Masanori HiranoPreferred Networks Inc.
Masanori Hirano
Doctor of Philosophy
About
89
Publications
6,239
Reads
How we measure 'reads'
A 'read' is counted each time someone views a publication summary (such as the title, abstract, and list of authors), clicks on a figure, or views or downloads the full-text. Learn more
181
Citations
Introduction
Masanori Hirano currently belongs to Preferred Networks, Inc., Japan. Masanori does research in Social Simulation, Text Mining, Data Mining, Financial Market, and Artificial Intelligence.
Additional affiliations
September 2020 - present
August 2019 - present
Education
September 2020 - September 2023
February 2020 - March 2020
April 2019 - September 2020
Publications
Publications (89)
This study focuses on creating a domain-specific Large Language Model (LLM) for the finance sector. We developed a specialized pre-training dataset containing financial reports, central bank documents, and corporate information. Then, the continual pre-training was conducted on base models like rinna/nekomata-14b and meta-llama/Meta-Llama-3-70B, an...
Inspired by Kolmogorov-Arnold Networks (KANs), we propose the KAN-based Option Pricing (KANOP) model to value American-style options, building on the traditional Least Square Monte Carlo (LSMC) algorithm. KANs leverage the Kolmogorov-Arnold representation theorem to deliver a data-efficient alternative to standard Multi-Layer Perceptrons, requiring...
This paper proposes a novel method for constructing instruction-tuned large language models (LLMs) for finance without instruction data. Traditionally, developing such domain-specific LLMs has been resource-intensive, requiring a large dataset and significant computational power for continual pretraining and instruction tuning. Our study proposes a...
The AI traders in financial markets have sparked significant interest in their effects on price formation mechanisms and market volatility, raising important questions for market stability and regulation. Despite this interest, a comprehensive model to quantitatively assess the specific impacts of AI traders remains undeveloped. This study aims to...
Electricity procurement of industrial consumers is becoming more and more complicated, involving a combination of various procurement methods due to electricity liberalization and decarbonization trends. This study analyzed and improved power procurement strategies for a factory to achieve carbon neutralization through a multi-agent model simulatin...
This study aims to evaluate the sentiment of financial texts using large language models (LLMs) and to determine whether LLMs exhibit firm-specific biases in sentiment assessment. In particular, we investigate the impact of general knowledge about firms on the sentiment measurement of texts by LLMs. First, we compare the sentiment scores of perform...
We introduce a novel generative adversarial network (GAN) designed to generate realistic trading orders for financial markets. Past models of GANs for creating synthesized trading orders have always been focused on continuous spaces because their architecture has constraints coming from the learning algorithm. Contrary to this, actual orders are pl...
Large language models (LLMs) are now widely used in various fields, including finance. However, Japanese financial-specific LLMs have not been proposed yet. Hence, this study aims to construct a Japanese financial-specific LLM through continual pre-training. Before tuning, we constructed Japanese financial-focused datasets for continual pre-trainin...
Option pricing, a fundamental problem in finance, often requires solving non-linear partial differential equations (PDEs). When dealing with multi-asset options, such as rainbow options, these PDEs become high-dimensional, leading to challenges posed by the curse of dimensionality. While deep learning-based PDE solvers have recently emerged as scal...
Financial documents are increasing year by year, and natural language processing (NLP) techniques are widely applied to process these documents. Specifically, Transformer-based pre-trained models such as BERT have been successful in NLP in recent years. These cutting-edge models have been adapted to the financial domain by pre-training with financi...
Deep Hedging is now a key technology that uses deep learning to hedge derivatives. Various models for the underlier price process have been used for it, and it is known that the model is important for hedging performances. In this study, we conducted Deep Hedging experiments using artificial market simulations for its underlier simulator. As an art...
Recent developments of machine learning techniques have made AI traders more prominent in financial markets, drawing attention to their market impact. We focus on the GARCH(1,1) model, a key financial time series model, to analyze the influence of AI traders. The GARCH model is the most common method for modeling conditional variance capable of rep...
In this paper, we focus on the residual returns that are not explained by the common factors in financial asset returns. We propose a novel method to extract well-behaved residual returns based on principal component analysis (PCA). Traditional PCA requires determining the number of common factors, presenting a trade-off: increasing the number redu...
DeepHedge, using deep learning and price time series simulation for better hedging, is noted for handling real-world market issues like trading fees, not just ideal markets. It's known that training gets tough with standard feedforward neural networks in Deep Hedging, but some settings have efficient structures like the No-Transaction Band Network....
With the recent development of large language models (LLMs), models that focus on certain domains and languages have been discussed for their necessity. There is also a growing need for benchmarks to evaluate the performance of current LLMs in each domain. Therefore, in this study, we constructed a benchmark comprising multiple tasks specific to th...
With the recent development of large language models (LLMs), the models focusing on certain domains and languages have been discussed in their necessity. There is also a growing need for benchmarks to evaluate the performance of current large language models in each domain. Therefore, in this study, we constructed a benchmark consisting of multiple...
大規模言語モデル (LLM) の発展とともに、分野や言語に特化した言語モデルの構築の必要性が議論されてきている。その中で、現在の大規模言語モデルがどの程度の性能を発揮するかを分野に特化して評価するベンチマークの必要性が高まっている。そこで、本研究では、日本語かつ金融分野に特化した複数タスクからなるベンチマークの構築を行い、主要なモデルに対するベンチマーク計測を行った。その結果、現時点では GPT-4 が突出していることと、構築したベンチマークが有効に機能していることを確認できた。
With electricity liberalization and decarbonization becoming increasingly popular, electricity procurement for industrial consumers becomes more and more complicated. This study analyzed power procurement strategies for a factory to achieve carbon neutrality. For the analysis, a multi-agent simulation for an electric power market was constructed. T...
Instruction tuning is essential for large language models (LLMs) to become interactive. While many instruction tuning datasets exist in English, there is a noticeable lack in other languages. Also, their effectiveness has not been well verified in non-English languages. We construct a Japanese instruction dataset by expanding and filtering existing...
This study constructed a Japanese chat dataset for tuning large language models (LLMs), which consist of about 8.4 million records. Recently, LLMs have been developed and gaining popularity. However, high-performing LLMs are usually mainly for English. There are two ways to support languages other than English by those LLMs: constructing LLMs from...
In this study, we performed LoRA tuning on large language models (LLM) based on both Japanese and English using Japanese instruction tuning and evaluated these models from both quantitative and qualitative perspectives. As a result of the evaluation, the effectiveness of tuning with Japanese instruction data was confirmed. Furthermore, we clarified...
Instruction tuning is essential for large language models (LLMs) to become interactive. While many instruction tuning datasets exist in English, there is a noticeable lack in other languages. Also, their effectiveness has not been well verified in non-English languages. We construct a Japanese instruction dataset by expanding and filtering existing...
This study constructed a Japanese chat dataset for large language models. The dataset contains approximately 8.4 million records and includes various tasks in chat format, such as translation and knowledge tasks. To confirm the benefits of our constructed dataset, we tuned an existing large language model and confirmed its performance qualitatively...
This study proposes a new efficient parameter tuning method for multi-agent simulation (MAS) using deep reinforcement learning. MAS is currently a useful tool for social sciences, but is hard to realize realistic simulations due to its computational burden for parameter tuning. This study proposes efficient parameter tuning to address this issue us...
本稿では,新しい人工市場シミュレーションプラットフォームのPAMS: Platform for Artificial Market Simulationsを示す. PAMSは,深層学習技術などとのシームレスな融合を前提におき,Pythonベースのアーキテクチャを採用しつつ,様々なシミュレーションが可能になるように,ユーザーが簡便にエージェントや環境をカスタマイズ可能になっている. 実際に,使用例として,本稿では,深層学習による価格予測を行うエージェントを用いた研究を行い,PAMSの有効性について示す.
Deep hedging is a deep-learning-based framework for derivative hedging in incomplete markets. The advantage of deep hedging lies in its ability to handle various realistic market conditions, such as market frictions, which are challenging to address within the traditional mathematical finance framework. Since deep hedging relies on market simulatio...
Deep hedging is a framework for hedging derivatives in the presence of market frictions. In this study, we focus on the problem of hedging a given target option by using multiple options. To extend the deep hedging framework to this setting, the options used as hedging instruments also have to be priced during training. While one might use classica...
This study proposes a new generative adversar-ial network (GAN) for generating realistic orders in financial markets. In some previous works, GANs for financial markets generated fake orders in continuous spaces because of GAN architectures' learning limitations. However, in reality, the orders are discrete. For example, the orders have minimum ord...
Multi-agent simulations are useful in social sciences but they encounter an evaluation difficulty in that many social phenomena are qualitative, and it is difficult to evaluate quantitatively the realness of simulations. Therefore, we propose a new quantitative evaluation method for multi-agent simulation in social sciences using a generative adver...
In this study, we performed LoRA tuning on large language models (LLM) based on both Japanese and English using Japanese instruction tuning and evaluated these models from both quantitative and qualitative perspectives. As a result of the evaluation, the effectiveness of tuning with Japanese instruction data was confirmed. Furthermore, we clarified...
Electricity procurement of industrial consumers is becoming more and more complicated, involving a combination of various procurement methods, due to electricity liberalization and decarbonization trends. This study analyzed power procurement strategies for a factory to achieve carbon neutralization through a multi-agent model simulating the electr...
This study constructed a Japanese chat dataset for tuning large language models (LLMs), which consist of about 8.4 million records. Recently, LLMs have been developed and gaining popularity. However, high-performing LLMs are usually mainly for English. There are two ways to support languages other than English by those LLMs: constructing LLMs from...
This study constructed a Japanese chat dataset for tuning large language models (LLMs), which consist of about 8.4 million records. Recently, LLMs have been developed and gaining popularity. However, high-performing LLMs are usually mainly for English. There are two ways to support languages other than English by those LLMs: constructing LLMs from...
Deep hedging is a framework for hedging derivatives in the presence of market frictions. In this study, we focus on the problem of hedging a given target option by using multiple options. To extend the deep hedging framework to this setting, the options used as hedging instruments also have to be priced during training. While one might use classica...
Electricity procurement of industrial consumers is becoming more and more complicated , involving a combination of various procurement methods, due to electricity liberalization and decarbonization trends. This study analyzed power procurement strategies for a factory to achieve carbon neutralization through a multi-agent model simulating the elect...
Deep hedging, a framework for hedging a portfolio of derivatives using deep learning and price time-series simulation, has been gaining popularity because it allows for more realistic trading strategies that consider market frictions, such as transaction costs. However, for deep hedging learning, a specific price process, such as the Heston process...
The application of natural language processing (NLP) to financial fields is advancing with an increase in the number of available financial documents. Transformer-based models such as Bidirectional Encoder Representations from Transformers (BERT) have been successful in NLP in recent years. These cutting-edge models have been adapted to the financi...
This study proposes a reinforcement-learning-based method for efficient parameter tuning in multi-agent simulations (MAS). Usually, MAS has a high computational burden because of the agents involved; thus, it is important to tune its parameters efficiently. Our proposed method is centered around actor-critic-based reinforcement learning methods, su...
This study analyzed various electricity procurement scenarios for a factory in terms of carbon neutrality, using multi-agent simulations. We performed a multi-agent simulation with power-consuming and power-generating agents to simulate the electric power market. Additionally , we developed a factory model reflecting the actual electricity consumpt...
This study analyzed the effect of order simultaneity in financial markets on data mining tasks, using multi-agent simulations. In financial markets, multiple orders are submitted almost simultaneously or within very quick succession; such orders are thought of as independent of one another. We call this phenomenon order simultaneity. If order simul...
This study proposes a new scheme for implementing actual data into artificial market simulations at the level of trader agents. Traditionally, agent design is performed by humans, so the reliability of trader agents depends on the sense of the model designer. Because humans can introduce bias or overlook the important features of actual traders, we...
This study proposes a new scheme for implementing actual data into artificial market simulations at the level of trader agents. Because humans can introduce bias or overlook the important features of actual traders, we implemented the actual data and automated the strategy learning (imitating) of agents using machine learning (ML). We then ran arti...
This study proposes a new generative adversarial network (GAN) for generating realistic orders in financial markets. In some previous works, GANs for financial markets generated fake orders in continuous spaces because of GAN architectures' learning limitations. However, in reality, the orders are discrete, such as order prices, which has minimum o...
In addition to the need for electricity consumers to take into account the complex behavior of the electricity market as a result of electricity deregulation, CO2 emissions from economic activities have also become an issue in response to recent calls for decarbonization, making the electricity procurement environment increasingly complex. In this...
Deep Hedging, which uses deep learning and price time-series simulations to optimize option hedging, has recently been in the spotlight because it enables more realistic hedging that can take into account frictions such as transaction fees (imperfect market). However, the situation of hedging an option by other options has never been addressed by d...
本研究では決算短信や有価証券報告書を用い,言語モデルの BERT と ELECTRA につ
いて,事前学習や追加で事前学習 (追加事前学習) を行いモデルを構築する.構築したモデルについて,金融ドメインのタスクによって汎用コーパスを用いたモデルとの性能を比較する.その際,ファインチューニングを行う層の数などパラメーターによる性能についても比較を行う.構築した一部のモデルについては一般に公開する.
Electricity procurement by electricity consumers is becoming more and more complicated due to the trend of electricity liberalization and decarbonization. In this study, we evaluate the cost of carbon neutrality in electricity procurement for a factory as a large electricity consumer using an electricity market multi-agent model and a factory elect...
This study demonstrates whether analysts' sentiment toward individual stocks is useful in predicting the macroeconomic index. This can be achieved by using natural language processing to create polarity indexes from analyst reports. In this study, the created polarity indexes were analyzed using the Vector Autoregressive model with various macroeco...
This study proposes a new model to reverse engineer and predict traders’ behavior for the financial market. This trial is essential to build a more reliable simulation because the reliability of models is a fundamental issue in the increasing use of simulations. Thus, we tried to build a behavior model of financial traders through the traders’ futu...
BERT を始めとする事前学習言語モデルは,様々な自然言語処理のタスクにおいて成果を上げている.これらのモデルの多くは Wikipedia やニュース記事などの一般的なコーパスを用いているため,専門的な単語が使用される金融分野においては十分な効果が得られない.本研究では決算短信や有価証券報告書から事前学習言語モデルを構築する.また金融ドメインのタスクによって汎用モデルとの性能を比較する.
本研究では, アナリストの個別銘柄に対するセンチメントが, マクロ経済指標の予測に役
立つかを実証する. これはアナリストレポートのテキスト情報を自然言語処理を使用して極性指標を作成することで実現可能となる. 本研究では, 作成した極性指標に対し, 各種マクロ経済指標を使用し,VAR モデルを用いた分析を行った. 結果, 極性指標から物価, 為替, 国債等の指標へのグレンジャー因果性があることが確認された. これにより, 極性指標が先行しており, マクロ経済指標の予測に役立つことが示唆された.
Recently, general-purpose language models pre-trained on large corpora such as BERT have been widely used. In Japanese, several pre-trained models based on Wikipedia have been published. On the other hand, general-purpose models may not be sufficiently effective in the financial domain because of the use of specialized phrases. In this study, we co...
This is an extension from a selected paper from JSAI2020. In this study, we propose a stochastic model for predicting the behavior of financial market traders. First, using real ordering data that includes traders’ information, we cluster the traders and select a recognizable cluster that appears to employ a high-frequency traders’ market-making (H...
This paper proposes a new model to reverse engineer and predict traders' behaviors for financial market. In this model, we used an architecture based on the transformer and residual block, and a loss function based on Kullback-Leibler divergence. In addition, we established a new evaluation metric, and consequently, succeeded in constructing a mode...
電力の自由化に伴い,工場などの大口の電力需要を持つ事業者は,電力調達の際に市場価格や需給の変動など不確定要素を新たに考慮する必要が出てきた.本研究では,価格高騰時のコスト削減や需給調節などのために需要側がとりうる手段の1つであるデマンドレスポンス(DR)について,電力市場におけるその効果を分析した.具体的にはまず,標準的な工場の用途別電力消費量時系列データから,主成分分析により特性を抽出し,これを基に工場の用途別電力消費モデルを構築した.次に,この工場エージェントに加え電力供給エージェントと需要エージェントが参加する,JEPX(日本卸電力取引市場)の1日前市場を模したマルチエージェントモデルを用いて,シミュレーション実験を行った.実験では,工場のDRシナリオについてそれぞれ市場に対するスケー...
In this paper, we propose a new scheme for implementing the machine-learned trader-agent model in financial market simulations based on real data. The implementation is only focused on the high-frequency-trader market-making (HFT-MM) strategy. We first extract order data of HFT-MM traders from the real order data by clustering. Then, using the data...
In this paper, we address the problem of unintentional price collusion, which happens due to auto pricing, such as systems using reinforcement learning. Firstly, Q-learning, sarsa, and deep Q-Learning models were used for auto pricing to test whether they cause collusion. To test them, we performed multi-agent simulations of a competitive market wi...
In this paper, we address the problem of unintentional price collusion, which happens due to auto pricing, such as systems using reinforcement learning. Previous work has pointed out the risk of unintentional collusion caused by auto pricing using Q-learning, one of the basic models of reinforcement learning, using a market simulation. This study e...
Recently financial markets have shown significant risks and levels of volatility. Understanding the sources of these risks require simulation models capable of representing adequately the real mechanisms of markets. In this paper, we compared data of the high-frequency-trader market-making (HFT-MM) strategy from both the real financial market and o...
In this study, we propose a stochastic model for predicting the behavior of financial market traders. First, using real ordering data that includes traders' information, we cluster the traders and select a recognizable cluster that appears to employ a high-frequency traders' market-making (HFT-MM) strategy. Then, we use an LSTM-based stochastic pre...
In this study, we assessed the impact of capital adequacy ratio (CAR) regulation in the Basel regulatory framework. This regulation was established to make the banking network robust. However, a previous work argued that CAR regulation has a destabilization effect on financial markets. To assess impacts such as destabilizing effects, we conducted s...
We propose a scheme for selecting stocks related to a theme. This scheme was designed to support fund managers who are building themed mutual funds. Our scheme is a type of natural language processing method and based on words extracted according to their similarity to a theme using word2vec and our unique similarity based on co-occurrence in compa...
In this paper, we compare data of the high-frequency-trader market-making (HFT-MM) strategy from both the real financial market and our simulation. For the former, we extracted trader clusters and identified one cluster whose statistical indexes indicated HFT-MM features. Subsequently, we analyzed the difference between these traders' orders and th...
This paper proposes a pointwise prediction for a sentence boundary detection task. The proposed pointwise prediction is combined with our original word embedding method and three-layered perceptron. It predicts whether the targeted words have the role of the beginning/end of a sentence or not by using word features around the targeted words. We tes...
本研究は,金融市場における高頻度取引(HFT)のマーケットメイク(MM)戦略と呼ばれる注文行動について分析を行うことを目的とした.株式会社日本取引所グループより提供を受けた,東京証券取引所の注文データを使用し,仮想サーバーの名寄せを前処理として行なった.その結果得られた,取引主体別の注文データを,いくつか指標を使うことで,クラスター分析を行い,高頻度マーケットメイク戦略(HFT-MM)を取っている取引主体を抽出し,それらの注文が,直近約定価格から何ティック離れたところに置かれているかについて計算した.その結果,HFT-MMとされる行動主体は,直近約定価格からかなり離れた位置(5-10ティック)のところにも注文を置いていることが明らかになった.この結果は,HFT-MMとされる取引主体が,マーケ...
This paper proposes a pointwise prediction for a sentence boundary detection task. The proposed pointwise prediction is combined with our original word embedding method and three-layered perceptron. It predicts whether the targeted words have the role of the beginning/end of a sentence or not by using word features around the targeted words. We tes...
本研究において,文書内における単語の共起を利用した上位下位概念の推定の手法を提案した.本手法に基づき,Wikipediaの記事を文書として使い,Wordnetに含まれる上位下位概念をデータとして実験を行った.
その結果,精度は低いものの,一定の有効性を確認することができた.本手法は非常に少ない訓練データで必要なパラメータのチューニングが可能であることもわかった.
We propose an extended scheme for selecting related stocks for themed mutual funds. This scheme was designed to support fund managers who are building themed mutual funds. In our preliminary experiments, building a themed mutual fund was found to be quite difficult. Our scheme is a type of natural language processing method and based on words extra...
We propose a method to select and rank stocks related to a given theme. The proposed method has two flows; obtaining related words, and selecting related stocks based on obtained related words. First, on the basis of the given theme word, the proposed method selects words with high similarity using an ensemble of word2vec models. Then, we modify th...
In this study, we assessed the impact of capital adequacy ratio (CAR) regulation like that in the Basel regulatory framework. To asses impact such as destabilizing effects (e.g., whether CAR regulation destabilizes markets or not), we conducted simulations of an artificial market, one of the computer simulations imitating real financial markets. An...
Basel regulatory framework, one of CAR (capital adequacy ratio) regulations, is said to make markets destabilized in a previous study. But the previous study included some inappropriate assumptions. So, this study assessed this destabilizing effects with a new model. In my model, FCN agents and 2 kinds of portfolio agents, CAR regulated ones and no...