Unified Inference for Predictive Mean and Quantile Regressions via Empirical Likelihood
The paper unveils a tuning-free empirical likelihood toolkit to test stock return predictability in mean and quantile regressions across any predictor persistence, allowing heteroskedasticity and an unrestricted intercept. A reverse-regression yields chi-squared limits and avoids overlapping returns at long horizons. Two versions: EL1, more powerful; EL2, sample-splitting and less efficient. Simulations show accurate size and power. U.S. data reveal predictability from inflation, consistent with IVX. A quantile test covers short horizons but needs i.i.d. errors.
What it examines
The paper develops empirical-likelihood tests for mean and quantile return predictability that stay valid whether predictors are stationary, nearly/unit-root, mildly integrated, or mildly explosive. It handles long horizons, conditional heteroskedasticity, and nonzero intercepts, using reverse regression with two implementations: sample splitting and a full-sample two-stage method.
What it concludes
Both EL tests deliver chi-squared limits, accurate size, and good power; the two-stage is more powerful, while sample-splitting is conservative. US data show inflation predicts long-run returns. Uses include asset allocation, risk management, and macro--finance forecasting. Limits: quantile EL needs i.i.d. errors or two-stage. Future work: multivariate predictors, richer heteroskedasticity, and structural change.
Evidence objects
Unified, tuning-free empirical likelihood tests stock returns in mean and quantile regressions, robust to predictor persistence, heteroskedasticity, and intercepts; reverse regression delivers chi-squared limits and sidesteps overlapping pitfalls at horizons.
key_findings bullet 1 · key_findings · validation V0
EL1 projects intercepts using full sample; EL2 removes them by large-lag differencing with sample splitting. Simulations show size and power; EL1 stronger, EL2 halves samples, improving robustness but reducing efficiency.
key_findings bullet 2 · key_findings · validation V0
U.S. data (1952--2025) reveal long-run return predictability from inflation, confirmed by EL1 and IVX. Short-horizon quantile test avoids instruments but requires i.i.d. errors; drawbacks include EL2 inefficiency and EL1 conservatism.
key_findings bullet 3 · key_findings · validation V0
This paper introduces a unified empirical likelihood framework for return predictability that stays valid from stationary to mildly explosive predictors, permits conditional heteroskedasticity, and handles growing long-horizon designs. It jointly covers mean and quantile regressions, relaxes intercept restrictions, and outperforms IVX/Bonferroni theory, with simulations and U.S. evidence underscoring practical impact.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
- … We study an empirical likelihood procedure for testing the predictability of asset returns in the mean and quantile. We propose two-stage and sample-splitting methods in …
Source row: 2097 · abstract type: snippet