Not logged in.
Quick Search - Contribution
Contribution Details
| Type | Master's Thesis |
| Scope | Discipline-based scholarship |
| Title | Predicting Financial Distress Using Quantitative Textual Analysis |
| Organization Unit | |
| Authors |
|
| Supervisors |
|
| Language |
|
| Institution | University of Zurich |
| Faculty | Faculty of Economics, Business Administration and Information Technology |
| Number of Pages | 71 |
| Date | 2014 |
| Abstract Text | Executive Summary This thesis uses the quantitative linguistic analysis approach to study the sentiment of over 24,000 news articles on 126 European companies. Thereby, eight different media tone variables are derived, which are in turn employed in an innovative bankruptcy prediction model. Structured into four main parts, the study analyzes theoretical concepts of bankruptcy prediction and textual analysis, as well as their application into practice. In a first part, a short introduction to five major bankruptcy models, grouped into accounting-based and market-based, is given. In a second part, the related research in linguistic analysis, particularly in the area of finance, is presented. In a third part, an empirical study is conducted, in which media sentiment variables are constructed in four steps: (a) selection of adequate news sources and information collection, (b) pre-processing of the textual documents, (c) application of a suited dictionary, and (d) adjustment with the tf/idf term weighting scheme. The model is empirically tested using a logit regression approach. In a fourth part, the study is concluded by summarizing the main findings and suggesting further areas of research. This study finds that the tone in news articles contributes incremental information on the financial stability of a company. Bankruptcy prediction accuracy ranges from 76.2% to 82.5% in the 90 day period prior to the bankruptcy event, up to 57.1% 180 days prior, up to 55.6% 270 days prior and up to 54% 360 days prior to the corporate default. The sentiment variables significantly differ in predicting quality, with the model specification using the negative sentiment category forecasting bankruptcy best. The necessity of using explorative approaches in this research points to the occurring difficulties, particularly with the application of textual variables in the bankruptcy prediction model. This thesis identifies four main issues with the methodologies: (1) the limited size of the population, which constrains the number of possible samples, (2) the time intensity and error-proneness of manual data collection, (3) the ambiguity in texts (e.g., negations, information quality), which restricts the extraction of meaningful sentiment variables, and (4) the occurring statistical issues (e.g., quasi-complete separation, complete data bias). |
| Export |
BibTeX
|