Arthur Thomas

• Bayesian Econometrics
• Noncausal Econometrics
• Structural Econometrics
• Energy economics
• Forecasting
• Machine Learning

I am currently a Teaching Fellow for Computer Science in ENSAE Paris. I am a graduate in statistical engineering from the National School for Statistics and Data Analysis (ENSAI). In 2017, to further pursue my academic interest in econometrics, I accepted a PhD scholarship from the IFP School and the University of Nantes under the supervision of Oliver Massol, Associate Professor at IFP School and City, University of London and Benoît Sèvi, Professor of Economics, University of Nantes. I defend my PhD in 2020. The member of my committee was:

• Karim Abadir, Professor of Financial Econometrics, Imperial College London (Reviewers)
• Derek Bunn, Professor of Decision Sciences, London Business School (Examiners)
• Dimitris Korobilis, Professor of Econometrics, University of Glasgow (Reviewers)
• Valérie Mignon, Professeur des Universités, Université Paris Nanterre (Examiners)

I have presented my research in international conferences, in different fields, Operations Research, Econometrics, Finance and Energy.

Positions

• Teaching Fellow for Computer Science in ENSAE Paris

• PhD Student in the University of Nantes / IFP School
  Read More

  • Subject: The Econometrics of Energy Demand: Identification and Forecast.
  • Supervisors:
  • Benoît Sévi, University of Nantes and Olivier Massol, City, University of London & IFP School.
  • Committee:
  • Karim Abadir, Professor of Financial Econometrics, Imperial College London
  • Derek Bunn, Professor of Decision Sciences, London Business School
  • Dimitris Korobilis, Professor of Econometrics, University of Glasgow
  • Valérie Mignon, Professeur des Universités, Université Paris Nanterre

References

Publications

• Are day-ahead prices enough to explain and predict the next day’s natural gas demand? Evidences from the French case, with Olivier Massol and Benoît Sévi (2020) accepted for publication in The Energy Journal. (Link to article)
  Abstract

  The purpose of this paper is to investigate, for the first time, whether the next day’sconsumption of natural gas can be accurately forecast using a simple model that solely incorporates theinformation contained in day-ahead market data. Hence, unlike standard models that use a numberof meteorological variables, we only consider two predictors: the price of natural gas and the sparkratio measuring the relative price of electricity to gas. We develop a suitable modeling approach thatcaptures the essential features of daily gas consumption and, in particular, the nonlinearities resultingfrom power dispatching and apply it to the case of France. Our results document the existence of along-run relation between demand and spot prices and provide estimates of the marginal impacts thatthese price variables have on observed demand levels. We also provide evidence of the pivotal role ofthe spark ratio in the short run which is found to have an asymmetric and highly nonlinear impact ondemand variations. Lastly, we show that our simple model is sufficient to generate predictions thatare considerably more accurate than the forecasts published by infrastructure operators

Working papers

• A Structural Non-causal VAR Model of the Global Oil Market: the Role of Oil SupplyNews Shocks with Zakaria Moussa (2020) (DRAFT AVAILABLE)
  Abstract

  Nonfundamentalness issue on the global oil market has been addressed by either augmenting small-scale VAR models by additional variables or latent factors, or using external instrument or proxies leading to more credible identification scheme. We tackle this problem by estimating a non-causalVAR model for standard global oil market variables. We identify the oil supply news shock as a shock that drives global oil production the most for a finite time horizon. First, Our findings highlight the prominent role of expectations in propagating the shock. Second, we show that a negative oil supply news shock results in abrupt and permanent reaction in global oil production, global economic activity and in oil inventory. However, the oil supply shock has only a limited effect on oil price. Finally, news shock about oil supply shortfalls do have macroeconomic consequences as it causes a substantial decline in US industrial production.
• Considering real-time demand to forecast the U.S. natural gas price in real-time: The role of temperature data, with Benoît Sévi and Zakaria Moussa (2020). (DRAFT AVAILABLE)
  Abstract

  This paper provides evidence of the pivotal role of temperature data to forecast natural gas prices at the Henry Hub in real-time. Considering a newly constructed temperature index as an additional exogenous variable in a Bayesian vector autoregressive (VAR) framework significantly increases the forecast accuracy at horizons up to 12 months. Our approach is new to the energy price forecasting literature as it considers both supply and demand at the same time and further includes the temperature as a proxy of real-time demand of natural gas with superior forecasting results.
• Production intermittence in spot electricity markets: a behavioral simulations approach, with Albert Banal-Estanol, Olivier Massol and Augusto Ruperez Micola (2020). (DRAFT AVAILABLE)
  Abstract

  This paper analyzes the influence of production intermittence on spot electricity markets. More specifically, we examine how the presence of a competitive fringe operating low-cost intermittent generation assets modifies the bidding behavior of the strategic players who own the conventional (reliable) power plants. We first use game theory to derive the market outcomes obtained with perfectly rational players. We then compare them with the ones obtained when the players behave as adaptative traders who follow the Camerer and Ho (1999) behavioral model. The simulation results show that, compared to the theoretical benchmark, intermittent technologies yield lower prices when incumbents have individual market power, but are higher when they do not have it. We also run the simulations for a series of alternative specifications. The results indicate that this finding happens under different intermittency and ownership configurations. We also observe that replacing high-cost assets with low-cost ones results in prices that are higher than when they are left to co-exist
• The role of expectations in predicting the real prices of oil: a non-causal analysis (2020). (DRAFT AVAILABLE)
  Abstract

  This paper revisits the predictive power of convenience yield for oil by incorporating expectations into an empirical specification through the estimation of Bayesian non-causal VAR. We empirically show that expectations play a significant role in the determination of oil prices. Second, we provide empirical evidence that real-time forecasts of real oil prices can be remarkably more accurate than the no-change forecast and significantly more accurate than real-time forecasts generated by existing structural models relying on Bayesian VAR. Beyond the traditional analysis at the monthly frequency, we further investigate the forecasting accuracy of our empirical specification at the daily and weekly frequency, resulting in interesting findings for potential investment purpose.

Academic conferences

2020

• Thé des économètres, Paris, France.
• 37th International Conference of the French Finance Association (AFFI), Nantes, France.
• 19èmeJournée d’Économétrie, Développements Récents de l’Econométrie Appliquée àla Finance, EconomiX, Nanterre, France.
• 2nd International Conference Environmental Economics: A Focus on Natural Re-sources,University of Orleans.

2019

• 13th International Conference on Computational and Financial Econometrics, London, UK
• INFORMS Annual meeting 2019, Seattle, USA
• 13th Annual Trans-Atlantic Infraday Conference, Washington, USA
• 18ème Journée d’Économétrie, Développements Récents de l’Econométrie Appliquée à la Finance, EconomiX, Nanterre, France.
• Séminaire CREST-ENSAI 2019, Rennes, France.
• Thé des économètres, Orléans, France.
• Workshop in Financial Econometrics, Nantes, France.
• The 3rd Commodity Markets Winter Workshop-Leibniz University, Hannover, Germany
• Workshop EDGE 2019, Rennes, France
• The 2nd International Conference The Economics of Natural Gas, University Paris-Dauphine, Paris, France.

2018

• 12th International Conference on Computational and Financial Econometrics, Pisa, Italy
• 41st edition of the IAEE international conference, Groningen, Netherland
• FAEE summer workshop, Mines ParisTech, Paris, France
• 29th European Conference On Operational Research. Valencia, Spain
• INFORMS 2018 Annual Meeting Phoenix, USA
• 11thAnnual Trans-Atlantic Infraday Conference, Washington, USA
• Commodities and Energy Market Organization in the Energy Transition Context, IFP Energies nouvelles, Reuil-Malmaison, France

Other research activities

Refereeing activities

Energy Journal, Energy Economics.

Conference organization

• 43rd IAEE International Conference, Paris, France.
• 37th International Conference of the French Finance Association (AFFI), Nantes, France

2021-

Teaching assistant of Deep Learning: Models and Optimization, Graduate level, ENSAE. 6h (In French)

Times series econometrics, Graduate level, M2-EEET: Université Paris-Saclay, l’Université Paris Nanterre , l’IFP-School, l’Ecole des Ponts ParisTech. L’Ecole des Mines ParisTech. 18h (In French)

2020-

Coordinator of computer science courses, Graduate level, ENSAE.

Introduction to the Python Computer Language, Graduate level, ENSAE.

Introduction to the R Computer Language, Graduate level, ENSAE.

2019

Applied Econometrics, 1st year, Pantheon-Sorbonne Master In Economics, Université Paris 1 Panthéon-Sorbonne. 48h (in English)

2017-2018

Energy econometrics, Master of Science in Statistics: Risk Management and Financial Engineering Specialization, “Diplôme d’ingénieur – France’s Grandes écoles”, ENSAI. 8h (In French)

2017

Times series modeling, 1st year Bachelor Statistique et Informatique Décisionnelle (STID), Université Paris-Descartes. 54h (In French)