A Sparse GARCH-MIDAS Approach with Minimax Concave Penalty for Crude Oil Volatility

Abstract Sparse GARCH-MIDAS specifications used in the volatility forecasting literature typically combine a Lasso-type penalty with a single information criterion, which leaves two issues unad-dressed: uniform shrinkage biases large coefficients, and the retained factor set can be sensitive to the particular data split. We replace the Lasso penalty with the Minimax Concave Penalty (MCP), solved by ADMM, and add a path-level inclusion-frequency check from 5-fold expanding-window cross-validation, keeping only factors that are both HQIC-optimal in sample and stable across folds. Applied to WTI crude oil between February 2010 and November 2025, with 53 candidate factors covering uncertainty and geopolitics, oil-market fundamentals, macro-finance, and market sentiment and stress, the rule retains six core drivers spanning all four categories: the Commodity Markets EMV Tracker, drilling of oil and gas wells (IPN213111S), narrowly defined climate policy uncertainty, fiscal policy uncertainty, the World Sentiment Index , and the real broad dollar index. Our results show that the six-factor model attains the lowest in-sample information criteria, the lowest out-of-sample losses at the daily, weekly, and monthly horizons, and a Mincer–Zarnowitz slope close to the theoretical value of one; a joint Wald test on the six loadings rejects the zero-loading null at the 1% level, and the Diebold– Mariano and Clark–West tests are consistent with this ranking. A dense benchmark chosen by an information criterion alone produces substantially higher out-of-sample losses and a negative out-of-sample R 2 , which shows that the cross-sample stability constraint, not just the penalty, accounts for much of the gain.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-13
DOI
https://doi.org/10.21203/rs.3.rs-10212852/v1
Primary Topic
Financial Risk and Volatility Modeling
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

A Sparse GARCH-MIDAS Approach with Minimax Concave Penalty for Crude Oil Volatility

Xiaogang Dong, Xiwen Qin, Liping Yuan, Hongzhen Jin et al.
1 links
Research Square
Financial Risk and Volatility Modeling
preprint

A Sparse GARCH-MIDAS Approach with Minimax Concave Penalty for Crude Oil Volatility

Xiaogang Dong, Xiwen Qin, Liping Yuan, Hongzhen Jin, Shuo Li
preprint en

Abstract

Abstract Sparse GARCH-MIDAS specifications used in the volatility forecasting literature typically combine a Lasso-type penalty with a single information criterion, which leaves two issues unad-dressed: uniform shrinkage biases large coefficients, and the retained factor set can be sensitive to the particular data split. We replace the Lasso penalty with the Minimax Concave Penalty (MCP), solved by ADMM, and add a path-level inclusion-frequency check from 5-fold expanding-window cross-validation, keeping only factors that are both HQIC-optimal in sample and stable across folds. Applied to WTI crude oil between February 2010 and November 2025, with 53 candidate factors covering uncertainty and geopolitics, oil-market fundamentals, macro-finance, and market sentiment and stress, the rule retains six core drivers spanning all four categories: the Commodity Markets EMV Tracker, drilling of oil and gas wells (IPN213111S), narrowly defined climate policy uncertainty, fiscal policy uncertainty, the World Sentiment Index , and the real broad dollar index. Our results show that the six-factor model attains the lowest in-sample information criteria, the lowest out-of-sample losses at the daily, weekly, and monthly horizons, and a Mincer–Zarnowitz slope close to the theoretical value of one; a joint Wald test on the six loadings rejects the zero-loading null at the 1% level, and the Diebold– Mariano and Clark–West tests are consistent with this ranking. A dense benchmark chosen by an information criterion alone produces substantially higher out-of-sample losses and a negative out-of-sample R 2 , which shows that the cross-sample stability constraint, not just the penalty, accounts for much of the gain.

Research Square
Changchun University of Technology (CN)
Financial Risk and Volatility Modeling
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon