Role of Metaheuristic Optimization in Portfolio Management for the Banking Sector: A Case Study

Role of Metaheuristic Optimization in Portfolio Management for the Banking Sector: A Case Study

Soumen Mukherjee (RCC Institute of Information Technology, India), Arpan Deyasi (RCC Institute of Information Technology, India), Arup Kumar Bhattacharjee (RCC Institute of Information Technology, India), Arindam Mondal (RCC Institute of Information Technology, India) and Anirban Mukherjee (RCC Institute of Information Technology, India)
Copyright: © 2019 |Pages: 23
DOI: 10.4018/978-1-5225-8103-1.ch009

Abstract

In this chapter, the importance of optimization technique, more specifically metaheuristic optimization in banking portfolio management, is reviewed. Present work deals with interactive bank marketing campaign of a specific Portugal bank, taken from UCI dataset archive. This dataset consists of 45,211 samples with 17 features including one response/output variable. The classification work is carried out with all data using decision tree (DT), support vector machine (SVM), and k-nearest neighbour (k-NN), without any feature optimization. Metaheuristic genetic algorithm (GA) is used as a feature optimizer to find only 5 features out of the 16 features. Finally, the classification work with the optimized feature shows relatively good accuracy in comparison to classification with all feature set. This result shows that with a smaller number of optimized features better classification can be achieved with less computational overhead.
Chapter Preview
Top

Introduction

Complexity in real-life problems demands optimized solution. The solution should be such that it will consider all the constraints while doing the mathematics and probably the trade-off between hostile objectives; where slight tailoring in the requirements or constraints will be reflected in the output. If the solution consists of dominated and non-dominated depending on a continuously updated algorithm, then it can be termed as Pareto optimization (Ngatchou et. al., 2005; Khoroshiltseva et. al., 2016), whereas in scalarization method, multi-objective functions can be made into one solution using the weight factors (Jahn 1985; Eichfelder 2009; Braun et. al., 2017). The weights in the later case may either be equal or rank-sum or rank order centroid (Braun et. al., 2017), determined by the dominance. The same problem can be solved for single objective or multi-objective depending on the demand or angle of inspection, a classic example is the route hazardous materials in transportation network (Erkut & Alp, 2007). The problem can be solved as single-objective solution by separating the boundary constraints like risk of transportation, road condition, distance between two end points, cost of transport, and risk for the population. Looking from a different angle, either from economical aspect, or from societal perspective, the problem may be converted into a multi-objective optimization (MOO) algorithm. In the latter case, Pareto-optimal set is calculated. Comparative analysis is therefore required as the need of the hour between multiple solution mechanisms for the multi-objective solutions of a single problem (Dey & Choudhury, 2016) considering the diversity, degree of convergence and objective space (Liang et. al., 2016). A few specific real-life problems are recently addressed and optimized in diverse fields like politics (Gunasekara et. al., 2014); energy-harvesting (Sessarego et. al., 2015) etc. Complexity of the problems in most of the cases demands adoption of scalarization methods.

Scalarization method is probably the most popular method among multi-objective optimization methods because of the advantages it offer in case-to-case basis where complexity and the input parameters play the crucial role. For convex problems, this method offers solution by linear scalarization method which is easier from computation point of view. As it is able to convert any multi-objective problem in a series of single objective problems, so the solutions are considered in the form of weighted sum. If weight is imported on a particular parameter/unit for all the objectives, then the linearization method becomes very useful. Under this condition, the solution leads to Pareto optimal point. Nonlinear scalarization methods are also incorporated when priori preference is assigned for some specific input parameters, and weights are redistributed according to that information. It is always good to form Pareto frontier at first by using posteriori, and in this context, evolutionary algorithm has been preferred by researchers (Khan et. al., 2019) in recent past, more precisely when information is not shared about the objective functions. Works are also available where gradient properties of multi-objective landscape are used (Kerschke & Grimme 2017) and claim is made for visualization of local optima superposition. This method is also applied in blast furnace iron making process (Mahanta & Chakraborti 2018) and Pareto trade-off is calculated. However, this method faces some criticism due to lack of convergence theory, or availability of a single Pareto optimal point. Henceforth, various approaches are reported on different problems with multiple objectives, and comparative analysis is also presented with that obtained by evolutionary algorithm.

Complete Chapter List

Search this Book:
Reset