Research

How does earnings news get priced into stock returns? I use a demand system approach to show that this passthrough depends on investor responses to both earnings and prices and that these sensitivities are heterogeneous across investors. A key identification challenge is that earnings news is rapidly incorporated into prices; as a result, it is difficult to distinguish whether investors react to the earnings news itself or the concurrent price change. Using a two-step procedure to isolate price from earnings responses, I identify an average asset-weighted earnings elasticity of 3, i.e. for a stock that beats earnings expectations by 1%, the average investor would increase his number of shares held by 3% if prices were held fixed. These estimates vary across sectors, with most institutional investors more earnings elastic and price inelastic compared to the residual ("household") sector. The stock-level sensitivities implied by their ownership account for heterogeneous earnings passthroughs, as stocks with higher earnings sensitivity and lower price sensitivity see larger return responses from the same earnings surprise. Extremes of price and earnings elasticities are also closely related to misreaction: a strategy that bets on subsequent reversal (momentum) in sensitive (insensitive) stocks in response to earnings news generates significant outperformance and alpha. These findings suggest that the pricing of earnings news is closely related to the ownership structure of stocks.

We explore the cross-sectional predictability of stock returns with lagged past returns through the lens of a hypothetical Bayesian researcher who begins with an initial prior that is neutral, showing no bias toward momentum, reversal, or other predictable patterns. This researcher considers a wide range of monthly lags as potential return predictors. By applying Gaussian process regression, which flexibly allows expected returns to depend on lagged returns, and using empirical Bayes shrinkage to guard against spurious anomaly discovery due to multiple testing and against mistaking the ex-post visible effects of investor learning as ex-ante expected returns, this researcher would have discovered prominent return-based anomalies—such as momentum and long-term reversal—well before the authors of the published studies analyzed the data. This suggests that these anomalies represented properties of ex-ante expected returns at the time of their academic discovery. However, tracking these anomalies in real-time, with posterior beliefs based on optimally weighted historical data, reveals that the ex-ante expected returns of many prominent anomalies diminish significantly around their publication dates. Based on these findings, there is little justification for viewing momentum, long-term reversal effects, or other return-based anomalies as permanent features of the cross-section of expected stock returns.

How do inflation expectations affect stock returns, and what accounts for this relationship? We directly measure investors' expectations using traded inflation-indexed contracts and show that, post-2000, stocks offer positive returns in response to higher expected inflation: unconditionally, a 10 basis point increase in 10-year breakeven inflation is associated with a 1.1% increase in the value-weighted stock index. Using a wide range of approaches, we show that this positive relationship is almost entirely due to aggregate variations in expected excess returns rather than changes in firm cash flows (e.g., due to higher mark-ups) or fluctuations in risk-free rates (e.g., due to expected monetary policy response). Overall, a risk premium "proxy" mechanism appears to explain this dominant role of expected excess returns: higher long-term inflation expectations signal stronger future economic growth and reduced volatility.

We use a New Keynesian model with an effective lower bound (ELB) and a general stochastic process for the natural rate to study optimal monetary policy. The central bank has perfect commitment and an interest rate smoothing term in its loss function. Despite the ELB binding occasionally and endogenously, we can derive a closed-form solution for the optimal interest rate: it is the maximum of zero and a weighted average of all past realizations of the output gap. This implies that the optimal interest rate (i) takes a simple form, (ii) is path dependent at all times, (iii) should be pre-emptively lowered when close to the ELB –or kept at zero if at the ELB– if and only if the weighted average of past output gaps is negative, and (iv) behaves very differently from the Taylor rule. We illustrate these insights by solving for key variables in the New Keynesian model using a neural network.