Multistage stochastic programming models for the portfolio optimization of oil projects
MetadataShow full item record
Exploration and production (E&P) involves the upstream activities from looking for promising reservoirs to extracting oil and selling it to downstream companies. E&P is the most profitable business in the oil industry. However, it is also the most capital-intensive and risky. Hence, the proper assessment of E&P projects with effective management of uncertainties is crucial to the success of any upstream business. This dissertation is concentrated on developing portfolio optimization models to manage E&P projects. The idea is not new, but it has been mostly restricted to the conceptual level due to the inherent complications to capture interactions among projects. We disentangle the complications by modeling the project portfolio optimization problem as multistage stochastic programs with mixed integer programming (MIP) techniques. Due to the disparate nature of uncertainties, we separately consider explored and unexplored oil fields. We model portfolios of real options and portfolios of decision trees for the two cases, respectively. The resulting project portfolio models provide rigorous and consistent treatments to optimally balance the total rewards and the overall risk. For explored oil fields, oil price fluctuations dominate the geologic risk. The field development process hence can be modeled and assessed as sequentially compounded options with our optimization based option pricing models. We can further model the portfolio of real options to solve the dynamic capital budgeting problem for oil projects. For unexplored oil fields, the geologic risk plays the dominating role to determine how a field is optimally explored and developed. We can model the E&P process as a decision tree in the form of an optimization model with MIP techniques. By applying the inventory-style budget constraints, we can pool multiple project-specific decision trees to get the multistage E&P project portfolio optimization (MEPPO) model. The resulting large scale MILP is efficiently solved by a decomposition-based primal heuristic algorithm. The MEPPO model requires a scenario tree to approximate the stochastic process of the geologic parameters. We apply statistical learning, Monte Carlo simulation, and scenario reduction methods to generate the scenario tree, in which prior beliefs can be progressively refined with new information.