In recent years, various researches in the field of economic prediction have been carried out using fundamental analysis and technical analysis with numerical information. Considering news articles containing not only numerical information but also textual information means that we can pay attention to public opinion, and thus we can make more accurate economic trend prediction which are difficult to predict only by numerical information. In this study, we propose a news text analysis for economic trend prediction using polarity dictionaries.
In an environment of high uncertainty, risks in corporate management are changing daily. However, risks do not occur independently, but are often caused by a variety of events (risk factors), including other risks. We define this as causal relationships between risk factors. In addition, a single risk factor may have multiple consequences. For example, when a risk factor such as the Covid-19 epidemic occurred in one company, some departments increased sales due to increased IT investment, while others suffered losses as a result of reduced advertising due to worsening business conditions (risk factors of worsening business environment). Therefore, when actually understanding and assessing the risks faced by a company, it is necessary not only to understand the risk factors, but also to understand how they will ultimately affect business performances. In other words, it is important to stratify the risk factors. In this study, we propose a method to obtain the causal content of extracted risk sentences and to understand the risk factors using a hierarchical structure.
We trained a classifier to determine whether a technology in the patent literature corresponds to a decarbonization-related technology or not, using BERT pre-trained on the patent documents and the corpus of decarbonization-related patents we collected. The trained BERT model was also used to conduct macro analysis of Japanese published patent gazettes applied after 2000 to visualize the decarbonizationrelated patent technologies in Japan.
In this paper, we propose a method that uses causal information extracted from economic texts to predict numerical indicators related to economic and financial fields, such as macroeconomic indicators and stock prices. The proposed method automatically detects whether each sentence in the economic text contains causal information or not, and if it does, it identifies the cause and effect expressions and stores them in our economic causality database. Furthermore, the proposed method calculates the similarity between the result expression of causal information contained in the economic causal database and the causal expression of another causal information, and generates causal chains from the given text data. Causal chains are used to predict how the numerical values of economic indicators will change in the future due to spillover effects.
The purpose of this study is to visualize nonlinear relationships which are quite ambiguous in the correlation diagram by the approach of explainable artificial intelligence (XAI). By using this approach for a practical marketing problem, we could visualize the nonlinear relationship between fund performances and money flows, which is asymmetric in inflows and out flows and is consistent with investors psychology based on the behavioral economics.
Recently, many researchers have studied foreign exchange trading using technical analysis. However, it is difficult to achieve profitability using this technique. Therefore, using Genetic Network Programming, we construct a model that considers the technical index signal strength for devising a profitable trading strategy. Finally, we confirmed the effectiveness of our model using historical data of the exchange market.
In the rollover of forward foreign exchange contracts, FX brokers generally selects tomorrow-next transaction because of higher liquidity and lower risk. However, it might be possible to obtain larger swap points by selecting longer forward transactions such as one-week or three- week forward in terms of the term premium. Therefore, we detect optimal timings to select longer forward transactions by machine learning techniques, and propose a mixed strategy that combines tomorrow-next and longer forward transactions. This timing might be affected by various factors such as global stocks, bonds, commodities, etc., and we could obtain larger swap points by the mixed strategy using the machine learning with these global factors.