Debiasing Production Forecasts Through Reference Class Forecasting

Abstract

This thesis investigates past performance of production forecasts provided by operators on the NCS at the time of project sanction. Utilising a dataset comprising annual forecasted and actual production from 1995 to 2017, we demonstrate that operators on the NCS exhibit considerable optimism and overconfidence biases in their production forecasts. To debias these production forecasts, we develop and implement a reference class forecasting (RCF) methodology with the goal of producing well-calibrated forecasts. The debiased forecasts that are generated from this process are evaluated through a series of tests, providing strong evidence for bias reduction and enhanced forecasting performance. Prior to applying RCF adjustments,only 33 % of all observations of actual productionin the first six years fall within the 80% confidence interval defined by the forecasts. Applying RCF significantly reduces the overconfidence bias as the adjusted 80% confidence interval now captures 77% of the actual production levels. Moreover, RCF increases the fraction of fields whose actual production exceed the P50 estimate from 37% to 47%, implying reduced optimism.