Small Sample Properties of Estimators of Distributed Lag Models

Teaching Regressions with a Lagged Dependent Variable and Autocorrelated Disturbances

“…(This pattern is similar to those disclosed in Maeshiro [ 1988]). However, the range of <1> 1 , over which the magnitude of the approximate bias is smaller than the magnitude of the dynamic effect (i.e ., the range over which the OLS estimator performs well precisely because of positive autocorrelation in the disturbances) , would be small even for a large value of A, except when the sample size is very small.…”

“…The second group consists of studies by Maddala and Rao ( 1973) and Maeshiro ( 1980), that employed both smoothly trended (e.g., U.S. real GNP series) and 80 JOURNAL OF ECONOMIC EDUCATION nontrended exogenous variables and obtained results from the trended variables that were quite different from those discussed above: consistently superior performance of the OLS estimator, confirming the role of factor 04. In particular, both studies revealed that the range of <j > 1 over which the OLS estimator performs better than an estimator that takes account of autocorrelation in the disturbances increases with A at an increasing rate, confirming the joint role of factors C2 and 02.…”

“…This is particularly true if an intercept is included. Maeshiro ( 1980), on the other hand. blames the detri-82 JOURNAL OF ECONOMIC EDUCATION mental effect of autoregressive transformation on smoothly trended exogenous…”

“…In other words, the OLS estimator should perform well over such a range precisely because of positive autocorrelation in the disturbances. This conjecture was substantiated, employing several different disturbance-generating processes, for the simple dynamic model (I) by using a Monte Carlo method in Maeshiro ( 1988) and also for the general dynamic model by using an approximate bias formula in Maeshiro ( 1994a). From these studies, one can safely conclude that when A is positive, the tradeoff between the two effects exists for any disturbance-generating process that creates positive correlation between U, and Y, .…”

A lagged dependent variable, autocorrelated disturbances, and unit root tests - peculiar OLS bias properties - a pedagogical note

“…Many Monte Carlo studies Dhrymes, 1971, Appendix;Hong and L'Esperance, 1973;Maddala and Rao, 1973; Maeshiro, 1980;Sargent, 1968, Wallis, 1967 have been performed on the following model:…”