Bridging Factor and Sparse Models

Inference in high-dimensional time series models

“…Dynamic Factor Models with Sparse VAR Idiosyncratic Components sense that it performs well regarding both prediction and model selection and in both uncorrelated and highly correlated cases. Furthermore, Fan, Masini, et al (2021) provide hypothesis tests to test whether after removing factors (as well as trends in a first step) the regressors possesses some pre-defined weakly correlated structure or not. Fan, Ke, et al (2020) and Fan, Masini, et al (2021) allow for time-dependent regressors, however they do not consider that the idiosyncratic part follows a sparse vector autoregressive model.…”

“…Furthermore, Fan, Masini, et al (2021) provide hypothesis tests to test whether after removing factors (as well as trends in a first step) the regressors possesses some pre-defined weakly correlated structure or not. Fan, Ke, et al (2020) and Fan, Masini, et al (2021) allow for time-dependent regressors, however they do not consider that the idiosyncratic part follows a sparse vector autoregressive model. Moreover, their assumption about the dependency of the idiosyncratic part excludes sparse vector autoregressive models where the cross-sectional sparsity can grow with the sample size.…”

Essays on Volatility and Returns Predictability

“…This paper carries out a combination of factor and shrinkage models, as a strategy to enhance standard single-step realized variance (RV) estimation procedures. Based on Fan et al (2021)'s FarmPredict three-step estimation approach, we are able to boost a single factor model, initially with a poor performance for a single-step approach, but with a much higher predictive capacity when the multiple-step method is implemented. Moreover, this method improves the performance of well-known RV forecasting methods, as Corsi ( 2009)'s heterogeneous autoregressive (HAR) model.…”

“…In this paper we contribute to the literature on shrinkage and factor models, focusing on the FAVAR. We implement a generalization of the results in Fan et al (2021). Their work assemble high-dimensional models in an unprecedented manner, by combining both factor and shrinkage models, and extracting its potential comparative advantages at different estimation steps.…”

“…Factor Augmented High-Dimension Vector Autoregressive Models: Application to a Panel of Realized Volatilities 38 variances are driven by a limited set of factors, which can be both observed and unobserved (latent). The goal of this exercise is to combine the stepwise method developed by Fan et al (2021) to a multivariate panel prediction model. This strategy allows us to model the volatility panel as a factor model, while assuming that the idiosyncratic error term follows a (weakly) sparse VAR structure.…”