The CAPM is rejected empirically because market beta is only weakly related to average returns. I show that the model's performance depends on where institutional investors concentrate their attention, not how much attention they pay. Combining firm-specific news from the Dow Jones Newswire — filtered with a large language model to strip out market-wide content — with Bloomberg institutional attention data, I find that when attention is concentrated on public firm-specific news, beta prices the cross-section and the security market line is steep and positive, on roughly 46% of trading days, even though aggregate attention is below average. A noisy rational-expectations model rationalizes this: because firm-specific news is informationally orthogonal to the aggregate factor, attention to it crowds out factor learning, raising posterior factor uncertainty and the equilibrium risk premium, and steepening the SML.

The empirical security market line is famously too flat: high-beta stocks earn far less than the CAPM predicts. I show that this flatness is tied to the information environment. When aggregate idiosyncratic news intensity is low, the cross-sectional relation between beta and average excess return is strongly positive and close to the theoretical benchmark (Panel B below); unconditionally, it nearly disappears (Panel A). Consistent with this, the estimated market premium comoves with idiosyncratic news intensity through time.

Presented at: AFFI 2026 · Future Finance Festival · Global Finance Conference 2026 · EFMA 2026 · FMARC 2026