Abstract:Correlation between microstructure noise and latent financial logarithmic returns is an empirically relevant phenomenon with sound theoretical justification. With few notable exceptions, all integrated variance estimators proposed in the financial literature are not designed to explicitly handle such a dependence, or handle it only in special settings. We provide an integrated variance estimator that is robust to correlated noise and returns. For this purpose, a generalization of the Forward Filtering Backward Sampling algorithm is proposed, to provide a sampling technique for a latent conditionally Gaussian random sequence. We apply our methodology to intra-day Microsoft prices, and compare it in a simulation study with established alternatives, showing an advantage in terms of root mean square error and dispersion.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to accurately estimate the daily Integrated Variance in the case where there is a correlation between microstructure noise and the underlying financial log - return in financial data. Specifically, most existing Integrated Variance estimators are designed without considering this correlation, or can only handle this correlation under specific conditions. This paper proposes a new Integrated Variance estimator, which is based on the theoretical framework of conditionally Gaussian random sequences and uses an extended Forward Filtering Backward Sampling (FFBS) algorithm to provide a robust estimation method capable of handling the correlation between microstructure noise and the underlying return. ### Background and Motivation - **Background**: In financial markets, high - frequency trading data usually contains microstructure noise, and there may be a correlation between this noise and the underlying financial log - return. This correlation can lead to bias and inconsistency in traditional Integrated Variance estimators. - **Motivation**: Some existing Integrated Variance estimators (such as the estimator proposed by Zhang et al. in 2005) perform well in some cases, but they usually assume that there is no correlation between microstructure noise and the underlying return. However, empirical studies (such as the study by Hansen and Lunde in 2006) show that this correlation does exist and has important economic significance. ### Method - **Theoretical Framework**: The paper adopts the theoretical framework of conditionally Gaussian random sequences, which is a statistical model that can handle the correlation between observations and the underlying state. - **Algorithm**: An extended FFBS algorithm, called G - FFBS (Generalized FFBS), is proposed. This algorithm can perform posterior sampling in more general cases, thus better handling the correlation between microstructure noise and the underlying return. ### Results - **Performance Evaluation**: Through simulation studies and practical applications (such as analyzing the intraday price of Microsoft stocks), the paper shows the advantages of the proposed estimator in terms of root - mean - square error and dispersion. - **Advantages**: Compared with existing estimators, the G - FFBS algorithm shows better robustness and accuracy in handling the correlation between microstructure noise and the underlying return. ### Main Contributions 1. **Algorithm Extension**: Extend from the standard state - space model to conditionally Gaussian random sequences, solving the filtering and smoothing problems in more general cases. 2. **Bayesian Estimator**: Propose a Bayesian Integrated Variance estimator that can handle the correlation between microstructure noise and the underlying return. To the best of the author's knowledge, this is the first Bayesian estimator with these characteristics. ### Formula Summary - **Recursive Equations of Conditionally Gaussian Random Sequences**: \[ \theta_{t + 1}=a_0(t)+a_1(t)\theta_t + b_1(t)\epsilon_1(t + 1)+b_2(t)\epsilon_2(t + 1) \] \[ \xi_{t + 1}=\tilde{A}_0(t)+\tilde{A}_1(t)\theta_{t + 1}+\tilde{B}_1(t)\epsilon_1(t + 1)+\tilde{B}_2(t)\epsilon_2(t + 1) \] - **Key Formulas in the G - FFBS Algorithm**: \[ m(t + 1)=m(t)+\frac{b_1(t)(\tilde{B}_1(t)+b_1(t))+\gamma(t)}{(\tilde{B}_1(t)+b_1(t))^2+\tilde{B}_2^2(t)+\gamma(t)}(\xi_{(t + 1)/T}-m(t)) \] \[ \gamma(t + 1)=\left(\gamma(t)+b_1^2(t)\right)-\frac{(b_1(t)

Conditionally Gaussian Random Sequences for an Integrated Variance Estimator with Correlation between Noise and Returns

Conditional Tail-Related Risk Estimation Using Composite Asymmetric Least Squares and Empirical Likelihood

High-Dimensional Volatility Matrix Estimation with Cross-Sectional Dependent and Heavy-Tailed Microstructural Noise

A Score-Driven Conditional Correlation Model for Noisy and Asynchronous Data: An Application to High-Frequency Covariance Dynamics

A Regularized High-Dimensional Positive Definite Covariance Estimator with High-Frequency Data

Large-Dimensional Positive Definite Covariance Estimation for High Frequency Data via Low-rank and Sparse Matrix Decomposition

Estimating financial risk using piecewise Gaussian processes

Exponential Weighting and Random-Matrix-Theory-Based Filtering of Financial Covariance Matrices for Portfolio Optimization

Bayesian Nonparametric Estimation of Ex-post Variance

Estimating correlations among elliptically distributed random variables under any form of heteroskedasticity

A Bayesian Learning Method for Financial Time-Series Analysis.

Weighted Least Squares Realized Covariation Estimation

On the non-stationarity of financial time series: impact on optimal portfolio selection

A micro-to-macro approach to returns, volumes and waiting times

Vector Autoregressions with Dynamic Factor Coefficients and Conditionally Heteroskedastic Errors

Integrated volatility estimation: the case of observed noise variables

Composite likelihood estimation of stationary Gaussian processes with a view toward stochastic volatility

Estimation of Conditional Covariance Matrices for a Class of High-Dimensional Varying Coefficient Factor-Generalized Autoregressive Conditional Heteroscedasticity Model

Importance sampling for option pricing with feedforward neural networks

Covariance estimation for multivariate conditionally Gaussian dynamic linear models

Semi-parametric Realized Nonlinear Conditional Autoregressive Expectile and Expected Shortfall