Forecasting methods in oil & gas sector. Optimised Theta model and application on annual oil and gas demand data of european countries

Abstract

In this thesis, forecasting methods in Oil&Gas industry are studied, relying on historical data. Focus is given on oil and gas yearly demand of European countries and the Theta Model is used, a time series extrapolative forecasting method that topped the M3-Competition 20 years ago, the largest empirical forecasting competition till that date; and performed very good in the recent M4-Competition in 2018. Annual historical data are used, taken from iea and Rystad, from 1990 to 2018 and the Theta method is applied to all time series using a simple excel spreadsheet to construct the model. The fit, as well as the accuracy of the method for the 2019 demand value, are measured with sMAPE and are compared to a benchmark of three simple commonly used methods (Naive, MA3, MA5) and two ready forecasting methods of excel (FRC.LIN, FRC.ETS). The Theta Model is then optimised for each time series using Solver in Excel in order to define the combination of parameters, number of θ lines, value of each θ and weight in the forecast, that offer the best fit, through sMAPE minimization. This combination is used to forecast the 2019 demand value and forecast error is measured. Finally, we forecast the annual demand until 2024 and compare it with the respective forecasts of Rystad that we have in our possession. The excel spreadsheet that simulates the above and adjusts the respective calculations on data entered by the user, can act as a general free tool for country annual oil&gas demand forecasting, achieving very good point forecasts, without complex calculations or effort need to be done by the user. This tool, in which annual data are simply placed as input and a button is pressed so that Solver automatically extracts the best describing Theta model for the data entered, minimizes error and gives us reliable forecast. Thus, it can be used by users who desire to forecast oil or gas annual demand. In each case, the model is driven by the amount and differentiation of annual input data and the minimization of total sMAPE of forecasts and results are verified by the optimisation tool Solver.