A growing literature, typically using “bags of words” dictionaries, examines the information content of text in financial accounting disclosures. We generate context for our text analysis to help predict effective tax rates using two approaches. First, we create tax-specific, expert-derived, dictionaries and, second, we generate the counts for those bags of words using text taken from tax-related discussions of the Form 10-K, as opposed to its entirety. We find that using expertise provides more information than simply using general accounting and finance dictionaries. In addition, we find that generating general accounting text variable values from tax-related content in the Form 10-K provides statistically significant improvement in model fit. Contrary to more generic accounting and finance word-based text analysis, we find that the signs on our positive and negative tax event dictionaries are different and are consistent with theoretical expectations through each of our modeled time periods.

You do not currently have access to this content.