Estimating the default probability for Microfinance institutitons

The objective of this research is to identify the main risk factors that affect microfinance institutions defaults and consequently develop a model able to predict the probability of such events. Data from 197 institutions all around the world were collected from December 2010 to September 2015. The organizations analyzed are part of the portfolio of the Belgian investment manager specialized in impact investing Incofin Investment Management. The study baseline intuition is the plausible existence of some variables influencing the microfinance institutions' defaulting portfolio which is considered ¿ and will be proved¿ to be one of the main dynamics of the default probability. Via Two-Stage-Least-Square regression method the percentage of female borrowers of the organizations will be identified as such variable having the power of improving the portfolio quality. Based on the Wald criterion, Likelihood Ratio test and Variance Inflator Factor, twenty of the initial 75 predictor variables considered are indicated as main factors affecting the default probability. The dataset is analyzed by implementing a binary logistic model, and a binary logistic model with fixed effects. Both the models confirm the key role played by the defaulting portfolio in the determination of the default risk of an institution. In addition the researcher proves that the type of institution impact on its repayment capacity and the same holds for the area in which the organization operates. On the other hand factors such as ROE, liquidity ratio and branch productivity are softening the default risk. Moreover the study shows correlation between the predicted probability and the institution's portfolio composition, in particular financing agricultural, craft and production activities or consumer credit would increase the default probability; whereas orientating the loan portfolio towards service and trade activity would lowering the default risk. The predicted probabilities arising from the models are assessed through predicting capability test and the evidence lead the researcher to opt of the classical logistic regression method. 2.36% is the average predicted probability resulting from the calculations, the result is consider to be reliable according to the high predicting ability and explanatory power. Goodness of fit test results revel 85% accuracy level and furthermore 88% of the default events are predicted correctly while only 0.37% are indicated as non default. The research could be used by the industry to improve the loan decision procedure by easily assessing the default probability with a relatively high level of confidence.