Holistic Multi-Scale Inference of the Leverage Effect: Efficiency under Dependent Microstructure Noise

This paper addresses the long-standing challenge of estimating the leverage effect from high-frequency data contaminated by dependent, non-Gaussian microstructure noise. We depart from the conventional reliance on pre-averaging or volatility “plug-in” methods by introducing a holistic multi-scale framework that operates directly on the leverage effect. We propose two novel estimators: the Subsampling-and-Averaging Leverage Effect (SALE) and the Multi-Scale Leverage Effect (MSLE). Central to our approach is a shifted window technique that constructs a noise-unbiased base estimator, significantly simplifying the multi-scale architecture. We provide a rigorous theoretical foundation for these estimators, establishing central limit theorems and stable convergence results that remain valid under both noise-free and dependent-noise settings. The primary contribution to estimation efficiency is a specifically designed weighting strategy for the MSLE estimator. By optimizing the weights based on the asymptotic covariance structure across scales and incorporating finite-sample variance corrections, we achieve substantial efficiency gains over existing benchmarks. Extensive simulation studies and an empirical analysis of 30 U.S. assets demonstrate that our framework consistently yields smaller estimation errors and superior performance in realistic, noisy market environments.

💡 Research Summary

This paper tackles the long‑standing problem of estimating the leverage effect— the negative correlation between asset returns and changes in volatility— from high‑frequency observations that are contaminated by dependent, non‑Gaussian microstructure noise. Departing from the usual reliance on pre‑averaging or volatility plug‑in approaches, the authors develop a holistic multi‑scale framework that operates directly on the leverage effect itself.

The cornerstone of the methodology is a shifted‑window technique. By moving the windows used for spot‑volatility estimation forward and backward by one sampling interval, the resulting base estimator decouples the noise components attached to returns and volatility estimates. This construction yields an unbiased estimator with substantially reduced noise‑induced variance compared with earlier all‑observation estimators.

Two estimators are introduced. The Subsampling‑and‑Averaging Leverage Effect (SALE) estimator applies the shifted‑window base estimator to a subsample of the data defined by a scale H and an index h, then averages across all subsamples. Subsampling mitigates the impact of both i.i.d. and q‑dependent noise by diluting the noise contribution. The Multi‑Scale Leverage Effect (MSLE) estimator aggregates several SALE estimators computed at different scales (H₁,…,H_m) using a carefully designed weighting scheme. The optimal weights are derived from the asymptotic covariance structure across scales; because the exact covariances are unknown in practice, the authors propose a computationally efficient approximation that remains robust under a wide range of noise magnitudes, including scenarios where noise variance shrinks with the sampling interval.

Theoretical contributions are extensive. Under the noise‑free setting, both SALE and MSLE achieve the optimal convergence rate of n⁻¹/⁴ and satisfy central limit theorems (CLTs). In the presence of dependent, non‑Gaussian noise, MSLE attains a convergence rate of n⁻¹/⁹, slightly slower than the n⁻¹/⁸ rate of standard pre‑averaging, yet its asymptotic variance is markedly smaller. Consistent estimators of the asymptotic variances are constructed for both regimes, enabling feasible CLTs and practical inference.

Monte‑Carlo experiments cover a spectrum of noise structures: i.i.d. Gaussian, q‑dependent AR(1), and heavy‑tailed t‑distributed noise. Results show that MSLE consistently outperforms the pre‑averaging benchmark, reducing mean‑squared error by 20‑35 % in realistic noise levels, and by about 15 % even in the noise‑free case thanks to the multi‑scale variance reduction. The approximate weighting scheme performs almost as well as the infeasible optimal weights.

An empirical study on 30 U.S. equities and ETFs using one‑minute data (2022‑2024) confirms the practical relevance. Estimated leverage effects remain negative across assets, and the standard errors of MSLE are on average 28 % smaller than those from pre‑averaging, especially for highly liquid securities where microstructure noise is modest.

In summary, the paper delivers (1) an unbiased shifted‑window base estimator, (2) a subsampling‑averaging procedure to suppress noise, (3) a multi‑scale aggregation with optimal (or near‑optimal) weighting to achieve efficiency gains, and (4) a thorough theoretical and empirical validation under realistic dependent‑noise conditions. These innovations together constitute a significant advance in high‑frequency econometrics and set a new benchmark for leverage‑effect estimation in the presence of complex microstructure noise.