Chapter 2· Submission version, Journal of Futures Markets
Seeing Through the ETF: Indicative NAV and Commodity Volatility Transmission
Intraday iNAV Reveals Jump-Driven ETF Volatility Transmission
How does volatility flow between commodity ETFs and their underlyings? The intraday indicative NAV shows transmission runs through jumps, not diffusion.
Simon-Pierre Boucher — contact@spboucher.ai
≈45M
Tick observations underlying the minute-level iNAV dataset, 2010–2023
4
Single-commodity ETFs studied (GLD, SLV, USO, UNG)
2×
1-minute transmission estimates up to twice as large as 30-minute ones
3
Sampling frequencies compared: 1, 5, and 30 minutes
13
Tables and 8 figures (realized volatility and IRF plots)
Abstract
This essay asks how volatility flows between commodity ETFs and their underlying assets, and what the intraday indicative NAV (iNAV) reveals that daily data cannot. It builds a novel minute-level iNAV dataset for four single-commodity ETFs — gold (GLD), silver (SLV), oil (USO), and natural gas (UNG) — from roughly 45 million tick observations spanning 2010 to 2023.
Realized variance is decomposed into continuous and jump components and modelled with HAR-X and HAR-CJ-X regressions at 1, 5, and 30 minutes, plus a Bayesian VAR. The iNAV yields a sharper image of the ETF–underlying volatility relationship: transmission runs primarily through jumps, sampling frequency matters (1-minute estimates are up to twice as large as 30-minute ones), and precious metals show unidirectional transmission while energy is bidirectional and asymmetric.
Key Results
Novel minute-level iNAV dataset
Constructs an original minute-frequency indicative NAV dataset for four single-commodity ETFs from roughly 45 million tick observations covering 2010 to 2023.
Sharper view of ETF–underlying dynamics
Shows the intraday iNAV gives a sharper image of the volatility relationship between ETFs and their underlying assets than daily data can provide.
Jumps as the transmission channel
Decomposing realized variance shows volatility transmission runs primarily through jump components rather than the continuous diffusion component.
Sampling frequency matters for inference
Comparing 1-, 5-, and 30-minute estimates shows 1-minute transmission coefficients can be up to twice as large as 30-minute ones.
Asset-class contrast in transmission direction
Precious metals show unidirectional iNAV-to-ETF transmission consistent with passive arbitrage, while energy ETFs display bidirectional and asymmetric volatility flows.
Data
Minute-level iNAV series, 2010–2023
A novel intraday indicative NAV dataset for four single-commodity ETFs, built from approximately 45 million tick observations.
Commodity ETF prices: GLD, SLV, USO, UNG
High-frequency prices for the gold, silver, crude oil, and natural gas ETFs, matched to their underlying assets at 1-, 5-, and 30-minute frequencies.
Methodology
HAR-X and HAR-CJ-X models
Heterogeneous autoregressive realized-volatility regressions with cross-market terms, estimated at 1-, 5-, and 30-minute sampling frequencies.
Jump decomposition (Barndorff-Nielsen–Shephard)
Realized variance is decomposed into continuous and jump components to identify which channel carries volatility between ETFs and underlyings.
Minnesota-prior Bayesian VAR
A Bayesian VAR with Minnesota prior traces impulse responses and the direction and asymmetry of volatility transmission between markets.
Reproducibility
- Full LaTeX source in phd_chap2_20260731/ is organized as main.tex plus a sections/ folder (introduction, data, methods, results, conclusion), with 13 tables, 8 figures, and master.bib.
- The folder is a frozen July 31, 2026 submission snapshot for the Journal of Futures Markets and includes the compiled main.pdf.
- The thesis-adapted version lives in these-ulaval/chapitre2/, with labels prefixed ch2: and references merged into the consolidated thesis bibliography.