1. Introduction
In recent years, the analysis of banking services’ prices in Italy has focused on loan interest rates
rather than deposit rates. This may be explained with the attention for the restrictive effects of monetary
policy and by the differences on variations in loan prices across Italian regions, particularly between the
South and the Center - North.2
In this paper we focus on bank deposit rates because they are still one of the main ways in which
Italian households invest their financial wealth. We consider the rates charged by banks in the provinces.
The empirical analysis is based on a large panel data set (more than 10,000 observations) for the years
1990-99.
Examining the determinants of deposit interest rates, this paper compares alternative econometric
packages for the estimation of the panel data. With our abundance of observations, many different
specifications have been estimated using the fixed-effects and the random-effects models. The purpose of
this work is to respond to the caveats about numerical accuracy raised by McCullogh and Vinod in the
June 1999 issue of the Journal of Economic Literature. The authors were very concerned about the scarce
attention paid to numerical accuracy in the selection of econometric packages. This choice might drive the
market by placing CPU (Central Processing Unit) performance before precision. Considering this
treacherous trade-off, we decided to compare the numerical values of the estimates of some popular
econometric software.
The paper is divided into six sections. Section two surveys the literature on factors influencing interest
rates on deposits, with particular reference to the USA, where the largest number of studies has been
conducted. Section three presents the data used in the regressions and the hypotheses we tested. Section
four discusses some econometric results of the determinants of interest rates on total deposits, current
accounts, savings accounts and certificates of deposit. Section five compares the numerical value of the
estimates of three of the most popular econometric packages featuring built-in panel data estimation
algorithms: LIMDEP, STATA and TSP. As a numerical benchmark we used Modeleasy Plus, a general-
purpose language containing matrix operations. Finally, Section six states the main conclusions. Appendix 1
documents some program listings we built. Appendix 2 describes the data used in the regressions.