Empirical comparison of hazard models in predicting SMEs failure

Abstract

This study aims to shed light on the debate concerning the choice between discrete-time and continuous-time hazard models in making bankruptcy or any binary prediction using interval censored data. Building on the theoretical suggestions from various disciplines, we empirically compare widely used discrete-time hazard models (with logit and clog-log links) and continuous-time Cox Proportional Hazards (CPH) model in predicting bankruptcy and financial distress of the United States Small and Medium-Sized Enterprises (SMEs). Consistent with the theoretical arguments, we report that discrete-time hazard models are superior to continuous-time CPH model in making binary predictions using interval censored data. Moreover, hazard models developed using failure definition based jointly on bankruptcy laws and firms’ financial health exhibit superior goodness of fit and classification measures, in comparison to models that employ failure definition based either on bankruptcy laws or firms’ financial health

Does size matter in predicting SMEs failure?

“…Holmes, Hunt, and Stone (2010) estimate hazard functions separately for microenterprise and SMEs and find that the effect of variables on the survival of these two types of firms is substantially different. Notwithstanding, the effort spent in assessing a firm's financial situation, it would appear that there is room in the literature (e.g., Gupta, Gregoriou, & Ebrahimi, 2018) for approaches to directly identify factors affecting both financial distress and bankruptcy across different size categories of SMEs. The relationship between SMEs asset size and insolvency risk is not linear, when controlling for company size using total asset value (Altman et al, 2010).…”

“…Following Gupta et al (2018), we then introduce each covariate in turn into the multivariate model in increasing order of the rank of their AME, the rationale being the higher the value of AME, the higher the change in the predicted probability due to a unit change in the covariate. The dependent variable has a binary outcome with financially distressed/bankrupt equal to "1" and "0" otherwise, whereas independent variables are the set of covariates found to be significant in the univariate regression analysis.…”

Does size matter in predicting hedge funds' liquidation?

“…These covariates are novel to hedge funds' failure literature and exploit the information content of funds' relative size, average growth, past performance, volatility of tail risk and past liquidation rate to predict hedge funds' liquidation for up to two years. Some of these covariates are adapted from previous studies on corporate bankruptcy by Gupta et al () and Campbell et al (), but are suitably modified to fit the scope of this study. Further details on all main covariates are as follows:…”

“…The standard error of a multivariate regression model increases with an increase in the number of covariates, and this subsequently adds to the numerical instability of the model (Hosmer Jr et al, ; Gupta et al, ). This also makes the model more dependent on the observed data (Hosmer Jr et al, ).…”

“…For the first time, we explore information content of funds' relative size ( WRSIZE ), growth in assets under management ( GROWTH ), total value of downside and upside returns ( DRATIO ), volatility of tail risk ( TRISK ) and past liquidation rate of hedge funds ( INDRISK ) in predicting liquidation likelihood of hedge funds. Some of these covariates are motivated from studies on corporate bankruptcy by Campbell et al () and Gupta et al (), however they are suitably modified to fit the purposes of this study. Following the existing literature, we also investigate the discriminatory power of funds' own performance ( RETURN ) and their past 60 month winning ratio ( WRATIO ) in predicting hedge funds' liquidation.…”