PhD Thesis
Three Essays in High-Frequency Finance
A Université Laval PhD thesis on return and volatility dynamics in commodity and financial futures, built on minute-level data around information events.
3
essays
188
pages
~45M
tick observations (Ch. 2)
148
FOMC events analyzed
271
consolidated references
26
macro announcement types (Ch. 1)
Overview
This thesis-by-articles at Université Laval's Faculty of Business Administration studies how information moves markets when you watch at one-to-five-minute resolution. Three essays, co-authored with Marie-Hélène Gagnon and Gabriel J. Power, share one lens: high-frequency return, volatility and liquidity dynamics around information events in commodity and financial futures markets.
Chapter 1 — revised for The Energy Journal — shows that speculative trading dampens, rather than amplifies, the impact of macroeconomic surprises on energy and metals futures, with money managers (not swap dealers) improving liquidity and price discovery. Chapter 2 builds a novel minute-level indicative NAV dataset for four commodity ETFs and finds that volatility transmission runs primarily through jumps.
Chapter 3 decomposes FOMC statements into policy tone and informational novelty using a dual-model NLP ensemble (MiniLM + BERT), then links them to one-minute futures data: tone predicts directional returns while novelty predicts volatility, with pre-announcement placebos coming up null. The assembled ULaval thesis compiles to 188 pages with zero errors and a consolidated 271-key bibliography.
Key Features
Speculation and macro surprises
5-minute futures data (2007–2024) across crude oil, natural gas, gold, silver, copper and palladium, conditioned on a CFTC-based speculation intensity proxy.
Speculators dampen, not amplify
Higher speculative intensity reduces the impact of macro surprises on price drift, volatility and bid-ask spreads — driven by money managers, not swap dealers.
A novel minute-level iNAV dataset
Roughly 45 million tick observations (2010–2023) for GLD, SLV, USO and UNG give a sharper image of ETF–underlying volatility transmission than daily data can.
Transmission runs through jumps
Barndorff-Nielsen–Shephard decomposition shows volatility flows via jumps, not diffusion; 1-minute estimates run up to 2x larger than 30-minute ones.
NLP on the Fed's own words
217 FOMC statements (2000–2025) decomposed into hawkish/dovish tone and informational novelty via a MiniLM + BERT ensemble with TSDAE+MNRL fine-tuning.
Tone moves returns, novelty moves volatility
A 1-sigma dovish shift builds to roughly +12 bps in equities within two hours; the stance-novelty interaction on VIX persists 5–120 minutes.
Econometrics at full depth
WLS-EWMA and GARCH event studies, HAR-X and HAR-CJ-X models, Minnesota BVARs, minute-level panels, Jorda local projections and five inference methods.
Identification taken seriously
Pre-announcement placebo tests are null across designs, confirming that measured effects are announcement-driven rather than spurious.
Reproducible LaTeX build
The full ulthese/memoir document compiles with latexmk to 0 errors, 0 undefined references and 0 missing citations, with an exhaustive assembly audit.
How It Works
Chapter 1 — Speculative trading in energy markets
Macro announcement surprises interacted with CFTC-based speculation intensity across six futures contracts; 14 tables, 6 figures, COVID and ZLB robustness appendices.
Chapter 2 — iNAV and volatility transmission
Minute-level iNAV construction, realized variance decomposition into continuous and jump components, HAR-X models at three frequencies, and a Bayesian VAR; 13 tables, 8 figures.
Chapter 3 — FOMC tone and novelty
NLP ensemble scoring of statements linked to 1-minute data on seven futures contracts via event regressions, panels and local projections; 24 tables, 19 figures, proofs appendix.
these-ulaval/ — the assembled thesis
French front matter with English abstracts, general introduction and conclusion, appendices, and a merged 271-key bibliography; article sources are frozen snapshots, never edited in place.
Tech Stack
Methods
NLP
Data
Document
Highlights
- Chapter 1 is revised for The Energy Journal; Chapter 2 is at submission stage with the Journal of Futures Markets.
- Precious-metals volatility transmission is unidirectional (iNAV to ETF, passive arbitrage) while energy is bidirectional and asymmetric.
- Policy stance moves realized volatility in 6 of 7 futures contracts at p < 0.01; the key VIX interaction carries t = -5.06.
- Markets covered span energy, metals, equity and rates: CL, NG, GC, SI, HG, PA futures plus ES, VX, ZN, ZF and DX.
- Funded by SSHRC and the Chaire Industrielle-Alliance Groupe financier.
- Each article folder is a frozen dated snapshot; every thesis adaptation is documented in an exhaustive INVENTAIRE.md audit.