Ground motion index-based parametric mechanisms are commonly used in the risk transfer industry. They depend on actual or estimated ground motion parameters, such as peak ground acceleration or pseudo-spectral acceleration values at certain periods of interest. Our research focuses on improving these mechanisms, particularly in reducing basis risk, i.e., the discrepancy between expected payouts and actual payouts triggered by parametric policies. In this paper we assess the sensitivity of a proposed ground motion index-based parametric formulation for Morocco to: i) the polynomial order used to represent the relationship between ground motion values and losses; and ii) to the data partitioning, calibrating distinct functions based on earthquake characteristics such as magnitude, depth, and location. To prevent overfitting, we introduce a data split between training and testing data, and a 5-fold cross-validation process. Results show that when 100% of data are used for training and no cross-validation is considered, parametric models with larger number of partitions relying on high-order polynomial functions produce the lowest errors, whereas the baseline approach consisting of a polynomial function of order 3 without data partition proves to be an excellent and robust model when data overfitting is minimized. The methodology is illustrated for Morocco, but it is general and applicable to any region with an available catastrophe model.