Assessment of model risk due to the use of an inappropriate parameter estimator

The purpose of this study is to assess model risk with respect to parameter estimation for a simple binary logistic regression model applied as a predictive model. The assessment is done by comparing the effectiveness of eleven different parameter estimation methods. The results from the historical credit dataset of a certain financial institution confirmed that using several optimization methods to address parameter estimation risk for predictive models is substantial. This is the case, especially when there exists a numerical optimization method that estimates the optimum parameters and minimizes the cost function among alternative methods. Our study only considers a univariate predictor with a static sample size of cases. This research work contributes to the literature by presenting different parameter estimation methods for predicting the probability of default through binary logistic regression model and determining optimum parameters that minimize the objective model's cost function. The Mini-Batch Gradient Descent method is revealed to be the better parameter estimator.