The Financing of Innovative SMEs: a multicriteria credit rating model `2 SILVIA ANGILELLA1 SEBASTIANO MAZZU

arXiv:1308.0889v1 [q-fin.RM] 5 Aug 2013

Department of Economics and Business University of Catania, Corso Italia, 55, Catania 95129, Italy August 6, 2013

Abstract Small Medium-sized Enterprises (SMEs) face many obstacles when they try to access credit market. These obstacles are increased if the SMEs are innovative. In this case, nancial data are insucient or even not reliable. Thus, when building a judgemental rating model, mainly based on qualitative criteria (soft information), it is very important to nance SMEs activities. Until now, there isnt a multicriteria credit risk model based on soft information for innovative SMEs. In this paper, we try to ll this gap by presenting a multicriteria credit risk model, specically, ELECTRE-TRI. To obtain robust SMEs assignments to the risk classes, a SMAA-TRI analysis is also implemented. In fact, SMAA-TRI incorporates ELECTRE-TRI by considering dierent sets of preference parameters with Monte Carlo simulations. Finally, we carry out some real case studies, with the aim of illustrating the multicriteria credit risk model proposed. Keywords: Multiple Criteria Decision Aiding; qualitative criteria; SMAA-TRI; innovative SMEs; judgmental credit rating model.

Introduction

Since small and medium-sized enterprises (SMEs) are the backbone of all economies, scholarly attention to their credit risk assessment has considerably increased. SMEs face many nancial restrictions to access the credit market (for example see Berger et al., 2005, Beck and Demirg ur c-Kunt, 2006 and Canales and Nanda, 2012). Moreover, if the SMEs are newly created rms innovative, these restrictions are more tight (see Brown et al., 2009). Consequently, there exists a nance gap within the SMEs sector, and such a condition is worsened when SMEs have to request credit to the banks. Banks usually adopt dierent lending approaches that can be grouped in four categories: nancial statement lending (based on the evaluation of information from balance-sheet data), asset-based lending (based on the provision of a collateral), credit scoring models (based on statistical techniques), and relationship lending (see Moro and Fink, 2013).
1 2

Email address: angisil@unict.it Email address: s.mazzu@unict.it

In nancing 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 nance SMEs on the basis of a long and strong activity of relationship banking, a type of lending mainly based on non nancial information (soft information) (see among others Moro and Fink, 2013 and Berger and Udell, 1996). On the other, there is a recent stream of research that asserts that big banks lend more easily to SMEs if lending techniques based on hard information are adopted (see for example Berger and Udell, 2006 and Beck et al., 2011). Following this recent research direction, Pederzoli et al., 2012 present a credit scoring model based on nancial ratios and innovation measures to predict the 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 rating advanced 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 rst component (top-down logic) or the second one (bottom-up logic). In the rst case, the rating is mechanical, and considers mainly nancial criteria (hard information) of the enterprise under evaluation, while in the second one, the assignment of enterprises to risk classes is based on the subjective analysis of non-nancial criteria (soft information). Recently, even if Basel III has introduced some tighter constraints in terms of banks capital requirements (e.g. leverage ratio, countercyclical buers, liquidity risk, revision of the components Tier 1 e Tier 2), it has left the regulation of Basel II unchanged giving freedom to the banks to determine the mix between the quantitative and qualitative components of a credit risk model. Consequently, owing to the SMEs lack of sucient or reliable track records, the judgmental rating, mainly relying on soft information, seems the most useful approach to evaluate the SMEs creditworthiness. In fact, the signicant role of the non-nancial criteria for the SMEs credit risk evaluation is supported 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 specically, for innovative SMEs see for example Shefer and Frenkel, 2005 and Czarnitzki and Hottenrott, 2011). 2

Starting from the broad literature on this topic, our paper aims at building a judgmental rating model to nancing innovative SMEs. Until now, to the best of our knowledge, there isnt a judgemental credit risk model for innovative SMEs. In this paper, we try to ll this gap by adopting a Multiple Criteria Decision Aid (MCDA) approach (Vincke, 1992). Since the nancing of innovative SMEs is an uncertain problem due to several and conicting aspects, the Multiple Criteria Decision Aid approach (also called the constructive approach Roy, 1993) seems the most useful (see Greco et al., 2013 for some useful comments on this topic). Presently, several multicriteria approaches have already been adopted to predict business failures, that is a typical sorting problem (see Zopounidis and Doumpos, 2002). Most of them rely on a utility function to classify enterprises into two categories: the default and non-defaulted (see among 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 outranking relation; for example, ELECTRE TRI (rstly, introduced in Yu, 1992) has been implemented for the rst time as a quantitative credit rating model in Zopounidis and Doumpos, 2011 and PROMETHEE II (Brans and Vincke, 1985) applied for banks rating evaluation (Doumpos and Zopounidis, 2010). Moreover, a rst attempt to estimate SMEs performance, based on a nancial ratio analysis, can be found in Voulgaris et al., 2000, where the multicriteria sorting approach, UTADIS (Jacquet-Lagr` eze, 1995) has been applied. However, none of these papers aims to develop a judgemental 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 to the risk classes, a SMAA-TRI analysis (Tervonen et al., 2007) has also been implemented. In fact, SMAA-TRI incorporates ELECTRE-TRI by considering dierent sets of preference parameters with Monte Carlo simulations. The second purpose of this paper is to illustrate the multicriteria methodology proposed by analyzing some real case studies, specically, some Italian innovative enterprises in the phase of start-up. Thus, by playing the role of dierent credit ocers, we have simulated the nancing of these innovative enterprises. Moreover, preliminary to the multicriteria analysis, we have evaluated the nancial soundness of the start-up considered outperforming a scenario analysis on the basis of their Business Plans. The nancial analysis has been conducted along two directions. On one hand, nancial ratios have been evaluated and compared with those of the sector to which each start-up belongs. On the other, a NPV for each scenario has been computed. The paper is organized as follows. In Section 2, we provide a brief description of ELECTRETRI and SMAA-TRI. In Section 3, we describe the main phases of the multicriteria rating model proposed to evaluate the riskiness of innovative SMEs. Some real case studies are carried out in 3

Section 4 to illustrate the judgemental multicriteria credit risk rating presented. Section 5 gives some useful managerial insights of the methodology considered and nally, some conclusions are included in Section 6.

A sorting model

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

2.1

An overview of ELECTRE-TRI method

Let A = {a1 , a2 , . . . , ai , . . . , am } be a nite set of m alternatives to be assigned on the basis of a consistent 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 proles: bp1 , , bk , , b1 . Each prole bk is the upper limit of the group k and the lower limit of group k + 1. For example, considering two ordered classes, C2 could represent the nanced enterprises, while C1 groups the ones non-funded. As a result, one risk prole b1 delimits two risk categories to which an enterprise has to be assigned. The alternatives are assigned to the risk classes on the basis of their comparisons with the risk proles exploiting the two outranking relations ai Sbk and bk Sai , which mean, respectively, that alternative ai is at least as good as the prole bk and vice versa. The exploitation of each outranking relation consists of two phases: the concordance and discordance test. Roughly speaking, the concordance test indicates the level of majority of criteria that supports the outranking relation ai Sbk (or bk Sai ), while the discordance test represents the strength of the minority of criteria that oppose a veto to the outranking relation ai Sbk (or bk Sai ). The concordance test is performed by the computation of the concordance index C (ai , bk ):
n

C (ai , bk ) =
j =1

wj cj (aij , bk ), 4

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 the prole bk . The partial concordance index cj (aij , bk ) is computed as follows: 0 if aij bkj pj aij bkj +pj if bkj pj < aij < bkj qj cj (aij , bk ) = p j q j 1 if aij bkj qj ,

where vj pj is a veto threshold expressing the smallest dierence aij bk which is incompatible with the statement ai Sbk . Finally, a credibility index of the outranking relation is computed by: (ai , bk ) = C (ai , bk )
j T

where pj qj 0 indicate the preference and indierence thresholds, respectively. Such thresholds take into account the imprecise criteria evaluations. The preference threshold represents the smallest dierence aij bk compatible with the preference of ai on criterion j , while the indierence threshold indicates the largest indierence aij bk that preserves indierence between ai and bk . Similarly, the concordance index C (bk , ai ) can also be dened. The second phase consists in the discordance test, that is performed by the computation of the discordance index dj (aij , bk ) with respect to criterion j . Such index is dened as follows: 0 if aij bkj pj bkj aij pj if bkj vj < aij < bkj pj dj (aij , bk ) = vj pj 1 if aij bkj vj ,

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 ai Sbk has to be defuzzyed on the the basis of a user-dened threshold [0.5, 1]. The relation ai Sbk is valid if and only if (ai , bk ) > . Similarly, the outranking relation bk Sai is validated if and only if (bk , ai ) > . Choosing a value 0.5, is equivalent to saying that at least 50% of the criteria considered are 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 the proles bp1 , b2 . . . b1 ; if ai Sbk , then ai is assigned to the category Ck+1 , otherwise ai is assigned to C1 , i.e. the worst category; 5

 Optimistic assignment (disjunctive logic): each alternative being compared to the proles b1 , b2 , . . . , bp1 ; if bk Sai ai Sbk , then ai is assigned to the category Ck and otherwise ai is assigned to the group Cp , i.e. the best category. Some troublesome situations of incomparability could arise if the two preference statements are veried: ai Sbk and bk Sai . In the present paper, we adopt the pessimistic rule that is most used in practice, assigning every alternative ai to the category Ck+1 if the following inequalities are veried: (ai , bk ) and (bk , ai ) < . If there arent veto criteria, it is easy to verify that
n

(1)

(ai , bk ) = C (ai , bk ) =
j =1

wj cj (ai , bk ).

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

n n

wj cj (ai , bk ) and
j =1 j =1

wj cj (bk , ai ) < .

(2)

2.2

SMAA-TRI

In this section, we introduce SMAA-TRI, rstly introduced in Tervonen et al., 2007 (see also Tervonen 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 required as input by ELECTRE-TRI. In this paper, SMAA-TRI will be used to analyze the robustness of ELECTRE-TRI based on the parameter stability. By performing Monte Carlo simulations, such a method generates a set of weights on a cutting level within an ELECTRE-TRI model. The input for SMAA-TRI are the following: the proles; the feasible set of weights of criteria dened as:
n

W = { w j R+ :
j =1

w j = 1 };

 the cutting level; the data and the other parameters of ELECTRE-TRI are supposed deterministic. In SMAA-TRI, a categorization function is dened to evaluate the category k to which an alternative ai is assigned as follows: k = F (i, ), where is the set of parameters of ELECTRE-TRI. It is also introduced the following category membership function: mk i = 1, if F (i, ) = k 0, otherwise. (4) (3)

k The category membership function is used to compute the category acceptability index i that is numerically a multidimensional integral over the preference parameter space. The category acceptability index, generally expressed in percentage-wise, evaluates the stability of the assignment of an alternative ai to a category Ck . The index is within the range [0, 1]; if it results 0, this means that the alternative ai is never assigned to category Ck on the basis of all the parameters randomly generated during the simulations; on the contrary if 1 the alternative ai is certainly assigned to category Ck , considering all the parameters randomly generated during the procedure. In this paper, SMAA-TRI has been implemented using JSMAA, an open source software in Java (www.smaa.fi; see Tervonen, 2012). The simulations considered in JSMAA are 10,000.

Description of the proposed model

Presently, the starting point of the riskiness evaluation process of an innovative SME is a Business Plan (BP), in which the principal used investment appraisal indicators based on cash-ow 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 rst screening of the enterprise under consideration. In the case studies, a scenario analysis has been outperformed by considering the base-scenario and two worst scenarios. Thus, for each enterprise we have evaluated some nancial 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 nancial analysis, with respect to the innovative SMEs, the only evaluation of the nancial factors, estimated in the business plan, is weakly signicant, since if their specic risks are not analyzed, their economic-nancial prospectives could be aected. 7

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-nancial (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 ocer of a bank, that is the Decision Maker (DM) that will nance the SMEs which will result classied in the best risk class (i.e. the less risky) after the multicriteria sorting decision procedure. At any rate, in this decision procedure an important role is also given to the experts, such as engineers, physicists or chemists, that may help the DM in selecting the proper criteria for evaluating an innovative enterprise, especially for detecting its technological risks. In the present paper, in order to evaluate the riskiness of innovative SMEs, we suggest the adoption of a multicriteria assignment model, ELECTRE TRI (Yu, 1992), that belongs to the family of ELECTRE methods, rstly introduced in Roy, 1968. Specically, ELECTRE TRI assigns an innovative SME to some predened risk categories by comparing it with some reference proles delimiting 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 the DMs preferences are not existent in her/his mind, but they are revealed during an interactive decision process to obtain a nal 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) denition 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 consistent family of criteria (Roy, 1985), dened 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 cannot be assigned to a lower class (consistent criteria );

 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-nancial criteria, considered in this study, will be gathered in four main groups on the basis of the risk areas specic of an innovation (see Mazz` u, 2008). The risk areas considered are described and listed hereafter. Development risk. The enterprises organization shows a weak competence to orient itself to the necessary changes to implement a process or product innovation. Such a risk may be related to some mistakes during the projecting phase of an innovation. Indicators of such a risk may be the management quality, the scientic skills for the patents, or the company prole. Technological risk. Of course, this risk is crucial for any innovative enterprise. Its weaknesses in the technological skills may result in the failure of the project. The indicators of such 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 market and of the key competitors of the innovation under consideration. For example, its possible indicators 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 aected by the non-exible company structure. 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 innovation indicators 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 the two 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 many empirical studies especially because they are easily measurable. In nancing innovative SMEs, we consider the criteria hierarchically structured as follows: at the rst 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 } (see gure 1).

Set of Criteria G

Development Risk G1
1 g(1) 1 g(2) 1 g(3) 1 g(4)

Technological Risk G2
2 g(5) 2 g(6) 2 g(7) 3 g(8)

Market Risk G3
3 g(9) 3 g(10)

Production Risk G4
4 g(11) 4 g(12)

Innovation indicators G5
5 g(13) 5 g(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 r r r r and denoted by Gr = {g(1) , g(2) , . . . g( q ) . . . , g(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 .
r For the sake of simplicity, g( 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 that all the criteria are dened on a common level; henceforth we consider the criteria at the second level.

3.1

Phase (ii): denition of the risk classes

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

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

j being the criterion index and whose coordinates indicate the performance of the prole in each of 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 dierent criteria evaluated on the basis of the alternatives under consideration. In the paper, we have considered ve risk classes that are ordered from the highest level risk (risk class 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 dicult tasks of MCDA is the determination of importance weights of criteria. In literature, several methods have been proposed to assess weights of criteria (see, for example Bana e Costa and Vansnick, Bana e Costa and Vansnick, 1997, Jacquet-Lagr` eze and Siskos, 1982 and Mousseau et al., 2001). 10

C1

C2

C3

C4

C5

r g(1) r g(2)

Criteria

r g( j)

. . .

. . . b1 b2 b3 b4
r g( n)

Figure 2: Risk classes dened by the limit proles.

In addition to the dierent preference models of elicitation of weights, an important question in MCDA is the relative importance of criteria (see Mousseau, 1995 and Roy and Mousseau, 1996). Indeed, there is a dierent meaning between weights in compensatory methods, like the Multi-Attribute Utility Theory (MAUT) (Keeney and Raia, 1976) and non-compensatory methods like ELECTRE (Roy, 1991) and PROMETHEE (Brans et al., 1984). In compensatory methods, weights of criteria are essentially trade-os between criteria and are constants dependent on their scale and range. On the other hand, in non-compensatory methods the weights of criteria have an intrinsic 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 the cards 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 by using a set of cards (one for each criterion). The DM can also insert some blank cards to separate the relative importance between criteria. If the DM inserts no white cards between criteria, this means that these criteria have not the same weights and their relative dierence can be taken as unit measure u for weights. With a similar reasoning, if the DM inserts one white card, two white cards, etc, this means respectively a dierence 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 to obtain ordinal information on the weights of the criteria and then to obtain indirectly the weights of criteria from the DMs ranking. Since in many banks the granting of a loan depends on the decision of a pool of credit ocers (the DMs), in the present paper it seems more realistic to assume dierent orders of importance of weights of criteria depending on the DMs preference attitudes. In Section 4.4, the case studies will be carried by considering, ve hypothetical dierent DMs. For each DM, we apply SMAA-TRI by varying the -cutting level represented by a stochastic variable uniformly distributed. Then, to take into account simultaneously all the DMs sets of weights, the evaluation 11

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 again SMAA-TRI, computing the category acceptability indices for each alternative.

Start-up case studies

In this section, we carry out some real case studies, with the aim of illustrating the multicriteria credit risk model proposed. The authors have contacted four innovative SMEs headquartered in Italy requesting them for their business plan. Furthermore, a structured questionnaire was also sent to the companies. The data collected in the questionnaire concern some general information on the company, distribution and customer networks, demand forecasting, supply chain information, the owners company CV and the awards possibly 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 analysis simulates the nancing of such companies. Companies 1. Company A is a biotechnological start-up, operating in the eld of the green economy. It has developed some biological systems based on plants and micro-organisms forming some eco-friendly barriers, to prevent soil from the hydrological destruction and the environmental pollution. The services provided by it, consist of realizing such biological barriers, also giving consultancy; furthermore its services are oered at lower prices than the ones of the traditional techniques. 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 sensor network for water leakage detection, based on energy harvesting. In particular, its technology can be applied not only in the water distribution eld, but also in other energy applications, such as heat and gas, or environmental monitoring. Its main innovation is in the green technology FluE (Fluid Energy) that is based on some sensors auto-charged by the energy produced by the uid 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-lm deposition processes. Its mission is providing breakthrough technology, in order to increase the eciency of PVD (Physical Vapour Deposition) processes, today used to produce microchips, MEMS (Micro Electro-Mechanical Systems), solar cells and other hi-tech devices. 12

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 on nano-particles and provide technical assistance and advice to businesses willing to use these technologies. The company is mainly focused in producing nano-engineered epoxy resins used in the manufacture of sport equipment, for example sailing boats or kiteboards.

4.1

Financial analysis

The starting point of the nancial analysis, presented in this section, is the enterprises business plan, on which we have out-performed a scenario analysis. From the base-case scenario, we have developed two dierent worst scenarios considering the cash-ows lowered by 20% and 40%. Then, we have addressed our analysis along two directions. On one hand, we have computed some nancial ratios. Thus, only for descriptive purposes we have compared them with those of the enterprises from the same industrial/economic sector. On the other, we have evaluated the NPVs for each scenario. First of all, let us remark that the enterprises have considered dierent planning horizons; company D has presented a three years BP, company A a four year BP, company B has considered a ve years BP and company C a six years BP. To the aim of detecting the Italian rms in the same sector of the four enterprises under evaluation, we have considered their ATECO codes, elaborated by ISTAT (www.istat.it), which identify their enterprises industrial/economic sectors. From AIDA dataset, we have reported dierent nancial ratios of the balance-sheets (from 2007 to 2011) of the sectors of companies A, B, C and D respectively, composed of 488, 13806, 946, and 869 SMEs (enterprises with a number of employees lower than 250). On the basis of the available nancial data, the following nancial ratios have been selected: ROA, Cash/Operating Revenue, EBIDTA/Total Asset, Short-term debt/Equity, Equity/Total Asset and Cash/Total Asset reecting the areas of protability, leverage and liquidity of each enterprise (some of these ratios have been already selected to evaluate the US SMEs in the paper of Altman and Sabato, 2007). Moreover, with respect to each enterprise the averages of each nancial ratio have been computed for the years from 2007 to 2011. Finally, we have evaluated the median value of each ratio for all the 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 nancial 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 dierent 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

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 the projects are protable.

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 been natural to adopt a binary codication, for all the others, for simplicity, we have used the same ordinal scale on a ve point scale. Let also assume that all the criteria evaluations are increasing, i.e. the more the better and consider ve dierent credit ocers (DMs) evaluating the SMEs creditworthiness. The family of criteria have been elaborated by the credit ocers with the help of the experts. Even more on the basis of the BP presented by the rms, the criteria evaluations have been given by the credit ocers in cooperation with the experts that can detect the specic risks of each enterprise. The criteria considered in the case studies are described hereafter (see also Table 1 for an overview of the criteria). Criteria related to the development risk.
1 ] The medium age of the management team. Recent literature has shown the advantages of in[g(1) 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 is considered innovative. For this reason, the following scale has been adopted: <40 medium age (1) or 40 medium age (0).

All the enterprises have a medium age of the management team less than 40 years old. In two cases (companies A and C) the medium age is lower than 30 years old.
1 [g(2) ] Awards: codes yes (1) or no (0).

The idea of company A has been certied by dierent awards. Among others the D2T Start Cup given by Trentino Sviluppo and it has also been classied rst in the competition AGRIstart Up. Company B has obtained the following awards: research grant Working Capital 2011, seed fund eCapital 2011 and ItaliaCamp 2012. Even company C has received dierent awards: both Italian and international. Among others company C has obtained the third place in the Innovact Campus Award 2012 in Reims (France). 14

Company D won the Business Plan Competition eCapital 2010, competition aimed at creating innovative startups. In 2011, it was nalist at the Italian National Innovation Award PNI Telecom Working Capital.
1 [g(3) ] Scientic skills: codes yes (1) or no (0).

Company A. The three principal partners are agronomists and are specialized in microbiology and in plant genetics. Company B is composed of three young engineers, a full professor, a strategic marketing expert and ArieLab s.r.l, a spin-o of the Universit` a Politecnica delle Marche. Company C. The management team is composed of a well qualied sta. For example, the CEO has a Master degree in Engineering, or the CTO has a Master degree in Physics and NanoScience. Company D. All the ve members are well qualied; some of them have a degree in engineering, others are specialized in nanotechnology and one of them is specialized in sailing prototypes that are the main products in which the company is investing its research and sale activities.
1 [g(4) ] Honeymoon period : codes yes (1) or no (0).

Many research papers have analyzed that the risk of failure is highest when an enterprise is a 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 signicantly lowered. Consequently, following the same reasoning we have included as an indicator of risk to detect if the innovative enterprise is within such period. All the four enterprises under consideration in the present study are within the honeymoon period. Company A has started in 2012, while companies B, C and D have been founded in 2011. Criteria related to the technological risk.
2 [g(5) ] Knowledge of the existing technologies: codes yes (1) or no (0). An innovative production technique 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 the enterprises demonstrate to know the other existing techniques. 2 [g(6) ] Upgrade capability: codes yes (1) or no (0). From their BP, all the enterprises show many good capabilities in upgrading their techniques.

15

2 [g(7) ] Pros of the techniques used in comparison to other similar already existing: a ve-points scale is adopted; one point means that the advantages are irrelevant, while ve point levels are given if they are signicant.

Company A has received three points, since there are other methods that can be used to prevent soil from the hydrological destruction and the environmental pollution. On the basis of this criterion, company B has received ve points since the only similar existing technique consists of putting a few Data Loggers in some points of the water pipelines to register the noise by a microphone. Anyway, such products are too expensive and detect the soundness of the water distribution una tantum. On the basis of this criterion, company C has received ve points since other existing techniques need too complex chemical processes to recover precious materials. Company D has received three point levels, since the retail prices of their epoxy resins are too high in comparison to other existing ones. Criteria related to the market risk.
3 [g(8) ] Presence of key competitors: a ve-points scale is adopted, ranging from one point, if there is a monopolist competitor in the market, to ve points, if the innovative enterprise is the only one competitor present in the market.

Company A has one competitor, that adopts a methodology much more expensive and with many dierent 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 four points. Company B doesnt have competitors in the market. The only similar solution, i.e. the Data Logger has many drawbacks. The DMs suggest an evaluation of ve points on this criterion. Even if company C has registered a patent, this doesnt represent a barrier to avoid that other enterprises could enter the market, by slightly modifying the system developed by company C. Some possible similar solutions could be: vacuum ad hoc chambers, static monitors and mechanical removing. Anyway, the above techniques need the support of complex chemical processes. On the basis of such criterion, The DMs suggest an evaluation of two points. Company D has dierent competitors in nanotechnology even if operating with dierent prices and quality in the their products. The DMs give an evaluation of two points on the basis of this criterion.
3 ] Potential market: a ve-points scale is adopted; a one point means that the market is reducing [g(9) while 5 means that the market is booming.

16

Company As potential market is limited by the traditional techniques, mainly based on chemical products, preferred to the innovative solution oered by Company A. In fact, there is a cultural resistance to accept this new type of methodology to prevent soil from the hydrological destruction and the environmental pollution, even if nowadays there is much more attention to environmental problems. On the basis of this criterion, a prudential evaluation of three has been considered. With respect to this point, the DMs believe that the Company Bs market is reducing (1 point). More precisely, even if the private and public companies running the water distribution in Italy need a system to monitor the water leakage, they are not nancially sound. Company C. The market of thin lm deposition is booming, with an estimation of market value 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 an evaluation of ve points. Company D. Within the nanotechnology market nanomaterials are a booming business with benets for all other sectors. Carbon nanotubes are in this case, together with some materials connected to them (e.g. graphene and fullerenes), the most important products of the entire sector. On this criterion, Company D has received an evaluation of ve points.
3 ] Dependence on suppliers: a ve-points scale is adopted. If the evaluation is one, this indicates [g(10) a high supplier dependence, while the evaluation of ve 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. Other suppliers can be found abroad, but they have not a sucient know-how to consider the specic plants used by Company A. For this reason, an evaluation of two points has been given to such criterion. The dependence from the suppliers is very high (1) for the Company B. In fact, to produce the sensors with the technique FluE some piezoelectric bers are necessary. Presently, only two companies abroad are available to supply such components. The Company Cs supplier dependence is very high, since a complex supplier network is needed; specically, 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 for its formulations raw materials purchased externally, but aims to internalize in the shortest possible time the production of the key ingredients: the nanocarbons. The DMs give an evaluation of one point on this criterion.

17

Criteria related to the production risk.

4 [g(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 the potential 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 rst prototype and tested it in the vacuum chamber of a thermal evaporator at the laboratory of CNR. The company Ds has started a pilot plant for the supplement of nanocarbons ller to thermosetting epoxy resins.
4 [g(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 of metals from thin-lm deposition facilities and also of an International application entitled: Palette modular device for collection of metals in this lm deposition equipment. The company Ds hasnt yet applied for a patent, but it aims to record strategic patents on its core product and a series of multiple patents that go to cover similar areas of research and development thus protecting future market segments. Innovation indicators
5 ] Intangible Assets/Fixed Assets; [g(13) 5 [g(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 proles

The limit proles are built by the DMs who determine the performance on each criterion with respect to every category. Table 3 shows the 4 limit proles for the ve under consideration categories. For simplicity, the preference and indierence thresholds have been set equal to zero.

18

Table 1: Family of criteria

G1 : 1 g(1) : 1 g(2) : 1 g(3) : 1 g(4) : G2 : 2 g(5) :
development risk Codes medium age awards scientic skills honeymoon period technological risk knowledge of the existing technologies 2 g(6) : upgrade capability

> 40(0), 40 (1) no (0), yes (1) no (0), yes (1) no (0), yes (1)
Codes

no (0), yes (1) no (0), yes (1) irrelevant (1) weakly signicant (2) signicant (3) strongly signicant (4) very signicant (5)
Codes

2 g(7) : pros of the technique

G3 : market risk

3 g(8) : key competitors

3 g(9) : potential market

3 g(10) : dependence on

suppliers

monopolist (1) numerous competitors (2) few competitors (3) one competitor (4) start-up (5) reducing (1) static (2) weakly rising (3) rising (4) booming (5) very high (1) high (2) medium (3) low (4) very low(5)
Codes

G4 : production risk 4 g(11) : availability of testing

and unit pilots 4 g(12) : owner of a patent G5 : innovation indicators
5 g(13) : intangible assets/xed assets 5 g(14) : R&D /sales

no (0), yes (1) no (0), yes (1) unit percentage percentage

4.4

Phase (iii): eliciting the set of weights

Let us assume a total preorder , i.e., a reexive binary relation on the set of criteria G satisfying two additional properties: transitivity and completeness. In this case, the preference relation can be decomposed into its symmetric part , called indierence, and into its asymmetric part , called strict preference, whose semantics are, in a multicriteria context, respectively: gj1 1 gj2 2 gj1 1 is equally important to gj2 2 , gj1 1 gj2 2 gj1 1 is less important than gj2 2 , 19
(r ) (r ) (r ) (r ) (r ) (r ) (r ) (r )

Company A B C D

1 g(1)

1 g(2)

1 g(3)

1 g(4)

Table 2: Evaluation matrix.

2 g(5) 2 g(6) 2 g(7) 3 g(8) 3 g(9)

3 g(10)

4 g(11)

4 g(12)

5 g(13)

5 g(14)

1 1 1 1

1 1 1 1

1 1 1 1

1 1 1 1

1 1 1 1

1 1 1 1

3 5 5 3

4 5 2 2

3 1 5 5

2 1 1 1

0 0 1 1

0 0 1 0

0.55 0.72 0.18 0.06

0.06 0.17 0.05 0.14

1 g(1)

1 g(2)

1 g(3)

1 g(4)

2 g(5)

Table 3: Limit proles. 2 2 3 3 3 g(6) g(7) g(8) g(9) g(10) 0 0 1 1 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4

4 g(11)

4 g(12)

5 g(13)

5 g(14)

b1 b2 b3 b4

0 0 1 1
(r )

0 0 1 1
(r )

0 0 1 1

0 0 1 1

0 0 1 1

0 0 0 1

0 0 1 1

0.2 0.4 0.6 0.9

0.03 0.05 0.07 0.1

where gj1 1 and gj2 2 are two distinct criteria from the set G. In the following, we will play the role of the DM (credit ocer), by assuming ve DMs preference attitudes. For the rst DM (DM1, say), the most important criterion assumed in the evaluation of an innovative enterprise is the existence of a unit pilot and the second most important criterion is the patent. DM1 gives this preference information on the criteria:
1 1 1 1 5 5 3 3 3 2 2 2 4 4 g(1) g(4) g(2) g(3) g(13) g(14) g(8) g(9) g(10) g(5) g(6) g(7) g(12) g(11) .

The second DM (DM2, say) expresses these preference statements:

1 1 4 5 5 1 1 2 2 2 4 3 3 3 g(1) g(4) g(12) g(13) g(14) g(2) g(3) g(5) g(6) g(7) g(11) g(10) g(8) g(9) .

In this case, the prevailing criteria in evaluating an innovative enterprise are the potential market and the presence of key competitors, i.e. the sub-criteria relative to the risk market. The DM3s preference information is the following:
1 1 3 3 3 5 5 4 4 1 1 2 2 2 g(1) g(4) g(10) g(8) g(9) g(13) g(14) g(12) g(11) g(2) g(3) g(5) g(6) g(7) .

In this case, the most important weights considered by the DM3 are the sub-criteria relative to the technological risk. 20

The DM4s preference information is given by:

1 1 5 5 1 1 3 3 3 2 2 2 3 4 g(1) g(4) g(13) g(14) g(2) g(3) g(8) g(10) g(11) g(5) g(6) g(7) g(9) g(12) .

In this case, the most important criterion is assumed to be the patents ownership and the second 1 1 most important one is the potential market. In all the four DMs, the criteria in the pairs (g(1) , g(4) ), 1 1 2 2 5 5 (g(2) , g(3) ), (g(6) , g(7) ) and (g(13) , g(14) ) are considered equally important, while the least important 1 1 criteria are the medium age (g(1) ) and the honeymoon period (g(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.
DM1 DM2 DM3 DM4 DM5
1 g(1) 0.021 0.018 0.017 0.020 0.071 1 g(2) 0.034 0.052 0.107 0.044 0.071 1 g(3) 0.034 0.052 0.107 0.044 0.071 1 g(4) 0.021 0.018 0.017 0.020 0.071 2 g(5) 0.071 0.059 0.116 0.093 0.071 2 g(6) 0.071 0.059 0.116 0.093 0.071 2 g(7) 0.149 0.080 0.125 0.134 0.071 3 g(8) 0.059 0.143 0.044 0.060 0.071 3 g(9) 0.059 0.148 0.044 0.142 0.071 3 g(10) 0.059 0.134 0.035 0.060 0.071 4 g(11) 0.171 0.127 0.077 0.068 0.071 4 g(12) 0.159 0.032 0.071 0.150 0.071 5 g(13) 0.046 0.039 0.062 0.036 0.071 5 g(14) 0.046 0.039 0.062 0.036 0.071

4.5

Computing the category acceptability indices

In this section, we have computed the category acceptability indices, according to the dierent DMs, 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 paper by 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 reecting 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 an histogram).

21

Table 5: Category acceptability indices according to the dierent DMs.

(a) DM1 (b) DM2

Company A B C D Company A B C D

C1 0% 0% 0% 0% C1 0% 0% 0% 0%

C2 0% 0% 0% 0% C2 0% 0% 0% 0%

C3 C4 28% 72% 51% 49% 0% 1% 61% 5% C3 0% 0% 0% 13%

C5 0% 0% 99% 34%

Company A B C D Company A B C D
(e) DM5

C1 C2 0% 0% 0% 53% 0% 0% 0% 0% C1 C2 0% 0% 0% 21% 0% 0% 0% 0% C5 35% 41% 81% 50%

C3 C4 7% 93% 12% 35% 58% 5% 72% 28% C3 C4 25% 75% 79% 0% 0% 0% 56% 4%

C5 0% 0% 37% 0% C5 0% 0% 100% 40%

(c) DM3

(d) DM4

C4 C5 40% 60% 35% 65% 0% 100% 8% 79% C1 0% 0% 0% 0% C2 0% 0% 0% 0%

Company A B C D

C3 C4 0% 65% 25% 34% 10% 9% 41% 9%

Table 6: Results of SMAA-TRI by considering all the DMs.

(a) Interval weights.
1 1 1 1 2 2 2 3 3 3 4 4 5 5 g(1) g(2) g(3) g(4) g(5) g(6) g(7) g(8) g(9) g(10) g(11) g(12) g(13) g(14) [0.02, 0.07] [0.03,0.11] [0.03,0.11] [0.02,0.07] [0.06,0.12] [0.06,0.12] [0.07,0.15] [0.04,0.14] [0.04,0.15] [0.03,0.13] [0.07,0.17] [0.03,0.16] [0.04, 0.07] [0.04,0.07]

(b) Category acceptability indices.

Company A B C D

Class 1 0% 0% 0% 0%

Class 2 Class 3 Class 4 Class 5 0% 8% 66% 27% 1% 21% 47% 31% 0% 13% 6% 81% 0% 46% 7% 47%

22

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

1.4 Category 1 Category 2 Category 3 Category 4 Category 5

1.2

Category accepatbilities indices

0.8

0.6

0.4

0.2

Company A

Company B

Company C

Company D

Discussion and managerial implications

In the previous section, all the simulations in SMAA-TRI have been out-performed by varying in the range [0.65, 0.8] to analyze the robustness of ELECTRE-TRI. We have reported in Table 7 the highest category acceptability for each DM obtained during the simulations. One can notice that the assignment of company C to the category C5 is very stable; also the assignment of Company A to the class C4 is quite stable. Some incompatibilities among the DMs arise for the enterprises B 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, but if the DMs try to reach a consensus, they dont manage to give a stable assignment to Company B. At this point, it could be useful to consider the nancial analysis performed in Section 4.1. On the basis of the comparison with the enterprises of the same sector, while the project of Company D is protable, Company B is not quite sound from a nancial point of view. The evaluation of the nancial soundness, as outlined in Section 4.1, could help the DMs in making the appropriate choice in the assignment of an enterprise to a risk class. Consequently, from this last enterprise one can observe that the judgemental rating model proposed suers of some limitations that can arise when the DMs dont agree in an unique system of weights. The last step of our analysis consists of xing the cutting level . As a result, SMAA-TRI coincides with ELECTRE-TRI, since for each DM the preference parameters are xed. 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

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 DMs A 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], we have decided to choose the extremes of the interval, i.e. = 0.6 and = 0.85. In this case, the results 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 C5 B C4 C4 C5 C3 C5 C C5 C5 C5 C5 C5 D 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 C4 B C4 C2 C4 C2 C3 C C4 C3 C5 C5 C3 D 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 three of the ve DMs that agree on the assignment of the same risk class. In Table 9, the same situation can 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 have a considerable impact on the decision procedure. In fact, let us suppose for example to consider 4 the criterion availability of testing and unit pilots (g(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 a 0 point in the evaluation according to this criterion), the criterion would be against the assignment 24

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 the enterprise 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.

Conclusions

Many recent studies have been dedicated to measuring SMEs innovation considered as a crucial factor for the development of a national economy (Brandolini and Bugamelli, 2009). To achieve a high level of innovation the main obstacle that the innovative SMEs face is a non straightforward access 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, nancial institutions could try to reduce such informative asymmetries. Presently, banks assess their credit risk by evaluating the business plan and considering some non-nancial information such as the age of the entrepreneurs, the market trend and the quality of the management team. Although nancing innovation isnt an easy task due to the lack of track records of innovative SMEs, we believe that a rating model based upon experts judgments could improve it. In nancing innovative SMEs, the role of the experts is crucial, since they can help the credit ocers in selecting the proper criteria, especially, in detecting their risks. In this paper, we have presented a multicriteria approach to sort innovative enterprises into risk classes, mainly on the basis of soft information. Specically, we describe a possible hierarchical structure of the innovation risks: development, production, market and technological ones. Such risk indicators have been considered together with two innovation indicators, mostly used in the literature that is intangible assets divided to xed assets and R & D expenditure divided by sales. The multicriteria approach proposed is ELECTRETRI based on an outranking preference relation comparing each innovation to some existing risk proles. As explained in the paper, the multidimensional and complex decisional framework of nancing innovation is well adapted to the aforementioned multicriteria method. In fact, the credit ocers with the help of the experts could dene the risk proles, the criteria weights, but could also tune some specic parameters such as the cutting level, the preference, indierence and veto thresholds on which the nal 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 making process. What is meant by the term critical here is the smallest change that might occur to a certain criterion in order increase the category acceptability of the better risk class to which each enterprise has been assigned. 25

 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 integral preference model among the multicriteria approaches representing interaction between criteria e.g. in Angilella et al., 2004 and its recent extension within a SMAA methodology in Angilella et al., 2012). Since the criteria adopted in this paper are hierarchally structured, it could be useful to apply the recent approach of the hierarchal Choquet integral presented in Angilella et al., 2013.

References
E.I. Altman and G. Sabato. Eect of the new capital accord on bank capital requirements for SMEs. 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-nancial information in small and medium-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 multicriteria decision aid. European Journal of Operational Research, 158(3):734744, 2004. S. Angilella, S. Corrente, and S. Greco. SMAA-Choquet: Stochastic multicriteria acceptability analysis for the Choquet integral. In S. Greco et al., editor, Advances in Computational Intelligence, volume 300 of Communications in Computer and Information Science, pages 248257. Springer Berlin Heidelberg, 2012. S. Angilella, S. Corrente, S. Greco, and R. Slowi nski. Multiple criteria hierarchy process for the Choquet integral. In R.C. Purshouse et al., editor, EMO 2013, volume 7811 of LNCS, pages 475489. Springer Berlin Heidelberg, 2013. H.V. Auken, G.H. C anovas, and A.M. Guijarro. Role of information in the debt nancing of technology-based rms in Spain. In The Entrepreneurial Society: How to Fill the Gap Between Knowledge and Innovation. Edward Elgar Publishing Inc., 2010. C.A. Bana e Costa and J.C. Vansnick. A theoretical framework for measuring attractiveness by a categorical based evaluation technique (MACBETH). In Proc. XIth Int. Conf. on Multicriteria Decision Making, 15-24, Coimbra, Portugal, August 1994.

26

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. Demirg ur c-Kunt. Small and medium-size enterprises: Access to nance as a growth constraint. Journal of Banking & Finance, 30(11):2931 2943, 2006. T. Beck, A. Demirg ur c-Kunt, and M.S. Mart nez Per a. Bank nancing for SMEs: Evidence across countries 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 nance. Journal of Banking & 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 organizational form? evidence from the lending practices of large and small banks. Journal of Financial Economics, 76(2):237 269, 2005. A. Brandolini and M. Bugamelli. Report on trends in the Italian productive system. Questioni di Economia e Finanza (Occasional Papers) 45, Bank of Italy, Economic Research and International Relations 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: A new family of outranking methods in multicriteria analysis, pages 408421. Elsevier Science Publishers B.V. (North-Holland), 1984. R.J. Brown, S.M. Fazzari, and C.B. Petersen. Financing innovation and growth: Cash ow, external equity, 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 of Italy, Economic Research and International Relations Area, April 2012.

27

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 nancing constraints of small and mediumsized rms. Small Business Economics, 36:6583, 2011. S. Damart, L. Dias, and V. Mousseau. Supporting groups in sorting decisions: Methodology and use of 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. Decision Support Systems, 50(1):5563, 2010. M. Doumpos, K. Kosmidou, G. Baourakis, and C. Zopounidis. Credit risk assessment using a multicriteria hierarchical discrimination approach: a comparative analysis. European Journal of Operational Research, 138(2):392 412, 2002. J. Figueira, S. Greco, B. Roy, and R. Slowi nski. Handbook of Multicriteria Analysis, chapter ELECTRE Methods: main features and recent developments. Springer, New York, 2010. S. Greco, B. Matarazzo, and R. Slowi nski. Comments on: Multicriteria decision systems for nancial problems. TOP, pages 17, 2013. J. Grunert, L. Norden, and M. Weber. The role of non-nancial 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 Business Finance & Accounting, 14(2):199213, 1987. E. Jacquet-Lagr` eze. 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 in Multicriteria Analysis, volume 5 of Nonconvex Optimization and Its Applications, pages 203211. Springer US, 1995. E. Jacquet-Lagr` eze and J. Siskos. Assessing a set of additive utility functions for multicriteria decision-making, the UTA method. European Journal of Operational Research, 10(2):151164, 1982. R.L. Keeney and H. Raia. Decisions with multiple objectives: Preferences and value tradeos. J. Wiley, New York, 1976.

28

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. Mazz` u. Il nanziamento dellinnovazione. Strumenti, rischi e modelli di valutazione. G. Giappichelli, 2008. M. Merad, T. Verdel, B. Roy, and S. Kouniali. Use of multi-criteria decision-aids for risk zoning and management of large area subjected to mining-indiced hazards. Tunnelling and Underground Space Technology, 19:125 138, 2004. D.C. Morais, A.T. de Almeida, and J.R. Figueira. A sorting model for group decision making: A case 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, pages 1743. Kluwer Academic Publishers, Dordrecht, 1995. V. Mousseau, J. Figueira, and J.P. Naux. Using assignment examples to infer weights for ELECTRETRI 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 of innovation measures. Journal of Financial Services Research, pages 119, 2012. B. Roy. Classement et choix en pr esence de points de vue multiples: La m ethode ELECTRE. Revue Francaise dInformatique et de Recherche Op erationnelle, 8:5775, 1968. B. Roy. M ethodologie Multicrit` ere dAide ` a la D ecision. 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

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, rm size and innovation: an empirical analysis. Technovation, 25(1):25 32, 2005. T. Tervonen. JSMAA: open source software for SMAA computations. International Journal of Systems Science, pages 113, 2012. T. Tervonen, R. Lahdelma, J. Almeida Dias, J.R. Figueira, and P. Salminen. SMAA-TRI. In I. Linkov, G.A. Kiker, and R.J. Wenning, editors, Environmental Security in Harbors and Coastal Areas, 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 method for 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 SMEs performance via multicriteria analysis of nancial ratios. Small Business Economics, 15(2):127 136, 2000. W. Yu. ELECTRE TRI: Aspects m` ethodologiques et manuel dutilisation. Document du LAMSADE No.74, Universit e Paris-Dauphine, June 1992. C. Zopounidis and M. Doumpos. Multicriteria classication and sorting methods: A literature review. 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

Appendix A
Table 10: Comparisons with the sector.
(a) Company As nancial analysis.
(b) Company Bs nancial analysis.

Ratios ROA Cash-Flow/Op. revenue EBITDA/Total Asset Short term debt/Equity Equity/Total Asset Cash/Total Asset

Average Median 2012-2015 sector 0.24 0.03 -0.03 0.04 0.26 0.04 0.18 1.73 0.87 0.17 0.74 0.08

Ratios ROA Cash-Flow/Op. revenue EBITDA/Total Asset Short term debt/Equity Equity/Total Asset Cash/Total Asset

Average Median 2012-2015 sector 0.03 0.05 0.02 0.05 0.14 0.07 0.12 1.36 0.58 0.24 0.51 0.07

(c) Company Cs nancial analysis.

(d) Company Ds nancial analysis.

Ratios ROA Cash-Flow/Operating revenue EBITDA/Total Asset Short term debt/Equity Equity/Total Asset Cash/Total Asset

Average Median 2012-2015 sector 0.94 0.08 0.24 0.05 0.34 0.08 0.30 1.23 0.46 0.26 0.56 0.06

Ratios ROA Cash-Flow/Operating revenue EBITDA/Total Asset Short term debt/Equity Equity/Total Asset Cash/Total Asset

Average Median 2012-2015 sector 0.52 0.01 0.33 0.08 0.6 0.01 -0.26 0.72 0.36 0.39 0.26 0.09

31

Table 11: Scenario analysis on the Company As cash-ows. First Scenario year 1 year 2 year 3 year 4 NPV (7.93%) cash-ow (e) -43,534.00 69,616.91 9,178.96 118,470.63 140,275.51 Second Scenario year 1 year 2 year 3 year 4 NPV (7.93%) cash-ow (e) -52,240.80 55,693.53 7,343.17 94,776.50 75,092.75 Third Scenario year 1 year 2 year 3 year 4 NPV (7.93%) cash-ow (e) -60,947.60 41,770.15 5,507.37 71,082.38 36,151.86

Table 12: Scenario analysis on the Company Bs First Scenario year 1 year 2 year 3 year 4 cash-ow (e) - 8,715.00 -15,528.00 52,196.00 58,422.00 Second Scenario year 1 year 2 year 3 year 4 cash-ow (e) - 10,458.00 - 18,633.60 41,756.80 46,737.60 Third Scenario year 1 year 2 year 3 year 4 cash-ow (e) - 12,201.00 - 21,739.20 31,317.60 35,053.20

cash-ows. year 5 NPV (7.93%) 57,472.00 102,405.74 year 5 NPV (7.93%) 45,977.60 73,362.71 year 5 NPV (7.93%) 34,483.20 44,319.68

Table 13: Scenario analysis on the Company Cs cash-ows.

First Scenario year 1 year 2 year 3 year 4 year 5 year 6 NPV (7.93%) -211,100.00 126,543.19 196,034.36 233,763.68 438,942.51 568,339.16 1,032,614.23 Second Scenario year 1 year 2 year 3 year 4 year 5 year 6 NPV (7.93%) -253,320.00 101,234.55 156,827.48 187,010.95 351,154.01 454,671.33 642,152.30 Third Scenario year 1 year 2 year 3 year 4 year 5 year 6 NPV (7.93%) -295,540.00 75,925.91 117,620.61 140,258.21 263,365.51 341,003.49 383,819.35

cash-ow (e)

cash-ow (e)

cash-ow (e)

32

Table 14: Scenario analysis on the Company Ds cash-ows. First Scenario year 1 year 2 year 3 NPV (7.93%) cash-ow (e) -62,272.07 204,057.11 1,094,740.87 988,208.95 Second Scenario year 1 year 2 year 3 NPV (7.93%) cash-ow (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-ow (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 dened: z is the ratio expressing how many times the last criterion is more important than the rst one in the ranking; er is the number of white cards between the rank r and r + 1; er = er + 1; u= e=
r =1 z 1 ; e n 1

er .

The non normalized weight k (r ) is computed by: k (r ) = 1 + u(e0 + + er1 ), with e0 = 0. Let
ki n ki the sum of the i=1

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

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

1 k. K i Within the Simos procedure, the preference information expressed by the DM1 is reported in Table 15(a). Thus, the normalized weights are those displayed in Table 15(b). 33

Table 15: Simos method

(a) DM1s set of weights with z = 8.

Rank r 1 2 3 4 5 6 7 8 Total
1 g(1) 1 g(2)

Criteria n. of white in the rank r cards 1 1 {g(1) , g(4) } 0 1 1 {g(2) , g(3) } 0 5 1 {g(13) , g(14) } 0 3 3 3 {g(8) , g(9) , g(10) } 0 2 2 {g(5) , g(6) } 5 2 {g(7) } 0 4 {g(12) } 0 4 {g(11) } e
1 g(3) 1 g(4)

Non-normalized weights k (r ) 1 1.0 1 1.6 1 2.2 1 2.8 6 3.3 1 6.8 1 7.4 8.0 12 K er

Total 2.0 3.2 4.3 8.3 6.7 6.8 7.4 8.0 46.7

ki

(b) Normalized weights for the DM1. 2 2 2 3 3 3 4 4 5 5 g(5) g(6) g(7) g(8) g(9) g(10) g(11) g(12) g(13) g(14) 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