Contact us | Terms of use | Privacy Policy | |


Oil Price Models

Objective & Purpose

Provide a short- to long-term outlook for oil prices allowing for the possibility to run different simulation exercises according to cyclical perspectives.

TAC uses two different approaches:

Medium-term

This approach relies on a strong theoretical background to determine the price of finite resources.

The model provides yearly average prices of Brent, from a series of relations and economic determinants. TAC uses econometric techniques (co-integration, error correction model), over a very long period. The variables are grouped into three distinct components:

  • "Hotelling" variables: a set of variables that captures, for each oil exporter, an optimal strategy of evolution of prices in the long-term depending on its "discount rate", or its degree of preference for present. This discount rate depends on the relationship between proven reserves and the country’s population.
  • "Adelman" variables: a set of variables that captures the substitution effects between energy sources based on prices, and the effects on energy efficiency of economic growth.
  • "Business cycle" variables: more traditional indicators of economic trends in the global economy

    •     TAC provides to customer:    

        Observed    
    2007
    70$
    72$
    2008
    82$
    97$
    2009
    55$
    62$
    2010
    74$
    80$
    2011
    109$
    111$
    2012
    112$
    112$

    Short-term

    This approach relies on an efficient statistical calibration in terms of advanced indicators for future dynamics. TAC combines many data mining models (fundamental and market variables), including:

  • Recursive Partitioning for the identification of critical thresholds and the selection of "critical" variables.
  • Random Forest and bootstrapping/bagging techniques for improving the stability of estimated models over time.
  • Self Organizing Maps to identify shapes on oil prices, related to economic patterns.
  • Support Vector Machine and Neural Network for the supervised learning (learning for past/historical prices).
  • ...
    • TAC provides to customer:
    Prediction vs actual

    Publication

  • 26th European Conference on Operational Research Presentation given by TAC ...