Premium
Use and misuse of unobserved components in economic forecasting
Author(s) -
Maravall Agustín
Publication year - 1994
Publication title -
journal of forecasting
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.543
H-Index - 59
eISSN - 1099-131X
pISSN - 0277-6693
DOI - 10.1002/for.3980130209
Subject(s) - estimator , econometrics , component (thermodynamics) , autoregressive integrated moving average , series (stratigraphy) , recession , economics , computer science , stock (firearms) , stylized fact , time series , statistics , mathematics , engineering , mechanical engineering , paleontology , physics , macroeconomics , biology , keynesian economics , thermodynamics
The paper deals with unobserved components in economic time series within a general model‐based approach. The component, its final estimator, and the preliminary one (which also includes the forecast) are seen to follow different ARIMA models, which can be expressed in terms of the series innovations. Analytical expressions are derived for the different types of associated errors. Two applications are presented. The first shows how the use of unobserved components can increase substantially forecasting precision, and how the model‐based approach can rigorously answer questions of applied concern. The second application illustrates the dangers of using unobserved components in some macroeconomic models. It is first shown how unobserved component estimators, such as a series seasonally adjusted with XII or with a model‐based procedure, will most likely be non‐invertible, and hence invertible models (for example, a VAR model) are not appropriate for them. Second, the recent Stock and Watson model aimed at forecasting recessions is used to illustrate how probabilities computed over the distribution of the component, of its final estimator, and of its preliminary one will be poor estimators of each other. As a consequence, the recession forecasts will be systematically biased.