Text Analysis of Bankruptcy Forecasting for Energy Companies

Research paper attempts to predict the bankruptcy of a leveraged industry such as the energy industry. The main goal of this work is to build a default prediction model with the following indicators: financial, macroeconomic, market, non-trivial factors such as text analysis. The basis of the research includes the analysis of news tonality of the selected companies, conducted using the Loughran and McDonald Word List dictionary, Thomson Reuters was selected as the database for news analysis. The work encompasses three models for predicting bankruptcy: logistic regression, decision tree, and random forest. Each specification has its own set of variables, but no significant correlation was found. The empirical base of the research consists of 125 North American companies from the energy industry, which filed for bankruptcy in the period from 01.01.2015 to 01.01.2020. Non-default analogues from the S&P information resource were chosen to conduct the research. Four hypotheses were put forward during the study. Three hypotheses were confirmed and one was partially confirmed. The first hypothesis of including macroeconomic factors into the financial and market analysis leads to the increase of the model accuracy. It has been confirmed. The second hypothesis about the inclusion of text analysis in the calculation of the probability of default was partially confirmed. The third hypothesis about the increase in the accuracy of the forecast with the increase in the period was proved. The fourth hypothesis, that the economic instability of the country has a positive impact on the probability of bankruptcy of companies, was also confirmed. Keywords: bankruptcy, text analysis, news, tonality analysis, news impact, random forest, logistic regression, machine learning.