Measurement Error and Seasonal Adjustment Error in Economic Time Series (Arma Models, Kalman Filtering).

It is well known that the presence of error in reported time series can distort estimates of relationships between the true variables. Owing much to the formidable methodological difficulties encountered when one recognizes the existence of errors, the most common econometric practice has been to ignore them. The effect of this misspecification, which will depend on the source of the error, should be assessed if one is interested in the design of estimators based on methodological, rather than pragmatic, criteria. For the case where errors are due to faulty measurement, we consider the effects of ignoring such error on the estimated parameters of some simple univariate and bivariate dynamic models. We find that such procedures typically produce estimates which seriously misstate the true parameters. For the prediction problem, the effect of ignoring measurement error is found to be less serious for the univariate model than the bivariate model. A maximum likelihood estimator which incorporates the measurement error is described, and demonstrated in an empirical example. When the error in our observations is due to errors in seasonal adjustment, the lack of clearly defined objectives of the adjustment procedure dictates a more limited approach. We present evidence that casts some doubt on the specification of the X-11 procedure as a moving average with constant weights. The results suggest that the linear characterization of the most commonly used seasonal adjustment procedure is incomplete.PhDEconomicsUniversity of Michiganhttp://deepblue.lib.umich.edu/bitstream/2027.42/161204/1/8702680.pd