The question of whether stock prices fully reflect all relevant information takes a direct shot at the efficient market hypothesis, stirring a lively debate. On one side, we have the believers in the rational expectation model. They see market participants as perfectly rational beings, always making decisions that maximize their economic benefit. On the other side, there's a growing camp that doubts this perfect rationality. They argue that humans are prone to irrationality and that stock prices might not always mirror the true underlying value. Behavioral finance stands firmly in this camp, suggesting that our psychological quirks and biases can significantly shape how we view and act in the financial markets.

In this project, we take a closer look at how visual cues, like stock charts, and verbal cues, such as outdated but still circulating information, influence investor decisions. By exploring these aspects, we aim to peel back the layers of how stock prices are formed, questioning the age-old belief that they reflect all available information. This approach isn't about pitting one theory against another; it's about understanding the complexities of financial markets through the lens of human behavior.

1. Verifying Visual Information

To quantitatively handle investors' visual information, this project focused on chart images. Since image information can be treated as numerical information in matrices, stock chart images, being represented as numerical matrices, can be analyzed within the framework of machine learning with neural networks. Specifically, in this research project, the following tasks were performed. First, we input 3.5 million chart images from the years 1993 to 2000. For these 3.5 million images, each was labeled according to its subsequent outcome. Using CNNs (Convolutional Neural Networks), we created a model that predicts the subsequent outcomes most accurately. This trained model was then used to diagnose chart images from the year 2000 to 2021 that were out of sample. As a result, it became clear that chart images contain information that can predict subsequent stock returns. For more details, please refer to the following working paper.

[https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4478013]

2. Verifying Known Information Published in the SHIKIHO (Japan company handbook)

The SHIKIHO is a traditional stock handbook published four times a year. A distinctive feature of the SHIKIHO, which has existed since before World War II, is that they include evaluations by journalists from Toyo Keizai Inc., which are neither buy-side nor sell-side. The objective of this project is to verify the efficient market hypothesis, which states that known information is already reflected in stock prices, along with clarifying whether the stock market reacts to stale information not in the short term but in the long term. The verification results showed that the evaluative information in the SHIKIHO is not already reflected in stock prices and is related to long-term corporate value. For more details, please refer to the following working paper.

[https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4535647]