Machine Learning in Capital Markets : Part I

Since the first scientific treatises on the research object of artificial intelligence in the late 1950s, the topic has become increasingly relevant. AI is now used in all newer areas of technology and is continuously being developed and optimized with regard to more efficient uses in the respective areas of application. AI, as such, presents itself as a complex area that on the one hand tries to understand intelligence and on the other hand develops evidence of understanding by creating intelligent, technical systems. Due to the complexity of the research object, AI currently encompasses a large number of different areas, the contents of which can be arranged on a spectrum between general (e.g. learning and perception) and highly specialized (e.g. the proof of mathematical assertions). Progressive developments in technology and science, which originally made simple activities increasingly complex and abstract, can no longer be carried out without extensive analyzes. The analysis of highly complex technical processes is often accompanied by an exponentially increasing volume of data. Accordingly, these analyzes are often based on a variety of factors. In this context, AI, in particular the area of ​​machine learning, offers an option to optimize analysis activities.

According to the common definition, machine learning is a learning principle in which computer systems are enabled through the application and research of processes to independently acquire and expand knowledge. The aim of this process research is to be able to solve an existing problem better than before. In contrast to manual revision through humanitarian capacities, machine learning processes enable complex data material to be processed efficiently. The financial market offers a clear example with regard to the optimization of analysis activities.

In this context, the following case study provides a more detailed explanation: imagine that you are a woman in your late 40s and would now like to apply for a loan for your home. Your partner has a permanent contract with a large machine manufacturer and you work as an official at the tax office in your district. These aspects should speak without problems that your loan application will be approved and that you can start building your own home without any detours. After the first talks with your advisor at the bank, who is confident about approving the loan application, there should be no obstacles to the home concept. However, a few days later they receive a message from the bank rejecting their loan application. This case study shows a fictitious situation in which, however, it was not people who made the decision to reject the loan application, but machines. All criteria of the loan application are compared with risk factors and patterns of other borrowers in order to predict a possible loan default. There are different learning approaches and methods to identify and compare such different patterns and risk factors.

The present work addresses these learning approaches and processes in relation to machine learning as a subcategory of AI. With regard to the outgoing thesis, it is to be discussed and exemplified to what extent machine learning processes are suitable for forecasting financial market-specific data. The work is divided into four main chapters. The first main chapter serves as an introduction to the topic of machine learning and outlines the general processes with regard to learning strategies, learning methods and areas of application. The possible areas of the financial market to be analyzed by machine learning are then presented. Based on these representations, the question arises to what extent machine learning can be used to forecast financial market-specific data.

The second part of this article will describes the use of machine learning to forecast financial market-specific data. It will provide information about whether machine learning as such can be successfully used on the stock market. In order to be able to make this statement, the various processes of machine learning are checked with regard to their analytical ability in relation to financial data. The process of machine learning is divided into five different processing stages: data collection and preprocessing, creation of features that can be learned, feature selection and classification. The results and concluding discussions regarding the usability of machine learning methods on the capital market will be discussed in the fourth part. A comprehensive final assessment will be given to shed light on the extent to which machine learning can be used on the stock market.

Generally, a large number of factors must be taken into account when analyzing capital markets, particularly the stock market, in order to be able to successfully predict stock prices. In view of the exponentially increasing volume of data, it is an elementary factor to check the quality of the data used, since only prepared and cleaned data sets for a classification procedure provide positive results and thus have a strong influence on the success factor with regard to the forecast. With regard to the conciseness of the aforementioned data premise, it is crucial that known and foreseeable exogenous factors are taken into account and integrated into the analysis of the database. This work also demonstrated that direct and indirect characteristics play a fundamental role in relation to the informative value of the object under consideration. The aim was to achieve a high level of generalization of the characteristics so that they could be used afterwards for further calculations and comparisons. In a subsequent classification, the data only had to be divided so that it could be used due to its predictability, controllability and adaptability.

[Will be continued to part II]