This first section gives a clear overview of the structure of this research. First, the occasion of the research and an introduction to the research question will be discussed. Second, the objectives of the research will be stated and hence the research question follows. Also the sample and the results will be argued. This section will be closed with the practical and academic relevance.
The most important document a company releases is the annual report. In the annual report is stated what the operations of the company are and how the financial conditions are. This annual report is a very important document for both internal and external users. Next to the annual report there is the 10-K report. The 10-K report is the enhanced version of the annual report without the glossy color pages and all the pictures. The 10-K report begins with a detailed description of the business, followed by risk factors, a rundown of any legal issues, and, finally, the numbers and financial notes in the back (Gad, n.d.). The annual report and the 10-k report are usually written by members of the management team or corporate attorneys (Goel & Gangoll, 2012). From now on the 10-K report will be referred to as the annual report.
Wondering whether or not to invest in a company, most users of the annual report will look at the quantitative context, such as the financial numbers. With that information investors can make their decision to invest in the company or not. Investing in a company which will go bankrupt is an investors’ or banks’ worst nightmare. Therefore, prior research has primarily focused on the quantitative context of the annual report in order to find a relationship between the financial numbers and a higher risk for bankruptcy. Obviously, the company wants to paint the picture in the annual report as positive as possible without violating any regulations. To do so, the writers of the annual report can have the tendency to creatively use several terminology to camouflage their true financial situation with a (more) positive representation by using linguistic tricks.
As a result, recent research has started to focus at the qualitative context of the annual report as well. They focus on the use of language in the annual report to see if there is any relationship with a higher risk for bankruptcy. This research will focus on the relationship of the use of language and several terminology in the annual report of a company and the risk of that company to go bankrupt. That leads to the following research question:
Can the qualitative context of an annual report predict bankruptcy?
It is important to know if there is any relationship between the use of language in an annual report and the risk of the company going bankrupt, because this can be of great importance to especially investors and banks. The information about the prediction of a company going bankrupt on the basis of the language used in the annual report can make a huge difference in their investment decisions. This information could be used to help investors screen out companies at risk for bankruptcy. Also this information can be useful for government agencies and regulators to determine which companies are at a higher risk for bankruptcy.
The remainder of this paper is organized as follows. Section 2 gives the literature review and hypotheses development. In section 3 the methodology is discussed and the sample selection specified. In section 4 the results are stated. Finally, in section 5 the conclusion and future research possibilities are discussed.
II. LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT
It is a relatively new phenomenon in the financial research streams to combine text-mining research and research into corporate financial events. This means that financial researchers now also focus on the nonfinancial (qualitative) context of the annual report to predict financial events instead of only the financial context. Both of these research streams are widespread, but there is little crossover research. The financial event this literature review will focus on is bankruptcy of a company. Most of the prior research about text analysis and bankruptcy focuses on the Management Discussion and Analysis (MD&A). The MD&A is a very important section of the annual report. It provides an overview of the previous year of operations and how the company fared in that time period. Although this section contains useful information, users of the MD&A should keep in mind that the section is unaudited (Investopedia).
In 2012, Mayew, Sethuraman & Venkatachalam found that both the management’s opinion about going concern stated in the MD&A and the linguistic tone of the MD&A together provide significant explanatory power in predicting whether a company will go bankrupt. The striking feature of these findings is that the information in the MD&A disclosures is more useful in predicting bankruptcy relative to financial ratios three years prior to bankruptcy. They tested the linguistic tone of the MD&A using the dictionary developed by Loughran and McDonald. The resulting frequency of positive and negative words captures the linguistic tone of the MD&A. Loughran and McDonald (2011) found that almost three-fourths of negative word counts in annual reports filings based on the Harvard dictionary are typically not negative in a financial context. Common words like depreciation, liability, foreign, and board are not tonal when occurring in an annual report. That is why they have created their own dictionary for business settings with different lists for different categories of words. There is a list for negative words, positive words, uncertainty words, litigious words and more.
Cecchini, Aytug, Koehler, & Pathak (2010) on the other hand did not assume a priori knowledge of the words or phrases important to financial events detection. They use computational linguistic tools to make their dictionary creation automatic. With those tools they created dictionaries of keywords that can help to predict fraud and bankruptcy, based on the text from the MD&A. They found that their dictionary is able to discriminate companies that will go bankrupt in one year against companies that will not go bankrupt on the basis of their MD&A 80% of the time with the dictionaries they created themselves. They have compared their results with the quantitative prediction methods and came to the conclusion that their methodology achieved superior results using the same data. In 2011, Davis, Piger, & Sedorfound found similar results using an established, textual-analysis program. That program counts words characterized by linguistic theory as optimistic and pessimistic used in quarterly earnings press releases. They subtract the latter count from the former to obtain a measure of the net signal. They stated that levels of net optimistic language in annual reports are predictive of firm performance in the future. They found that net optimistic language in earnings press releases is positively associated with future return on assets (ROA) and generates a market response.
In 1996, Abrahamson & Amir have shown that also the president’s letter contains useful information about the future of the company and not just about past performance. These findings emphasize the importance of nonfinancial information relative to the widely used financial information such as earnings and book values. This together leads to hypothesis 1:
Pessimistic language in the annual report can predict bankruptcy in one year.
Shirata & Sakagami (2008) found that certain nonfinancial key words in financial reports can be used to evaluate a company’s corporate financial position. They verified that it is possible to assess whether a company will go bankrupt or not based on qualitative data, without looking at financial numbers. Their results show that nonfinancial information published by non-bankrupt companies shows a unique tendency. The keywords that appear with high frequencies among non-bankrupt companies reflect their sound financial trends. A few years later Shirata, Takeuchi, Ogino, & Watanabe (2011) found similar results. They state that if some particular expressions appear together with the word ‘dividends’ or ‘retained earnings’ in the same section of the annual report, they were effective in distinguishing between bankrupt companies and non-bankrupt companies. Next to that, Li (2010) found that in annual reports, managers tend to refer to themselves more frequently when the firm performance is better. This leads to hypothesis 2:
The number of times managers refer to themselves can predict bankruptcy in one year.
First the bankrupt companies are selected and for each bankrupt companies there is a control non-bankrupt company selected. To answer the hypotheses the MD&As of the selected bankrupt and non-bankrupt companies are used. A similar kind of method as Shirata, Takeuchi, Ogino, & Watanabe in 2011 used, is used in this research. The chi-square test will be used to analyze the data as this test is the most commonly used significance test in corpus linguistics (Goel & Gangolly, 2012). From those MD&As the frequencies of the words characterized in existing linguistic theories are used (Loughran & McDonald, 2011). Words as optimistic and pessimistic are counted and then the differences in frequencies between the bankrupt companies and the non-bankrupt companies are calculated. Those results construct a measure of managers’ net optimistic language for each annual report (Davis, Piger, & Sedor, 2011). Next to that, the number of self-referring words in each annual report are counted, and the results of the two groups compared with each other.
...(download the rest of the essay above)