Models of firm dynamics and the hazard rate of exits: reconciling theory and evidence using hazard regression models

    Research output: Working paper

    Abstract

    This paper considers empirical work relating to models of firm dynamics. We show that a hazard regression model for firm exits, with a modification to accommodate age-varying covariate effects, provides an empirical framework accommodating many of the features of interest in studies on firm dynamics. Modelling implications of some of the popular theoretical models are considered and a set of empirical procedures for verifying testable implications of the theoretical models are proposed. The proposed hazard regression models can accommodate negative effects of initial size that go to zero with age (active learning model), negative initial size effects that fall with age but stay permanently negative (passive learning model), conditional and unconditional hazard rates that decrease with age at higher ages, and adverse effects of macroeconomic shocks that decrease with age of the firm. The methods are illustrated using data on quoted UK firms. Consistent with the active learning model, the effect of initial size is significantly negative for a young firm and falls to zero with age. The hazard function conditional on size, other firm- and industry-level characteristics, and macroeconomic conditions decreases with age only at higher ages, but shows the weaker property of Increasing Mean Residual Life over its entire life-duration. Instability in exchange rates affects survival of very young firms strongly, and the effect decreases to insignificant levels for older firms.
    Original languageEnglish
    PublisherEconWPA
    Publication statusPublished - 2005

    Publication series

    NameEconometrics

    Keywords

    • firm exit
    • learning
    • firm dynamics
    • non-proportional hazards
    • hazard regression models

    Fingerprint Dive into the research topics of 'Models of firm dynamics and the hazard rate of exits: reconciling theory and evidence using hazard regression models'. Together they form a unique fingerprint.

    Cite this