Page 5 - QuantScan-User Guide

Professional investors only
Classification – Statistical learning
Logistic regression models the probability of classifiers with the logistic function.
Decision tree models the probability of classifiers with the Classification And Regression Tree algorithm.
Gradient boosted trees model the probability of classifiers by iterative training of decision trees.
Nearest neighbors analyze the nearest neighbors in the feature space to model the probability of classifiers.
Neural network is composed of layers of neurons which it processes to model the probability of classifiers.
Random forest uses an ensemble of decision trees to model the probability of classifiers.
Y (hat) is the estimated classifier of the fund versus observed returns of the benchmark (respectively Min, Mean and Max return of the benchmark).
AUC-ROC is the abbreviation of « Area Under the Curve-Receiver Operating Characteristics ». ROC is a probability curve indicating how well a model separates the classifiers. The higher the AUC-ROC, the better the model.
A perfect model has an AUC-ROC equal to 1.
Cohen's Kappa measures the inter-rater agreement of the classifiers. In general, Kappa <= 0.20 = poor agreement, Kappa <= 0.40 = fair agreement, Kappa <= 0.60 = moderate agreement, Kappa <= 0.80 = good agreement and Kappa <= 1 = very good agreement. Perfect agreement is 1. Kappa < 0 indicates less agreement than would be expected just by luck.
Classification – Classifier probability
The colors on the graphs show the excess return probability associated to the benchmark return (green indicates outperformance probability and red indicates underperformance probability; green and red add up to 1).
5/8