This paper presents a comparison of three distributed lag estimators: OLS, the Almon procedure, and the Hannan inefficient method. Each method is compared for sample sizes of 50 and 100 for several alternative distributed lag shapes and residual process structures. The results not only reveal the relative performance of these estimators, but also provide evidence on each method's performance under misspecification with respect to lag length and the residual process.

Thomas F. Cargill, Michael M. Hutchinson, Political Business Cycles with Endogenous Election Timing: Evidence from Japan, The Review of Economics and Statistics, Vol. 73, No. 4 (Nov., 1991), pp. 733-739

The stability of the demand function for money has received extensive attention over the past two decades. However, there is no precise meaning of the term stability in the literature. The issue is most often discussed in reference to time-series estimates of the function and generally based on three characteristics: 1) The demand for money can be explained by a small set of variables as determined by various statistical tests; 2) the function does not exhibit marked shifts over time; 3) the function is capable of generating reasonable forecasts outside of the interval of estimation. Stephen Goldfeld (1973) and John Boorman in exhaustive surveys conclude that relatively simple formulations of the demand for money yield stable shortand long-run functions. Despite some negative evidence (see William Poole), stability of the demand function has been fairly well accepted, at least up to the last few years (Goldfeld, 1976). The overwhelming majority of evidence is based on time-series models using constant coefficient estimation procedures. Yet, arguments can be developed to show that estimating a demand function for money via constant coefficient methods amounts to misspecification. The time varying characteristics of the demand function should be explicitly recognized in the estimation procedure to properly investigate the stability issue. This study is organized around two objectives: First, to use a theoretical model of risk preferences to develop the opportunity cost aspect of the demand function for money implying time varying coefficients and second, to provide within and outside sample comparisons of constant and variable coefficient estimates of various demand function specifications.