Estimating financial distress with a dynamic model: Evidence from family owned enterprises in a small open economy

Abstract

Employing earnings shortfall as a financial distress indicator, we formulate a dynamic nonlinear model, implementing Wooldridge's conditional maximum likelihood estimator and accounting for potentially endogenous covariates. Likewise, we not only achieve a significant improvement in consistency and classification accuracy over static approaches, but we also manage to understand better the evolution of the financial distress process. In our sample of Greek listed firms the higher the positive performance and the lower the leverage at the initial period the greater the chance that a company enters financial distress further down the road, possibly due to manager–owner overconfidence and debt-imposed discipline by company's creditors.

Introduction

Estimating the onset of corporate financial distress (FD) has assumed indisputably wide-ranging usefulness and applicability, stemming from its significant ramifications for company stakeholders in cases of failure to meet debt obligations and payment default, on the one side, and to corporate restructuring, asset sales and debt workouts, on the other² (Hotchkiss et al., 2008). Vis-à-vis its importance, there has been extensive efforts to understand its economic drivers, performance and arising conflicts of interest, both in and out of court. Herein, we propose a new methodology to better understand FD on the one hand and to accurately predict its occurrence on the other, employing a dynamic nonlinear model which incorporates valuable information regarding past FD records.

The majority of research effort has been devoted so far in understanding and predicting bankruptcy and debt payment legal default, which paves the way for asset and debt restructurings, in or out-of-court, notwithstanding the potential for severe conflicts arising between managers, shareholders and creditors (see Hotchkiss et al., 2008 for a survey). Nevertheless, numerous researchers, including Wruck (1990), Asquith et al. (1994), Andrade and Kaplan (1998), Platt and Platt (2006), Jostarndt and Sautner (2008) and Pindado et al. (2008) have focused on obtaining a measure of FD likelihood (FDL), not necessarily entailing a formal bankruptcy filing or payments in arrears, but rather a situation of distress recognized by evidence of financial shortcomings from published accounts. Defining FDL this way includes as stress indicators, indispensably an earnings (either EBITDA – earnings before interest, tax, depreciation and amortization adjusted – or EBIT – earnings before interest and taxes) shortfall to cover financial expenses (FE), coupled in some cases with additional indicators, such as layoffs, reducing market valuation, negative EBIT or negative net income before special items. Its usefulness lies in the fact that it is independent of the eventual outcome, but consistent with an ex-ante approach (Pindado et al., 2008). Moreover, upon its diagnosis, FDL incorporates the potential for continuous reassessments prior to the occurrence of its ultimate resolution, facilitating the ex-ante corporate or stakeholder planning of possible remedies. The helpfulness of doing so may be seriously justified by the fact that firms in FD are considerably more probable to go bankrupt or be acquired.³

It is worth noting that there exist several distinctive features between the aforementioned definition of FDL and bankruptcy or payments default: the latter (bankruptcy and payments default) are (i) more closely related to corporate death than the former, and (ii) more closely resemble a one-off incident, modeled as an “absorbing barrier”⁴; in contrast, FDL may last for several years, especially in the case when it coexists with relative economic under-performance (Kahl, 2002). Moreover, the past record towards FD does matter, as there is an early warning of a potentially forthcoming distress-related bankruptcy or acquisition many years before its ultimate formal resolution (Ro et al., 1992). By defining likewise FDL, analysts are provided with an early distress warning tool, useful for an ex-ante FD estimation approach.

The majority of such estimation approaches have used static estimation techniques (e.g., the typical panel logit or probit) that fail to account for the full sample FD evolution dynamics; a solution proposal is the employment of hazard models that do use the full past record of FD and model the eventual bankruptcy (Campbell et al., 2008). Nevertheless, this very definition and staying nature of FD (an FE-EBIT shortfall may persist in several accounting years) sets estimation techniques via hazard models questionable for modeling the dynamics of the process. Up to corporate disappearance (the absorbing barrier) hazard models assume a different process than further on, where observations are treated as a new beginning of the process, not accounting for a dependence on previous FD history.

Given the importance of accounting for the FD dynamics, we employ a dynamic nonlinear panel model specification proposed by Wooldridge (2005) (conditional maximum likelihood estimator) which accounts for unobserved heterogeneity in a dynamic discrete choice framework, which may be easily formulated in widely available software. In the existing FDL literature, the strong dynamic dependence of FD on previous-year outcomes has been (partly) circumvented by limiting its definition to earnings shortfalls lasting two years (Platt and Platt, 2006, Pindado et al., 2008, Jostarndt and Sautner, 2008), in essence de-trending its one-period state-persistence. Nevertheless, a conceptual setback in employing such a biennial FDL indicator is the unavoidable treatment of the rest of the observations as a “non-FDL group”. This categorization fails to differentiate within this group between companies not experiencing biennial financial shortcomings, those managing to overcome FD problems of the previous period with a positive turnaround, and those that have only recently (for just one period) entered FD.

In our study we explore the degree to which a dynamic nonlinear panel model may help us to improve our understanding of the evolution of FD within firms as well as accurately classifying a firm's FD or FDL. The data we used consist of a panel with the entire listing of non-financial, non-government-owned firms on the Athens Stock Exchange, from 1993 to 2004 and from 2005 to 2009 inclusive The reason for this data partitioning is the mandatory introduction of International Financial Reporting Standards (IAS) for listed companies, beginning on 2005, which has altered the information content of reported accounts vis-à-vis the domestic accounting standards previously used. Likewise we shall be able to better evaluate the consistency of our results comparing the two subsamples and understand the validity of our approach under the new information environment. Furthermore, we screened our panel for reported bankruptcies, FD-related trading suspensions and supervisions by stock exchange authorities as well as announcements of significant corporate restructurings; likewise, we strived to establish which of the two definitions closely follows the more salient FD outcomes for the investment community.

As a next step, we formulate a dynamic nonlinear model for FD with explicit dependence on the initial period (Wooldridge, 2005), correcting for possible endogenous dependence between random error and covariates (Mundlak, 1978). We use the same explanatory variables used in previous studies of FDL (Pindado et al., 2008) namely profitability, leverage, retained earnings and time-dummies. The fact that previous studies of FDL have focused on major worldwide markets (US and G7 countries) (Pindado et al., 2008) coupled with the necessity to develop a consistent and stable model of FDL for different countries and time periods (Grice and Dunkan, 2001), renders the Greek stock market, vis-à-vis its particularities, an excellent opportunity to test previous methodologies “out of sample” and develop an understanding of managerial handling of FD under “family capitalism” (Morck and Steier, 2005). Moreover, the validity of the new econometric methodology we propose, namely a dynamic versus static formulation, depends on whether our dependent variable – financial distress – exhibits state dependency; given the indisputably persistent nature of financial distress evidenced in mainstream markets (Ro et al., 1992) we are convinced that our econometric approach is applicable in a wide range of environments globally.

Our introduction of covariates accounting for the inter-temporal process dynamics, such as initial period exogenous covariates, is expected to help us better understand the historical evolution of FD and the economic reasons behind it. In particular, the effect of initial conditions of the process on later developments may provide us with helpful insights into the implications of early managerial decisions down the corporate road. Vis-à-vis managerial agency conflicts with shareholders (Jensen and Meckling, 1976) and possibly problems caused by overconfidence (Hackbarth, 2002), the deeper into corporate history we dig the more likely we might uncover useful pieces of information regarding FD evolution.

Finally, we perform model simulations with data up to one year preceding the last (including 2003 for our first pre-IAS sample, and 2008 for our second post-IAS sample), and measure the accuracy of simulated responses on our last data year (2004 and 2009, respectively). Hence, we check with our real-world data whether our suggested approach is helpful in predicting FDL.

The following consists of Part 2 which describes our data and the characteristics of the Greek stock exchange, Part 3, describing our modeling framework, Part 4, providing and discussing our results, and Part 5, with our conclusions and suggestions for further research.