논문

One of the central aims of research in econometrics or social sciences is to
define basic principles whereby available information may be used to forecast
future economic and social conditions or parameters. This paper presents an
example of how quantitative methods can be applied in forecasting research.
To more effectively demonstrate the ability of forecast, it employs the real retail
pork price data that could be obtained through USDA.

In the following, the step by step development of two alternative price models is
outlined. These alternatives result from the statistical weakness observed in
simple OLS pre-test model which was initially investigated. The revised models
prescribed here, a restricted first-order lag model and a Durbin two-stage
feasible generalized least squares model, are derived from two different
assumptions about the origin of the statistical problems in the simple OLS model.
Even though they make these different assumptions about the source of error, the
specifications of both the lag and FGLS models originate from an identical, more
general structure, which was used as a pretest for each of these final models.

Ideally one of these models would be preferred in part due to better statistical
properties and at least partially on the basis of some knowledge about the pork
economy and relationships within meat markets. In this investigation, however,
after the models were specified, they were compared and evaluated primarily on
the basis of their forecast abilities. The same general strategy employed to arrive
at and assess these pork price forecast models can be used in a wide spectrum
of other research issues in numerous fields of study.