Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
PREDICTIVE MODELS OF DEFAULT RISK IN LOCAL GOVERNMENTS. A LOGISTIC REGRESSION Pina Puntillo Researcher in Business Economics University of Calabria Department of Business Science Phone +39984492247 Fax +39984492277 E mail [email protected] PREDICTIVE MODELS OF DEFAULT RISK IN LOCAL GOVERNMENTS. A LOGISTIC REGRESSION Key words – Financial risk, local government, logistic regression Abstract With the activation of fiscal federalism in Italy it has become an urgent matter to assess the risks of local authority insolvency and to find the appropriate tools to do this. In other word the fiscal federalism has led to a growing interest in local government bankruptcy prediction models. In this context the present research proposes to model the risk of local authorities’ insolvency on the basis of accounting variables. The purpose of this research is to test the usefulness of financial data for predicting local government failure in Italy. After a brief literature review on predictive model of failure the study will then go on to focus on the main drivers of fiscal risk and their effects in accounting terms. The aim of this phase of the research is to identify the variables to be then used as explanatory variables in the quantitative models for estimating or predicting insolvency. The predictive models are an application of regression models that assess the likelihood of the occurrence of an event. From the various models in the literature identifying financial crises, type of regression model with a binary dependent variable: it includes the logit and probit regression model. These are non linear models that use standard or logistic distribution functions. The explanatory variables of the model are the accounting variables identified in the first phase of the work. Once constructed on a theoretical basis, the model will then be tested empirically on a sample of local bodies in Italy.. 2 PREDICTIVE MODELS OF DEFAULT RISK IN LOCAL GOVERNMENTS. A LOGISTIC REGRESSION Summary of contents Introduction; 1. Financial Distress in Italian Local Governments; 2. Literature Review; 3 Research Methodology; 3.1 The explanatory variables; 3.2 Dataset;3.3 Hypothesis; 3.4 Logit regression model; 3.5 Probit regression model; 4 Results; Conclusions. 1. Financial distress in Italian local governments The reforms In Italy in the 1990s, culminating in the modification of Section V of the Constitution law, changed the economic and financial structure of local government by affirming the principle of financial autonomy for all levels of government to guarantee cover for the public functions carried out by local bodies following the acceptance of the principle of subsidiarity. Thus, the system of public finance evolved towards a decentralised system characterised by the progressive reduction of government transfers, by the replacement of the principle of past spending with a criterion based on expected spending required to cover essential levels of assistance and services (Borgonovi, 2009: p. 15; Paoloni et al., 2007: pp. 577-615). In this new system internal resources and the ability to find other sources of finance through recourse to capital markets will become increasingly important. Financial autonomy, in fact, encourages local bodies to act responsibly with regard to the cost of decision-making and internal competition (Borgonovi, 2005: p. 162). In view of the foregoing, it is argued, therefore, monitoring of all financial aspects of the institution produces reflections becomes crucial for the survival of local authorities (Mele et al., 2005: p. 487); in other words, to avoid situations of financial distress of local authorities, it is necessary to develop a greater "risk sensitivity" (Speca, 2002; p.772-821), and identify an appropriate methodology for financial risk assessment. With the introduction of the fiscal federalism the policy makers have become increasingly concerned with the question of the ‘financial sustainability’ of individual local councils. According to Deal (2007) the financial health, the fiscal stress and municipal bankruptcy must be studied in order to avoid future filings of municipal bankruptcy as well as to improve the 3 financial health of local governments. Predictive models of risk of failure for local governments have been tested mainly in the United States and Australia. In Europe there is not currently great empirical literature, excluding research on rating models adopted by international agencies such as Moody's, Standard & Poor's, etc. These rating agencies stress the importance of data from company accounts in the evaluation of the overall credit worthiness of a business; furthermore, studies carried out on this question have shown the existence of a significant relationship between ratings and quantitative variables (Cannata, 2001; p. 37 onwards). the scenario likely to emerge from the full activation of fiscal federalism will require an ever more careful evaluation of the risk of insolvency and the use of reliable models to achieve this. Given the current state of public finances, quantitative predictive models for assessing the risk of insolvency of public bodies have come to play a key role. Only with the ability to identify quickly and accurately potential crisis situations, will it become possible to adopt effective corrective measures. Starting from these assumptions, therefore, the work will examine, from a theoretical perspective, predictive models of financial distress for local governments. In the second part of the work the model, developed on a theoretical level in the first part, will be tested empirically in order to demonstrate its validity. 2. Literature Review Over the years, failure prediction or financial distress models have been much discussed in accounting and credit management literature. From the late 1960s, when Beaver (1967) and Altman (1968) published their first failure prediction model, an enormous number of academic researchers from all over the world have been developing good failure prediction models based on various modelling techniques. There are numerous examples: Altman et al. (France, 1974), Taffler and Tisshaw (UK, 1977), Knight (Canada, 1979), Fischer (Germany, 1981), Ooghe and Verbaere (Belgium, 1982), Ko (Japan, 1982), Fernandez (Spain, 1988), Swanson and Tybout (Argentina, 1988). For a more complete list, see Zavgren (1983), Altman (1984), Taffler (1984), Jones (1987), 4 Keasey & Watson (1991), Ooghe et al. (1995), Dimitras et al. (1996), Altman and Narayanan (1997). Altman & Saunders (1998). Moreover in many papers some attention has been paid to the comparison of different scoring techniques (for example logit analysis, neural networks and decision trees) on the same dataset have been published, for example Bell et al. (1990), Altman et al. (1993), Curram and Mingers (1994), Joos, Ooghe and Sierens (1998), Kankaanpää and Laitinen (1999)., or to the performance of different types of failure prediction models (Mossman et al., 1998). However most of researches have focused, from both a theoretical and empirical point of view with reference to the private sector. This study aims to contribute, from an empirical point of view, in this field of research with reference to local governments. Beaver (1967a) was the pioneer in building a corporate failure prediction model with financial ratios. He was the first researcher to apply a univariate model – a “univariate Discriminant analysis model” – on a number of financial ratios of a paired sample of failing and non-failing companies in order to predict company failure. In his studies Beaver demonstrated the predictive ability of accounting data (Baever HW, 1966; Beaver HW et al., 1968). Univariate analysis is a very simple technique that classifies a company as healthy if the value assumed by an accounting index is above the critical value, also called the cut off point; on the contrary, if its value is below the cut off point the company is considered at risk (Lachenbruch, 1975). Discriminant analysis can also be expressed graphically: Fig.1 - Estimate of model through linear discriminant analysis Bankrupt healthy Source: Barontini (2000), The evaluation of credit risk, p. 67. 5 The cut off, in other words the critical point that determines the company’s classification, corresponds, therefore, to the equidistant value between the averages of the two groups. This methodology provides the same results as those obtained through a multiple linear regression, in which the dependent variable assumes a dichotomous value (1 in the case of a company crisis and 0 for the healthy company)1. The Univariate Discriminant analysis is appealing in his simplicity (for each ratio, one simply compares the ratio value for the firm with a cut-off point and decides on the classification accordingly), but his main disadvantage lies in his inability to account for the coexisting effects of many different indicators of default and provides largely arbitrary risk-metrics (Aaro Hazak, Kadri Männasoo 2007). Moreover the Univariate analysis is based on the stringent assumption that the functional form of the relationship between a measure or ratio and the failure status is linear. This assumption is often violated in practice, where many ratios show a non-linear relationship with the failure status (Keasey & Watson, 1991). Moreover, firm classification can only occur for one ratio at a time, which may give inconsistent and confusing classifications results for different ratios on the same firm (Altman, 1968; Zavgren, 1983). This problem is called the ‘inconsistency problem’. When using financial accounting ratios in a univariate model, it is difficult to assess the importance of any of the ratios in isolation, because most variables are highly correlated (Cybinski, 1998). In the same context, the univariate model contradicts with reality in that the financial status of a company is a complex, multidimensional concept, which can not be analysed by one single ratio. Finally, the optimal cut-off points for the variables are chosen by ‘trial and error’ and on an ‘ex post’ basis, which means that the actual failure status of the companies in the sample is known (Bilderbeek, 1973). Consequently, the cut-off points may be sample specific and it is possible that the classification accuracy of the univariate model is (much) lower when the model is used in a predictive context (i.e. ‘ex ante’). In response to Beaver, Tamari (1966) realized a ‘risk index’. It is 1 Fisher has demonstrated that there is a close analogy between discriminant analysis and linear regression, given that the coefficients estimated with the two methodologies are in perfect proportion; only the estimate of the constant differs. For more information on this matter see Maddala G. Limited Dependent and Qualitative Variables in Econometrics, Cambridge University Press, 1983. 6 a simple ‘point system’, which includes different ratios, generally accepted as measures of financial health. Each firm is attributed a certain number of points, between 0 and 100, according to the values of the ratios for the firm. A higher total of points indicates a better financial situation. This approach shares the same weaknesses of univariate analysis and provides largely arbitrary riskmetrics. In 1968, Altman (1968) introduced the Multiple Discriminant Analysis (MDA) which is “a statistical technique used to classify an observation into one of several a priori groups dependent upon the observation’s individual characteristics… attempts to derive a linear [or quadratic] combination of these characteristics which ‘best’ discriminates between the groups (Altman, 1968, p. 592)”. In his study he estimated a model called the ‘Z-score model’ used in an enormous volume of studies. After the 1980s, the MDA method is frequently used as a ‘baseline’ method for comparative studies (Altman & Narayanan, 1997). An MDA model consists of a linear combination of variables, which provides the best distinction between the group of failing and the group of non-failing firms. For example, Altman’s Z-score model is a linear combination of the following ratios: working capital / total assets, retained earnings / total assets, earnings before interest and taxes / total assets, market capitalization / total debts and sales / total assets (Altman, 1968). The linear discriminant function is the following (Lachenbruch, 1975): Di = D0 + D1Xi1 + D2Xi2 + … + DnXin (1) with Di = discriminant score for firm i (between -∞ and +∞), Xij = value of the attribute Xj (with j = 1, …, n) for firm i, Dj = linear discriminant coefficients with j = 0, 1,…, n. In an MDA model, several (mostly financial) characteristics or ‘attributes’ of a company are combined into one single multivariate discriminant score Di. This discriminant score is a one dimensional measure which has a value between -∞ and +∞ and gives an indication of the financial health of the firm. The score is seen as a risk measure. In most studies, a low discriminant score 7 indicates a poor financial health. In 1980s MDA method has been replaced by conditional probability models’ (Zavgren, 1983; Zavgren, 1985; Doumpos & Zopoudinis, 1999) as logit analysis (LA), probit analysis (PA) and linear probability modelling (LPM). Ohlson (1980) pioneered in using logit analysis on financial ratios in order to predict company failure while Zmijewski (1984) was the pioneer in applying probit analysis (PA). On the theoretical the logistic regression is a more appropriate instrument than linear regression, since it allows to define two distinct classes (good and bad risks) (Hand DJ and Henley WE, 1997, p. 533). In this study, it allows you to define two distinct classes: local authorities with financial risk and local authorities with no financial risk. In the author’s opinion, logit and probit models can be successfully employed in the analysis of the risk of insolvency of local authorities, because they allow researchers to estimate the probability that a crisis occurs, given the values of the account variables, that constitute the explanatory variables of the model, i.e. X. The econometric methodology of the logit analysis was also chosen to avoid some know problems associated with the Multivariate Discriminant Analysis (MDA). Some of the problems of this technique are: 1) there are some statistical requirements imposed on the properties of the distribution of predictors: for example, the variance-covariance matrix of the predictors should be the same for both groups (failed and no-failed firms); 2) the requirement of normally distributed predictors certainly reduces the use of binary independent variables (Eisenbeis, 1977; Joy and Tollefson, 1975). The MDA approach has long been the leading method for predictions of business failure. The main drawback is certainly the assumption of normally distributed regressors: since generally the financial ratios are not normally distributed, the maximum likelihood methods and in particular the logit and probit were the most frequently used (Martin, 1977; Demirgüc-Kunt, 1989; Lennox, 1999; Trussel J. M.. Patrick P. A, 2009). There are also some problems with the "matching procedures" that are generally used in the MDA: the failed and no-failed firms are matched based on criteria such as size and sector, and these tend to be somewhat arbitrary (Ohlson JA, 1980). The use of logit analysis, instead, avoids essentially all of the problems discussed in relation to MDA. 8 The fundamental problem of the estimate may be reduced simply to the following statement: “as the company belong to a prespecified population, what is the probability that the company failed within a prespecified period of time?” With logit analysis no assumptions should be made regarding the prior probability of failure and/or distribution of the predictors. These are the main advantages compared to discriminant analysis (Ohlson JA, 1980). Conditional probability models allow to estimate the probability of company failure conditional on a range of firm characteristics by a non-linear maximum likelihood estimation. The models are based on a certain assumption concerning the probability distribution. The logit models assume a logistic distribution (Maddala, 1977; Hosmer & Lemeshow, 1989), while the probit models assume a cumulative normal distribution (Theil, 1971). In the linear probability models, the relationship between the variables and the failure probability is assumed to be linear (Altman et al., 1981; Gloubos & Grammatikos, 1988). Both logit and probit models provide the probability of occurrence of an outcome described by a dichotomous (or polytomous) dependent variable using coefficients of the independent variables (Zavgren, 1985; Vranas, 1992). The main difference between logit and probit analysis is that the former uses the cumulative logistic probability function, while the latter is based on the cumulative standard normal distribution function. A significant advantage of these models over discriminant analysis is that they do not require the independent variables to be multivariate normal (Keasey, K. and Watson, R., 1991). Furthermore, they provide much more useful information to the financial/credit analysts in bankruptcy prediction, since except for the classification of the firms into bankrupt and non-bankrupt ones, they also provide the probability of failure of a firm. Besides the fact that logit analysis has no assumptions concerning the distribution of the independent variables and the prior probabilities of failure, there are some other important advantages of LA. First, the output of the LA model, the logit score, is a score between zero and one, which immediately gives the ‘failure probability’ of the company (Ohlson, 1980; Ooghe et al., 1993). Second, the estimated coefficients in a LA model can be interpreted separately as the importance or significance of each of the independent variables in the explanation of the estimated 9 failure probability (Ohlson, 1980; Mensah, 1984; Zavgren, 1985), provided that there is no multicollinearity among the variables. Third, LA models allow for qualitative variables with categories rather than continuous data. In this case, dummies are used (Ohlson, 1980; Keasey & Watson, 1987; Joos et al, 1998a). The formula of the linear probability model is that of the multiple linear regression model: Yi = β0 + β1X1i + β 2X2i +…. + β kXki + ui; where Yi is binary, so that Pr(Y = 1|X1;X2;…. ;Xk) = β0 + β1X1i + β 2X2i +…. + β kXki + ui The linear probability model can also be expressed graphically. Fig. 2 - Linear probability model Stock J. H., Watson M. W. (2005) Introduction to econometrics The linearity that ensures the linear model is easy to use can also be its major drawback, however. These models run into problems of inference, and the assumptions of normality/homoschedasticity of the errors are violated (i.e., the remainders are dichotomous and heteroschedastic). The probit regression model with multiple regressors takes the following form: Pr(Y = 1|X1,X2, ……Xk) = Ф (β0 + β1X1 + β2X2 +…. + βkXki) where Ф is the normal distribution function. 10 If β is positive then an increase in X increases the probability that Y = 1. If β is negative then an increase in X reduces the probability that Y = 1. The logit regression model with multiple regressors takes the following form: Pr(Y = 1|X1,X2, ……Xk) = F (β0 + β1X1 + β2X2 +…. + β kXki) = = 1/ 1+e -(β0 + β1X1 + β2X2 +…. + βkXki) Where F is the standard logistic distribution function. If β Is positive then an increase in X increases the probability that Y = 1. If βis negative then an increase in X reduces the probability that Y = 1. Logit and probit simulation often produce similar results2. Fig. 3 - Model Probit Logit Stock J. H., Watson M. W. (2005) Introduction to econometrics The three models (linear probability, logit and probit) are approximations of the unknown population regression function E( Y/X ) = Pr ( Y= 1/X ). 2 For more information on the econometric models used see Stock J.H., Watson M.W., 2005. 11 The linear model is probably the easiest to use and read, but fails to capture the nonlinear nature of the true regression function of the population. The logit and probit regressions model the nonlinearity in probability. The classic static-econometric methods can be considered the most commonly used methods for developing business models to forecast failure. In addition to these traditional statistical methods, academic researchers are beginning to use several alternative methods to analyze and predict business failure (Balcaen S. and Ooghe H., 2004). These methods are the result of strong technological progress and the so-called artificial intelligence (AI). The best-known alternative models that have produced a considerable number of studies on the prediction of business failure are the survival analysis, decision trees and neural networks (Balcaen S. and Ooghe H., 2004); in the following table are shortly mentioned the main alternative - even the lesser known - indicating the academic contributions on the subject of study developed in this research and summarize the salient characteristics of themselves: 12 Table 1 - Alternative methods applied for the prediction of business failure Methods Authors Description of the methods First-generation Merton (1974) This approach stems from the option pricing models Black et al. (1976) model originally developed by Black and Cox in Geske (1977) 1973, and that finds a first application to the risk Vasicek (1984) of insolvency with the work of Merton (1974).The Crouhy & Galai (1994) model is based on structural variables of the firm Jones, Mason & Rosenfeld in which the event of default resulting from the (1984) evolution of the assets of the company. According to the adaptation of Merton and subsequent authors, insolvency occurs when the value of the business is less than the value of liabilities. The corporate debt is modeled as a call option on assets with a strike price equal to liabilities, the option will be exercised until the value of the activity is highest in liabilities. Default occurs if the option is exercised before it expires. Second-generation Kim, Ramaswamy & This evolutionary approach simplifies the first models Sundaresan (1993) class of models both explicitly exogenously cash Nielsen, Saà-Requejo, Santa flows in the event of default that simplifying the Clara (1993) process of default. This happens when the value of Hull & White (1995) the assets of the company reaches a certain limit; Longstaff & Schuwarz (1995) what changes of the elements considered is the recovery rate which is exogenous and independent from the value of the business and therefore from the event of default. Despite these efforts in line with the Merton model, the second-generation models have several disadvantages which are the cause of the limited empirical application. Reduced form models Littermann & Iben (1991) While the structural models observe default as the Madan & Unal (1995) result of a gradual process of deterioration of asset Jarrow & Turnbull (1995) values, the reduced form models (also called Jarrow, Lando & Turnbull intensity-based models) represent the default as an (1997) unexpected event (sudden surprise). Lando (1998) These models are highly empirical, and do not Duffie & Singleton (1999) provide a stochastic process that generates the Duffie (1999) default, but they tend to decompose observed credit spreads on the debt to ensure both the probability of default, which is an unexpected event that the LGD (Loss Given Default), calculated as a complement to the recovery rate. Expert Systems Messier & Hansen (1988) Expert systems are the result of the strong Hawley et al. (1990) development of the artificial intelligence. The method of expert systems has been applied by Messier and Hansen (1988) for the prediction of failure. An expert system depends on the knowledge representation of experts on bankrupt, and such knowledge is transformed into a set of rules to be used for the prediction of business failure. Decision Trees Frydman et al. (1985) A decision tree is a classification tree that can help (or method of Joos et al. (1998) to identify, through an evaluation form, applicants recursive requesting a loan with a low or high risk of 13 partitioning) Survival analysis Lane et al. (1986) Luoma & Laitinen (1991) Kauffman & Wang (2001) Method of linear programming Gupta et al. (1990) Forward Looking Montesi and Papiro (2003) Catastrophes theory (or chaos theory) Scapens et al. (1981) Lindsay & Campbell (1996) insolvency and is built in order to support action over the granting of credit . Therefore, it can be used to assess the risk of business failure and the financial risk of local authorities. Survival analysis is a statistical methodology that allows us to study the risk that an individual has to live or not a certain event in a certain timeframe. In the case object of study, this method allows us to study the risk that a firm (or a local authority) has to fail (declare bankruptcy) in a certain period of time. In contrast to classical statistical models, a model of survival analysis not assume a dichotomous dependent variable (dummy). The method of linear programming (LGP) is one of several techniques derived from mathematical programming. The model formulates intragroup and intergroup differences between risky business and not and, on the basis of these differences, it calculates a score for each firm and a limit (cutoff) for the discrimination group. This creates a hyper-plane, which is used to distinguish between the group of risky business and the group of healthy companies. The cut-off or limit is determined (1) maximizing the weighted sum of the distances between the observations and the cut-off and (2) minimizing the weighted sum of violations of the cut-off limit. The forward looking, proposed by Montesi and Papiro (2003) is another method used to calculate the probability of default. The logic of this method it not proposed to accurately predict the value that may take some variable in the future, but to estimate what may be the true value within a range of possible values according to a probability distribution. For each test is generated a scenery of the company, which includes the development of a complete budget estimate for each forecast period and for each period; is determined then the margin of solvency: it reconstructs the evolution of solvency requirements in relation to uncertainty assumed. It is estimated that the financial fragility of the company against future unforeseen events. Scapens et al. (1981) were the first researchers who considered the business failure as a catastrophic event and who have used catastrophes theory to explain business failure. The theory implicitly assumes that firms are deterministic and predictable, but only in short periods of time. A second assumption of the model is that the healthy and no risky firms are more inclined to fail than unhealthy and risky firms. It is clear that a model of chaos theory requires an appropriate measure of chaos. Lindsay & Campbell (1996) measured the amount of chaos 14 Neural networks Odom & Sharda (1990) Cadden (1991) Coats & Fant (1991) Raghupathi et al. (1991) Tam e Kiang (1992) Chilanti M. (1993) Coats & Fant (1993) Fletcher & Goss (1993) Udo (1993) Weymaere & Martens (1993) Altman et al. (1994) Rosenberg e Gleit (1994) Wilson & Sharda (1994) Boritz et al. (1995) Lenard et al. (1995) Back et al. (1996) Bardos & Zhu (1997) Sironi e Marsella (1998) Yang et al. (1999) Atiya (2001) Neophytou et al. (2001) Daubie & Meskens (2002) of the companies using the so-called “Lyapunov exponent”: the larger this exponent, as soon as the company becomes unpredictable, so it is more risky. These models are inspired by research in biology based on the structure of the human brain, whose the primary computational units can be considered the neuron (Rosenblatt 1961; Minsky and Papert, 1969); it was thought that artificial neurons to be able to solve problems complexes such as the risk of insolvency (F. D'Annunzio, G. Falavigna, 2004). Neural networks are created using a set of input and output nodes that are connected to intermediate nodes called hidden-layer nodes. The hidden layers allow the network to generate a number of mapping functions so that the desired output can be produced using a given set of inputs. When applied to the analysis of the business failure, a neural network allows to assess the risk of business failure, based on an array of information input and on an output vector. In order to build one neural network for predicting bankruptcy of firms, the researcher must use a certain algorithm; different algorithms are proposed, the most common is the backpropagation algorithm (Rumelhart et al., 1986).The neural network model usually nonlinear interactions in a data set much better than the statistical procedures of analysis, especially when the data distribution is unknown (Maher JJ, Tarun K., Sen TK, 1997). The alternative methods mentioned above are clearly more sophisticated and computationally more complex than the classical statistical methods such as discriminant analysis, logit analysis, probit and lineare (Balcaen S. and Ooghe H., 2004). An interesting question to ask is whether these alternative methods produce bankruptcy prediction models higher performance than traditional statistical methods. The question of which forecast method produces the best results has been relieved in several papers; there are many empirical studies that compare the results and/or predictive capability of bankruptcy prediction models based on different techniques. Unfortunately, there is no study that systematically compares all possible methods and that arrives to identify which method is the best (Balcaen S. and Ooghe H., 2004). Perhaps the reason is to be found in the fact that no method is the “best” in absolute. What is the best depend on the details of the problem: the structure and the availability of data, the characteristics used, the objective of the 15 research. With reference to selection of predictive model to be used in this study, it is argued that not only has no generally agreed approach has been achieved, but that the inherent difficulties in designing a satisfactory method of measuring sustainability make any consensus in future most unlikely. As a result, this study only considers failure prediction models estimated with classical statistical techniques, such logistic regression. A second reason why we focus on logistic regression technique is because it is used in most failure prediction research, both in the earlier versions and in the most recent ones. Multiple discriminant analysis is by far the dominant classical statistical method, followed by logit analysis (Altman & Saunders, 1998). 3 Research Methodology The terms of ‘bankruptcy’, ‘failure’, ‘(cash) insolvency’, ‘liquidation’ and ‘(loan) default’ are commonly used and sometimes refer to the same failure concept. An overview of the meaning of these terms can be found in Altman (1993) and in Argenti (1976). It is clear that this study is based on the ‘legal definition’ of failure. According the Text of the laws on local government, approved by Legislative Decree 18 August 2000, No. 267 in Articles 244-269 "The financial distress occurs when the body, municipality or province, is no longer able to perform the essential functions and services defined, or if there are credits against the organizing third parties to whom are unable to cope with the ordinary means of restoring fiscal balance or the debt instrument with off balance sheet" .If it occurs any of these conditions, the Authority has to declare bankruptcy and start a series of procedures to bring itself to the financial recovery through the previous zero debt3, and then return to the condition of being "healthy". The regression models with binary dependent variable allow us to interpret the regression as a probability model whose dependent variable is equal to one; in other words Y is a variable dummy or dichotomy in the sense that it can only assume two values 1 or 0. Since in this research we will study the probability of financial distress of local authorities, in function of some financial ratios, it 3 For debt means the sum of the deficit of the administration to final account last year prior to bankruptcy and debt off balance sheet which occurred before the reference year of distress, recognized as meeting the institutional purposes of local government. 16 is appropriate to use the regression model with binary dependent variable: y is a dummy that takes the value 1 if the local authority is ruined, 0 otherwise. As the dependent variable Y is binary, the regression function of the population corresponds to the probability that the dependent variable is equal to one, given X. The coefficient β1 associated with a regressor (predictor variable) X is the variation in the probability that Y=1 associated to a unitary variation in X (Palomba G., 2008, p. 112). To overcome these problems we use non-linear models specifically designed for binary dependent variables, in other words probit and logit regression models. Many of the published studies are characterized by the application of logit analysis (including probit and logit model), which calculates the conditional probabilities (or logit scores), between 0 and 1, on a sigmoidal curve, that in the holding an event occurring (Hosmer and Lemeshow, 1989). In this research, the logit analysis estimates the probability that a crisis occurs, given the values assumed by the variables in the budget. A particular feature of this methodology is that it does not require the normality hypothesis as regards the distribution of the variables considered: even if the independent variables do not satisfy this condition, as is often the case with budget indexes, the logit analysis still determines consistent estimates (Maddala, 1983; Maddala, 1992). Estimation of the logit or probit model requires the identification of parameters β maximising the function; this takes place by solving a system of non linear equations, obtained by setting at zero the prime derivative of likelihood function with respect to the single coefficients to be estimated. The logit model assumes a logistic distribution; if, on the other hand, a normal distribution is assumed the probit model is obtained. In other words the relation between the probability of the occurrence of an event and the explanatory variables is, in most cases, non linear. Hence, recourse is made to cumulative distribution functions (in logit models) or normal (in probit models). Since the regression with a binary dependent variable Y models the probability that Y=1, it is reasonable to adopt a non linear formulation that constrains the chosen values to assume values 17 between zero and one. In the logit and probit regressions, therefore, cumulative distribution functions or CDF are used, because they produce a probability between one and zero, the normal distribution function for the probit regression, and the logistic CDF for the logit regression, also referred to as logistic regression. Thereby apply a logit-probit model, the purpose of this research is to estimate the risk of insolvency of the Italian local authorities. After identifying the explanatory variables and thus which account values to insert in the econometric model estimating the risk of insolvency, we go on to the empirical application of the model on a sample. Once the local authorities that have experienced problems of insolvency in a given time period have been identified, other local bodies, that can be considered “healthy” (in the sense that their accounts are in order) are then selected at random. The budget values of n bodies, that will be used in the model as regressors, are then measured in order to construct a matrix on which to base the evaluation. The next phase will be the estimation of the model (probit or logit), that is to say evaluation of the unknown parameters; the model will then be validated through statistical procedures. The following section will focus, therefore, on factors of financial risk to be considered as explanatory variables in the econometric models estimating insolvency risks. 3.1 The explanatory variables The accounting values representing financial dynamics highlight anomalies or situations that could lead to financial crisis are as follows: financial autonomy; recovery time of outstanding payments; internal income resources; spending on personnel; total income; outstanding arrears. 18 This study basically maintains that the above-mentioned indexes can be considered as valid proxies for the financial structure of local authorities. In particular, financial autonomy and recovery time of payments synthesise tax and rates policies. Spending levels on personnel synthesise the level of rigidity in the budget, and arrears represent spending commitments that have not be fulfilled for lack of resources. We shall now explain why these indexes have been chosen as explanatory variables in the model. In the evaluation of local authority financial risk tax policy has two aspects. Above all the presence of considerable sums in taxes or rates indicates that the authority is equipped with a certain financial flexibility to cope with any future budgetary demands. Since control and collection are two quite distinct legal parts of the process of guaranteeing income, the means and speed of collecting taxes are very important. The regulation of collection is, in fact, a determining element in the evaluation of a local body’s credit worthiness (Mussari, 2002; p. 27-77). An inefficient system of recovery generates permanent arrears. If these arrears cannot be transformed into earnings in a reasonable time span, they can have a negative effect on the direction and value of the administration’s overall turnover. The reform process aimed at granting autonomy to local bodies hinges on the recognition and importance of their decision taking authority, especially with reference to budget management. This implies that local authorities can either deal with debt recovery themselves or delegate the task to outside agents. In the first case the direct responsibility for recovery can increase the predictability of incoming cash flows (Carnevale, 2006; p. 39-343), but it also has organisational implications (staffing, procedural, logistic) and hence related costs. As regards outstanding debts, the local body can assess the feasibility of adopting coercive measures or of writing off the debts. Through the Factoring Institute local bodies can subcontract debt collection to a third party (Piscino 200.). 19 Recourse to a third party implies legal costs for the service, but avoids the costs connected with coercive measures. It is clear that each authority must make a careful assessment of the costs and benefits of the two options. In the econometric model therefore we must include an index of financial autonomy and recovery time of outstanding debts. Index of financial autonomy: Title 1 + (Title 111/Current earnings*100) This index relates the authority’s earnings (from local taxes/rates and other income) with the total current income identifying to what extent the authority manages to be financially autonomous with respect to the total transfers of income from central and regional government. The function of this is to highlight the level of guaranteed current income from internal sources and to what extent the resources required come from the State. The higher the value of this index the greater the autonomy of the authority (i.e. less dependent on transfers from central government) and the consequent lessening of financial risk. Recovery time of arrears Recovery (Title 1 + Title 111) / Controls (Title 1 + Title 111) In general high percentages indicate structural efficiency and lack of problems regarding debt collection and low risk of financial difficulties. Low percentages indicate instead lack of or adequate technical and human resources, likely difficulties in recovering arrears, negative consequences and, therefore, higher risk of financial difficulties. Spending on personnel is calculated in the following way: Spending on Personnel/Current spending (Title 1) This indicator measures the proportion of overall current spending dedicated to personnel. In the last five years the Italian parliament has passed various laws regarding spending on personnel in local government, in an attempt to reduce the overall level of spending and the ratio of spending on 20 human resources vis-à-vis total spending. In the accounts of public bodies this spending is a fixed item that creates inflexibility and, above all, takes up the greater part of overall income. The Court of Auditors has, indeed, identified the ability to reduce the proportion of the budget spent on staffing as a parameter of virtuous behaviour. The collection of arrears is certainly a crucial factor in the financial management of local authorities. As regards the monitoring of arrears, it is fundamental to discover the nature of the debt; in fact, if the repayment of outstanding arrears gives rise to excessive repayment times, this could have an adverse knock on effect on financial stability of the local authority. In the econometric model the incidence of outstanding arrears must be treated as an independent variable. Level of outstanding arrears Total outstanding arrears/Total commitments)*100 One risk factor of a financial nature linked to the debt position of the authority concerns the contraction of debts not covered by the budget. When the body does not respect the stipulated budget procedure, the commitment undertaken is known as a debt “below the line” (Nobile, 2004). This involves unlicensed budget items that constitute a violation of accounting principles; the Observer for the financing and accounting of local bodies refers to such payments as monetary obligations, relating to the provision of a public service, that is valid from a legal point of view, but is not covered financially. Article 194 of the legislation on Local Authorities (Testo Unico Enti Locali) attributes to the governing council of the local body the right to recognise the legitimacy of such “below the line” undertakings and of reintegrating such sums into the budget. The types of debts “below the line” recognised in Article 194, Section1, derive from executive decisions, cover for deficits of consortiums, specialised companies or institutions, operating within the limits of their statute, convention or founding purpose, provided that the obligation to balance the budget is respected (Article 114), and the deficit incurred by management decisions be covered by the application of political and financial policies (Articles 31 and 114), by the recapitalisation of the 21 companies set up to undertake local public services, by procedures of compulsory purchase or occupation concerning works of legally authorised public utility (Dpr. N. 327, 2001), for the acquisition of goods or services, in violation of the obligations laid down in Sections 1 2 and 3 of Article 191 (any spending, even if carried out incorrectly, provided that it satisfied the needs of the body and is covered by the assets of the body is considered a “below the line debt” that can be reintegrated in the budget). It would be very useful to insert all the possible “below the line” debts in the econometric model. However, since the relevant information is not easily available, all one can include in the explanatory variables are those “below the line” debts that have been recognised. Nevertheless, it would certainly be more fruitful if we could tract down the information on unrecognised “below the line” payments, especially as it is these payments that, being without financial cover, often lead to financial deficits. Unrecognised “below the line” debts escape the attention of official accounting procedures and thus remain an unknown quantity. Moreover, the model will contain control variables- as regressors- to capture the influence that other factors of an economic or territorial nature have on the dependent variable. In order to avoid producing distorted estimates, therefore, it is considered preferable to include in the regression model, as independent variables, other factors that influence the risk of bankruptcy. These are supplementary variables that do not interact with the independent variables in the model. If we want to estimate the causal effect of the variable of interest on the dependent variable correctly it is necessary, in most cases, to control (i.e. neutralise) the disruptive action exercised by one or more supplementary variables, defined for this reason as control variables. The following control variables are included in the model: Total Income, Size of Population, Size of Area, Per capita Income, Demographic Classification. Table 2 – Explanatory variables in the model Explanatory variables Financial Autonomy (X1) Recovery time of arrears (X2) Acronym used afg vrep Description Measures size of own resources (Tit. 1 + Tit. 3) on current income (Tit. 1 + Tit. 2 + Tit. 3). Relation between arrears collected and estimates regarding items 1,2 and 3 of income 22 Proportion of spending on personnel (X3) Incidence of arrears isc rp Measures size of spending on personnel in relation to overall spending Measures size of spending commitments not paid on 31/12 Total Income (X5) te Overall financial resources of the authority. In accounting terms they are equal to the sum of the first four items in the budget; part 5 excluded as contents represent nether income nor expenditure Average earnings (X6) redm Average income Population (X7) abitanti Number of inhabitants in the council considered Area Kmq Total area of authority in kilometres 2KMQ. Demographic classification (X9) cd Councils grouped in three classes 1A, 2 A,,3 A. 3.2 Dataset The sample chosen for the research includes local government authorities in Italy- some healthy from a financial point of view, and others in financial crisis- classified on the basis of different demographic groupings. The list of authorities whose bad financial state had been declared was taken from the Ministry of the Interior while, as regards the healthy authorities, the sample was drawn in a random matter from local authorities throughout the whole country, utilising the web site: Comuni Italiani. For the construction of the accounting indexes we used the official budgetary data from the website of the Ministry of the Interior (link Finanza Locale). The following classification was then used for the demographic groupings: a. councils with less than 5000 inhabitants belong to the first grouping; b. councils with a number of inhabitants between 5,000 and 15,000 belong to the second group; c. councils with a population above 15,000 belong to the third group. The information supplied by the above source enabled us to select in a causal manner, and for each of these we extracted the following information: population, area in square kilometres, average income. From the Ministry of the Interior, and specifically from the item “Bilanci Consuntivi” we collected accounting and other data that allowed us to calculate the explanatory variables, in other words the determinants of financial crisis: revenues, receipts from current transfers, non-tax revenue 23 (assessed and collected), total revenue and fees for staff, running costs, residual income, total expenditure. The analysis was carried out for three years 200, 2003 2006. The dataset is reported in Attachment 1. 3.3 Hypothesis The hypotheses to be demonstrated are the following: H1: the increase in financial autonomy, and the rapidity of access to internal sources of funding decreases the probability of financial crisis; H2: the increase in arrears increases the probability of financial crisis; H3: the increase in the proportion of spending on personnel increases the probability of financial crisis; H4: membership of the largest demographic group (3) reduces the probability of financial crisis. In order to determine which of the explanatory variables considered is statistically significant regarding the probability of financial crisis, logit and probit models were employed, on the understanding that their results would not be dissimilar from each other. 3.4 Logit regression model The logit model assumes the following formulation: Pr(Y = 1|X1,X2, X3 X4, X5, X6, X7, X8 X9) = F (β0 + β1X1 + β2X2 + β3X3 + β4X4 + β5X5 +β6X6 + β7X7 + β8X8 + β9X9i) Where F is the standard logistic distribution function. The Xs are the explanatory variables illustrated in Table 1. Y is the dependent variable “crisis/non crisis” that assumes the value of 1 for the authorities in crisis, and 0 for the healthy authorities. 24 If β is positive, then an increase in X increases the probability that Y=1. If β is negative then an increase in X decreases the probability that Y=1. In this study a 5% significance threshold is assumed. Before the econometric estimation, carried out using STATA software, two variable temporal dummies were created, one for the year 2003 and the other for the year 2006, while 2000 was considered the base year. The temporal dummies are generated both to verify whether the probability of a declaration of financial crisis varies from year to year, and also to evaluate the existence of a correlation between the years. In Table 2 the statistics are reported, while in 3 and 4 the results of the regressions are found. Table 3 - Descriptive statistics Below are reported the results obtained, using non linear models, i.e. logit and probit. With these models the probability of financial crisis is not estimated directly; instead it is necessary to relate the derivative of the probability of crisis to the derivative of the regression (though the control Stata, “Mfx Compute), in this way one can measure the marginal impact of the single regressor on the probability of financial crisis (Pampel, 2000). In the following table the estimates of the logit model are reported. 25 Table 4 - Logit model estimates 3.5Probit regression model The probit model assumes the following formulation: Pr(Y = 1|X1,X2, X3 X4, X5, X6, X7, X8 X9) = Ф (β0 + β1X1 + β 2X2 + β3X3 + β4X4 + β5X5 +β6X6 + β7X7 + β8X8 + β9X9i) Where Ф is the normal distribution function. The Xs are the explanatory variables illustrated in Table 1. Y is the dependent variable: crisis/healthy assumes the value 1 for authorities in crisis and 0 for healthy authorities. 26 If β is positive, then an increase in X increases the probability that Y=1. If β is negative then an increase in X decreases the probability that Y=1. The estimates of the Probit Model are reported in the table below. Table 5 - Probit model estimates 4 Results Tables 4 and 5 illustrate the results achieved using a logit and a probit estimator, respectively. In both cases the estimate was adjusted for heteroschedasticity (Robust) in order to obtain robust standard errors with respect to the potential heteroschedasticity of the errors. In the Logit Model the demographic grouping was significant at 10% level, while population was significant at 5%. The result for demographic grouping (also in the case of Probit) is interesting, in that it indicates that the increase in demographic grouping reduce the probability of bankruptcy. With reference to the 27 variables of interest, we note that the variable for outstanding debts has, surprisingly, only marginal significance at 10% level, while there is a 5% significance level for speed of recovery of income and the incidence of spending on personnel, thereby confirming the expected direction. On the basis of the results obtained, therefore, it is clear that - a unitary increase in the speed of recovery of local sources of income has a negative effect on the probability of financial crisis, reducing it by around 0.71; - a unitary increase in the proportion of spending on personnel determines an increase in the probability of financial crisis of 0.10. The estimates carried out with the Probit Model are as follows: the variables Population, Rp, Vrep and Isc are statistically significant at 5%. In this case as well, the more interesting variables (Vrep and Isc) confirm the expected direction. In particular: - a unitary increase in the speed of recovery of local sources of income (Vrep) has a negative effect on the probability of financial crisis of approximately 0.09; - a unitary increase in the proportion of spending on personnel (Ics) determines an increase in the probability of financial crisis of 0.14. In this study has been demonstrated the following hypothesis: H1: the rapidity of access to internal sources of funding decreases the probability of financial crisis; H3: the increase in the proportion of spending on personnel increases the probability of financial crisis; H4: membership of the largest demographic group (3) reduces the probability of financial crisis. It is significant that, in both estimations, speed of recovery of local sources of income is found to be more significant than financial autonomy (i.e., the proportion of local sources of income on overall current income. It appears, therefore, that the ability to recover money rapidly, rather than the ability to control one’s sources of income is of greater import as regards the probability of bankruptcy. In other words, if the authority acquires the right to control a certain amount of their own resources but does not have the ability to collect them in a short time, then the probability of 28 financial crisis rises. One can state, to sum up, that authorities are more likely to be declared bankrupt, if they are unable to collect their income in a reasonable time, even if they enjoy greater control over their resources, than authorities with less financial autonomy, but a better capability to collect money owed in a short time. From an accounting perspective the results of the estimates determine the following dynamics: - slowness in income collection determines cash flow shortages, that can lead to an increase in debt, and consequently an increase in spending on servicing the debt; such shortages also tend to increase the likelihood of “below the line” spending, in that spending cannot be authorised as there is no financial cover at the time of the operation; - an increase in spending on personnel increases the level of rigidity in the budget and hence reduces the level of financial cover for other expenditure. It is easy to understand how the results, obtained with non-linear models, are similar. This is an important aspect as it not only confirms Econometric Theory, but also and above all it confirms the hypotheses the model rests on. Conclusions The results of the econometric analysis are especially interesting in the first place in the light of the current situation of public finances in Italy. The implementation of a decentralised system of public finance through the implementation of fiscal federalism, has handed financial management over to local authorities. Moreover, the reduction of central government transfers imposes on local authorities greater responsibility in the running of their own incomes and the management of spending. Legislation regarding both these aspects provides further confirmation of their importance. As regards the management of local government sources of income, there have been various pieces of legislation concerning the activity of local bodies, in particular regulating other means of raising income, apart from direct taxation. A regulating authority (Albo) has been set up by the Finance Ministry- DM 11/9/2000 n. 289- to establish the technical, financial and organizational 29 qualifications and competence of members. Moreover, as regards spending on personnel the government has ruled that public bodies must ensure a reduction of spending in this area, both in relative and absolute terms. Within the overall goal of reducing the level of financial risk, it is important to establish a financial dynamic that ensures a closer correspondence between income and spending in terms of time in order to obviate shortages of cash flow. Furthermore, it would be advisable to assess sums earmarked in the budget on the basis of their “sensitivity”, that is to say on the reaction of such spending to variations in market interest rates over a set time span (Speca, 20002; p. 772-821). The results of the research are unique in terms of the accounting implications too. The main consideration in this regard is that accounting data can be evaluated in terms of utility and that utility can be defined in terms of predictive ability This study represents the first empirical study in Italy applying an econometric model predictive of the risk of failure of local authorities. The implications for future research includes the modeling techniques with the second generation. REFERENCES ALTMAN EI (1968), "Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy" The Journal of Finance, 23 (4): 589-609. ALTMAN EI (1973), "Predicting Railroad Bankruptcies in America", Bell Journal of Economics and Management Science, 4 (1): 184-211. ALTMAN EI, HALDEMAN R., NARAYANNAN P. (1977), “Zeta Analisys. A new model to identify bankruptcy risk of corporations”, Journal of Banking and Finance: 29-54. ALTMAN EI, LORRIS B. (1976), "A Financial Early Warning System for Over-the-Counter Broker Dealers", The Journal of Finance, 31 (4): 1201-1217. ALTMAN EI, MARCO G., VARETTO F. (1994), “Corporate distress diagnosis: Comparisons using linear discriminant analysis and neural networks (the Italian experience)”, Journal of Banking and Finance, 18: 505-529. ALTMAN EI, MARGAINE M., SCHLOSSER M., VERNIMMEN P. (1974), “Statistical credit analysis in the textile industry, a French experience”, Journal of Financial and Quantitative Analysis. ALTMAN EI, McGOUGH T. (1974), "Evaluation of a Company as a Going Concern", Journal of Accountancy, December: 50-57. ANSELMI L. (2003), Percorsi aziendali per le pubbliche amministrazioni, Torino: Giappichelli. 30 ANSELMI L., GIOVANELLI L., ET AL. (2005), Principi e metodologie economico aziendali per gli enti locali - L'azienda comune, Milano: Giuffrè. APILADO VP, WARNER DC, DAUTEN JJ (1974), “Evaluative techniques in consumer finance”, J. Finan. Quant. Anal.: 275-283. ATIYA AF (2001),“Bankruptcy prediction for credit risk using neural networks: a survey and new result”, IEEE Transactions on neural networks, 12 (4): 929-935. BACK B., LAITINEN T., SERE K. (1996), “Neural networks and bankruptcy prediction: funds flows, accrual ratios and accounting data”, Advances in Accounting, 14: 23-37. BALCAEN S., OOGHE H. (2004), Alternative methodologies in studies on business failure: do they produce better results than the classical statistical methods?, Working Paper n. 249, Faculteit Economie en Bedrijfskunde, Universiteit Gent. BARDOS M., ZHU W. (1997), “Comparison of discriminant analysis and neural networks: Application for the detection of company failures”, in: Biometric approaches in management science, Kluwer Academic Publishers: 25. BARONTINI R. (2000), La valutazione del rischio di credito. I modelli di previsione delle insolvenze, Bologna: Il Mulino. BEAVER HW (1966), “Financial Ratios As Predictors of Failure”, Journal of Accounting Research, 4: 71111. BEAVER HW, KENNELLY JW, VOSSSOURCE WM (1968), “Predictive Ability as a Criterion for the Evaluation of Accounting Data”, The Accounting Review, 43 (4): 675-683. BEAVER WH (1967), “Financial Ratios as predictors of failure. Empirical Research in Accounting Selected Studies”, in Supplement to The Journal of Accounting Research: 71-111. BEAVER WH (1968a), "Alternative Accounting Measures as Predictiors of Failure", The Accounting Review, 43 (1): 113-122. BEAVER WH (1968b), "Market Prices, Financial Ratios and the Prediction of Failure", Journal of Accounting Research, 6 (2): 179-192. BEAVER WH, MCNICHOLS MF, RHIE JW (2005), “Have Financial Statements Become Less Informative? Evidence from the Ability of Financial Ratios to Predict Bankruptcy”, Review of Accounting Studies, 10: 93–122. BELL T., RIBAR G., VERCHIO J. (1990), Neural nets versus logistic regression: a comparison of each model’s ability to predict commercial bank failures, Paper presented at the “Cash Flow Accounting Conference”, Nice (France): 1-8. BELLESIA M. (1998), Enti locali. Analisi di bilancio, Milano: Ipsoa. BENEDETTI L. (2009), I derivati finanziari degli enti locali e lo schema di regolamento ministeriale del 21.9.2009: una breve riflessione, ww.dt.tesoro.it/.../documenti...finanziaria/.../COMUNE_DI_SIENA.pdf BLACK, FISCHER, COX JC (1976), “Valuing Corporate Securities: Some Effects of Bond Indenture Provisions”, Journal of Finance, XXXI, 2: 351-367. 31 BLUM M. (1974), "Failing Company Discriminant Analysis", Journal of Accounting Research, 12 (1): 125. BLUM MP (1969), The Failing Company Doctrine, Ph.D. diss., Columbia University. BORGONOVI E. (2005), Principi e sistemi aziendali per le pubbliche amministrazioni, Milano: Egea. BORGONOVI E. (2009), “Il concetto di costo standard: valutazione sui possibili effetti derivanti dalla sua applicazione”, in: ANSELMI L. (a cura di), La misurazione delle performance nelle pubbliche amministrazioni, http://151.1.143.150/share/progetti/81/Anselmi%20%20Misurazione%20delle%20performance.pdf BORITZ JE, KENNEDY DB, ALBUQUERQUE A. (1995), “Predicting corporate failure using a neural network approach”, Intelligent Systems in Accounting, Finance and Management, 4 (2): 95-111. BRUNETTI G., CODA V., FAVOTTO F. (1990), Analisi, previsioni, simulazioni economico – finanziarie d’impresa, Milano: EtasLibri. CADDEN D. (1991), Neural networks and the mathematics of chaos – An investigation of these methodologies as accurate predictions of corporate bankruptcy, Paper presented at the First International Conference on Artificial Intelligence Applications on Wall Street, New York. CANNATA F. (2001), “Rating esterni e dati di bilancio: un’analisi statistica”, Studi e Note di Economia, 3: 37-65. CARNEVALE R. (2001), “La riscossione dei tributi assume un ruolo centrale nel rating degli enti locali europei”, Banche e banchieri, 28 (4): 339-343. CHILANTI M. (1993), “Analisi e previsione delle insolvenze: un approccio neurale”, Finanza Imprese e Mercati, 3. COATS PK, FANT LF (1991), “A neural network approach to forecasting financial distress”, The journal of Business Forecasting, 10 (4): 9-12. COATS PK, FANT LF (1993), “Recognising financial distress patterns using a neural network tool”, Financial Management, 22 (3): 142-155. COPPOLA FS, PANARO A., TIZZANO P. (2005), “La finanza pubblica locale nel Mezzogiorno ed il ruolo del sistema bancario. Province e comuni”, La Finanza Locale, 25 (11): 66-89. CROUHY M., GALAI D. (1994), “The interaction between the financial and investment decisions of the firm: The case of issuing warrants in a levered firm”, Journal of Banking and Finance, 18: 861-880. CURRAM S., MINGERS J. (1994), “Neural Networks, Decision tree induction and discriminant analysis: an empirical comparison”, Operational Research Society, 45 (4): 440-450. D’ANNUNZIO N., FALAVIGNA G. (2004), Modelli di analisi e previsione del rischio di insolvenza. Una prospettiva delle metodologie applicate, Ceris. Cnr n°17. DANIELLI EK, PITTALIS MG (2010), Il dissesto finanziario degli enti locali alla luce del nuovo assetto normativo, Studi e Ricerche condotte dal Ministero dell’Interno, Dipartimento per gli affari interni e territoriali, www.finanzalocale.interno.it/docum/istrc.html 32 DAUBIE M., MESKENS N. (2002), Business failure prediction: a review and analysis of the literature, Working Paper, Department of Productions and Operations Management, Catholic University of Mons, Belgium: 1-15. DEAKIN EB (1972), "A Discriminant Analysis of Predictors of Business Failure", Journal of Accounting Research, Spring: 167-179. DEMIRGÜC-KUNT A. (1989), “Deposit institution failures: A review of the empirical literature”, Economic Review, 25 (4): 2-18. DUFFIE D. (1999), “Credit swap valuation”, Financial Analysts Journal, 55: 73-87. DUFFIE D., SINGLETON KJ (1999), “Modeling term structures of defaultable bonds”, Review of Financial Studies, 12: 687-720. DURAND D. (1941), Risk Elements in Consumer Instalment Financing, New York: National Bureau of Economic Research. EDMISTER RO (1972), "An Empirical Test of Financial Ratio Analysis for Small Business Failure Prediction", Journal of Financial and Quantitative Analysis, 7: 1477-1493. EISENBEIS RA (1977), “Pitfalls in the application of discriminant analysis in business, finance, and economics”, J. Finan., 32: 875-900. EISENBEIS RA (1978), “Problems in applying discriminant analysis in credit scoring models”, J. Bank. Finan., 2: 205-219. ESPAHBODI P. (1991), “Identification of problem banks and binary choice models”, Journal of Banking and Finance, 15: 53-71. FARNETI G. (2009), “Il principio contabile n. 1. La programmazione nel sistema di bilancio”, Azienditalia,16 (6): 7-13. FARNETI G. (2009), “Programmare e controllare? È necessario, lo impone il federalismo, ma non solo. I risultati di una ricerca”, Azienditalia, 16 (4): 277-285. FARNETI G., MAZZARA L., SAVIOLI G. (1996), Il sistema degli indicatori negli enti locali, Torino: Giappichelli Editore. FARNETI G., PADOVANI E. (2003), Il check-up dell’ente locale, Milano: Il Sole 24 Ore. FERNANDEZ AI (1988), “A Spanish model for credit risk classification”, Studies in Banking and Finance, 7: 115-125. FERRARA L. (2003), “Credit risk management: la variabile organizzativa”, Contabilità Finanza e Controllo, 26 (1): 89-95. FLETCHER D., GOSS E. (1993), “Forecasting with neural networks: An application using bankruptcy data”, Information and Management, 24 (3): 159-167. FRYDMAN H., ALTMAN EI, KAO DL (1985), “Introducing recursive partitioning for financial classification: The case of financial distress”, Journal of Finance, 40 (1): 269-291. GALLO L., SIMONETTO M. (2004), Gli indicatori di bilancio per l’ente locale, Rimini. Maggioli Editore. GESKE R. (1977), “The Valuation of Corporate Liabilities as Compound Options”, Journal of Financial and Quantitative Analysis, 12 (4): 541-552. 33 GIOVANNELLI L. (2005), “Il sistema di bilancio come strumento di governo dell’ente locale”, in: ANSELMI L. (a cura di) Principi e metodologie economico aziendali per gli enti locali. L’azienda comune, Milano: Giuffrè. GRABLOWSKY BJ, TALLEY WK (1981), “Probit and discriminant functions for classifying credit applicants: a comparison”, J. Econ. Bus., 33: 254-261. GUPTA Y., RAO R., BAGCHI P. (1990), “Linear goal programming as an alternative to multivariate discriminant analysis: A note”, Journal of Business Finance and Accounting, 17 (4): 593-598. HAND DJ, HENLEY WE (1997), “Statistical classification methods in consumer credit scoring: A review”, Journal of the Royal Statistical Society, Series A, 160 (3): 523–541. HAWLEY DD, JOHNSON JD, RAINA D. (1990), “Artificial neural systems: A new tool for financial decision-making”, Financial Analysts Journal, 46 (6): 63-72. HOSMER DW, LEMESHOW S. (1989), Applied Logistic Regression. New York: John Wiley & Sons. HULL JC, WHITE A. (1995), “The Impact of Default Risk on Options and Other Derivative Securities”, Journal of Banking and Finance, 19 (2): 299-322. JARROW R., LANDO D., TURNBULL S. (1997), “A Markov model for the term structure of credit spreads”, Review of financial Studies, 10: 481-523. JARROW RA, TURNBULL SM (1995), “Pricing derivatives on financial securities subject to credit risk”, Journal of Finance, 50: 53-85. JONES EP, MASON SP, ROSENFELD E. (1984), “Contingent Claims Analysis of Corporate Capital Structures: an Empirical Investigation", Journal of Finance, XXXIX, 3: 611-625. JOOS P., VANHOOF K., OOGHE H., SIERENS N. (1998), Credit classification: A comparison of logit models and decision trees, Proceedings Notes of the “Workshop on Application of Machine Learning and Data Mining in Finance, 10th European Conference on Machine Learning”, Chemnitz (Germany): 59-72. JOOS PH, OOGHE H., SIERENS N. (1998), “Methodologie bij het opstellen en beoordelen van kredietclassificatiemodellen”, Tijdschrift voor Economie en Management, 43 (1): 3-48. JOY MO, TOLLEFSON JO (1975), "On the Financial Applications of Discriminant Analysis", Journal of Financial and Quantitative Analysis, 10: 723-739. KAUFFMAN RJ, WANG B. (2001), The success and failure of Dotcoms: A multi-method survival analysis, Working Paper nr. 01-09, Management Information Systems Research Center, Carlson School of Management, University of Minnesota, Minneapolis: 1-7. Keasey, K. and Watson, R., 1991. Financial distress prediction models: a review of their usefulness. British Journal of Management 2, pp. 89–102 KIM IJ, RAMASWAMY K., SUNDARESAN S. (1993), “Does default risk in coupons a.ect the valuation of corporate bonds?: A contingent claim model”, Financial Management, 22: 117-131. LACHENBRUCH PA (1975), Discriminant Analysis, London: Hafner Press, McMillan. LAITINEN T., KANKAANPÄÄ M. (1999), “Comparative analysis of failure prediction methods: the Finnish case”, The European Accounting Review, 8 (1): 67-92. 34 LANDO D. (1998), “On Cox Processes and Credit Risky Securities”, Review of Derivatives Research, 2: 99120. LANE S. (1972), “Submarginal credit risk classification”, J. Finan. Quant. Anal., 7: 1379-1385. LANE WR, LOONEY SW, WANSLEY JW (1986), “An application of the Cox proportional hazards model to bank failure”, Journal of Banking and Finance, 10: 511-531. LENARD MJ, ALAM P., MADEY GR (1995), “The application of neural networks and a qualitative response model to the auditor’s going concern uncertainty decision”, Decision Sciences, 26 (2): 209-227. LENNOX C. (1999), “Identifying failing companies: A reevaluation of the logit, probit and DA Approaches”, Journal of Economics and Business, 51: 347-364. LEV B. (1971), "Financial Failure and Informational Decomposition Measures", in: Accounting in Perspective Contributions to Accounting Thoughts by Other Disciplines, edited by Sterling RR and Bentz WF: 102-11. Cincinnati: Southwestern Publishing Co. LIBBY R. (1975), "Accounting Ratios and the Prediction of Failure: Some Behavioral Evidence", Journal of Accounting Research, 13 (1): 150-161. LINDSAY DH, CAMPBELL A. (1996), “A chaos approach to bankruptcy prediction”, Journal of Applied Business Research, 12 (4): 1-9. LITTERMAN R., IBEN T. (1991), “Corporate Bond Valuation and the Term Structure of Credit Spreads”, The Journal of Portfolio Management, 17 (3): 52-64. LO AW (1986), “Logit versus discriminant analysis: A specification test and application to corporate bankruptcies”, Journal of Econometrics, 31: 151-178. LOMBRANO A. (2001), “Il sistema contabile a supporto dell’attività di controllo direzionale”, in: LOMBRANO A. (a cura di), Il controllo di gestione negli Enti Locali, Rimini: Maggioli. LONGSTAFF FA, SCHWARTZ ES (1995), “A simple approach to valuing risky fixed and floating rate debt”, Journal of Finance, 50 (3): 789-819. LUOMA M., LAITINEN EK (1991), “Survival analysis as a tool for company failure prediction”, Omega International Journal of Management Science, 19 (6): 673-678. MADAN D., UNAL H. (1995), Pricing the Risks of Default, Working Paper, University of Maryland. MADDALA GS (1983), Limited-Dependent and Qualitative Variables in Econometrics, New York: Cambridge University Press. MADDALA GS (1992), Introduction to Econometrics, New York: Maxwell, MacMillan. MAHER JJ, TARUN K., SEN TK (1997), “Predicting Bond Ratings Using Neural Networks: A Comparison with Logistic Regression”, Intelligent Systems in Accounting, Finance and Management, 6: 59-72. MARRADI A. (1993), L'analisi monovariata, Milano: FrancoAngeli. MARTIN D. (1977), “Early warning of bank failure”, Journal of Banking and Finance, 1: 249-276. MARTINIELLO L. (2005), “Risk management e Public Sector comparator nelle Partenership Pubblico Private”, Rivista Italiana di Ragioneria e Economia Aziendale, nov.- dic: 678-689. MAZZOLENI M. (1989), “L’analisi di bilancio negli enti pubblici territoriali”, Azienda Pubblica, 2. 35 MERTON RC (1974), “On the pricing of corporate debt: The risk structure of interest rates”, Journal of Finance, 29: 449-470. MESSIER WF, HANSEN JV (1988), “Including rules for expert system development: An example using default and bankruptcy data”, Management Science, 34 (2): 1403-1415. MINSKY ML, PAPERT S. (1969), Perceptrons, MIT: Cambridge, MA. MONTESI G., PAPIRO G. (2003), “Un approccio forward looking per la stima della Probabilità di Default”, Amministrazione e Finanza, 13. MONTRONE A. (2002), Elementi di metodologie e determinazioni quantitative d’azienda, Milano: FrancoAngeli. MONTRONE A. (2005), Il sistema delle analisi di bilancio per la valutazione dell’impresa, Milano: FrancoAngeli. MORI M. (2001), “Finanza locale e ruolo del rating”, La Finanza Locale, 21 (4): 485-501. MOSES D., LIAO SS (1987), “On developing models for failure prediction”, J. Commrcial Bank Lend., 69: 27-38. MOSSMAN CE, BELL GG, TURTLE H., SWARTZ LM (1998), “An empirical comparison of bankruptcy models”, The Financial Review, 33: 35-54. MOYER R. (1977), "Forecasting Financial Failure: A Re-Examination", Financial Management, 6 (1): 1117. MULAZZANI M. (2006), Economia delle aziende e delle amministrazioni pubbliche, vol. II, Padova: Cedam. MUSSARI R. (2002), Economia dell’azienda pubblica locale, Padova: Cedam. MUSSARI R. (2003), “La contabilità pubblica locale in Europa: tendenza in atto e difficoltà operative”, Azienditalia, 10 (10): 596-600. MYERS JH, FORGY EW (1963), “The development of numerical credit evaluation systems”, J. Am. Statist. Ass., 58: 799-806. NEOPHYTOU E., CHARITOU A., CHARALAMBOUS C. (2001), Predicting corporate failure: empirical evidence for the UK, Working Paper, University of Southampton, Department of Accounting and Management Science: 1-29. NIELSEN LT, SAA-REQUEJO J., SANTA-CLARA P. (1993), Default risk and interest rate risk: the term structure of default spreads, Working paper, INSEAD. NOBILE R. (2004), “I debiti fuori bilancio dei Comuni e delle Province ed i profili della loro riconoscibilità”, La Finanza Locale, 24 (4): 33-64. ODOM M., SHARDA R. (1990), A neural network model for bankruptcy prediction. Proceedings of the Second IEEE International Joint Conference on Neural Networks, San Diego, USA, Vol. II: 163-168. New York: IEEE: 63-68. OHLSON JA (1980), “Financial Ratios and the Probabilistic Prediction of Bankruptcy”, Journal of Accounting Research, 18 (1): 109-131. 36 OOGHE H. and BALCAEN S. (2002), Are Failure Prediction Models Transferable From One Country to Another? An Empirical Study Using Belgian Financial Statements, Working Paper n.132, Faculteit Economie en Bedrijfskunde, Universiteit Gent. OOGHE H., JOOS PH., DE VOS D. (1991), Failure prediction models, Working Paper, Department of Corporate Finance, Ghent University, Belgium. OOGHE H., VERBAERE E. (1982), Determinanten van faling: verklaring en predictive, Unpublished paper, Department of Corporate Finance, Accountancy and Management Information, Ghent University, Belgium. ORIANI M. (2003), La Finanza Innovativa, http://www.bilanci.net/archivio/bilanci2003. PALOMBA G. (2008), Modelli a variabili dipendenti qualitative, DEA Univpm. PAMPEL FC (2000), Logistic regression: a primer, USA: Thousand Oaks, Sage. PAOLONI M., GRANDIS FG (2007), La dimensione aziendale delle amministrazioni pubbliche, Torino: Giappichelli. PISCINO E. (2004), “Il rating negli enti locali”, La Finanza Locale, 24 (11): 109-116. PISCINO E. (2005), “Il factoring pubblico”, La Finanza Locale, 25 (11): 45-65. POHRAT D. (2004), Estimating probabilities of default for German savings banks and credit cooperatives, Discussion Paper Series 2: Banking and Financial Supervision, No 06/2004. POZZOLI S. (2007), “Ma i rischi vanno messi a bilancio”, Il Sole 24Ore, 19 novembre. POZZOLI S. (2008), “Doppia trasparenza sui contratti derivati”, Il Sole 24Ore, 28 gennaio. PUDDU L. (2005), Lezioni di ragioneria pubblica, Milano: Giuffrè. PUNTILLO P. (2007), Gli indicatori per l’analisi della dinamica finanziaria degli enti locali, Milano: Franco Angeli. RAGHUPATHI, SCHKADE, RAJU (1991), “A neural network approach to bankruptcy prediction”, NN in Finance and Investing: Using AI toimprove real-world performance, TRIPPI/TURBAN Irwin Professional Publishing: 227-241. ROSENBERG E., GLEIT A. (1994), “Quantitative Methods in Credit Management: A Survey”, Operations Research, 42 (4): 589-614. ROSENBLATT F. (1961), “Principles of neurodynamics”, Spartan, Washington, DC. RUMELHART DE, HINTON GE, WILLIAMS RJ (1986), “Learning representations by back-propagating errors”, Nature, 323, 9: 533-536. SANTOMERO A., VINSO JD (1977), "Estimating the Probability of Failure for Commercial Banks and the Banking System", Journal of Banking and Finance, 1 (2): 185-205. SCAPENS RW, RYAN RJ, FLECHER L. (1981), “Explaining corporate failure: A catastrophe theory approach”, Journal of Business Finance and Accounting, 8 (1): 1-26. SCARAMUCCI P. (2005), “Gli enti locali e il rischio finanziario”, Azienditalia, 12 (4): 245-253. SCIANCA S. (2008), “I derivati degli enti locali: origine, dimensione e criticità”, Istituto di Studi e Analisi Economica. 37 SERVE S. (2000), Assessment of local financial risk : the determinants of the rating of European local authorities - An empirical study over the period 1995-1998, EFMA Lugano Meetings, Lugano, www.odu.edu SFERRA G. (2008), “Relazione sulla gestione finanziaria degli enti locali”, Corte dei conti. SIRONI A., MARSELLA M. (1998), La misurazione e la gestione del rischio di credito. Modelli, strumenti e politiche, Bancaria editrice. SPECA M. (2002), Gli interest rate swap negli enti locali: rischi, opportunità e nuovo profilo giuridico, http://www-1.unipv.it/websiep/wp/157.pdf. STOCK JH, WATSON MW (2005), Introduzione all'econometria, Milano: Pearson Education Italia. SWANSON E., TYBOUT J. (1988), “Industrial bankruptcy determinants in Argentina”, Studies in Banking and Finance, 7: 1-25. TAFFLER RJ, TISSHAW HJ (1977), “Going, going, gone - Four factors which predict”, Accountancy: 5054. TAM KY, KIANG MY (1992), “Managerial applications of neural networks: the case of bank failure predictions”, Management Science, 38 (7): 926-947. TENUTA P. (2007), Crisi finanziaria e strumenti di previsione del risk management nelle aziende pubbliche locali, Milano: Franco Angeli. TENUTA P. (2008), “Gli strumenti finanziari derivati. Origine, tipologie e possibili effetti sul future management degli enti locali”, La Finanza Locale, 28 (10): 35-44. Trussel J. M.. Patrick P. A, (2009). A predictive model of fiscal distress in local governments, 21 (4), Journal of Public Budgeting, Accounting & Financial Management. UDO G. (1993), “Neural network performance on the bankruptcy classification problem”, Computers and Industrial Engineering, 25: 377-380. VAN WYMEERSCH C., WOLFS A. (1996), La “trajectoire de faillite” des entreprises: une analyse chronologique sur base des comptes annuels, Papers n. 172, Notre-Dame de la Paix, Sciences Economiques et Sociales. VASICEK, OLDRICH A. (1984), Credit Valuation, Working Paper, KMV Corporation, http://www.kmv.com. VESCI M., BOTTI A., MELE R. (2005), “La governance degli enti locali”, Economia e diritto del terziario, 17 (2): 468-487. WEYMAERE N., MARTENS JP (1993), Financial distress analysis with neural networks. Internal paper, Electronics Laboratory, Ghent University: 1-12. WHITE RW, TURNBULL M. (1975a), The Probability of Bankruptcy: American Railroads, Working paper, Institute of Finance and Accounting, London University Graduate School of Business. WHITE RW, TURNBULL M. (1975b), The Probability of Bankruptcy for American Industrial Firms, Working paper. WILCOX J. (1973), "A Prediction of Business Failure Using Accounting Data", Empirical Research in Accounting: Selected Studies, Supplement to Journal of Accounting Research, 11: 169-173. WILSON RL, SHARDA R. (1994), “Bankruptcy prediction using neural networks”, Decision Support Systems, 11 (5): 545-557. YANG ZR, PLATT MB, PLATT HD (1999), “Probabilistic neural networks in bankruptcy prediction”, Journal of Business Research, 44 (2): 67-74. ZAVGREN C. (1983), “The prediction of corporate failure: the state of the art”, Journal of Accounting Literature, 2: 1-33. ZIRIUOLO A. (2000), Il supporto informativo – contabile degli enti locali nel processo di programmazione e controllo, Torino: Giappichelli. 38