# Portfolio selection with liability and affine interest rate in the HARA utility framework.

1. IntroductionIn practice of investments, it is all well known that some investors or financial institutions are often confronted with liability factor. It is very clear that considering liability factor in the portfolio selection problems will be more practical. In recent years, portfolio selection problems with liability have inspired literally hundreds of extensions and applications. For example, Chiu and Li [1] supposed liability process to be driven by a geometric Brownian motion and studied the efficient strategy and the efficient frontier under mean-variance criterion. Xie et al. [2] studied a mean-variance model with random liability, which is assumed to be governed by the drifted Brownian motion. Zeng and Li [3] investigated a mean-variance model with a jump diffusion liability process. Chen et al. [4] studied the asset and liability management (ALM) problems with regime switching, which described the effect of the changes in macroeconomic conditions. Chiu and Wong [5] investigated an ALM problem under the assumption that risky assets were cointegrated. Those models studied the ALM problems with exogenous liabilities, which cannot be controlled. Along another line, the ALM problems with endogenous liabilities have been paid more and more attention nowadays. Being different from exogenous liabilities, endogenous liabilities can be controlled by various financial instruments and investors' decisions. The interested readers can refer to the works of Leippold et al. [6] and Yao et al. [7].

However, the above mentioned research results were generally achieved under the assumption of constant interest rate. As a matter of fact, interest rate is always changing with time and can be delineated by some term structure models, for example, the Vasicek model [8] or the CIR model [9]. Therefore, some scholars began to be concerned with the portfolio selection problems with stochastic interest rate. For example, Korn and Kraft [10] investigated the portfolio selection problems with stochastic interest rates and proved the verification theorem. Deelstra et al. [11] studied the optimal investment strategies in the affine interest rate environment. Gao [12] applied dynamic programming principle and Legendre transform to study the DC pension problems with affine interest rate. Liu [13] and Chang and Rong [14] considered the optimal investment and consumption strategy with stochastic interest rate and stochastic volatility. These models were all studied in the utility maximization framework and the optimal investment strategies were established under power utility or logarithm utility.

Power utility, logarithm utility, and exponential utility are all special cases of HARA utility, which used to be studied by Grasselli [15], Canakoglu and Ozekici [16], and Jung and Kim [17]. Due to the complexity of HARA utility, there is little work on HARA utility in the existing literature. In recent years, some scholars found that Legendre transform is an effective method in solving sophisticated portfolio section problems. One can refer to the works of Jonsson and Sircar [18], Gao [12], and Jung and Kim [17].

This paper introduces liability process into a continuous-time dynamic portfolio selection problem and assumes interest rate to be driven by affine interest rate model. The objective of the investor is to look for an optimal investment strategy to maximize the expected utility of the terminal surplus. Considering the generality of HARA utility, we take HARA utility for our analysis. Applying dynamic programming principle, we obtain the Hamilton-Jacobi-Bellman (HJB) equation for the value function. Due to the nonlinearity of HARA utility, it is very difficult for us to conjecture the form of the solution for the HJB equation. Therefore, we apply Legendre transform to change the nonlinear HJB equation into a linear dual one. What is more, the form of the solution for the dual equation is linear and is easy to construct. Finally, we establish the explicit expression of the optimal investment strategy for HARA utility. Some special cases are also discussed. There are three highlights: (i) we study an asset and liability management problem under the affine interest rate model; (ii) interest rate is supposed to be governed by the affine interest rate model, while liability is assumed to be driven by the drifted Brownian motion; (iii) the optimal investment strategies for HARA utility are obtained explicitly.

This paper proceeds as follows. In Section 2, an asset and liability management problem with affine interest rate is formulated. In Section 3, we use dynamic programming principle to obtain the HJB equation and apply Legendre transform to change the HJB equation into a linear dual one. In Section 4, we derive the optimal investment strategy for HARA utility explicitly. Section 5 discusses some special cases of the optimal policies. Section 6 provides a numerical example to illustrate our results. Some interesting results are concluded in Section 7.

2. Model and Assumptions

In this paper, we assume that the financial market consists of three assets: one risk-free asset, one risky asset, and one zero-coupon bond. [0, T] represents certain investment horizon, where T < [infinity]. An investor is confronted with two key sources of risk, that is, interest rate risk and stock information risk, which are described by two one-dimensional standard and adapted independent Brownian motion [W.sub.r](t) and [W.sub.s](t). [W.sub.r](t) and [W.sub.s](t) are all defined on complete probability space ([OMEGA], F, {[F.sub.t]}, P).

The price process of the risk-free asset (i.e., cash), denoted by [S.sub.0](t), is given by

d[S.sub.0] (t) = r (t) [S.sub.0] (t) dt, [S.sub.0] (0) = 1, (1)

where r(t) is the risk-free interest rate. In this paper, r(t) is supposed to be stochastic process and is governed by

dr (t) = (a - br (t)) dt - [square root of [k.sub.1] r (t) + [k.sub.2]]d[W.sub.r] (t), r (0) = [r.sub.0] > 0, (2)

where the parameters a, b, [k.sub.1], and [k.sub.2] are positive real constants.

Remark 1. Notice that (2) consists of the Vasicek model (resp., the CIR model) as special cases, when [k.sub.1] (resp., [k.sub.2]) is equal to zero. Under these dynamics, the term structure of interest rate is affine.

The price process of the risky asset (i.e., the stock), denoted by [S.sub.1](t), can be described by the following stochastic differential equation (SDE) (referring to Deelstra et al. [11]):

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (3)

where [[lambda].sub.1], [[lambda].sub.2] (resp., [[sigma].sub.1], [[sigma].sub.2]) are constants (resp., positive constants).

The third asset is one zero-coupon bond with maturity T, whose price process is denoted by B(t, T). Then B(t, T) evolves (referring to Deelstra et al. [11])

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (4)

where [[sigma].sub.B](t, r(t)) = h(t) [square root of [k.sub.1] r(t) + [k.sub.2]] and h(t) is given by

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (5)

Remark 2. Equations (3) and (4) have been investigated by Deelstra et al. [11] and Gao [12], but their results were only achieved under power utility or logarithm utility. From (3), we can see that the price process of the stock is affected by uncertainty of interest rate, which is consistent with reality. In addition, we find that the price process of zero-coupon bond and interest rate is driven by the same Brownian motion. It implies that the zero-coupon bond is reduced to the risk-free asset when interest rate is reduced to a constant.

Assume that the liability process satisfies the following SDE:

dL (t) = udt + vd[W.sub.s](t), L(0) = [l.sub.0] > 0, (6)

where u and v are positive constants.

Assume that the initial capital of an investor is given by [w.sub.0] and the initial liability is denoted by [l.sub.0]; then the initial surplus is [w.sub.0] - [l.sub.0], which is denoted by [x.sub.0]. The investor wishes to invest the surplus in the financial assets to maximize the expected utility of the terminal surplus. The amount invested in the stock and zero-coupon bond is denoted by [[pi].sub.1](t) and [[pi].sub.2](t), respectively. Therefore, the surplus process X(t) at time t under the strategy [pi](t) = ([[pi].sub.1](t), [[pi].sub.2](t)) satisfies the following SDE:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (7)

with the initial surplus X(0) = [x.sub.0] > 0.

Definition 3 (admissible strategy). An investment strategy [pi](t) = ([[pi].sub.1](t), [[pi].sub.2](t)) is said to be admissible if the following conditions are satisfied:

(i) [[pi].sub.1](t) and [[pi].sub.2](t) are all [F.sub.t]-measurable;

(ii) E([[integral].sup.T.sub.0] ([([[pi].sub.1](t)[[sigma].sub.1] - v).sup.2] + [([[pi].sub.1](t)[[sigma].sub.2] [square root of [k.sub.1]r(t) + [k.sub.2]] + [[pi].sub.2](t)[[sigma].sub.B]).sup.2]) dt) < +[infinity];

(iii) SDE (7) has a unique solution for all [pi](t) = ([[pi].sub.1](t), [[pi].sub.2](t)).

Assume that the set of all admissible strategies is denoted by n. The investor wishes to look for an optimal strategy to maximize the expected utility of the terminal surplus; that is,

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (8)

where U(*) represents utility function, which is strictly concave and satisfies the Inada conditions: [??](0) = +[infinity] and [??](+[infinity]) = 0.

In this paper, we choose hyperbolic absolute risk aversion (HARA) utility function for our analysis. The HARA utility function with parameters [eta], p, and q is given by

U (x) = U ([eta], p, q, x] = 1 - p/qp [(q/1 - p x + [eta]).sup.p], q > 0, p < 1, p [not equal to] 0. (9)

As a matter of fact, HARA utility function recovers power utility and exponential utility as special cases.

(i) If we choose [eta] = 0 and q = 1 - p, then we have

U (0, p, 1 - p, x) = [x.sup.p]/p = [U.sub.power] (x). (10)

(ii) If we choose [eta] = 1 and p [right arrow] -[infinity], then we have

U (1, p, q, x] = - [e.sup.-qx]/q = [U.sub.exp] (x). (11)

3. HJB Equation and Legendre Transform

According to dynamic programming principle, we define the value function as

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (12)

with boundary condition given by V(T, r, x) = U(x).

By applying the principle of optimality, we obtain the following proposition.

Proposition 4. If the value function V(t, r, x) [member of] [C.sup.1,2,2] ([0, T] x R x R), then V(t, r, x) satisfies the following HJB equation:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (13)

where A is a variational operator, and by letting [[sigma].sub.r] = [square root of [k.sub.1] r(t) + [k.sub.2]] we obtain

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (14)

Here, [V.sub.t], [V.sub.x], [V.sub.xx], [V.sub.r], [V.sub.rr], and [V.sub.rx] represent first-order and second-order partial derivatives with respect to the variables t, r, and x.

Proof. The proof is standard. The interested readers can refer to the work of Korn and Kraft [10].

Assume that H(t, r, x) is a solution of the HJB equation (13); then we get

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (15)

Putting (15) into (13), we derive

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (16)

with boundary condition given by H(T, r, x) = ((1 - p)/qp) [((q/(1 - p))x + [eta]).sup.p].

Noting that (16) is a nonlinear second-order partial differential equation and it is not easy for us to conjecture the structure of a solution to (16) for HARA utility, we introduce the following Legendre transform to change (16) into a linear second-order partial differential equation such that we can obtain the explicit solution to (16).

Definition 5. Let f : [R.sup.n] [right arrow] R be a concave function. For all z > 0, Legendre transform can be defined as follows:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (17)

and then the function L(z) is called the Legendre dual function of f(x) (cf. [12, 17, 18]).

If f(x) is strictly concave, the maximum in (17) will be attained at just one point, which we denoted by [x.sub.0]. We can attain at the unique solution by the first-order condition

df (x)/dx - z = 0. (18)

So we have

L (z) = f ([x.sub.0]) - z[x.sub.0]. (19)

Following Jonsson and Sircar [18], Gao [12], and Jung and Kim [17], Legendre transform can be defined by

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (20)

where z > 0 denotes the dual variable to x. The value of x where this optimum is attained is denoted by g(t, r, z), so we have

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (21)

The relationship between [??](t, r, z) and g(t, r, z) is given by

g (t, r, z) = -[[??].sub.z] (t, r, z). (22)

Hence, we can choose either one of two functions g(t, r, z) and [??](t, r, z) as the dual function of H(t, r, x). In this paper, we choose g(t, r, z). Moreover, we have

[H.sub.x] = z, [??] (t, r, z) = H (t, r, g) - zg, g (t, r, z) = x. (23)

Differentiating (23) with respect to t, r, and x, we get

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (24)

Notice that H(T, r, x) = U(x); then at the terminal time T, we can define

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (25)

So we have g(T, r, z) = [([??]).sup.-1] (z), where [([??]).sup.-1] (z) is taken as the inverse of marginal utility.

Substituting (24) back into (16) yields

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (26)

Differentiating (26) with respect to z and using (22), we obtain

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (27)

with boundary condition given by g(T, r, z) = [([??]).sup.-1](z).

Under HARA utility function (9), we have

g (T, r, z) = 1 - p/q ([z.sup.1/(p-1)] - [eta]). (28)

4. The Optimal Portfolio

Assume that a solution of (27) is conjectured as follows:

g (t, r, z) = 1 - p/q [z.sup.1/(p-1)] f (t, r) - 1 - p/q [eta]h (t, r) + J (t, r), (29)

with boundary conditions given by f(T, r) = 1, h(T, r) = 1, and J(T, r) = 0.

Further, the partial derivatives of g(t, r, z) with respect to t, r, and z are as follows:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (30)

Plugging (30) into (27), after some simple calculations, we derive

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (31)

Equation (31) can be decomposed into the following three equations:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (33)

[h.sub.t] - rh + (a - br + [[lambda].sub.2][[sigma].sup.2.sub.r]) [h.sub.r] + 1/2 [[sigma].sup.2.sub.r] [h.sub.rr] = 0, h (T, r) = 1. (34)

Lemma 6. Assume that a solution of (32) is conjectured as [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] with boundary conditions given by [D.sub.1] (T) = 0 and [D.sub.2](T) = 0; then under the condition of p < min{1, [b.sup.2] /(2[k.sub.1] + [(b - [[lambda].sub.2][k.sub.1]).sup.2])} and p [not equal to] 0, [D.sub.1](t) and [D.sub.2](t) are given by (44) and (42), respectively.

Proof. Putting [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] into (32) and takng [[sigma].sub.r] = [square root of [k.sub.1]r(t) + [k.sub.2]] into consideration, we obtain

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (35)

Comparing the coefficients yields

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (36)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (37)

For the quadratic equation

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (38)

it is easy to calculate its discriminant, which is given by

[[DELTA].sub.1] = p/1 - p (1/p [b.sup.2] - 2[k.sub.1] - [(b - [[lambda].sub.2][k.sub.1]).sup.2]). (39)

(i) If 0 < p < 1, then when p < [b.sup.2] /(2[k.sub.1] + [(b - [[lambda].sub.2][k.sub.1]).sup.2]), that is, 0 < p < min{1, [b.sup.2] /(2[k.sub.1] + [(b - [[lambda].sub.2][k.sub.1]).sup.2])}, we have [[DELTA].sub.1] > 0.

(ii) If p < 0, we find that [[DELTA].sub.1] > 0 always holds.

Hence, under the condition of p < min{1, [b.sup.2] /(2[k.sub.1] + [(b - [[lambda].sub.2][k.sub.1]).sup.2])} and p [not equal to] 0, we have [[DELTA].sub.1] > 0. In addition, two different roots of (38) are given by

[m.sub.1,2] = [[b + (p / (1 - p)) [[lambda].sub.2][k.sub.1]]/[k.sub.1]] [+ or -] [square root of [[DELTA].sub.1]/[k.sub.1]. (40)

Further, (36) can be rewritten as

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (41)

After easy calculations, we have

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (42)

By using (37) x [k.sub.1] - (36) x [k.sub.2], we get

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (43)

Integrating (43) from t to T, we derive

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (44)

As a result, Lemma 6 is completed.

Lemma 7. Assume that a solution of (33) is given by the structure J(t, r) = (u - [[lambda].sub.1]v) [[integral].sup.T.sub.t] [??](s, r)ds; then [??](t, r) satisfies the following PDE:

[[??].sub.t] - r[??] + (a - br + [[lambda].sub.2][[sigma].sup.2.sub.r]) [[??].sub.r] + 1/2 [[sigma].sup.2.sub.r] [[??].sub.rr] = 0, [??] (T, r) = 1. (45)

Proof. Introducing the following variational operator on any function J(t, r):

[[nabla]J (t, r) = -rJ + (a - br + [[lambda].sub.2][[sigma].sup.2.sub.r]) [J.sub.r] + 1/2 [[sigma].sup.2.sub.r][J.sub.rr], (46)

(33) can be rewritten as

[[partial derivative]J (t, r)/[partial derivative]t] + [nabla]J (t, r) + u - [[lambda].sub.1] v = 0, J (T, r) = 0. (47)

On the other hand, we have

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (48)

Putting (48) into (47), we get

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (49)

Therefore, we obtain

[[partial derivative][??] (s, r)/[partial derivative]s] + [nabla][??](s, r) = 0, [??](T, r) = 1. (50)

That is, (45) holds.

Lemma 8. Assume that a solution of (45) is of the form [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], with terminal conditions [D.sub.3](T) = 0 and [D.sub.4](T) = 0; then [D.sub.4](t) and [D.sub.3] (t) are given by (56) and (58), respectively.

Proof. Plugging [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] into (45), we derive

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (51)

Comparing the coefficients on both sides of (51), we get the following two equations:

[[??].sub.4] (t) - 1 + ([[lambda].sub.2][k.sub.1] - b) [D.sub.4] (t) + 1/2 [k.sub.1][D.sup.2.sub.4] (t) = 0, [D.sub.4] (T) = 0; (52)

[[??].sub.3] (t) + ([[lambda].sub.2][k.sub.2] + a) [D.sub.4] (t) + 1/2 [k.sub.2][D.sup.2.sub.4] (t) = 0, [D.sub.3] (T) = 0; (53)

For (52), we find that the discriminant of the quadratic equation

- 1/2 [k.sub.1][D.sup.2.sub.4] (t) - ([[lambda].sub.2][k.sub.1] - b) [D.sub.4] (t) + 1 = 0 (54)

is given by [[DELTA].sub.2] = [([[lambda].sub.2][k.sub.1] - b).sup.2] + 2[k.sub.1] > 0. Therefore, two different roots of (54) are given by

[m.sub.3,4] = [[b - [[lambda].sub.2][k.sub.1]]/[k.sub.1]] [+ or -] [[square root of [[DELTA].sub.2]/[k.sub.1]]. (55)

Using the same approach as (36), we obtain

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (56)

By applying (53) x [k.sub.1] - (52) x [k.sub.2], we get

[[??].sub.3] (t) = [k.sub.2]/[k.sub.1] [[??].sub.4] (t) - [[a[k.sub.1] + b[k.sub.2]]/[k.sub.1]] [D.sub.4] (t) - [k.sub.2]. (57)

Further, after easy integration, we obtain

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (58)

Therefore, the proof is completed.

Lemma 9. Suppose that[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] is a solution of (34), with boundary conditions [D.sub.5](T) = 0 and [D.sub.6](T) = 0; then one has [D.sub.5](t) = [D.sub.3](t) and [D.sub.6](t) = [D.sub.4](t).

Proof. Investigating (34), we find that (34) and (45) have the same solutions. Hence, the conclusion holds.

Applying (24), (29), (30), and Lemmas 6-9, we have

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (59)

Namely, we obtain

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (60)

Meantime, considering (29) and letting g(t, r, z) = x, we derive

z = [([q/[1 - p]](x - J(t, r)) + [eta]h(t, r)).sup.p-1] [f.sup.1-p] (t, r). (61)

By using [H.sub.x] = z, we obtain the following optimal value function:

[H.sup.*] (t, r, x)

= 1 - p/qp [(q/1 - p (x - J(t, r)) + [eta]h(t, r)).sup.p] [f.sup.1-p] (t, r). (62)

In conclusion, we can summarize the above results in the following theorem.

Theorem 10. For HARA utility (9), if a solution to the HJB equation (16) is given by H(t, r, x), then, under the condition of p < min{l, [b.sup.2] /(2[k.sub.1] + [(b - [[lambda].sub.2][k.sub.1]).sup.2])} and p [not equal to] 0, the optimal investment strategies for the problem (8) are given by

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (63)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (64)

where [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], and [D.sub.2](t) are given by Lemmas 8, 9, and 6, respectively. Moreover, [partial derivative]J(t, r)/[partial derivative]r in (64) is given by

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (65)

The following theorem verifies that the optimal strategy given by Theorem 10 is optimal for the problem (8).

Theorem 11 (verification theorem). If H(t, r, x) is a solution to (13), that is, H(t, r, x) satisfies

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (66)

then for all admissible strategies [pi](t) = ([[pi].sub.1](t), [[pi].sub.2](t)) [member of] [PI], one has V(t, r, x) [less than or equal to] H(t, r, x); if [[pi].sup.*](t) = ([[pi].sup.*.sub.1](t), [[pi].sup.*.sub.2](t)) satisfies

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (67)

then one has V(t, r, x) = H(t, r, x), and it implies that [[pi].sup.*] (t) = ([[pi].sup.*.sub.1](t), [[pi].sup.*.sub.2](t)) [member of] [PI] is the optimal investment strategy of the problem (8).

Proof. Considering H(t, r, x) [member of] [C.sup.1,2,2]([0, T] x R x R) and using Ito formula from t to T for H(t, r, x), we obtain

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (68)

For H(t, r, x) is a solution to (13); that is, we have [sup.sub.[pi](t)[member of][PI]] [A.sup.[pi](t)] H(s, r(s), X(s)) = 0. So we obtain

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (69)

The last three terms in (69) are local martingales and their expected values are equal to zero. Thus, we get

E (H (T, r (T), X (T)) | r (t) = r, X (t) = x) [less than or equal to] H (t, r, x). (70)

Maximizing (70) for all admissible strategies [pi](t) = ([[pi].sub.1](t), [[pi].sub.2](t)) [member of] [PI], we derive

V (t, r, x) [less than or equal to] H (t, r, x). (71)

When [pi](t) = [[pi].sup.*] (t), all inequalities become equalities; that is, H(t, r, x) = V(t, r, x), and [[pi].sup.*](t) = ([[pi].sup.*.sub.1](t), [[pi].sup.*.sub.2](t)) is the optimal investment strategy of the problem (8).

The proof is completed.

Remark 12. Some interesting conclusions can be seen from (63) and (64). (i) The optimal amount [[pi].sup.*.sub.1](t) invested in the stock is affected by the parameters a, b, [k.sub.1], [k.sub.2], [[lambda].sub.1], [[sigma].sub.1], [[lambda].sub.2], u, and v but does not depend on the parameter [[sigma].sub.2]. In fact, [[sigma].sub.2] has an influence on the dynamics of stock price, which can be observed from (3). (ii) The optimal amount [[pi].sup.*.sub.2](t) invested in the zero-coupon bond depends on all the model parameters a, b, [k.sub.1], [k.sub.2], [[lambda].sub.1], [[sigma].sub.1], [[lambda.sub.2], u, v, and [[sigma].sub.2]; however, the price dynamics of zero-coupon bond is only impacted by b, [k.sub.1], [k.sub.2], and [[lambda].sub.2].

According to Theorem 10, we derive the following three corollaries.

Corollary 13. If [eta] = 0 and q = 1 - p, HARA utility (9) is degenerated into power utility (10). Therefore, under the condition of p < min{1, [b.sup.2] /(2[k.sub.1] + [(b - [[lambda].sub.2][k.sub.1]).sup.2])} and p [not equal to] 0, the optimal investment strategies for power utility are given by

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (72)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (73)

where [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] and [D.sub.2](t) are given by Lemmas 8 and 6, respectively. Moreover, [partial derivative]J(t, r)/[partial derivative]r in (73) is still given by

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (74)

Proof. It is very easy to derive that

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].(75)

Therefore, (72) and (73) hold.

Corollary 14. If [eta] = 1 and p [right arrow] -[infinity], HARA utility (9) is reduced to exponential utility (11); then the optimal policies for exponential utility are determined by

[[pi].sup.*.sub.1] (t) = [[lambda].sub.1]/[[sigma].sub.1] x 1/q h (t, r) + v/[[sigma].sub.1], (76)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (77)

where [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] are given by Lemmas 8 and 9, respectively. Moreover, [partial derivative]J(t, r)/[partial derivative]r in (77) is still given by

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (78)

Proof. When p [right arrow] -[infinity], then we have [D.sub.2](t) [right arrow] [D.sub.4](t) and [D.sub.6](t) = [D.sub.4](t). Moreover, we derive that

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (79)

On the other hand, we have

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (80)

Further, we obtain [partial derivative][D.sub.6](t)/[partial derivative]p = [partial derivative][D.sub.4](t)/[partial derivative]p = 0 and

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (81)

Using L'Hopital's rule, (80) is equal to

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (82)

Let [eta] = 1; we obtain that (76) and (77) hold.

Corollary 15. If [eta] = 0, p [right arrow] 0, and q [right arrow] 1, HARA utility is reduced to logarithm utility [U.sub.log](x) = ln x; then the optimal policies for logarithm utility are given by

[[pi].sup.*.sub.1] (t) = [[lambda].sub.1]/[[sigma].sub.1] x (X (t) - J (t, r)) + v/[[sigma].sub.1], (83)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (84)

where [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] is given by Lemma 8. Moreover, [partial derivative]J(t, r)/[partial derivative]r in (83) is still given by

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (85)

Proof. If p [right arrow] 0, then we have [D.sub.2](t) [right arrow] 0. Therefore, we derive

[H.sub.x]/[H.sub.xx] = - (X (t) -J (t, r)), [H.sub.rx]/[H.sub.rr] = -[partial derivative]J (t, r)/[partial derivative]r. (86)

Therefore, (83) and (84) hold.

5. Special Cases

In this section, we give several special cases for HARA utility, power utility, exponential utility, and logarithm utility, respectively.

In our model, if we do not consider liability factor, that is, u = v = 0, then we have J(t, r) = 0. Therefore, we obtain the following special cases.

Special Case 1. Under HARA utility (9), if there is no liability and p satisfies the conditions p < min{1, [b.sup.2]/(2[k.sub.1] + [(b - [[lambda].sub.2][k.sub.1]).sup.2])} and p [not equal to] 0, then the optimal policies for the problem (8) are given by

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (87)

where [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] and [D.sub.2](t) are given by Lemmas 9 and 6, respectively.

Special Case 2. If utility function is given by [U.sub.power](x) = [x.sup.p]/p, P < 1 and p [not equal to] 0, and p satisfies the conditions p < min{1, [b.sup.2]/(2[k.sub.1] + [(b - [[lambda].sub.2][k.sub.1]).sup.2])} and p [not equal to] 0, then the optimal policies for the problem (8) under the condition of no liability are given by

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (88)

where [D.sub.2](t) is given by Lemma 6.

Special Case 3. If utility function is given by [U.sub.exp] (x) = -[e.sup.-qx]/q, q > 0, then the optimal policies for the problem (8) under the condition of no liability are

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (89)

where [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] are given by Lemmas 8 and 9, respectively.

Special Case 4. If utility function is given by [U.sub.log](x) = ln x, then the optimal policies for the problem (8) with no liability are

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (90)

where [D.sub.2](t) are given by Lemma 6.

On the other hand, if we consider the problem (8) in the constant interest rate environments, that is, a = b = [k.sub.1] = [k.sub.2] = 0, then the price dynamics of zero-coupon bond is degenerated into that of risk-free asset. As a result, the optimal amount invested in the zero-coupon bond is zero; that is, [[pi].sup.*.sub.2](t) = 0. And it leads to some new expressions:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (91)

Therefore, we can derive the optimal policy of the problem (8) with liability in the following special cases.

Special Case 5. Under constant interest rate model, if utility function is given by HARA function (9), then the optimal policy of the problem (8) with liability is given by

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (92)

Special Case 6. If utility function is given by [U.sub.power](x) = [x.sup.p]/p, p < 1 and p [not equal to] 0, then the optimal policy of the problem (8) with liability and constant interest rate is

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (93)

Special Case 7. If utility function is given by [U.sub.exp](x) = -[e.sup.-qx]/q, q > 0, then the optimal policy of the problem (8) with liability and constant interest rate is

[[pi].sup.*.sub.1] (t) = [[lambda].sub.1]/[[sigma].sub.1] x 1/q [e.sup.-(T-t)r] + v/[[sigma].sub.1]. (94)

Special Case 8. If utility function is given by [U.sub.log](x) = ln x, then the optimal policy of the problem (8) with liability and constant interest rate is

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (95)

6. Numerical Analysis

In this section, we provide a numerical example to illustrate the effect of market parameters on the optimal investment strategy and compare our results with those in the existing literature. Throughout this section, unless otherwise stated, the basic parameters are given by a = 0.18, b = 0.23, [k.sub.1] = 0.7, [k.sub.2] = 0.9, r(0) = 0.05, [[lambda].sub.1] = 0.6, [[lambda].sub.2] = 0.8, [[sigma].sub.1] = 1.9, [[sigma].sub.2] = 1.7, u = 0.6, v = 0.6, p = -0.2, q = 2, [eta] = 0.02, t = 0, T = 1, and [x.sub.0] = 100.

By applying the above conclusions obtained, some optimal investment policies are calculated in Tables 1, 2, and 3. From Tables 1-3, we draw some conclusions as follows.

(a1) The optimal policy under HARA utility is roughly equivalent to that under power utility but is markedly different from that under exponential utility or logarithm utility.

(a2) In the stochastic interest rate environments, the optimal amount invested in the stock and zero-coupon bond under HARA utility is less than that under exponential utility but is more than that under logarithm utility. This situation in the constant interest rate environments is contrary to that in the stochastic interest rate environments.

(a3) The optimal amount with liability and affine interest rate invested in the risky assets (including a stock and a zero-coupon bond) is less than that only with affine interest rate but is more than that with liability and constant interest rate.

In order to illustrate the impact of model parameters on the optimal investment strategy with liability and affine interest rate, we depict the following graphs. In Figures 1 and 2, the optimal amount invested in the stock is depicted by the dashed line and is denoted by [[pi].sup.*.sub.1] (t); the optimal amount invested in the zero-coupon bond is depicted by the orange line and is denoted by [[pi].sup.*.sub.2] (t); the optimal amount invested in the cash is depicted by the thick line and is denoted by [[pi].sup.*.sub.0] (t).

Form Figures 1 and 2, we can draw some conclusions as follows.

(b1) [[pi].sup.*.sub.0] (t) and [[pi].sup.*.sub.1] (t) decrease with respect to the parameter [[sigma].sub.1], while [[pi].sup.*.sub.2] (t) increases with respect to [[sigma].sub.1]. In fact, as the value of [[sigma].sub.1] is increasing, the volatility risk of the stock is also increasing. It leads to that the investors would invest less money in the stock. In order to hedge interest rate risk, the investors need to invest much more money in the zero-coupon bond and less money in the cash.

(b2) [[pi].sup.*.sub.1] (t) does not depend on the parameter [[sigma].sub.2], while [[pi].sup.*.sub.2] (t) decreases and [[pi].sup.*.sub.0] (t) increases with respect to [[sigma].sub.2]. It shows that the investors need to invest less money in the zero-coupon bond and more money in the cash when the value of [[sigma].sub.2] increases.

(b3) [[pi].sup.*.sub.0] (t) and [[pi].sup.*.sub.1] (t) increase with respect to the parameter v, while [[pi].sup.*.sub.2] (t) decreases with respect to v. As a matter of fact, as the value of v becomes larger, the volatility of liability is increasing. This means that the investors need to invest more money in the stock and cash in order to hedge the risk that resulted from liability.

(b4) [[pi].sup.*.sub.0] (t) and [[pi].sup.*.sub.1] (t) increase with respect to the parameter p, while [[pi].sup.*.sub.2] (t) decreases with respect to p. This shows that the larger the value of p, the more the amount invested in the stock and cash. It leads to the fact that less money is invested in the zero-coupon bond.

7. Conclusions

This paper investigates the optimal investment strategy for an ALM problem in the HARA utility framework, where interest rate is supposed to be driven by an affine interest rate model, while liability process follows Brownian motion with drift. By applying dynamic programming principle and Legendre transform, we obtain the explicit expressions of the optimal investment strategies. Some special cases are discussed. Finally, we illustrate the impact of model parameters on the optimal policy by providing a numerical example. Some interesting conclusions are found as follows: (i) although the parameter [[sigma].sub.2] has an influence on the dynamics of stock price, the optimal amount invested in the stock does not depend on the value of [[sigma].sub.2]; (ii) the optimal amount under HARA utility is roughly equivalent to that under power utility but is markedly different from that under exponential utility and logarithm utility.

As far as we know, there is little work on the ALM problems with stochastic interest rate in the existing literature. However, our work has also some limits: (i) we only consider the liability process driven by the drifted Brownian motion, which is the simplest stochastic process; (ii) we study an ALM problem in the utility framework and do not consider it in a continuous-time mean-variance framework. We would leave those problems for future research.

http://dx.doi.org/10.1155/2014/312640

Conflict of Interests

The authors declare that there is no conflict of interests regarding the publication of this paper.

Acknowledgments

This research is supported by Humanities and Social Science Research Youth Foundation of Ministry of Education of China (no. 11YJC790006), Center for Research of Regulation and Policy of Zhejiang Province of China (no. 13JDGZ03YB), the Project of National Statistical Science (no. 2013LY125), and Higher School Science and Technology Development Foundation of Tianjin (no. 20100821).

References

[1] M. C. Chiu and D. Li, "Asset and liability management under a continuous-time mean-variance optimization framework," Insurance: Mathematics and Economics, vol. 39, no. 3, pp. 330-355, 2006.

[2] S. Xie, Z. Li, and S. Wang, "Continuous-time portfolio selection with liability: mean-variance model and stochastic LQ approach," Insurance: Mathematics and Economics, vol. 42, no. 3, pp. 943-953, 2008.

[3] Y. Zeng and Z. Li, "Asset-liability management under benchmark and mean-variance criteria in a jump diffusion market," Journal of Systems Science and Complexity, vol. 24, no. 2, pp. 317-327, 2011.

[4] P. Chen, H. Yang, and G. Yin, "Markowitz's mean-variance asset-liability management with regime switching: a continuous-time model," Insurance: Mathematics and Economics, vol. 43, no. 3, pp. 456-465, 2008.

[5] M. C. Chiu and H. Y. Wong, "Mean-variance principle of managing cointegrated risky assets and random liabilities," Operations Research Letters, vol. 41, no. 1, pp. 98-106, 2013.

[6] M. Leippold, F. Trojani, and P. Vanini, "Multiperiod mean-variance efficient portfolios with endogenous liabilities," Quantitative Finance, vol. 11, no. 10, pp. 1535-1546, 2011.

[7] H. Yao, Y. Lai, and Y. Li, "Continuous-time mean-variance asset-liability management with endogenous liabilities," Insurance: Mathematics and Economics, vol. 52, no. 1, pp. 6-17, 2013.

[8] O. Vasicek, "An equilibrium characterization of the term structure," Journal of Financial Economics, vol. 5, no. 2, pp. 177-188, 1977.

[9] J. C. Cox, J. E. Ingersoll, and S. A. Ross, "A theory of the term structure of interest rates," Econometrica, vol. 53, no. 2, pp. 385-408, 1985.

[10] R. Korn and H. Kraft, "A stochastic control approach to portfolio problems with stochastic interest rates," SIAM Journal on Control and Optimization, vol. 40, no. 4, pp. 1250-1269, 2002.

[11] G. Deelstra, M. Grasselli, and P. Koehl, "Optimal investment strategies in the presence of a minimum guarantee," Insurance: Mathematics and Economics, vol. 33, no. 1, pp. 189-207, 2003.

[12] J. Gao, "Stochastic optimal control of DC pension funds," Insurance: Mathematics and Economics, vol. 42, no. 3, pp. 1159-1164, 2008.

[13] J. Liu, "Portfolio selection in stochastic environments," Review of Financial Studies, vol. 20, no. 1, pp. 1-39, 2007

[14] H. Chang and X. Rong, "An investment and consumption problem with CIR interest rate and stochastic volatility," Abstract and Applied Analysis, vol. 2013, Article ID 219397, 12 pages, 2013.

[15] M. Grasselli, "A stability result for the HARA class with stochastic interest rates," Insurance: Mathematics and Economics, vol. 33, no. 3, pp. 611-627, 2003.

[16] E. Canakoglu and S. Ozekici, "Portfolio selection in stochastic markets with HARA utility functions," European Journal of Operational Research, vol. 201, no. 2, pp. 520-536, 2010.

[17] E. J. Jung and J. H. Kim, "Optimal investment strategies for the HARA utility under the constant elasticity of variance model," Insurance: Mathematics and Economics, vol. 51, no. 3, pp. 667-673, 2012.

[18] M. Jonsson and R. Sircar, "Optimal investment problems and volatility homogenization approximat ions," in Modern Methods in Scientific Computing and Applications, vol. 75 of NATO Science Series II, pp. 255-281, Springer, Berlin, Germany, 2002.

Hao Chang, (1) Kai Chang, (2) and Ji-mei Lu (3)

(1) School of Science, Tianjin Polytechnic University, Tianjin 300387, China

(2) School of Finance, Zhejiang University of Finance & Economics, Hangzhou 310018, China

(3) School of Electrical Engineering and Automation, Tianjin Polytechnic University, Tianjin 300387, China

Correspondence should be addressed to Hao Chang; ch8683897@126.com

Received 15 March 2014; Revised 18 May 2014; Accepted 27 May 2014; Published 17 June 2014

Academic Editor: Simone Marsiglio

TABLE 1: The optimal policy with liability and affine interest rate. HARA Power Exponential Logarithm utility utility utility utility Stock 26.6064 26.6049 0.390775 31.8627 Bond 37.8308 37.8343 99.4978 24.7162 Cash 35.5629 35.5608 0.111445 43.4211 TABLE 2: The optimal policy only with affine interest rate. HARA Power Exponential Logarithm utility utility utility utility Stock 26.3173 26.3158 0.0749854 31.5789 Bond 38.3303 38.3338 100.06 25.2023 Cash 35.3525 35.3504 -0.134829 43.2187 TABLE 3: The optimal policy with liability and constant interest rate. HARA Power Exponential Logarithm utility utility utility utility Stock 26.573 26.57 0.465984 31.8208 Cash 73.427 73.43 99.534 68.1792

Printer friendly Cite/link Email Feedback | |

Title Annotation: | Research Article |
---|---|

Author: | Chang, Hao; Chang, Kai; Lu, Ji-mei |

Publication: | Abstract and Applied Analysis |

Article Type: | Report |

Date: | Jan 1, 2014 |

Words: | 6914 |

Previous Article: | Derivatives of meromorphic functions with multiple zeros and small functions. |

Next Article: | Cluster projective synchronization of fractional-order complex network via pinning control. |

Topics: |