Vector Autoregressions: Forecasting and Reality
介绍了一种基于六变量向量自回归模型的美国宏观经济预测方法，重点解决数据发布滞后、季度与月度数据匹配等实际应用中的技术难题，为商业和政府预测提供可复现的模型框架。
Constructing forecasts of the future path for economic series such as real gross domestic product growth, inflation, and unemployment forms a large part of applied economic analysis for business and government. Model-based forecasts are easier to replicate and validate by independent researchers than forecasts based on expert opinion alone. In addition, the forecaster can formally investigate the source of systematic errors in model forecasts, and a forecast model s performance can be established before it is used by a decision maker. ; The authors of this article describe a particular model-based forecasting approach, a vector autoregression comprising six U.S. macroeconomic variables. They focus attention on the technical hurdles that must be addressed in a real-time application and methods for overcoming those hurdles, such as conditional forecasting to handle the staggered release of data and matching quarterly with monthly data. ; By emphasizing the practical problems of forecasting economic data using a statistical model, the authors draw on experience in using such a model at the Federal Reserve Bank of Atlanta. Although the model studied is small and highly aggregated, it provides a convenient framework for illustrating several practical forecasting issues. The focus on a simple model provides potential users with a road map of how one might implement such a forecasting model in specific applications.