FORECASTING ABILITIES OF INDIVIDUAL PETROLEUM EXPLORERS: PRELIMINARY FINDINGS FROM CROWDSOURCED PROSPECT ASSESSMENTS

Abstract

Petroleum explorers regularly make numerous forecasts for prospects and wells, but the quality of their predictions has inadequate documentation. Here I discuss the forecasting abilities of individual explorers inferred from a survey about future exploration wells, some of which were drilled in late 2018 and 2019. A total of 104 petroleum explorers provided 7,068 answers about eleven wells and about themselves, and subsets from this dataset were used to study their predictions. The survey participants were diverse, and most had M.Sc. or Ph.D. degrees with >16 years of industry experience working as individual contributors and/or middle managers for oil companies. Assessments of the geological probability of success (PoS) by different explorers for the same well were highly variable, and average assessments poorly discriminated between future discoveries and dry holes. Point (binary or multiple choice) forecasts by explorers participating in the survey were, on average, only slightly better than random guessing. The participants' highest academic degree had little apparent influence on the quality of the point forecasts. Years of experience resulted in only slightly better correlation with increasing quality of forecasts. Survey participants more familiar with the prospects (those who generated them or evaluated proprietary company data) made, on average, worse exploration forecasts than those who studied only the limited publicly available information provided with the survey. The duration of time spent on the survey did not affect the quality of forecasts. The aggregated opinion of all explorers (wisdom of the crowd) may be beneficial in assessments of the geological PoS, but perhaps not so when forecasted outcomes have binary or multiple possibilities. The generally poor forecasting abilities of individual petroleum explorers may surprise some decision-makers and investors. However, many of the study results are based on relatively small datasets and should be treated as preliminary. Further research with larger datasets is necessary to replicate, validate and explain the findings of this study.