Large Non-Stationary Noisy Covariance Matrices: A Cross-Validation Approach

We introduce a novel covariance estimator that exploits the heteroskedastic nature of financial time series by employing exponential weighted moving averages and shrinking in-sample eigenvalues through cross-validation. Our is model-agnostic in we make no assumptions on distribution random entries matrix or structure matrix. Additionally, show how Random Matrix Theory can provide guidance for automatic tuning hyperparameter which characterizes scale dynamics estimator. By attenuating noise from both cross-sectional time-series dimensions, empirically demonstrate superiority our over competing estimators are based exponentially-weighted uniformly-weighted matrices.