Normative considerations in the development of a software package for econometric estimation

Econometric Software

“…Only certain individual developers have ventured into print to any significant degree [19,63,64,113,115,160,163,202,203,208,209,225,230,244]. Furthermore, although the user guides and reference manuals commonly provided with individual programs often provide some aspects of their history, these accounts tend to be presented selectively, ordinarily without technical details.…”

A benchmark-free approach to access the accuracy of an OLS model calculation in an econometrically uncertain world

“…A comparison between the accuracy of single precision (real*4) and double precision (real*8) calculation in imposing the restriction E(e, x i ) = 0 will be illustrated using two representative data sets. One is the GE equation of the famous Grunfeld [16] data, discussed by Theil [24] and recently used as the basis for a number of benchmarks of various regression diagnostic tests by Renfro [17], which involves two variables on the right plus the constant. This problem appears to have minimal The logically equivalent command if(x.eq.0.0d+00)write(6,*)'x is in fact 0.0' will not work properly.…”

Econometric software: The first fifty years in perspective

“…The work at the Brookings Institution in the 1960s incorporated several related themes, ranging from the creation of databases by James Craig [79]-based upon programming by Mark Eisner and later partially utilizing the NBER database created by Charlotte Boschan and others [31,276], to the creation of software to condense Input-Output tables by McCarthy and Renfro [212], to the early development of econometric modeling packages such as MODLER [274,281,283] and PLANETS [32,33]. The first task to which MODLER was applied, and which initially caused its creation, was the estimation of the approximately 200 equation "condensed" Brookings model [115,116].…”

“…The increasing separation between users and developers of econometric software, as these two roles become more distinct, carries with it a number of implications. In addition to insuring that packages are numerically reliable, there is in particular both a need for this software to be readily useable [52,283] and the requirement to consider the implications of the way in which econometric techniques are implemented [227]. Some years ago, Kenneth Berk pointed out that although statistical "software should [have] the features to do readily what needs to be done," that in fact there is a "tendency for the user to do what is readily available in the software," that "…packages have enormous influence over…analysis, especially over [that of] the less sophisticated users" [25].…”