Let’s agree to disagree: Mutual fund holdings-based disagreement and the time series of stock returns

We propose a new measure of investor disagreement on the market factor based on US asset allocation fund portfolio data. Compared to disagreement metrics based on, e.g., analyst forecasts, our measure has the advantages of (1) capturing revealed, rather than stated, investor preferences and (2) being invariant to disagreement on stock-specific risks and expressing only the systematic component. Our results show that the dispersion, and not the mean level, of total equity weights among funds with a market timing objective significantly and positively (negatively) predicts excess market returns (Treasury returns) over the next quarter both in and out of sample and contains information that is not spanned by commonly used predictors. In fact, holdings-based disagreement is the best out-of-sample predictor for both the equity risk premium and Treasury returns, and it is the only variable in our sample that can explain the joint dynamics. This is consistent with a model of priced difference-of-opinion risk on equities, for which investors demand compensation, and for which Treasuries are a hedge.