Academic journal article Fuzzy Economic Review

How Do Different Time Spans Affect the Prediction Accuracy of Business Failure?

Academic journal article Fuzzy Economic Review

How Do Different Time Spans Affect the Prediction Accuracy of Business Failure?

Article excerpt

(ProQuest: ... denotes formulae omitted.)


The global financial crisis that started in 2007 in the United States of America, and the consequent increase of bankruptcy firms since then in many countries, has revived the interest in predicting companies' failure.

Even though business failure has been widely tackled in financial literature, most authors focused on applying new methodologies and selecting the most suitable financial ratios and other macro-economic variables to improve prediction.

However, there is much less attention on study the time dimension in business failure prediction. This can be approached through two perspectives: (1) in terms of number of years taken as inputs for prediction; and (2) with regard to the bankruptcy time horizon, i.e. the number of years between the end of the analyzed period and the bankruptcy.

In relation to the first approach, most studies use data from 1 to 4 years previous to the bankruptcy time horizon (Bellovary et al., 2007) but other authors (Blum, 1974) demonstrated that using financial data of 8 years improves the prediction accuracy.

On the other hand, as du Jardin (2015) has pointed out, it is possible to obtain positive prediction results using a bankruptcy time horizon of only one year, but these results are getting significantly worse when a medium term prediction is used. Therefore, a bankruptcy horizon of 1 year is the common approach (Bellovary et al., 2007). Nonetheless, other authors as Dwyer (1992) built methods with high success percentage using a 3-year bankruptcy horizon. In a similar way, du Jardin and Séverin (2011) also have proved that Self-Organizing Maps (SOM) can increase the period of bankruptcy prediction up to 3-years without losing predictive accuracy in business failure.

This study focuses on analyzing the first approach: the impact of the input data timeframe on the prediction accuracy of business failure when SOM models are applied in classifying healthy and bankrupt companies.

The paper is aimed to compare the error rate in identifying healthy and bankrupt companies from the Spanish chemical industry, through SOM methodology, using 3 years previous to the bankruptcy time horizon (2005-2007), by using 6 years (2002-2007) and 9 years (1999-2007). That will enable us to analyze the consequences of longer or shorter timeframe data sample, and to observe if the inclusion of former information provides added value or distorts the obtained results.

The remainder of this paper is organized as follows: Section 2 briefly reviews prior studies on business failure. In Section 3 we describe the data and methodology used. In Section 4 the empirical results are discussed. In the final section, we present the conclusions of the study.


Creditors, owners or managers of a company have always been concerned about business failure prediction. Back in the early part of the 20th century, some researchers tried to respond this problem by individual ratio analysis and subsequent comparison between bankrupt and healthy companies (FitzPatrick, 1932; Merwin, 1942; Smith and Winakor, 1935).

Nevertheless, was in the mid 1960's when the studies of Beaver (1966) by applying univariate techniques and Altman (1968) by using multi variant discriminant analysis awake the interest on this topic. Altman formulated two scenarios, taking data both from 1 and 2 years previously to bankruptcy.

In the 1970's and 1980's a large number of works were made applying different statistic techniques: Libby (1975) combined the principal components analysis (PCA) with discriminant analysis; Ohlson (1980) proposed a logit model to predict business failure; Zmijewski (1984) was pioneer on using a probit model by applying profitability related ratios only; or West (1985) that combined factorial analysis with a logit model. Most of papers in this period were designed using data that are measured over a period t to achieve a prediction over a period

t+1. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.