The Forefront of Financial Data Analysis
Approximately 30 years have passed since the academic field of financial engineering, which attempts to solve various financial problems using engineering methods, was established in Japan. The importance of engineering approaches in finance continues to grow. I will introduce specific examples of this through the research on stock management conducted in my laboratory.
Traditionally, quantitative analysis of stock management has typically used data such as financial figures on profitability and stability disclosed by companies, as well as historical stock prices and trading volumes. With about 3,000 listed companies in Japan alone, analyzing this data to extract information that leads to future profits requires parsing large datasets. However, in recent years, the availability of affordable high-performance PCs and large-scale servers, along with new technologies like text analysis, has made it possible to use new types of large-scale data (alternative data) that were previously unavailable. Examples include textual information from news flashes and internet forums, data on inter-company transaction relationships, executive career information, and satellite imagery. By promptly analyzing this alternative data, there is a potential to extract more profit-generating information than ever before, and financial institutions and researchers worldwide are exploring and analyzing such data.
Here, I will introduce two of my studies that use alternative data. The first involves inter-company transaction information, which is data on the business partners of each company. Previously, data on a company's annual or quarterly performance was available, but now it is possible to identify which transactions generated specific profits. The left panel of Figure 1 illustrates transaction relationship information aggregated by industry; such networks are constructed among tens of thousands of companies globally. These inter-company relationships also affect stock price movements. In our research, we used a method called network analysis to analyze the transaction network structure and extracted information that influences future returns on stock investments. The right panel of Figure 1 shows the results of verifying these findings in the global stock market. An investment simulation using historical data demonstrated stable stock price performance that surpassed the market benchmark.
The second study is a text analysis of the *Kaisha Shikiho* (Japan Company Handbook), a corporate performance information magazine. While various text sources like news, newspapers, and forums exist, the *Kaisha Shikiho* is a performance information magazine that attracts the attention of many investors. The left panel of Figure 2 shows the performance information for Toyota Motor Corporation in 2020, which includes corporate analysts' future outlooks as performance comments. In our research, we proposed a method to extract investment information by applying text analysis techniques to the comments for all listed companies. The right panel of Figure 2 shows the verification of its effectiveness in the domestic stock market. Here too, the investment simulation achieved stable stock price performance exceeding the market benchmark.
As you can see, by appropriately analyzing alternative data, it is possible to extract useful information for stock investment. However, conducting such analysis requires not only a wide range of knowledge about companies and investments but also skills in programming and statistical analysis for handling large-scale data. In recent years, individuals with these skills are called data scientists and are highly valued in financial institutions and other organizations.
As technology advances, the amount of information that can be handled in the world of finance is growing daily. The engineering ability to freely handle such information and extract what is important will continue to be indispensable.