Outlier Detection Using Cluster Analysis for Fixed Income Bonds

摘要

In fixed income asset classes, detection of outliers is a fundamental issue, as it is a tedious task of detection and removal of anomalous objects from gigabytes of data. Outliers can arise due to irregular behavior, incomplete data from source, incorrect capturing of information due to human errors, system error while processing data etc. In this paper we will propose a methodology of outlier detection by using a K-Means [1] clustering [2] technique for fixed income bonds [10], which is commonly known by the name of reference database in investment banking sector [11]. Reference data in investment banking is generally collection of data about different securities like shares, bonds, debentures, loans, fixed income, credit derivatives etc. This methodology is being tested on the problem of detecting bonds which are behaving irregular and may requires attention by subject matter experts.