Forecasting using a large number of predictors: Is Bayesian regression a valid alternative to principal components?

By Christine De Mol, Domenico Giannone, L. Reichlin. 2008

140 citations recorded by the source corpus.

Subfields: Management Science and Operations Research, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/6ce8b68e36ff5f1137e35c4d15dc2d714581f162

Citations in the corpus

• A network analysis of the volatility of high dimensional financial series (2017, unknown): “It has been shown (see for example De Mol et al. (2008)) that forecasts that are obtained via penalized regression are highly unstable in the presence of collinearity.”
• Global commodity cycles and linkages: a FAVAR approach (2011, unknown): “A recent approach that has been found to be effective especially for forecasting purposes is the so-called Bayesian shrinkage (De Mol et al. 2008; Banbura et al. 2010; Carriero et al. 2011).”
• Hierarchical shrinkage priors for dynamic regressions with many predictors (2011, unknown): “Similarly, De Mol, Giannone, and Reichlin (2008) compare in one integrated setting the shrinkage and model selection properties of Bayesian LASSO and ridge regression estimators as opposed to principal component shrinkage.”
• Generalized Shrinkage Methods for Forecasting Using Many Predictors (2012, unknown): “This conclusion complements Bańbura, Giannone, and Reichlin (2010) and De Mol, Giannone, and Reichlin (2008), who reached a similar conclusion concerning many-predictor models specified in terms of the original variables instead of the factors.”
• Monetary Policy and Bank Profitability in a Low Interest Rate Environment (2018, unknown): “…compounded by the                                                              20 For the estimation of the VAR, we address the high-dimensional data problem (eleven variables, five lags, and a quarterly sample starting in 1999:Q1) by using Bayesian shrinkage, as suggested in De Mol et al. (2008).”
• Sparse and Stable Portfolio Selection With Parameter Uncertainty (2015, unknown): “L1 norm penalized regressions have been popular in statistics (e.g., Tibshirani 1996) and econometrics (e.g., Bai and Ng 2008; De Mol, Giannone, and Reichlin 2008).”
• Real-Time Inflation Forecasting in a Changing World (2009, unknown): “We choose λ = 10 in (15), as De Mol, Giannone, and Reichlin (2008) showed that the degree of shrinkage should be proportional to the number of regressors to achieve the best forecasting performance in a data-rich context.”
• Time Varying Dimension Models (2012, unknown): “See, among many others, Banbura, Giannone, and Reichlin ( 2010 ), De Mol, Giannone, and Reichlin ( 2008 ), George, Sun, and Ni ( 2008 ), Korobilis ( 2011 ), Koop, Leon-Gonzalez, and Strachan ( 2009 ), and Groen, Paap, and Ravazzolo ( 2010 ).”
• Testing for Granger Causality in Large Mixed-Frequency VARs (2016, unknown): “Hence, in contrast to De Mol et al. (2008), we can already assume that, for a given sample size, λ needs to grow with the size of the system in order for the data to have any influence on the estimates also when m is large.”
• Real-Time Forecasting With a Mixed-Frequency VAR (2015, unknown): “…assessment and Giannone, Lenza, and Primiceri 2014 for a recent assessment) and there is strong evidence that they perform well in high-dimensional environments if estimated with shrinkage estimation techniques (see, e.g., De Mol, Giannone, and Reichlin 2008; Banbura, Giannone, and Reichlin 2010).”
• Large Bayesian vector auto regressions (2010, unknown): “A comparison of forecasts based, alternatively, on Bayesian regression and principal components regression has recently been performed by De Mol et al. (2008) and Giacomini and White (2006).”
• Now-Casting and the Real-time Data Flow (2013, unknown): “…if data are collinear, there is no major difference in the forecasting performance of models based on selected or all available predictors (see De Mol, Giannone, and Reichlin, 2008).14 Finally and most importantly, because of collinearity among predictors, variable selection is inherently…”
• Regularized estimation in sparse high-dimensional time series models (2013, unknown): “In a recent work De Mol et al. (2008) have shown that lasso and ridge penalized estimates often provide forecasting accuracy that is comparable to principal component regressions.”
• Non-Standard Monetary Policy Measures and Monetary Developments (2010, unknown): “The so-called ‘curse of dimensionality’ is addressed using Bayesian shrinkage methods, as suggested in De Mol et al. (2008) and Banbura et al. (2010).”
• Theory and Practice of GVAR Modeling (2016, unknown): “Large scale Bayesian VARs have been explored, among others, by Giacomini and White (2006), De Mol, Giannone, and Reichlin (2008), Carriero, Kapetanios, and Marcellino (2009), and Banbura, Giannone, and Reichlin (2010).”