Risk Management for Pension Funds A Continuous Time Approach with Applications in R 1st Edition Francesco Menoncin instant download
Risk Management for Pension Funds A Continuous Time Approach with Applications in R 1st Edition Francesco Menoncin instant download
https://ebookmeta.com/product/risk-management-for-pension-funds-
a-continuous-time-approach-with-applications-in-r-1st-edition-
francesco-menoncin/
Francesco Menoncin
Risk Management
for Pension Funds
A Continuous Time Approach
with Applications in R
EURO Advanced Tutorials on Operational
Research
Series Editors
M. Grazia Speranza, Brescia, Italy
José Fernando Oliveira, Porto, Portugal
The EURO Advanced Tutorials on Operational Research are a series of short books
devoted to an advanced topic—a topic that is not treated in depth in available
textbooks. The series covers comprehensively all aspects of Operations Research.
The scope of a Tutorial is to provide an understanding of an advanced topic to young
researchers, such as Ph.D. students or Post-docs, but also to senior researchers and
practitioners. Tutorials may be used as textbooks in graduate courses.
Risk Management
for Pension Funds
A Continuous Time Approach
with Applications in R
Francesco Menoncin
Department of Economics and Management
University of Brescia
Brescia, Italy
This Springer imprint is published by the registered company Springer Nature Switzerland AG.
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.1 The Structure of the Book .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 4
1.2 The R Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 5
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 8
2 Decision Theory Under Uncertainty . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 11
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 11
2.2 Decision Theory (Without Risk) . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 11
2.3 Decision Theory (With Risk) . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 13
2.4 Critics to the Expected Utility . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 16
2.5 Risk Aversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 18
2.6 The Stone–Geary Utility Function .. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 24
2.7 Certainty Equivalent on Financial Markets . . . . . . .. . . . . . . . . . . . . . . . . . . . 25
2.8 Utility and Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 29
2.9 A First Pension Model.. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 31
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 36
3 Stochastic Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 37
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 37
3.2 Deterministic Linear Differential Equation.. . . . . .. . . . . . . . . . . . . . . . . . . . 37
3.3 Stochastic Linear Differential Equation . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 38
3.4 Stochastic Models Used in Finance .. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 41
3.5 Parameter Estimation .. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 43
3.6 The Interest Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 45
3.7 Simulation.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 51
3.8 The State Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 53
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 55
4 The Financial Market . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 57
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 57
4.2 Financial Assets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 57
4.3 Portfolio and Wealth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 58
v
vi Contents
Conclusions .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 239
Chapter 1
Introduction
Since the seminal papers of Markowitz (1952) and Merton (1969, 1971), the
literature about optimal asset allocation and risk management has been developing
fast and now takes into account many possible frameworks and applications. Here,
we deal with the application of the asset allocation problem to a more recent topic:
the pension funds.
Unlike analyses dedicated to non-actuarial institutional investors (a general
framework can be found, for instance, in Lioui and Poncet 2001), the case of a
pension fund requires the introduction of two new characteristics: (1) the different
behaviour of the fund wealth during the accumulation phase (hereafter APh) when
contributions are paid by the member and the distribution phase (hereafter DPh)
when the pension is paid to the member, and (2) the mortality risk.
The link between contributions and pensions can be established inside one of
the two following frameworks: the so-called defined-benefit pension plan (hereafter
DB) or the so-called defined-contribution pension plan (hereafter DC). In a DB
plan benefits are fixed in advance by the sponsor and contributions are set in order
to maintain the fund in balance. In a DC plan contributions are fixed and benefits
depend on the returns on fund portfolio (of course, also mix structures are allowed in
the real world). DC pension schemes are becoming more and more important in the
pension systems of most industrialised countries and are replacing the DB schemes
that were more frequent in the past. It is well known that the financial risk is mainly
faced by the sponsor in DB schemes and by the member in DC schemes (see, for
instance, Knox 1993).
Contributions and pensions must be linked by a so-called “feasibility” condition
guaranteeing it is convenient for both the pension fund and the member to
underwrite the pension scheme. Such a condition is also present, for instance, in
Sundaresan and Zapatero (1997) and Josa-Fombellida and Rincón-Zapatero (2001).
The latter work, in particular, examines the problem of a firm which must pay
both wages (before its workers retire) and pensions (after they retire). Thus, a
“feasibility” condition implies the equality between the total expected value of
wages and pensions paid with the total expected value of worker productivity
(according to the usual economic rule equating the optimal wage with the marginal
product of labour).
The demographic dimension is introduced via a survival probability for the
member of the pension fund whose time of death τ is stochastic. Furthermore, in
order to take into take into account a longevity risk, we also assume that the force
of mortality of τ is stochastic itself (in a so-called double stochastic framework).
Let us remark that the mortality risk supported by a single subscriber is much
more important than the one supported by the fund, and, nevertheless, the longevity
risk cannot be effectively managed through a mutualisation technique. The only
effective tool for hedging against longevity risk is a derivative on human life (like,
for instance, a longevity bond).
Here, we take into account a fully funded pension fund where pensions are paid
using only the contributions paid by each member. This means that we do not take
into account any overlapping generation (see, for instance, Haberman and Sung
1994).
The existing literature dealing with the asset allocation problem for a pension
fund in a mixed actuarial and financial framework, mainly neglects the longevity
risk, while some works deal with the mortality risk (see, for instance, the sem-
inal paper of Richard 1975 and Charupat and Milevsky 2002 and Young and
Zariphopoulou 2002 for a more recent reference).
The optimal investment strategy in the accumulation phase (i.e. prior to retire-
ment) in a DC framework has been derived in the literature with a variety of
both objective functions (mainly maximisation of expected utility of final wealth)
and financial market structures (see, among others, Boulier et al. 2001; Vigna and
Haberman 2001; Haberman and Vigna 2002; Deelstra et al. 2003; Devolder et al.
2003; Battocchio and Menoncin 2004; Xiao et al. 2007; Battocchio et al. 2007; Gao
2008; Di Giacinto et al. 2011).
There is relatively little literature on mean-variance portfolio selection in long-
term investment planning and in pension funds (see also Steinbach 2001). Mean-
variance problems for DC plans are solved in He and Liang (2013), Yao et al. (2013),
Yao et al. (2014), and Vigna (2014); for DB plans they are solved in Josa-Fombellida
and Rincón-Zapatero (2008) and Delong et al. (2008). In continuous time the mean-
variance optimisation problem has been solved for the first time analytically by
Richardson (1989), and then by Bajeux-Besnainou and Portait (1998), both through
the martingale approach. The quadratic utility presents some drawbacks that will be
shown in this work while describing how to represent agent’s preferences.
Despite the relevant and increasing hedging need of pension funds and annuity
providers, the market for longevity risk, i.e. the risk of unexpected changes in the
mortality of a group of individuals, is not sufficiently liquid yet.
Many reasons may have contributed to undermine a rapid development of the
market, such as the lack of standardisation, informational asymmetries, and basis
risk. Nevertheless, recent developments provide a sound technology for modelling
the systematic randomness in mortality (see e.g. Lee and Carter 1992), for designing
1 Introduction 3
and evaluating hedging instruments (Blake et al. 2006 and Denuit et al. 2007) and
for managing longevity risk (Barrieu et al. 2012).
Furthermore, the transfer of longevity risk from pension funds to re-insurers
has become more and more common, although on an over-the-counter basis. For
instance, the volume of outstanding UK longevity swaps has reached 50 billion
pounds as of the end of 2014, with a prevalence of very large deals, such as the 16
billion pounds swap between BT Pension Scheme and Prudential and the 12 billion
Euros Delta Lloyd/RGA Re index-based transaction. Investment banks have been
also actively in the transactions. Between 2008 and 2014, alongside reinsurance
specialists, JP Morgan, Credit Suisse, Goldman Sachs, Deutsche Bank and Société
Générale were involved in longevity deals (Luciano and Regis 2014).
Longevity-linked products should be of interest to asset managers for at least
two reasons: their low correlation to other asset classes (at least in the short run, see
Loeys et al. 2007), and their effectiveness in hedging individual investors against the
unexpected fluctuations of their subjective discount factors, which take into account
lifetime uncertainty (Yaari 1965, Merton 1971, Huang et al. 2012).
The aim of this work is to present and study the optimal portfolio choices of a
pension fund subject to longevity risk during both the APh and the DPh. The fund
can invest in a friction-less, arbitrage free, and complete financial market where both
traditional assets (bonds and stocks) and a longevity bond are listed. We consider a
fixed deterministic retirement age, in contrast with Farhi and Panageas (2007) and
Dybvig and Liu (2010) for instance, who consider an endogenous retirement choice.
An extensive literature has explored consumption and investment decisions when
mortality contingent claims are present. In particular, Huang and Milevsky (2008)
analyse the decisions of families in the presence of income risk and life insurance.
Explicit solutions are also obtained by Pirvu and Zhang (2012) with stochastic asset
prices drifts and inflation risk and by Kwak and Lim (2014) with constant relative
risk aversion (CRRA) preferences. All these papers consider a deterministic force
of mortality, while we model it as a stochastic process. We describe longevity risk
by means of a doubly stochastic process whose intensity follows a continuous-
time diffusion (as in Dahl 2004). This process may be correlated with the other
state variables. With stochastic mortality, both individuals and annuity/life insurance
sellers are exposed to unexpected changes in the force of mortality, implying under
or over reserving.
The optimal investment problem of pension funds in the accumulation phase has
been studied for instance by Battocchio and Menoncin (2004) and Delong et al.
(2008), while Battocchio et al. (2007) propose a unified model for describing both
the APh and the DPh. The role of longevity-linked assets in investor’s optimal
portfolio has been addressed first by Menoncin (2008). Maurer et al. (2013), solving
a life-cycle portfolio investment problem with longevity risk, assess the importance
of variable annuities to smooth consumption, while Horneff et al. (2010) analyse the
role of deferred annuities. They find that these products should optimally account
for 78% of the financial wealth of a retiree.
While insurance products are non-marketable, longevity assets on the market
allow individuals to dynamically hedge against mortality fluctuations (we abstract
4 1 Introduction
from transaction costs). Cocco and Gomes (2012) analyse, in the context of a life-
cycle model, the demand for a perfect hedge against shocks in the life expectancy
of a CRRA agent. They study the optimal investment in a longevity bond, which
is akin to our zero-coupon longevity asset. In their numerical simulations, they find
that individuals—at old ages and especially approaching retirement—should invest
a relevant fraction of their wealth in the longevity asset.
Here, we show how to obtain a closed form solution to the problem of a fund,
under Hyperbolic Absolute Risk Aversion (HARA) preferences, when mortality
intensity is stochastic. We also provide a calibrated application, which allows to
appreciate the relevance of longevity products in the optimal portfolio.
The aim of this work is to show how to solve the asset allocation problem of a
pension fund which aims to maximise the expected utility of the wealth remaining
at the death time of a representative pensioner. Of course, this framework is akin to
that of a pension fund which works on a cohort of workers.
If we call τ the stochastic death time of a representative pensioner, wt the vector
containing the fund’s portfolio and ρt the rate used by the fund for discounting
future cash flows, the problem of the fund can be written as
τ
− ρ ds
max Et0 U (Rτ ) e t0 s ,
wt∈[t0 ,τ ]
where Rτ is the wealth of the fund at the death time, U (•) is the utility function
and Et0 [•] is the expected value operator conditional to the information available at
time t0 . The fund’s wealth dynamics is driven by both the values of the assets held
by the fund in its portfolio and the external cash flows given by the contributions and
the pensions. Finally, the fund must face the risks modelled by many state variables
such as: interest rate risk, mortality and longevity risk (stochastic force of mortality),
inflation risk. In our general framework we will take into account any (finite) number
of state variables, but we will also present some particular cases.
Each element of this optimisation problem is studied in this work according to
the following schedule.
1. The utility function is useful for taking into account the risk aversion that the
pension fund is assumed to inherit from its subscribers (the workers/pensioners).
In the first chapter we show how to interpret the shape of the utility function for
representing investor’s preferences.
2. The second chapter presents the stochastic processes that will be used in the
work in order to model the state variables of the optimisation problem. Some
estimation techniques will be shown too.
1.2 The R Software 5
3. The third chapter shows how a financial market can be modelled through the
stochastic processes presented in the previous chapter. In particular, we show
how to model a portfolio of risky and risk-less asset and how to price assets
through the fundamental theorem of asset pricing in an arbitrage free financial
market. The issue of an incomplete financial market will also be shown.
4. Chapter 4 will show how to use stochastic processes for modelling the actuarial
variables, like the survival probability or the mortality intensity.
5. In Chap. 5 the fundamental theorem of asset pricing will be used in a version
suitable for pricing actuarial assets thought of as derivatives written on the
force of mortality which is a stochastic process itself (modelled in the previous
chapter).
6. The Chap. 6 presents the theoretical computation of the optimal portfolio for a
pension fund by using the so-called “martingale method”. We demonstrate that
this optimal portfolio is formed by a speculative and a hedging component and
their roles and characteristics are described in full details.
7. In Chap. 7 we perform a full numerical example with US data and we show the
values of the optimal portfolio containing a risk-less asset, a stock, a rolling zero
coupon bond, and a rolling longevity zero coupon bond. The dynamics of the
optimal asset shares are shown and commented.
8. The last chapter is devoted to the case of a pure accumulation fund that receives
contributions and pays a final amount of money at retirement, without any
mortality/longevity risk (i.e. all the wealth matured for the contributors is paid to
the heirs in case of death). This case is show just as a benchmark for underlining
in a deeper way how the longevity risk affects the optimal asset allocation of a
pension fund.
In this book we will use the R free software https://www.r-project.org/ and its
free interface RStudio https://www.rstudio.com/. A full introduction to the R
programming language is out of the scope of this book. Here, we just out-
line the main features of R. Other (and similar) software like Matlab (or its
freeware clones like Scilab—https://www.scilab.org—or Octave—https://www.gnu.
org/software/octave/), use mainly a vector/matrix approach to computations, while
R also and mainly works with data frames and lists.
In this work we use the package Knitr for LATEX (http://yihui.name/knitr/) which
allows to code and execute R scripts directly on a LATEX document without using
externally the R software.
In the following code we show how to create three sets of data (with command
c) and show the first one.
6 1 Introduction
The number between brackets is the number of the first element of the row. In
this example, the first element of the first row of A is the element number 1.
All the sets that have been created can be put together into a data frame through
the following commands, where we also show the whole set and a subset of it.
Finally, the mean of the second subset is computed. In order to recall a subset of a
data frame, the dollar symbol is needed. In the following example, the data frame
X contains the subsets A, B, and C which are identified as X$A, X$B, and X$C
respectively.
When two matrices are multiplied through the command *, the product is applied
element wise. In order to apply the matrix product, we must use the command %*%,
as we can see in the following example.
If the command rbind is used on a matrix and a single number, a new row
is appended to the matrix and all the elements of this row are equal to that single
number.
SQUIRE GEOFFREY
TREASON