WEAI/AERE 2012 - Individual Paper Abstract
Title: Learning about Regime Shifts in Resource Management
Author(s): Craig BOND, Department of Agricultural and Resource Economics, Colorado State University, B311 Clark, Fort Collins, CO 80523-1172, USA, 970-491-6951, 970-491-2067, craig.bond at colostate dot edu [Photo credit: based on Retrieval of zebra mussel-encrusted Vector Averaging Current Meter near Michigan City, IN. Lake Michigan, June 1999. Photo by M. McCormick]
Abstract:
Management of complex ecosystems is often fundamentally a problem of choosing actions in the face of large degrees of uncertainty distinct from mere stochasticity of the underlying physical processes that generate environmental outcomes. For example, managers may have competing theories regarding the underlying structure of the system, such as the shallow lakes model in [1], may be uncertain about a key structural parameter or parameters, as in the climate change model considered by [2], or may be subject to a partially-observable process in which an underlying state cannot be observed, as in the invasive species scenario considered by [3]. In essence, problems of this sort essentially introduce a new margin over which managers must trade off; namely, the potential for endogenous learning about the future through deviations in the management strategies that would have otherwise been 'optimal' if current beliefs were to persist. In this manner, the manager can experiment by taking actions that generate a beneficial data series in terms of information content, and process this information to reduce uncertainty about the state of the system in the future, thus providing a tangible economic benefit [4]. In systems characterized by thresholds and irreversibilities, however, the uncertainty may be even deeper, in that the optimizing agent(s) may not know for certain which regime is generating the ultimate outcomes, and the potential for experimentation and data generation may be limited by the fact that there is an 'absorbing process' that cannot change once the threshold is crossed (or that can switch back to an original regime according to some other process).
In this paper, a renewable resources model with thresholds and irreversibility is analyzed under conditions where the decision-maker is not certain of the regime state of the system, due to either a stochastic data generation process or the lack of observability of the switching event. Beliefs over the unknown state are modeled in a state-space Bayesian framework, rendering the problem as a partially-observable Markov decision model.
We find that the relaxation of the assumption of resolution of past regime uncertainty changes the major conclusion in [5]; i.e., that precautionary behavior is not necessarily optimal when uncertainty about the regime persists. Further, we offer an additional explanation for non-monotonicity in precautionary behavior similar to that in [6]. A final, more general conclusion is that the information regime under which agents operate, and the ability to either exogenously or endogenously manipulate this regime in the future, has important implications for ex ante optimal policy in the presence of potential regime shifts.