Welcome

Abstract

Prior research replaces investor cash flow expectations with analyst forecasts to derive implied cost of capital metrics (ICCs). These ICCs are known to track expected returns poorly. I develop a parsimonious Bayesian model that estimates investors’ earnings expectations directly, filtering noisy analyst signals toward a fundamental prior when the signal’s uncertainty is high. The empirical proxy for this uncertainty – the squared common surprise (SCS) – identifies firms that are currently distressed but realize the largest improvements in profitability over the following year. The model anticipates this earnings reversion, departing most from analyst forecasts precisely for high-uncertainty firms and, notably, when analysts agree the most. Among low-dispersion firms (where forecasts appear most reliable), analyst-based ICCs are nearly flat across SCS quintiles, assigning almost the same cost of capital to low- and high-SCS firms whose average realized future returns differ by 19 percentage points; ICCs built from my model’s earnings instead project this spread far more closely (11% versus 0% for analyst ICCs). In predictability regressions, the model’s ICCs further deliver slopes near one and significantly higher explained variation. Moreover, three of five ICCs derived from the model’s earnings are reliable under common measurement-error criteria, and – contrary to prior evidence – the composite ICC is reliable in both cross-sectional and time-series designs.