Multiple regression-based models for accurate credit risk management Marwa HASNI * , Safa BHAR LAYEB University of Tunis El Manar, National Engineering School of Tunis UR-OASIS: Optimization & Analysis of Service and Industrial Systems BP 37 Le Belvédère, 1002, Tunis, Tunisia *Marwa.Gharbi@enit.rnu.tn Abstract— In banking system, credit risk represents the expectation of losses stemming from inability of a borrower to repay a loan. In the academic literature, it has been shown that inadequate management of credit risk is a driving factor of financial crises. To accurately control them, banks seek for developing portfolios of information on their customers. To achieve that, the econometric theory advocates using mathematic models (e.g. scoring), which are not only restricted to collecting information on borrowers characteristics, but also, anticipates the quantitative credit risk. In this work, we show that multiple regression-based models are accurate tools to be explored. Specifically, we focused on a real life case study of a private Tunisian bank. In fact, we designed a multiple regression model which enables the bank system to predict the total turnover of a company (i.e. a customer), within a specific time horizon, with regard to eventual changes in financial, macroeconomic and microeconomic data/variables. Accuracy of our approximation is established due to extensive simulation experimentation. Keywords—Forecasting; Management of Credit Risk; Banking System; Loan; Multiple Regression Models. I. INTRODUCTION Credit risk represents eventual loss resulting from non- performance of financial contracts. For commercial banks, several driving factors of credit risk may be the bonds, short-term debt securities and derivatives. Amongst, loans are typically viewed as the mostly influencing. Besides, according to Mačcimskein et al. [1], country risk and settlement risk, which are macroeconomic indicators, are also regarded as credit risks. [2] stated that it is essential for a banking organization to effectively control credit risk in order to ensure its long-term success and evenly, to better cope with inevitable defaults due to global financial crisis. Traditional econometric tools for credit risk management may include rating methods [3], expert systems or neural networks. Relevant examples of these approaches may be found in [4]. All of these rely upon qualitative data analysis (e.g. assessment of future business strategies or appraisals of a business data) and produce estimates based on knowledge gained in the past. A significant example may be found in [5], where a multi- criteria model for credit risk assessment is developed, based on the capability of incorporating value judgements.Mačcimskein et al. [1] established that it was not until 2007, date of the global financial crisis in the USA, that researches started to advocate incorporation of quantitative analysis methods to optimize the risk/return ratio. The relevance of quantitative criteria was firstly highlighted in the work of Saunders and Allen[6], who developed new approaches such as the optional pricing models (e.g. KMV), the VAR models (e.g. CreditMetrics) and time varying models (e.g. CreditPortfolio View). Following this line of research, the concern of our study is to design a new statistical analysis model based on multiple regressions, which provides a private Tunisian bank with accurate information on its credit market including 18 companies. Specifically, for each company, the total turnover is modelled with regard to eventual changes in financial and macroeconomic data/variables. Then, underlying model is used to predict future total turnovers for three consecutive years. The reminder of this paper is organized as follows. In section 2, we briefly describe the case study, then, we sketch the methodological approach being processed in section 3. Details on the experimental design and underlying computation results are included in section 4 and the final conclusions are outlined in section 5. II. THE CASE STUDY: A BRIEF DESCRIPTION The company under study consists of a private Tunisian commercial bank in which the credit process generates, annually, about 60% of its net banking income. Yet, this process involves developing portfolios of information on all its customers as a way to efficiently allocate funds and avoid repayment defaults. For each company, underlying portfolio of information is constructed by means of judgmental estimations of its future financial solvency (i.e. ability to meet its long-term financial obligations), its future liquid assets (i.e. funds, such as stocks or bonds that can easily be converted into cash to meet financial obligations), its rate of return and its margin rate. Thereby, the decision rule of whether a loan should be leaded or not, relies upon comparison of estimates of these ratios with minimal prefixed respective, thresholds. It should be noted, however, that accurately collecting this information is a challenging task, especially because the risk of borrowers is hazard and their performance is difficult to