Ratio based failure prediction models for the Small-Medium Enterprises (SMEs) in the UK

Although SMEs represent over 99% of companies in the UK, there has been limited researchinto SME failure prediction modelling. This study develops failure prediction modelsspecifically for liquidated SMEs using financial ratios. Despite the fact that there is a cleardistinction between small sized (SEs) and medium sized (MEs) companies (in terms of assets,turnovers and employees size), the majority of studies developed models for SMEs as onegroup, but not as two different groups.In order to capture the predictive power of ratios in each group, there should be differentprediction models for both sizes. The main question this study aims to answer: are there anydifferences between SEs and MEs predictive variables. To answer this question the study will1) identify the financial variables for each group and examine their predictive power, 2)compare the classification accuracy of three different statistical techniques namely, MultipleDiscriminant Analysis (MDA), Logistic Regression (LR), and Probabilistic Neural Network(PNN). And finally 3) testing reliability level and validation of the different ratio basedprediction models.In order to include all categories of ratios, the data sample consists of 560 SMEs thatdisclosed full financial statements, and were liquidated in the period 2000-2007. The samplewas divided into three groups; SEs, MEs and a combined SME group for size-specificpredictions. A range of financial ratios were selected and examined, two sample procedureswere tested to validate the results.Profitability is found to be the most important predictor for the SEs group, while liquidity isfor the MEs. The overall accuracy of the three methods (MDA, LR, PNN) for each model is:SEs model (70%, 71%, 81%), MEs model (74%, 76%, 87%), and combined SME model(72%, 74%, 84%). For the first time, we reported size-specific financial ratio predictors forpredicting SMEs failure. The results support our main question that there are differencesbetween SEs and MEs predictive variables. Further studies are needed to explore the natureof these findings.

Level dc:type.qualificationlevel

Doctoral (Level 8)

Year dc:date.issued

2013

Author dc:creator

• Khanji, IMJ

Language dc:language

en

Identifier

oai:salford-repository.worktribe.com:1338176

OAI identifier oai:identifier

oai:salford-repository.worktribe.com:1338176