Forecast and Uncertainty Analysis of Production Decline Trends with Bootstrap and Monte Carlo Modeling

Abstract A new multi-step procedure was developed and tested for the analysis and forecast of production decline curves. It includes multidimensional grid search over values of nonlinear least squares errors, nonlinear least squares approximation, Monte Carlo and block bootstrap simulation of production trends. Several decline curve models were tested in grid search and iterative minimization: SEPD, extended Hyperbolic, Duong and the Power-Exponential. Grid search and iterative minimization worked equally well on all tested models. Multidimensional grid search finds a starting point for nonlinear iterative process. Then, iterative minimization finds parameters of an optimum approximating model. The approximating model is used for the forecasting of production trends. Block bootstrap and new Monte Carlo simulation methods produce third-level aggregated forecasts of production rate. In addition, these two methods characterize the range of possible values of predicted production (uncertainty range). Comparative analysis of Monte Carlo and block bootstrap simulation indicates that these methods are characterized by different widths of the uncertainty regions and by a certain mutual shift of predicted production curves. Thus, the joint use of both techniques may result in a more reliable forecast of future production.