Sorcingové modely ekonomických agentů

Scoring models can be used to quantify the probability of an event, e.g. default. In this sense it can be used to evaluate economic agents quantified probability allows to assess where a particular subject can be assigned to which segment or group. Nowadays the scoring modeling is widely developed field where a wide spectrum of optimization techniques is being used in the context of statistical machine learning. Whereas the problematic of the scoring is complex with overlaps to many fields, it is necessary to examine this topic with connections and consequences of particular approaches, methods, and techniques of optimizations. However current literature doesn´t provide sufficient insight thus this topic is even more important. The another problem of the current theoretical background of scoring modeling is that it doesn´t provide any summarizing econometrical test for verification of stability and performance. This research thus focuses on these topics and in context explains connections between methods with an empirical application on big data to illustrate modeling practice. Reasoning and derivation of the logistic regression with Weigh of Evidence transformation, decision trees with boosting or random forests or neural networks are the pillars of the research. The Test of Performance Relevance is a derived algorithm with statistical machine learning usage by the author, which is based on the concept of the research and can be seen as an overlap between techniques data mining, segmentation, and optimization. In the scoring practice, the crucial point is not only separation ability of a model, the distinguishing power among economic agents, however, the calibration correctness is important too. Thus next layer of the modeling is a macroeconomic component which provides better quality and flexibility of modeling results and at the same moment, it reflects the current economic condition of particular economy or portfolio. The research is finishing with an empirical application on big data and shows connections and options of particular techniques.