An analysis of the petroleum industry’s inability to deliver on early production forecasts : shortcomings in probabilistic modeling

Date

2019-12

Authors

Petutschnig, David

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Previous studies have confirmed that production forecasts in the oil and gas industry are exposed to a variety of biases. This thesis extends those previous findings by investigating the quality of production forecasts for oil fields on the Norwegian Continental Shelf, which were approved between 1995 and 2017. The research focuses on optimism and overconfidence biases. Both biases are observable in the production forecasts provided by the Norwegian Petroleum Directorate. By comparing annual production data with production forecasts, it is possible to draw conclusions pertaining to the quality of those forecasts. A variety of methods are applied to investigate and illustrate the magnitude of those biases. The findings illustrate that the reason operators do not attain set project goals is because of aforementioned biases rather than unexpected events. The systemic inability to deliver on what was promised is observable through the lack of forecasting quality improvement over time. Two correction processes are proposed to reduce the encountered biases. A reference class is established to put past outcomes in a distributional setting. Uplift and scaling factors are drawn from the class to adjust the biased production forecasts. The results show a clear improvement in the quality of production forecasts through the use of reference class forecasting. A second process is introduced in which a Bayesian framework is suggested to calculate updated production forecasts. The same reference class is used to provide a prior distribution, which is then updated by the initial forecast (signal) to determine a posterior distribution. The posterior distribution exhibts on average a greater variance and a lower mean than the initial forecast. Therefore, the updated production forecasts are better calibrated and the impact of the biases is reduced. Limitations arise regarding the availability of additional data, however preliminary results from the analyses are encouraging. Drawing on past experience to debias production forecasts is of paramount importance

Description

LCSH Subject Headings

Citation