The Division of Science and Technology (DST) welcomed two guest speakers on 4 December. The first speaker was Dr Wang Dan, who is an Assistant Professor of Finance at New York University Shanghai (NYU Shanghai) and also a Global Network Assistant Professor, New York University. Following her was Research Fellow at the School of Data Science at Fudan University, Dr Chen Zhao.

Wang Dan 1

Dr Wang Dan

Chen Zhao 1

Dr Chen Zhao

Dr Wang’s talk was titled ‘Security Selection Based on High-Frequency Sharpe Ratio’. She explained that in portfolio allocation, the classical Mean-Variance model in Markowitz (1952) relies heavily on the covariance structure among assets. As the number and types of assets increase rapidly, traditional methods to estimate the covariance matrix and its inverse suffer from the common issues in high-dimensional analysis.

She explained that to avoid the issue of estimating the covariance matrix with ultra-high dimensional data, she proposed a procedure to reduce dimension based on a new risk/return measure constructed from intra-day high-frequency data and select assets via Sure Explained Variability and Independence Screening (SEVIS).

Wang Dan 2

Dr Wang stressed that with the assets selected through SEVIS, we will build a portfolio that earns more excess return compared with several existing portfolio allocation methods. She illustrated this advantage of the asset selection method with real data from the stock market.

Following Dr Wang was Dr Chen Zhao, who gave a talk that was titled ‘Ultrahigh Dimensional Precision Matrix Estimation via Refitted Cross-Validation’. He talked about how his paper has developed a new estimation procedure for ultrahigh dimensional sparse precision matrix, the inverse of the covariance matrix.

He explained how regularisation methods have been proposed for sparse precision matrix estimation, but they may not perform well with ultrahigh dimensional data due to spurious correlation. He proposed a refitted cross-validation (RCV) method for sparse precision matrix estimation based on its Cholesky decomposition. The proposed RCV procedure can be easily implemented with existing software for ultrahigh dimensional linear regression.

Chen Zhao 2

Dr Chen mentioned how they had established the consistency of the proposed RCV estimate and show that the rate of convergence of the RCV estimate without assuming banded structure is the same as those assuming the banded structure in Bickel and Levina (2008b). Monte Carlo studies were conducted to access the finite sample performance of the RCV estimate. His numerical comparison shows that the RCV estimate can outperform existing ones in various scenarios and he further applied the RCV estimate for an empirical analysis of asset allocation.

Following each talk was a question and answer session where various faculty members expressed interest in each speaker's talk.

Reporter/Photographer: Samuel Burgess
Editors: Deen He
(from MPRO with thanks to the ELC)