\section{Introduction}  
\label{ch1:sec:introduction}

%SETUP: EASIER TO TRADE COMMO FUTURES, IT’S BIG BUSINESS PROFITABLE, IT MATTERS
The appeal of energy and other commodities as an asset class has grown since the Commodity Futures Modernization Act of 2000 (CFMA). By partially deregulating derivatives, the CFMA has made it easier to trade commodity futures contracts for investment purposes. Rather than invest in physicals or in shares of commodity-linked firms, investors can use futures to gain exposure to "commodity beta" \citep{boons2014price}. This evolution is particularly relevant for energy markets, where futures trading volume has grown tremendously. For instance, the total trading volume of commodity derivatives was 137.3 bn contracts in 2023, which is 64\% more than in 2022 (Futures Industry Association, 2024).
An important reason for this trend is that commodities have periodically benefited from bull cycles (most notably in 2004-2008), attracting a growing number of speculators and institutional investors. As a result, the commodities asset class has become an important but volatile revenue source for trading firms and investment banks. %\footnote{Gross margins from commodities trading reached about \$57 bn during the 2007-09 period. While profits were more modest in the following years, they increased once more from \$36 bn in 2018 to \$100 bn in 2021 as a result of increased worldwide demand and supply chain bottlenecks, and up to \$150 bn in 2022. A substantial portion of these profits came from energy trading, particularly during periods of high oil price volatility.}
Three firms, Goldman Sachs, Citi, and Macquarie earned together \$20 bn from commodities trading in 2022, much of it from energy-related contracts.\footnote{Sources: The Financial Times, Bloomberg, S\&P Global and Euronews.}


%ARE FINANCIAL INVESTORS BAD? PRESSURE ON POLICYMAKERS, DESPITE WEAK EVIDENCE
Is the presence of more financial investors harmful to traditional market participants, such as hedgers? While many think so,  the evidence is unclear. The idea that poorly informed investors can disrupt markets has a long history \citep{shleifer1990noise} and has been revived in a recent theoretical literature on financialization  \citep{basak2016model,goldstein2022commodity}. These papers are motivated by the commodity price run-up of 2004-2008, which occurred shortly after the CFMA was passed \citep{domanski2007financial}.  Critics argue that the activities of financial investors, who are not directly involved in producing or processing commodities, can distort prices and increase volatility.\footnote{A high-profile example is \citet{masters2009testimony}, who testified before the U.S. Congress in 2008 and before the CFTC in 2009 about ``Ending excessive speculation in commodity markets.'' While his argument is not supported by empirical evidence, as shown by \citet{irwin2012testing}, the Masters hypothesis reflects beliefs held at the time by many market participants.} These critics claim that energy and commodity markets have become more sensitive to financial market fluctuations, and less to supply and demand fundamentals. 
Energy futures markets have attracted attention due to the popular perception that large price swings affect the real economy (e.g., through higher gasoline and heating costs) \citep{cheng2014financialization}. Research, however, generally does not support this claim \citep{baumeister2014oil}. Whether or not these fears are justified, policymakers have taken notice and the CFTC has progressively implemented rule changes such as new position limits.%\footnote{In 2009 and 2010, the CFTC held public hearings to review the impact of speculative trading on market integrity. In 2010, the CFTC proposed new regulations to impose position limits on commodity derivatives (e.g., number of contracts held by any single trader). The ``Final rule on position limits'' was only adopted on March 15, 2021.}


%COMMODITY BULL CYCLE WHO IS TO BLAME, NOT CLEAR

A large empirical literature debates the causes of periodic price and volatility run-ups in commodity and energy markets. Researchers emphasize the importance of differentiating between speculative traders and passive investors (e.g., index traders). The latter category of traders is more recent and trades energy and commodity contracts for diversification purposes rather than for speculative profit. While \citet{singleton2014investor} suggests that financial investors may be to blame for higher energy prices, \citet{kilian2014role} use a structural model to show that speculation can be ruled out as a cause of the oil price surge during 2003-2008 -- even though speculative demand played a role in previous oil price spikes. Further evidence against the hypothesis that index traders are responsible for the sharp increase in commodity prices is provided by \citet{irwin2011index} and \citet{irwin2012testing}. In a different strand of the literature, \citet{buyukcsahin2011speculators} use Granger causality tests and daily data to investigate whether speculators increase crude oil futures prices. They find little evidence to support that claim. Also using daily position-level data, \citet{brunetti2016speculators} show that speculators reduce price volatility in commodity and energy markets. Reviewing this early literature, \citet{fattouh2013role} conclude that speculation is unlikely to explain the commodity and energy bull cycle of 2004-2008.


%RECENT FINANCIALIZATION PAPERS THEORY AND EMPIRICAL
Recent research provides new theoretical grounds to establish how trading activity could affect energy and commodity prices \citep{basak2016model,goldstein2022commodity}. The subsequent empirical literature, however, does not reach a consensus. \citet{henderson2015new} use data on commodity-linked notes to show that uninformed trading flows affect commodity prices, but \citet{ready2022order} argue that the economic magnitude of this effect is too small to matter. Other recent papers find instances of futures price overshooting, reversals, and greater noise in markets \citep{da2024financialization}. They also find that commodities seem to display higher correlations with equities and with each other \citep{kang2023financialization}.  


%POSITIONING OUR PAPER AS ABOUT SPECULATIVE TRADING, NOT FINANCIALIZATION (REFEREE 2) %MH J'ai collé un petit paragraphe du haut avec celui-ci. Je pense que ça va et que c'est ensemble. 



%Although financialization has been linked to the trading activities of any non-traditional, financial investors in commodity markets (e.g., hedge funds), a more commonly accepted definition focuses on the activities of passive (index) traders \citep{tang2012index}. In contrast with speculation, which is likely to reflect informed trades, index traders are considered uninformed. Since this paper is motivated by the economic role played by all non-commercial traders, we will refer to speculative trading as a general description of this broad class of market participants.  It is important to distinguish, however, speculators from index or institutional investors. It is the latter who are the more recent financial actors in energy and commodity markets \citep{irwin2011index}. Our paper therefore investigates the impact of speculative trading as well as sub-categories of non-commercial traders.%, rather than the narrower class of index traders.  


%WHAT WE DO AND WHY IT IS INNOVATIV
Thus, our main contribution is to provide sharply identified evidence on the impact of speculative trading on energy (crude oil and natural gas) and metal markets (gold, silver, copper, and palladium), with additional evidence on sub-categories of traders. We focus on speculation rather than financialization, which has been linked to the trading activities of passive (index) traders \citep{tang2012index}. Specifically, our paper investigates the impact of speculative trading of sub-categories of non-commercial traders.  Our rationale is that speculative trading is likely to reflect informed trades, which is our focus, in contrast to index traders who are considered uninformed.   Energy commodities serve as our baseline case due to their economic importance and high trading volumes, while metals offer a useful comparison, particularly as gold is perceived as a safe-haven asset. We use high-frequency (5-minute) data to measure the instantaneous reaction of commodity futures returns, volatility, and bid-ask spreads to the surprise component in macroeconomic announcement releases \citep{andersen2007real, kurov2019price}. The data runs from April 4th, 2007, to February 11th, 2024. Our framework also accounts for the time-varying intensity of speculative trading activity. This study builds on \citet{kilian2011energy}, who find no evidence, at a daily frequency, that energy prices react to macroeconomic announcements. By using intraday data, we can better identify the impact of specific macro surprises. We also avoid a common criticism of event study methods, namely that using daily frequency data may reduce the power of statistical tests and could lead the researcher to misattribute the effect of a specific announcement, as other market events occur the same day \citep*{kothari2007econometrics}. 


%OUR FINDINGS, ONE BY ONE, BRIEFLY
We find evidence of beneficial effects (price stability and market efficiency) from increased trading activity in energy and commodity markets. Our first finding is a damping effect on price reactions: while macro surprises generate a positive abnormal return for good news (and negative for bad news), the magnitude of this reaction is significantly weaker when speculative trading is higher. Second, we find a similar damping effect on volatility reactions. While all surprises (good or bad) generate a volatility increase, this reaction is lessened when the futures market shows more speculative trading. Third, we document lower bid-ask spreads when speculative trading is higher, controlling for the surprise environment. Fourth and last,  these beneficial effects are linked to the trading activities of money managers. In contrast, increased trading by swap dealers appears to have an amplifying effect on reactions to macro surprises. These new insights are made possible by investigating this issue using a new angle, namely their sensitivity to macroeconomic surprises, and with high-frequency data. Our findings have important implications for energy market investment and regulation, and to the broader debate about speculation in energy and commodity markets. 


By investigating a broad range of traders in energy and commodity markets, our findings also extend the work of \citet{brunetti2016speculators}. They find that financial investors, especially money managers and hedge funds, help commodity markets by supplying liquidity, reducing volatility, and generally improving market efficiency. Moreover, our results relate to \citet{cheng2015convective} who show that financial investors, being better informed about markets, contribute to price discovery and liquidity. This is particularly relevant for energy markets, where accurate price discovery is crucial for physical market participants and investors. Thus, speculative traders help markets by distributing and assimilating new information into prices. These insights are valuable given the ongoing energy transition and the importance of efficient price discovery in energy markets. Two papers are probably closest to ours. First, \citet{brunetti2016speculators} who find that hedge funds add liquidity to commodity markets, resulting in more efficient prices and lower volatility. They argue that it is  merchant positions (i.e., hedgers) that are linked to greater volatility, and that the presence of hedge funds allows for faster and more efficient price discovery. Second, using daily data, \citet{kilian2011energy} study how energy prices react to macroeconomic announcements. Our paper extends this work to high frequency data.









%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Background}
% personnellement, je pense qu'on peut enlever le paragraphe en entier. le reste de la section s'applique bien je pense. et on en a parlé dans l'intro de ça.  
%\alert{\subsection{Speculative trading activity in energy and commodity futures markets}
%In the literature, financialization usually refers to the process by which an asset class, such as commodities, attracts significantly more attention from financial investors. As a result, prices and volatility may no longer be determined only by commodity supply and demand \citep{cheng2014financialization}. \citet{brunetti2016speculators} refers to financialization as ``the changing mix of participant positions''. These financial investors include hedge funds, commodity trading firms, swap dealers and index traders, but specifically exclude hedgers. Traders may take commodity futures positions for purposes of speculation, diversification, or factor exposure to commodity beta.  REVISE TO AVOID FINANCIALIZATION.} 


Price discovery in commodity markets occurs mainly in futures markets and is affected by informational frictions around supply and demand. Thus, risk sharing and information discovery represent a potential channel for financial investors to generate distortions in energy and commodity markets \citep{cheng2014financialization}. In their model, \citet{basak2016model}  predict that the increased presence of financial actors can increase commodity futures volatility, as well as correlations between commodity and equity returns. \citet{goldstein2022commodity} also argue that under some conditions, a greater presence of financial investors can be harmful to commodity markets. 
  Theory shows how a change in the participant mix in commodity markets could generate undesirable distortions, but the empirical literature is far from settled. \citet{singleton2014investor} argues that trading activity by financial investors creates informational frictions, leading commodity prices to become more volatile and to diverge from their fundamental values. Using a no-arbitrage argument, however,  \citet*{hamilton2014risk} show that the positions of commodity traders included in index funds cannot be used to achieve excess returns in futures markets.    \citet{ready2022order} show that while index traders do have a positive price impact, it is much too small to explain the apparent price distortions or bull cycles observed since 2004. In addition, financial investors do not have a uniform impact on market liquidity. Investors affect liquidity risk by either providing liquidity to meet the hedging needs of other traders or consuming liquidity when they trade for their own needs \citep{kang2020tale}. Indeed, \citet{brunetti2014commodity} show using data on commodity trader positions that index traders provide insurance against price risk.  

\subsection{Macroeconomic announcements}
Surprises in macroeconomic announcements affect financial markets, whether in stocks \citep*{scholtus2014speed} or in bonds \citep{fleming1997moves}. In a key study, \citet*{balduzzi2001economic} find that 17 public news releases affect bond prices, trading volume, and bid-ask spreads. \citet{karali2014macro} show that energy futures markets exhibit asymmetric responses to macroeconomic news, with significant volatility spillovers between natural gas and crude oil markets. \citet{cao2024us} document a time-varying relationship between U.S. monetary policy and crude oil prices, finding that unexpected oil price increases can push monetary policy from expansionary to restrictive stance. \citet{kang2020economic} further show that after 2004, short-term oil price volatility is driven by industrial production, term spreads, and credit spreads, along with traditional market factors.


The literature on commodity-specific announcements is smaller and less conclusive. \citet*{hollstein2020volatility} look at how different economic variables affect the term structure of commodity futures volatility. They show that speculation and jobs-related macro variables have the largest impact on volatility. \citet{zhu2022oil} further show that stock market anomalies can be explained to some extent by oil price shocks, separately from the effect of other macroeconomic variables and investor sentiment.
While the literature finds a clear impact of macroeconomic announcements on stock and bond prices, there is no clear answer as to whether they affect commodity futures prices, or whether  increased trading by financial participants accentuates these reactions. This issue is especially relevant for energy markets, given their macroeconomic importance. Our research provides new insights by using high-frequency data, expanding the set of announcements, and considering a time-varying measure of speculative trading intensity to capture trading activities for each of the commodities in the sample.










%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Data}

We now present a detailed description of our data. Since this paper relies on several types of data, we describe: i) how to obtain the macroeconomic announcement surprises, ii) the commodity futures data, and iii) how to capture speculative trading activity.%, and iv) a PCA analysis for financial investor activity.

\subsection{Data on macroeconomic announcements}

The macroeconomic announcement release data are obtained from Bloomberg and Refinitiv Eikon. We collect information on 22 announcements that are standard to the literature \citep[see e.g.,][]{andersen2003micro}. Our sample for macroeconomic announcements is matched to our high-frequency data and therefore runs from April 2nd, 2007 to February 11th, 2024. The announcements belong to ten categories: Income, Employment, Industrial Activity, Investment, Consumption, Housing Sector, Government, Net Exports, Inflation, and Forward-looking. Most of the announcements are released on a monthly basis. Table \ref{ch1:tab:stat1} summarizes the announcements and provides more detail such as the number of observations, release frequency, source, unit of measure, and time of release. Bloomberg provides analyst forecasts for all announcements, as well as the actual value of the announcement release. For all announcements except the Consumer Price Index and Initial Jobless Claims releases, a positive surprise will be interpreted by investors as signaling a strong economy \citep*{fleming1997moves}.
%Based on a survey of U.S. announcements and their impacts on financial markets, the only ones that  signal a weaker economy are  \citep*{fleming1997moves}. For the other announcements, a positive surprise will be interpreted by investors as signaling a strong economy.
In addition, we include energy sector-specific announcements published by the U.S. Energy Information Administration. The first is the weekly crude oil storage report, which provides an update on the quantity of crude oil held in storage in the U.S. The second is the weekly natural gas storage report. We do not include OPEC announcements, as they cannot be reliably used in a high-frequency econometric design \citep{10.1257/aer.20190964}.\footnote{There are a few issues with the OPEC announcements: First, they are not released at a specific time. Second, it is impossible to know precisely when a given OPEC announcement was made available to investors. Third, OPEC's influence has weakened since the 1980s.}

It is common practice in this literature to use the standardized surprise of an announcement rather than its realized value to quantify the unexpected component of the release. To calculate surprises, we follow \citet*{balduzzi2001economic}. Let $A_{kt}$ be the realized value (i.e., release) of macroeconomic announcement $k$ at time $t$, and let $E_{kt}$ be the median value of all Bloomberg analyst forecasts for announcement $k$ at time $t$. To standardize the surprise, we divide the raw surprise $(A_{kt} - E_{kt})$ by $\sigma_k$, the sample standard deviation of the surprise for announcement $k$. Thus, equation (\ref{ch1:eqn:SURPRISE}) describes the standardized surprise for announcement $k$ at time $t$:

\begin{equation}\label{ch1:eqn:SURPRISE}
S_{kt} = \frac{A_{kt} - E_{kt}}{\sigma_k}
\end{equation}

The sample period is used to compute $\sigma_k$, as in \citet{balduzzi2001economic} and \citet{kurov2019price}.\footnote{The literature argues that measuring $\sigma_k$ in this way is reasonable because the standardized surprise is not used for forecasting purposes. Using raw surprises is not recommended due to scaling issues, nor is using analyst dispersion for $\sigma_k$ because announcement coverage sometimes involves only a few analysts. For robustness, we also estimate our models using surprises where $\sigma_k$ is computed using only past observations. The main findings are unchanged. In this case, we exclude the first $M$ observations (e.g., $M=10$) to get a reasonable sample size for $\sigma_k$.} Table \ref{ch1:tab:stat2} presents the minimum, 1st quartile, median, mean, 3rd quartile, and maximum of the surprise for each announcement.










\subsection{Commodity futures price data}

For intraday data on commodity futures prices, we use Barchart's API.\footnote{See the \url{https://www.barchart.com/futures} website.} Our dataset for prices contains some of the most economically significant commodity futures contracts traded in the U.S. We use a high-frequency price series that runs from April 2nd, 2007 to February 11, 2024. Among these contracts, crude oil and natural gas are pro-cyclical, while gold and silver behave as safe havens. High-grade copper and palladium are industrial metals used in the manufacturing of consumer products.

For each of the commodities in our sample, price returns $R_t$ are calculated as the log return over a 5-minute period $(\tau=5)$ beginning at time $t$. The database provides the futures contract close price ($p_{t}^{close}$) of each 5-minute period. Thus, $R_t$ is obtained as in equation (\ref{ch1:eqn:RETURN}):

\begin{equation}\label{ch1:eqn:RETURN}
R_t^{t+\tau} = \ln \left( \frac{p_{t+\tau}^{close}}{p_{t}^{close}} \right) = \ln (p_{t+\tau}^{close}) - \ln(p_{t}^{close})
\end{equation}

Descriptive statistics for the 5-minute log returns are presented in Table \ref{ch1:tab:stat4}. The most extreme outlier observations belong to crude oil, while gold has the fewest outliers.\footnote{The main findings are robust to using different window lengths by estimating equation \ref{ch1:eqn:RETURN} using 30-minute returns.}

\subsection{Measures of speculative trading activity and trader categories}
%To measure the impact of financialization, we need a measure that captures the intensity of speculation in commodity markets. 


%We investigate the impact of speculative trading activity increasing relative to productive activity. % We consider proxies designed to capture the intensity of speculation in commodity markets and we use them as indicators of financial investor activity. 
%For instance, swap dealers are not speculators, but they belong to our broad definition of financialization. 
The index of speculative trading is constructed using data in the \emph{Commitment of Traders (CoT) Report} published weekly by the Commodity Futures Trading Commission (CFTC). The data provided by the CFTC includes the number of positions held by different types of participants in commodity markets. The CFTC separates trader types as follows: \textit{Commercials} refer to trader-reported futures positions which the trader claims are used for hedging purposes, while  \textit{Non-Commercials} is obtained by subtracting the total long and short commercial positions from the total open interest.\footnote{The CFTC defines commercial traders as participants in commodity markets who primarily use futures contracts to hedge their business activities (e.g., buying or selling commodities). All traders who are not classified as Commercial are automatically classified as Non-Commercial traders. To obtain the number of long positions held by Non-Commercial traders, we subtract the total long Commercial positions from the total open interest. For the number of short positions held by Non-Commercial traders, we subtract the total short Commercial Positions from the total open interest.} 
%Given that $CoT$  reports are not available in real time, they are not considered to be sources of information that traders could act upon, but rather a way to capture the state of the market. 
We use the following information presented in the $CoT$ report: for a futures contract $i$, the number of long and short positions held by Non-Commercial traders are $SL_i$ and $SS_i$, respectively, while for Commercial traders they are $HL_i$ and $HS_i$.\footnote{In an earlier draft, we also reported results based on two alternative proxies as well as a proxy constructed using principal component analysis. The alternative proxies are Working's $T$ \citep{working1960speculation} and the market share of non-commercial traders  \citep*{buyukcsahin2014speculators}. These results, which are available upon request, are consistent with our main findings and do not change the paper's implications.}

%The first proxy we consider to assess levels of speculative and hedging activity is Working’s $T$ \citep{working1960speculation}. This index compares the activity levels of Non-Commercial commodity futures traders (e.g., speculators) to those of Commercial traders (e.g., hedgers).  Working’s $T$ measures the extent to which speculation exceeds the level required to offset any unbalanced hedging at the market clearing price. This index, denoted $WT_i$, is computed as follows:
%Historically this is true but today more nuanced Typically, Commercial traders take short positions in futures contracts while Non-Commercial traders take long positions.

%\begin{equation} \label{ch1:eqn:Working}
%WT_i = \left\{
%\begin{matrix}
%1 + \frac{SS_i}{HL_i + HS_i} & \mbox{if} & HS_i \ge HL_i \\
%1 + \frac{SL_i}{HL_i + HS_i} & \mbox{if} & HS_i < HL_i
%\end{matrix}
%\right.
%\end{equation}

 %\citet*{buyukcsahin2014speculators} suggest a measure that emphasizes the \emph{market share of Non-Commercial traders (MSCT)}. This ratio is expressed as the sum of the short and long positions of Non-Commercial traders over twice the total open interest in a market:
%%Instead of Working’s $T$,
%
%
%\begin{equation} \label{ch1:eqn:MSCT}
%MSCT_i = \frac{SL_i + SS_i}{2 \times OI_i}
%\end{equation}

The specific measure we use follows \citet{hedegaard2011margins}, who suggests an index of speculative activity computed as the ratio of net long speculative positions over total open interest ($NLS_i$):

\begin{equation} \label{ch1:eqn:NLS}
NLS_i = \frac{SL_i - SS_i}{OI_i}
\end{equation}

In addition to computing $NLS$ using the full sample data, we use disaggregated data from the CFTC to compute the NLS index separately for money manager (MM) and swap dealer (SD) positions, which allows for additional empirical analysis. The data on money manager and swap dealer positions come from Quandl's API.\footnote{See the \url{https://data.nasdaq.com/data/CFTC-commodity-futures-trading-commission-reports} website.}
\emph{Money managers} typically refer to Non-Commercial market participants who are involved in managing funds and investing in commodity futures and options markets \citep{fishe2012identifying}.\footnote{This category is also called ``Managed money.'' The CFTC writes that they are ``registered commodity trading advisor (CTA); a registered commodity pool operator (CPO); or an unregistered fund identified by CFTC.'' There is some overlap between Money managers and hedge funds, but they are distinct.} Their activities are influenced by financial and economic factors related to commodities. Money managers are often considered to be more informed investors because they actively manage portfolios and adjust their positions based on market information and analysis.
% factors such as speculative activity, imperfect information about real economic activity as well as supply, demand, and inventory accumulation in commodity markets \citep{singleton2014}.
\emph{Swap dealers} are considered as Non-Commercial traders by the CFTC. They typically use futures contracts to hedge risk generated by their swap positions. Swap dealers have been studied in relation to index investors, as their positions are distinct from those of other market participants. Their activities are influenced by the need to manage large exposures and to facilitate trading for clients. While money managers regularly take long or short futures positions, swap dealers mainly take long positions \citep{fishe2012identifying}. Hedgers, who we exclude from the analysis, tend to take short positions.
%often leading them to take significant speculative positions in the market.
%and contribute to price discovery and liquidity \citep{brunetti2016speculators}
%\citep{sandrs2016bubbles}


%\subsection{A financialization proxy using principal component analysis}
%
%Descriptive statistics for the three financialization variables are shown in Table \ref{ch1:tab:stat5} and computed separately for each of the six commodities in our sample, based on the number of open positions for a given futures contract. The variables themselves are scale-free, unlike the number of open positions. The MSCT variable fluctuates between 0 and 0.5, while NLS varies between -0.4 and 0.8, and Working's $T$ between 1 and 2.  Since the literature does not show that one is superior to the others, and for brevity’s sake, we use Principal Component Analysis to construct a new proxy using information from the three variables. This step allows us to present results based on a single variable.\footnote{In an earlier draft, we showed results using the three different variables and found that our main findings were similar.}
%Our approach is as follows: First, since the variables have different scales, we standardize them to have a mean of zero and a variance of one. Second, we compute the covariance matrix. Third, we calculate the eigenvalues and eigenvectors. Fourth, we obtain the principal components as the eigenvectors of the covariance matrix, and we use the first principal component as our new proxy for financialization. Table \ref{ch1:tab:PCA} shows the outcome of the PCA analysis and in particular, that the first principal component explains a large part of the individual variances.








%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Econometric framework and methods}

\subsection{Modeling the impact of surprises on returns}\label{ch1:return}

Our high-frequency regression model is based on \citet{kurov2019price}.\footnote{In unreported results, we run the regressions using the approach shown in \citet{andersen2003micro}. The results are similar.} We run the following regression using the specification in equation (\ref{ch1:eq:Model 1}):

\begin{equation}\label{ch1:eq:Model 1}
R_{t}^{t+\tau}=\alpha+\sum_{m=1}^{22} \gamma_m S_{m,t}+ \delta X_{j} + \sum_{m=1}^{22} \theta_m (S_{m,t} \cdot X_j)+\beta R_{t-\tau}^{t}+\epsilon_{t} 
\end{equation}
%%% MH :donc on enlève les notes en rouge??? 
where $R_{t}^{t+\tau}$ is the continuously compounded futures return from time $t$ to $t+\tau$, $S_{mt}$ is the surprise for macroeconomic announcement $m$ published at time $t$, and $X_{j}$ is the NLS speculative trading intensity variable, which is updated at a weekly frequency, with $j$ the index for the week.
%$X_{t}$ is the value of the speculative intensity proxy using the $NLS$ variable. 
The impact of macro announcements on commodity futures returns can be assessed by looking at the $\gamma_m$ coefficient in the mean equation, while the $\delta$ coefficient controls for the level of speculative trading intensity as it relates to futures returns. The key coefficient to help answer our main research question  is  $\theta_m$, which relates the effect of time-varying speculative trading intensity on the impact of the news release.% This last coefficient is the most important one to help answer our main research question.
The regression is estimated using a two-step weighted least squares (WLS) procedure. %For robustness, appendix A presents results using the NLS variable.
To account for heteroskedasticity, we construct a volatility estimate by means of an exponential moving average, using the regression residuals obtained in the first step. This auxiliary regression is presented in equation~(\ref{ch1:eqn:auxiliary 2}), with a smoothing parameter $\alpha=0.9$ and a starting parameter value set to $\sigma_1=\epsilon_t$:

\begin{equation}\label{ch1:eqn:auxiliary 2}
\sigma_t=\alpha \sigma_{t-1}+(1-\alpha) \mid \epsilon_t \mid 
\end{equation} 

After obtaining $\sigma_t$ for each observation, we apply the transformation $w_t = \hat{\sigma_t}^{-2}$ to obtain the WLS regression weight. Then, we multiply each variable by $w_t$ and run an OLS regression to estimate the model.





\subsection{Modeling the impact of surprises on volatility}\label{ch1:variance}

To estimate the volatility equation, we use a GARCH specification, as it is well known that the variance of commodity futures  returns displays time variation and clustering \citep*[see e.g.,][]{brunetti2014commodity}. We specify a GARCH (1,1) model and extend the equation by including our NLS speculative intensity proxy as well as the macroeconomic news surprise variables. 
First, we estimate the mean equation (\ref{ch1:eqn:MeanEqn}):

\begin{equation}\label{ch1:eqn:MeanEqn}
%R_{t}^{t+\tau}=\alpha+\sum_{m=1}^{22} \gamma_m S_{m,t}+\beta R_{t-\tau}^{t}+\epsilon_{t} ICI À VOUS AVEC sp PARCE QUE JE NE VOIS PAS CE QUI A CHANGÉ
R_{t}^{t+\tau}=\alpha+\sum_{m=1}^{22} \gamma_m S_{m,t}+\beta R_{t-\tau}^{t}+\epsilon_{t}
\end{equation}

Then, we estimate the following equation for conditional variance:

\begin{equation}\label{ch1:eqn:VarianceEqn}
%\sigma_{t}^2=\alpha_0+\alpha_1 \sigma_{t-1}^2+\alpha_2 \epsilon_t^2 +\sum_{m=1}^{22} \Phi_m D_{m,t}+\beta X_{t}+\sum_{k=1}^n \phi_k I_{kt} + \sum_{h=1}^{23} \rho_h D_h
\sigma_{t}^2=\alpha_0+\alpha_1 \sigma_{t-1}^2+\alpha_2 \epsilon_t^2 +\sum_{m=1}^{22} \Phi_m D_{m,t}+\beta X_{j}+\sum_{k=1}^n \phi_k I_{kt} + \sum_{h=1}^{23} \rho_h D_h 
\end{equation}
%j,ai change le j en bas



where $I_{k,t}=D_{m,t} \cdot X_{j}$ and $D_{m,t}$ is a dummy variable for macro announcement $m$. The latter equals 1 if an announcement takes place at time $t$ (5-minute frequency) and equals 0 otherwise. $X_{j}$ is the NLS speculative intensity variable as defined earlier. The $\rho_h$ coefficient captures intraday periodicity, while the $D_h$ dummy equals 1 at hour $h$ and 0 otherwise. The impact of macro announcement $m$ on conditional variance is captured by the $\Phi_m$ coefficient in equation~(\ref{ch1:eqn:VarianceEqn}), while the $\beta$ coefficient shows the impact of the speculative intensity variable $X_{j}$. Finally, the $\phi_k$ coefficient shows the interaction effect from speculative trading and the macro surprise $m$.
The standard errors are computed using the Newey-West heteroskedasticity and autocorrelation consistent (HAC) estimator with automatic lag selection, following the procedure outlined in \citet{newey1994automatic}.%\footnote{This approach accounts for both heteroskedasticity and serial correlation in the residuals, which is standard practice in high-frequency financial data analysis.}  
  This methodology mirrors the approach used in \citet{andersen2003micro, andersen2007real} and \citet{kurov2019price} to study announcement effects in other asset classes.  In unpublished results, we consider the mixed-data sampling (MIDAS) approach proposed by \citet{ghysels2004midas}. We find that the results are similar.%\footnote{We thank a reviewer for suggesting a mixed-frequency approach. Results are available upon request.}

% Among the announcements in our sample, all but one are ``good news.'' Only a positive surprise in Initial Jobless Claims indicates a deterioration in economic conditions. Therefore, the surprise coefficient is expected to be positive for all pro-cyclical commodities (i.e., all but gold and silver) for all announcements except Initial Jobless Claims, for which it should be negative (since a positive surprise is ``bad news’’). In the case of gold and silver, which are safe-haven commodities, the reverse is expected for coefficient signs.









\subsection{Modeling the impact on bid-ask spreads}
 Speculative trading could make markets more efficient by improving information. We test this hypothesis by measuring the effect of macro surprises on the futures price bid-ask spread in high-frequency regressions.
%We aim to investigate whether increased financialization affects the responsiveness of commodity futures returns to macroeconomic surprises by analyzing the impact on the bid-ask spread. 
The bid-ask spread is widely recognized as a measure of market efficiency. A narrower spread suggests less uncertainty about the asset's true value and reflects lower transaction costs, improved liquidity, and lower information asymmetry \citep{Roll1984}. 
%A narrower bid-ask spread typically indicates higher liquidity and better informational efficiency, suggesting less uncertainty about the asset's value. 
Furthermore, \citet{chordia2008liquidity} show that the bid-ask spread is an indicator of market quality and market efficiency. Since a smaller spread is associated with a more efficient price discovery process, an increase in the quality of market information should decrease the spread. Therefore, we estimate the following equation, where the relative bid-ask spread is defined as $(Ask_t-Bid_t)/Mid_t$:

\begin{equation}\label{ch1:eq:Model 2}
%\text{Spread}_{t}=\alpha+\sum_{m=1}^{22} \gamma_m D_{m,t}+ \delta X_{t} + \sum_{m=1}^{22} \theta_m (D_{m,t} \cdot X_t)+\beta \text{Spread}_{t-\tau}+\epsilon_{t}
\text{Spread}_{t}=\alpha+\sum_{m=1}^{22} \gamma_m D_{m,t}+ \delta X_{j} + \sum_{m=1}^{22} \theta_m (D_{m,t} \cdot X_j)+\beta \: \text{Spread}_{t-\tau}+\epsilon_{t}. 
\end{equation} 

In equation ~(\ref{ch1:eq:Model 2}) , $\text{Spread}_{t}$ is the relative bid-ask spread measured at 5-minute frequency $t$ using the high-frequency data and $\text{Spread}_{t-\tau}$ is the lagged spread. In addition, $\sum_{m=1}^{22} \gamma_m D_{m,t}$ accounts for the macroeconomic announcements, where each dummy variable $D_{m,t}$ is multiplied by its respective coefficient $\gamma_m$. We also include the speculative trading intensity variable $X_{j}$  with its coefficient $\delta$. The interaction terms $\sum_{m=1}^{22} \theta_m (D_{m,t} \cdot X_j)$ capture the effect of speculative trading intensity on the bid-ask spread at the time of a release, with each term multiplied by its respective coefficient $\theta_m$. To test whether increased speculative trading activity improves informational efficiency at the time of a macroeconomic announcement, we check whether the sign on $\theta_m$ is negative and significant, thus reducing the spread. %Indeed, the interaction terms $\theta_m (D_{m,t} \cdot X_t)$ capture the effect of financialization on the  spread at the time of a release.% macroeconomic announcements.
%(measured at daily frequency, where $j$ represents the trading day)










%\subsection{Impact according to the type of non-commercial trader}
%\alert{TOO SHORT AS A STAND ALONE SECTION. FIGURE OUT HOW TO INTEGRATE IT IN THE TEXT.}



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Results} \label{ch1:sec:result}

\subsection{The impact of surprises on cumulative abnormal returns}

We begin by documenting the impact of macroeconomic announcement surprises on commodity futures returns. The impact of surprises is shown in the following graphs of high-frequency cumulative abnormal returns (CARs). Figures \ref{ch1:fig:cl} to \ref{ch1:fig:ng} show CARs for each commodity as measured over a window of 60 minutes before to 60 minutes after a macroeconomic announcement release.  The figures are constructed similarly to those shown in  \citet{kurov2019price}. The magnitude of the CARs after announcement releases is comparable to those shown in their paper for stock index and Treasury futures. Unlike them, however, we do not see evidence of a pre-announcement drift. 


A red line denotes the average CAR for announcement releases that are seen as negative surprises (i.e., worse than anticipated news), while a green line denotes the average CAR for positive surprises.
For brevity, we discuss only the CARs for crude oil and gold, as they are representative of pro-cyclical energy markets and safe haven assets. Figure \ref{ch1:fig:cl} shows the average CAR for crude oil futures. The CAR increases following a positive surprise and decreases following a negative one, confirming that crude oil is a pro-cyclical commodity.  In contrast, figure \ref{ch1:fig:gc} shows that gold futures react in the opposite manner. The red line indicates that CAR is positive after bad news, while the green line shows that CAR is negative after good news. %These results support the hypothesis that gold is a safe haven asset.%finanthe one presented for stocks in their paper.\footnote{A difference is that the commodity markets do not seem to display pre-announcement drift whereas \citet{kurov2019price} documented one for the S\&P 500 futures.}
%The increase (decrease) in CAR in reaction to a positive (negative) surprise confirms that crude oil is a pro-cyclical commodity. 




\subsection{Macroeconomic surprises, speculative trading activity and futures returns}

Table~\ref{ch1:tab:macro_fin_nls_fut_returns_full} presents the results of high-frequency regressions that explain commodity futures returns immediately after a macroeconomic announcement release. Our discussion focuses on coefficients that are statistically significant at the 5\% level.\footnote{In an earlier draft, we also reported results for different sub-periods, such as the Zero Lower Bound period, the 2008-2010 financial crisis and Great Recession, and the COVID-19 period. These results do not materially affect our findings or conclusions, and they are available upon request. } %For brevity, the table only reports results for regressions in which the PCA proxy is used.\footnote{That being said, there is no material difference in the results if we use one of the other measures.}
We begin with energy commodities, which serve as our baseline case. The $\gamma_m$ coefficient shows the immediate impact of a macro surprise on commodity futures returns. For crude oil futures, we find that several macroeconomic announcements exhibit significant effects. Consider for instance Initial Jobless Claims, for which a greater than expected value indicates bad economic news. The table shows that for this announcement, $\gamma_m$ is negative, indicating that crude oil prices tend to drop in response to unexpected increases in jobless claims. This result is consistent with the expectation that higher jobless claims signal weaker economic conditions, which reduce the demand for crude oil. The corresponding $\theta_m$ coefficient for Initial Jobless Claims is positive, however, suggesting that increased speculative trading mitigates the negative impact of bad macroeconomic news on crude oil prices. This damping effect suggests that markets are better informed as a result of the increased participation of speculators. Indeed, the announcement release creates a smaller surprise and a smaller shock.

%TOUT CA EST A CHANGER AVEC NLS car les coefficients changent de signes parfois et ne sont plus significatifs au mêmes endroits. For natural gas, the ADP Employment announcements show a pattern similar to crude oil, with a positive $\gamma_m$ coefficient implying that better than expected employment figures boost energy prices, reflecting increased economic activity and demand. The negative $\theta_m$ coefficient indicates that speculative trading dampens this positive reaction. The CB Consumer Confidence and Advance Retail Sales announcements also show positive $\gamma_m$ coefficients and negative $\theta_m$ coefficients for both energy commodities, supporting the claim that increased speculative trading intensity smooths out the market's response to macroeconomic surprises.

In the case of natural gas futures, we find that the results across announcements are less frequently significant. Increased speculative trading tends to increase the magnitude of the surprise's effects, when the results are significant. An exception is the natural gas market-specific announcement release (inventories), for which the coefficient is negative but not significant. Overall for natural gas futures, we do not find as much support that speculative trading dampens the effect of macroeconomic announcements on returns. The reason why the results for natural gas futures are less conclusive is most likely that this futures contract displays greater volatility, that it has a lower trading volume \citep{irwin2012testing}, and it has less speculative trading activity (thus, fewer informed traders) \citep{buyukcsahin2014speculators}. 
%results confirm the specific nature of natural gas, in terms of lower trading volume , and less speculative trading activity such as international traders \citep{buyukcsahin2014speculators}.

Copper is a pro-cyclical, industrial commodity, so it is expected that the results for copper futures should resemble those for crude oil. The results are highly significant for many announcements. If we look at announcements such as ADP Employment and CB Consumer Confidence, which are ``good news'', the $\gamma_m$ coefficients are positive, confirming copper's pro-cyclical nature and its ties to industrial production. The corresponding $\theta_m$ coefficients are negative, consistent with the damping effect that we document for energy commodities. Thus, the main finding for copper futures is that, as with crude oil, increased speculative trading intensity has the effect of weakening the impact of a macro surprise.
%Comparing these results with other commodities, the behavior of copper futures closely resembles that of crude oil -- indeed, both are pro-cyclical. For


Gold is considered to be a safe haven asset \citep{baur2010gold}. Therefore, it is expected that the reaction of gold futures returns to macro surprises will be the opposite to what we have found for crude oil, natural gas, and copper. This is indeed what we find:  gold futures returns are lower after "good news" and higher after "bad news". These results support the idea that energy commodities are pro-cyclical, while gold is a safe haven asset. In particular, the positive $\gamma_m$ coefficient for Initial Jobless Claims indicates that gold prices rise in response to unexpected increases in jobless claims, as investors seek safety in gold positions during economic uncertainty. The $\theta_m$ coefficient is negative, indicating that speculative trading activity tempers this flight to safety, leading to less pronounced price increases.%Turning to gold futures, the reaction to macroeconomic surprises is opposite to what we find for our energy baseline case:
In the case of the ADP Employment release, the $\gamma_m$ coefficient is negative for gold futures, suggesting that strong employment figures reduce gold prices as investors move away from safe-haven assets towards pro-cyclical assets, such as energy commodities. The positive $\theta_m$ coefficient suggests that speculative trading reduces the extent of this price drop. The CB Consumer Confidence and Advance Retail Sales announcements also generate negative $\gamma_m$ coefficients for gold futures and positive $\theta_m$ coefficients at a 10 percent level, providing additional evidence of a moderating influence of speculative trading on the reaction of gold to economic news. The main finding is therefore that whether a commodity is pro-cyclical or a safe haven asset, increased speculative trading has a damping effect on the reactions to macro surprises. 

The last two commodities in our sample, silver and palladium futures, behave more like gold futures. We find that for Initial Jobless Claims for palladium and ADP Employment in silver, significant $\gamma_m$ coefficients are found in directions consistent with their status as safe haven assets. The $\theta_m$ coefficients for silver and palladium also indicate a damping effect on price reactions. Taken together, our results show that the damping effect of speculative trading is stronger in energy markets, suggesting that the beneficial effects of financial participants  are particularly important for energy commodities.






\subsection{Macroeconomic surprises, speculative trading and volatility}

Table~\ref{ch1:tab:macro_fin_nls_var_full} presents regression results to explain the conditional variance of high-frequency commodity futures returns after macroeconomic announcements. We first examine our baseline assets, energy commodities, as volatility in energy markets is of particular concern given their economic importance and direct impact on consumer prices.
For crude oil,  macroeconomic surprises generally lead to an increase in conditional variance, as the $\Phi_m$ coefficients are consistently positive. For instance, a surprise in Initial Jobless Claims significantly increases crude oil volatility. Natural gas exhibits similar patterns, with some variations in magnitude and a greater impact on inventory-related announcements. This result suggests that unexpected economic news generates greater uncertainty and price fluctuations in energy markets.
In contrast, the interaction coefficients $\phi_m$, which inform us about the impact of speculative trading, tend to be negative for the two energy commodities. The implication is that increased trading activity by speculative traders lowers volatility following a macro surprise. Taking Initial Jobless Claims as an example, we find that higher levels of speculation reduce the  impact of news on volatility  in crude oil futures markets. Thus, speculative trading can act as a stabilizing force by damping the heightened volatility that  occurs after a surprise in macroeconomic news. %\textcolor{red}{This stabilizing effect is especially valuable in energy markets, where excessive volatility can have significant economic consequences}.

Comparing these results with those for other commodities, we find that gold, copper, silver, and palladium also show positive $\Phi_m$ coefficients across various announcements, indicating increased volatility following macro surprises. However, the magnitude of these effects is generally smaller than what we see in energy markets, especially in the case of precious metals. The corresponding $\phi_m$ coefficients, measuring the damping effect of financial investor activity, are negative across commodities and announcements, and the strongest effects are found in crude oil futures markets. This pattern holds for other macroeconomic announcements as well. The coefficients for ADP Employment, CB Consumer Confidence, and Advance Retail Sales, for instance, generally indicate increased volatility after surprises, as shown by the positive $\Phi_m$ coefficients, while the corresponding $\phi_m$ coefficients are negative, supporting the finding of a stabilizing effect of increased speculative trading. %\textcolor{red}{The consistency of these results across different types of announcements is particularly noteworthy in energy markets, where price stability is crucial for both consumers and industrial users}.

Therefore, our findings highlight the valuable role of speculative traders in reducing volatility, particularly in crude oil futures markets, where price stability has important implications for the broader economy. While macroeconomic news tends to increase volatility across commodity markets, the damping effect of speculative trading appears strongest in energy markets. We show that this relationship is consistent  across different types of announcements and remains robust when controlling for various market conditions.




\subsection{Macroeconomic surprises, speculation and bid-ask spreads}

Table~\ref{ch1:tab:return-fin-full} shows our results for the impact of macroeconomic surprises and speculative trading intensity on futures prices bid-ask spreads. This empirical analysis provides a test of informational efficiency. We first focus on energy markets. In the bid-ask spread regressions, the $\gamma_m$ coefficient denotes the effect of surprises on the spread, while the $\theta_m$ coefficient shows the interaction effect between surprises and speculative trading intensity. The main hypothesis is whether $\theta_m < 0$, which would indicate that greater speculative activity improves informational efficiency through narrower spreads. Such a finding would be consistent with what we have documented above for futures returns and volatility.  In the case of crude oil futures, we find that $\gamma_m$ is negative for the initial jobless claims announcement, which means that the market becomes more efficient immediately after a news release. This is consistent with the resolution of uncertainty. The $\theta_m$ coefficient is always negative when it is significant, indicating that the bid-ask spread narrows even more (implying greater informational efficiency) after a macro announcement if crude oil futures markets benefit from greater speculative activity relative to hedging activity, as measured by the NLS proxy. The results for natural gas futures are similar to those for crude oil. The relationship between speculative trading and market efficiency is clearest during periods of higher trading volume, such as the release of storage reports and weather-related announcements.
%\textcolor{red}{CHECK TO MAKE SURE THAT THIS IS STILL ACCURATE.}

Comparing these results to those obtained for the other commodities in our sample, we find that the results for $\theta_m$ (speculative intensity) are less often significant, but that they are negative when they are significant. The $\theta_m$ coefficient being negative suggests that greater speculative trading intensity improves market efficiency, as bid-ask spreads tend to be lower when the coefficient is significant. This improvement in market efficiency appears to be most pronounced in energy markets, where accurate price discovery is particularly important given their relevance for the real economy.
%The stronger effects we observe in energy markets may reflect their greater integration with the broader financial system and higher trading volumes, which allow for more efficient price discovery. This is particularly relevant given the increasing importance of energy price stability for economic policy and planning}.
Overall, the results for the bid-ask spread provide additional support for our claim that greater trading activity has beneficial effects on commodity derivatives markets, with these benefits being especially notable in energy markets where efficient price discovery has important implications for both market participants and the broader economy.




\subsection{Differences in results according to trader type}

To investigate whether differences in trader type are relevant in explaining our findings, we provide disaggregated results in this section. To this end, we estimate equations (\ref{ch1:eqn:MeanEqn}), (\ref{ch1:eqn:VarianceEqn}) and (\ref{ch1:eq:Model 2}) for two categories of Non-Commercial traders, namely, swap dealers (SD) and money managers (MM). For each of the two, the CFTC reports the number of long and short positions in their disaggregated Commitment of Traders (COT) reports. We compute the NLS index  for each trader category over time, allowing us to separately quantify the intensity of trading activity by money managers and swap dealers.%\footnote{We use the NLS variable for this disaggregated analysis because it is the only one we can construct using the available data on MM and SD positions.}


%This section shows disaggregated results for the two types of financial participants that are reported separately, namely swap dealers and money managers. %We continue to use the NLS variable, but it is computed separately for MM and SD futures positions. %\footnote{The main reason is that the MSCT and Working's $T$ variables cannot be computed only for MM or SD because the calculations involve the positions of commercial traders.} 
 First, we examine the returns equation for money manager positions, as shown in table \ref{ch1:tab:macro_fin_mm_fut_returns_full}. Increased trading activity by money managers has the same effect as in our baseline results. If we consider crude oil futures, for example,  the $\gamma_m$ macro surprise coefficient is positive while the $\theta_m$  coefficient for speculative trading is negative. Since $\theta$ has the opposite sign to $\gamma$, the implication is that increased futures trading activity by money managers lowers the impact of surprises on futures returns. This is similar to our baseline, aggregate findings. 
  Next, table \ref{ch1:tab:macro_fin_sd_fut_returns_full} presents results using only swap dealer positions. Here we find a notable difference relative to money managers. This table shows that the speculative intensity coefficient $\theta$ for swap dealers has the same sign as the macro surprise coefficient $\gamma$. This result means that increased swap dealer trading activity seems to amplify the reaction of futures returns to macro surprises. The exception to these results is in the case of natural gas futures, where increased trading by swap dealers for some announcements appears to  have the opposite effect on returns. To contrast this finding with prior research,  \citet{brunetti2016speculators} find, using daily data, that the positions of swap dealers are not correlated with contemporaneous returns and volatility in commodity futures markets. They further show that hedge funds decrease, and hedgers increase, volatility. Our empirical analysis extends their findings using high-frequency data and the setting of macro surprises as a source of new information affecting energy and commodity futures markets.

Tables \ref{ch1:tab:macro_fin_mm_var_full} and \ref{ch1:tab:macro_fin_sd_var_full} show our disaggregated results for the variance equation using sample data for money managers and swap dealers, respectively. The money manager results are similar to what we find in the aggregate sample, namely that both good and bad surprises increase volatility (as shown by $\Phi_{mm}>0$), while greater money manager trading activity lowers the impact of news on volatility (as shown by  $\phi_{mm}<0$). This result strengthens prior evidence in the literature about beneficial effects of speculators, which were based on daily data \citep{brunetti2016speculators}. In contrast, our results suggest that increased activity by swap dealers appears to increase volatility after macro news (since $\phi_{sd}>0$). Together, the results line up with our findings for the returns equation and with our main message, which is that informed traders stabilize markets by contributing new information. Swap dealers, as intermediaries,  typically do not trade based on information but rather following the needs of their clients. This economic motivation can explain why our results suggest that their trading activities amplify price and volatility reactions to news.
%. that the signs for $\gamma_m$ and $\theta_m$ are equal. Therefore, unlike money managers, it seems that more trading activity by swap dealers causes even more strong market reactions to macroeconomic news announcements. Swap dealers' trading operations are motivated by supporting their clients' needs rather than information.
%Findings for the variance equation using only money managers and swap dealers can be found in tables  respectively. We find that whereas for swap dealers the financialization interaction coefficient $\phi_k$ is positive, for money managers it is negative. These findings support the economic interpretation of previous findings for returns. Although rising trade by money managers lessens the impact of macro surprises on volatility, a higher presence of swap dealers seems to enhance the influence of surprises on volatility. Therefore, the trading actions of money managers help to minimize market volatility after market news; yet, the activities of swap dealers exacerbate this effect. 
Although our discussion  centers on crude oil as a benchmark commodity, the results for the other pro-cyclical commodities support our interpretation of the findings. In the case of gold futures, the disaggregated results continue to support  a safe haven interpretation \citep{erb2013golden}. Indeed, gold has features of a commodity and a currency, but earlier studies have also found that its value increases with investor risk aversion, since it is perceived as a safe haven during times of economic uncertainty and market volatility.


%\subsubsection{Bid-ask spread analysis for money managers and swap dealers}

%The examination of bid-ask spreads following macroeconomic announcements is expanded to include a NLS financialization proxy computed using just money manager (MM) or swap dealer (SD) positions, to account for potential differences in the impact of  their trading activities. 
Lastly, we examine whether the effects on bid-ask spreads differ between trader types. 
Table \ref{ch1:tab:macro_fin_mm_fut_spread_full} presents the results for money managers. As the $\theta_m$ coefficients tend to be negative, it seems that money managers enhance market efficiency by lowering bid-ask spreads. The results for swap dealers are shown in Table \ref{ch1:tab:macro_fin_sd_fut_spread_full}. Unlike for money managers, we find that $\theta_m$ coefficients for swap dealers tend to be positive, indicating that greater swap dealer activity increases the bid-ask spread after a macroeconomic release. The results for swap dealers therefore suggest a decline in market efficiency due to their increased trading activities.  %meaning that a higher degree of activity by swap dealers reduces market efficiency by increasing bid-ask spreads following a macroeconomic announcement, indicating a decline in market efficiency.
These findings are in line with our main results and further support the claim that traders in energy and commodity markets do not all have the same effect on market efficiency. The results point to the importance of information acquisition and investor attention as an economic channel. 
%These results also relate to  \citet{rakowski2021twitter} who show that Twitter activity, by serving as a distinct form of investor attention, has significant effects on trading volume and asset prices.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%Important: keep this in LaTex format; use % to comment out; keep references as is
\section{Discussion and implications}



Our findings contribute new insights to an unsettled literature on the impact of different types of financial participants in energy and commodity markets \citep{ready2022order}. In addition to being an important alternative asset class, energy commodities are central to economic activity, while energy futures trading is important for price stability. Being closely related to macroeconomic risk \citep{cheng2015convective}, energy commodities therefore play a key role in macro-finance. By taking a novel angle of high-frequency market reactions to macroeconomic surprises, our results contribute a more nuanced picture of the impact of speculative trading and help to reconcile previous findings in the literature. 
%Our analysis builds on the models proposed by \citet{goldstein2022commodity} and \citet{basak2016model}, which explain how institutional investors in particular can influence commodity prices, volatility and market efficiency.

%Our paper is motivated by recent theoretical advances. 
\citet{goldstein2022commodity} develop a theoretical model to examine how financial traders improve information transmission between futures and spot markets. They show how market efficiency should benefit from the presence of informed financial traders, whose actions lead to increases in pricing transparency and to lower information asymmetry. Our empirical results provide support for this model. We show that increased speculative trading activity lowers bid-ask spreads after a macro news release, in addition to reducing the magnitude of price and volatility reactions to macro surprises. This improvement in informational efficiency is especially valuable in energy markets, where price discovery has direct implications for industrial users and consumers. We find that this effect is driven by money managers. In addition, our results point to smaller, negative effects linked to swap dealers. The difference between the two sets of disaggregated empirical results can be explained by the fact that they trade for different purposes. Money managers aim to make a profit based on information, while swap dealers are intermediaries who, while perhaps informed, primarily provide services to customers such as institutional investors, in addition to hedging their swap positions. In line with this interpretation, our results also build on \citet{fishe2012identifying}, who show that some financial traders (e.g., money managers) are better informed than others in commodity markets. 
%\textcolor{red}{This information advantage appears particularly valuable in energy markets, where complex supply-demand dynamics and geopolitical factors make price discovery especially challenging}.

%DO NOT INCLUDE THIS PART SINCE WE WANT TO DOWNPLAY INDEX TRADERS
%Our findings also provide evidence to support the predictions in the model proposed by \citet{basak2016model}. They argue that financialization should increase commodity futures prices and volatility, especially for indexed futures. They argue that the activities of institutional investors enhance the link between commodities and equities. This relationship is particularly relevant in energy markets, where crude oil and natural gas prices have become increasingly correlated with financial markets. We find some evidence to this effect, as increased trading by swap dealers, who tend to be long and deal with index funds, increases price reactions and volatility. However, our results suggest that overall, these effects are offset by the stabilizing presence of informed traders such as money managers.

Our findings also extend and build on the results of an earlier empirical literature that uses daily-level data. This literature includes \citet{brunetti2016speculators} who find that speculative traders, especially money managers and hedge funds, are helpful to energy and commodity markets, as well as \citet{buyukcsahin2011speculators} and \citet{alquist2013role} who show that the futures positions of financial firms such as hedge funds do not predict next-day changes in crude oil prices. 
Moreover, our results relate to \citet{cheng2015convective} who show that financial investors, being better informed about markets, contribute to price discovery and liquidity. The key message is therefore that speculative trading is helpful to markets by distributing and assimilating new information into prices.
%This is particularly relevant for energy markets, where accurate price discovery is crucial not only for market participants but also for energy policy and planning}. 
The findings shown in this paper have important implications for energy markets regulation and policy. First, they suggest that attempts to limit speculative trading in energy markets could, in fact, increase price volatility and reduce informational efficiency and price discovery. Second, they indicate that different types of financial participants have distinct effects on market quality, suggesting that regulatory frameworks should pay careful attention to market composition. Third, they highlight the importance of maintaining a robust price discovery mechanism in energy markets, given their crucial role in the economy.
%REDONDANT
%To summarize, the evidence we present suggests that financial investors play an important role in improving the functioning of energy markets, which is especially important given current challenges including the energy transition, geopolitical tensions, and the need for stable energy prices to support economic growth.


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%\section{Robustness in sub-periods}
%
%To rule out the possibility that our main results are driven by a specific period, we present in the following robustness checks our results for different sub-sample periods, namely the Zero Lower Bound period from December 22nd, 2008 to December 21st, 2015, and the COVID-19 pandemic period from January 31st, 2020 to June 10th, 2022. The start and end dates are selected following the literature and sources such as the NBER and the CDC. Our main finding is that the sub-sample results are consistent with our full sample results, despite some minor discrepancies. Therefore, our main findings are not explained by any particular period, but rather hold for the entire sample period from 2007 to 2024.
%
%% The full tables are presented in appendices B and C, respectively. In order to investigate whether our results are period-specific, we now present the results for different sub-periods. The results presented in the appendix are for specific sub-periods, namely the Zero Lower Bound (ZLB) (2008-12-22 to 2015-12-21) and COVID-19 sub-periods (2020-01-31 to 2022-06-10). The main finding is that the sub-period results are consistent with the full-sample results presented earlier.

%\subsection{Price reactions during the Zero Lower Bound and COVID-19 sub-periods}
%
%Tables~\ref{ch1:tab:macro_fin_fut_returns_covid} and \ref{ch1:tab:macro_fin_fut_returns_zlb} present the results of high-frequency regressions to explain returns after macroeconomic announcements, estimated separately for the Zero Lower Bound (ZLB) and COVID-19 sub-periods. We discuss a few major announcements here. Looking at Initial Jobless Claims, the surprise coefficient $\gamma_m$  is negative for crude oil and positive for gold in both sub-periods, consistent with the full sample results.
%%indicating that more jobless claims than expected decrease crude oil prices and increase gold prices. 
%
%The financialization $\theta_m$ coefficient is negative for crude oil during the COVID-19 sub-period and positive during the ZLB sub-period, suggesting that the interaction effect of financialization varies to some degree over the years. In the case of ADP Employment, $\gamma_m$ is positive for crude oil and copper while negative for gold in both sub-periods. The $\theta_m$ coefficient has the opposite sign to $\gamma_m$, similar to our baseline results.  In the case of CB Consumer Confidence, $\gamma_m$ is negative for gold during the ZLB sub-period but is not significant during COVID-19, while the $\theta_m$ coefficients have the opposite sign, consistent with the damping effect we discuss. Lastly, for Advance Retail Sales, $\gamma_m$ is significant for crude oil, gold, copper, and silver in both sub-periods while $\theta_m$ generally has the opposite sign to $\gamma_m$. Thus, our sub-period results are for the most part similar to the full sample results, and we do not see any important differences in the economic interpretation of the coefficient signs.
%
%\subsection{Volatility reactions during the Zero Lower Bound and COVID-19 sub-periods}
%
%Table \ref{ch1:tab:macro_fin_fut_var_covid} and Table \ref{ch1:tab:macro_fin_fut_var_zlb} present the results of regressions to estimate the reaction of conditional variance to macro surprises in different sub-periods. This analysis confirms our full sample results, namely that surprises increase volatility while financialization acts to lower this impact. This can be seen from the $\Phi_m$ surprise coefficients, which remain positive, while the financialization coefficients $\phi_m$  are negative. Thus, our main findings are not driven by a particular sub-period.
%
%\subsection{Bid-ask spreads during the Zero Lower Bound and COVID-19 sub-periods}
%
%%This section examines the impact of macroeconomic surprises and financialization on the bid-ask spread during the Zero Lower Bound (ZLB) and COVID-19 sub-periods, aiming to determine whether the relationship between financialization and market efficiency, as measured by the bid-ask spread, remains consistent across different economic environments.
%
%Tables \ref{ch1:tab:macro_fin_fut_spread_zlb} and \ref{ch1:tab:macro_fin_fut_spread_covid} present our regression results for  the bid-ask spread during the ZLB and COVID-19 sub-periods. We find that in both sub-periods the $\gamma_m$ coefficients tend to be negative, indicating that macroeconomic surprises generally reduce the bid-ask spread across our sample of commodities. This is the case, for instance, for Initial Jobless Claims for crude oil and copper futures in the COVID-19 sub-period, which implies increased efficiency. Furthermore, the $\theta_m$ coefficients tend to be negative in the two sub-periods, suggesting that financial investor activity improves efficiency by narrowing spreads after the release of macro news. 
%
%%during the COVID-19 period are generally negative, indicating that increased financialization amplifies the efficiency improvements induced by macroeconomic surprises. Specifically, for crude oil, the negative $\theta_m$ coefficient suggests that the participation of financial investors, such as money managers, enhances the market's ability to process new information, further reducing the bid-ask spread.
%%% POURQUOI ON PARLE DE MONEY MANAGERS ICI?
%%shows the results for the ZLB period. The $\gamma_m$ coefficients are mostly negative, consistent with the full-sample findings, indicating that macroeconomic surprises during the ZLB period also tend to reduce the bid-ask spread, enhancing market efficiency. The $\theta_m$ coefficients are negative for most commodities, similar to the COVID-19 period, suggesting that financialization continues to improve market efficiency by further reducing the bid-ask spread following macroeconomic announcements.
%
%While the overall patterns in the $\gamma_m$ and $\theta_m$ coefficients are similar to the full sample results, the coefficients tend to be less statistically significant in these sub-periods. This may be due to smaller sample sizes or to  unique market conditions during these periods. 
%%That being said, the results are consistent with our main findings for bid-ask spreads and suggest there are no meaningful differences in sub-periods.%, the general trend indicates that macroeconomic surprises reduce the bid-ask spread, and financialization enhances this effect, thereby improving market efficiency.
%
%That being said, we find that the beneficial impact of financialization on market efficiency, as measured by a narrower bid-ask spread, is robust across different economic conditions.
%
%%% JE NE CROIS PAS QUE LES PHRASES SUIVANTES AJOUTENT BEAUCOUP; ON SE REPETE. 
%%The consistent negative $\theta_m$ coefficients imply that the presence of financial investors, particularly money managers, helps markets process information more efficiently, regardless of the broader economic environment. This supports the argument that financialization contributes to improved market functioning by enhancing the dissemination and incorporation of new information into asset prices. Overall, the sub-period analysis reinforces our full-sample results, highlighting the beneficial role of financialization in commodity markets, particularly in terms of market efficiency as reflected in narrower bid-ask spreads following macroeconomic surprises.
%
%\subsection{Sub-period analysis for money managers and swap dealers}
%
%We discuss in this section a robustness check for the disaggregated analysis using only positions of money managers and swap dealers. In short, these results confirm the baseline finding that money managers bring more stability to markets through a damping effect after a macro surprise, while swap dealers have the opposite impact. The results for money managers are shown in   table \ref{ch1:tab:macro_fin_mm_fut_returns_covid} (returns equation, COVID-19 period),  table \ref{ch1:tab:macro_fin_mm_var_covid} (variance equation, COVID-19 period), \ref{ch1:tab:macro_fin_mm_fut_returns_zlb} (returns equation, ZLB period), and table  \ref{ch1:tab:macro_fin_mm_var_zlb} (variance equation, ZLB period). The regression results for swap dealers, in the same order, are presented in tables \ref{ch1:tab:macro_fin_sd_fut_returns_covid},  \ref{ch1:tab:macro_fin_sd_var_covid},  \ref{ch1:tab:macro_fin_sd_fut_returns_zlb},  and \ref{ch1:tab:macro_fin_sd_var_zlb}.
%
%%The results during the ZLB sub-period reveal that the $\gamma_m$ coefficients for money managers are generally positive for crude oil and copper and negative for gold, similar to the full-sample findings. This suggests that the presence of money managers mitigates the impact of macroeconomic surprises on commodity futures returns. The $\theta_m$ coefficients are negative, further indicating that financialization by money managers dampens the effects of macro surprises. For swap dealers, the $\gamma_m$ coefficients are positive for most commodities, suggesting that their activities amplify market reactions to macroeconomic announcements. The $\theta_m$ coefficients for swap dealers are positive, supporting the notion that increased trading by swap dealers leads to greater market reactions to surprises.
%
%%During the COVID-19 sub-period, the results are consistent with the ZLB findings. The $\gamma_m$ coefficients for money managers show a similar pattern, with positive values for crude oil and copper and negative values for gold. The $\theta_m$ coefficients remain negative, confirming that financialization by money managers continues to reduce the impact of macroeconomic surprises on returns. For swap dealers, the $\gamma_m$ coefficients are again positive for most commodities, and the $\theta_m$ coefficients are positive, indicating that swap dealers' activities exacerbate market reactions to economic news. The variance equation results during both sub-periods show that the $\phi_m$ coefficients for money managers are negative, indicating that their presence reduces volatility following macroeconomic announcements. In contrast, the $\phi_m$ coefficients for swap dealers are positive, suggesting that their trading activities increase volatility in response to macro surprises.
%
%Overall, the sub-period analysis reinforces our full-sample findings. It highlights how money managers and swap dealers have different impacts on the market's reactions to macroeconomic announcements. While money managers tend to enhance market stability by damping the effects of surprises, swap dealers appear to increase market volatility and amplify the reactions to economic news.


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Conclusion} \label{ch1:sec:conclusion}

This paper investigates the impact of speculative trading on the real economy and energy markets through a new angle, namely high-frequency surprises in macroeconomic announcement releases, which allows for better-identified effects. %We study energy commodities as our baseline case, given their economic importance.
We empirically test whether increased speculative trading activity amplifies or dampens the impact of macro surprises on prices and volatility in commodity futures markets. 
%ce paragraphe est bon, mais la conslusion est très longue et selon les commentaires on devrait focuser sur les trader types.
%In fact, it is well known that the equity and bond markets react to these surprises. Suppose that a consequence of increased speculative trading intensity is to make energy and commodity futures behave more like financial assets. Then, we should see futures prices display greater reactions to macro surprises. This hypothesis is the basis of our investigation. Moreover, since the increased involvement of financial investors cannot be easily categorized into before and after sub-sample, we measure the intensity of speculative trading by means of a time-varying and commodity-specific proxy. 
Our results suggest that increased speculative trading activity has beneficial effects for energy and commodity markets. This is accomplished by reducing volatility and improving price discovery, as indeed price stability and efficient price discovery are crucial for market participants and the broader economy. We find that a greater intensity of speculative trading does not  amplify the effects of macro announcement surprises on prices or volatility. On the contrary, an increase in speculative trading in a given commodity has a damping effect: prices and volatility react \emph{less} to macro surprises when speculative trading is higher. This stabilizing effect of speculation is especially valuable to energy markets, where price volatility can have significant economic consequences. 

What is more, our findings are consistent with information diffusion economic arguments. Our analysis of bid-ask spreads in futures contracts further confirms that speculative trading tends to improve market efficiency. 
% j'ai ajouté le début du paragraphe pour mettre l'emphase sur le main results
Our results show that these effects are mostly linked to the trading activities of money managers rather than swap dealers. Thus, we contribute to a literature that emphasizes how non-commercial market participants such as money managers are beneficial to commodity markets by supplying liquidity, reducing volatility, and generally improving market efficiency. This finding is particularly relevant for energy markets, which have seen substantial increases in trading volume and complexity. %The results we present are robust to the use of a non-parametric variance estimator, different proxies for speculative trading intensity, and to alternative empirical specifications (e.g., regression specification, high-frequency window, etc.). 

Our findings have important implications for energy market regulation and policy. The damping effect on volatility shocks documented in this paper implies that speculative traders contribute to market stability, which is essential for energy security and economic planning. This stability could also facilitate investment in energy infrastructure and support the ongoing energy transition. By lowering the magnitude of volatility shocks, and thus reducing the real option value of delaying investments, our findings suggest that a greater involvement by speculative traders may also help with sustainability efforts to finance a green energy transition, alongside other instruments such as green bonds and portfolio screens for sustainable investments.
Looking forward, our results suggest several promising avenues for future research in energy markets. First, the role of financial investors (speculators as well as passive investors) in facilitating the energy transition warrants further investigation. Second, the connection between speculative trading and energy market regulation remains an important area for study, given that we find different impacts for money managers and swap dealers. Finally, the impact on energy price discovery of new trading technologies and market participants  is an emerging research frontier. These questions are particularly relevant given the increasing importance of energy markets in addressing climate uncertainty and in ensuring economic stability.