Stock exchanges are making many changes to become a more desirable market. One of these efforts is changing the tick size. This represents the unit of order price when an investor places an order. In this study, we estimated the change point for trading volume in the period before and after this change. As a result, it became clear that the change point exists before the actual tick size change date.

This paper proposes an Understanding of non-Financial Objects in Financial Reports (UFO) task. The UFO task aims to develop techniques for extracting structured information from tabular data and documents, focusing on annual securities reports. We will provide a dataset based on annual securities reports and organize an evaluation-based workshop for participants. The UFO task consists of two subtasks: table data extraction (TDE) and text-to-table relationship extraction (TTRE). The table data extraction subtask aims to extract the correct entries and values in the tables of the annual securities reports. The text-to-table relationship extraction subtask aims to link the values contained in the tables with the relevant statements in the text. In this paper, we describe an overview of the UFO task.

This paper targets to predict overnight stock movement by taking contextualized news and stock information into account, using the Pre-trained Language Model (PLM) that was recently popular in Natural Language Processing (NLP) field. We proposed a model in which, given a piece of news and a stock code, the model can predict its overnight stock movement by utilizing combined news-stock embedding. Such embedding consists of (1) the contextualized embedding that contains the semantics of such a piece of news produced by a language model trained on a set of news and its paired stock movement. (2) The contextualized embedding is produced by a PLM trained on the information of stocks. Moreover, we introduce news augmentation on multiple pieces of news for the input and study its effect, respectively.

Autoregressive integrated moving average (ARIMA) is a widely used linear model withgreat performance for time series forecasting problems. Supplemented by support vector regression (SVR), an effective method to solve the nonlinear problem with a kernel function, ARIMA-SVR model captures both linear and nonlinear patterns in stock price forecasting. However, it does not have high accuracy and parameter selection speed when its parameters are chosen by the traditional method. Therefore, in this study, we applied genetic algorithm (GA) to optimize the parameter selection process of SVR to improve the performance of the ARIMA-SVR model. Subsequently, we built the ARIMA-GA-SVR model by integrating ARIMA with optimized SVR. Finally, we used actual stock price data to compare the forecasting accuracy of the proposed model, ARIMA and ARIMA-SVR models using error functions. The result shows that the proposed ARIMA-GA-SVR model outperforms other models.

The Hawkes process is a flexible and versatile model that can accommodate the self-exciting nature of occurrences of events in natural and social sciences. Recently, a nonlinear version of this model has been applied to describe financial markets' intermittent and clustering behaviour. On the other hand, analytical characters of the nonlinear Hawkes process have not been studied well due to its nonlinear and non-Markovian nature. In this talk, we present our solution to a broad class of nonlinear Hawkes processes via the field master equations based on our previous publications (K. Kanazawa and D. Sornette, PRL 2020 and PRL 2021). We find that the power-law relationship is ubiquitously found in the intensity distributions in nonlinear Hawkes processes. This character would be helpful for data calibration to financial data, particularly from the viewpoint of the power-law price movement statistics.

Selection of product portfolio and determination of its inclusion ratio is fundamental management issues in e-commerce (EC) logistics business. EC logistics business is a form of business in which products are stocked in advance. Upon receiving an order from a customer via the Internet, the company allocates the stocked products and ships them to the customer. The problem is to control logistics costs by taking into account risks, such as seasonality of individual products, changes in trends, and sudden fluctuations in demand, to increase expected profits continuously. We investigate a method of product portfolio optimization using Markowitz's mean-variance model as a starting point for solving this problem. The general computational complexity of the mean-variance model scales with the target number of items n as ∝ n3. Since n is of order a million, a significant issue is whether it can efficiently find an optimal or good solution using such an extensive data set in a finite amount of time. We seek a method to obtain a feasible solution to this problem within an acceptable time frame for business operations using computer resources that are currently relatively easy to acquire by an average business. In this study, we report on our investigation of classical methods such as divide-and-conquer, compact decomposition, and multi-factor models, as well as relatively new methods such as quantum annealing.