A B S T R A C T
The prediction reliability is of primary concern in many clinical studies when the objective is to develop new predictive models or improve existing risk scores. In fact, before using a model in any clinical decision making, it is very important to check its ability to discriminate between subjects who are at risk of, for example, developing certain disease in a near future from those who will not. To that end, the time-dependent receiver operating characteristic curve (ROC) is the most commonly used method in practice. Several approaches have been proposed in the literature to estimate the ROC non-parametrically in the context of survival data. But, except one recent approach, all the existing methods provide a non-smooth ROC estimator whereas, by definition, the ROC curve is smooth. In this article we propose and study a new non-parametric smooth ROC estimator based on a weighted kernel smoother. More precisely, our approach relies on a well-known kernel method used to estimate cumulative distribution functions of random variables with bounded supports. We derived some asymptotic properties for the proposed estimator. As bandwidth is the main parameter to be set, we present and study different methods to appropriately select one. A simulation study is conducted, under different scenarios, to prove the consistency of the proposed method and to compare its finite sample performance with a competitor. The results show that the proposed method performs better and appear to be quite robust to bandwidth choice. Furthermore, we illustrate the method using a real data example.

"Machine Learning for Mixed Frequency Data" by Jonas Striaukas, LFIN

A B S T R A C T
In this article, we propose a machine learning method for high-dimensional time-series data that are sampled at different frequencies. We extend the commonly used MIDAS (mixed frequency data sampling) regression model to the high-dimensional setting and study the estimator that builds on the sparse-group LASSO algorithm introduced in Simon et al. (2013). We establish non-asymptotic and asymptotic properties of the estimator with dependent data under mild mixing conditions. Our empirical application to nowcasting the US GDP growth shows that the estimator compares favorably to other alternatives.

"Tail dependence in a Hüsler-Reiss Markov tree" by Stefka Kirilova Asenova, ISBA

A B S T R A C T
Tail dependence modeling in multivariate extremes involves high dimensional functions which often become intractable analytically and numerically as well. Conditional independence models also called graphical models, because of their representability as a graph, offer the possibility of dimension reduction. A particular case are tree graphical models where joint distributions can be factorized in bivariate functions. We present a so called Hüsler-Reiss Markov (HRM) tree, a graphical model suitable for studying tail dependence of a graphical model on a tree. Practical interest in such structures may arise when dealing with variables measured on a river network, such as water level, flow or chemical concentration. The model is a special case of a regularly varying Markov tree in Segers (2019). The logarithm of the excesses over a threshold of a HRM tree have multivariate normal distribution as limiting distribution, conditional on one of the variables exceeding a high threshold. The limiting distribution also represents a Gaussian Graphical Model. We treat the case of latent variables in the tree, which is a frequent situation in practice. We apply the Hu¨sler–Reiss Markov tree to the real data of Seine river, France with the scope to study the goodness of fit of the HRM model as well as to study the extremal dependence between key locations throughout the Seine network.