Review of Economic Studies (1990) 517, 167-182
© 1990 The Review of Economic Studies Limited
0034-6527/90/00100167 $02.00
, The Non-Parametric
Identification of Generalized
Accelerated Failure-Time Models
GEERT RIDDER
Rijksuniuersiteit Groningen
First uersion received August 1988; final version accepted November 1989 i Eds.)
We consider a class of models that generalizes the popular Mixed Proportional Hazard
(MPH) model for duration data: the Generalized Accelerated Failure-Time (GAFT) model. We
show that the GAFf model is non-parametrically identified (up to a normalization). We then
reconsider the non-parametric identification of the MPH model. We show that the class of MPH
models is not closed under normalization. This implies that a finite mean of the mixing distribution
is a necessary condition for (non-parametric) identification of the MPH model. It is impossible
to test this hypothesis without imposing arbitrary restrictions on the base-line hazard and/ or the
regression function.
1. INTRODUCTION
An econometric model usually characterizes the (conditional) distribution of some vari-
able(s) of interest by specifying the mean and variance of that distribution. The obvious
example is the linear regression model and its extensions. Even econometric models for
limited dependent variables are derived from latent regression models which specify the
(conditional) mean and variance of some latent variable. In contrast, econometric models
for duration data start from a specification of the hazard rate of the duration distribution.
There are several reasons for this. First, empirical distributions of duration data are
usually skewed, and their mean and standard deviation are of the same order of magnitude.
The simplest model that has been used to describe such data, the exponential distribution,
has a constant hazard rate. Second, economists are often interested in the variation of
the hazard rate with the elapsed duration and with explanatory variables. Economic
theories, e.g. job search theory, provide testable restrictions on the duration dependence
of the hazard rate. Third, time-varying explanatory variables can be easily included in
a specification of the hazard rate.
However, particularly in biomedical and life-testing applications, duration data have
been analysed with standard regression methods. The resulting duration models are called
Accelerated Failure-Time (AFT) models. Of course, regression methods have a strong
attraction to econometricians. If durations can be modelled by (non)linear regression
models, duration analysis becomes a part of "mainstream" econometrics, and simultaneity,
measurement error etc. can be studied in a more traditional way. In this paper we discuss
a class of duration models that can be seen as a generalization of the Accelerated
Failure-Time models. This class of models, the Generalized Accelerated Failure-Time
(GAFT) models, contains as special cases the popular Proportional Hazards (PH) and
the Mixed Proportional Hazards (MPH) models that have been used extensively in
economic research.
167
Downloaded from https://academic.oup.com/restud/article/57/2/167/1551576 by guest on 14 October 2021