5. Comparison between some econometric packages
The last two decades have shown a rapid development of software packages featuring complex
estimation procedures, without the need for in-depth knowledge of the numerical algorithms and computer
science. Most of the econometric and statistical software packages commonly used in academic and
business environments have both an interactive and a programmable interface. The main selection criteria
for these products are:
1) usability;
2) richness of the commands most frequently required;
3) cost.18
To the authors’ knowledge there has been only one Panel Data software review in the recent past;
this is the paper “Software review” by Pierre Blanchard, in “Econometrics of Panel Data”, 1992. The
justification for our work is twofold:
a) the impressive pace at which these packages have developed in the last decade;
b) the lack of numerical applications in Blanchard’s paper.
Unlike Blanchard, we did not take into account the RATS software package because it does not
address the panel data estimation problem in a straightforward way.
Quite often, for a number of different reasons, the numerical accuracy of the answer supplied by the
software is overlooked or even neglected. McCullogh and Vinod (1999) gave some examples in which the
numerical answers produced by alternative econometric and statistical packages are completely different19.
Given the importance of this issue, it has become the economist’s responsibility to validate the results. The
natural question that arises is how to evaluate the numerical accuracy of a complex algorithm. The very
simple idea, adopted in this paper, is to compare the results obtained using three different econometric
packages available in the research department of the Bank of Italy. The packages compared are:
18 In this paper we do not address this issue.
19 The authors do not mention the name of the packages used.