Development of a model to predict financial distress of companies listed on the JSE

Muller, Grant Henri (2008-03)

Thesis (MBA (Business Management))--University of Stellenbosch, 2008.

Thesis

ENGLISH ABSTRACT: To date, there has been significant research completed on the topic of corporate financial distress. Two pioneering researchers in the field of predicting financial distress was Beaver in 1966 and Altman in 1968. More recent research, based on companies listed on the JSE has been that of Steyn-Bruwer and Hamman (2006). This project, based on the latter authors’ work, has been formulated with one main research objective and two subordinate research objectives. The main research objective is to prove that different modelling techniques provide better prediction accuracies than others. The two subordinate research objectives are firstly to prove that there is a difference in the overall predictive accuracy if the data (provided by Steyn-Bruwer and Hamman) is subdivided according to “year before failure” and not according to economic period and secondly to prove that more optimised, independent variables would provide a better overall predictive accuracy. This research report summarises several significant papers on the topic; and draws the conclusion that research on financial distress is fragmented with very little consensus on any of the major definitions, assumptions and findings. In order to contextualise these differences; this research report defines and discusses corporate financial distress and considers the major issues associated with the field of research. An interesting observation from the literature survey was the fact that existing literature does not readily take consideration of the number of Type I and Type II errors made. As such, this research report introduces a novel concept (not seen in other research) called the “Normalised Cost of Failure” (NCF) which takes cognisance of the fact that a Type I error typically costs 20 to 38 times that of a Type II error. In order to satisfy the main research objective several different modelling techniques were selected based on their popularity in the literature surveyed. They are: Multiple Discriminant Analysis (MDA), Recursive Partitioning (RP), Logit Analysis (LA) and Neural Networks (NN). A summary of each of the different techniques is provided in Chapter 4 of this research report. The research by Steyn-Bruwer and Hamman forms the departure point for this research and their work is summarised in Chapter 5 of this report. Chapters 6, 7 and 8 use the data from Steyn-Bruwer and Hamman along with the above mentioned modelling techniques to verify the main and subordinate objectives. In terms of the main research objective, the results of these chapters show that the different analysis techniques definitely produce different predictive accuracies. Here, the MDA and RP techniques correctly predict the most “failed” companies; and consequently have the lowest NCF. This research report also shows that LA and NN provide the best overall predictive accuracy. In terms of the first subordinate research objective; this research shows that using the year before failure rather than the economic period as a subdivision provides superior predictive accuracy. With regard to the second subordinate research objective: there is no difference in the predictive accuracies if the independent variables are further optimised. These results were disappointing and consequently disprove the second subordinate objective that widening the number of input variables actually improves the predictive accuracy. In fact, the results indicate that the information contained in the independent variables seems to saturate after the most important (key predictor) independent variables have been included in the model. It is important to take cognisance of the fact that each predictive technique has its own strength and weakness. It is proposed by the author that the strengths and weaknesses of these predictive techniques be combined to provide a better overall predictive methodology.

AFRIKAANSE OPSOMMING: Heelwat betekenisvolle navorsing oor die onderwerp van maatskappye se finansiële verknorsing is tot op hede voltooi. Twee baanbreker-navorsers op die gebied van vooruitskatting van finansiële verknorsing was Beaver in 1966 en Altman in 1968. Meer onlangse navorsing, gebaseer op maatskappye wat op die JSE genoteer is, was dié van Steyn-Bruwer en Hamman (2006). Hierdie navorsingsverslag, gebaseer op die voorgenoemde outeurs se werk, is geformuleer met een hoofnavorsingsdoelwit en twee ondergeskikte navorsingsdoelwitte. Die hoofnavorsingsdoelwit is om te bewys dat verskillende modelleringstegnieke beter voorspellingsakkuraatheid as andere het. Die twee ondergeskikte navorsingsdoelwitte is, eerstens, dat daar ʼn verskil is in die oorhoofse voorspellingsakkuraatheid as die data (verskaf deur Steyn-Bruwer en Hamman) onderverdeel word volgens die “jaar voor mislukking” eerder as volgens die ekonomiese tydperk; en tweedens, om te bewys dat meer geoptimiseerde, onafhanklike veranderlikes kan lei tot ʼn beter oorhoofse voorspellingsakkuraatheid. Ten einde hierdie verskille te konseptualiseer, het hierdie navorsingsverslag finansiële mislukkings van maatskappye bespreek en gedefinieer en aandag geskenk aan die belangrikste aspekte geassosieer met die navorsingsveld. ʼn Interessante waarneming uit die literatuurstudie was die feit dat die huidige literatuur selde indien enige, oorweging skenk aan die aantal Tipe I- en Tipe II-foute wat gemaak word. As sulks het hierdie navorsingsprojek ʼn nuwe begrip (nog nie in ander navorsing gesien nie) ontwikkel, wat beskryf word as die “Genormaliseerde Kostefaktor”; wat die feit dat ʼn Tipe I-fout tipies 20 tot 38 maal die koste van ʼn Tipe II-fout beloop, in ag neem. Ten einde te voldoen aan die hoofnavorsingsdoelwit is verskillende modelleringstegnieke wat op grond van hul gewildheid in die literatuur voorgekom het, gekies. Hulle is: Meervoudige Diskriminantanalise (MDA), Herhalende Verdeling (RP), Logit-Analise (LA) en Neurale Netwerke (NN). ʼn Opsomming van elk van hierdie verskillende tegnieke word in Hoofstuk 4 van hierdie navorsingsverslag verskaf. Die navorsing wat deur Steyn-Bruwer en Hamman gedoen is, vorm die vertrekpunt van hierdie navorsing en hulle werk is gevolglik in Hoofstuk 5 van hierdie verslag opgesom. Hoofstukke 6, 7 en 8 gebruik die data van Steyn-Bruwer en Hamman tesame met die bovermelde modelleringstegnieke ten einde die hoof- en ondergeskikte doelwitte te bewys. In terme van die hoofnavorsingsdoelwit, het die resultate van hierdie hoofstukke getoon dat die verskillende analitiese tegnieke definitief verskillende voorspellingsakkuraatheid oplewer. Hier het die MDA- en RP-tegnieke die grootste aantal mislukte maatskappye korrek voorspel, en gevolglik die laagste Genormaliseerde Kostefaktor gehad. Die navorsingsverslag toon ook dat LA en NN die beste oorhoofse akkuraatheid van voorspelling het. In terme van die eerste ondergeskikte navorsingsprobleem het hierdie navorsing getoon dat, om die jaar voor mislukking te gebruik as onderverdeling, eerder as die ekonomiese tydperk, beter voorspellingsakkuraatheid het. Wat die tweede ondergeskikte navorsingsdoelwit betref, is daar bevind dat daar geen verskille in die voorspellingsakkuraatheid bestaan as die individuele veranderlikes verder geoptimaliseer word nie. Hierdie resultate was teleurstellend en het gevolglik die tweede ondergeskikte probleem, naamlik dat as die aantal inset-veranderlikes sou vergroot word, dit die vooruitskattingsakkuraatheid behoort te kan verhoog, verkeerd bewys. Tewens, die resultate het getoon dat die inligting soos vervat in die onafhanklike veranderlikes klaarblyklik versadiging bereik nadat die belangrikste (hoof-vooruitskatter) onafhanklike veranderlikes in die model opgeneem is. Dit is belangrik om kennis te neem van die feit dat elke vooruitskattingstegniek sy eie sterk en swak punte het. Die skrywer stel dus voor dat hierdie sterk- en swakpunte gekombineerd gebruik word om ʼn beter oorhoofse vooruitskattingsmetodologie daar te stel.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/823
This item appears in the following collections: