the financing of innovative smes: a multicriteria credit rating model

8/22/2019 The Financing of Innovative SMEs: a multicriteria credit rating model

1/34

arXiv:1308

.0889v1[q-fin.RM

]5Aug2013

The Financing of Innovative SMEs: a multicriteria credit rating model

SILVIA ANGILELLA1 SEBASTIANO MAZZU2

Department of Economics and BusinessUniversity of Catania,

Corso Italia, 55, Catania 95129, ItalyAugust 6, 2013

Abstract

Small Medium-sized Enterprises (SMEs) face many obstacles when they try to access creditmarket. These obstacles are increased if the SMEs are innovative. In this case, financialdata are insufficient or even not reliable. Thus, when building a judgemental rating model,mainly based on qualitative criteria (soft information), it is very important to finance SMEsactivities. Until now, there isnt a multicriteria credit risk model based on soft information forinnovative SMEs. In this paper, we try to fill this gap by presenting a multicriteria credit riskmodel, specifically, ELECTRE-TRI. To obtain robust SMEs assignments to the risk classes,a SMAA-TRI analysis is also implemented. In fact, SMAA-TRI incorporates ELECTRE-TRIby considering different sets of preference parameters with Monte Carlo simulations.

Finally, we carry out some real case studies, with the aim of illustrating the multicriteriacredit risk model proposed.

Keywords: Multiple Criteria Decision Aiding; qualitative criteria; SMAA-TRI; innovativeSMEs; judgmental credit rating model.

1 Introduction

Since small and medium-sized enterprises (SMEs) are the backbone of all economies, scholarlyattention to their credit risk assessment has considerably increased.

SMEs face many financial restrictions to access the credit market (for example see Berger et al.,2005, Beck and Demirgurc-Kunt, 2006 and Canales and Nanda, 2012). Moreover, if the SMEs arenewly created firms innovative, these restrictions are more tight (see Brown et al., 2009).

Consequently, there exists a finance gap within the SMEs sector, and such a condition isworsened when SMEs have to request credit to the banks. Banks usually adopt different lendingapproaches that can be grouped in four categories: financial statement lending (based on the

evaluation of information from balance-sheet data), asset-based lending (based on the provision ofa collateral), credit scoring models (based on statistical techniques), and relationship lending (seeMoro and Fink, 2013).

1Email address: [email protected] address: [email protected]
http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1http://arxiv.org/abs/1308.0889v1


2/34

In financing SMEs, there is a recent research debate on which is the best type of lending

technique.From one hand, there are several papers supporting the idea that banks prefer to financeSMEs on the basis of a long and strong activity of relationship banking, a type of lending mainlybased on non financial information (soft information) (see among others Moro and Fink, 2013 andBerger and Udell, 1996).

On the other, there is a recent stream of research that asserts that big banks lend moreeasily to SMEs if lending techniques based on hard information are adopted (see for exampleBerger and Udell, 2006 and Beck et al., 2011). Following this recent research direction, Pederzoli et al.,2012 present a credit scoring model based on financial ratios and innovation measures to predictthe Probability of Default (PD) of an innovative SME.

This issue is also linked to what has been established for the banks rating systems within Basel

II (Basel Committee on Banking Supervision, 2006).The rating system is based either on rating agencies (standardized approach) or on internal

ratings. The latter can be performed via an internal rating based approach or an internal ratingadvanced approach (Altman and Sabato, 2005).

Moreover, internal ratings can be decomposed into the following two parts:

a statistical-based rating,

a judgemental rating.

According to the supervisory system, in the rating process banks are free to let prevail the

first component (top-down logic) or the second one (bottom-up logic). In the first case, the ratingis mechanical, and considers mainly financial criteria (hard information) of the enterprise underevaluation, while in the second one, the assignment of enterprises to risk classes is based on thesubjective analysis of non-financial criteria (soft information).

Recently, even if Basel III has introduced some tighter constraints in terms of banks capitalrequirements (e.g. leverage ratio, countercyclical buffers, liquidity risk, revision of the componentsTier 1 e Tier 2), it has left the regulation of Basel II unchanged giving freedom to the banks todetermine the mix between the quantitative and qualitative components of a credit risk model.

Consequently, owing to the SMEs lack of sufficient or reliable track records, the judgmentalrating, mainly relying on soft information, seems the most useful approach to evaluate the SMEscreditworthiness.

In fact, the significant role of the non-financial criteria for the SMEs credit risk evaluation issupported by several research papers (see among others Altman et al., 2010, Auken et al., 2010,Berger et al., 2005 and Grunert et al., 2005; and specifically, for innovative SMEs see for exampleShefer and Frenkel, 2005 and Czarnitzki and Hottenrott, 2011).

2


3/34

Starting from the broad literature on this topic, our paper aims at building a judgmental rating

model to financing innovative SMEs.Until now, to the best of our knowledge, there isnt a judgemental credit risk model for innovativeSMEs. In this paper, we try to fill this gap by adopting a Multiple Criteria Decision Aid (MCDA)approach (Vincke, 1992). Since the financing of innovative SMEs is an uncertain problem dueto several and conflicting aspects, the Multiple Criteria Decision Aid approach (also called theconstructive approach Roy, 1993) seems the most useful (see Greco et al., 2013 for some usefulcomments on this topic).

Presently, several multicriteria approaches have already been adopted to predict business fail-ures, that is a typical sorting problem (see Zopounidis and Doumpos, 2002). Most of them relyon a utility function to classify enterprises into two categories: the default and non-defaulted (seeamong these the multicriteria hierarchical discrimination approach proposed in Doumpos et al.,

2002).Very few recent papers have also developed some credit rating models based on an outrank-

ing relation; for example, ELECTRE TRI (firstly, introduced in Yu, 1992) has been implementedfor the first time as a quantitative credit rating model in Zopounidis and Doumpos, 2011 andPROMETHEE II (Brans and Vincke, 1985) applied for banks rating evaluation (Doumpos and Zopoun2010). Moreover, a first attempt to estimate SMEs performance, based on a financial ratio anal-ysis, can be found in Voulgaris et al., 2000, where the multicriteria sorting approach, UTADIS(Jacquet-Lagreze, 1995) has been applied. However, none of these papers aims to develop a judge-mental rating model to assign innovative SMEs into risk classes.

The aim of this paper is twofold. First, we propose ELECTRE TRI to evaluate the SMEs

credit risk, mainly on the basis of qualitative criteria. To obtain robust SMEs assignments tothe risk classes, a SMAA-TRI analysis (Tervonen et al., 2007) has also been implemented. In fact,SMAA-TRI incorporates ELECTRE-TRI by considering different sets of preference parameterswith Monte Carlo simulations.

The second purpose of this paper is to illustrate the multicriteria methodology proposed byanalyzing some real case studies, specifically, some Italian innovative enterprises in the phase ofstart-up. Thus, by playing the role of different credit officers, we have simulated the financing ofthese innovative enterprises.

Moreover, preliminary to the multicriteria analysis, we have evaluated the financial soundnessof the start-up considered outperforming a scenario analysis on the basis of their Business Plans.The financial analysis has been conducted along two directions. On one hand, financial ratios have

been evaluated and compared with those of the sector to which each start-up belongs. On theother, a NPV for each scenario has been computed.

The paper is organized as follows. In Section 2, we provide a brief description of ELECTRE-TRI and SMAA-TRI. In Section 3, we describe the main phases of the multicriteria rating modelproposed to evaluate the riskiness of innovative SMEs. Some real case studies are carried out in

3


4/34

Section 4 to illustrate the judgemental multicriteria credit risk rating presented. Section 5 gives

some useful managerial insights of the methodology considered and finally, some conclusions areincluded in Section 6.

2 A sorting model

In this paper, the multicriteria approach ELECTRE-TRI (Yu, 1992) has been proposed as a creditrating model to assign innovative SMEs into risky categories. ELECTRE-TRI has been imple-mented in the framework of SMAA-TRI, to take into account uncertainty and imprecision in itspreference parameters. In the next sections, we provide a brief description of ELECTRE-TRI andSMAA-TRI.

2.1 An overview of ELECTRE-TRI method

Let A = {a1, a2, . . . , ai, . . . , am} be a finite set of m alternatives to be assigned on the basis of aconsistent family of n criteria G = {1, 2, . . . , n} to p risk ordered categories Cp Ck C1,where Ck+1 consists of a group of alternatives, better than those in Ck.An alternative ai A evaluated on the set of n criteria is denoted by:

ai = (ai1, , aij , ain),

where aij indicates the evaluation of the alternative ai on criterion j.Every group of alternatives is separated from the others by means of risk profiles:bp1, , bk, , b1. Each profile bk is the upper limit of the group k and the lower limit of groupk + 1. For example, considering two ordered classes, C2 could represent the financed enterprises,while C1 groups the ones non-funded. As a result, one risk profile b1 delimits two risk categoriesto which an enterprise has to be assigned. The alternatives are assigned to the risk classes on thebasis of their comparisons with the risk profiles exploiting the two outranking relations aiSbk andbkSai, which mean, respectively, that alternative ai is at least as good as the profile bk and viceversa.

The exploitation of each outranking relation consists of two phases: the concordance and dis-cordance test. Roughly speaking, the concordance test indicates the level of majority of criteriathat supports the outranking relation aiSbk (or bkSai), while the discordance test represents the

strength of the minority of criteria that oppose a veto to the outranking relation aiSbk (or bkSai).The concordance test is performed by the computation of the concordance index C(ai, bk):

C(ai, bk) =n

j=1

wjcj(aij, bk),

4


5/34

where wj indicates the weight of criterion j with the weights summing up to 1 and cj(ai, bk) is the

partial concordance index with respect to criterion j of the assertion ai is at least as good as theprofile bk. The partial concordance index cj(aij, bk) is computed as follows:

cj(aij , bk) =

0 if aij bkj pjaijbkj+pj

pjqjif bkj pj < aij < bkj qj

1 if aij bkj qj ,

where pj qj 0 indicate the preference and indifference thresholds, respectively. Such thresholdstake into account the imprecise criteria evaluations.

The preference threshold represents the smallest difference aij bk compatible with the prefer-ence of ai on criterion j, while the indifference threshold indicates the largest indifference aij bkthat preserves indifference between ai and bk. Similarly, the concordance index C(bk, ai) can alsobe defined.

The second phase consists in the discordance test, that is performed by the computation of thediscordance index dj(aij, bk) with respect to criterion j. Such index is defined as follows:

dj(aij , bk) =

0 if aij bkj pjbkjaijpj

vjpjif bkj vj < aij < bkj pj

1 if aij bkj vj,

where vj pj is a veto threshold expressing the smallest difference aij bk which is incompatiblewith the statement aiSbk.

Finally, a credibility index of the outranking relation is computed by:

(ai, bk) = C(ai, bk)jT

1 dj(aij, bk)

1 C(ai, bk),

where T is the set of criteria such as dj(aij , bk) > C(ai, bk). The outranking relation aiSbk has tobe defuzzyfied on the the basis of a user-defined threshold [0.5, 1].

The relation aiSbk is valid if and only if (ai, bk) > . Similarly, the outranking relation bkSaiis validated if and only if (bk, ai) > .

Choosing a value 0.5, is equivalent to saying that at least 50% of the criteria consideredare in favour of the assignment of ai to class Ck.

Then, two assignment procedures could be adopted; the pessimistic and the optimistic ones.The above assignment rules are formulated as follows:

Pessimistic assignment (conjunctive logic): each alternative is compared successively to theprofiles bp1, b2 . . . b1; ifaiSbk, then ai is assigned to the category Ck+1, otherwise ai is assignedto C1, i.e. the worst category;

5


6/34

Optimistic assignment (disjunctive logic): each alternative being compared to the profiles

b1, b2, . . . , bp1; if bkSai aiSbk, then ai is assigned to the category Ck and otherwise ai isassigned to the group Cp, i.e. the best category.

Some troublesome situations of incomparability could arise if the two preference statements areverified: aiSbk and bkSai.

In the present paper, we adopt the pessimistic rule that is most used in practice, assigning everyalternative ai to the category Ck+1 if the following inequalities are verified:

(ai, bk) and (bk, ai) < . (1)

If there arent veto criteria, it is easy to verify that

(ai, bk) = C(ai, bk) =n

j=1

wjcj(ai, bk).

As a consequence, the aforementioned decisional rule (1) is simplified as follows:

nj=1

wjcj(ai, bk) andn

j=1

wjcj(bk, ai) < . (2)

2.2 SMAA-TRI

In this section, we introduce SMAA-TRI, firstly introduced in Tervonen et al., 2007 (see alsoTervonen et al., 2009). SMAA-TRI, is a SMAA (Stochastic Multicriteria Acceptability Analysis)method that can take into account uncertainty and imprecision on the set of parameters requiredas input by ELECTRE-TRI.

In this paper, SMAA-TRI will be used to analyze the robustness of ELECTRE-TRI based onthe parameter stability. By performing Monte Carlo simulations, such a method generates a set ofweights on a cutting level within an ELECTRE-TRI model.

The input for SMAA-TRI are the following:

the profiles;

the feasible set of weights of criteria defined as:

W = {wj R+ :

nj=1

wj = 1};

6


7/34

the cutting level;

the data and the other parameters of ELECTRE-TRI are supposed deterministic.

In SMAA-TRI, a categorization function is defined to evaluate the category k to which analternative ai is assigned as follows:

k = F(i, ), (3)

where is the set of parameters of ELECTRE-TRI.It is also introduced the following category membership function:

mki = 1, if F(i, ) = k

0, otherwise.

(4)

The category membership function is used to compute the category acceptability index ki thatis numerically a multidimensional integral over the preference parameter space.

The category acceptability index, generally expressed in percentage-wise, evaluates the stabilityof the assignment of an alternative ai to a category Ck. The index is within the range [0, 1]; if itresults 0, this means that the alternative ai is never assigned to category Ck on the basis of all theparameters randomly generated during the simulations; on the contrary if 1 the alternative ai iscertainly assigned to category Ck, considering all the parameters randomly generated during theprocedure. In this paper, SMAA-TRI has been implemented using JSMAA, an open source softwarein Java (www.smaa.fi; see Tervonen, 2012). The simulations considered in JSMAA are 10,000.

3 Description of the proposed model

Presently, the starting point of the riskiness evaluation process of an innovative SME is a BusinessPlan (BP), in which the principal used investment appraisal indicators based on cash-flow estimates,like the Net Present Value (NPV) are computed. In the evaluation process of an innovative SME,such indicators, as it will be shown in the case studies, could be useful in a first screening of theenterprise under consideration. In the case studies, a scenario analysis has been outperformed byconsidering the base-scenario and two worst scenarios.

Thus, for each enterprise we have evaluated some financial ratios and then we have compared to

the median values for each sector to which the enterprises, considered in the case studies, belong.At the same time, we have also estimated their NPV under each scenario.Besides the need of a preliminary financial analysis, with respect to the innovative SMEs, the

only evaluation of the financial factors, estimated in the business plan, is weakly significant, sinceif their specific risks are not analyzed, their economic-financial prospectives could be affected.

7


8/34

As a consequence, the problem of sorting innovative SMEs by intensity of risk requires to

consider several aspects such as multiple criteria; in particular, the non-financial (qualitative)criteria that are the most relevant ones as it has been pointed out in Section 1.First of all, in a multicriteria problem it is very important to identify all the actors present. The

principal actor involved in our decision problem is the credit officer of a bank, that is the DecisionMaker (DM) that will finance the SMEs which will result classified in the best risk class (i.e. theless risky) after the multicriteria sorting decision procedure.

At any rate, in this decision procedure an important role is also given to the experts, suchas engineers, physicists or chemists, that may help the DM in selecting the proper criteria forevaluating an innovative enterprise, especially for detecting its technological risks.

In the present paper, in order to evaluate the riskiness of innovative SMEs, we suggest theadoption of a multicriteria assignment model, ELECTRE TRI (Yu, 1992), that belongs to the family

of ELECTRE methods, firstly introduced in Roy, 1968. Specifically, ELECTRE TRI assigns aninnovative SME to some predefined risk categories by comparing it with some reference profilesdelimiting the categories.

ELECTRE TRI is implemented within the framework of a constructive approach (Roy, 1993)since its main operational phases are determined by an active participation of the DMs where theDMs preferences are not existent in her/his mind, but they are revealed during an interactivedecision process to obtain a final recommendation.

The main phases of the multicriteria judgemental rating model proposed, i.e. ELECTRE-TRI,are listed hereafter:

Phase(i) selection of the evaluation criteria;

Phase(ii) definition of the risk classes;

Phase(iii) elicitation of importance weights of criteria.

Phase (i): selection of the evaluation criteria

Firstly, in a multicriteria problem the alternatives have to be evaluated on the basis of a consistentfamily of criteria (Roy, 1985), defined as follows:

two alternatives with the same criteria evaluations have to be assigned to the same risk class

(exhaustive criteria);

an innovative alternative on which a criterion value is decreased in terms of lower risk cannotbe assigned to a lower class (consistent criteria);

8


9/34

an innovative alternative cannot be assigned to a risk class when one criterion is dropped

from the family of criteria (non-redundant criteria).

The non-financial criteria, considered in this study, will be gathered in four main groups on thebasis of the risk areas specific of an innovation (see Mazzu, 2008). The risk areas considered aredescribed and listed hereafter.

Development risk. The enterprises organization shows a weak competence to orient itselfto the necessary changes to implement a process or product innovation. Such a risk may berelated to some mistakes during the projecting phase of an innovation. Indicators of sucha risk may be the management quality, the scientific skills for the patents, or the companyprofile.

Technological risk. Of course, this risk is crucial for any innovative enterprise. Its weak-nesses in the technological skills may result in the failure of the project. The indicators ofsuch a risk are mainly related to the knowledge of the existing technologies.

Market risk. Such a risk is tied to the continuous monitoring of the potential marketand of the key competitors of the innovation under consideration. For example, its possibleindicators may be the adoption of customer evaluation models for the innovations demand,or knowledge of the competitors.

Production risk. The production of innovation may be affected by the non-flexible companystructure. The production risk indicators may be the adequacy of the production structure,dependence on the suppliers for raw materials, and the availability of testing and unit pilots.

Beyond the qualitative criteria (soft information), we also take into account some innovationindicators estimated on the basis of the Business Plans presented by the enterprises.

In literature, there are several innovation indicators. Among others, we have chosen to adopt thetwo following innovation indicators: Intangible Assets/Fixed Assets (see Brandolini and Bugamelli,2009) and R&D/Sales (see Kleinknecht et al., 2002). Both indicators are the most used in manyempirical studies especially because they are easily measurable.

In financing innovative SMEs, we consider the criteria hierarchically structured as follows:

at the first level, the criteria are denoted by Gr

where r = 1, 2, 3, 4 and 5 indicate, respectively,the criteria relative to the development (G1), technological (G2), market (G3) and production(G4) risk areas and the innovation related criteria (G5), i.e. G = {G1, G2, G3, G4, G5} (seefigure 1).

9


10/34

Set of Criteria

G

Development Risk

G1Technological Risk

G2Market Risk

G3Production Risk

G4Innovation indicators

G5

g1(1) g1(2) g

1(3) g

1(4) g

2(5) g

2(6) g

2(7) g

3(8) g

3(9) g

3(10) g

4(11) g

4(12) g

5(13) g

5(14)

Figure 1: A hierarchical family of criteria in the risk evaluation of an innovative SME.

at the second level, the set of sub-criteria relative to every group of criteria are considered

and denoted by Gr

= {gr

(1), g

r

(2), . . . g

r

(q) . . . , g

r

(nr)} with q = 1, 2, . . . , nr.At this level, we denote the set of all the sub-criteria by G where |G| = n = n1+n2+n3+n4+n5.

For the sake of simplicity, gr(j) will used to indicate any sub-criterion belonging to G with

j = 1, 2, , n. As is well-known (see Figueira et al., 2010), the ELECTRE methods require thatall the criteria are defined on a common level; henceforth we consider the criteria at the secondlevel.

3.1 Phase (ii): definition of the risk classes

Within ELECTRE-TRI, the reference profile, delimiting the risk categories, is given by the vector

bk = {bk1, bk2, . . . , bkj},

j being the criterion index and whose coordinates indicate the performance of the profile in eachof its criteria. It delimits the lower limit of category Ck+1 and the upper limit of category Ck.Delimiting the risk classes is also a DMs task. Determining categories depends on the DM, how(s)he perceives the different criteria evaluated on the basis of the alternatives under consideration.In the paper, we have considered five risk classes that are ordered from the highest level risk (riskclass 1: weak enterprises) to the lowest (risk class 5: very good enterprises).

3.2 Phase (iii): the importance weights of criteria

One of the most important and difficult tasks of MCDA is the determination of importance weightsof criteria. In literature, several methods have been proposed to assess weights of criteria (see, for ex-ample Bana e Costa and Vansnick, Bana e Costa and Vansnick, 1997, Jacquet-Lagreze and Siskos,1982 and Mousseau et al., 2001).

10


11/34

C1 C2 C3 C4 C5gr(1)

gr(2)...gr(j)

...

gr(n)b4b3b2b1

Criteria

Figure 2: Risk classes defined by the limit profiles.

In addition to the different preference models of elicitation of weights, an important ques-tion in MCDA is the relative importance of criteria (see Mousseau, 1995 and Roy and Mousseau,1996). Indeed, there is a different meaning between weights in compensatory methods, like theMulti-Attribute Utility Theory (MAUT) (Keeney and Raiffa, 1976) and non-compensatory meth-ods like ELECTRE (Roy, 1991) and PROMETHEE (Brans et al., 1984). In compensatory methods,weights of criteria are essentially trade-offs between criteria and are constants dependent on theirscale and range. On the other hand, in non-compensatory methods the weights of criteria have anintrinsic value and do not change on the basis of the scale of criteria used.

Among these, we recall the revised Simos method (Roy and Figueira, 2002), known also as thecards method, that will be used in the case studies illustrated in the next section.

Within the Simos procedure, the DM ranks the criteria from the least to the most important byusing a set of cards (one for each criterion). The DM can also insert some blank cards to separatethe relative importance between criteria. If the DM inserts no white cards between criteria, thismeans that these criteria have not the same weights and their relative difference can be taken asunit measure u for weights. With a similar reasoning, if the DM inserts one white card, two whitecards, etc, this means respectively a difference of weights of two times u, three times u, etc.

In the next section, where the case studies are presented, the cards method has been adopted toobtain ordinal information on the weights of the criteria and then to obtain indirectly the weights ofcriteria from the DMs ranking. Since in many banks the granting of a loan depends on the decision

of a pool of credit officers (the DMs), in the present paper it seems more realistic to assume differentorders of importance of weights of criteria depending on the DMs preference attitudes. In Section4.4, the case studies will be carried by considering, five hypothetical different DMs. For each DM,we apply SMAA-TRI by varying the -cutting level represented by a stochastic variable uniformlydistributed. Then, to take into account simultaneously all the DMs sets of weights, the evaluation

11


12/34

of each criterion ranges in an interval delimited by the maximum and the minimum weight for all

DMs. Since the sets of of weights are uncertain, we outperform againSMAA-TRI

, computing thecategory acceptability indices for each alternative.

4 Start-up case studies

In this section, we carry out some real case studies, with the aim of illustrating the multicriteriacredit risk model proposed. The authors have contacted four innovative SMEs headquartered inItaly requesting them for their business plan.

Furthermore, a structured questionnaire was also sent to the companies. The data collectedin the questionnaire concern some general information on the company, distribution and customer

networks, demand forecasting, supply chain information, the owners company CV and the awardspossibly received by the company. For the sake of privacy, we have renamed the companies as A,B, C and D. On the basis of the multicriteria rating model proposed in Section 3, our analysissimulates the financing of such companies.

Companies

1. Company A is a biotechnological start-up, operating in the field of the green economy. Ithas developed some biological systems based on plants and micro-organisms forming someeco-friendly barriers, to prevent soil from the hydrological destruction and the environmentalpollution. The services provided by it, consist of realizing such biological barriers, also givingconsultancy; furthermore its services are offered at lower prices than the ones of the traditionaltechniques.

2. Company B is a technological company with a strong expertise in digital communications.The innovative idea of the company is to develop a Water-MeMo that is a wireless sensornetwork for water leakage detection, based on energy harvesting. In particular, its technologycan be applied not only in the water distribution field, but also in other energy applications,such as heat and gas, or environmental monitoring. Its main innovation is in the greentechnology FluE (Fluid Energy) that is based on some sensors auto-charged by the energyproduced by the fluid itself trough some piezoelectric foils.

3. Company C is a high-technological start-up, an R & D mechanical design and service provider

company with a focus on material recovery systems applied to thin-film deposition processes.Its mission is providing breakthrough technology, in order to increase the efficiency of PVD(Physical Vapour Deposition) processes, today used to produce microchips, MEMS (MicroElectro-Mechanical Systems), solar cells and other hi-tech devices.

12


13/34

4. Company D is a high-technological start-up that aims to produce and commercialize graphene

and carbon nanotubes for industrial use and for research, to develop new materials based onnano-particles and provide technical assistance and advice to businesses willing to use thesetechnologies. The company is mainly focused in producing nano-engineered epoxy resins usedin the manufacture of sport equipment, for example sailing boats or kiteboards.

4.1 Financial analysis

The starting point of the financial analysis, presented in this section, is the enterprises businessplan, on which we have out-performed a scenario analysis. From the base-case scenario, we havedeveloped two different worst scenarios considering the cash-flows lowered by 20% and 40%.

Then, we have addressed our analysis along two directions. On one hand, we have computed

some financial ratios. Thus, only for descriptive purposes we have compared them with those of theenterprises from the same industrial/economic sector. On the other, we have evaluated the NPVsfor each scenario.

First of all, let us remark that the enterprises have considered different planning horizons;company D has presented a three years BP, company A a four year BP, company B has considereda five years BP and company C a six years BP.

To the aim of detecting the Italian firms in the same sector of the four enterprises underevaluation, we have considered their ATECO codes, elaborated by ISTAT (www.istat.it), whichidentify their enterprises industrial/economic sectors.

From AIDA dataset, we have reported different financial ratios of the balance-sheets (from 2007to 2011) of the sectors of companies A, B, C and D respectively, composed of 488, 13806, 946, and869 SMEs (enterprises with a number of employees lower than 250).

On the basis of the available financial data, the following financial ratios have been selected:ROA, Cash/Operating Revenue, EBIDTA/Total Asset, Short-term debt/Equity, Equity/Total As-set and Cash/Total Asset reflecting the areas of profitability, leverage and liquidity of each enter-prise (some of these ratios have been already selected to evaluate the US SMEs in the paper ofAltman and Sabato, 2007).

Moreover, with respect to each enterprise the averages of each financial ratio have been computedfor the years from 2007 to 2011. Finally, we have evaluated the median value of each ratio for allthe enterprises under evaluation.

Summing up, the sector observations of companies A, B, C and D are, respectively, 14,640,

28,380, 414,180 and 26,070.The financial ratios of the enterprises compared to the median values of their relative sector are

displayed in Tables 10(a), 10(b),10(c) and 10(d) in the Appendix A.The NPV under the different scenarios have been computed by using a risk free rate of 7 .93%,

commensurated with the performance of Italian public debt securities in date 30/9/2012 (Base

13


14/34

informativa pubblica, Banca dItalia, 2013).

The results are reported for each enterprise in the Tables 11, 12,13, and 14 in the Appendix A.In all the enterprises under each scenario, all the NPV result positive and consequently all theprojects are profitable.

4.2 Phase (i): building a family of criteria

Since most of the criteria, considered in this paper, are qualitative, for some of them it has beennatural to adopt a binary codification, for all the others, for simplicity, we have used the sameordinal scale on a five point scale. Let also assume that all the criteria evaluations are increasing,i.e. the more the better and consider five different credit officers (DMs) evaluating the SMEscreditworthiness.

The family of criteria have been elaborated by the credit officers with the help of the experts.Even more on the basis of the BP presented by the firms, the criteria evaluations have been givenby the credit officers in cooperation with the experts that can detect the specific risks of eachenterprise.

The criteria considered in the case studies are described hereafter (see also Table 1 for anoverview of the criteria).

Criteria related to the development risk.

[g1(1)] The medium age of the management team. Recent literature has shown the advantages of in-

cluding the criterion age in credit risks evaluation (see Grunert et al., 2005 and Altman et al.,

2010). We have decided to adopt this rule: the more the working team is young the more isconsidered innovative. For this reason, the following scale has been adopted:


15/34

Company D won the Business Plan Competition eCapital 2010, competition aimed at creating

innovative startups. In 2011, it was finalist at the Italian National Innovation Award PNI -Telecom Working Capital.

[g1(3)] Scientific skills: codes yes (1) or no (0).

Company A. The three principal partners are agronomists and are specialized in microbiologyand in plant genetics.

Company B is composed of three young engineers, a full professor, a strategic marketingexpert and ArieLab s.r.l, a spin-off of the Universit a Politecnica delle Marche.

Company C. The management team is composed of a well qualified staff. For example, theCEO has a Master degree in Engineering, or the CTO has a Master degree in Physics and

NanoScience.Company D. All the five members are well qualified; some of them have a degree in engi-neering, others are specialized in nanotechnology and one of them is specialized in sailingprototypes that are the main products in which the company is investing its research and saleactivities.

[g1(4)] Honeymoon period: codes yes (1) or no (0).

Many research papers have analyzed that the risk of failure is highest when an enterprise isa start-up, and decreases with the age of a company (see the liability of newness theory).On the contrary, in Hudson, 1987 it has been highlighted that a newly formed enterprise can

have a honeymoon period of around two years during which the risk of failure is significantlylowered. Consequently, following the same reasoning we have included as an indicator of riskto detect if the innovative enterprise is within such period.

All the four enterprises under consideration in the present study are within the honeymoonperiod. Company A has started in 2012, while companies B, C and D have been founded in2011.

Criteria related to the technological risk.

[g2(5)] Knowledge of the existing technologies: codes yes (1) or no (0). An innovative productiontechnique shouldnt replicate an already existing method. Let us observe that the four case

studies presented in this paper are introducing a process innovation. In their BP, all theenterprises demonstrate to know the other existing techniques.

[g2(6)] Upgrade capability: codes yes (1) or no (0). From their BP, all the enterprises show manygood capabilities in upgrading their techniques.

15


16/34

[g2(7)] Pros of the techniques used in comparison to other similar already existing: a five-points scale

is adopted; one point means that the advantages are irrelevant, while five point levels aregiven if they are significant.

Company A has received three points, since there are other methods that can be used toprevent soil from the hydrological destruction and the environmental pollution.

On the basis of this criterion, company B has received five points since the only similarexisting technique consists of putting a few Data Loggers in some points of the water pipelinesto register the noise by a microphone. Anyway, such products are too expensive and detectthe soundness of the water distribution una tantum.

On the basis of this criterion, company C has received five points since other existing tech-niques need too complex chemical processes to recover precious materials.

Company D has received three point levels, since the retail prices of their epoxy resins aretoo high in comparison to other existing ones.

Criteria related to the market risk.

[g3(8)] Presence of key competitors: a five-points scale is adopted, ranging from one point, if there isa monopolist competitor in the market, to five points, if the innovative enterprise is the onlyone competitor present in the market.

Company A has one competitor, that adopts a methodology much more expensive and withmany different drawbacks, for example the plants used are not very suitable for the soil.

Consequently, on the basis of such criterion Company A has obtained an evaluation of fourpoints.

Company B doesnt have competitors in the market. The only similar solution, i.e. the DataLogger has many drawbacks. The DMs suggest an evaluation of five points on this criterion.

Even if company C has registered a patent, this doesnt represent a barrier to avoid that otherenterprises could enter the market, by slightly modifying the system developed by companyC. Some possible similar solutions could be: vacuum ad hoc chambers, static monitors andmechanical removing. Anyway, the above techniques need the support of complex chemicalprocesses. On the basis of such criterion, The DMs suggest an evaluation of two points.

Company D has different competitors in nanotechnology even if operating with different prices

and quality in the their products. The DMs give an evaluation of two points on the basis ofthis criterion.

[g3(9)] Potential market: a five-points scale is adopted; a one point means that the market is reducingwhile 5 means that the market is booming.

16


17/34

Company As potential market is limited by the traditional techniques, mainly based on

chemical products, preferred to the innovative solution offered by Company A. In fact, thereis a cultural resistance to accept this new type of methodology to prevent soil from thehydrological destruction and the environmental pollution, even if nowadays there is much moreattention to environmental problems. On the basis of this criterion, a prudential evaluationof three has been considered.

With respect to this point, the DMs believe that the Company Bs market is reducing (1point). More precisely, even if the private and public companies running the water distributionin Italy need a system to monitor the water leakage, they are not financially sound.

Company C. The market of thin film deposition is booming, with an estimation of marketvalue amounting to 22.4 billion USD in 2015. The use of precious materials (Au, Ag, Pt and

Pd) is very common in optics and electronics. On this criterion, Company C has received anevaluation of five points.

Company D. Within the nanotechnology market nanomaterials are a booming business withbenefits for all other sectors. Carbon nanotubes are in this case, together with some materialsconnected to them (e.g. graphene and fullerenes), the most important products of the entiresector. On this criterion, Company D has received an evaluation of five points.

[g3(10)] Dependence on suppliers: a five-points scale is adopted. If the evaluation is one, this indicatesa high supplier dependence, while the evaluation of five stands for a low level dependence.

Company A has to be supplied continuously with numerous plants used in the biological

barriers. Presently, such plants can be obtained only from two Italian suppliers. Othersuppliers can be found abroad, but they have not a sufficient know-how to consider thespecific plants used by Company A. For this reason, an evaluation of two points has beengiven to such criterion.

The dependence from the suppliers is very high (1) for the Company B. In fact, to producethe sensors with the technique FluE some piezoelectric fibers are necessary. Presently, onlytwo companies abroad are available to supply such components.

The Company Cs supplier dependence is very high, since a complex supplier network isneeded; specifically, it is based on facilities, components and material suppliers. In this case,one point has been given.

The Company Ds supplier dependence is very high since currently Company D uses forits formulations raw materials purchased externally, but aims to internalize in the shortestpossible time the production of the key ingredients: the nanocarbons. The DMs give anevaluation of one point on this criterion.

17


18/34

Criteria related to the production risk.

[g4(11) ] Availability of testing and unit pilots: codes yes (1) or no (0).

Company A has not yet realized any pilot installations that can be useful to show to thepotential clients the advantages of the technique proposed compared to the traditional ones.

Company B hasnt yet produced any unit pilots, even if it is one of its immediate projects.

Company C has developed a first prototype and tested it in the vacuum chamber of a thermalevaporator at the laboratory of CNR.

The company Ds has started a pilot plant for the supplement of nanocarbons filler to ther-mosetting epoxy resins.

[g4(12)] Owner of a patent: codes yes (1) or no (0).

Company A hasnt yet applied for a patent even if it is one of its projects.

Company B hasnt yet registered the patent for the FluE sensor, even if it is one of its goals.

The company C has a right of an Italian patent entitled: Palette system for the recovery ofmetals from thin-film deposition facilities and also of an International application entitled:Palette modular device for collection of metals in this film deposition equipment.

The company Ds hasnt yet applied for a patent, but it aims to record strategic patents onits core product and a series of multiple patents that go to cover similar areas of research anddevelopment thus protecting future market segments.

Innovation indicators

[g5(13) ] Intangible Assets/Fixed Assets;

[g5(14) ] R&D/Sales.

Summing up, for each enterprise the scores on the considered criteria are showed in Table 2.

4.3 Phase (ii): determining the limit profiles

The limit profiles are built by the DMs who determine the performance on each criterion withrespect to every category. Table 3 shows the 4 limit profiles for the five under considerationcategories.

For simplicity, the preference and indifference thresholds have been set equal to zero.

18


19/34


20/34


21/34

The DM4s preference information is given by:

g1(1) g1(4) g

5(13) g

5(14) g

1(2) g

1(3) g

3(8) g

3(10) g

3(11) g

2(5) g

2(6) g

2(7) g

3(9) g

4(12).

In this case, the most important criterion is assumed to be the patents ownership and the secondmost important one is the potential market. In all the four DMs, the criteria in the pairs (g1(1), g

1(4)),

(g1(2), g1(3)), (g

2(6), g

2(7)) and (g

5(13), g

5(14)) are considered equally important, while the least important

criteria are the medium age (g1(1)) and the honeymoon period (g1(4)). The last DM5 considers all

the criteria equally important.As explained in Section 3.2, the method adopted to assess the criteria weights is the Simos

procedure (the computations relative to the DM1s weights are reported in the Appendix). For all

the DMs, the criteria weights obtained are showed in Table 4.

Table 4: Weights for each DM.g1(1) g

1(2) g

1(3) g

1(4) g

2(5) g

2(6) g

2(7) g

3(8) g

3(9) g

3(10) g

4(11) g

4(12) g

5(13) g

5(14)

DM1 0.021 0.034 0.034 0.021 0.071 0.071 0.149 0.059 0.059 0.059 0.171 0.159 0.046 0.046DM2 0.018 0.052 0.052 0.018 0.059 0.059 0.080 0.143 0.148 0.134 0.127 0.032 0.039 0.039DM3 0.017 0.107 0.107 0.017 0.116 0.116 0.125 0.044 0.044 0.035 0.077 0.071 0.062 0.062DM4 0.020 0.044 0.044 0.020 0.093 0.093 0.134 0.060 0.142 0.060 0.068 0.150 0.036 0.036DM5 0.071 0.071 0.071 0.071 0.071 0.071 0.071 0.071 0.071 0.071 0.071 0.071 0.071 0.071

4.5 Computing the category acceptability indicesIn this section, we have computed the category acceptability indices, according to the differentDMs, by using the JSMAA software (the results are showed in Table 5). In all the computations,the -cutting level has been represented by a stochastic variable uniformly distributed in the range[0.6, 0.85].

To take into account the preferences of all the DMs, we consider, as already done in the paperby Morais et al., 2012, the minimum and the maximum weight for each criterion for all the DMs.

In this way, the weight of each criterion ranges in an interval reflecting all the DMs attitudes.The results obtained are showed in Table 6.

By performing again the JSMAA software, we compute the category acceptabilities presented

in Table 6(b) (see also Figure 3 for a representation of category acceptabilities in terms of anhistogram).

21


22/34


23/34

Figure 3: Histogram of the category acceptabilities for all the DMs.

Company A Company B Company C Company D

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Categoryaccepatbilitiesindices

Category 1

Category 2

Category 3

Category 4

Category 5

5 Discussion and managerial implications

In the previous section, all the simulations in SMAA-TRI have been out-performed by varying inthe range [0.65, 0.8] to analyze the robustness of ELECTRE-TRI. We have reported in Table 7 thehighest category acceptability for each DM obtained during the simulations. One can notice thatthe assignment of company C to the category C5 is very stable; also the assignment of CompanyA to the class C4 is quite stable. Some incompatibilities among the DMs arise for the enterprisesB and D. More precisely, Company D should be assigned to class C3 or C5, while the assignment

of Company B to a class risk is very uncertain. This might depend on which DM is prevailing, butif the DMs try to reach a consensus, they dont manage to give a stable assignment to CompanyB. At this point, it could be useful to consider the financial analysis performed in Section 4.1. Onthe basis of the comparison with the enterprises of the same sector, while the project of CompanyD is profitable, Company B is not quite sound from a financial point of view. The evaluation ofthe financial soundness, as outlined in Section 4.1, could help the DMs in making the appropriatechoice in the assignment of an enterprise to a risk class.

Consequently, from this last enterprise one can observe that the judgemental rating modelproposed suffers of some limitations that can arise when the DMs dont agree in an unique systemof weights.

The last step of our analysis consists of fixing the cutting level . As a result, SMAA-TRIcoincides with ELECTRE-TRI, since for each DM the preference parameters are fixed.

On the choice of the cutting level , there are several discussions in multicriteria literature(see among others Damart et al., 2007 and Merad et al., 2004). Anyway, should be at least 0.5,meaning that at least 50% of the criteria should be in favour of the entry of an enterprise into the

23


24/34

Table 7: Results of SMAA-TRI for each DM with varying in the range [0.65, 0.8].

Company DM1 DM2 DM3 DM4 DM5 all the DMsA C4(72%) C4(93%) C5(60%) C4(75%) C4(65%) C4(66%)B C3(51%) C2(53%) C5(65%) C3(79%) C5(41%) C4(47%)C C5(99%) C3(58%) C5(100%) C5(75%) C5(81%) C5(81%)D C3(61%) C3(72%) C5(79%) C3(56%) C5(50%) C5(47%)

class under consideration. Since in the previous section we have taken in the range [0.6, 0.85], wehave decided to choose the extremes of the interval, i.e. = 0.6 and = 0.85. In this case, theresults obtained for = 0.6 and = 0.85, are showed, in Tables 8 and 9, respectively.

Table 8: Results of ELECTRE-TRI with = 0.6.Company DM1 DM2 DM3 DM4 DM5

A C4 C4 C5 C4 C5B C4 C4 C5 C3 C5C C5 C5 C5 C5 C5D C5 C4 C5 C5 C5

Table 9: Results of ELECTRE-TRI with = 0.85.Company DM1 DM2 DM3 DM4 DM5

A C3 C3 C4 C3 C4B C4 C2 C4 C2 C3C C4 C3 C5 C5 C3D C4 C3 C3 C3 C3

From Table 8, one can notice that even if the DMs would assign company B to a risk class,there isnt a consensus of a majority of DMs. In all the other companies, there are at least threeof the five DMs that agree on the assignment of the same risk class. In Table 9, the same situationcan be observed for Companies B and C.

Moreover, it is still important to emphasize that in the judgemental credit risk model considered,

we have never introduced any veto criterion. The introduction of veto criteria is expected to havea considerable impact on the decision procedure. In fact, let us suppose for example to considerthe criterion availability of testing and unit pilots (g4(11)) as a veto criterion. One implication of the

inclusion of this veto criterion is that, when an enterprise doesnt have a unit pilot (thus receiving a0 point in the evaluation according to this criterion), the criterion would be against the assignment

24


25/34

of such an enterprise to a good risk class. Such enterprise would be assigned to the class C1, even

if it has a good evaluation on the majority of criteria. In the case studies, for example only theenterprise Company C has a good evaluation on this criterion. As a result, in many real situations,it may be very relevant to impose some veto criteria.

6 Conclusions

Many recent studies have been dedicated to measuring SMEs innovation considered as a crucialfactor for the development of a national economy (Brandolini and Bugamelli, 2009). To achieve ahigh level of innovation the main obstacle that the innovative SMEs face is a non straightforwardaccess to credit. For example in Italy, such asymmetric information is due to the small dimension

of most of the innovative SMEs (Bugamelli et al., 2012). To help innovative enterprises, financialinstitutions could try to reduce such informative asymmetries. Presently, banks assess their creditrisk by evaluating the business plan and considering some non-financial information such as the ageof the entrepreneurs, the market trend and the quality of the management team. Although financinginnovation isnt an easy task due to the lack of track records of innovative SMEs, we believe that arating model based upon experts judgments could improve it. In financing innovative SMEs, therole of the experts is crucial, since they can help the credit officers in selecting the proper criteria,especially, in detecting their risks.

In this paper, we have presented a multicriteria approach to sort innovative enterprises into riskclasses, mainly on the basis of soft information.

Specifically, we describe a possible hierarchical structure of the innovation risks: development,

production, market and technological ones. Such risk indicators have been considered together withtwo innovation indicators, mostly used in the literature that is intangible assets divided to fixedassets and R & D expenditure divided by sales. The multicriteria approach proposed is ELECTRE-TRI based on an outranking preference relation comparing each innovation to some existing riskprofiles. As explained in the paper, the multidimensional and complex decisional framework offinancing innovation is well adapted to the aforementioned multicriteria method.

In fact, the credit officers with the help of the experts could define the risk profiles, the criteriaweights, but could also tune some specific parameters such as the cutting level, the preference,indifference and veto thresholds on which the final ratings depend.

Finally, we envisage some possible research lines for the future:

It could be interesting to detect the most critical criteria governing the decision makingprocess. What is meant by the term critical here is the smallest change that might occurto a certain criterion in order increase the category acceptability of the better risk class towhich each enterprise has been assigned.

25


26/34

ELECTRE-TRI is a non-compensatory method; maybe it could be useful to consider also

other multicriteria models including interaction between criteria (see the Choquet integralpreference model among the multicriteria approaches representing interaction between cri-teria e.g. in Angilella et al., 2004 and its recent extension within a SMAA methodology inAngilella et al., 2012).

Since the criteria adopted in this paper are hierarchally structured, it could be useful to applythe recent approach of the hierarchal Choquet integral presented in Angilella et al., 2013.

References

E.I. Altman and G. Sabato. Effect of the new capital accord on bank capital requirements forSMEs. Journal of Financial Services Research, 28:1542, 2005.

E.I. Altman and G. Sabato. Modelling credit risk for SMEs: Evidence from the U.S. market.Abacus, 43(3):332357, 2007.

E.I. Altman, G. Sabato, and N. Wilson. The value of non-financial information in small andmedium-sized enterprise risk management. The Journal of Credit Risk, 6(2):133, 2010.

S. Angilella, S. Greco, F. Lamantia, and B. Matarazzo. Assessing non-additive utility for multicri-teria decision aid. European Journal of Operational Research, 158(3):734744, 2004.

S. Angilella, S. Corrente, and S. Greco. SMAA-Choquet: Stochastic multicriteria acceptabilityanalysis for the Choquet integral. In S. Greco et al., editor, Advances in Computational Intel-ligence, volume 300 of Communications in Computer and Information Science, pages 248257.Springer Berlin Heidelberg, 2012.

S. Angilella, S. Corrente, S. Greco, and R. Slowinski. Multiple criteria hierarchy process for theChoquet integral. In R.C. Purshouse et al., editor, EMO 2013, volume 7811 of LNCS, pages475489. Springer Berlin Heidelberg, 2013.

H.V. Auken, G.H. Canovas, and A.M. Guijarro. Role of information in the debt financing oftechnology-based firms in Spain. In The Entrepreneurial Society: How to Fill the Gap BetweenKnowledge and Innovation. Edward Elgar Publishing Inc., 2010.

C.A. Bana e Costa and J.C. Vansnick. A theoretical framework for measuring attractiveness by acategorical based evaluation technique (MACBETH). In Proc. XIth Int. Conf. on MulticriteriaDecision Making, 15-24, Coimbra, Portugal, August 1994.

26


27/34

C.A. Bana e Costa and J.C. Vansnick. Applications of the MACBETH approach in the framework

of an additive aggregation model. Journal of Multi-criteria Decision Analysis, 6:107114, 1997.Basel Committee on Banking Supervision. International Convergence of Capital Measurement and

Capital Standards. www.bis.org, June 2006.

T. Beck and A. Demirgurc-Kunt. Small and medium-size enterprises: Access to finance as a growthconstraint. Journal of Banking & Finance, 30(11):2931 2943, 2006.

T. Beck, A. Demirgurc-Kunt, and M.S. Martnez Pera. Bank financing for SMEs: Evidence acrosscountries and bank ownership types. Journal of Financial Services Research, 39(1-2):3554, 2011.

A.N. Berger and G.F. Udell. Financial system design: the case for universal banking. Irwin (Richard

D), Burr Ridge, chapter Universal Banking and the Future of Small Business Lending, pages 559 627. 1996.

A.N. Berger and G.F. Udell. A more complete conceptual framework for SME finance. Journal ofBanking & Finance, 30(11):2945 2966, 2006.

A.N. Berger, N.H. Miller, M.A. Petersen, R.G. Rajan, and J.C. Stein. Does function follow or-ganizational form? evidence from the lending practices of large and small banks. Journal ofFinancial Economics, 76(2):237 269, 2005.

A. Brandolini and M. Bugamelli. Report on trends in the Italian productive system. Questioni diEconomia e Finanza (Occasional Papers) 45, Bank of Italy, Economic Research and InternationalRelations Area, April 2009.

J.P. Brans and Ph. Vincke. A preference ranking organization method. Management Science, 31(6):647656, 1985.

J.P. Brans, B. Mareschal, and Ph. Vincke. Operational Research 84, chapter PROMETHEE: Anew family of outranking methods in multicriteria analysis, pages 408421. Elsevier SciencePublishers B.V. (North-Holland), 1984.

R.J. Brown, S.M. Fazzari, and C.B. Petersen. Financing innovation and growth: Cash flow, externalequity, and the 1990s R&D boom. The Journal of Finance, 54(1):151185, 2009.

M. Bugamelli, L. Cannari, F. Lotti, and S. Magri. The innovation gap of Italys production system,roots and possible solutions. Questioni di Economia e Finanza (Occasional Papers) 121, Bank ofItaly, Economic Research and International Relations Area, April 2012.

27


28/34

R. Canales and R. Nanda. A darker side to decentralized banks: Market power and credit rationing

in SME lending. Journal of Financial Economics, 105(2):353 366, 2012.D. Czarnitzki and H. Hottenrott. R & D investment and financing constraints of small and medium-

sized firms. Small Business Economics, 36:6583, 2011.

S. Damart, L. Dias, and V. Mousseau. Supporting groups in sorting decisions: Methodology and useof a multi-criteria aggregation/disaggregation DSS. Decision Support Systems, 43(4):14641475,2007.

M. Doumpos and C. Zopounidis. A multicriteria decision support system for bank rating. DecisionSupport Systems, 50(1):5563, 2010.

M. Doumpos, K. Kosmidou, G. Baourakis, and C. Zopounidis. Credit risk assessment using amulticriteria hierarchical discrimination approach: a comparative analysis. European Journal ofOperational Research, 138(2):392 412, 2002.

J. Figueira, S. Greco, B. Roy, and R. Slowinski. Handbook of Multicriteria Analysis, chapterELECTRE Methods: main features and recent developments. Springer, New York, 2010.

S. Greco, B. Matarazzo, and R. Slowinski. Comments on: Multicriteria decision systems for financialproblems. TOP, pages 17, 2013.

J. Grunert, L. Norden, and M. Weber. The role of non-financial factors in internal credit ratings.Journal of Banking & Finance, 29:509531, 2005.

J. Hudson. The age, regional, and industrial structure of company liquidations. Journal of BusinessFinance & Accounting, 14(2):199213, 1987.

E. Jacquet-Lagreze. An application of the UTA discriminant model for the evaluation of R &D projects. In P. M. Pardalos, Yannis Siskos, and Constantin Zopounidis, editors, Advances inMulticriteria Analysis, volume 5 ofNonconvex Optimization and Its Applications, pages 203211.Springer US, 1995.

E. Jacquet-Lagreze and J. Siskos. Assessing a set of additive utility functions for multicriteriadecision-making, the UTA method. European Journal of Operational Research, 10(2):151164,

1982.

R.L. Keeney and H. Raiffa. Decisions with multiple objectives: Preferences and value tradeoffs. J.Wiley, New York, 1976.

28


29/34

A. Kleinknecht, K. Van Montfort, and E. Brouwer. The non-trivial choice between innovation

indicators. Economics of Innovation and New Technology, 11(2):109121, 2002.S. Mazzu. Il finanziamento dellinnovazione. Strumenti, rischi e modelli di valutazione. G. Giap-

pichelli, 2008.

M. Merad, T. Verdel, B. Roy, and S. Kouniali. Use of multi-criteria decision-aids for risk zoningand management of large area subjected to mining-indiced hazards. Tunnelling and UndergroundSpace Technology, 19:125 138, 2004.

D.C. Morais, A.T. de Almeida, and J.R. Figueira. A sorting model for group decision making: Acase study of water losses in Brazil. Group Decision and Negotation, 2012.

A. Moro and M. Fink. Loan managers trust and credit access for SMEs. Journal of Banking &Finance, 37(3):927 936, 2013.

V. Mousseau. Advances in Multicriteria Analysis. Nonconvex Optimization and its Applications,volume 5, chapter Eliciting information concerning the relative importance of criteria, pages1743. Kluwer Academic Publishers, Dordrecht, 1995.

V. Mousseau, J. Figueira, and J.P. Naux. Using assignment examples to infer weights for ELECTRE-TRI method: Some experimental results. European Journal of Operational Research, 130(2):263275, 2001.

C. Pederzoli, G. Thoma, and C. Torricelli. Modelling credit risk for innovative SMEs: the role ofinnovation measures. Journal of Financial Services Research, pages 119, 2012.

B. Roy. Classement et choix en presence de points de vue multiples: La methode ELECTRE. RevueFrancaise dInformatique et de Recherche Operationnelle, 8:5775, 1968.

B. Roy. Methodologie Multicritere dAide a la Decision. Economica, Paris., 1985.

B. Roy. The outranking approach and the foundations of ELECTRE methods. Theory and Decision,31(1):4973, 1991.

B. Roy. Decision science or decision-aid science? European Journal of Operational Research, 66(2):

184203, 1993.B. Roy and J.R. Figueira. Determining the weights of criteria in the ELECTRE type methods with

a revised Simos procedure. European Journal of Operational Research, 139:317326, 2002.

29


30/34

B. Roy and V. Mousseau. A theoretical framework for analyzing the notion of relative importance

of criteria. Journal of Multi-criteria Decision Analysis,, 5:145159, 1996.D. Shefer and A. Frenkel. R & D, firm size and innovation: an empirical analysis. Technovation,

25(1):25 32, 2005.

T. Tervonen. JSMAA: open source software for SMAA computations. International Journal ofSystems Science, pages 113, 2012.

T. Tervonen, R. Lahdelma, J. Almeida Dias, J.R. Figueira, and P. Salminen. SMAA-TRI. InI. Linkov, G.A. Kiker, and R.J. Wenning, editors, Environmental Security in Harbors and CoastalAreas, NATO Security through Science Series, pages 217231. Springer Netherlands, 2007.

T. Tervonen, J.R. Figueira, R. Lahdelma, D.A. Juscelino, and P. Salminen. A stochastic methodfor robustness analysis in sorting problems. European Journal of Operational Research, 192(1):236242, 2009.

Ph. Vincke. Multicriteria decision-aid. Wiley, New York, 1992.

F. Voulgaris, M. Doumpos, and C. Zopounidis. On the evaluation of Greek industrial SMEsperformance via multicriteria analysis of financial ratios. Small Business Economics, 15(2):127136, 2000.

W. Yu. ELECTRE TRI: Aspects methodologiques et manuel dutilisation. Document du LAMSADENo.74, Universite Paris-Dauphine, June 1992.

C. Zopounidis and M. Doumpos. Multicriteria classification and sorting methods: A literaturereview. European Journal of Operational Research, 138:229246, 2002.

C. Zopounidis and M. Doumpos. A multicriteria outranking modeling approach for credit rating.Decision Sciences, 42(3):721742, 2011.

30


31/34


32/34


33/34

Table 14: Scenario analysis on the Company Ds cash-flows.

First Scenarioyear 1 year 2 year 3 NPV (7.93%)

cash-flow (e) -62,272.07 204,057.11 1,094,740.87 988,208.95Second Scenario

year 1 year 2 year 3 NPV (7.93%)

cash-flow (e) -74,726.48 163,245.68 875,792.69 767,488.48

Third Scenario

year 1 year 2 year 3 NPV (7.93%)

cash-flow (e) - 87,180.89 122,434.26 656,844.52 546,768.00

Appendix B

To assess the criteria weights within a Simosprocedure, the following variables are defined:

z is the ratio expressing how many times the last criterion is more important than the firstone in the ranking;

er is the number of white cards between the rank r and r + 1;

er = e

r + 1;

u = z1e

;

e =n1r=1

er.

The non normalized weight k(r) is computed by:

k(r) = 1 + u(e0 + + er1),

with e0 = 0.

Let ki = k(r) be the weight relative to the criterion i and let K =

ni=1

ki the sum of the

non-normalized weights. The normalized weight ki is computed by:

ki =1

Kki.

Within the Simos procedure, the preference information expressed by the DM1 is reported inTable 15(a). Thus, the normalized weights are those displayed in Table 15(b).

33


34/34

Table 15: Simos method(a) DM1s set of weights with z = 8.

Rank r Criteria n. of white er Non-normalized Totalin the rank r cards weights k(r)

1 {g1(1), g1(4)} 0 1 1.0 2.0

2 {g1(2), g1(3)} 0 1 1.6 3.2

3 {g5(13), g1(14)} 0 1 2.2 4.3

4 {g3(8), g3(9), g

3(10)} 0 1 2.8 8.3

5 {g2(5), g2(6)} 5 6 3.3 6.7

6 {g2(7)} 0 1 6.8 6.8

7 {g4(12)} 0 1 7.4 7.4

8 {g4

(11)} 8.0 8.0Total e 12 K 46.7

(b) Normalized weights for the DM1.

g1(1) g1(2) g

1(3) g

1(4) g

2(5) g

2(6) g

2(7) g

3(8) g

3(9) g

3(10) g

4(11) g

4(12) g

5(13) g

5(14)

ki 0.021 0.034 0.034 0.021 0.071 0.071 0.149 0.059 0.059 0.059 0.171 0.159 0.046 0.046

34

the financing of innovative smes: a multicriteria credit rating model

Documents