi rate regulation and find both of them can lead heavier market impact in the market with group behavior biases.

In this paper, we propose a method of assigning polarity to causal information extracted from summary of financial statements of companies. Our method assigns polarity (positive or negative) to causal information in accordance with business performance, e.g. "Orders of semiconductor manufacturing equipments were strong". First, we assigns polarity to extended clue expressions to be used to extract causal information. Using them, our method automatically generates training data and assigns polarity to causal information by deep learning. We evaluated our method and confirmed that it attained 86.7% precision and 95.4% recall of assigning polarity positive, and 90.0% precision and 73.9% recall of assigning polarity negative, respectively.

現状，単語単位での極性辞書自動構築に関する既存研究はあるものの，似たような意味の単語がひとまとまりになった形での極性辞書，極性概念辞書の構築に関する研究はほとんどされていない．本研究では，極性概念辞書の構築手法を提案すると共にその有用性を検証した．まず，ロイターニュース記事・ヤフーファイナンス掲示板のテキスト情報，専門家によって与えられた約 100 単語の極性値，及び構造化されたニューラルネットワークモデルを用いて意味的にクラス分けされた極性概念辞書を作成した．その後，極性概念辞書を用いて市場動向の分析を試みた．

Calculating similarity score for monolingual text is a popular task since it could be used for various text mining system. However seldom research is focusing on multilingual text resources. On the other hand, machine learning based algorithms such as CBOW word embedding and clustering are widely used in extracting features of text. In this research, we develop and train a model that could calculate the similarity of the two finance news reports, by utilizing CBOW, spherical clustering, bi-graph extraction as well as the Siamese-LSTM deep learning model. In the end, we train the model by feeding news data that is closely related in the financial domain to help us to analyze the relationship among news reports written in different languages.

This paper reports on our ongoing work to construct sentiment lexicons in the financial domain. Our approach takes advantages of news headlines and a given financial variable, such as stock prices, so as to generate initial sentiment lexicons. The initial lexicons are then filtered based on their co-occurrences with financial seed words and are subsequently expanded by analogical reasoning by using distributed representation of words. Evaluative experiments on around 12 years' worth of news data show that the resulting lexicons are mostly reasonable. As a possible application of the lexicons, trading simulation is also carried out, showing promising results.

In this research, we aim to predict start pages of proposals stated in notice of the meeting of shareholders and classify which proposal the page is. We propose two methods that classification method of proposals. The first method heuristically predicts the page on which the proposal is described. Moreover our method extracts specialized terms of each proposal and assigns weights to them. After that, our method classifies proposals by specialized terms. The second method classifies proposals using deep learning. Each methods were evaluated, and the effectiveness of each methods was verified.

It is necessary to build a comprehensive polarity dictionary specialized for financial policy to improve the accuracy of lexicon-based sentiment analysis in evaluating texts written on financial policy. In this research, we acquire distributed representation of words using feature learning of dependency network of words and create the polarity dictionary by bootstrap method using the distributed representation of words.

Central bank's monetary policy is one of the major interests for market participants. In this paper, we clarified Reserve Bank of Australia's monetary policy reaction function, predicted its policy change,and applied them to investment strategy. First of all, assuming perfect foresight by the central bank, we estimated an extended Taylor rule using bidirectional Recurrent Neural Network. Next, we combined it with distributed representation of Monetary Policy Committee minutes to develop a classifier of interest rate decision. Both the extended Taylor rule and the classifier showed improved performance. Finally, we formulated profitable Foreign Exchange strategy based on the classifier's prediction and market economists' forecasts.

金融市場を取り巻く情報は日々増加している。物価指数や産業活動指数など数値データもあれば、企業の有価証券報告書、ロイターなどの配信している経済ニュースなどのテキストデータも含まれている。この膨大な量の情報から、投資家やトレーダーは、それぞれの投資戦略に従い、必要な情報を選び出し、投資判断を下している。ニューラルネットワークや遺伝的アルゴリズム等を数値データに用いて市場分析を行うものがあり、一定の成果を上げている[1]。一方で、近年、データマイニング技術を用いて、市場動向を分析する研究が多く行われている。ニュースのヘッドラインを分析し、為替市場の短期予測を行った研究[2]や、中央銀行の発行するテキストを用いた研究などがある。日本銀行のテキストを対象にしたものでは、ニューラルネットワークを用いることでレポートを指数化する研究[3]、政策変更確率を予測する研究[4]、米国のFOMCのテキストを対象にしたものでは、トピックモデルによる分類と極性語による指数化に関する研究[5]がある。本研究では、日本銀行の景気に対するセンチメントを日本銀行の発行するテキストに対して、トピックモデルおよびニューラルネットワーク用いて、トピック事に分解した指数化を行った。今後の展望として、よりリアルタイム性を出すためにテキストの発行以前において、マクロ経済指数を外挿することによりリアルタイムな指数の予測を目指す。

For algorithmic trading, it is important to reduce market impact and opportunity costs that closely related to market liquidity. In this work, we propose a tick-based approach to prediction of the liquidity. Our method utilizes order data encoded according to its exibility and a Long Short-Term Memory(LSTM) that predict a next order. Accuracy of the model outperforms by a large margin maximum occurrence ratio of order labels. Furthermore, we examine the embedding layer of the trained model and find out that it obtains difference and similarity between each order.

Due to high uncertainty in the stock market, it is difficult to predict the future uctuations of stock prices even if we use the state-of-the-art techniques of machine learning, such as Deep Learning. However, in some cases with choosing an appropriate machine learning algorithm, feature values and outputs for the prediction, we can have desirable predicted results, especially on short-term stock uctuations about some market indices. Some initial reliable results have been achieved in our related work, by using Soft Confidence-Weighted (SCW) Leaning, which is one of online learning. In this paper, we propose a predicting method using two-level mapping and SCW. We will focus on feature transformations using the two-level mapping. The first one is based on the mathematical concept of the Singular Value Decomposition (SVD), to get strong convergence and higher accuracy. The second one is to make the predicted Fluctuation Strength (FS) more precisely, in which we use pre-learned outputs and do relearning.

Market participants place their limit/market orders by taking into account both the trajectory and current status of the limit order book. This behavior is based on the policy that the shape of the limit order book is quite informative for predicting future direction of a traded asset. In this paper, we employ Support Vector Machine combined with conformally transformed Gaussian RBF kernel to forecast the mid price dynamics. Our empirical studies show that the conformal transform methods improved the precision more than 3% in average compared to the standard Gaussian RBF kernel.

Researchers and practitioners in the economic and financial field recently have a keen interest in discovering new ideas by making full use of large-scale data, such as in the form of document data of company valuation in online news and the form of numerical data of company financial indices. One promising approach to analyzing such large-scale data is topic modeling, typically by Maximum Entropy Discrimination LDA (MedLDA). MedLDA is a supervised topic model that can improve accuracy of latent topic estimation by making use of the side information associated with each document. In this paper, we generalize Multi-task MedLDA (MultiMedLDA) that simultaneously addresses classification and regression tasks in an extension of MedLDA. In this paper, we evaluate the effectiveness of MultiMedLDA through experiments with enterprise evaluation documents associated with continuous labels of change rate of operating incomes and discrete labels of categories of business, and discuss it compared with single-task MedLDA.

The purpose of th is discussion is to illustrate in practice the investment risk analysis of geopolitical events such as Brexit, China shock and similar events, which traditionally have been forecasted using qualitative approaches I discuss what quantitative method s are available for modeling and foreca sting geopolitical risk from t he perspective of investment prac tit ioners, and provide us age case examples.

機械学習の利点は人間が分からない特徴の分析ができるところ。 今回の研究ではUSD/JPYの2005年から2015年のデータを取得して、ニューラルネットで学習させたところどのような、勝ちパターンが見えてくるのかを研究しました。検証は2016年のデータで実施しました。

Network embedding is one of the approaches to effectively analyzinge the network data. Almost all the existing network embedding methods adopt shallow models without having deep architecture that is commonly used in deep learning studies. However, shallow models cannot capture highly non-linear network structures that are often observed in real-world, complex networks. To solve this problem, Structural Deep Network Embedding (SDNE) was proposed as a deep model for network embedding. In this paper, we focus on Generative Stochastic Network (GSN) for network embedding, in an extension of Autoencoder. GSN robustly captures latent features of data by adding random noises in the process of learning. The framework to capture the latent structure of network is similar to that of SDNE. As a target network in this study, we focus on the time-dependent networks. In order to address the dependency between time intervals and to capture the tendency of previous time interval, we propose time-dependent pretraining that uses the parameters learned from the previous time interval as initial states of the current time interval while in the learning process. In the experiments, we use time-dependent financial network data, where each node (or vertex) represents a bank and each link (or directed edge) represents a per-month transaction between a pair of banks, resulting in a series of per-month networks.