Using unstructured and qualitative disclosures to explain accruals

We examine the usefulness of support vector regressions (SVRs) in assessing the content of unstructured, qualitative disclosures by relating MD&A-based SVR-accrual estimates (MD&A accruals) to actual accruals. We find that MD&A accruals explain a statistically and economically significant portion of firm-level accruals and identify more persistent accruals. We find that the explanatory power of MD&A accruals is higher for more readable 10-Ks, thereby providing evidence for the construct validity of the readability measures. To highlight the flexibility of the SVR method, we apply it to other dependent variables and disclosures. We find that MD&A-based cash-flow forecasts produced by SVR predict next period's cash flows. We apply SVR to conference call transcripts and find accruals estimates have similar explanatory power to MD&A accruals. Finally, the explanatory power of MD&A accruals increases between 1994 and 2013.