• automatic trading
• machine learning
• pricing models
• pricing template
• trading
• trading platform

This thesis investigates how data mining algorithms can be used for the prediction of financial instrument types in the context of an electronic trading platform. The exploration of data produced by a testing environment provides a categorization of instruments based on the type template manually assigned to each instrument during its insertion. The intuition is to use these business categories extracted from data to develop and train a machine learning model able to automatically assign a type template to a new financial instrument. The development of the model is conducted following two approaches that differ in the feature selection method and in the target class concept. The first approach evaluated is focused on models trained with automatically selected features and data extracted business categories as target classes. The second approach is focused on models trained with manually selected business features and target classes obtained by grouping similar business data extracted categories. The algorithm used to merge similar categories was developed specifically for the scope and constitutes a novel approach introduced in this thesis work. The best models obtained, trained with the second approach, reach significant performance and provide visualization on the logic behind business classification of instruments. This work is intended to be a pilot research project leading to the implementation of a more versatile and refined model to include in the electronic trading platform.