\chapter*{Abstract}             % ne pas numéroter
\label{chap-abstract}           % étiquette pour renvois
\phantomsection\addcontentsline{toc}{chapter}{\nameref{chap-abstract}} % inclure dans TdM

% BROUILLON À RÉVISER PAR L'AUTEUR.
\begin{otherlanguage*}{english}
  This dissertation consists of three essays that use high-frequency data to
  study return and volatility dynamics in commodities and financial futures
  markets. The common thread is the minute-level identification of the
  mechanisms through which information --- macroeconomic announcements, trader
  positioning, and central bank communications --- is transmitted to prices
  and volatility, mechanisms that daily data cannot disentangle.

  The first essay asks whether the intensity of speculative trading in energy
  futures markets amplifies or dampens the reaction of prices to macroeconomic
  news. Using 5-minute data on six U.S. futures contracts (crude oil, natural
  gas, gold, silver, copper, and palladium) from April 2007 to February 2024,
  standardized surprises from 26 macroeconomic announcements, and a
  time-varying measure of speculative intensity built from disaggregated CFTC
  positions, the essay shows that higher speculative activity \emph{dampens}
  the response of returns and volatility to surprises and narrows bid-ask
  spreads. These stabilizing effects --- stronger for procyclical commodities
  than for gold, a safe haven --- are driven by money managers rather than
  swap dealers, suggesting that speculative trading improves liquidity and
  price discovery.

  The second essay builds a novel dataset of minute-level indicative Net Asset
  Value (iNAV) observations for four commodity exchange-traded funds (ETFs)
  --- gold, silver, crude oil, and natural gas --- and measures volatility
  transmission between each ETF and its underlying basket. Decomposing
  realized variance into continuous and jump components and combining HAR
  models with Bayesian vector autoregressions, the essay establishes that
  transmission runs primarily through jumps rather than diffusion, that it is
  substantially larger in 1-minute than in 30-minute data, and that its
  direction depends on the arbitrage mechanism: unidirectional from iNAV to
  ETF for physically backed precious metals, bidirectional and asymmetric for
  futures-based energy ETFs.

  The third essay decomposes FOMC statements into two semantic dimensions ---
  policy tone (hawkish/dovish) and informational novelty relative to the
  previous statement --- using an ensemble of language models (MiniLM and
  BERT) trained on Federal Reserve communications with data-driven, PCA-based
  reference selection. Across 148 FOMC announcements (2008--2025) and 1-minute
  data for seven futures contracts, the essay shows that tone predicts
  directional returns while novelty predicts volatility, with the
  tone--novelty interaction on VIX being the most robust effect. The two
  channels display opposite dynamics --- rapid uncertainty resolution for
  volatility, gradual two-hour repricing for returns --- a pattern
  inconsistent with a single information channel.

  Together, the essays show that intraday granularity is not a technical
  refinement but a condition for identification: the magnitude, direction, and
  even sign of the documented effects would be invisible or biased at the
  daily frequency. The results inform policy debates on speculative position
  limits, index-product design, and central bank communication strategy by
  identifying who transmits information to prices, through which channel, and
  at what horizon.
\end{otherlanguage*}