Using MD&A to Improve Earnings Forecasts

Abstract

In this paper, we develop techniques that combine text with financial variables to generate explicit firm-level forecasts. We find that text-enhanced models are more accurate than models using quantitative financial variables alone, providing evidence on the predictive value of the MD&A section. Firms with smaller changes in current performance, larger changes in future performance, negative changes in future performance, higher accruals, greater market capitalization, lower Z-scores, higher audit quality, shorter and more readable MD&A text, and higher incentive-based compensation have more informative MD&As. MD&A is more informative in the period following regulatory reforms but less informative in the period covering the recent financial crisis. Finally, we show that analysts lose their forecasting superiority over text-enhanced statistical models for smaller firms and those with lower analyst following.