Financial Early Warning System Model and Data Mining Application for Risk Detection 2012 Expert Systems With Applications

Expert Systems with Applications 39 (2012) 6238–6253

Contents lists available at SciVerse ScienceDirect

Expert Systems with Applications

journal homepage: www.elsevier .com/locate /eswa

Financial early warning system model anddata mining application for risk detection

Ali Serhan Koyuncugil a,⇑, Nermin Ozgulbas b

a Capital Markets Board of Turkey, Research Department, Eskisehir Yolu 18.km., Ankara, Turkeyb Baskent University, School of Health Sciences, Department of Healthcare Management, Eskisehir Yolu 20.km., Ankara, Turkey

a r t i c l e i n f o

Keywords:CHAIDData miningEarly warning systemsFinancial riskFinancial distressSMEs

0957-4174/$ - see front matter � 2011 Elsevier Ltd. Adoi:10.1016/j.eswa.2011.12.021

⇑ Corresponding author. Tel.: +90 5326657084; faxE-mail address: [email protected] (A.S. Koy

a b s t r a c t

One of the biggest problems of SMEs is their tendencies to financial distress because of insufficientfinance background. In this study, an early warning system (EWS) model based on data mining for finan-cial risk detection is presented. CHAID algorithm has been used for development of the EWS. DevelopedEWS can be served like a tailor made financial advisor in decision making process of the firms with itsautomated nature to the ones who have inadequate financial background. Besides, an application ofthe model implemented which covered 7853 SMEs based on Turkish Central Bank (TCB) 2007 data. Byusing EWS model, 31 risk profiles, 15 risk indicators, 2 early warning signals, and 4 financial road mapshas been determined for financial risk mitigation.

� 2011 Elsevier Ltd. All rights reserved.

1. Introduction

All enterprises especially SMEs need to think about globaldimensions of their business earlier than ever. Especially in devel-oping countries, in addition to the administrative insufficiencies,competition, economical conditions, the permanent threat towardsSMEs from globalization, and financial crisis have caused distressand affect firms’ performance.

SMEs are defined as enterprises in the non-financial businesseconomy (NACE, Nomenclature statistique des activités économi-ques dans la Communauté européenne (Statistical classification ofeconomic activities in the European Community)) that employ lessthan 250 persons. The complements of SMEs – enterprises that em-ploy 250 or more persons – are large scale enterprises (LSEs). With-in the SME sector, the following size-classes are distinguished:

� Micro enterprises, employing less than 10 persons.� Small enterprises, employing at least 10 but less than 50

persons.� Medium-sized enterprises that employ between 50 and 250

persons.

This definition is used for statistical reasons. In the Europeandefinition of SMEs two additional criteria are added: annual turn-over should be less than 50 million €, and balance sheet totalshould be less than 43 million € (Commission Recommendation,2003/361/EC).

SMEs play a significant role in all economies and are the keygenerators of employment and income, and drivers of innovation

ll rights reserved.

: +90 3122466670.uncugil).

and growth. Access to financing is the most significant challengesfor the creation, survival and growth of SMEs, especially innova-tive ones. The problem is strongly exacerbated by the financialand economic crisis as SMEs have suffered a double shock: adrastic drop in demand for goods and services and a tighteningin credit terms, which are severely affecting their cash flows(OECD, 2009). As a result, all these factors throw SMEs in financialdistress.

The failure of a business is an event which can produce substan-tial losses to all parties like creditors, investors, auditors, financialinstitutions, stockholders, employees, and customers, and itundoubtedly reflects the economics of the countries concerned.When a business with financial problems is not able to pay itsfinancial obligations, the business may be driven into the situationof becoming a non-performing loan business and, finally, if theproblems cannot be solved, the business may become bankruptand forced to close down. Those business failures inevitably influ-ence all businesses as a whole. Direct and indirect bankruptcy costsare incurred which include the expenses of either liquidating or anattempting to reorganize businesses, accounting fees, legal feesand other professional service costs and the disaster broadens toother businesses and the economics of the countries involved(Ross, Westerfield, & Jordan, 2008; Terdpaopong, 2008; Warner,1977).

The awareness of factors that contribute to making a businesssuccessful is important; it is also applicable for all the related par-ties to have an understanding of financial performance and bank-ruptcy. It is also important for a financial manager of successfulfirms to know their firm’s possible actions that should be takenwhen their customers, or suppliers, go into bankruptcy. Similarly,firms should be aware of their own status, of when and where theyshould take necessary actions in response to their financial

http://dx.doi.org/10.1016/j.eswa.2011.12.021

mailto:[email protected]

http://dx.doi.org/10.1016/j.eswa.2011.12.021

http://www.sciencedirect.com/science/journal/09574174

http://www.elsevier.com/locate/eswa

A.S. Koyuncugil, N. Ozgulbas / Expert Systems with Applications 39 (2012) 6238–6253 6239

problems, as soon as possible rather than when the problems arebeyond their control and reach a crisis.

Therefore, to bring out the financial distress risk factors intoopen as early warning signals have a vital importance for SMEsas all enterprises. There is no specific method for total preventionfor a financial crisis of enterprises. The important point is to set thefactors that cause the condition with calmness, to take correctiveprecautions for a long term, to make a flexible emergency plantowards the potential future crisis.

The aim of this paper is to present an EWS model based on datamining. EWS model was developed for SMEs to detect risk profiles,risk indicators and early warning signs. Chi-Square AutomaticInteraction Detector (CHAID) Decision Tree Algorithm was in thestudy as a data mining method. Remaining of this paper is orga-nized as follows: Section 2 presents definition of EWS. Section 3contains data mining model for risk detection and early warningsystem. Implementation of data mining for risk detection and earlywarning signals is presented in Section 4. Concluding remarks andstrategies were suggested in Section 5.

2. Financial early warning systems

An early warning system (EWS) is a system which is using forpredicting the success level, probable anomalies and is reducingcrisis risk of cases, affairs transactions, systems, phenomena, firmsand people. Furthermore, their current situations and probablerisks can be identified quantitatively (Ozgulbas & Koyuncugil,2010). Financial EWS is a monitoring and reporting system thatalerts for the probability of problems, risks and opportunities be-fore they affect the financial statements of firms. EWSs are usedfor detecting financial performance, financial risk and potentialbankruptcies. EWSs give a chance to management to take advan-tage of opportunities to avoid or mitigate potential problems.Nearly, all of the financial EWSs are based on financial statements.Balance sheets and income tables are the data sources that reflectthe financial truth for early warning systems. In essence, the earlywarning system is a financial analysis technique, and it identifiesthe achievement analysis of enterprise due to its industry withthe help of financial ratios.

The efforts towards the separation of distressed enterprisesstarted with the z-score that are based on the usage of ratios byBeaver (1966) for single and multiple discriminant analysis of Alt-man in 1968. The examples of other important studies that usedmulti variable statistical models, are given by Deakin (1972), Alt-man, Haldeman, and Narayanan (1977), Taffler and Tisshaw(1977) with the usage of multiple discriminant model; are alsogiven by Zmijewski (1984), Zavgren (1985), Jones (1987), Panta-lone and Platt (1987), with the usage of logit and probit models;are at the same time given by Meyer and Pifer (1970) with theusage of multiple regression model. Beside the business distressedstudies, researchers focused on monitoring ongoing situations todetect sudden changes or unexpected risk factors of enterprises.These attempts made important the early warning systems forresearch. Some of previous studies conducted in SMEs, banks,insurers, i.e., and their research methods are presented below.

Brockett and Cooper (1990) developed an EWS by using neuralnetwork method. The model was developed with 24 variablesfirstly, and then the numbers of variables were decreased to 8.These variables were equities, capitalization ratio, return on assets,turnover of assets, account receivables to equities, changing ofloses, and debt to current assets.

Lee and Urrutia (1996) compared the models of logit, hazard,neural networks and discriminant for developing an early warningsystem. They found different indicators or signs for each model.Also they determined that forecast power of all models were same.

Barniv and Hathorn (1997) developed an early warning modelbased on logistic regression by evaluating the studies of Triesch-mann and Pinches (1973), Ambrose and Seward (1998), and Barnivand McDonald (1992) in insurance firms.

Laitinen and Chong (1999) presented a model for predicting cri-ses in small businesses using early-warning signals. Study summa-rized the results of two separate studies carried out in Finland (with72% response) and the UK (26%) on the decision process of corpo-rate analysts (Finland) and bank managers (UK) in predicting thefailure of small and medium-sized enterprises (SMEs). Both studiesconsisted of seven main headings and over 40 sub-headings of pos-sible factors leading to failure. Weighted averages were used forboth studies to show the importance of these factors. There weresignificant similarities in the results of the two studies. Manage-ment incompetence was regarded as the most important factor, fol-lowed by deficiencies in the accounting system and attitudetowards customers. However, low accounting staff morale was con-sidered a very important factor in Finland but not in the UK.

Yang, Ling, Hai, and Jing (2001) used artificial neural networks(ANN) for detecting financial risk of banks as an early warning,and tested the method.

Salas and Saurina (2002) compared the determinants of prob-lem loans of Spanish commercial and savings banks in the period1985–1997, taking into account both macroeconomic and individ-ual bank level variables. The GDP growth rate, firms, and familyindebtedness, rapid past credit or branch expansion, inefficiency,portfolio composition, size, net interest margin, capital ratio, andmarket power are variables that explain credit risk. The findingsraised important bank supervisory policy issues: the use of banklevel variables as early warning indicators, the advantages of bankmergers from different regions, and the role of banking competi-tion and ownership in determining credit risk.

Edison (2003) developed an operational early warning system(EWS) that can detect financial crises. The system monitored sev-eral indicators that tend to exhibit an unusual behavior in the peri-ods preceding a crisis. When an indicator exceeded (or falls below)a threshold, then it was said to issue a ‘‘signal’’ that a currency cri-sis may occur within a given period. The model was tested in 1997/1998 crises, but several weaknesses to the approach were identi-fied. The paper also evaluated how this system can be applied toan individual country. The results suggested that an early warningsystem should be thought of as a useful diagnostic tool.

El-Shazly (2003) investigated the predictive power of an empir-ical model for an early warning system of currency crises. EWS em-ployed qualitative response models within a signals frameworkthat monitors the behavior of key economic variables and issuesa warning when their values exceed certain critical levels. Authorconducted a case study in Egypt. Results showed that this model,and in particular the extreme value model, captured to a good ex-tent the turbulence in the foreign exchange market and the onsetof crises.

Jacobs and Kuper (2004) presented an EWS for six countries inAsia. Financial crises were distinguished in three types; currencycrises, banking crises, and debt crises. The significance of the indi-cator groups was tested in a multivariate logit model on a panel ofsix Asian countries for the period 1970–2001. Author founded thatsome currency crises dating schemes outperform others by usingEWS.

Berg, Borensztein, and Pattillo (2004), developed early warningsystem models of currency crisis for Mexican and Asian crisis.Since the beginning of 1999, IMF staff has been systematicallytracking, on an ongoing basis, various models developed in-houseand by private institutions, as part of its broader forward-lookingvulnerability assessment. This study examined in detail at the per-formance of these models in practice. The forecasts of the in-housemodel were statistically and economically significant predictors of

6240 A.S. Koyuncugil, N. Ozgulbas / Expert Systems with Applications 39 (2012) 6238–6253

actual crises. On the whole, the short-horizon private sector mod-els examined performed poorly out of sample, despite stellarin-sample performance.

Brockett, Golden, Jang, and Yang (2006) examined the effect ofstatistical model and neural network methods to detect financiallytroubled life insurers. They considered two neural network meth-ods; back-propagation (BP), and learning vector quantization(LVQ), two statistical methods; multiple discriminant analysis,and logistic regression analysis. The results showed that BP andLQV outperform the traditional statistical approaches.

Abumustafa (2006) detected early warning signs for predictingcurrency crises in Egypt, Jordan and Turkey. The study proposedreal exchange rate, exports, imports, trade balance/gross domesticproduct (GDP), foreign liabilities/foreign assets, domestic realinterest rate, world oil prices, and government consumption/GDPas indicators to predict currency risk. The results showed that allcrises were predictable, and EWSs should use for detecting crises.

Kyong, Tae, Chiho, and Suk (2006) presented the constructionprocess of a daily financial condition indicator (DFCI), which canbe used as an early warning signal using neural networks and non-linear programming. The procedure of DFCI construction was com-pleted by integrating three sub-DFCIs, based on each financialvariable, into the final DFCI. The study, then examined the predict-ability of alarm zone for the financial crisis forecasting in Korea.

Katz (2006) proposed to use EWS and early warning signs.Study listed often common warning signs and the best ways tosolve the problems. These are: payroll taxes, sales tax, and otherfiduciary obligations; communications with executive manage-ment and company leaders; accounts receivable; customers andproduct profitability; accounts payable; inventory, management;for capital-intensive or manufacturing operations; and checks asan indicator of problems.

Koyuncugil and Ozgulbas (2007a) aim to develop d a financialearly warning model for the SMEs listed in Istanbul Stock Exchange(ISE) in Turkey by using data mining. Authors conducted anotherstudy (2007b) and detected early warning signs for financial risk.A data mining method, Chi-Square Automatic Interaction Detector(CHAID) Decision Tree Algorithm, was used in the study for finan-cial profiling and detecting signs. The study covered 697 SMEslisted in ISE between 2000 and 2005. As a result of the study, thecovered SMEs listed in ISE were categorized into 19 financial pro-files and it was determined that 430 of them had poor financialperformance, in other words 61.69%. According to the profiles ofSMEs in financial distress, return on equity (ROE) will be a financialearly warning signal for SMEs listed in ISE.

Koyuncugil and Ozgulbas (2008a) emphasized the affect andimportance of operational risk in financial distressed of SMEs,beside the financial risk. Authors developed an early warning mod-el that qualitative (operational) and quantitative (financial) data ofSMEs taken into consideration. During the formation of system; aneasy to understand, easy to interpret and easy to apply utilitarianmodel that is far from the requirement of theoretical backgroundwas targeted by the discovery of the implicit relationships betweenthe data and the identification of effect level of every factor. Thismodel was designed by data mining.

Koyuncugil and Ozgulbas (2009a) developed a financial earlywarning model that detected operational risk factors for hedgingfinancial risk. For this purpose study used CHAID (Chi-Square Auto-matic Interaction Detector) Decision Trees. The study covered6.185 firms in Organized Industrial Region of Ankara in 2008. Itwas found that firms should emphasize the educational back-ground of managers, status of managers, annual turnover, operat-ing length of firms, makers of financial strategies, expenditure ofenergy, knowledge about BASEL-II, quality standards, and usageof credit as operational risk factors for hedging operational riskand raising financial performance.

Koyuncugil and Ozgulbas (2009b) to develop an intelligentfinancial early warning system model based on operational andfinancial risk factors by using data mining for SMEs in Turkey. Thismodel was aimed to not remain in theoretical structure, be practi-cable for SME, and available for the utilization of SMEs managers.According to model, financial data of Turkish SMEs was obtainedby means of financial analyses of balance sheets and income state-ments through Turkish Central Bank. Operational data couldn’t beaccess by balance sheets and income statements was collected by afield study from SMEs. Next step of model was analyzed thefinancial and operational data by data mining and detecting earlywarning signs.

Davis and Karim (2008a) successful predicted a majority ofbanking crises in emerging markets and advanced countries in1970–2003. Karim also, suggested that logit was the most appro-priate approach for global EWS and signal extraction for countryspecific EWS.

Davis and Karim (2008b) searched to assess whether earlywarning systems based on the logit and binomial tree approacheson the UK and US economies could have helped to warn about thecrisis. The study suggested that a broadening of approaches ofmacro prudential analysis was appropriate for early warning.

It can be seen that risk detection oriented early warningsystems have a very large implementation domain from the pic-ture given above. Furthermore, last generation Business Intelli-gence approach data mining accelerated the accuracy of thosesystems. Operational logic of early warning systems is based onfinding unexpected and extraordinary behaviors in subject area.On the other hand, data mining is the way of uncover previouslyunknown, useful and valuable knowledge, patterns, relations frombig amount of data via sophisticated evolutionary algorithms ofclassical techniques such as statistics, pattern recognition, artificialintelligence, machine learning. The definitions of EWS and datamining given lead an interesting similarity.

An EWS developed for SMEs must design according to the needsof SMEs managers. Therefore, system must be easy to understandand easy to use, must design according to financial and operationalrisk factors (as banks and BASEL II requirements), and must beintelligence for using update data.

3. Data mining model for risk detection and early warningsystem

The identification of the risk factors by clarifying the relation-ship between the variables defines the discovery of knowledge.Automatic and estimation oriented information discovery processcoincides the definition of data mining. Data mining is the processof sorting through large amounts of data and picking out relevantinformation. Frawley, Piatetsky-Shapiro, and Matheus (1992) hasbeen described data mining as ‘‘the nontrivial extraction of impli-cit, previously unknown, and potentially useful information fromdata’’. Also, Hand, Mannila, and Smyth (2001) described data min-ing as ‘‘the science of extracting useful information from large datasets or databases’’. Data mining, the extraction of hidden predictiveinformation from large databases, is a powerful new technologywith great potential to help companies focus on the most impor-tant information in their data warehouses. Data mining tools pre-dict future trends and behaviors, allowing businesses to makeproactive, knowledge-driven decisions. The automated, prospec-tive analyses offered by data mining move beyond the analysesof past events provided by retrospective tools typical of decisionsupport systems. Data mining tools can answer business questionsthat traditionally were too time consuming to resolve. They scourdatabases for hidden patterns, finding predictive information thatexperts may miss because it lies outside their expectations (Thear-ling, 2004).

Fig. 1. Data flow diagram of the EWS.


Koyuncugil and Ozgulbas (2010) described the data mining as‘‘collection of evolved statistical analysis, machine learning andpattern recognition methods via intelligent algorithms which areusing for automated uncovering and extraction process of hiddenpredictional information, patterns, relations, similarities or dissim-ilarities in (huge) data’’.

Data mining is used by business intelligence organizations, andfinancial analysts to get information from the large data sets. Datamining in relation to enterprise resource planning is the statisticaland logical analysis of large sets of transaction data, looking forpatterns that can aid decision making (Monk & Wagner, 2006).Today, data mining technology integrated measurement of differ-ent kinds of is moving into focus to measure and hedging risk. Datamining techniques have been successfully applied like fraud detec-tion and bankruptcy prediction by Tam and Kiang (1992), Lee, Han,and Kwon (1996), Kumar, Krovi, and Rajagopalan (1997), strategicdecision-making by Nazem and Shin (1999) and financial perfor-mance by Eklund, Back, Vanharanta, and Visa (2003), Hoppszallern(2003), Derby (2003), Chang, Chang, Lin, and Kao (2003), Kloptch-enko et al. (2004), Magnusson, Arppe, Eklund, and Back (2005).Also, some earlier studies of Koyuncugil and Ozgulbas (2006a,2006b, 2006c, 2007a, 2007b, 2008a, 2008b, 2009a, 2009b), Ozgul-bas and Koyuncugil (2006, 2009, 2010) conducted on financial per-formance, financial risk and operational risk of Small and MediumEnterprises (SMEs) and hospitals by data mining.

Fayyad, Piatetsky-Shapiro, and Symth (1996), proposed mainsteps of DM:

� Retrieving the data from a large database.� Selecting the relevant subset to work with.� Deciding on the appropriate sampling system, cleaning the data

and dealing with missing fields and records.� Applying the appropriate transformations, dimensionality

reduction, and projections.� Fitting models to the preprocessed data.

Data mining techniques can yield the benefits of automation onexisting software and hardware platforms, and can be imple-mented on new systems as existing platforms are upgraded andnew products developed. When data mining tools are imple-mented on high performance parallel processing systems, theycan analyze massive databases in minutes. The most commonlyused techniques in data mining are (Koyuncugil, 2006; Thearling,2004):

� Artificial neural networks: Non-linear predictive models thatlearn through training and resemble biological neural networksin structure.� Decision trees: Tree-shaped structures that represent sets of

decisions. These decisions generate rules for the classificationof a dataset. Specific Decision Tree methods include Classifica-tion and Regression Trees (CART) and Chi Square AutomaticInteraction Detection (CHAID).� Genetic algorithms: Optimization techniques that use process

such as genetic combination, mutation, and natural selectionin a design based on the concepts of evolution.� Nearest neighbor method: A technique that classifies each

record in a dataset based on a combination of the classes ofthe k record (s) most similar to it in a historical dataset. Some-times called the k-nearest neighbor technique.� Rule induction: The extraction of useful if-then rules from data

based on statistical significance.

Decision trees are tree-shaped structures that represent sets ofdecisions. The Decision Tree approach can generate rules for theclassification of a data set. Specific Decision Tree methods include

Classification and Regression Trees (CART) and Chi Square Auto-matic Interaction Detection (CHAID). CART and CHAID are DecisionTree techniques used for classification of a data set. They provide aset of rules that can be applied to a new (unclassified) data set topredict which records will have a given outcome. CART typicallyrequires less data preparation than CHAID (Lee & Siau, 2001).

During the developing EWS; an easy to understand, easy tointerpret and easy to apply utilitarian model that is far from therequirement of theoretical background is targeted by the discoveryof the implicit relationships between the data and the identifica-tion of effect level of every factor. Because of this reason, data min-ing is the ideal method for financial early warning system.

Developing an EWS for SMEs focused segmentation methods.The main approach in analysis is discovering different risk levelsand identifying the factors affected financial performance. Bymeans of Chi-Square metrics CHAID is able to separately segmentthe groups classified in terms of level of relations. Therefore, leavesof the tree have not binary branches but as much branches as thenumber of different variables in the data. So, it was deemed conve-nient to use CHAID algorithm method in the study.

CHAID modeling is an exploratory data analysis method used tostudy the relationships between a dependent measure and a largeseries of possible predictor variables those themselves may inter-act. The dependent measure may be a qualitative (nominal or ordi-nal) one or a quantitative indicator. For qualitative variables, aseries of chi-square analyses are conducted between the depen-dent and predictor variables. For quantitative variables, analysisof variance methods are used where intervals (splits) are deter-mined optimally for the independent variables so as to maximizethe ability to explain a dependent measure in terms of variancecomponents (Thearling, 2004).

Model of EWSThe model of EWS based on data mining and data flow diagram

of the EWS is shown in Fig. 1.The steps of the EWS are:

� Step I: Preparation of data collection.� Step II: Implementation of DM method.


� Step III: Determination of risk profiles.� Step IV: Identification for current situation of SMEs from risk

profiles and early warning signs.� Step V: Description of roadmaps for SMEs.

The details of the EWS are given below:

Step I: Preparation of data collection
Financial data that are gained from balance sheets: Items of bal-ance sheets will be entered as financial data and will be used tocalculate financial indicators of system: � Calculation of financial indicators like in Table 1.� Reduction of repeating variables in different indicators to
solve the problem of collinearity/multicollinearity.� Imputation of missing data.� Solution of outlier and extreme value problem.

Step II: Implementation of DM method
In the scope of the methods of data mining � Logistic regression.� Discriminant analysis.� Cluster analysis.� Hierarchical cluster analysis.� Self Organizing Maps (SOM),� Classification and Regression Trees (C& RT).� Chi-Square Automatic Interaction Detector (CHAID).can be the principal methods, in addition to this several classi-
fication/segmentation methods can be mentioned. However, dur-ing the preparation of an early warning system for SMEs, one ofthe basic objectives is to help SME administrators and decisionmakers, who does not have financial expertise, knowledge of datamining and analytic perspective, to reach easy to understand, easyto interpret, and easy to apply results about the risk condition oftheir enterprises. Therefore, Decision Tree algorithms that areone of the segmentation methods can be used because of their easyto understand and easy to apply visualization. Although, severalDecision Tree algorithms have widespread usage today, CHAID isseparated from other Decision Tree algorithms because of thenumber of the branches that are produced by CHAID. Other Deci-sion Tree algorithms are branched in binary, but CHAID manifestsall the different structures in data with its multi-branched charac-teristic. Hence, the method of Chi-Square Automatic InteractionDetector (CHAID) is used in the scope of this study.

Assume that X1,X2, . . . ,XN�1,XN denote discrete or continuousindependent (predictor) variables and Y denotes dependent vari-

Table 1Financial indicators.

Financial variables

Current ratioQuick ratio (liquidity ratio)Absolute liquidityInventories to current assetsCurrent liabilities to total assetsDebt ratioCurrent liabilities to total liabilitiesLong term liabilities to total liabilitiesEquity to assets ratioCurrent assets turnover rateFixed assets turnover rateDays in accounts receivablesInventories turnover rateAssets turnover rateEquity turnover rateProfit marginReturn on equityReturn on assets

able as target variable where X1 2 [a1,b1], X2 2 [a2,b2], . . . ,XN

2 [aN,bN] and Y 2 {Poor,Good}. While ‘Poor’ shows poor financialperformance in red1 bar and ‘Good’ shows good financial perfor-mance in green bar. CHAID Decision Tree is given in Fig. 2.

In Fig. 2 we can see that,� Only 3 variables of N have a statistically significant relation-

ship with the target Y,� X1 has most statistically significant relation with target Y,� X2 has statistically significant relation with X1 where,

X1 6 b11.� X3 has statistically significant relation with X1 where

b11 < X1 6 b12.

Step III: Determination of risk profiles
CHAID algorithm organizes Chi-square independency test amongthe target variable and predictor variables, starts from branchingthe variable which has the strongest relationship and arranges sta-tistically significant variables on the branches of the tree due to thestrength of the relationship. An example of a CHAID Decision Treeis seen in Fig. 2. As it is observed from Fig. 2, CHAID has multi-branches, while other Decision Trees are branched in binary. Thus,all of the important relationships in data can be investigated untilthe subtle details. In essence, the study identifies all the differentrisk profiles. Here the term risk means the risk that is causedbecause of the financial failures of enterprises.Fig. 2 shows that there are six risk profiles:Profile B1 shows thatThere are n11 samples where X1 6 b11 and X2 6 b21% m111 has poorfinancial performance,% m211 has good financial performance.Profile B2 shows thatThere are n12 samples where X1 6 b11 and X2 > b21% m112 has poorfinancial performance,% m212 has good financial performanceProfile C1 shows thatThere are n21 samples where b11 < X1 6 b12 and X3 6 b31% m121 haspoor financial performance,% m221 has good financial performance.Profile C2 shows thatThere are n21 samples where b11 < X1 6 b12 and X3 > b31% m122 haspoor financial performance,% m222 has good financial performance.Profile D shows thatThere are n3 samples where b12 < X1 6 b13% m13 has poor financialperformance,% m23 has good financial performance.Profile E shows thatThere are n4 samples where X1 > b13% m14 has poor financial perfor-mance,% m24 has good financial performance.If all of the profiles are investigated separately,Profile B1 shows that if any firm’s variables X1 and X2 have valueswhere X1 6 b11 and X2 6 b21, poor financial performance rate or inanother words risk rate of the firm will be RB1 = m111.Profile B2 shows that if any firm’s variables X1 and X2 have valueswhere X1 6 b11 and X2 > b21, poor financial performance rate or inanother words risk rate of the firm will be RB2 = m112.Profile C1 shows that if any firm’s variables X1 and X3 have valueswhere b11 < X1 6 b12 and X3 6 b31 poor financial performance rateor in another words risk rate of the firm will be RC1 = m121.Profile C2 shows that if any firm’s variables X1 and X3 have valueswhere b11 < X1 6 b12 and X3 > b31 poor financial performance rateor in another words risk rate of the firm will be RC2 = m122.Profile D shows that if any firm’s variable X1 have values whereb12 < X1 6 b13 poor financial performance rate or in another wordsrisk rate of the firm will be RD = m13.Profile E shows that if any firm’s variable X1 have values where
1 For interpretation of color in Fig. 2, the reader is referred to the web version of
this article.

Fig. 2. CHAID Decision Tree.


X1 > b13 poor financial performance rate or in another words riskrate of the firm will be RE = m14.

Step IV: Identification for current situation of SME according torisk profiles and early warning signs

The part of study until this point is based on the identification ofrisk profiles from all of the data. In the scope of the data that isabout the past of SMEs, the part of the study until this point definesthe relationships between financial risk and variables, and also therisk profiles.At this step, risk profiles that all of the firms belong to are identi-fied in the study. This identification is realized with taking thegroup of variables in the risk profiles into consideration.All of the firm will look at the values of their own enterprises, inthe light of the statistically significant variables in the DecisionTree. According to Fig. 2 these variables are X1, X2 and X3. The firmcompares the values of X1, X2 and X3 between decision tree andfirms. Then, they can identify their risk profile. For example ifany firm has X1 > b13. Therefore, the risk profile of the firm mustbe Profile E.According to the risk profiles of SMEs, it is possible to detect theearly warning signs that show highest financial risk.

Step V: Description of roadmap for SMEs
According to Fig. 2 we can easily determine the risk grades of thefirms. Assume that, the risk rates of the firms in the order ofE > D > C2 > C1 > B2 > B1. Therefore, the best risk profile will beB1. Then, every firm tries to be in Profile B1. There are two vari-ables X1 and X2 related with profile B1. If any firm want to be inProfile B1, the firm must make arrangements to make valuesX1 6 b11 and X2 6 b21.Enterprise will identify the suitable road map after defining its riskprofile. The enterprise can identify the path to reach upper levelrisk profile and the indicators that require privileged improvementin the light of the priorities of the variables in the roadmap. Fur-thermore, enterprise can pass to upper level risk profiles step bystep at the same time can reach to a targeted risk profile in theupper levels for improving indicators due to this target. For exam-ple, any firm in Profile E has the biggest risk rate. The firm must berehabilitating first the variable X1 to decrease it between (b12,b13).Therefore, the firm will be in profile D and so on.
4. Application of model for risk detection and early warningsigns

Application of our model, early waning signs, financial roadmaps and other results are presented below.

Step I: Data preparation
Application of our model covered SMEs in Turkey in 2007. Data offirms was obtained from Turkish Central Bank (TCB) after permis-sion. Total number of firms had financial data were 8.979 in TCB in2007. Since scope of our study only covered micro, medium, andsmall-scaled enterprises, which are often referred to as SMEs, those7.853 firms were classified to identify the firms, which can be cat-egorized as a SME. We based on SME definition of the EU in anattempt to participate to Turkey’s efforts to align with the EUacquits and to ensure comparability of the analysis providedherein. The thresholds used to classify SME on basis of the EU’sSME definitions are €50 million.Financial data that are gained from balance sheets and incomestatements was used to calculate financial indicators of system.Steps of preparation of data:
� Calculation of financial indicators like in Table 2.� Reduction of repeating variables in different indicators

to solve the problem of collinearity/multicollinearity.� Imputation of missing data.� Solution of outlier and extreme value problem.

Financial ratios as financial risk indicators were calculated withvariables collected from balance sheets of SMEs. These indicatorsand their definitions are presented in Table 2.

Step II: Implementation of data mining method (CHAID)During the preparation of an early warning system for SMEs, one
of the basic objectives is to help SME managers and decisionmakers, who does not have financial expertise, knowledge ofdata mining and analytic perspective, to reach easy to under-stand, easy to interpret, and easy to apply results about the riskcondition of their enterprises. Therefore, Decision Tree algo-rithms that are one of the segmentation methods can be usedbecause of their easy to understand and easy to apply visualiza-tion. CHAID algorithms used in this study are developed on basisof two groups of variables, namely target variable and predictor

Table 2Financial variables and definitions.

Code Variables Definition

A Liquidity ratiosA1 Current ratio Current assets/current liabilitiesA2 Quick ratio (liquidity ratio) (Cash, banks, marketable securities, account receivables)/current liabilitiesA3 Absolute liquidity (Cash, banks, marketable securities)/current liabilitiesA4 Inventories to current assets Total inventories/current assetsA5 Inventories to total assets Total inventories/total assetsA6 Inventory dependency ratio (Short-term liabilities – (liquid assets + marketable securities))/inventoriesA7 Current account receivables to Total assets Current account receivables/total assetsA8 Short-term receivables to total assets total assets Short-term receivables/total assets total assets

B Ratios of financial positionB1 Total loans to total assets (Short-term liabilities + long-term liabilities)/total assets (leverage ratio)B2 Own funds to total assets Own funds/total assetsB3 Own funds to total loans Own funds/(short-term liabilities + long-term liabilities)B4 Short-term liabilities to total liabilities Short-term liabilities/ total liabilitiesB5 Long-term liabilities to total liabilities Long-term liabilities/ total liabilitiesB6 Long-term liabilities to long-term liabilities and own

fundslong-term liabilities/(long-term liabilities and own funds)

B7 Tangible fixed assets to own funds Tangible fixed assets/own fundsB8 Tangible fixed assets to long-term liabilities Tangible fixed assets/long-term liabilitiesB9 Fixed assets to total loans Fixed assets/(short-term liabilities + long term liabilities)B10 Fixed assets to own funds Fixed assets/own fundsB11 Fixed assets to long term loans + own funds fixed assets/(long term loans + own funds)B12 Short-term liabilities to total loans Short-term liabilities/ total loansB13 Bank loans to total assets Bank loans/total assetsB14 Bank loans to short-term liabilities (Short-term bank loans + principal installments and interest payments of long-term bank loans)/short-

term liabilitiesB15 Bank loans to total loans (Short-term bank loans + principal installments and interest T payments of long-term bank loans + long-

term bank loans)/(short-term liabilities + long term liabilities)B16 Current assets to total assets Current assets/total assetsB17 Tangible fixed assets to total assets Tangible fixed assets/total assets

C Turnover ratiosC1 Inventory turnover Cost of goods sold (current year)/(previous year’s inventory + current year’s inventory)/2C2 Receivables turnover Net sales/(short-term trade receivables + long-term)C3 Working capital turnover Net sales/current assetC4 Net working capital turnover Net sales/(current assets–short-term liabilities)C5 Tangible fixed assets turnover Net sales/tangible fixed assets (net)C6 Fixed assets turnover Net sales/fixed assetsC7 Own funds turnover Net sales/own fundsC8 Total assets turnover Net sales/total assets

D Profitability ratiosD1 Ratios relating profit to capitalD1a Net profit to own funds Net profit (profit after tax)/own fundsD1b Profit before tax to own funds Profit before tax/own fundsD1c Profit before interest and tax to profit before

tax + financing expensesProfit before interest and tax/(profit before tax + financing expenses)

D1d Net profit to total assets Net profit/total assetsD1e Operating profit to assets used in carrying out of the

operationsOperating profit/total assets-financial fixed assets

D1f Cumulative profitability ratio Reserves from retained earnings/ total assetsD2 Ratios relating profit to salesD2a Operating profit to net sales Operating profit/net salesD2b Gross profit to net sales Gross profit/net salesD2c Net profit to net sales Net profit/net salesD2d Cost of goods sold to net sales Cost of goods sold/net salesD2e Operating expenses to net sales Operating expenses/net salesD2f Interest expenses to net sales Interest expenses/net salesD3 Ratios relating profit to financial obligationsD3a Profit before interest and tax to profit before

tax + financing expensesProfit before interest and tax/(profit before tax + financing expenses)

D3b Net profit and interest expenses netprofit + financing expenses to interest expenses

(Net profit and interest expenses net profit + financing expenses)/interest expenses


variables that will explain the target variable. In this studyfinancial performance is explained by means of all financial vari-ables of a SME. Therefore, the financial performance indicator isconsidered as the target variable and all financial variables (seeTable 2) are considered as the predictor variables. Fig. 3 showsthe CHAID Decision Tree and all profiles obtained from CHAID.Tables 3 and 4 are explained and summarized the profiles inFig. 3.

Step III: Determination of risk profiles
CHAID has multi-branches, and all of the important relationshipsin data can be investigated until the subtle details. As can be seenfrom Fig. 3 and its partitions Figs. 4–7 which explain SMEs profilingand financial performance statuses based on CHAID method,although it was possible to superficially categorize the coveredSMEs into two groups as SMEs with good financial performanceand with poor financial performance with CHAID method it was

Fig. 3. CHAID Decision Tree and financial profiles of SMEs.


possible to categorize the covered SMEs in 31 different profiles interms of level of financial performance. These profiles show uswhat financial indicators should focus on for good financial perfor-mance as well as those profiles those SMEs should take example toimprove their financial performances.As required under CHAID method SMES profiling is based on profitbefore tax to own funds ratio (D1B), which has the strongest rela-tion with the financial performance (p < 0.001). SMEs with thisratio lower than and equal to 0 are grouped in the 1st profile.According to this profile, all of 1718 SMEs in this profile, or21.88% of all covered SMEs in the study have poor financial perfor-mance.Those SMEs with profit before tax to own funds ratio between 0and 0.20 are grouped in 2nd to 5th profiles given in Fig. 4. Also, itwas determined that return on equity ratio (D1A, p < 0.001) andcumulative profitability ratio (D1F, p = 0.001) affected financialrisk of SMEs in these profiles. 2nd profile comprises 27 SMEs,or 0.34% of total covered SMEs with return on equity ratio lowerthan and equal to 0, and all those SMEs have poor financial per-formance. 3rd profile with return on equity ratio between 0 and0.02 and cumulative profitability ratio lower than and equal to0.0000002 covered total 101 SMEs. In this profile 78.22% (79SMEs) have good financial performance and remaining 21.78%(22 SMEs) have poor financial performance. On the other hand,4th profile comprises 13 SMEs with return on equity ratio higherthan 0.02 and cumulative profitability ratio lower than and equalto 0.0000002, and 46.15% of which have good financial perfor-mance and 53.85% of which have poor financial performance. Lastprofile with profit before tax to own funds ratio between 0 and0.20 is 5th profile. In this profile, return on equity ratio is higherthan 0 and cumulative profitability ratio is higher than0.0000002. 90.12% (447 SMEs) of SMEs have good financial per-formance and remaining 9.88% (49 SMEs) have poor financial per-formance in this profile.

SMEs with profit before tax to own funds ratio between 0.20 and0.36 are grouped in 6th to 21st profiles. Profiles 6th to 16th givenin Fig. 5 and profiles 17th to 21st given in Fig. 6. Beside this ratio,return on equity ratio (D1A, p < 0.001), cumulative profitabilityratio (D1F, p = 0.001), short-term liabilities to total loans (B12,p = 0.0001), and total loans to total assets (B1, p = 0.0230) affectedfinancial risk of SMEs in 6th profile to 9th profile. 6th profile com-prises 8 SMEs, or 0.1% of total covered SMEs with return onequity ratio lower than and equal to 0, and all those SMEs havepoor financial performance. 7th profile includes SMEs with returnon equity ratio higher than 0, cumulative profitability ratio lowerthan and equal to 0.0000002, short-term liabilities to total loanslower than and equal to 0.86, and total loans to total assets lowerthan and equal to 0.20. In this profile, 20% (1 SMEs) of SMEs havegood financial performance, and 80% (4 SMEs) of SMEs have poorfinancial performance. 8th profile contains SMEs with return onequity ratio higher than 0, cumulative profitability ratio lowerthan and equal to 0.0000002, short-term liabilities to total loanslower than and equal to 0.86, and total loans to total assetshigher than 0.20. In this profile, 77.45% (285 SMEs) of SMEs havegood financial performance, and 22.55% (83 SMEs) of SMEs havepoor financial performance. 9th profile includes SMEs with returnon equity ratio higher than 0, cumulative profitability ratio lowerthan and equal to 0.0000002, and short-term liabilities to totalloans higher than 0.86. In this profile, 88.57% (341 SMEs) of SMEshave good financial performance, and 11.43% (44 SMEs) of SMEshave poor financial performance.In 10th to 16th profiles, beside profit before tax to own fundsratio (D1B, p < 0.001), return on equity ratio (D1A, p < 0.001),cumulative profitability ratio (D1F, p = 0.001), interest expensesto net sales (D2F, p = 0.0011), fixed assets to long term loan-s + own funds (B9, p = 0.0027), and long-term liabilities to totalliabilities (B5, p < 0.0001) affected financial risk of SMEs. 10thprofile includes SMEs with return on equity ratio higher than 0,

Table 3Profiles of SMEs according to CHAID Decision Trees.

Profiles Nodes Financial indicators

D1B D1A D1F D2F B12 B1 B9 B5 D2B B6 B13 C7 A8 D2E C2

1 0, 1 602 0, 2, 5 0–

0.2060

3 0, 2, 6,11, 20

0–0.20

0–0.02

60.0000002

4 0, 2, 6,21

0–0.20

>0.02 60.0000002

5 0, 2, 6,12

0–0.20

>0 >0.0000002

6 0, 3, 7 0.20–0.36

60

7 0, 3, 8,13, 22,36

0.20–0.36

>0 60.0000002 60.86 60.20

8 0, 3, 8,13, 22,37

0.20–0.36

>0 60.0000002 60.86 >0.20

9 0, 3, 8,13, 23

0.20–0.36

>0 60.0000002 >0.86

10 0, 3, 8,14, 24

0.20–0.36

>0 0.0000002–0.04

60

11 0, 3, 8,14, 25,38

0.20–0.36

>0 0.0000002–0.04

0–0.000005

60.74

12 0, 3, 8,14, 25,39

0.20–0.36

>0 0.0000002–0.04

0–0.000005

0.74–0.95

13 0, 3, 8,14, 25,40

0.20–0.36

>0 0.0000002–0.04

0–0.000005

>0.95

14 0, 3, 8,14, 26

0.20–0.36

>0 0.0000002–0.04

0.000005–0.06

15 0, 3, 8,14, 27,41

0.20–0.36

>0 0.0000002–0.04

>0.06 60.22

16 0, 3, 8,14, 27,42

0.20–0.36

>0 0.0000002–0.04

>0.06 >0.22

17 0, 3, 8,15, 28,43

0.20–0.36

>0 >0.04 60.14 60.52

18 0, 3, 8,15, 28,44

0.20–0.36

>0 >0.04 0.14–0.38

60.52

19 0, 3, 8,15, 28,45

0.20–0.36

>0 >0.04 >0.38 60.52

20 0, 3, 8,15, 29,46

0.20–0.36

>0 >0.04 60.13 >0.52

21 0, 3, 8,15, 29,47

0.20–0.36

>0 >0.04 >0.13 >0.52

22 0, 4, 9,16, 30,48

>0.36 60.75 60.26 60.015

23 0, 4, 9,16, 30,49

>0.36 60.75 60.26 60.015

24 0, 4, 9,16, 30,50

>0.36 60.75 60.26 60.015 >0.03

25 0, 4, 9,16, 31

>0.36 60.75 60.26 >0.015

26 0, 4, 9,17, 32

>0.36 >0.75 60.26 60.03

27 0, 4, 9,17, 33,51

>0.36 >0.75 60.26 >0.03 60.02

28 0, 4, 9,17, 33,52

>0.36 >0.75 60.26 >0.02

29 0, 4, 10, >0.36 >0.26 60.05


1830 0, 4, 10,

19, 34>0.36 >0.26 60.0000006 >0.05

31 0, 4, 10,19, 35

>0.36 >0.26 >0.0000006

Table 4Financial situation of SMEs.

Profiles Financial performance

Good Poor Totaln % n % n %

1 0 0.00 1718 100.00 1718 21.882 0 0.00 27 100.00 27 0.343 79 78.22 22 21.78 101 1.294 6 46.15 7 53.85 13 0.175 447 90.12 49 9.88 496 6.326 0 0.00 8 100.00 8 0.107 1 20.00 4 80.00 5 0.068 285 77.45 83 22.55 368 4.699 341 88.57 44 11.43 385 4.90

10 247 89.82 28 10.18 275 3.5011 202 97.12 6 2.88 208 2.6512 20 76.92 6 23.08 26 0.3313 86 94.51 5 5.49 91 1.1614 1274 88.41 167 11.59 1441 18.3515 269 88.49 35 11.51 304 3.8716 90 70.31 38 29.69 128 1.6317 1065 94.16 66 5.84 1131 14.4018 218 88.26 29 11.74 247 3.1519 35 100.00 0 0.00 35 0.4520 31 93.94 2 6.06 33 0.4221 18 64.29 10 35.71 28 0.3622 15 100.00 0 0.00 15 0.1923 3 75.00 1 25.00 4 0.0524 101 100.00 0 0.00 101 1.2925 236 91.83 21 8.17 257 3.2726 27 100.00 0 0.00 27 0.3427 3 33.33 6 66.67 9 0.1128 93 80.17 23 19.83 116 1.4829 66 67.35 32 32.65 98 1.2530 1 25.00 3 75.00 4 0.0531 132 85.71 22 14.29 154 1.96

Total 5391 68.65 2462 31.35 7853 100.00


cumulative profitability ratio between 0.0000002 and 0.04, andinterest expenses to net sales lower than and equal to 0. In thisprofile, 89.82% (247 SMEs) of SMEs have good financial perfor-mance, and 10.18% (28 SMEs) of SMEs have poor financial perfor-mance. 11th profile covers SMEs with return on equity ratiohigher than 0, cumulative profitability ratio between 0.0000002and 0.04, interest expenses to net sales between 0 and0.000005, and fixed assets to long term loans + own funds lowerthan and equal to 0.74. In this profile, 97.12% (202 SMEs) of SMEshave good financial performance, and 2.88% (6 SMEs) of SMEshave poor financial performance. 12th profile contains SMEs withreturn on equity ratio higher than 0, cumulative profitability ratiobetween 0.0000002 and 0.04, interest expenses to net salesbetween 0 and 0.000005, and fixed assets to long term loan-s + own funds ratio between 0.74 and 0.95. In this profile,76.92% (20 SMEs) of SMEs have good financial performance, and23.08% (6 SMEs) of SMEs have poor financial performance. 13thprofile comprises SMEs with return on equity ratio higher than0, cumulative profitability ratio between 0.0000002 and 0.04,interest expenses to net sales between 0 and 0.000005, and fixedassets to long term loans + own funds ratio higher than 0.95. Inthis profile, 94.51% (86 SMEs) of SMEs have good financial perfor-mance, and 5.49% (5 SMEs) of SMEs have poor financial perfor-mance. 14th profile with return on equity ratio higher than 0,cumulative profitability ratio between 0.0000002 and 0.04, andinterest expenses to net sales between 0.000005 and 0.06 coverstotal 1441 SMEs. In this profile, 88.41% (1274 SMEs) of SMEs havegood financial performance, and 11.59% (167 SMEs) of SMEs havepoor financial performance. 15th profile includes SMEs withreturn on equity ratio higher than 0, cumulative profitability ratiobetween 0.0000002 and 0.04, interest expenses to net saleshigher than 0.06, and long-term liabilities to total liabilities lowerthan and equal to 0.22. In this profile, 88.49% (269 SMEs) havegood financial performance and remaining 11.51% (35 SMEs) havepoor financial performance. 16th profile contains SMEs withreturn on equity ratio higher than 0, cumulative profitability ratiobetween 0.0000002 and 0.04, interest expenses to net saleshigher than 0.06, and long-term liabilities to total liabilitieshigher than. In this profile, 70.31% (90 SMEs) have good financialperformance and remaining 29.69% (38 SMEs) have poor financialperformance.In 17th to 21st profiles, profit before tax to own funds ratio (D1B,p < 0.001), return on equity ratio (D1A, p < 0.001), cumulativeprofitability ratio (D1F, p = 0.001), long-term liabilities to total lia-bilities (B5, p < 0.0001), gross profit to net sales (D2B, p = 0.0332),and bank loans to total assets (B13, p < 0.0012) affected financialrisk of SMEs. 17th profile with return on equity ratio higher than0, cumulative profitability ratio higher than 0.04, long-term liabil-ities to total liabilities lower than and equal to 0.14, and bankloans to total assets lower than and equal to 0.52 contains total1131 SMEs. In this profile, 94.16% (1065 SMEs) of SMEs have goodfinancial performance, and 5.84% (66 SMEs) of SMEs have poorfinancial performance. 18th profile comprises SMEs with returnon equity ratio higher than 0, cumulative profitability ratio higherthan 0.04, long-term liabilities to total liabilities between 0.14and 0.38, and bank loans to total assets lower than and equal

to 0.52. In this profile, 88.26% (218 SMEs) of SMEs have goodfinancial performance, and 11.74% (29 SMEs) of SMEs have poorfinancial performance. All of 35 SMEs in profile 19th have goodfinancial performance. This profile covers SMEs with return onequity ratio higher than 0, cumulative profitability ratio higherthan 0.04, long-term liabilities to total liabilities higher than0.38, and bank loans to total assets lower than and equal to0.52. 20th profile contains SMEs with return on equity ratiohigher than 0, cumulative profitability ratio higher than 0.04,gross profit to net sales lower than and equal to 0.13, and bankloans to total assets higher than 0.52. In this profile, 93.94% (31SMEs) have good financial performance and remaining 6.06%(38 SMEs) have poor financial performance. 21st profile containsSMEs with return on equity ratio higher than 0, cumulative prof-itability ratio higher than 0.04, gross profit to net sales higherthan 0.13, and bank loans to total assets higher than 0.52. In thisprofile, 64.29% (18 SMEs) have good financial performance andremaining 35.71% (10 SMEs) have poor financial performance.SMEs profit before tax to own funds ratio higher than 0.36 aregrouped in 22nd to 31st profiles given in Fig. 7. Beside this ratio,total loans to total assets (B1, p = 0.0230), inventory dependency


ratio (B6, p < 0.0001), bank loans to total assets (B13, p < 0.0012),own funds turnover (C7, p = 0.0432), short-term receivables tototal assets total assets (A8, p = 0.0121), operating expenses tonet sales assets (D2E, p = 0.0149), receivables turnover (C2,p < 0.0001) affected financial risk of SMEs in these profiles.In 22nd to 25th profiles, total loans to total assets ratio is lowerthan and equal to 0.75, and inventory dependency ratio is lowerthan and equal to 0.26. Beside these ratios, 22nd profile containsSMEs with bank loans to total assets ratio lower than and equalto 0.015. All of 15 SMEs have good financial performance in thisprofile. 23rd profile covers SMEs with bank loans to total assetsratio lower than and equal to 0.015, and receivables turnoverhigher lower and equal to 0.03. In this profile, 75 % (3 SMEs) havegood financial performance and 25% (1 SMEs) have poor financialperformance. All of 101 SMEs have good financial performance in24th profile. This profile contains SMEs with bank loans to totalassets ratio lower than and equal to 0.015, and receivables turn-over higher 0.03. 25th profile contains SMEs with bank loans tototal assets ratio higher than 0.015. In this profile, 91.83% (236SMEs) have good financial performance and 8.17% (21 SMEs) havepoor financial performance.In 26th to 28th profiles, total loans to total assets ratio is higherthan 0.75 and inventory dependency ratio is lower than and equalto 0.26. Beside these ratios, 26th profile contains SMEs with ownfunds turnover lower than and equal to 0.03. All of 27 SMEs havegood financial performance in 26th profile. 27th profile coversSMEs with own funds turnover higher than 0.03, and short-termreceivables to total assets ratio lower than and equal to 0.02. Inthis profile, 33.33% (3 SMEs) have good financial performanceand 66.67% (6 SMEs) have poor financial performance. Last profilein this group covers SMEs with short-term receivables to total

Fig. 4. CHAID Decision Tree

assets ratio higher than 0.02. In 28th profile, 80.17% (93 SMEs)have good financial performance and 19.83% (23 SMEs) have poorfinancial performance.

and 2nd a

Step IV:Identification forcurrent situation of SME from risk pro-files and early warning signsAccording to the CHAID, it is possible to indicate the finan-cial position or situation with risk profiles that all of thefirms belong to, determine the profiles with highest finan-cial risk, identify financial indicators that affect financialdistress of SMEs, and detect early warning signs.As you can see in Fig. 3 and Table 4, SMEs are classified in31 different profiles, according to indicators that affectedtheir financial performance and financial situation. It wasdetermined that 5391 SMEs (68.6%) out of 7853 coveredSMEs had good financial performance while 2462 of themhad poor financial performance. Results showed that 31.4%of the covered SMEs financially distress. These distressfirm are in 27 different profiles except 19th, 22nd, 24th,and 26th profiles depend on different financial indicators.All of SMEs in profiles 1st, 2nd, and 6th have poor financialperformance and these SMEs are exactly distressed firms.These profiles contain SMEs with highest financial risk.All of SMEs in profiles 19th, 22nd, 24th, and 26th havegood financial performance and these SMEs are exactlynon distressed firms.Results of the study revealed that there are 15 indicators(in total 41 ratios), which had effects on financial perfor-mance or in other words distress of the covered SMEs. Asseen in Table 5, these are profit before tax to own fundsratio (D1B, p < 0.001), return on equity ratio (D1A,

nd 5th financial profiles of SMEs.

Fig. 5. CHAID Decision Tree and 6th and 16th financial profiles of SMEs.

Fig. 6. CHAID Decision Tree and 17th and 21st financial profiles of SMEs.


Fig. 7. CHAID Decision Tree and 22nd and 31st financial profiles of SMEs.


p < 0.001), cumulative profitability ratio (D1F, p = 0.001),short-term liabilities to total loans (B12, p = 0.0001), totalloans to total assets (B1, p = 0.0230), interest expenses tonet sales (D2F, p = 0.0011), fixed assets to long term loan-s + own funds (B9, p = 0.0027), long-term liabilities to totalliabilities (B5, p < 0.0001), gross profit to net sales (D2B,p = 0.0332), bank loans to total assets (B13, p < 0.0012),inventory dependency ratio (B6, p < 0.0001), own fundsturnover (C7, p = 0.0432), short-term receivables to totalassets total assets (A8, p = 0.0121), operating expenses tonet sales assets (D2E, p = 0.0149), receivables turnover(C2, p < 0.0001).We determined that 15 indicators affected financial riskand distress position of SMEs. When we consider risk pro-files and risk indicators together, only 2 indicators can beidentified as early warning signals. These are profit beforetax to own funds and return on equity (ROE).

Table 5Financial indicators affected financial distress.

Code Financial indicators p

D1B Profit before tax to own funds <0.0001D1A Return on equity <0.0001D1F Cumulative profitability ratio =0.0001

(i) If profit before tax to own funds was lower than andequal to 0.

(ii) If ROE was lower than and equal to 0.(iii)If profit before tax to own funds was between 0 and 0.20,and ROE was lower than and equal to 0.(iv)If profit before tax to own funds was between 0.20 and0.36, and ROE was lower than and equal to 0 financial dis-tress were indispensable for SMEs.

B12 Short-term liabilities to total loans =0.0001B1 Total loans to total assets =0.0230D2F Interest expenses to net sales =0.0011B9 Fixed assets to long term loans + Own funds =0.0027B5 Long-term liabilities to total liabilities <0.0001D2B Gross profit to net sales =0.0332B13 Bank loans to total assets =0.0012B6 Inventory dependency ratio <0.0001C7 Own funds turnover =0.0432A8 Short-term receivables to total assets total assets =0.0121D2E Operating expenses to net sales =0.0149C2 Receivables turnover <0.0001

Step V:Description of roadmap for SMEsWhen we determine the best risk profile, we can suggestthis profile for benchmarking as a road map. Financial roadmaps are tools for decision making and give information asinputs in decision process. SMEs can identify the path toreach upper level risk profile and the indicators thatrequire privileged improvement in the light of the priori-ties of the variables in the roadmap. Furthermore, enter-prise can pass to upper level risk profiles step by step at

the same time can reach to a targeted risk profile in theupper levels for improving indicators due to this target.

According to our model, the best risk Profiles are 19th,22nd, 24th, and 26th that contained SMEs without risk.Then, every firm tries to be in these profiles. The bestprofiles and road maps to reach best performance arepresented in Table 6.1st road map contains 19th profileand their indicators. If any SME wants to be in Profile 19,the SME must make arrangements to make values of profitbefore tax to own funds between 0.20 and 0.36, return onequity higher than 0, cumulative profitability ratio higherthan 0.04, long term liabilities to total liabilities higherthan 0.38, and bank loans to total assets lower than andequal to 0.52.

2nd road map contains 22nd profile and their indicators. If anySME wants to be in Profile 22 the SME must make arrangements tomake values of profit before tax to own funds higher than 0.36, to-tal loan to total assets lower than and equal to 0.75, inventorydependency ratio lower than and equal to 0.26, and bank loan tototal assets lower than and equal to 0.02.

Table 6Road maps.

Rodmaps

Profiles Probabilityof no risk(%)

Financial indicators

D1B D1A D1F B1 B5 B6 B13 C2 C7

Profit beforetax to ownfunds

Returnonequity

Cumulativeprofitabilityratio

Total loansto Totalassets

Long termliabilities to Totalliabilities

Inventorydependencyratio

Bank loansto totalassets

Receivablesturnover

Ownfundsturnover

1 19 100 0.20–0.36 >0 >0.04 >0.38 60.522 22 100 >0.36 60.75 60.26 60.023 24 100 >0.36 60.75 60.26 60.02 >0.034 26 100 >0.36 >0.75 60.26 >0.03


3rd road map contains 24th profile and their indicators. If anySME wants to be in Profile 24 the SME must make arrangementsto make values of profit before tax to own funds higher than0.36, total loan to total assets lower than and equal to 0.75, inven-tory dependency ratio lower than and equal to 0.26, bank loan tototal assets lower than and equal to 0.02, and receivables turnoverhigher than 0.03.

4th road map contains 26th profile and their indicators. If anySME wants to be in Profile 26 the SME must make arrangementsto make values of profit before tax to own funds higher than0.36, total loan to total assets higher than 0.75, inventory depen-dency ratio lower than and equal to 0.26, own funds turnover high-er than 0.03.

5. Conclusion

Financial early warning system is a technique of analysis that isused to predict the achievement condition of enterprises and todecrease the risk of financial distress. By the application of thistechnique of analysis, the condition and possible risks of an enter-prise can be identified with quantity. Risk management hasbecome a vital topic for all institutions, especially for SMEs, banks,credit rating firms, and insurance companies. The financial crisishas pushed all firms to active risk management and control finan-cial risks. All enterprises need EWS to warn against risks and pre-vent from financial distress. But, when we consider the issues ofpoor business performance, insufficient information and insuffi-ciencies of managers in finance education, it is clear that EWS is vi-tal for SMEs. Benefits of an EWS can summarize as early warningbefore financial distress, road maps for good credit rating, betterbusiness decision making, and greater likelihood of achieving busi-ness plan and objectives.

Developing practical solutions will not only help to SMEs butalso to the economies of countries. Having information about theirfinancial risk, monitoring this financial risk and knowing the re-quired roadmap for the improvement of financial risk are veryimportant for SMEs to take the required precautions. Data mining,that is the reflection of information technologies in the area of stra-tegically decision support, develops a system for finding solutionsto the financial administration as one of the most suitable applica-tion area for SMEs as the vital point of economy.

In this study, we developed a financial EWS based on financialrisk by using data mining. As results of the study we classified7853 SMEs in 31 different risk profiles via CHAID. Results showedthat 31.4% of the covered SMEs financially distress. All of SMEs inprofiles 1st, 2nd, and 6th have poor financial performance andthese SMEs are exactly distressed firms. These profiles containSMEs with highest financial risk. All of SMEs in profiles 19th,22nd, 24th, and 26th have good financial performance and theseSMEs are exactly non distressed firms.

According to these profiles, we identified that profit before taxto own funds ratio, return on equity ratio, cumulative profitabilityratio, short-term liabilities to total loans, total loans to total assets,interest expenses to net sales, fixed assets to long term loans + own

funds, long term liabilities to total liabilities, gross profit to netsales, bank loans to total assets, inventory dependency ratio, ownfunds turnover, short-term receivables to total assets total assets,operating expenses to net sales assets, receivables turnover affectfinancial performance or in other words distress of the coveredSMEs. When we consider risk profiles and these 15 risk indicatorstogether, only 2 indicators can be identified as early warning sig-nals. Financial early warning signs for covered SMEs are profitbefore tax to own funds and return on equity (ROE). If profits be-fore tax to own funds and ROE ratios are lower than and equal to0, financial distress is indispensable for SMEs.

Beside these findings we determine financial road maps for riskmitigation and improve financial performance. Financial roadmaps can use for decision making process as inputs. According toour study findings, we developed 4 financial road maps. All of 4road maps provide risk indicators and their values for successfulmanagement and risk hedging.

EWSs should develop and implement in every business, to pro-vide information relating to the actions of individual officers,supervisors, and specific units or divisions. In deciding what infor-mation to include in their early warning system, business shouldbalance the need for sufficient information for the system to becomprehensive with the need for a system that is not too cumber-some to be utilized effectively. The system should provide supervi-sors and managers with both statistical information anddescriptive information about the function of business.

In case of using our EWS model by SMEs, some of expected con-tributions can be summarized as:

� Determine financial performance and position of firms.� Determine financial strategies by minimum level of finance

education and information.� Financial and operational risk detection.� Roadmaps for risk mitigation.� Prevent for financial distress.� Decrease the possibility of bankruptcy.� Decrease risk rate.� Efficient usage of financial resources.� By efficiency in resources;� Increase the competition capacity.� New potential for export.� Decrease the unemployment rate.� More taxes for government.� Adaptation to BASEL II capital accord

Developing a financial EWS based on financial risk is not en-ough for to understand and manage the financial risks that cancause insolvency and distress. Managers need also to manageoperational risks that can arise from execution of a company’sbusiness functions, and strategically risks that can underminethe viability of their business models and strategies or reducetheir growth prospects and damage their market value. For thisreason we suggest to develop EWS that contain all kind of riskfactors.


Acknowledgment

This research was funded by The Scientific and TechnologicalResearch Council of Turkey (TUBITAK).

References

Abumustafa, N. I. (2006). Development of an early warning model for currencycrises in emerging economies: An empirical study among Middle Easterncountries. International Journal of Management, 23(3), 403.

Altman, E. (1968). Financial ratios, discriminant analysis and the prediction ofcorporate bankruptcy. Journal of Finance, September, 589–609.

Altman, E. I., Haldeman, G., & Narayanan, P. (1977). Zeta analysis: A new model toidentify bankruptcy risk of corporations. Journal of Banking and Finance, June,29–54.

Ambrose, J. M., & Seward, J. A. (1998). Best’s ratings financial ratios and priorprobabilities in insolvency prediction. Journal of Risk and Insurance, 55, 229–244.

Barniv, R., & Hathorn, J. (1997). The merger or insolvency alternative in theinsurance industry. Journal of Risk and Insurance, 64(1), 89–113.

Barniv, R., & McDonald, J. B. (1992). Identifying financial distress in the insuranceindustry: A synthesis of methodological and empirical issues. Journal of Risk andInsurance, 59, 543–573.

Beaver, W. (1966). Financial ratios as predictors of failure. Journal of AccountingResearch, 4, 71–111.

Berg, A., Borensztein, E., & Pattillo, C. (2004). Assessing early warning systems: Howhave they worked in practice? IMF Working Paper, March 2004. Retrieved April,2009, from <http://www.ksri.org/bbs/files/research02/wp0452.pdf>.

Brockett, P. L., & Cooper, W. W. (1990). Report to the State Auditor and the StateBoard of Insurance on Early Warning Systems to Monitor the Performance ofInsurance Companies in Texas. Office of the State Auditor, Austin, TX.

Brockett, P. L., Golden, L. L., Jang, J., & Yang, C. (2006). A comparison of neural network,statistical methods and variable. Journal of Risk and Insurance, 73(3), 397–419.

Chang, S., Chang, H., Lin, C., & Kao, S. (2003). The effect of organizationalattributes on the adoption of data mining techniques in the financial serviceindustry: An empirical study in Taiwan. International Journal of Management,20(1), 497–503.

Commission Recommendation of 6 May 2003 Concerning the Definition of Micro,Small and Medium-sized Enterprises (2003/361/EC), L 124/36 2003. OfficialJournal of the European Union,Retrieved August 2010, from <http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=OJ:L:2003:124:0036:0041:en:PDF>.

Davis, E. P., & Karim, D. (2008a). Comparing early warning systems for bankingcrises. Journal of Financial Stability, 4(2), 89–120.

Davis, E. P., & Karim, D. (2008b). Could early warning systems have helped topredict the sub-prime crisis? National Institute Economic Review, 206(1), 35–47.

Deakin, E. B. (1972). A discriminant analysis of predictors of business failure. Journalof Accounting Research, 10(1), 167–179.

Derby, B. L. (2003). Data mining for improper payments. The Journal of GovernmentFinancial Management, 52(1), 10–13.

Edison, H. J. (2003). Do indicators of financial crises work? An evaluation of an earlyearning system. International Journal of Finance and Economics, 8(1), 11–53.

Eklund, T., Back, B., Vanharanta, H., & Visa, A. (2003). Using the self-organizing mapas a visualization tool in financial benchmarking. Information Visualization, 2(3),171–181.

El-Shazly, A. (2003). Early warning of currency crises: An econometric analysis forEgypt. The Middle East Business and Economic Review, 18(1), 34–48.

Fayyad, G., Piatetsky-Shapiro, P., & Symth, P. (1996). From data mining toknowledge discovery in databases. AI Magazine, 17(3), 37–54.

Frawley, W., Piatetsky-Shapiro, G., & Matheus, C. (1992). Knowledge discovery indatabases: An overview. AI Magazine: Fall, 213–228.

Hand, D., Mannila, H., & Smyth, P. (2001). Principles of data mining. Cambridge: MITPress.

Hoppszallern, S. (2003). Healthcare benchmarking. Hospitals & Health Networks, 77,37–44.

Jacobs, L. J. & Kuper, G. H. (2004) Indicators of financial crises do work! An early-warning system for six Asian countries. CCSO Working Paper 13. Department ofEconomics, University of Groningen, the Netherlands.

Jones, F. (1987). Current techniques in bankruptcy prediction. Journal of AccountingLiterature, 6, 131–164.

Katz, M. (2006). Multivariable analysis: A practical guide for clinicians. New York:Churchill-Livingstone.

Kloptchenko, A., Eklund, T., Karlsson, J., Back, B., Vanhatanta, H., & Visa, A. (2004).Combining data and text mining techniques for analyzing financial reports.Intelligent Systems in Accounting Finance and Management, 12(1), 29–41.

Koyuncugil, A. S. (2006). Fuzzy Data Mining and its application to capital markets.Unpublished doctoral dissertation, Ankara University, Ankara.

Koyuncugil, A. S., & Ozgulbas, N. (2006a). Financial profiling of SMEs: An applicationby data mining. In Proceedings Book of the European Applied Business Research(EABR) Conference, Clute Institute for Academic Research, pp. 1–22.

Koyuncugil, A. S., & Ozgulbas, N. (2006b). Is there a specific measure for financialperformance of SMEs? The Business Review, Cambridge, 5(2), 314–319.

Koyuncugil, A. S., & Ozgulbas, N. (2006c). Determination of factors affected financialdistress of SMEs listed in ISE by data mining. In Proceedings Book of 3rd Congressof SMEs and Productivity, KOSGEB and Istanbul Kultur University, Istanbul.

Koyuncugil, A. S., & Ozgulbas, N. (2007a). Developing financial early warning systemvia data mining. In Proceedings Book of 4th Congress of SMEs and Productivity,Istanbul, pp. 153–166.

Koyuncugil, A. S., & Ozgulbas, N. (2007b). Detecting financial early warning signs inIstanbul Stock Exchange by data mining. International Journal of BusinessResearch VII (3), 188–193.

Koyuncugil, A. S., & Ozgulbas, N. (2008a). Early warning system for SMEs asa financial risk detector. In Hakikur Rahman (Ed.), Data mining applicationsfor empowering knowledge societies (pp. 221–240), New York: Idea Group Inc.

Koyuncugil, A. S., & Ozgulbas, N. (2008b). Strengths and weaknesses of SMEs listedin ISE: A CHAID Decision Tree application. Journal of Dokuz Eylul University,Faculty of Economics and Administrative Sciences, 23(1), 1–22.

Koyuncugil, A. S., & Ozgulbas, N. (2009a). Measuring and hedging operational riskby data mining. In Proceedings Book of World Summit on Economic-Financial Crisisand International Business, Washington, pp. 1–6.

Koyuncugil, A. S., & Ozgulbas, N. (2009b). An intelligent financial early warningsystem model based on data mining for SMEs. In Proceedings of the InternationalConference on Future Computer and Communication, Kuala Lumpur, Malaysia.doi:10.1109/ICFCC.2009.118, pp. 662–666.

Koyuncugil, A. S., & Ozgulbas, N. (2010). Social Aid fraud detection system andpoverty map model suggestion based on data mining for social risk mitigation.In A. S. Koyuncugil, & N. Ozgulbas (Eds.). Surveillance Technologies and EarlyWarning Systems: Data Mining Applications for Risk Detection (pp. 173–193), NewYork: Idea Group Inc.

Kumar, N., Krovi, R., & Rajagopalan, B. (1997). Financial decision support withhybrid genetic and neural based modeling tools. European Journal of OperationalResearch, 103, 339–349.

Kyong, J. O, Tae, Y. K, Chiho, K., & Suk, J. L. (2006). Using neural networks to tune thefluctuation of daily financial condition indicator for financial crisis forecasting.Advances in Artificial Intelligence, DOI:10.1007/11941439_65 Volume 4304/2006.

Laitinen, K., & Chong, H. G. (1999). Early warning system for crisis in SMEs:Preliminary evidence from Finland and the UK. Journal of Small Business andEnterprise Development, 6(1), 89–102.

Lee, K. C., Han, I., & Kwon, Y. (1996). Hybrid neural network models for bankruptcypredictions. Decision Support Systems, 18, 63–73.

Lee, S. J., & Siau, K. (2001). A review of data mining techniques. IndustrialManagement & Data Systems, 101(1), 41–46.

Lee, S. H., & Urrutia, J. L. (1996). Analysis of insolvency prediction in the property-liability insurance industry: A comparison of logit and hazard models. Journal ofRisk and Insurance, 63, 121–130.

Magnusson, C., Arppe, A., Eklund, T., & Back, B. (2005). The language of quarterlyreports as an indicator of change in the company’s financial status. Information& Management, 42, 561–570.

Meyer, P. A., & Pifer, H. W. (1970). Prediction of bank failures. Journal of Finance,XXV(4), 853–886.

Monk, E., & Wagner, B. (2006). Concepts in Enterprise Resource Planning.SecondEdition. Boston: Thomson Course Technologies.

Nomenclature statistique des activités économiques dans la Communautéeuropéenne (NACE). (2009). Retrieved August 2010, from <http://epp.eurostat.ec.europa.eu/statistics_explained/index.php/SMEs>.

Nazem, S., & Shin, B. (1999). Data mining: New arsenal for strategic decisionmaking. Journal of Database Management, 10(1), 39–42.

OECD (2009). Policy responses to the economic crisis: Investing in innovation forlong-term growth. Retrieved August 2010, from <http://www.oecd.org/dataoecd/59/45/42983414.pdf>.

Ozgulbas, N., & Koyuncugil, A. S. (2006). Profiling and determining the strengths andweaknesses of SMEs listed in ISE by the Data Mining Decision Trees AlgorithmCHAID. In 10th National Finance Symposium, Izmir.

Ozgulbas N., & Koyuncugil A. S. (2009). Developing road maps for financial decisionmaking by CHAID Decision Tree. In Proceeding of International Conference onInformation Management and Engineering (pp. 723–727), IEEE Computer SocietyPress.

Ozgulbas, N., & Koyuncugil, A. S. (2010) Financial early warning system for riskdetection and prevention from financial crisis. In Surveillance Technologies andEarly Warning Systems: Data Mining Applications for Risk Detection. In A. S.Koyuncugil, & N. Ozgulbas (Eds.). Surveillance Technologies and Early WarningSystems: Data Mining Applications for Risk Detection (pp. 76–108). New York:Idea Group Inc.

Pantalone, C., & Platt, M. (1987). Predicting failures of savings and loan associations.AREUEA Journal, 15, 46–64.

Ross, S. A., Westerfield, R., & Jordan, B. D. (2008). Fundamentals of corporate finance.New York: McGraw-Hill Publishing Company.

Salas, V., & Saurina, J. (2002). Credit risk in two institutional regimes: Spanishcommercial and savings banks. Journal of Financial Services Research, 22(3),203–224.

Taffler, R. J., & Tisshaw, H. (1977). Going, going, gone-four factors which factorswhich predict. Accountancy, March, 50–54.

Tam, K. Y., & Kiang, M. Y. (1992). Managerial applications of neural networks: Thecase of bank failure predictions. Decision Sciences, 38, 926–948.

Terdpaopong, K. (2008). Financially distressed, small and medium-sizedenterprises’ characteristics and discriminant analysis model: Evidence fromthe Thai market. In: Proceedings of the Small Enterprise Association of Australia &New Zealand (SEAANZ) Conference 2008, Sydney, Australia, ISBN: 978-0-646-50668-5 p.125-165. Retrieved August 2010, from <http://www.seaanz.org/documents/SEAANZ2008ConferenceProceeding_000.pdf>.

http://www.ksri.org/bbs/files/research02/wp0452.pdf

http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=OJ:L:2003:124:0036:0041:en:PDF

http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=OJ:L:2003:124:0036:0041:en:PDF

http://dx.doi.org/10.1109/ICFCC.2009.118

http://epp.eurostat.ec.europa.eu/statistics_explained/index.php/SMEs

http://epp.eurostat.ec.europa.eu/statistics_explained/index.php/SMEs

http://www.oecd.org/dataoecd/59/45/42983414.pdf

http://www.oecd.org/dataoecd/59/45/42983414.pdf

http://www.seaanz.org/documents/SEAANZ2008ConferenceProceeding_000.pdf

http://www.seaanz.org/documents/SEAANZ2008ConferenceProceeding_000.pdf


Thearling, K. (2004). Information about data mining and analytic technologies.Retrieved August 2010, from <http://www.thearling.com/>.

Trieschmann, J. S., & Pinches, G. E. (1973). A multivariate model for predictingfinancially distressed property-liability insurers. Journal of Risk and Insurance,40, 327–338.

Warner, J. (1977). Bankruptcy costs: some evidence. Journal of Finance, 32, 337–347.

Yang, B., Ling, X. L., Hai, J., & Jing, X. (2001). An early warning system for loan riskassessment using artificial neural networks. Knowledge Based Systems, 14(5–6),303–306.

Zavgren, C. (1985). Assessing the vulnerability to failure of American industrialfirms: A logistics analysis. Journal of Accounting Research, 22, 59–82.

Zmijewski, M. E. (1984). Methodological issues related to the estimation of financialdistress prediction models. Journal of Accounting Research, Supplement, 59–82.

http://www.thearling.com/

Financial Early Warning System Model and Data Mining Application for Risk Detection 2012 Expert Systems With Applications

Documents

nancial problems

nancial crisis

nancial institutions

nancial advisor

nancial obligations

nancial risk detection

nancial risk mitigation

growth of smes