We present a system that uses Memetic Algorithms to perform long-term rebalancing of financial portfolios. This allows for a greater resilience to changes in the market. In our experiments, we achieved a number of portfolios which show stable return values even during market crash environments.
The portfolio optimizations are generally to determine the proportion-weighted combination in the static asset portfolio. It means that the assets included in the portfolio have already been given before the optimizing process. However, it is hard to determine the proportion-weighted combination for the optimal portfolio consisting of the static large number of assets. In order to avoid this problem, we propose a method that optimizes the portfolio consisting of not only the given static assets but also the dynamically selected assets in this paper. Our method consists of the following two steps, Steps A and B. Step A is to move the valuable assets expected to have good influence on the objective function from all assets into the portfolio through a GA. Step B is to remove the less-valuable assets expected to have not-so-good influence on the objective function from the portfolio through a GA. In the numerical experiments, we apply the proposed method to creating the index funds for the Tokyo Stock Exchange.
We present an advanced three-body model in markets for numerical simulation of Japanese stock market. The original three-body model was introduced by Yoon in 2001. It is an artificial market model to simulate intraday price fluctuation, and the model is composed of three different types of agents; daytraders, market-makers, and investors. They have different time spans in terms of investment return from each other, and the complex nature of the price fluctuation is explained as resulting from their cross-interaction. In our advanced model, we add the investor agents a new function to observe VWAP for benchmark price, and it makes the price fluctuation very familiar to the real one in the Japanese stock market.
Recently, a lot of studies on the eigenvalue analysis are performed in order to investigate the statistical characteristics of multi-assets correlation in financial markets. In this study, we analyzed the eigenvalue of the cross-correlation matrix of the stock prices listed in the Tokyo Stock Exchange. A filtered cross-correlation matrix is built by removing the noise mode. Comparing the network graphs visualized from normal cross-correlation matrix with the one from the filtered cross-correlation matrix, the latter reflects the characteristic of the market in a more insightful manner. In addition, we build a portfolio from the filtered cross-correlation matrix, and carry out a backtesting by doing a simulated investment. As a result, the portfolio made from the filtered cross-correlation matrix shows better performance than the normal one. Our result suggests that eigenvalue analysis is useful for both understanding the market structure and improving the portfolio for investment.
We conduct a novel virtual stock market experiment that aims to investigate the motives behind short-term investment behavior at the individual decision-making level. In particular, we focus on individual investors' trading strategies in response public information ? about prices, macroeconomic news, and relevant individual-stock information. The distinguishing feature of our experiment is the use of factual contemporaneous news items directly related to the stocks in subjects' portfolios. We find that more information leads among our experiment participants to more frequent trading; majority of it is positive-feedback following individual stock prices and the market as a whole. Our subjects are driven by psychological motives when deciding their orders; in particular, regret aversion is a habitually common reason for trading and for not trading ? through the disposition effect.
The generation of profitable trading rules for Foreign Exchange (FX) investments is a difficult but popular problem. The use of Machine Learning in this problem allows us to obtain objective results by using information of the past market behavior. In this paper, we propose a Genetic Algorithm (GA) system to automatically generate trading rules based on Technical Indexes. Unlike related researches in the area, our work focuses on calculating the most appropriate trade timing, instead of predicting the trading prices.
We examine that the models which are inputted very different factors and are learning same stock returns select similar stocks each other. We make 2 type models, neural network model and linear regression model. In both cases, the predictions of returns are more similar, as learning term is shorter. The reasons may be that the past stock returns which all models learned are same, and there are positive feedback mechanisms in the markets.
We are to analyze relations between stock index and stock BBS. Previous studies in US find that the stock BBS posting can predict market volatility and trading volume. We will develop hypotheses based on the results of these analyses, and apply statistical analysis to the data of companies that are observed the large amount of message posted in Yahoo! stock topics in 2005-2006. Based on the messages, we will analyze the contents of posting using natural language processing and machine learning. In concerning future stock return, the number of posting is significantly correlated with the volatility and the trading volume, and that significant correlations are observed between the amount of bullish opinion and the stock return.
This paper presents our system for mining relations between texts and stock prices. We developed databases for news texts, stock prices, and company names, whose data were extracted from various sources including web articles and Wikipedia. We implemented the interface called MarketSearcher on these databases, which helps users to search and analyze texts related to various types of trends of stock prices.
This paper analyses the influence text information gives to the credit market, focusing on Headline News, a source of information that has immediate influence to the money market, and also which is regarded as one of the important information when making investment decisions. Text automatic classification algorithm has been applied for analysis. After classifying the headline news into several types using this algorithm, verification were made whether there were any characteristic changes in the credit market. As a result, (1) It is possible to build a headline news algorithm by an accuracy of 80% using the automatic classification algorithm. (2)The information extracted from the headline news gives influence to the bond market, similar to the equity market. (3)Difference could be observed in the market before and after the headline news was broadcasted.
In this study, we proposed a new text-mining methods for long-term market analysis. Using our method, we analyzed monthly price data of financial markets; Japanese government bond market, Japanese stock market, and the yen-dollar market . First we extracted feature vectors from monthly reports of Bank of Japan. Then, trends of each market were estimated by regression analysis using the feature vectors. As a result, determination coefficients were over 75%, and market trends were explained well by the information that was extracted from textual data. We compared the predictive power of our method among the markets. As a result, the method could estimate JGB market best and the stock market is the second. Finally, we carried out sequential prediction test in changing the forecast horizon. As a result, the effective forecast horizon of our method was shorter than 3 months.