Browsing by Author "Qian, Skyla"

For investors covering a stock, one critical event occurs every quarter: earnings day. On this day, a company issues an earnings release, detailing information about its recent financial performance. Along with releasing its earnings report, the company oftentimes hosts an earnings call, during which company management describes the business’s financial results, current projects, and market headwinds and tailwinds. The earnings call also consists of a Q&A segment, where equity analysts probe management with questions relating to the company’s overall health and future outlook. Despite the availability of earnings information, successful stock picking is notoriously difficult, with stock prices subject to a variety of factors.

This paper investigates the impact of EPS surprise, revenue surprise, EBIT surprise, EBITDA surprise, earnings transcript polarity, and earnings transcript subjectivity on stock price movement. The research explores the directional correlations between the feature variables and stock price movement and utilizes linear regression, lasso regression, ridge regression, random forest regression, and logistic regression to generate both regression and classification predictor models. The model results indicate that EPS surprise and earnings transcript polarity are the most important features and that EBITDA surprise and earnings transcript subjectivity are less important. The research produces a logistic regression model with 62% accuracy and a random forest regression model with a mean absolute error of 6.63. Moreover, an analysis of different scaling methods demonstrates that a signed log transformation produces the best evaluation metrics, and an examination of outliers indicates that future stock price research should focus on percentage movement rather than numerical movement when creating predictor models.