Towards A Canonical Particle Swarm Optimized
Direct Least Squares Prioritization Method for Ratio
Pairwise Comparison: An Application of Mutual
Fund Selection
Kevin Kam Fung Yuen
Department of Computer Science and Software Engineering,
Xi’an Jiaotong-Liverpool University, Suzhou, China
E-MAIL: kevinkf.yuen@gmail.com,
Abstract—Pairwise comparison is a popular method for
decision making. Analytic Hierarchy Process (AHP) is based on
pairwise comparison (PC), which the rating results of PC forms
pairwise reciprocal matrices (PRM) further to be derived as the
priority values by the Eigen method. Prioritization method using
Eigen method is still open to discuss, although many applications
using this method in AHP have been made. This study proposes
the Canonical Particle Swarm Optimized Direct Least Squares
Prioritization (CPSO-DLSP) method for deriving the priority
values from a PRM. The direct least squares function leads to the
least total square variance between ground PRM and the derived
priority values. The Canonical Particle Swarm Optimization
(CPSO) algorithm is one of the promising methods to effectively
solve the direct least squares optimization problem. An
application of mutual fund selection using the conventional AHP
method is revised and demonstrated by the proposed method.
Keywords—Pairwise comparisons; Particle Swarm
Optimization; Prioritization; Mutual Fund Selection; Analytic
Hierarchy Process.
I. INTRODUCTION
The first use of Pairwise Comparisons (PC) is often
attributed to Ramon Llull, the 13th-century mystic and
philosopher [3]. The PC method was not the main subject for
the scientific investigation but rather its use until Thurstone [6]
proposed “The Law of Comparative Judgments,” in 1927 [3].
The study [7] had a considerable impact on PC research and
led to the Analytic Hierarchy Process (AHP) becoming a
proprietary eponym for PC [3].
Analytic Hierarchy Process is a popular decision tool to
rank the aggregated values from prioritizing the pairwise
reciprocal matrices formed by the rating scores of the pairwise
comparisons performed by the users. Whilst there are a lot of
debates for the validity of the AHP, for example, [1-3,9-10],
many authors of the AHP application papers seem not to be
aware of the problems of AHP in various aspects. In addition
to the scale issues [9-10], AHP’s Eigen method for the
prioritization is one of the controversial topics leading to
misleading outcomes [1-3].
The direct least squares method, an optimization model
proposed by [1], is one of ideal alternative methods which
produces better results than the AHP’s Eigen method in views
of the least square variances between ground pairwise
reciprocal matrix and ratio matrix of the derived priority values.
However, no closed form solution is found for this method.
PSO is one of the recent methods to solve this optimization
problem.
Particle swarm optimization (PSO) firstly developed by [4]
is a population based stochastic optimization technique to find
global optimistic solution. PSO employs a population of
particles, which the number of particles is typically far less
than in the usual evolutionary algorithm, and most researchers
use twenty to fifty particles in a population [11]. There are
various versions of PSO. Canonical Particle Swarm Algorithm
(CPSO) [5] is one of the feasible ones. This research proposes
the structure on the basis of CPSO to solve direct least squares
optimization problem for the prioritization.
The rest of the article is organized as follows. Section II
presents the direct least squares optimization model to solve
prioritization problem. Section III proposes Canonical Particle
Swarm Optimized Direct Least Squares Prioritization (CPSO-
DLSP) method to derive the priority values from pairwise
reciprocal matrix. Section IV demonstrates the proposed
method by revising a financial investment problem using
Saaty’s Eigen method. Section IV concludes this study.
II. DIRECT LEAST SQUARES PRIORITIZATION
The pairwise comparison is conducted in a survey and the
survey results are formed as a ratio pairwise reciprocal matrix
(PRM), denoted by
,
,
,
where
is a number to estimate the relative importance of
object i over object j.
A priority values vector
such that
is derived from a pairwise reciprocal matrix
by