An overview of computational modeling in agricultural and resource economics

Publication Type

Journal contribution (peer reviewed)

Authors

Nolan, J., Parker, D., Van Kooten, G. C., Berger, T.

Year of publication

2009

Published in

Canadian Journal of Agricultural Economics/Revue canadienne d' agroeconomie

Band/Volume

57 (4)/

DOI

10.1111/j.1744-7976.2009.01163.x

Page (from - to)

417-429

Keywords

sustainability

A major pillar of the field of agricultural and resource economics (referred to hereafter as agricultural economics) is its tradition of interdisciplinarity, especially in linking socioeconomic and biophysical processes. In contrast to economic analysis in other fields, agricultural economists are more likely to broaden their experimental perspective to include interaction or feedback between humans and the natural world. This orientation has led in large part to increased interest in studies of economic processes over both time and space, using both dynamic optimization and spatial analysis (Kennedy 1986;Miranda and Fackler 2002;Nelson 2002).

A second pillar of agricultural economics is its tradition of empirical testing of carefully derived hypotheses. However, when modeling human-environment interactions, economics in general has had difficulty linking traditional deductive theoretical models, which include just a few state variables and feedbacks for tractability, with inductive statistical models that include many independent variables but often exclude explicit representations of the underlying processes. Current analytical models are also limited in their ability to represent human learning and adaptation, a factor that is particularly important when future conditions depend heavily on the actions of other economic decision makers.

Agricultural economists recognize that individual and environmental heterogeneity are key components of dynamic human-natural systems. However, theoretical and econometric models remain somewhat limited in terms of their ability to portray heterogeneous decision-making individuals in a heterogeneous environment and in terms of modeling significant interactions between economic agents, where economic interaction can be generated by activities, such as resource transfers through local markets and imitative behavior, as well as through spatial externalities. In addition, current spatial models are often founded on the assumption that neighborhood conditions are fixed and that the supply or demand decision of a particular neighbor will not alter the spatial environment for a given individual, an assumption that rarely holds in reality. Relevant examples in agriculture include a decision to extract groundwater resources by one farmer that affects water availability for other farmers in a catchment area, or a decision by a farmer to rent out or sell land that will certainly affect the land rental options, sales, and production choices of neighboring farmers.

Concurrent with these issues, it is also the case that optimization-based farm and resource management models, often operating under short time scales with purely financial objectives, have become increasingly sophisticated over the past half century in part because of technical advances in computer hardware and software combined with improved training of students in mathematical modeling. Researchers continue to try to advance the ability of these models to capture economic and ecosystem uncertainty, irreversible thresholds (e.g., bankruptcy, destruction of shallow lakes due to excessive nutrients), as well as interpersonal (interhousehold, interfirm) and dynamic natural resource management challenges.

As the latter models have become more realistic and sophisticated, operating over longer time scales and incorporating higher degrees of human-environment feedback, they too have become more difficult to solve analytically. Ultimately, the response of the profession to all of the shortcomings described above has been the development of the field of computational economics.¹ As such, the field encompasses both numerical optimization (Glover and Laguna 1993) as well as simulation methods.

In this special issue, we examine recent advancements in two computational economics modeling approaches used within agricultural economics—stochastic dynamic programming (SDP) and agent-based modeling (ABM). SDP is used to address uncertainty in applied research, while ABM is a simulation methodology that is increasingly used increasingly in other social sciences (Berry et al 2002;Hernandez et al 2008;Waldrop 2009), particularly in cases where agents are heterogeneous and the system may be out of equilibrium. One goal of this special issue is to familiarize the wider agricultural economics profession with these powerful tools, while also providing a context with which many readers are likely familiar.

This introductory paper provides some background to these methods, albeit mostly in an informal and nontechnical fashion. In the next section, we provide a brief overview of how uncertainty is treated using SDP, as well as outline more recent advances in this field. In third section, ABM is described in more detail. Since these are computational methods, the fourth section briefly discusses software issues. Section fifth provides an overview of the applications of computational economics represented in this special issue, while sixth section concludes by listing some of the challenges facing modelers who may choose to use these computational methods.