• Explore the loan data set to inform model building. Justify your model and the choice of predictor variables. Explore predictor variables beyond those provided in the SBA data set, in particular time-varying risk factors. Also consider linear and nonlinear model alternatives.
• Select and implement an appropriate method for fitting the model parameters.
• Rigorously test predictive performance (in- and out-of-sample) of your model alternatives using a Receiver-Operating-Characteristic (ROC) curve and other appropriate metrics.
• Explain the fitting results and the fitted model. Which variables are important and why?
• Next develop, fit, and evaluate a model for the loss at default.
• Then, estimate the distribution of total loss on a portfolio of 500 randomly selected loans over one and five year periods (state the loan selection method). Measure the risk in terms of the VaR and the Average VaR (also known as expected shortfall) at the 95% and 99% levels (include confidence bands for your estimates).
• Finally, estimate the distributions for the one and five year losses of an investor who has bought a [5%,15%] tranche backed by the chosen portfolio. Also, consider a [15%, 100%] senior tranche. Interpret and compare the distributions from a risk management perspective.
References