Big Data Science in Finance

Chapter 4Modeling Human Behavior with Semi-Supervised Learning

Introduction

One of the main challenges of data science is translating the human capital and know-how of expert professionals into computer models, often referred to as artificial intelligence programming applications. For example, a master trader may have a record-winning strategy, but is about to retire. How do you copy his decision-making? Alternatively, an industry analyst may have a successful ability to predict the content of upcoming earnings announcements. How do you immortalize his brain in a computer program? Is that even possible?

Supervised frameworks discussed in Chapter 3 covered models that used structured data neatly organized into rows in columns. An example of such data may be the corporate financials such as the figures in quarterly and annual regulatory filings, required and published by the U.S. Securities and Exchange Commission (SEC). By contrast, most of the data and news consumed by humans arrive as text. For example, news articles, social commentary, and regulatory updates that may directly affect asset prices comprise textual items. Traditionally, the text was read and interpreted by trained analysts, who collected and thought about various press releases and news articles about a specific firm and industry, alongside product and competitor information. Analysts next wrote and published opinion pieces on the upcoming changes in the given stock or bond price, industry forecasts, or the ...