On the Spatial Decomposition of Forecasts
Jackson, Randall W., Sonis, Michael, Geographical Analysis
This paper uses historical socioeconomic data to evaluate the elasticities of a Dendrinos-Sonis one-population/two-locations nonlinear dynamic comparative advantage model for determining subregional shares of aggregate regional forecasts. The fractal dimension properties of the discrepancies between the actual and simulated data are used to enhance the forecasting framework. The analysis focuses in the first phase on total population, total personal income, and earnings by sector for the Columbus, Ohio Metropolitan Statistical Area, and Delaware County, one of its component counties. For the second phase of the analysis, these data for nondurable goods and for services are used, along with monthly data for total employment in Columbus, Cincinnati, and Cleveland, Ohio. Annual data for the analysis are drawn from the U.S. Bureau of Economic Analysis, Regional Economic Information System, while the monthly data come from the Ohio Bureau of Employment Securities. The nonlinear dynamic model is shown to outperform conventional approaches for the majority of socioeconomic stocks.
Defining regions for quantitative analysis is a fundamental task in geographical analysis. Haggett, Cliff, and Frey (1977) included an extended discussion of types of regions and considerations in selecting regional definitions more than two decades ago, and regional analysts continue to confront the regional definition issue each time a new model is constructed. Statistical problems associated with analytical scale and modifiable a real units (Fotheringham and Rogerson 1993) accentuate the importance of regional definition to any quantitative analysis. The problem of regional definition is central to a wide range of problem domains.
In modeling regional economic systems, analysts have traditionally attempted to define functional economic regions (Berry and Horton 1970). Functional, or nodal, regional boundaries are drawn so as to maximize the ratio of internal (within region) to external (between region) bonds. To delineate a functional economic region, boundaries are placed in such a way as to maximize the ratio of internal to external economic interactions. These objectives are often sought, however, within an environment constrained both by political considerations and data availability. In the first instance, policy relevance can dictate that a model provide results for an administrative region, while in the second, economically rational regional definitions often do not coincide with data reporting units.
The combination of objectives and constraints can present the analyst with a variety of challenging problems. This paper focuses on one of these challenges, and assesses the utility of a particular approach to its solution. We address the general problem of providing economic projections for administrative regions smaller than the functional economic region in which they are located. The approach we consider is a discrete nonlinear relative dynamic model of comparative advantage.
The problem context is elaborated in section 1 of the paper, using a discussion of the Columbus, Ohio, Metropolitan Statistical Area (MSA) as a point of departure and vehicle for identifying the dimensions of the problem. Although the Columbus MSA provides an empirical context for this research, the Columbus economy, per se, is not the focus of the analysis. Our primary intent is to explore and assess the features and characteristics of the comparative advantage model as an indicator of its potential applications to a broader range of problem domains.
Section 2 provides a formal presentation of the theoretical model. An exploratory analysis based on the classic nonlinear model formulation is presented in the third section. Section 4 presents an extension of the classic model, and is followed in the final section by discussion, summary and conclusions.
1. PROBLEM CONTEXT
The Columbus MSA, shown in Figure 1, is typical of many nodal regions in the United States and elsewhere. …