You are on page 1of 25

Journal of Computational and Applied Mathematics 330 (2018) 228–252

Contents lists available at ScienceDirect

Journal of Computational and Applied


Mathematics
journal homepage: www.elsevier.com/locate/cam

Optimal pension fund management in a jump–diffusion


environment: Theoretical and empirical studies
Charles I. Nkeki
Department of Mathematics, Faculty of Physical Sciences, University of Benin, P. M. B. 1154, Benin City, Edo State, Nigeria

article info a b s t r a c t
Article history: This paper considers a theoretical and an empirical study of an optimal pension fund in
Received 28 March 2017 an inflation environment in which the consumption–portfolio selection problem of an
Received in revised form 21 June 2017 investor who faces both diffusion and jump risks was analyzed. Since the time horizon
of a pension fund management is relatively long, we put into consideration four back-
JEL classification:
ground risks which include inflation, interest rate, investment and income risks. A pension
G11
plan member (PPM) is expected to contribute continuously a time-consistent proportion
G12
C02 of his income into the scheme. These contributions are invested into a market that is
C22 characterized by multiple risk-free assets (which include riskless bonds and bank deposit
C61 accounts), stocks and index bonds. The risky assets (stocks and index bonds), interest rates
and income process are assumed to follow a jump–diffusion process. Real wealth for the
Keywords: plan member is considered. The resulting Hamilton–Jacobi–Bellman equation was solved
Optimal pension fund
using dynamic programming approach. From which, the optimal consumption and optimal
Inflation environment
Consumption
investments with jump risks were obtained. Empirical data were collected from Nigeria
Jump risks Stock Exchange; National Bureau of Statistics, Nigeria; and Bursary Department, University
Income risk of Benin, Nigeria. The analyses of the data were carried out using SPSS package in order
to obtain the needed information for the parameters in our derived models. The resulting
models for optimal portfolio and consumption were solved using MatLab and Mathematica
software. We found that inflation, interest rate and income risks have significant influence
on the investor’s portfolio values in the risky assets. We also found that as the risk averse
coefficient increases, consumption increases and vice versa. Furthermore, we found that
an increase in income can lead to an increase in the portfolio risks and vice versa.
© 2017 Elsevier B.V. All rights reserved.

1. Introduction

This paper focuses on solving problems faced by countries in Africa and many other countries of the World that are
adopting defined contribution pension scheme. Today, there are general increase in prices of goods and services, lack of
price control, varying interest rates by financial institutions, price processes experiencing jump–diffusions and so many
other problems. All of these will go a long way to affect the plan member savings in pension scheme. We intend to provide
both theoretical and empirical analyses of these problems as it relates to pension plan.
In Nigeria for instance, consumer prices in recent time are experiencing continuous jumps from January 2016 to June
2016. In January 2016, the inflation rate was 9.6%, in February, inflation rate was 11.4%, March, 12.8%, April, 13.7% and as
at June 15, inflation rate has risen to 15.6%. It is the highest inflation rate in over 6 years, as cost of items such as food,
housing, utilities and transport surged as a result of 67 percent increase in the prices of gasoline, see [1]. Fig. 1 shows bar
chart of inflation rate in Nigeria from June 2015 to June 2016. At the moment, Nigeria is ranked number 7 highest inflation

E-mail address: iwebuke.nkeki@uniben.edu.

http://dx.doi.org/10.1016/j.cam.2017.07.018
0377-0427/© 2017 Elsevier B.V. All rights reserved.
C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252 229

Fig. 1. Nigeria Inflation Rate in past one year.

Fig. 2. Nigeria Consumer Price Index from June 2015 to June 2016 as by the National Bureau of Statistics, Nigeria as at June of 2016.

rate in Africa and number 12 in the World. The highest in the World is South Sudan with an inflation rate of about 290.30%,
followed by Venezuela with 180.90%, according to the National Bureau of Statistics, Nigeria as at June of 2016. In Nigeria, the
Consumer Price Index (CPI) measures changes in the prices paid by consumers in Nigeria for a basket of goods and services
over time in day-to-day living. CPI in Nigeria increased to 198.30 Index Points as at June, 2016 from 192.99 Index Points in
April of 2016. CPI in Nigeria averaged 77.79 Index Points from 1995 to 2016, reaching an all time high of 198.30 Index Points
in June of 2016 and a record low of 14.36 Index Points in January of 1995 as reported by the National Bureau of Statistics,
Nigeria in June 2016. Fig. 2 shows the bar chart of CPI as reported by the National Bureau of Statistics, Nigeria as at June of
2016. Since inflation rate and stocks are generally experiencing jumps, we must model and assume that our CPI, interest
rate, index bonds and stocks follow jump–diffusion processes. This paper aims at providing both theoretical and empirical
studies of pension fund management in which the environment is made up of general jump in CPI, index bonds, interest
rate, income and stocks prices.
We now give the highlights of our research work as follows:

1. time-dependent contribution rate is considered


2. jump–diffusion interest rate and jump–diffusion income process are considered
3. consumer price index is allowed to follow jump–diffusion process
4. multiple riskless assets, multiple index bonds and multiple stocks are considered
5. optimal investment and optimal consumption are obtained
6. empirical data were used to analyze the resulting models
7. real wealth and consumption plan for a PPM are considered
8. four background risks: interest rate, inflation, investment and income risks are considered
230 C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252

9. theoretical and empirical study of our models are carried out.

The remainder of this paper is structured as follows. In Section 2, we present the literature review on our problem.
Section 3 presents the model formulation, jump–diffusion interest rate, nominal income and asset price processes. In
Section 4, we present the wealth and consumption processes. The real wealth dynamics is presented in Section 5. In Section 6,
we present investment strategies. Section 7 presents the optimal policies. In Section 8, we present the empirical results of
our problems. Section 9 concludes the paper.

2. Literature review

This paper is centered on the following: interest rate, inflation index, asset allocation, and jump–diffusion process. Hence,
the literature review on our problem is structured in the following three subsections:

• first, is the interest rate dynamics and inflation index,


• second, asset allocation,
• last, jump–diffusion process.

2.1. Interest rate and inflation index

A classical dynamic optimization model proposed by [2], assumes a market structure with constant interest rate. But, for
pension funds being a long term management, the assumption of constant interest rates will be unrealistic in practice. [3]
considered a simple diffusion model for a risk-free interest rate that depends upon one factor or source of randomness. [4]
considered the stochastic short term interest rate process and assumed that the process follows the Cox–Ingersoll–Ross
dynamics. [5] and [6] choose the interest rate dynamics that follows [7] specification of the term structure, while [8]
chooses the affine term structure by using the methodology of [4]. [9] proposed the benefits with stochastic behavior of
other variables, such as interest rate and salaries. [10] used a stochastic dynamic programming approach to model a DC
pension fund in a complete financial market with stochastic investment opportunities and two background risks: income
risk and inflation risk. He gave a closed form solution to the asset allocation problem and analyzed the behavior of the optimal
portfolio with respect to income and inflation. [11] modeled inflation index that involves inflation uncertainty. They assumed
that the inflation rate follows the Ornstein–Uhlenbeck process.

2.2. Asset allocation

[12] studied the rate of return guarantee under defined contribution pension plans. They derived an explicit formulae
to value guarantees using martingale pricing theory. They concluded that when the pension plan provides rate of return
guarantees, the plan provider cannot ignore the impact of the participants frequent modification behaviors. [13] considered
the formulation and studied a continuous time stochastic model of optimal asset allocation for a DC pension fund with a
minimum guarantee. There are extensive literature that exist on the area of accumulation phase of DC pension plan and
optimal investment strategies. This can be found in [5,10,14–23]. For optimal portfolio and life-cycle of a PPM consumption
plan, see [24–28] and [29].

2.3. Jump–diffusion

The prices of goods and services in most part of the World today are experiencing jumps. In Nigeria for instance, interest
rate, bond prices, income process, stock prices and consumers price index are following jump–diffusion processes, hence
the purpose of this study. Many authors have studied jump–diffusion processes as it relates to financial and non-financial
disciplines. Our focus is literature on jump–diffusion as it relates to finances. [30] considered jump–diffusion processes in
continuous time. They emphasized mainly on the jump-amplitude distributions and development of more suitable models
using parameter estimation for the market in one phase. They further applied the resulting model to a stochastic optimal
investment in a second phase. The proposed developments include uniform jump-amplitude distributions and time-varying
market parameters. In their paper, they considered a riskless asset, a stock (with the stock was allowed to follow a jump–
diffusion process), interest rate, drift and volatility were assumed to be constants. [31] considered an optimal consumption
and portfolio control strategies for a jump–diffusion stock process with log–normal jumps. They found that a computational
solution can be obtained for an optimal consumption and portfolio policy problem in which the underlying stock satisfies
a geometric jump–diffusion and both the diffusion and jump amplitude are log-normally distributed. Their objective was
to maximize the expected, discounted utility of terminal wealth and the cumulative discounted utility of instantaneous
consumption. It was asserted that the jump–diffusion allows for a more realistic distribution, skewed toward negative jumps
and having leptokurtic behavior in which the tails are thicker so that the distribution will be more slender around the peak
than normal. They considered a stock with the stock allowing to follow jump–diffusion process and drift, volatility, jump
size and jump intensity were all time-dependent. [32] emphasized on jump-amplitude distributions by developing more
appropriate models using parameter estimation for the market. They proposed a method of parameter estimation, this time
C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252 231

for weighted least squares of the difference between theoretical and experimental bin frequencies. The weights were chosen
by the theory of jump–diffusion simulation which was applied to bin frequencies. In their paper, they considered a stock
with the stock allowing to follow jump–diffusion process and drift, volatility, jump size and jump intensity were all time-
dependent as well. [33] applied jump–diffusion process to the growth and treatment of brain tumors using a distributed
parameters model. The model includes three coupled reaction diffusion equations which involve the tumor cells, normal
tissue and the drug concentration. [34] proposed an alternative option pricing model. The model involved the stock prices
that follow a diffusion model with square root stochastic volatility while the jump model follows log-uniformly distributed
jump amplitudes in the stock price process. They further allowed stochastic-volatility to follow a square-root and mean-
reverting diffusion process. They further considered a stock price process that follows a jump–diffusion process with constant
interest rate.
[11] derived the optimal time-dependent portfolio allocation strategy by putting into consideration a long-term stock
market, term structure as well as inflation risk for a defined contribution pension scheme. They further benchmark common
long term defined contribution asset allocation strategies relative to the optimum in order to the assess welfare implications
for varying investment horizons and risk aversions. They found that the outcome of utility heavily depends on whether the
risk aversion of the investor is estimated appropriately. [35] analyzed the consumption and portfolio selection problems
of an investor that are exposed to both Brownian motion and jump risks. They introduced orthogonal decompositions to
determine the optimal portfolio in a closed form. They established the optimal policy that will enable the investor focus on
controlling his exposure to the jump risk. In their paper, they considered a riskless asset and a stock with the stock allowing
to follow jump–diffusion process and drift, volatility, jump size and jump intensity were all time-invariant. [36] considered
dividend optimization problem for an insurer with a jump–diffusion risk process in the presence of fixed and proportional
transaction costs. They constructed and obtained explicitly the value function together with the optimal policy, under a
risk-neutral assumption for the insurer. They further discussed the expected time to the first dividend payment when the
optimal strategy is employed. The underlying asset was allowed to follow a jump–diffusion. [37] considered the optimal
portfolio problems based on the asset price process satisfying a jump–diffusion stochastic differential equation. They derived
the efficient frontier of the optimal portfolio selection problem. They provided a generalization of the work of [38] in the
case of stochastic process and then incorporated jump in their assets prices. But, [38] did not considered asset price process
with a jump–diffusion equation. They considered a riskless asset and multiple stocks with time-dependent interest rate.
[39] developed an optimal timing for annuitization along three approaches. These approaches include the mutual fund in
which the individual invests before annuitization, and it was modeled by a jump diffusion process, the stopping time was
used to maximized the market value of future cash-flows and a solution was proposed in terms of expected present value
operators. They showed that the non annuitization region was either delimited by a lower or upper boundary, in the domain
of time-assets return. They further gave necessary conditions under which these mutually exclusive boundaries exist. They
also proposed a method of computing the probability of annuitization. [40] compares the two different kinds of retirement
plans: a defined benefit (DB) and a defined contribution (DC) plan. They established the tradeoffs and compare these two
plans in a utility-based framework.
None of the above papers address a pension fund management with time-dependent contribution rate, jump–diffusion
interest rate and jump–diffusion income process, jump–diffusion consumer price index, multiple riskless assets, multiple
index bonds, multiple stocks, real wealth and consumption processes. In this paper, we intend to address these problems as
it affects pension plan in the World today.

3. The models

In this section, we present the jump–diffusion interest rate, financial market models in a jump–diffusion and inflation
environment, jump–diffusion income and nominal and real wealth processes.

3.1. Financial market models

In this subsection, we begin by describing our financial models. Let (Ω , F , P) be a probability space and W̄ (t) =
(Wr (t), WI (t), WS (t), WY (t))′ = (W (t), WY (t))′ defined on a given filtered probability space (Ω , F , F(F ), P), where
W (t) = (Wr (t), WI (t), WS (t)),
Ft = σ (W (s), WY (s) : s ≤ t) and F(F ) = {Ft : t ∈ [0, T ]}, is a j + m + n + 1-dimensional Brownian motions with
respect to interest rate risks, inflation risks, stock market risks and an income risk, respectively at time, t and W (t) is a
j + m + n-dimensional Brownian motions with respect to interest rate source of risks Wr (t), an j-dimensional Brownian
motions, inflation source of risks WI (t), an m-dimensional Brownian motions and stock market source of risks WS (t),
n-dimensional Brownian motions, at time t. P denotes the real world probability measure, T the retirement time, the sign
‘‘′ ’’, denotes transpose and, 1 = [1, 1, . . . , 1] ∈ Rm .
Let Pr (t), PI (t) and PS (t) be Lévy pure jump processes with respect to interest rates, index bond and stock, respectively with
Lévy measure λl,r νl,r (dzl,r ), l = 1, 2, . . . , Mr , λk,I νk,I (dzk,I ), k = 1, 2, . . . , MI and λi,S νi,S (dzi,S ), i = 1, 2, . . . , M, where zl,r is the
jump size of interest rate l, zk,I is the jump size of index bond k, zi,S is the jump size of stock i, λl,r ≥ 0, l = 1, 2, . . . , Mr are fixed
parameters representing the jump intensities of interest rates, λk,I ≥ 0, k = 1, 2, . . . , MI are fixed parameters representing
232 C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252

the jump intensities of index bonds and λi,S ∫≥ 0, i = 1, 2, . . . , M are fixed parameters representing the jump intensities of
stocks. The measures ∫νl,r , νk,I and νi,S satisfy R min(1, |zl,r |)νl,r (dzl,r ) < ∞, l = 1, 2, . . . , Mr , R min(1, |zk,I |)νk,I (dzk,I ) < ∞,

k = 1, 2, . . . , MI and R min(1, |zi,S |)νi,S (dzi,S ) < ∞, i = 1, 2, . . . , M, it implies the jumps have finite variation. Now, for any
measurable subsets Ql,r ∈ R, Qk,I ∈ R and Qi,S ∈ R, λl,r νl,r (Ql,r ) = λl,r Q νl,r (dzl,r ), λk,I νk,I (Qk,I ) = λk,I Q νk,I (dzk,I ) and
∫ ∫
l ,r k,I
λi,S νi,S (Qi,S ) = λi,S νi,S (dzi,S ), measures the expected number of jumps, per unit of time, whose sizes belong to Ql,r , Qk,I

Qi,S
and Qi,S , respectively. If νk,I (Qk,I ) = ∞, it is necessarily for the case of small jumps, since νk,I (Qk,I ) < ∞ provided Qk,I does
not include 0, so also for interest rates and stocks. The economy-wide jump amplitude
Pr (t)
[ ]
P(t) = PI (t) ,
PS (t)

with Pr (t) = [P1,r (t), P2,r (t), . . . , PMr ,r (t)]′ , PI (t) = [P1,I (t), P2,I (t), . . . , PMI ,I (t)]′ and PS (t) = [P1,S (t), P2,S (t), . . . , PM ,S (t)]′ ,
scaled respectively on interest rates basis by the scaling factors Jl,r (t , r), l = 1, 2, . . . , Mr , index bond basis by the scaling
factors Jk,I (t , r), k = 1, 2, . . . , MI and on stock prices by the scaling factors Ji,S (t , r), i = 1, 2, . . . , M.
Since occurrence of jump process and Brownian motion are independent, we therefore assume that the individual jump
processes in P(t) and the individual Brownian motions in W (t) are mutually independent.
There is a special case where by ∑Nthe vector P(t)∑ is a compound Poisson∑ process. This occurs where there are a finite number
r (t) NI (t) NS (t)
of jumps of all sizes and Pr (t) = j=1 Z j , P I (t) = m=1 Z m , and PS (t) = n=1 Zn , where Nr (t) is a scalar Poisson process with
intensity parameter λr = (λ1,r , . . . , λMr ,r ) , NI (t) a scalar Poisson process with intensity parameter λI = (λ1,I , . . . , λMI ,I )′ and

NS (t) a scalar Poisson process with intensity parameter λS = (λ1,S , . . . , λM ,S )′ , and the Zj′ s are independent and identically
distributed (i.i.d.) random jump amplitudes with law νr (dzr ) independent of Nr (t). It therefore implies that the interest

rate processes follow a jump–diffusion. Similarly, the Zm s are independent and identically distributed (i.i.d.) random jump
amplitudes with law νI (dzI ) independent of NI (t). It also follows that the inflation-linked bond price processes follow a jump–
diffusion. In the same vain, Zn′ s are i.i.d. random jump amplitudes with law νS (dzS ) independent of NS (t). It also follows that
the stock price processes follow a jump–diffusion.

Remark 1. Note that dtdW (t) = dtdP(t) = dP(t)dW (t) = 0, (dP)m (t) = dP(t), dW (t)dWY (t) = ρY (t)dt, dP(t)dPI (t) =
dP(t)dPY (t) = dP(t).

We give the generalized form of interest rate dynamics in the following subsection.

3.1.1. Jump–diffusion interest rate dynamics


The assumption of constant or stochastic interest rate for a long time investment such as pension plan, to some extent
is unrealistic. Interest rate may experience jump in one time or the other. Hence, in this paper, we assume that our interest
rate is a jump–diffusion process. Let each Pl,r (t), l = 1, . . . , Mr be a Lévy pure jump process with Lévy measure λl,r νl,r (dzl,r ),
λl,r > 0, l = 1, . . . , Mr for interest rates, where zl,r is the jumpsizes of interest rate l. We assume that the Lévy pure jump
processes and the Brownian motions are mutually independent and that the sum of the jumps has support on (−1, ∞). The
function Jrl (t_, τ ) is scaling of the l′ s jump at time t.
We assume an j-dimensional interest rate since most financial institutions, have different interest rates and risk measures
for any deposits. For instance, First bank of Nigeria PLC pays interest of about 5%, Eco Bank 6.5%, Unity Bank 8% for an amount
below ten million naira and 10% for an amount above ten million naira, United Bank of Africa PLC 6%, Guarantee Trust
Bank of Nigeria PLC 5.5%, Fidelity Bank 7.5%, ASUU-UNIBEN Multipurpose Co-operative Society 10% and UNIBEN Faculty
of Agriculture Multi-Purpose Co-operative Society Ltd 10%. The value of these rates may vary over time due to some macro-
economic and micro-economic factors such as inflation, nature disasters, government policy etc.
We consider first a general j-factor model for the forward interest rates fw (t_, τ ), w = 1, 2, . . . , j that follows a jump–
diffusion process and the dynamics is given by:
j Mr w
∑ ∑
dfw (t , τ ) = κw (t , τ )dt + σfw,q (t , τ )dWr ,q (t) + Jrw,q (t_, τ )dPw,r ,q (t),
(1)
q=1 q=1
τ ∈ [0, T ], fw (0, τ ) = f0,w , w = 1, 2, . . . , j,
where κw (t , τ ) is the expected growth rate of the forward interest rate w at time τ , σfw,q (t , τ ), q = 1, 2, . . . , j is the volatility
of forward interest rate w at time τ and Jrw,q (t_, τ ) is the scaling of the w jump for forward interest rate at time τ .
If f (t , τ ) = (f1 (t , τ ), f2 (t , τ ), . . . , fj (t , τ ))′ satisfies (1), then the j numbers of spot rates satisfy

dr(t_) = ψ (t , τ )dt + η(t , τ )dW (t) + Jr (t_, τ )dP(t), (2)

where Jr (t_, τ ) = (Jr1 (t_, τ ), 0m , 0n ), 0m and 0n are Mr by MI and Mr by M zero matrices, Wr (t) = (Wr ,1 (t), Wr ,2 (t),
. . . , Wr ,j (t))′ , WI (t) = (WI ,1 (t), WI ,2 (t), . . . , WI ,m (t))′ , WS (t) = (WS ,1 (t), WS ,2 (t), . . . , WS ,n (t))′ , ψ (t , τ ) = (κ1 (t , τ ), κ2 (t , τ ),
C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252 233

. . . , κj (t , τ ))′ , η(t , τ ) = (σr (t , τ ), 0jm , 0jn ), 0jm and 0jn are j by m and j by n zero matrices, σr (t , τ ) a j by j matrix, Jr1 (t_, τ ) a j
by j diagonal matrix, and

{
ψ (t , τ ) = f (t , τ )|τ =t + ψ (t , t),
∂τ
σr (t) = σf (t , t).
Suppose that Jr1 (t_, τ ) is a zero matrix i.e., interest rates do not experience any jumps, then (2) reduces to the following
j-factor stochastic interest rates:

dr(t) = ψ (t , τ )dt + η(t , τ )dW (t). (3)

3.1.2. The asset price processes


Here, we consider the asset return, price index dynamics and Merton’s problem of maximizing the infinite-horizon
expected utility of consumption by investing in n stocks, m index bonds and j riskless assets at times t ∈ [0, ∞). We assume
j riskless assets because most financial institutions, have different interest rates for any deposits.
The fund manager manages the jump–diffusion fund contributed by a PPM in the planning interval [0, T ], by means of a
portfolio characterized by

B1jB (t , r) = denote the price process of a riskless bond jB at time t,


B2jD (t , r) = denote the price process of a bank deposit jD at time t,
Zk (t , r) = denote the price process of an index bond k at time t, and
Si (t , r) = denote the price process of a stock i at time t, .
I(t , r) = denote the price index at time t driven by a jump–diffusion process
and has the dynamics

dI(t , r) = I(t_, r)(µI (t , r)dt + σ1 (t , r)dWr (t) + σ2 (t , r)dWI (t)


(4)
+ σ3 (t , r)dWS (t) + J1,I (t , r)dPI (t)), I(0) = I0 > 0,
where the function J1,I (t , r) is scaling jump at time t. Since a country should be measured by a single price index, we take the
average of the interest rates, r(t) and r̄(t). We now express µI (t , r) = 1j r(t)1 − 1j r̄(t)1 + σ1 (t , r)θr (t , r) + σ2 (t , r)θI (t , r) +
σ3 (t , r)θS (t , r) which is the expected inflation index at time t, σ1 (t , r) = (σ1,r (t , r), σ2,r (t , r), . . . , σm,r (t , r)) is the volatility
of price index with respect to source risk, Wr (t), σ2 (t , r) = (σ1,I (t , r), σ2,I (t , r), . . . , σm,I (t , r)) is the volatility of price index
with respect to source of inflation risks, WI (t) at time t, σ3 (t , r) = (σ1,S (t , r), σ2,S (t , r), . . . , σn,S (t , r)) is the volatility of price
index with respect to source of stock market risks, WS (t) at time t, θr (t , r) = (θr ,1 (t , r), θr ,2 (t , r), . . . , θr ,j (t , r))′ is the market
price of interest rate risks at time t, θI (t , r) = (θI ,1 (t , r), θI ,2 (t , r), . . . , θI ,m (t , r))′ is the market price of inflation risks at time
t, θS (t , r) = (θS ,1 (t , r), θS ,2 (t , r), . . . , θS ,n (t , r))′ is the market price of stock risks at time t, rl (t) the instantaneous interest
rate for riskless asset l at time t and there is no default risk, and satisfies (2), r̄l (t) real interest rate for riskless asset l at
time t,
θr (t , r)
( )
θ (t , r) = θI (t , r) (5)
θS (t , r)
is the market price of risks and σZ (t , r) = (σ1 (t , r), σ2 (t , r), σ3 (t , r)) with the volatility vector of the price index correlated
with interest rate σ1 (t , r) ∈ [Rj × [0, T ]], inflation, σ2 (t , r) ∈ [Rm × [0, T ]] and volatility vector of price index correlated
with stock price, σ3 (t , r) ∈ [Rn × [0, T ]]. Here, we assume that the consumer price index may be affected by the fluctuation
of interest rates, inflation and stock prices. Note that if consumer price index is affected by inflation alone, it then follows
that σ1 (t , r) = 0 for all t, r and θS (t , r) = 0 for all t, r. Again, if there are no jumps, then J1 (t , r) = 0. (4) can be expressed as
follows:
dI(t , r) = I(t_, r) (µI (t , r)dt + σZ (t , r)dW (t) + JZ (t , r)dP(t)) ,
(6)
I(0) = I0 > 0,
where JZ (t , r) = (0Mr , J1,I (t , r), 0M ), 0Mr is zero vectors of Mr dimensions, 0M is zero vector of M dimensions, J1,I (t , r) =
(J1 (t , r), 0, . . . , 0) ∈ RMI .
The index bond (also referred to as inflation-linked bond) defined in terms of I(t , r) is
1
∫T
Z (t , r) = e− m τ r̄(t)1ds I(t , r) (7)
which pays the price index at maturity. Finding the differential of both sides of (7), we have
1
dZ (t , r) = Z (t , r)[( r(t)1 + σZ (t , r)θ (t , r))dt + σZ (t , r)dW (t) + JZ (t , r)dP(t)],
m (8)
Z (0) = z0 .
Here, (8) represents the dynamics for a single index bond.
234 C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252

But, according to Nigerian Pension Reform Act 2014, some proportion of the PPMs investment must be allocated to bonds
(both risky and riskless bonds), stocks and fixed deposits in Nigerian banks, see Part XII, Section 86, Subsections (a)–(i),
Cap.I24, LFN, 2004 of Nigerian Pension Reform Act, 2014. In this paper, we are interested in the spread of the proportion
allocated to index bond to multiple index bonds, i.e., m numbers of index bonds, j numbers of riskless bonds, n numbers of
stocks and the remainder in fixed deposit accounts. We will give the dynamics of the multiple riskless assets (riskless bonds
and bank fixed deposits), stocks and index bonds at time t in the following subsection.

3.1.3. Riskless assets, index bonds and stocks dynamics


The dynamics of riskless assets (which include jB riskless bonds, B1 (t) with interest rate r1 (t) and for jD bank fixed deposits,
B2 (t) with interest rate r2 (t)), m index bonds and n stocks are given respectively as follows:

dB(t , r) = r(t)B(t , r)dt , B(0) = B0 , (9)

where B(t) = (B1 (t), B2 (t)), r(t) = (r1 (t), r2 (t)), r1 (t) = (r11 (t), r12 (t), . . . , r1jB (t)), r2 (t) = (r21 (t), r22 (t), . . . , r2jD (t)),
B1 (t , r) ∈ RjB and B2 (t , r) ∈ RjD . Note that r1 (t) and r2 (t) satisfy (2).
The stock price process Si (t , r), i = 1, 2, . . . , n at time t which is governed by a jump–diffusion stochastic differential
equation (SDE) with time-dependent and interest rate-dependent coefficients is given as
j m
∑ ∑
dSi (t , r) = Si (t_, r)(ϕi (t , r)dt + σri,q (t , r)dWr ,q (t) + σBi,k (t , r)dWI ,k (t)
q=1 k=1
n M
(10)
∑ ∑
+ σSi,l (t , r)dWS ,l (t) + JSl,i dPl,S (t)), i = 1, 2, . . . , n
l=1 l=1

with Si (0) = Si0 , i = 1, 2, . . . , n, Si (t , r) > 0 for each i, where ϕi (t , r) is the diffusive drift term of stock i at time t, σri,q (t , r)
is the volatility of stock i in state q arising from interest rate risks, σBi,k (t , r) is the volatility of stock i in state k arising from
index bonds risks, σSi,l (t , r) is the volatility of stock i in state l arising from stock market risks, each Pl,S (t) is a Lévy pure jump
process with Lévy measure λl,S νl,S (dzS ), l = 1, . . . , M, where zS is the jump-size of the stock price and JSl,i , i = 1, 2, . . . , n,
l = 1, 2, . . . , M is the scaling factor.
Using (8), we have the k number of index bond price as
j

dZk (t , r) = Zk (t_, r)((rk (t) + σZ r k,q (t , r)θI ,q (t , r)
q=1
m n
∑ ∑
+ σZ I k,k (t , r)θI ,k (t , r) + σZ S k,i (t , r)θS ,i (t , r))dt
k=1 i=1
j m (11)
∑ ∑
+ σZ r k,q (t , r)dWr ,q (t) + σZ I k,k (t , r)dWI ,k (t)
q=1 k=1
n MI
∑ ∑
+ σZ S k,i (t , r)dWS ,i (t) + Jl,Ik dPl,I (t)),
i=1 l=1

with Zk (0) = zk0 , Zk (t , r) > 0 for each k = 1, 2, . . . , m.


Re-writing (10) and (11) in compact form, we have:
multiple stock dynamics:
dS(t , r) = S(t_, r) (ϕ (t , r)dt + σ (t , r)dW (t) + J(t , r)dP(t)) ,
(12)
S(0) = S0 ∈ Rn ,
where the vector ϕ (t , r) = (ϕ1 (t , r), ϕ2 (t , r), . . . , ϕn (t , r))′ is expected growth rate of stocks at time t, J(t , r) = [0Mr ×Mr ,
0MI ×MI , JS (t , r)], σ (t , r) = (σr (t , r)ρr (t), σB (t , r)ρB (t), σS (t , r)ρS (t)) is the diffusive volatility matrix vector and ρr (t), ρB (t)
and ρS (t) are the correlation coefficients at time t with respect to interest rates, index bonds and stocks, 0Mr ×Mr is an Mr by
Mr zero matrix, 0MI ×MI is an MI by MI zero matrix, σr (t , r), σB (t , r) and σS (t , r) are the volatility of stocks with respect to
interest rates, index bonds and stocks at time t, σr (t , r)ρr (t) is an n by j matrix, σB (t , r)ρB (t) is an n by m matrix, σS (t , r)ρS (t)
is an n by n matrix and JS (t , r) is an n by n diagonal matrix.
multiple index bonds denoted by Z̃ (t , r):

dZ̃ (t , r) = Z̃ (t_, r) ((r(t) + σL (t , r)θ (t , r))dt + σL (t , r)dW (t) + JL (t , r)dP(t)) ,


(13)
Z (0) = z0 ∈ Rm ,
where σL (t , r) = (σZ r (t , r), σZ I (t , r), σZ S (t , r)) and JL (t , r) = (0Mr ×Mr , JI (t , r), 0M ×M ), σZ r (t , r) is an m by j matrix, σZ I (t , r) is
an m by m matrix, σZ S (t , r) is an m by n matrix and JI (t , r) is an m by m diagonal matrix.
C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252 235

3.2. The nominal jump–diffusion income process

In a defined contribution scheme, the plan member contributes continuously a time-dependent proportion of his or her
jump–diffusion income to the pension scheme, and the contributions are in turn invested in some suitable underlying assets
available in the market. Let Y (t , r) be the nominal income process of a PPM that is correlated with m number of inflation-
linked bonds and n number of stocks at time t, then Y (t , r) is driven by the following jump–diffusion stochastic differential
equation:
dY (t , r) = Y (t_, r) (β (t , r)dt + σY (t , r)dW (t) + JY (t , r)dP(t) + aY (t , r)dA(t)) ,
(14)
Y (0) = y0 > 0,
where β (t , r) := q(t , r) + βY (t , r) + βw (t , r) ∈ R+ is the expected growth rate of the income and is made up of three compo-
nents where the first part q(t , r) is caused by the inflation, second part βY (t , r) is the mandatory yearly growth rate (for those
salary earners) and the other part βw (t , r) is resulted from other factors, such as economy growth or increased welfare. The
volatility σY (t , r) = (σY1 (t , r), σY2 (t , r), σY3 (t , r)), where σY1 (t , r) = (σY11 (t , r), σY12 (t , r), . . . , σY1j (t , r)) is volatility vector of
the income of a PPM arising from the uncertainty of interest rates, Wr (t), σY2 (t , r) = (σY21 (t , r), σY22 (t , r), . . . , σY2m (t , r)) is
volatility vector of the income of a PPM arising from the uncertainty of inflation markets, WI (t), σY3 (t , r) = (σY31 (t , r), σY32
(t , r), . . . , σY3n (t , r)) is volatility vector of the income of a PPM arising from the uncertainty of stock markets, WS (t), σY4 (t , r)
the volatility of the income process arising from the source of income risk, WY (t) a 1-dimensional Brownian motion, at time
t. The jump process of the PPM’s salary is given by JY (t , r) = (JY1 (t , r), JY2 (t , r), JY3 (t , r)), where JY1 (t , r) ∈ RMr is the jump
process arising from interest rate, JY2 (t , r) ∈ RMI is the jump process arising from inflation markets, JY3 (t , r) ∈ RM is the
jump process arising from stock markets at time t. The vector aY (t , r) = (σY4 (t , r), JY4 (t , r)) with the entries σY4 (t , r) ∈ R and
JY4 (t , r) ∈ R, where σY4 (t , r) is the volatility of PPM’s salary with respect to WY (t) and JY4 (t , r) ∈ R is the jump component
of the income at time t with respect to the jump income PY (t) is a 1-dimensional process and A(t) = (WY (t), PY (t))′ .
Suppose that the time-dependent proportion k(t , r) ∈ R+ of the income process is a contribution rate of the PPM into
the scheme at time t, then c(t , r)Y (t , r) is the amount of fund contributed into the scheme by a PPM at time t.
The asset price processes that make up the portfolio components are discussed in the following subsection.
We consider in the next section, the wealth process.

4. The wealth and consumption processes

Let K (t , r) be the investment process at time t. The PPM consumes continuously at the rate C (t , r) at time t.

Definition 1. The amount of fund invested in stocks, S(t , r) and index bonds Z̃ (t , r) at time, t are define respectively as

∆S (t , r) = (∆S ,1 (t , r), ∆S ,2 (t , r), . . . , ∆S ,n (t , r))


and

∆I (t , r) = (∆I ,1 (t , r), ∆I ,2 (t , r), . . . , ∆I ,m (t , r)).


Define the weight vector

φ (t) = (φB (t), φS (t)),


where φB (t) ∈ Rm is the fraction of K (t , r) invested in bond classes (riskless and risky bonds) at time t and φS (t) ∈ Rn is the
fraction of K (t , r) invested in stock classes (bank fixed deposits and risky stocks) at time t, it then follows that

K (t , r)φB (t) − ∆I (t , r) = ∆B0 (t , r)

is the amount invested in riskless bonds at time t. Note that if ∆B0 (t , r) = 0m , it implies that the entire amount allocated to
bond classes is invested in index bonds. In this paper, we allow ∆B0 (t , r) > 0m . The quantity

K (t , r)φS (t) − ∆S (t , r) = ∆S0 (t , r)

is the amount invested in bank deposit accounts at time t, and

K (t , r) = K (t , r)φB (t) + K (t , r)φS (t),

satisfies Nigerian Pension Reform Act, 2004. It implies that,

∆0 (t , r) = K (t , r)φB (t) + K (t , r)φS (t) − ∆I (t , r) − ∆S (t , r)


is invested in riskless assets at time t, where we assume that n = m = j = jB = jD and ∆(t , r) = (∆I (t , r), ∆S (t , r)).
236 C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252

4.1. The wealth process

Before expressing the dynamics of the wealth process of a PPM, we will first give the following definition.

Definition 2. Let C be the set of admissible strategies and C∆ and Cc with respect to investment and consumption, respectively
be subsets of C . We say:
(i) ∆ ∈ C∆ if
∫ T
E ∆(t , r)∆(t , r)′ dt < ∞, (15)
0

where E is the expectation operator and ∆(t , r) is a control process adapted to filtration {Ft }t ≥0 , Ft -measurable.
(ii) C ∈ Cc if
∫ T
E C (t , r)2 dt < ∞. (16)
0

The process C (t , r) is also a control process adapted to filtration {Ft }t ≥0 , Ft -measurable. We now give the dynamics of
investment and wealth processes of the PPM at time t.

4.1.1. Investment and wealth processes


In the absence of any income flows from outside PPM’s contributions, the wealth, starting with the initial endowment k0 ,
K (t , r) at time t is given by the following dynamics:
dK (t , r) = (K (t_, r)φ (t)r(t) + ∆(t , r)α (t , r))dt + ∆(t , r)Σ (t , r)′ dW (t)
(17)
+ ∆(t , r)Jm (t , r)′ dP(t), K (0) = k0 ∈ R+ ,
where Σ (t , r) = (σL (t , r), σ (t , r)), Jm (t , r) = (JL (t , r), J(t , r)) and α (t , r) = (σL (t , r)θ (t , r), ϕ (t , r) − r2 (t))′ .
Since a PPM start to make flow of contributions k(t , r)Y (t , r) into the scheme at time t and assume continuous
consumption at the rate C (t , r) from the wealth generated at time t, then we have the following definition:

Definition 3. The wealth process of a PPM is defined as

X (t , r) = K (t_, r) + k(t , r)Y (t , r) − C (t , r). (18)


Since it is expected that at the beginning of the planning horizon there will be no outflow of consumption i.e., C (0) = 0,
we now have the following definition.

Definition 4. The outflow of consumption is defined as


∫ t
C (t , r) = C (s, r)ds.
0

Unlike the assumption in literature that the contribution rate of a plan member in pension scheme is always constant, in
this paper, we assume that the contribution rate is time and interest rate dependent.

Proposition 1. The dynamics of the nominal wealth of the investor is


dX (t , r) = (X (t , r)φ (t)r(t) + ∆(t , r)α (t , r) + k(t , r)Y (t_, r)(β (t , r) − φ (t)r(t))
− (1 − φ (t)r(t))C (t , r))dt + (∆(t , r)Σ (t , r)′ + k(t , r)Y (t_, r)σY (t , r))dW (t)
(19)
+ (∆(t , r)Jm (t , r)′ + k(t , r)Y (t_, r)JY (t , r))dP(t) + k(t , r)Y (t_, r)aY (t , r)dA(t),
X (0) = x0 ∈ R+ .

Proof. Finding the differential of both sides of (18) and using (14) and (17), we have the wealth dynamics as follows:
dX (t , r) = (K (t_, r)φ (t)r(t) + ∆(t , r)α (t , r) + k(t , r)Y (t_, r)β (t , r) − C (t , r))dt
+ (∆(t , r)Σ (t , r)′ + k(t , r)Y (t_, r)σY (t_, r))dW (t) + (∆(t , r)Jm (t , r)′
(20)
+ k(t , r)Y (t_, r)JY (t , r))dP(t) + k(t , r)Y (t_, r)aY (t , r)dA(t),
X (0) = x0 ∈ R+ .
Using (18), we have the following:
dX (t , r) = (X (t , r)φ (t)r(t) + ∆(t , r)α (t , r) + k(t , r)Y (t_, r)(β (t , r) − φ (t)r(t))
− (1 − φ (t)r(t))C (t , r))dt + (∆(t , r)Σ (t , r)′ + k(t , r)Y (t_, r)σY (t , r))dW (t)
(21)
+ (∆(t , r)Jm (t , r)′ + k(t , r)Y (t_, r)JY (t , r))dP(t) + k(t , r)Y (t_, r)aY (t , r)dA(t),
X (0) = x0 ∈ R+ . □
C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252 237

(21) represents the nominal wealth of a PPM at time t. But, pension plan is relatively long period of time and considering
nominal wealth will be unrealistic since the investor cares about the utility of the real wealth level as the purchasing power
of her nominal wealth is diminished by the inflation. Hence, the real income and wealth processes are discussed in the
following sections.

5. The real wealth dynamics

In this section, we consider the real wealth for a PPM. Here, we assume that the fund manager is interested in the utility
of her real wealth level since the purchasing power of the nominal wealth is diminished by the inflation. The real wealth is
simply the ratio of the nominal wealth to the price index. Again, we will express the portfolio process as fraction of wealth,
since it is advantageous to express the pension fund dynamics in terms of fraction of wealth instead of amount of shares
invested in each asset.

Definition 5. The real wealth of the investor, X̄ (t , r) at time t is defined as the ratio of the nominal wealth to the price index.
Mathematically,
X (t , r)
X̄ (t , r) = .
I(t , r)

Proposition 2. The dynamics of the real wealth of the investor at time t is


X (t , r)
( )
dX̄ (t , r) = d = X̄ (t , r)[φ (t)r(t) + π (t , r)α (t , r)
I(t , r)
+ k(t , r)y(t , r)(β (t , r) − φ (t)r(t)) − (1 − φ (t)r(t))c(t , r) − µI (t , r)
+ σZ (t , r)σZ′ (t , r) − (π (t , r)Σ (t , r)′ + k(t , r)y(t , r)σY (t , r))σZ′ (t , r)
− ρY (t)k(t , r)y(t , r)σY4 (t , r)σZ (t , r)′ ]dt + X̄ (t , r)(π (t , r)Σ (t , r) (22)
+ k(t , r)y(t , r)σY (t , r) − σZ (t , r))dW (t) + X̄ (t , r)[(1 − J1 (t , r))
(π (t , r)Jm (t , r)′ + k(t , r)y(t , r)JY (t , r)) + JZ (t , r)JZ (t , r)′
X (0)
− JZ (t , r)]dP(t) + X̄ (t , r)k(t , r)y(t , r)aY (t , r)dA(t), = x̄0 ∈ R+ ,
I(0)
with ρY (t) = (ρY1 (t), ρY2 (t), ρY3 (t)).

Proof. We commence by re-write (21) as follows:


dX (t , r)
= (φ (t)r(t) + π (t , r)α (t , r) + k(t , r)y(t , r)(β (t , r) − φ (t)r(t))
X (t_, r)
− (1 − φ (t)r(t))c(t , r))dt + (π (t , r)Σ (t , r)′ + k(t , r)y(t , r)σY (t , r))dW (t) (23)
+ (π (t , r)Jm (t , r) + k(t , r)y(t , r)JY (t , r))dP(t) + k(t , r)y(t_, r)aY (t , r)dA(t),

X (0) = x0 ∈ R+ ,
∆(t ,r)
where π (t , r) = indicates the fraction of portfolio given by the jump–diffusion wealth process at time t ∈ [0, T ] and
X (t ,r)
π (t , r) = (πB (t , r), πS (t , r)) and πB (t , r) = ∆XI(t(t,,r)r) and πS (t , r) = ∆XS(t(t,,r)r) are fraction of portfolio invested in index bonds and
Y (t ,r)
stocks respectively, at time t. y(t , r) = X (t ,r) indicates the fraction of wealth given by the jump–diffusion income at time
t ∈ [0, T ].
C (t ,r)
c(t , r) = X (t ,r) indicates the fraction of wealth taken by consumption at time t.
By the application of Itô formula for quotient rule and using (6) and (23), we have the following
X (t , r)
( )
dX̄ (t , r) = d = X̄ (t , r)[φ (t)r(t) + π (t , r)α (t , r)
I(t , r)
+ k(t , r)y(t , r)(β (t , r) − φ (t)r(t)) − (1 − φ (t)r(t))c(t , r) − µI (t , r)
+ σZ (t , r)σZ′ (t , r) − (π (t , r)Σ (t , r)′ + k(t , r)y(t , r)σY (t , r))σZ′ (t , r)
− ρY (t)k(t , r)y(t , r)σY4 (t , r)σZ (t , r)′ ]dt + X̄ (t , r)(π (t , r)Σ (t , r) (24)
+ k(t , r)y(t , r)σY (t , r) − σZ (t , r))dW (t) + X̄ (t , r)[(1 − J1 (t , r))
(π (t , r)Jm (t , r)′ + k(t , r)y(t , r)JY (t , r)) + JZ (t , r)JZ (t , r)′
X (0)
− JZ (t , r)]dP(t) + X̄ (t , r)k(t , r)y(t , r)aY (t , r)dA(t), = x̄0 ∈ R+ ,
I(0)
238 C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252

where ρY (t) = (ρY1 (t), ρY2 (t), ρY3 (t)). The correlation coefficient between interest rate risks and income risks is given by
ρY1 (t) ∈ Rj , between inflation risks and income risks is given by ρY2 (t) ∈ Rm and between stock risks and income risks is
given by ρY3 (t) ∈ Rn . □

Note that the fraction of wealth invested in the riskless assets is given by

π0 (t , r) = (π0B (t , r), π0S (t , r)) = (1 − k(t , r)y(t , r) + c(t , r))φ (t) − π (t , r), (25)

where π0B (t , r) and π0S (t , r) are fraction of wealth invested in riskless bonds and banks fixed deposits respectively, at
time t.
For the sake of simplicity, from now on we will not indicate the functional dependences, unless it is necessary to do so. It
now follows that (24) will become:
( )
X
dX̄ = d = X̄ [φ r + πα + ky(β − φ r) − (1 − φ r)c(t , r) − µI + σZ σZ′
I
− (π Σ ′ + kyσY )σZ′ − ρY kyσY4 σZ′ ]dt + X̄ (π Σ ′ + kyσY − σZ )dW (t)
(26)
+ X̄ [(1 − J1 )(π Jm′ + kyJY − JZ )]dP(t) + X̄ kyaY dA(t),
X (0)
= x̄0 ∈ R+ .
I(0)

6. Optimal investment strategies

It is expected that PPMs should consume continuously throughout his lifetime. Then the goal of the fund manager
is to select a portfolio and consumption weight processes (c(s, r), π (s, r))t ≤s≤∞ that maximizes the expected utility of
consumption in an infinite horizon, where discounted rate of consumption is ξ and it measures the PPMs preference rate of
consumption. Now, the expected utility of consumption is given by
[∫ ∞ ]
U(t , X̄ , r) = sup Et e −ξ s
V (c(s, r))ds , (27)
c(s,r),π (s,r);t ≤s≤∞ t

subject to the dynamics of his real wealth (26), and with X̄ (t , r) given and total interest rate dynamics 1r(t) using (2)and
1 ∈ Rm a unit vector. In other words, we solve the following problem:
⎧ [∫ ∞ ]

⎪U(t , X̄ , r) = sup E t e−ξ s
V (c(s , r))ds ,
c(s,r),π (s,r);t ≤s≤∞

t


⎨s.t .



[ ] [ ] [ ] (28)
1r dW dP
Φ ,

d = mdt + M +


X̄ dWY dPY





X̄ (0) = x̄0 , 1r(0) = 1r0 ,

where

⎡ ⎤

⎣X̄ [φ r + π α + ky(β − φ r) − (1 − φ r)c(t , r) − µI + σZ σZ′ ⎦ ,


⎢ ⎥
m=⎢ ⎥

−(π Σ ′ + kyσY )σZ′ − ρY kyσY4 σZ′ ]


[ ]
0
M= ,
X̄ [π Σ ′ + kyσY − σZ ] X̄ kyσY4

[ ]
1Jr 0
Φ= .
X̄ [(1 − J1 )(π Jm

+ kyJY − JZ )] X̄ kyJY4

The scalar variables 1r and X̄ = x represent the two state variables and the elements of π (t , r) represent the m and n
control variables and c(t , r) represents a control variable.
Here, we adopt the stochastic dynamic programming and the suitable form of Itó’s formulae for semi-martingale
processes to obtain our Hamilton–Jacobi–Bellman equation that characterized the optimal solutions to the fund manager’s
C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252 239

problem. It is obtained as follows:

0 = Ut (t , x) + 1ψ Ur + x[φ r + πα + ky(β − φ r) − (1 − φ r)c(t , r) − µI + σZ σZ′


1
− (π Σ + kyσY )σZ′ − ρY kyσY4 σZ′ ]Ux + 1ηη′ 1′ Urr + x(π Σ ′ + kyσY − σZ )η′ 1′ Urx
2
1 1
+ x (π Σ + kyσY − σZ )(π Σ + kyσY − σZ )′ Uxx + x2 k2 y2 σY24 Uxx
2 ′ ′
2
∫ ∞ 2
+ [U(t , x, 1r + 1Jr1 ) − U(t , x, 1r)]λr νr (dJr ) (29)
−1
∫ ∞
+ [U(t , x + x((1 − J1 )(π Jm′ + kyJY − JZ ))) − U(t , x)]λν (dJ)
−1
∫ ∞
+ λY [U(t , x + xkyJY4 ) − U(t , x)]νY (dJY ) + e−ξ t V (c(t , r))
−1

with the transversality condition limt −→∞ E [U(t , X̄ , r)] = 0, see [2] and [35], where λν (dJ) = (λr νr (dJr ), λI νI (dJI ),
λS νS (dJS ))′ ,
λr νr (dJr ) = (λ1,r ν1,r (dJr ), λ2,r ν2,r (dJr ), . . . , λj,r νjr ,r (dJr )),
λI νI (dJI ) = (λ1,I ν1,I (dJI ), λ2,I ν2,I (dJI ), . . . , λm,I νm,I (dJI ))′ ,
∂U ∂U ∂2U ∂U
λS νS (dJS ) = (λ1,S ν1,S (dJS ), λ2,S ν2,S (dJS ), . . . , λn,S νn,S (dJS ))′ , λY ∈ R. We denote Ut ≡ ∂t
, Ur ≡ ∂r
, Urr ≡ ∂r2
, Ux ≡ ∂x
,
∂2U ∂2U
Urx ≡ ∂r∂x
and Uxx ≡ ∂ x2
.

The standard time-homogeneity argument for the case of infinite-horizon problems stipulates that
[∫ ∞ ]
eξ t U(t , X̄ , r) = sup Et e−ξ (s−t) V (c(s, r))ds
{c(s,r),π (s,r):t ≤s≤∞} t
[∫ ∞ ]
= sup Et e−ξ u V (c(s + u, r))du
{c(t +u,r),π (t +u,r):0≤u≤∞} 0
[∫ ∞ ]
= sup E0 e−ξ u V (c(u, r))du
{c(u,r),π (u,r):0≤u≤∞} 0

≡ L(t , X̄ ),

which is independent of time and the optimal control is Markov. Hence, U(t , X̄ ) = e−ξ t L(t , X̄ ) and (29) reduces to the
following equation for the time-homogeneous value function L:

0 = Lt + 1ψ Lr + x[φ r + πα + ky(β − φ r) − (1 − φ r)c(t , r) − µI + σZ σZ′


1
− (π Σ ′ + kyσY )σZ′ − ρY kyσY4 σZ′ ]Lx + 1ηη′ 1′ Lrr + x(π Σ ′ + kyσY − σZ )η′ 1′ Lrx
2
1 1
+ x2 (π Σ ′ + kyσY − σZ )(π Σ ′ + kyσY − σZ )′ Lxx + x2 k2 y2 σY24 Lxx
2 2
∫ ∞
+ [U(t , x, 1r + 1Jr1 ) − U(t , x, 1r)]λr νr (dJr ) (30)
−1
∫ ∞
+ [L(t , x + x((1 − J1 )(π Jm′ + kyJY − JZ ))) − L(t , x)]λν (dJ)
−1
∫ ∞
+ λY [L(t , x + xkyJY4 ) − L(t , x)]νY (dJY ) + V (c(t , r)) − ξ L(t , X̄ ),
−1

with the transversality condition limt −→∞ E [e−ξ t L(t , x)] = 0.


The maximization problem in (30) separates into one for c(t , r) , with first-order condition

∂ V (c(t , r)) ∂ L(t , x)


= (1 − φ r)
∂ c(t , r) ∂x
240 C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252

and one for π :


0 = sup{xπ α Lx − xπ Σ ′ σZ′ Lx + xπ Σ ′ η′ 1′ Lrx + x2 π Σ ′ (π Σ ′ + kyσY − σZ )′ Lxx
π
∫ ∞ (31)
+ [L(t , x + x((1 − J1 )(π Jm′ + kyJY − JZ ))) − L(t , x)]λν (dJ)}.
−1

It is imperative to note that Lxx must be strictly negative. Again, the second order condition holds if the following Hessian
matrix

H = x2 Lxx ΣΣ ′

is negative definite. Since ΣΣ ′ is a variance–covariance matrix, we have that ΣΣ ′ is a positive definite matrix. It then follows
that H can only be negative definite matrix if and only if Lxx < 0. This shows that as the utility function V is concave in x, the
value function L becomes a strictly concave function in x (see [9]). It follows that the matrix ΣΣ ′ is invertible.
Let c ∗ (t , r) denote the optimal fraction of wealth consume at time t and π ∗ (t , r) = (πB∗ (t , r), πS∗ (t , r)) denote the tensor
of optimal fraction of wealth invested in the index bonds and stocks at time t. From (30), we obtain the following: Given
wealth x, the optimal consumption choice is therefore obtained as follows:
)−1 (
∂V ∂ L(t , x)
( )
c ∗ (t , r) = (1 − φ r) . (32)
∂c ∂x
In the pure diffusive case, λZ = 0 and we obtain the variational form of Merton solutions
′∗ −(α − Σ ′ σZ′ )Lx − Σ ′ η′ 1′ Lrx − xΣ ′ (kyσY − σZ )′ Lxx
π = (Σ ′ Σ )−1 . (33)
xLxx
We must be specific about the utility function V so as to determine the optimal portfolio weights, wealth and value function.
1−δ eh(t ,r)
Consider an investor with the following value functions, L(t ; x, r) = 1−γ 1
x1−γ eh(t ,r) and V (c(t , r)) = c 1−δ with CRRA
coefficients γ , δ ∈ (0, 1) ∪ (1, ∞), where δ and γ are constant relative risk aversion parameters with respect to consumption
1−δ
and wealth processes respectively and L(T ; x, r) = 1−γ 1
x1−γ and V (c(T , r)) = c1−δ .
Next, we find the partial derivatives of L(t , x, r) with respect to t, r, x, rx, rr and xx as follows:
1
Lt (t , x, r) = x1−γ ht (t , r)eh(t ,r) = L(t , x, r)ht (t , r) (34)
1−γ

1
Lr (t , x, r) = x1−γ hr (t , r)eh(t ,r) = L(t , x, r)hr (t , r) (35)
1−γ

1
Lrr (t , x, r) = x1−γ [h2r (t , r) + hrr (t , r)]eh(t ,r) = L(t , x, r)[h2r (t , r) + hrr (t , r)] (36)
1−γ

Lx (t , x, r) = x−γ eh(t ,r) (37)

Lxx (t , x, r) = −γ x−γ −1 eh(t ,r) (38)

Lrx (t , x, r) = x−γ −1 hr (t , r)eh(t ,r) (39)

L(t , x + x((1 − J1 )(π Jm



+ kyJY − JZ )), r)
1 (40)
= x1−γ [(1 − J1 )(π Jm

+ kyJY − JZ )]1−γ eh(t ,r)
1−γ

1
L(t , x + xkyJY4 , r) = x1−γ [x + xkyJY4 ]1−γ eh(t ,r) (41)
1−γ

1 1
L(t , x, r + 1Jr ) = x1−γ eh(t ,r +1Jr ) = x1−γ eh(t ,r) eh(t ,1Jr ) (42)
1−γ 1−γ

h(t , r + 1Jr ) := h(t , r) + h(t , 1Jr ). (43)


C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252 241

Fig. 3. Contribution rate of a PPM who chooses to invest in First Bank of Nigeria PLC and Federal Government of Nigeria Bond. The information used in
plotting this graph are given in Table 1.

Substituting the partial derivatives (34)–(39) and the value functions (40)–(43) into (30), we have the following partial
differential equation (PDE):
ht (t , r) + 1ψ hr (t , r) + [φ r + πα + ky(β − φ r) − (1 − φ r)c(t , r)
−µI + σZ σZ′ − (π Σ ′ + kyσY )σZ′ − ρY kyσY4 σZ′ ](1 − γ )
1
+ 1ηη′ 1′ [h2r (t , r) + hrr (t , r)] + x−1 (π Σ ′ + kyσY − σZ )η′ 1′ (1 − γ )hr (t , r)
2
1 1
− (π Σ ′ + kyσY − σZ )(π Σ + kyσY − σZ )′ γ (1 − γ ) − k2 y2 σY24 γ (1 − γ )
2 2
∫ ∞ ∫ ∞ (44)
+ [eh(t ,1Jr1 ) − 1]λr νr (dJr ) + λY [(1 + kyJY4 )1−γ − 1]νY (dJY )
−1 −1
∫ ∞
+ [(1 + ((1 − J1 )(π Jm′ + kyJY − JZ )))1−γ − 1]λν (dJ)
−1
(1 − γ )c(t , r)1−δ
+ − ξ = 0.
(1 − δ )x1−γ

For the explicit solution of (44), see Appendix.


But, the PDE (44) depends on the state variable x. To eliminate it, the following interesting result must hold.

Definition 6. The contribution rate of a PPM at time t is given as

(σZ (t , r) − π ∗ (t , r)Σ (t , r)′ )σY−1 (t , r)


k(t , r) = . (45)
y(t , r)
Definition 6 tells us that the contribution of a PPM depends on the inflation risk, optimal investment in both index bonds
and stocks, and income risks at time t. It implies that PPM will want to know all of this at a particular time before he or
she decides on what to contribute into the scheme at time t. This confirms our earlier assumption that the contribution rate
depends on time, interest rate and other market forces.
Fig. 3 shows the contribution rate of a PPM for every value of portfolio in Federal Government of Nigeria Bond and First
Bank of Nigeria PLC.

7. Optimal policies

To be more specific, consider for simplicity a model in which the common jumps are identically distributed, that is,
νq (dzq ) = ν (dz) for all q = 1, . . . , Mr , νk (dzk ) = ν (dz) for all k = 1, . . . , MI and νl (dzl ) = ν (dz) for all l = 1, . . . , M . We
define m̄λ = (Mr λr , MI λI , M λS )′ such that λr is the arrival rate of jump for interest rate with Mr of jumps, λI is the arrival
rate of jump for inflation with MI number of jumps and λS the arrival rate of jump for stocks with M number of jumps.
242 C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252

Definition 7. The optimal consumption rate at time t is defined as


c ∗ (t , r) = arg max gc (t , r),
c(t ,r)

(1−γ )c(t ,r)1−δ


where gc (t , r) = −(1 − φ r)c(t , r) + (1−δ )x1−γ
= 0.

Definition 8. The optimal policy for the portfolio weights π (t , r) is defined as

π ∗ (t , r) = arg max g(π (t , r)), (46)


{π (t ,r)}

where the functions


Σ ′ η′ 1′ hr (t , r)
g(π , r) = π α − π Σ ′ σZ′ + π − xπγ Σ ′ (π Σ ′ + kyσY − σZ )′
x (47)
+ G(((1 − J1 )(π Jm′ + kyJY − JZ ))m̄λ) = 0,
∫∞
and G(((1 − J1 )(π Jm

+ kyJY − JZ ))m̄λ) = − −1
[1 − (1 + ((1 − J1 )(π Jm′ + kyJY − JZ )))1−γ ]λν (dJ).
Since the objective function g is time dependent, so it follows that any optimal solution will be time dependent.
Furthermore, the objective function is state dependent, so any optimal solution will also be state dependent. In other words,

any optimal π ∗ (t , r) will not be time invariant that is, it will be dependent on time and state. Finally, the objective function
g is strictly concave and hence always has a unique maximizer.

7.1. Optimal consumption

Proposition 3. Let c ∗ (t , r) be the optimal consumption rate, then


1
c ∗ (t , r) = [(1 − φ (t)r(t))x(t)−γ eh(t ,r) ]− δ . (48)
( ∂ V )−1 1 ∂L
Proof. As to the optimal consumption policy, with ∂c
(ỹ) = ỹ− δ and ∂x
= x−γ eh(t ,r) , we have
1
c ∗ (t , r) = [(1 − φ (t)r(t))x(t)−γ eh(t ,r) ]− δ . □ (49)
It then follows that the terminal consumption (i.e., when t = T ) will be
1
c ∗ (T , r) = [(1 − φ (T )r(T ))x(T )−γ ]− δ , (50)
since h(T , r) = 0 as we shall see later in this work. At t = 0, we have
1
c ∗ (0, r) = [(1 − φ (0)r0 )x̄0 eh(0,r) ]− δ .
−γ
(51)

7.1.1. Sensitivity analysis of c ∗ (T , r)


We now carry out the sensitivity analysis of c ∗ (T , r) by finding its partial derivatives with respect to γ and δ :
1
∂ c ∗ (T , r) [x(T )−γ (1 − φ (T )r(T ))]− δ log(x(T ))
= . (52)
∂γ δ
1
∂ c ∗ (T , r) [x(T )−γ (1 − φ (T )r(T ))]− δ log(x(T ))−γ (1 − φ (T )r(T ))
= . (53)
∂δ δ2
From (52), we have that
1
∂ c ∗ (T , r) (1 − φ (T )r(T ))− δ log(x(T )) ∂ c ∗ (T , r)
(i) lim = ; (ii) lim = 0. (54)
γ −→0 ∂γ δ γ −→∞ ∂γ
(54)(i) shows that if individual investor’s relative risk averse coefficient is zero, individual investor will still be willing to
consume even above her wealth level, provided all other parameters remain fixed. (54)(ii) tells us that if individual investor’s
relative risk averse coefficient is infinite, there will be no consumption at that level. That is, as individual investor’s relative
risk averse coefficient increases, rate of consumption decreases and vice versa.
Similarly,
∂ c ∗ (T , r) ∂ c ∗ (T , r)
(i) lim = +∞; (ii) lim = 1. (55)
δ−→0 ∂δ δ−→∞ ∂δ
(55)(i) tells us that as individual investor’s preference rate of consumption decreases, rate of consumption increases and
vice versa. (55)(ii) shows that no matter how large individual investor’s preference rate of consumption will be, rate of
consumption will be bounded by unit. Observe in Fig. 4 that the eight graphs behave in a similarly manner.
C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252 243

Fig. 4. The optimal final consumption level given the value of the final wealth and interest rate, where γ = 0.2, δ = 0.25 and φB = (0.1, 0.1, 0.11, 0.11),
φS = (0.1, 0.12, 0.16, 0.2).

7.1.2. The transversality condition


We now check that the transversality condition holds. By substituting in the optimizers X̄ ∗ (t , r) and c ∗ (t , r) into (27), and
then taking expectations of both sides, we find
[∫ ∞ ]
lim E [U(t , X̄ , r)] = lim E

e −ξ s
V (c (s, r))ds = 0.

(56)
t −→∞ t −→∞ t

7.2. Optimal portfolio strategies

In this subsection, we consider the optimal investment strategies for a plan member at time t.

Proposition 4. The optimal investment strategies of the investor are


Jm m̄λ(ΣΣ ′ )−1 Jm

m̄λ
( )
′∗ 1
π = 1− ((ΣΣ ′ )−1 α + (1 − γ )(Σ −1 σZ )′ )
γ Σ −1 m̄λ(Σ −1 m̄λ)′
Jm m̄λ(ΣΣ ′ )−1 Jm

m̄λ
( )
1 hr
+ 1− ((Σ −1 η)′ + γ ky(Σ −1 σY )′ ) (57)
γ Σ m̄λ(Σ m̄λ)
−1 − 1 ′ x
(kyJY − JZ )m̄λ(ΣΣ ′ )−1 (Jm′
m̄λ) a(ΣΣ ′ )−1 (Jm′
m̄λ)
+ + .
Σ −1 m̄λ(Σ −1 m̄λ)′ (1 − J1 )Σ −1 m̄λ(Σ −1 m̄λ)′

Proof. We commence by obtaining the first-order condition. The first-order condition for π is obtained as
Σ ′ η′ 1′ hr (t , r)
α − Σ ′ σZ′ + − γ Σ ′ (π Σ ′ + kyσY − σZ )′
x (58)
+ (1 − J1 )Jm′ m̄λĠ(((1 − J1 )(π Jm′ + kyJY − JZ ))m̄λ) = 0.
We now define the scalar a = ((1 − J1 )(π Jm

+ kyJY − JZ ))m̄λ. Then substituting
[ ]
a
− γ Σ′ − (kyJY − JZ )m̄λ (m̄λ)−1 (Jm′ )−1 Σ = −γ Σ ′ π ′ Σ
1 − J1
in (58), we have that a must satisfy the following:
1 1
− (1 − J1 )((ΣΣ ′ )−1 α )(Jm m̄λ) + (1 − J1 )(Σ −1 σZ )′ (Jm′ m̄λ)
γ γ
1 h r (t , r)
− (1 − J1 )(Σ −1 η)′ (Jm′ m̄λ) + a + (1 − J1 )(kyJY − JZ )m̄λ
γ x (59)
− (1 − J1 )ky(Σ −1 σY )′ (Jm′ m̄λ) + (1 − J1 )(Σ −1 σZ )′ (Jm′ m̄λ)
1
− (1 − J1 )2 Σ −1 m̄λ(Σ −1 m̄λ)′ Ġ(a) = 0,
γ
244 C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252

then use (58) to solve for π :

′∗ 1 1−γ 1 hr
π = (ΣΣ ′ )−1 α − (Σ −1 σZ )′ + (Σ −1 η)′
γ γ γ x
(60)
1
− ky(Σ −1
σY ) + ′
(ΣΣ ) ′ −1
(1 − J1 )Jm m̄λĠ(a).

γ
From (59), we have that

((ΣΣ ′ )−1 α )(Jm



m̄λ) (1 − γ )(Σ −1 σZ )′ (Jm

m̄λ)
Ġ(a) = − +
(1 − J1 )Σ −1 m̄λ(Σ −1 m̄λ)′ (1 − J1 )Σ −1 m̄λ(Σ −1 m̄λ)′
(Σ −1
η) (Jm m̄λ)hr (t , r)
′ ′
γa
− + (61)
x(1 − J1 )Σ −1 m̄λ(Σ −1 m̄λ)′ (1 − J1 )2 Σ −1 m̄λ(Σ −1 m̄λ)′
γ (kyJY − JZ )m̄λ γ ky(Σ −1 σY )′ (Jm′ m̄λ)
+ + .
(1 − J1 )Σ m̄λ(Σ m̄λ)
− 1 −1 ′ (1 − J1 )Σ −1 m̄λ(Σ −1 m̄λ)′

Substituting (61) into (60), we have the following


) {(ΣΣ ′ )−1 α + (1 − γ )(Σ −1 σ )′ }
Jm m̄λ(ΣΣ ′ )−1 Jm

m̄λ
(
′∗ 1 Z
π = 1− ′ hr
γ Σ −1 m̄λ(Σ −1 m̄λ)′ +(Σ −1
η)
+ γ ky(Σ −1
σY )′
x (62)
(kyJY − JZ )m̄λ(ΣΣ ′ )−1 (Jm

m̄λ) a(ΣΣ ) (Jm m̄λ)
′ −1 ′
+ + . □
Σ −1 m̄λ(Σ −1 m̄λ)′ (1 − J1 )Σ −1 m̄λ(Σ −1 m̄λ)′

Observe that (62) is made up of four parts. The first part is the Merton’s portfolio with jump, the second part is the
jump risks hedging portfolio, the third part is a hedging portfolio value of the two classes of risky assets and the last part
is proportional to a, which is a control parameter that measures how much of this portfolio value the investor is willing to
take. We now give the break down and comprehensive interpretations of every component of (62):

Jm m̄λ(ΣΣ ′ )−1 Jm

m̄λ 1−γ Jm m̄λ(ΣΣ ′ )−1 Jm

m̄λ
( ) ( )
′∗ 1
π = (ΣΣ ′ )−1 α 1− + (Σ −1 σZ )′ 1−
γ Σ −1 m̄λ(Σ −1 m̄λ)′ γ Σ −1 m̄λ(Σ −1 m̄λ)′
     
δ1 δ2

Jm m̄λ(ΣΣ ′ )−1 Jm

m̄λ Jm m̄λ(ΣΣ ′ )−1 Jm

m̄λ
( ) ( )
1 hr
+ (Σ −1 η)′ 1− + ky(Σ −1
σY )

1−
γ x Σ −1 m̄λ(Σ −1 m̄λ)′ Σ −1 m̄λ(Σ −1 m̄λ)′ (63)
     
δ3 δ4

kyJY m̄λ(ΣΣ ) ′ −1
(Jm m̄λ)

JZ m̄λ(ΣΣ ) ′ −1
(Jm m̄λ)

a(ΣΣ ) ′ −1 ′
(Jm m̄λ)
+ − + .
Σ −1 m̄λ(Σ −1 m̄λ)′ Σ −1 m̄λ(Σ −1 m̄λ)′ (1 − J1 )Σ −1 m̄λ(Σ −1 m̄λ)′
     
δ5 δ6

It shows that it is optimal to invest the PPM’s contributions in a portfolio that comprises of seven components:

1. a speculative portfolio δ1 proportional to the market price of risk corresponding to the two risky assets, stocks and
index bonds through the relative risk averse index γ1 ,
2. an inflation diffusion risk hedging portfolio δ2 proportional to the diffusion term of the inflation index through the
1−γ
relative risk averse index γ ,
3. an interest rate diffusion risk hedging portfolio δ3 proportional to the diffusion term of the interest rate and inversely
proportional to the optimal wealth process through the cross derivative of function h with respect to the interest rate
r and the relative risk averse index γ1 ,
4. an income diffusion risk (or a preference-free) hedging component δ4 proportional to the diffusion term of the income
risks and the contribution process,
5. an income jump risk hedging component δ5 proportional to the jump term of the income risks and the contribution
process,
6. an inflation jump risk hedging portfolio δ6 proportional to the jump term of the inflation index,
m a(ΣΣ ′ )−1 (J ′ m̄λ)
7. a hedging portfolio value (1−J )Σ −1 m̄λ(Σ −1 m̄λ)′ of the two classes of risky assets and a a control parameter that measures
1
how much of this portfolio value the investor is willing to take,
8. risk-free funds holding the riskless assets only.
C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252 245

Corollary 1. Suppose that the portfolio is free from interest rate risk i.e., σr (t) ≡ 0 or η(t) ≡ (0, 0, 0), then (63) becomes

Jm m̄λ(ΣΣ ′ )−1 Jm

m̄λ
( )
′∗ 1
π = 1− [(ΣΣ ′ )−1 α + (1 − γ )(Σ −1 σZ )′ ]
γ
Σ −1 m̄λ(Σ −1 m̄λ)′
(64)
(kyJY − JZ )m̄λ(ΣΣ ′ )−1 (Jm′
m̄λ) a(ΣΣ ′ )−1 (Jm

m̄λ)
+ + .
Σ m̄λ(Σ m̄λ)
− 1 − 1 ′ (1 − J1 )Σ m̄λ(Σ −1 m̄λ)′
−1

The following proposition gives an explicit solution of the parameter a.

Proposition 5. Suppose that the Lévy measure generates asymmetric positive and negative jumps, then
1
[ ]
1 (22 3 (3 + A2 + 3B(λ− − λ+ ))) 1
a= −2A + 1
+ (4Φ ) 3 . (65)
6 Φ3

Proof. The Lévy measure is allowed to generate asymmetric positive and negative jumps. By following [35] approach, we
consider a Lévy measure such that

λ+ dz /z, if z ∈ (0, 1]
{
ν (dz) = (66)
− λ− dz /z, if z ∈ [−1, 0)
λ λ
where λ+ = m̄S λ+ and λ− = m̄S λ− with λS + > 0 and λS − > 0. Setting γ = 2, we have that G(a) = −λ+ log(1 + a) −λ− log(1 − a)
and (59) becomes a cubic equation in a:

f ′ (a) = a + A + B[−λ+ (1 + a)−1 + λ− (1 − a)−1 ] = 0, (67)

with solvency constraint |a| < 1, where A = − 21 (1 − J1 )((ΣΣ ′ )−1 α )(Jm m̄λ) + 1
2
(1 − J1 )(Σ −1 σZ )′ (Jm′ m̄λ) − 1
2
(1 −
h (t ,r)
J1 )(Σ η) (Jm m̄λ
−1 ′ ′
+ (1 − J1 )(kyJY − JZ )m̄λ − (1 − J1 )ky(Σ σY ) (Jm m̄λ) + (1 − J1 )(Σ σZ ) (Jm m̄λ) and B =
) rx −1 ′ ′ −1 ′ ′ 1
2
(1 −
J1 )2 Σ −1 m̄λ(Σ −1 m̄λ)′ . The optimal solution to (59) is indeed solvable in closed form under the solvency constraint:
1
[ ]
1 (22 3 (3 + A2 + 3B(λ− − λ+ ))) 1
a= −2A + 1
+ (4Φ ) 3 , (68)
6 Φ3
where Φ = (−2A3 + 27B(λ− + λ+ ) + 9A(2 + B(λ+ − λ− )) +

(−4(3 + A2 + 3B(λ− − λ+ ))3 + ((−3 + A)(2A(3 + A) + 9Bλ− ) − 9(3 + A)Bλ+ )2 )).



We observed that if
− 21 ((ΣΣ ′ )−1 α )(Jm m̄λ) + 12 (Σ −1 σZ )′ (Jm′ m̄λ) − 21 (Σ −1 η)′ (Jm′ m̄λ) hr (tx ,r) + (kyJY − JZ )m̄λ− ky(Σ −1 σY )′ (Jm′ m̄λ) + (Σ −1 σZ )′ (Jm′ m̄λ)
= − 21 (1 − J1 )(λ− − λ+ )Σ −1 m̄λ(Σ −1 m̄λ)′ , it implies that the investor has no exposure to jump risk. This is due to the fact
that jumps in one investment are used to hedge or offset jump risks in another investment.

8. Empirical results

In this section, we give the numerical results of our problem. One of the major challenges faced by fund managers is what
fraction of wealth should be invested in one portfolio or the other. Nigeria Pension Reform Act stipulates that more of the
investment portfolio should be invested in bonds and fixed deposits and small past of the portfolio should be invested in
stock. In this paper, we allow 75% of the investment fund to be invested in bonds (both risky and riskless bonds) and the
remaining 25% in stocks (including bank deposit accounts) and each of the investment class is not equally weighted. Inflation
data from the International Monetary Fund from 1980 to June 14, 2016 were collected and analyzed, data for four banks,
which include First Bank of Nigeria PLC (FBN), Guarantee Trust Bank of Nigeria PLC (GTB), United Bank of Africa PLC (UBA),
Zenith Bank of Nigeria PLC (Zenith) as well as four government of Nigeria bonds, referred to as: BOND 1, BOND 2, BOND 3 and
BOND 4 were collected from Nigerian Stock Exchange, Sapele Road Benin City, Edo State, Nigeria from March 2006 to June
16, 2016. In order to obtain the expected growth rate and salary risks of a PPM, data from the salary chart of Academic Staff
of Nigeria University, 2009 was obtained from Bursary Department, University of Benin, Benin City, Edo State, Nigeria. The
data were analyzed using the SPSS package. Table 1 shows the mean and standard deviation (SD) of the bonds and stocks
and Table 2 shows the jump sizes. We assume that for every investment in risky bond there are corresponding riskless
bonds, risky stocks and bank fixed deposit accounts. This to a great extent will diversify risks associated with the investment
(which is our hedging strategy). Note that in this section, we assume that the portfolio strategies are free from interest rate
risks.
246 C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252

Table 1
Mean and SD of bonds and stocks.
Bond Mean SD Stock Mean SD
BOND 1 12.0733 1.64051 FBN 28.9614 13.36671
BOND 2 14.4008 1.45157 GTB 19.0368 8.34097
BOND 3 12.5142 1.79795 UBA 21.7460 15.58661
BOND 4 12.8608 1.37605 Zenith 25.1496 14.43704
Interest rate: FBN 5%,GTB 5.5%, UBA 6% and Zenith 6.5%.

Table 2
Jump size of bonds and stocks.
Bond Jumpsize Jump rate Stock Jump size Jump rate
BOND 1 4.35 0.189130 FBN 06.68 0.058087
BOND 2 5.84 0.253913 GTB 11.70 0.146250
BOND 3 5.91 0.256957 UBA 24.48 0.249796
BOND 4 4.69 0.203913 Zenith 11.80 0.119192

For inflation:

• Inflation: SD = 16.96334%, expected inflation rate= 19.2417%,


• φB = (0.1875, 0.1875, 0.1875, 0.1875), φS = (0.0625, 0.0625, 0.0625, 0.0625),
• where φ = (φB , φS ), φB and φS are weights for bonds and stocks respectively,
• ρY = (0.010, 0.010, 0.010), inflation jumpsize = 5.4,
• number of jump MI = 35, λ1 = 0.154286,
• JY4 = 0.0102 is the expected rate of income jump.

The covariance symmetric matrix σS of the four banks is obtained as


0.108 −0.131 −0.069 −0.065
⎛ ⎞
⎜−0.131 1.035 −0.316 −0.301⎟
σS = ⎝ . (69)
−0.069 −0.316 0.455 −0.256⎠
−0.065 −0.301 −0.256 0.503
The coefficient of correlation matrix is obtained as follows:
1.000 0.390 −0.309 −0.279
⎛ ⎞
⎜ 0.390 1.000 −0.461 −0.417⎟
ρ=⎝ . (70)
−0.309 −0.461 1.000 −0.536⎠
−0.279 −0.417 −0.536 1.000
It then follows that
0.108000 0.051090 0.021321 0.018135
⎛ ⎞
⎜0.051090 1.035000 0.145676 0.125517⎟
σS ρS = ⎝ . (71)
0.021321 0.145676 0.455000 0.137216⎠
0.018135 0.125517 0.137216 0.503000
The matrix σS ρS is not vanishing since its determinant is 0.0214538 which is not equal to zero. Therefore,
0.535070 −0.413097 −0.263651 −0.168769
⎛ ⎞
⎜−0.413097 1.044650 −0.262586 −0.174154⎟
(σS ρS )−1 =⎝ . (72)
−0.263651 −0.262586 2.475230 −0.600201⎠
−0.168769 −0.174154 −0.600201 2.201350
The covariance matrix for the bonds is
0.077 0.012 −0.072 −0.007
⎛ ⎞
⎜ 0.012 0.184 −0.094 −0.083⎟
σB = ⎝ . (73)
−0.072 0.094 0.219 −0.100⎠
−0.007 −0.083 −0.100 0.230
The coefficient of correlation matrix for bonds is obtained as follows:
1.000 0.103 −0.557 −0.055
⎛ ⎞
⎜ 0.103 1.000 −0.469 −0.404⎟
ρB = ⎝ . (74)
−0.557 −0.469 1.000 −0.450⎠
−0.055 −0.404 −0.450 1.000
C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252 247

Fig. 5. Optimal terminal consumption versus optimal wealth for γ = 0.5, δ = 0.6.

The values for σB ρB are obtained as


0.077000 0.001236 0.040104 0.000385
⎛ ⎞
⎜0.001236 0.084000 0.044086 0.033532⎟
σB ρB = ⎝ . (75)
0.040104 0.044086 0.217000 0.045000⎠
0.000385 0.033532 0.045000 0.230000
The matrix σB ρB is not vanishing and has the determinant of 0.026787. It implies that the matrix σB ρB is invertible.
Fig. 5 shows the relationship between optimal consumption and optimal real wealth at the terminal time. We observe
that as wealth increases, consumption increases alongside.
Table 3 shows the index bonds and stocks portfolio values, portfolio values in riskless bonds and in bank fixed deposits
accounts, as coefficient of investors relative risk averse varies, for c = 0.1, k = 0.5, y = 0.4 and a = 5 remain fixed. We
found that:

1. the higher the value of γ the lower the portfolio values in Bonds 1, 2, 3 and 4, FBN, GTB, UBA and Zenith and higher the
portfolios in the riskless assets, and vice versa. This implies that as the investor’s coefficient of CRRA increases more
of the fund should remain in the riskless assets and less fund should be in the risky assets
2. the higher the value of γ , for the class of riskless Bond: RB 3 yields the highest portfolio values, followed by RB 2, then
RB 1 and RB 4 have the least values. It implies that if investors must invest in risky bond, Bond 4 must be considered
first, followed by Bond 1 and so on. For the class of bank deposit: More fund should be fixed with FBN, followed by
UBA, then Zenith, last GTB. Looking at the poor returns from stocks as γ increases, GTB remains viable up to the point
γ = 0.5. We therefore remark that if investors must invest in stocks, GTB should be considered first
3. individual investors with lower coefficient of risk aversion should consider investing in FBN, GTB, UBA, Zenith, Bonds
1, 2, 3 and 4, otherwise their investment should remain in the riskless assets while individual investors with high
coefficient of risk averse should consider investing in riskless assets otherwise the investment should remain in the
risky assets
4. overall, RB 3 and GTB are the most promising assets, hence more of the PPM’s contributions should be invested into
these assets to ensure maximum returns at retirement
5. those who must invest in FBN must ensure that their investment remain in FBN fixed deposits account.

Table 4 shows change in portfolio values in both the riskless and risky assets as a (a control parameter that measures how
much of this portfolio value the investor is willing to take) changes for c = 10%, k = 50%, y = 20% and γ = 3.0 remain
fixed. We found the following:

1. as the value of a increases, the portfolio values in the risky assets increase and vice versa
2. as the value of a increases, investment in the riskless assets decreases and vice versa.

Table 5 shows the portfolio values in risky and riskless assets as the contribution rate changes for y = 40%, γ = 3.0,
γ = 3.0 remain fixed. It is observed that as contribution rate increases, the portfolio values in the risky assets increase and
the values of the riskless assets decrease, and vice versa. It was further observed that the portfolios in the riskless bonds
remain positive as contribution rate increases.
248 C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252

Table 3
Portfolio value for different values of γ .
γ Bond πB Stock πS RB πB0 FD πS0
0.5 BOND 1 5.5862 FBN 2.5599 RB 1 −5.4089 FBN −2.5122
BOND 2 5.8248 GTB 0.8064 RB 2 −5.6556 GTB −0.7479
BOND 3 7.0197 UBA 1.0468 RB 3 −6.8595 UBA −0.9838
BOND 4 5.0557 Zenith 1.1423 RB 4 −4.8874 Zenith −1.0865
1.0 BOND 1 0.9281 FBN 0.7740 RB 1 −0.7508 FBN −0.7263
BOND 2 0.9596 GTB 0.3600 RB 2 −0.7904 GTB −0.3015
BOND 3 1.1687 UBA 0.4326 RB 3 −1.0085 UBA −0.3696
BOND 4 0.8517 Zenith 0.4894 RB 4 −0.6834 Zenith −0.4336
2.0 BOND 1 −1.4010 FBN −0.1189 RB 1 1.5783 FBN 0.1666
BOND 2 −1.4730 GTB 0.1368 RB 2 1.6422 GTB −0.0783
BOND 3 −1.7568 UBA 0.1254 RB 3 1.9170 UBA −0.0624
BOND 4 −1.2502 Zenith 0.1629 RB 4 1.4185 Zenith −0.1071
3.0 BOND 1 −2.1774 FBN −0.4166 RB 1 2.3547 FBN 0.4643
BOND 2 −2.2839 GTB 0.0624 RB 2 2.4531 GTB −0.0039
BOND 3 −2.7319 UBA 0.0230 RB 3 2.8921 UBA 0.0400
BOND 4 −1.9509 Zenith 0.0541 RB 4 2.1192 Zenith 0.0017
4.0 BOND 1 −2.5655 FBN −0.5654 RB 1 2.7428 FBN 0.6131
BOND 2 −2.6893 GTB 0.0252 RB 2 2.8585 GTB 0.0333
BOND 3 −3.2195 UBA −0.0282 RB 3 3.3797 UBA 0.0912
BOND 4 −2.3012 Zenith −0.0003 RB 4 2.4695 Zenith 0.0561
5.0 BOND 1 −2.7985 FBN −0.6547 RB 1 2.9758 FBN 0.7024
BOND 2 −2.9326 GTB 0.0029 RB 2 3.1018 GTB 0.0556
BOND 3 −3.5121 UBA −0.0589 RB 3 3.6723 UBA 0.1219
BOND 4 −2.5114 Zenith −0.0329 RB 4 2.6797 Zenith 0.0887
6.0 BOND 1 −2.9537 FBN −0.7142 RB 1 3.1310 FBN 0.7619
BOND 2 −3.0947 GTB −0.0120 RB 2 3.2639 GTB 0.0705
BOND 3 −3.7071 UBA −0.0793 RB 3 3.8673 UBA 0.1423
BOND 4 −2.6516 Zenith −0.0547 RB 4 2.8199 Zenith 0.1105
7.0 BOND 1 −3.0646 FBN −0.7567 RB 1 3.2419 FBN 0.8044
BOND 2 −3.2106 GTB −0.0226 RB 2 3.3798 GTB 0.0811
BOND 3 −3.8464 UBA −0.0940 RB 3 4.0066 UBA 0.1570
BOND 4 −2.7517 Zenith −0.0702 RB 4 2.9200 Zenith 0.1260
11.0 BOND 1 −3.3066 FBN −0.8495 RB 1 3.4839 FBN 0.8972
BOND 2 −3.4633 GTB −0.0458 RB 2 3.6325 GTB 0.1043
BOND 3 −4.1504 UBA −0.1259 RB 3 4.3106 UBA 0.1889
BOND 4 −2.9701 Zenith −0.1041 RB 4 3.1384 Zenith 0.1599
RB stands for Riskless Bond and FD stands for Fixed Deposit. We take consumption to be 10%, k = 50%, y = 40% and a = 5.

Table 6 shows the portfolio values for varying value of the income rate for k = 0.5, γ = 3.0, a = 5 remain fixed.

1. It is observed that an increase in income can lead to an increase in the portfolio risks and vice versa. We found that
a 1% increase in income rate can lead to 0.4665% increase in the portfolio value in Bond 1 and 0.549% decrease in
riskless Bond 1. Also, a 1% increase in income rate can lead to 0.49% increase in portfolio value in Bond 2 and 0.51%
decrease in the portfolio value in riskless bond 2. Bonds 3 and 4 increase by 0.59% and 0.42% as income rate increases
by 1% and the corresponding riskless assets decrease respectively by 0.60% and 0.44%.
2. Similarly, we found that a 1% increase in income rate can lead to 0.1265% increase in the portfolio value in FBN and
0.1326% decrease in the corresponding fixed deposit. Again, 1% increase in income rate can lead to 0.042% increase in
portfolio value in GTB and 0.0170% decrease in the portfolio value in the corresponding fixed deposit. UBA and Zenith
increase by 0.0225% and 0.082% respectively as income rate increases by 1% and the corresponding riskless assets
decrease by 0.0295% and 0.106%.
3. We found that inflation risk and income risk contribute tremendously to the reduction of the portfolio values in the
riskless assets.

Table 7 is made up of three compartments and it shows the optimal terminal consumption of the investor at varying values
of optimal terminal real wealth (X ∗ (T )), investor’s relative risk averse coefficient γ and preference rate of consumption δ
provided k = 0.15, y = 0.4, a = 0.2 are held fixed. The following were found:

1. from the first compartment, as optimal wealth increases, consumption increases alongside;
2. from the second compartment, as γ increases, consumption increases drastically;
3. from compartment three, as δ increases, consumption decreases and vice versa;
4. as the risk averse coefficient increases, consumption increases. This shows that as PPM’s taste to consume decreases
optimal consumption decreases and vice versa.
C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252 249

Table 4
Portfolio value for different values of a.
a Bond $πB Stock πS RB π0B FD π0S
5 BOND 1 −2.1751 FBN −0.4165 RB 1 2.3524 FBN 0.4642
BOND 2 −2.2830 GTB 0.0627 RB 2 2.4522 GTB −0.0042
BOND 3 −2.7311 UBA 0.0239 RB 3 2.8913 UBA 0.0391
BOND 4 −1.9504 Zenith 0.0543 RB 4 2.1187 Zenith 0.0015
10 BOND 1 −2.1728 FBN −0.4165 RB 1 2.3501 FBN 0.4642
BOND 2 −2.2821 GTB 0.0630 RB 2 2.4513 GTB −0.0045
BOND 3 −2.7302 UBA 0.0247 RB 3 2.8904 UBA 0.0383
BOND 4 −1.9498 Zenith 0.0545 RB 4 2.1181 Zenith 0.0013
15 BOND 1 −2.1705 FBN −0.4164 RB 1 2.3478 FBN 0.4641
BOND 2 −2.2812 GTB 0.0633 RB 2 2.4504 GTB −0.0048
BOND 3 −2.7292 UBA 0.0256 RB 3 2.8894 UBA 0.0374
BOND 4 −1.9492 Zenith 0.0547 RB 4 2.1175 Zenith 0.0011
20 BOND 1 −2.1681 FBN −0.4164 RB 1 2.3454 FBN 0.4641
BOND 2 −2.2803 GTB 0.0636 RB 2 2.4495 GTB −0.0051
BOND 3 −2.7283 UBA 0.0265 RB 3 2.8885 UBA 0.0365
BOND 4 −1.9486 Zenith 0.0549 RB 4 2.1169 Zenith 0.0009
25 BOND 1 −2.1658 FBN −0.4163 RB 1 2.3431 FBN 0.4640
BOND 2 −2.2794 GTB 0.0639 RB 2 2.4486 GTB −0.0054
BOND 3 −2.7274 UBA 0.0273 RB 3 2.8876 UBA 0.0357
BOND 4 −1.9481 Zenith 0.0551 RB 4 2.1164 Zenith 0.0007
30 BOND 1 −2.1635 FBN −0.4163 RB 1 2.3408 FBN 0.4640
BOND 2 −2.2786 GTB 0.0642 RB 2 2.4478 GTB −0.0057
BOND 3 −2.7265 UBA 0.0282 RB 3 2.8867 UBA 0.0348
BOND 4 −1.9475 Zenith 0.0553 RB 4 2.1158 Zenith 0.0005
35 BOND 1 −2.1612 FBN −0.4163 RB 1 2.3385 FBN 0.4640
BOND 2 −2.2777 GTB 0.0645 RB 2 2.4469 GTB −0.0060
BOND 3 −2.7256 UBA 0.0290 RB 3 2.8858 UBA 0.0340
BOND 4 −1.9469 Zenith 0.0555 RB 4 2.1152 Zenith 0.0003
We take consumption to be 10%, k = 50%, y = 20% and γ = 3.0.

Table 5
Portfolio value for different values of k.
k Bond πB Stock πS RB π0B FD π0S
0.05 BOND 1 −3.0144 FBN −0.6442 RB 1 3.2272 FBN 0.7014
BOND 2 −3.1618 GTB 0.0433 RB 2 3.3648 GTB 0.0269
BOND 3 −3.7846 UBA −0.0170 RB 3 3.9768 UBA 0.0926
BOND 4 −2.7046 Zenith 0.0175 RB 4 2.9066 Zenith 0.0494
0.1 BOND 1 −2.9212 FBN −0.6189 RB 1 3.1300 FBN 0.6751
BOND 2 −3.0641 GTB 0.0454 RB 2 3.2634 GTB 0.0235
BOND 3 −3.6675 UBA −0.0125 RB 3 3.8562 UBA 0.0867
BOND 4 −2.6208 Zenith 0.0216 RB 4 2.8190 Zenith 0.0441
0.2 BOND 1 −2.7347 FBN −0.5683 RB 1 2.9356 FBN 0.6224
BOND 2 −2.8689 GTB 0.0498 RB 2 3.0606 GTB 0.0165
BOND 3 −3.4334 UBA −0.0034 RB 3 3.6150 UBA 0.0748
BOND 4 −2.4532 Zenith 0.0298 RB 4 2.6439 Zenith 0.0334
0.3 BOND 1 −2.5481 FBN −0.5177 RB 1 2.7412 FBN 0.5697
BOND 2 −2.6736 GTB 0.0541 RB 2 2.8578 GTB 0.0096
BOND 3 −3.1993 UBA 0.0057 RB 3 3.3737 UBA 0.0629
BOND 4 −2.2856 Zenith 0.0380 RB 4 2.4688 Zenith 0.0228
0.4 BOND 1 −2.3616 FBN −0.4671 RB 1 2.5468 FBN 0.5169
BOND 2 −2.4783 GTB 0.0584 RB 2 2.6550 GTB 0.0027
BOND 3 −2.9652 UBA 0.0148 RB 3 3.1325 UBA 0.0510
BOND 4 −2.1180 Zenith 0.0461 RB 4 2.2938 Zenith 0.0121
0.5 BOND 1 −2.1751 FBN −0.4165 RB 1 2.3524 FBN 0.4642
BOND 2 −2.2830 GTB 0.0627 RB 2 2.4522 GTB −0.0042
BOND 3 −2.7311 UBA 0.0239 RB 3 2.8913 UBA 0.0391
BOND 4 −1.9504 Zenith 0.0543 RB 4 2.1187 Zenith 0.0015
We take consumption to be 10%, a = 5, y = 20% and γ = 3.0.

From Tables 3–6, we observe that the portfolio values in the risky bonds remain negative (except when 0 < γ ≤ 1
in Table 3) and portfolio values in the riskless bonds remain promising. It therefore follows that all the proportion of the
investment allocated to the risky bonds should be withdrawn and invested in the riskless bonds most especially when γ > 1.
250 C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252

Table 6
Portfolio value for different values of y.
y Bond πB Stock πS RB π0B FD π0S
0.2 BOND 1 −3.0144 FBN −0.6442 RB 1 3.2272 FBN 0.7014
BOND 2 −3.1618 GTB 0.0433 RB 2 3.3648 GTB 0.0269
BOND 3 −3.7846 UBA −0.0170 RB 3 3.9768 UBA 0.0926
BOND 4 −2.7046 Zenith 0.0175 RB 4 2.9066 Zenith 0.0494
0.4 BOND 1 −2.9212 FBN −0.6189 RB 1 3.1300 FBN 0.6751
BOND 2 −3.0641 GTB 0.0454 RB 2 3.2634 GTB 0.0235
BOND 3 −3.6675 UBA −0.0125 RB 3 3.8562 UBA 0.0867
BOND 4 −2.6208 Zenith 0.0216 RB 4 2.8190 Zenith 0.0441
0.6 BOND 1 −2.8279 FBN −0.5936 RB 1 3.0328 FBN 0.6487
BOND 2 −2.9665 GTB 0.0476 RB 2 3.1620 GTB 0.0200
BOND 3 −3.5505 UBA −0.0079 RB 3 3.7356 UBA 0.0807
BOND 4 −2.5370 Zenith 0.0257 RB 4 2.7315 Zenith 0.0388
0.8 BOND 1 −2.7347 FBN −0.5683 RB 1 2.9356 FBN 0.6224
BOND 2 −2.8689 GTB 0.0498 RB 2 3.0606 GTB 0.0165
BOND 3 −3.4334 UBA −0.0034 RB 3 3.6150 UBA 0.0748
BOND 4 −2.4532 Zenith 0.0298 RB 4 2.6439 Zenith 0.0334
1.0 BOND 1 −2.6414 FBN −0.5430 RB 1 2.8384 FBN 0.5960
BOND 2 −2.7712 GTB 0.0519 RB 2 2.9592 GTB 0.0131
BOND 3 −3.3164 UBA 0.0011 RB 3 3.4944 UBA 0.0689
BOND 4 −2.3694 Zenith 0.0339 RB 4 2.5564 Zenith 0.0281
We take consumption to be 10%, a = 5, k = 10% and γ = 3.0.

Table 7
Optimal terminal consumption.
X ∗ (T ) C ∗ (T , r) γ C ∗ (T , r) δ C ∗ (T , r)
50 26.2384 0.10 4.1651 0.2 1.8064 × 109
100 46.7514 0.40 294.5156 0.4 4.2502 × 104
500 178.7598 0.50 1217.90 0.6 1.2179 × 103
1000 318.5138 0.55 2476.60 0.8 206.1591
5000 1217.90 0.60 5036.10 1.0 71.0169

9. Conclusion

This paper presented theoretical and empirical studies of optimal pension fund in the presence of inflation, interest rate,
income and equity risks. The optimal consumption and portfolio strategies for an investor who faces both diffusion and
jump risks were analyzed. The risky and the riskless assets are classified into two classes: the class of risky assets which
comprises of risky bonds and stocks, and class of riskless assets which is made up of riskless bonds and bank deposits. A
time-consistent proportion of income contributed into the scheme was obtained. The optimal investment with uniform
jumps and optimal consumption are obtained. It was found that the portfolio process is made up of four parts: the first
part is the Merton’s portfolio with jump, the second part is the jump risks hedging portfolio, the third part is a hedging
portfolio value of the two classes of risky assets and the last part is the component that is proportional to a which is a control
parameter that measures how much of this portfolio value the investor is willing to take. We also found that given γ = 2, if
− 21 ((ΣΣ ′ )−1 α )(Jm m̄λ) + 12 (Σ −1 σZ )′ (Jm′ m̄λ) − 21 (Σ −1 η)′ (Jm′ m̄λ) hr (tx ,r) + (kyJY − JZ )m̄λ− ky(Σ −1 σY )′ (Jm′ m̄λ) + (Σ −1 σZ )′ (Jm′ m̄λ) =
− 21 (1 − J1 )(λ− − λ+ )Σ −1 m̄λ(Σ −1 m̄λ)′ , the investor will no longer be exposed to jump risks. In this paper, we found from
the empirical study that

1. inflation and income risks contribute tremendously to the reduction in the investor’s portfolio values in riskless assets
2. as PPM’s taste to consume decreases, optimal consumption decreases and vice versa
3. as the investor’s coefficient of CRRA increases, more of the fund should remain in riskless assets and less fund should
be in the risky assets
4. overall, RB 3 and GTB are the most promising assets, hence more of the PPMs contributions should be invested into
these assets to ensure maximum returns at retirement.

Appendix. The HJB Equation

Here, we provide analytical solution to the highly non-linear partial differential equation (PDE) (i.e., our HJB equation).
In general, there is no existing analytical method to solving a highly nonlinear PDE. But, in this paper, we are going to solve
it by making some assumptions.
C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252 251

Suppose that Definition 6 holds, then our HJB equation (44) becomes

ht (t , r) + 1ψ hr (t , r) + [φ r + πα + (σZ − π Σ ′ )σY−1 (β − φ r) − (1 − φ r)c(t , r)


1
− µI − ρY (σZ − π Σ ′ )σY−1 σY4 σZ′ ](1 − γ ) + 1ηη′ 1′ [h2r (t , r) + hrr (t , r)]
2
∫ ∞
1
[eh(t ,1Jr1 1 ) − 1]ν (dJr )

− ((σZ − π Σ )σY ) σY4 γ (1 − γ ) + λr
′ −1 2 2
2 −1
∫ ∞
−1 1−γ (76)
+ λY [(1 + (σZ − π Σ )σY )JY4 − 1]νY (dJY )
∫ ∞−1
+ [(1 + ((1 − J1 )(π Jm′ + (σZ − π Σ )σY−1 )JY ) − JZ )1−γ − 1]λZ νZ (dJZ )
−1
(1 − γ )c(t , r)1−δ
+ − ξ = 0.
(1 − δ )x1−γ
(76) can be re-expressed as follows:
1
ht (t , r) + 1ψ hr (t , r) + 1ηη′ 1′ [h2r (t , r) + hrr (t , r)] + F (t , π ) = 0, (77)
2
where F (t , π ) = −ξ + [φ r + πα + (σZ − π Σ ′ )σY−1 (β − φ r) − (1 − φ r)c(t , r) − µI − ρY (σZ − π Σ ′ )σY−1 σY4 σZ′ ](1 − γ ) −
((σZ − π Σ ′ )σY−1 )2 σY2 γ (1 − γ ) + λr −1 [eh(t ,1Jr1 1 ) − 1]ν (dJr ) + λY −1 [(1 + (σZ − π Σ )σY−1 )JY − 1]νY (dJY ) + −1 [(1 + ((1 −
1
∫∞ ′ ∫∞ 1−γ ∫∞
2 4 4
(1−γ )c(t ,r)1−δ
J1 )(π Jm

+ (σZ − π Σ )σY−1 )JY ) − JZ )1−γ − 1]λZ νZ (dJZ ) + (1−δ )x1−γ
.
We now set hrr = (hr )2 , so that (77) becomes

ht (t , r) + 1ψ hr (t , r) + 1ηη′ 1′ hrr (t , r) + F (t , π ) = 0. (78)


2
Solving the following ODE hrr = (hr ) , we have a solution of the form:

h(t , r) = A(t) − ln[D(t) + E(t)r ]


{
(79)
h(T , r) = 0,

where A(t), D(t), E(t) ∈ R and D(t) + E(t)r > 0.


Finding the partial derivative of h(t , r) in (79) with respect to t, r and rr, and then substituting into (78), we have the
following ordinary differential equation (ODE):
D′ (t) E ′ (t)r 1ηη′ 1′ E(t)2
A′ (t) − − +
D(t) + E(t)r D(t) + E(t)r (D(t) + E(t)r)2
(80)
1ψ D(t) 1
− + F = 0, E(T ) = , A(T ) = 0, D(T ) = 0.
D(t) + E(t)r r
We now set A′ (t) = 0 and D′ (t) = 0, so that (80) now becomes
E ′ (t)r 1ηη′ 1′ E(t)2 1ψ D(t) 1
− + − F = 0, E(T ) = , (81)
cD + E(t)r (cD + E(t)r)2 cD + E(t)r r
where A(t) = cA and D(t) = cD . Solving (81), we have
T
1ηη′ 1′ E(s)2
∫ ( )
1 1
E(t) = + 1ψ E(s) − − F (cD + E(s)r) ds.
r r t cD + E(s)r
It then follows that
T
1ηη′ 1′ E(s)2
[ ∫ ( ) ]
h(t , r) = cA − ln cD + 1 + 1ψ E(s) − − F (cD + E(s)r) ds . (82)
t cD + E(s)r
We now have that

h(T , r) = cA − ln[cD + 1].

It follows that h(T , r) = 0 if and only if cA = 0 and cD = 0. It therefore follows that


T
1r ψ − 1ηη′ 1′ − Fr 2
[ ∫ ( ) ]
h(t , r) = − ln 1 + E(s)ds .
t r
252 C.I. Nkeki / Journal of Computational and Applied Mathematics 330 (2018) 228–252

References

[1] Inflation, average consumer prices. International Monetary Fund, World Economic Outlook Database, April, 2015.
[2] R.C. Merton, Optimal consumption and portfolio rules in a continuous-time model, J. Econom. Theory 3 (1971) 373–413.
[3] A. Cairns, Some notes on the dynamics and optimal control of stochastic pension fund models in continuous time, Astin Bull. 30 (2000) 19–55.
[4] G. Deelstra, M. Grasselli, P. Koehl, Optimal investment strategies in a CIR framework, J. Appl. Probab. 37 (2000) 936–946.
[5] J.F. Boulier, S.J. Huang, G. Taillard, Optimal management under stochastic interest rates: The case of a protected defined contribution pension fund,
Insurance Math. Econom. 28 (2001) 173–189.
[6] B.A. Jensen, C. Sørensen, Paying for minimum interest guarantees. who should compensate who? Eur. Financ. Manag. 7 (2001) 183–211.
[7] O.E. Vasicek, An equilibrium characterization of the term structure, J. Financ. Econom. 5 (1977) 177–188.
[8] G. Deelstra, M. Grasselli, P. Koehl, Optimal Investment Strategies for Defined Contribution Funds, Working Paper, 2001.
[9] P. Battocchio, Optimal Portfolio Strategies with Stochastic Wage Income: The Case of a Defined Contribution Pension Plan, Working Paper, IRES,
Universite Catholique de Louvain, 2002.
[10] P. Battocchio, F. Menoncin, Optimal pension management in a stochastic framework, Insurance Math. Econom. 34 (2004) 79–95.
[11] R.H. Maurer, C. Schlag, M.Z. Stamos, Optimal Life-Cycle Strategies in the Presence of Interest Rate and Inflation Risk, Working Paper, Department of
Finance, Goethe University, Frankfurt, Germany, 2008.
[12] T. Chun-Hua, The Study of Rate of Return Guarantee under Defined Contribution Pension Plans, Ph.D thesis, Department of Finance, National Taiwan
University, 2007.
[13] F. Gozzi, M.D. Giacinto, S. Federico, S.B. Goldy, Optimal management of pension funds: a stochastic control approach, in: RICAM workshop on
optimization and optimal control, 2008.
[14] P. Devolder, M. Bosch Princep, I.D. Fabian, Stochastic optimal control of annuity contracts, Insurance Math. Econom. 33 (2003) 227–238.
[15] A.J.G. Cairns, D. Blake, K. Dowd, Stochastic lifestyling: Optimal dynamic asset allocation for defined contribution pension plans, J. Econom. Dynam.
Control 30 (2006) 377–843.
[16] D. Blake, D. Wright, Y. Zhang, Optimal Funding and Investment Strategies in Defined Contribution Pension Plans under Epstein-Zin Utility, Discussion
Paper, The pensions Institute, Cass Business School, City University, UK, 2008.
[17] J. Gao, Stochastic optimal control of DC pension funds, Insurance Math. Econom. 42 (2008) 1159–1164.
[18] M. Di Giacinto, S. Federico, F. Gozzi, Pension funds with a minimum guarantee: a stochastic control approach, Finance Stoch. 15 (2) (2011) 297–342.
[19] E. Vigna, On efficiency of mean–variance based portfolio selection in DC pension schemes, in: Collegio Carlo Alberto Notebook, Vol. 154, 2010.
[20] C.I. Nkeki, C.R. Nwozo, Variational form of classical portfolio strategy and expected wealth for a defined contributory pension scheme, J. Math. Finance
2 (2012) 132–139.
[21] C.I. Nkeki, C.R. Nwozo, Optimal investment under inflation protection and optimal portfolios with stochastic cash flows strategy, IAENG Int. J. Appl.
Math. 43 (2) (2013) IJAM 43 2 02.
[22] C.I. Nkeki, Optimal portfolio strategy with discounted stochastic cash inflows, J. Math. Finance 3 (2013a) 130–137.
[23] C.I. Nkeki, Meanvariance portfolio selection problem with time dependent salary for defined contribution pension scheme, Financ. Math. Appl. 2 (1)
(2013b) 1–26.
[24] C.I. Nkeki, Mean–variance portfolio selection with inflation hedging strategy: a case of a defined contributory pension scheme, Theor. Appl. Math.
Comput. Sci. 2 (2) (2012) 67–82.
[25] C.R. Nwozo, C.I. Nkeki, Optimal portfolio and strategic consumption planning in a life-cycle of a pension plan member in a defined contributory pension
scheme, IAENG Int. J. Appl. Math. 41 (4) (2011) 299–309.
[26] S. Haberman, E. Vigna, Optimal investment strategies and risk measures in defined contribution pension schemes, Insurance Math. Econom. 31 (2002)
35–69.
[27] R. Gerrard, S. Haberman, E. Vigna, Optimal investment choices post retirement in a defined contribution pension scheme, Insurance Math. Econom.
35 (2004) 321–342.
[28] B. Højgaard, E. Vigna, Mean–Variance Portfolio Selection and Efficient Frontier for Defined Contribution Pension Schemes, Technical Report R-2007-13,
Department of Mathematical Sciences, Aalborg University, 2007.
[29] R. Josa-Fombellida, J. Rincón-Zapatero, Mean–variance portfolio and contribution selection in stochastic pension funding, European J. Oper. Res. 187
(2008) 120–137.
[30] F.B. Hanson, J.J. Westman, Optimal consumption and portfolio control for jump–diffusion stock process with log–normal jumps, in: Proceedings of
American Control Conference, 8-10 May 2002, pp. 4256-4261.
[31] F.B. Hanson, J.J. Westman, Portfolio optimization with jump–diffusions: Estimation of time-dependent parameters and application, in: Proceedings of
2002 Conference on Decision and Control, Las Vegas, 9–13 December 2002b, pp. 1–6.
[32] F.B. Hanson, J.J. Westman, Jump–diffusion stock-return model with weighted fitting of time-dependent parameters, in: The Proceedings of 2003
American Control Conference, Paper ACC03-IEEE0493, 2003.
[33] S.P. Chakrabarty, F.B. Hanson, Optimal control of drug delivery to brain tumors for a distributed parameters model, in: Proceedings of American Control
Conference, June 2005.
[34] G. Yan, F.B. Hanson, Option pricing for a stochastic-volatility jump-diffusion model with log-uniform jump-amplitudes, in: Proceedings of American
Control Conference, 13 September 2006, pp. 1-7.
[35] Y. Ait-Sahalia, J. Cacho-Daiz, T.R. Hurd, Portfolio choice with jumps: a closed-form solution, Ann. Appl. Probab. 19 (2) (2009) 556–584.
[36] J. Zou, Z. Zhang, J. Zhang, Optimal dividend payout under jump-diffusion risk process, Stoch. Models 25 (2009) 332–347.
[37] J.Liua.T. Zhaob, P. Zhaoc, Portfolio problems based on jump-diffusion models, Published by Faculty of Sciences and Mathematics, University of Niŝ,
Serbia, 26 (3) (2012) 573-583.
[38] X. Zhou, D. Li, Continuous-time mean–variance portfolio selection: A stochastic LQ framework, Appl. Math. Optim. 42 (2000) 19–33.
[39] D. Hainaut, G. Deelstra, Optimal Timing for Annuitization, Based on Jump Diffusion Fund and Stochastic Mortality, Working Paper, ESC Rennes Business
School and CREST, France, Department of Mathematics, Université Libre de Bruxelles, Belgium, 2014.
[40] A. Chen, F. Uzelac, Portability, salary and asset price risk: A continuous-time expected utility comparison of DB and DC pension plans, Risks 3 (2015)
77–102.