Essays on Some Recent Penalization Methods with Applications in Finance and Marketing Defended on Thursday, 11 September 2008

The subject of this PhD research is within the areas of

Econometrics and Artificial Intelligence. More concretely, it

deals with the tasks of statistical regression and classification

analysis. New classification methods have been proposed, as well

as new applications of established ones in the areas of Finance

and Marketing.

The bulk of this PhD research centers on extending standard

methods that fall under the general term of loss-versus-penalty

classification techniques. These techniques build on the premises

that a model that uses a finite amount of available data to be trained on

should neither be too complex nor too simple in order to possess a

good forecasting ability. New proposed classification techniques

in this area are Support Hyperplanes, Nearest Convex Hull

classification and Soft Nearest Neighbor.

Next to the new techniques, new applications of some standard

loss-versus-penalty methods have been put forward. Specifically,

these are the application of the so-called Support Vector Machines

(SVMs) for classification and regression analysis to financial

time series forecasting, solving the Market Share Attraction model

and solving and interpreting binary classification tasks in

Marketing.

In addition, this research focuses on new efficient solutions to

SVMs using the so-called majorization algorithm. This algorithm

provides for the possibility to incorporate various so-called loss

functions while solving general SVM-like methods.

Keywords

regularization, instance-based learning, kernel methods, econometrics and machine learning, financial time-series forecasting, binary problems in marketing