Incomplete Information Revelation through Mining of Semantic Structures of Financial Statements and Unstructured Financial Disclosure

The beliefs of market inefficiency against the well-known Efficient Market Hypothesis (EMH), which believes that the market price is able to fully reflect all public information of a firm. Incomplete Revelation Hypothesis (IRH) asserts that ``statistics that are more costly to extract from public data are less completely revealed in market prices". Investigating unrevealed information has become a mainstream of research in firm valuation and risk management. Qualitative data contains information that is not fully reflected in quantitative data, such as the generating functions of financials, managerial characteristics, managers' incentive, etc. Such information is able to help investors and analysts better understand the managers' decisions and firms' performances. This dissertation investigates the financial information from both structured and unstructured financial disclosures. First, by studying the similarity between financial statement semantic structures, we propose a novel structural financial statement comparability measure and conclude that the structural comparability mediates the relation between accounting function comparability and analysts' coverage, forecast accuracy, and forecast dispersion. This comparability measure is able to efficiently identify industry boundaries. Second, we identify risk information from firms' textual risk disclosures using topic modeling which is able to extract specific risk factors from the disclosures. Using US banks' risk disclosures, we document that banks' risk disclosures provide some useful information pertaining to specific bank risk-taking behaviors. We further detect the significant impact of the textual risk factors on the subsequent stock return after the public release of the financial disclosures. Third, we propose a novel investment strategy using both the semantic structures and the textual risk factors. The result demonstrates that the proposed portfolio strategy based on qualitative information indicators can produce superior risk-adjusted profits. Overall, this dissertation introduces novel approaches to access and quantify fundamental information from qualitative financial disclosures. It contributes to the firm comparability literature by introducing a structural comparability measure, and it also contributes to the firm valuation literature by revealing incomplete information from the unstructured financial disclosures. We conclude that qualitative financial disclosures provide additional information which can better evaluate firms' fundamental values.