A standard real business-cycle model with external habit and capital adjustment costs matches a long list of asset price and business-cycle moments: equity, firm value, and risk-free rate volatility; the equity premium; excess return predictability; consumption growth predictability; basic moments of consumption, output, and investment; among others. The model also generates endogenous consumption volatility risk. Precautionary savings motives make consumption sensitive to shocks in bad times, leading to countercyclical volatility, even with homoscedastic technology shocks. Habit acts as countercyclical leverage, which amplifies this channel. Habit also implies high risk aversion, which amplifies the stock price response.

A simple general equilibrium production economy matches moments of the value premium and equity premium. Value firms have low productivity, but will eventually produce high cash flows. The present value of these temporally distant cash flows is especially sensitive to equity premium movements. The value premium is the reward for bearing this sensitivity. Capital adjustment costs are important. Without these costs, value firms would disinvest heavily, leading to high cash flows today, low cash-flow growth going forward, and little exposure to discount rate shocks. Empirical evidence verifies that value firms have higher cash-flow growth and supports other predictions.

We develop an estimator for publication bias-adjusted returns and apply it to 156 published long-short portfolios. Our adjustment uses only in-sample data and provides sharper inferences than out-of-sample tests. Bias-adjusted returns are only 12.3% smaller than in-sample returns with a standard error of 1.7 percentage points. The small bias comes from the dispersion of returns across predictors, which is too large to be explained by data-mined noise. The bias is much smaller than post-publication decay (p-value ¡.0001), suggesting mispricing is important. Our results offer a different perspective about recent papers that find most published predictors are likely false.

A standard real business-cycle model with external habit and capital adjustment costs matches a long list of asset price and business-cycle moments: equity, firm value, and risk-free rate volatility; the equity premium; excess return predictability; consumption growth predictability; basic moments of consumption, output, and investment; among others. The model also generates endogenous consumption volatility risk. Precautionary savings motives make consumption sensitive to shocks in bad times, leading to countercyclical volatility, even with homoscedastic technology shocks. Habit acts as countercyclical leverage, which amplifies this channel. Habit also implies high risk aversion, which amplifies the stock price response. Received April 21, 2016; editorial decision February 3, 2017 by Editor Stijn Van Nieuwerburgh.

Suppose that the 300+ published asset pricing factors are all spurious. How much p ‐hacking is required to produce these factors? If 10,000 researchers generate eight factors every day, it takes hundreds of years. This is because dozens of published t ‐statistics exceed 6.0, while the corresponding p ‐value is infinitesimal, implying an astronomical amount of p ‐hacking in a general model. More structure implies that p ‐hacking cannot address 100 published t ‐statistics that exceed 4.0, as they require an implausibly nonlinear preference for t ‐statistics or even more p ‐hacking. These results imply that mispricing, risk, and/or frictions have a key role in stock returns.

We characterize the structure and origins of missingness for 159 cross-sectional return predictors and study missing value handling for portfolios constructed using machine learning. Simply imputing with cross-sectional means performs well compared to rigorous expectation-maximization methods. This stems from three facts about predictor data: (1) missingness occurs in large blocks organized by time, (2) cross-sectional correlations are small, and (3) missingness tends to occur in blocks organized by the underlying data source. As a result, observed data provide little information about missing data. Sophisticated imputations introduce estimation noise that can lead to underperformance if machine learning is not carefully applied.