考虑非同步交易影响的金融高频协方差阵的估计及应用

Estimation and Application Study on Covariance Matrix of High Frequency Data Considering the Effect of Nonsynchronous Transaction

在大数据时代,高维高频金融数据的协方差阵在投资组合中扮演着重要角色。但当高维资产的交易频率存在显著差异时,非同步交易会导致"Epps"效应,严重影响协方差阵的估计效率。本文将结构矩阵填充模型(SMC)与VAR-LASSO模型相结合,建立SMC-VAR-LASSO模型来估计和预测高维高频数据的协方差阵。该模型将资产按照交易频率的高低分组,避免直接估计和预测所有资产间以及交易频率低的资产间的协方差阵,在解决非同步交易问题的同时,大大降低了数据损失量,提高了协方差阵的估计效率。将SMC-VAR-LASSO模型应用在投资组合中,并将其与VAR-LASSO模型进行比较后,发现:SMC-VAR-LASSO模型在投资组合中的应用效果更好,其提高了投资者的收益和经济效用,降低了风险。

In the era of big data,Covariance of High-dimensional&high frequency financial data plays an important role in the portfolio.But when the transaction frequency of high-dimensional assets is significantly different,asynchronous transaction will lead to"Epps"effect,which seriously affects the estimation efficiency of covariance matrix.In this paper,structured matrix completion(SMC)model and VAR-LASSO model are combined,and the SMC-VAR-LASSO model is established to estimate and predict the covariance matrix of large dimensional high frequency data.The model divides the assets into groups according to the frequency of the transaction,and avoids estimating and predicting of covariance matrix between all assets and low trading assets directly,it not only solves the problem of asynchronous transaction,but also greatly reduces the amount of data loss and improves the estimation efficiency of covariance matrix.We apply the SMC-VAR-LASSO model in portfolio,and compare it with VAR-LASSO model,we find that the SMC-VAR-LASSO model has better performance in the portfolio,which improves the investor’s income and reduces the risk.