Using Unstructured and Qualitative Disclosures to Explain Accruals

We use MD&A disclosures to predict current-year firm-level accruals using support-vector regressions. We call these predictions big-data accruals. Our aim is to measure the explanatory power of MD&A disclosures for liquidity and critical accounting choices. We find that big-data accruals explain a statistically and economically significant portion of firm-level accruals and identify more persistent accruals. They have less incremental explanatory power when 10-K readability is low and more when earnings are more difficult to predict using fundamentals. In addition, we find that big data accruals are incrementally useful in predicting next period's cash flows. We apply our technique to conference calls and find that they have similar explanatory power for accruals. Our technique can be applied to a variety of unstructured and qualitative disclosures to assess their narrative content