Wenbo Wu

Identifying important features associated with right-censored survival time in ultrahigh-dimensional survival data is a challenging task due to the curse of dimensionality and information loss caused by censoring. To address these challenges, we propose a two-stage feature screening framework consisting of an imputation step and a feature screening step. The use of Buckley–James imputation leverages information from censoring and can therefore enhance the overall screening performance, particularly when the censoring rate is relatively high. We establish the sure screening properties of the two screening procedures proposed under this framework and illustrate their advantages through simulations. A real-world example is also provided to demonstrate the practical usefulness of the proposed approach.

In the context of life sciences, the rapid biotechnical development leads to the creation of huge amounts of biological data. The use of such data naturally brings concerns on human genetic privacy breaches, which also discourage biological data sharing. Prior studies have investigated the possibility of the privacy issues associated with individuals’ trait data. However, there are few studies on quantitatively analyzing the probability of the privacy risk. In this paper, we fill this void by proposing a scheme for systematically breaching genomic privacy, which is centered on quantifying the probability of the privacy risk of continuous trait data. With well-designed synthetic datasets, our theoretical analysis and experiments lead to several important findings, such as: (i) The size of genetic signatures and the sensitivity (true positive rate) significantly affect the accuracy of re-identification attack. (ii) Both the size of genetic signatures and the minor allele frequency have a significant impact on distinguishing true positive and false positive matching between traits and genetic profiles. (iii) The size of the matching quantitative trait locus dataset has a large impact on the confidence of the privacy risk assessment. Validation with a real dataset shows that our findings can effectively estimate the privacy risks of the continuous trait dataset.

We apply four machine learning methods to cross-sectional return prediction for hedge fund selection. We equip the forecast model with a set of idiosyncratic features, which are derived from historical returns of a hedge fund and capture a variety of fund-specific information. Evaluating the out-of-sample performance, we find that our forecast method significantly outperforms the four styled Hedge Fund Research indices in almost all situations. Among the four machine learning methods, we find that deep neural network appears to be overall most effective. Investigating the source of methodological advantage of our method using a case study, we find that cross-sectional forecast outperforms forecast based on time series regression in most cases. Advanced modeling capabilities of machine learning further enhance these advantages. We find that the return-based features lead to higher returns than the benchmark of a set of macroderivative features, and our forecast method yields best performance when the two sets of features are combined. This paper was accepted by David Simchi‐Levi, finance.

Estimation of a general multi-index model comprises determining the number of linear combinations of predictors (structural dimension) that are related to the response, estimating the loadings of each index vector, selecting the active predictors and estimating the underlying link function. These objectives are often achieved sequentially at different stages of the estimation process. In this study, we propose a unified estimation approach under a semi-parametric model framework to attain these estimation goals simultaneously. The proposed estimation method is more efficient and stable than many existing methods where the estimation error in the structural dimension may propagate to the estimation of the index vectors and variable selection stages. A detailed algorithm is provided to implement the proposed method. Comprehensive simulations and a real data analysis illustrate the effectiveness of the proposed method.

Often the research interest in causal inference is on the regression causal effect, which is the mean difference in the potential outcomes conditional on the covariates. In this paper, we use sufficient dimension reduction to estimate a lower dimensional linear combination of the covariates that is sufficient to model the regression causal effect. Compared with the existing applications of sufficient dimension reduction in causal inference, our approaches are more efficient in reducing the dimensionality of covariates, and avoid estimating the individual outcome regressions. The proposed approaches can be used in three ways to assist modeling the regression causal effect: to conduct variable selection, to improve the estimation accuracy, and to detect the heterogeneity. Their usefulness are illustrated by both simulation studies and a real data example.