CLASSIFYING THE I TALIAN PENSION FUNDS VIA GARCH DISTANCE Edoardo Otranto 1 , Alessandro Trudda 1 1 Dipartimento di Economia, Impresa e Regolamentazione Universit` a degli Studi di Sassari (e-mail: eotranto@uniss.it, atrudda@uniss.it) ABSTRACT: The adoption of pension funds in the Italian social security policy has increased the offer of several investment funds. The workers have to decide what kind of investment to perform, the funds having a different composition and a subsequent different degree of risk. In this paper we propose to use a distance between GARCH models as a measure of different volatility of some funds, with the purpose to classify a set of funds in terms of different risk. An application on thirteen Italian funds is performed. KEYWORDS: Agglomerative algorithm, Cluster analysis, Maximum risk profile 1 Introduction The main aim of a pension fund is to rise workers’ savings and to invest them accord- ing to an accurate policy of asset allocation, in order to give back the hoarded capital as a life annuity. Therefore, the most important index pension fund refers to the global asset return since it influences both periodic contributions and future benefits. For this reason the valuation of a pension fund is often related to the performance rather than to the risk level of asset portfolio. According to Ryan and Fabozzi (2003) the USA pension shortfalls after 2001 are not the consequence of a poor market perfor- mance, but the inevitable result of actuarial and fiscal accounting practices. In fact, in the long term the expected asset growth must be higher for equities than for bonds. Bader (2003) has an opposite position and argues that risky investment like equity has no place in portfolios pension funds. Such a rule should be applied particularly for pension funds when they concern public and not complementary pension. In Italy the pension plans of private sector workers are run by private funds controlled by government. Trudda (2005) studies actuarial balances and dynamics for independent consultants pensions (ICP) plans. The analysis of each random variable shows that a marginal increase of global asset return gives an important reduction of default proba- bility. In his analysis Trudda (2005) puts in evidence that this kind of funds belongs to the first pillar pensions group and they can not be assimilated as complementary pen- sions fund random variables due to technical and social reasons. Therefore the global asset return analysis should refers to a risk benchmark rather than a return benchmark, recognizing the maximum risk profile. The performance of these ICP funds is var- iegated depending on their asset composition. Some of them have only real or bond investment while some others have high stock market component. The 2006 annual