Portfolio adjusting optimization under credibility measures

Journal of Computational and Applied Mathematics 234 (2010) 1458–1465 Contents lists available at ScienceDirect Journal of Computational and Applied Mathematics journal homepage: www.elsevier.com/locate/cam Portfolio adjusting optimization under credibility measures Xili Zhang a , Wei-Guo Zhang a,∗ , Ruichu Cai b a School of Business Administration, South China University of Technology, Guangzhou, 510641, PR China b School of Computer Science and Engineering, South China University of Technology, Guangzhou, 510641, PR China article info Article history: Received 25 June 2009 Received in revised form 31 October 2009 MSC: 91B06 91G10 Keywords: Portfolio adjusting Possibility theory Credibility measure Transaction costs Sequential quadratic programming method abstract This paper discusses portfolio adjusting problems for an existing portfolio. The returns of risky assets are regarded as fuzzy variables and a class of credibilistic mean–variance adjusting models with transaction costs are proposed on the basis of credibility theory. Under the assumption that the returns of risky assets are triangular fuzzy variables, the optimization models are converted into crisp forms. Furthermore, we employ the sequential quadratic programming method to work out the optimal strategy. Numerical examples illustrate the effectiveness of the proposed models and the influence of the transaction costs in portfolio selection. © 2010 Elsevier B.V. All rights reserved. 1. Introduction A portfolio selection decision in a real investment process is made in an uncertain environment. In conventional portfolio analysis, a financial asset is usually characterized as a random variable with a probability distribution over its returns [1]. However, the portfolio selection environment is often subject to fuzziness and thus the assumption of only deterministic parameters or stochastic parameters can no longer be applied. Zhang and Nie [2] and Zhang et al. [3] studied admissible efficient portfolio selection problems under the assumption that the expected return and risk of an asset have admissible errors, to reflect the uncertainty in real investment actions, and gave an analytic derivation of an admissible efficient frontier when short sales are not allowed on any risky assets. The fuzzy set is a powerful tool used to describe an uncertain financial environment where not only the financial markets but also the investment decision makers are subject to vagueness, ambiguity or fuzziness. Possibility theory has been proposed in [4] and advanced in [5], where fuzzy variables are associated with possibility distributions. Recently, a number of researchers investigated fuzzy portfolio selection problem, such as [6–10]. There exist important differences between the possibility measure and the probability measure. But neither the possibility measure nor the necessity measure is self-dual. Credibility theory has been newly proposed in [11–13] where each fuzzy variable is associated with a credibility distribution in a similar way to how each random variable is associated with a probability distribution. As the average of a possibility measure and a necessity measure, the credibility measure is self-dual and weights possibility and necessity measures harmoniously. In this respect, the credibility measure shares some properties with the probability measure. Huang [14,15] proposed the credibilistic mean–variance model and mean–semivariance models for fuzzy portfolio selection. At the end of a typical time period, an investor considers adjusting the existing portfolio by buying or selling assets in response to conditions changing in financial markets. In practical situations, there is transaction cost associated with buying and selling an asset. Transaction cost is one of the main concerns for portfolio managers [16,17]. ∗ Corresponding author. Tel.: +86 20 87114121; fax: +86 20 22236282. E-mail addresses: [email protected], [email protected] (W.-G. Zhang). 0377-0427/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.cam.2010.02.022 X. Zhang et al. / Journal of Computational and Applied Mathematics 234 (2010) 1458–1465 1459 In this study, the returns of risky assets are taken as fuzzy variables. We use the credibility theory to measure the return and risk of the portfolio. By using the credibilistic expected value and variance, we discuss and develop a class of portfolio adjusting problems for an existing portfolio with transaction costs. The rest of the paper is organized in the following manner. In Section 2, we review some preliminary knowledge about fuzzy variables in credibility theory. In Section 3, we propose a class of credibilistic portfolio adjusting models with transaction costs and further convert these models into crisp forms under the assumption that the returns for risky assets are triangular fuzzy variables. In Section 4, we present a sequential quadratic programming for solving the proposed problems. In Section 5, numerical examples are given to illustrate the application of the proposed models and the influence of the transaction costs. Section 6 gives conclusions. 2. Preliminaries In this section, we are going to review some concepts of credibilistic theory which will be used in the remainder of the paper. Let ξ be a fuzzy variable with membership function µ and u be a real number. Then the possibility, necessity and credibility measures of {ξ ≤ u} are defined as Pos{ξ ≤ u} = sup µ(x), x ≤u Nec {ξ ≤ u} = 1 − Pos{ξ > u} = 1 − sup µ(x), x>u Cr {ξ ≤ u} = 1 2 (Pos{ξ ≤ u} + Nec {ξ ≤ u}), respectively, where Pos and Nec are a pair of dual fuzzy measures, and Cr is self-dual, i.e., Cr {ξ ≤ u} + Cr {ξ > u} = 1. On the basis of the credibility measure, the definitions of the expected value and variance of a fuzzy variable can be given as follows: Definition 1 ([13]). Let ξ be a normalized fuzzy variable. The expected value of ξ is defined by Ec [ξ ] = Z ∞ Cr {ξ ≥ r }dr − 0 Z 0 Cr {ξ ≤ r }dr −∞ provided that at least one of the two integrals is finite. Definition 2 ([12]). If the fuzzy variable ξ has a finite expected value e, then its variance is defined as Vc [ξ ] = Ec [(ξ − e)2 ]. The expected value and variance of a fuzzy variable satisfy the following properties. Theorem 1 ([12]). If ξ is a fuzzy variable with finite expected value, then for any real numbers λ1 and λ2 , we have Ec [λ1 ξ + λ2 ] = λ1 Ec [ξ ] + λ2 . Theorem 2 ([12]). If ξ is a fuzzy variable with finite expected value, then for any real numbers λ1 and λ2 , we have Vc [λ1 ξ + λ2 ] = λ21 Vc [ξ ]. In particular, if ξ = (b, α, β) is a triangular fuzzy number with center b, left width α > 0 and right width β > 0, then its membership function has the following form:  u − (b − α)  ,    α µ(u) = (b + β) − u  ,   β  0, if b − α ≤ u ≤ b, if b ≤ u ≤ b + β, otherwise. From the definition of credibility measure, the credibility of event {ξ ≤ u} is as follows: 0,    u−b+α    , 2α Cr {ξ ≤ u} =  u + β − b,    2β  1, if u ≤ b − α if b − α ≤ u ≤ b, if b ≤ u ≤ b + β, otherwise. 1460 X. Zhang et al. / Journal of Computational and Applied Mathematics 234 (2010) 1458–1465 From Definition 1, the credibilistic expected value of the triangular fuzzy number ξ is given as 4b − α + β Ec [ξ ] = . 4 From Definition 2, the credibilistic variance of the triangular fuzzy number ξ is obtained as Vc [ξ ] = 33M 3 + 21M 2 m + 11Mm2 − m3 384M where M = max{α, β} and m = min{α, β}. , 3. Credibilistic portfolio adjusting models Suppose an investor holds an existing portfolio. Let x0 = (x01 , . . . , x0n ) represent the investor’s current holdings of assets, where x0i is the holding of the risky asset i (i = 1, 2, . . . , n) owned by the investor before portfolio adjusting. Due to changes of situations in financial markets and investors’ preferences as regards risk, the investor intends to reallocate the holdings of assets. We assume that the investor allocates his/her wealth exclusively among these n risky assets offering fuzzy returns − ri (i = 1, . . . , n). Let x+ i and xi denote respectively the amounts purchased and sold of asset i. So the holding of risky asset i after adjusting xi can be expressed as − xi = x0i + x+ i − xi , + − xi xi = 0, x− x+ i ≥ 0, i ≥ 0, (1) i = 1, . . . , n, − which also implies that both x+ i and xi are nonnegative variables and the investor can either purchase or sell the asset i. Let the transaction cost rates of purchasing and selling asset i be pi ≥ 0 and si ≥ 0 (i = 1, . . . , n), respectively. So purchasing transaction costs are measured by the amount increased from x0i , that is, pi x+ i . Selling transaction costs are measured by the decrease from x0i , i.e., si x− . Because either purchasing or selling could be chosen, the transaction cost of i risky asset i can be calculated using the following formula: − Ci (xi ) = pi x+ i + s i xi , i = 1, . . . , n. Hence, the total transaction costs of n risky assets are obtained using C (x) = n X − (pi x+ i + si xi ). (2) i=1 We assume that the investor does not invest the additional capital during the portfolio adjusting process. Thus, we have the following equation: n X i=1 xi + n n X X − (pi x+ x0i . i + si x i ) = i=1 i=1 Pn The return on the portfolio x = (x1 , . . . , xn ) without transaction costs can be expressed as rp = i=1 xi ri . As we know, the sum of fuzzy variables is also a fuzzy variable and the product of a fuzzy variable and a scalar number is also a fuzzy variable. Therefore, rp isP still a fuzzy variable. The credibilistic expected return and variance of the portfolio x are respectively given Pn n by Ec [rp ] = E [ i=1 ri xi ] and Vc [rp ] = Vc [ i=1 ri xi ]. From Theorem 1, the credibilistic net expected return of the portfolio after paying transaction costs is given as Ec [rp − C (x)] = Ec = Ec " " n X i=1 X n i=1 # n X + − (pi xi + si xi ) r i xi − r i xi # i=1 − n X − (pi x+ i + si xi ). (3) i=1 From Theorem 2, the credibilistic variance of the portfolio after paying transaction costs is obtained by using Vc [rp − C (x)] = Vc " n X i=1 = Vc " n X i=1 n X − ri xi − (pi x+ i + s i xi ) # ri xi . i=1 # (4) According to the theory of the mean–variance criterion, the investor seeks an optimal investment strategy x = (x1 , . . . , xn ) for minimizing the variance when the growth rate is given. On the basis of the above discussions, the credibilistic 1461 X. Zhang et al. / Journal of Computational and Applied Mathematics 234 (2010) 1458–1465 portfolio adjusting problem is formulated as follows: min Vc x s.t . Ec " n X " n X n X i=1 r i xi i=1 ri xi i=1 # # n X − (pi x+ i + si xi ) ≥ µ0 , − i=1 n n X X − x0i , (pi x+ + s x ) = xi + i i i (5) i=1 i=1 − xi = x0i + x+ i = 1, . . . , n, i − xi , 0 ≤ xi ≤ ui , i = 1, . . . , n, − x+ i = 1, . . . , n, i xi = 0 , + xi ≥ 0, i = 1, . . . , n, x− i = 1, . . . , n, i ≥ 0, where µ0 is a required rate of return of the portfolio. Nonnegativity constraints xi ≥ 0 are included to prohibit short selling assets and ui is the upper bound on the asset i holding, i = 1, . . . , n. Pn 0 In the special case of pi = si ≥ 0, the total transaction costs can be written as C (x) = i=1 pi |xi − xi |. Then the model (5) can be simplified as min Vc x s.t . Ec " r i xi i=1 " n X X n n X ri xi i=1 xi + i=1 # # X n n X − pi |xi − x0i | ≥ µ0 , i=1 X (6) n pi |xi − x0i | = i=1 x0i , i=1 0 ≤ x i ≤ ui , i = 1, . . . , n. Let the return of risky asset ri = (bi , αi , βi ) (i = 1, . . . , n) be triangular fuzzy variables. It seems reasonable to choose the risky assets in such a way that the right width of the fuzzy number is greater than or equal to the left width, that is, βi ≥ αi , i = 1, . . . , n. It is known from [11,12] that the sum of independent triangular fuzzy variables ξ1 = (b1 , α1 , β1 ) and ξ2 = (b2 , α2 , β2 ) is also a triangular fuzzy variable and ξ1 + ξ2 = (b1 + b2 , α1 + α2 , β1 + β2 ), and that the product of a triangular fuzzy variable ξ1 = (b1 , α1 , β1 ) and a scalar number λ ≥ 0 is also a triangular fuzzy variable and λξ1 = (λb1 , λα1 , λβ1 ). Therefore, for any real numbers xi ≥ 0, i = 1, . . . , n, the return of a portfolio is still a triangular fuzzy variable in the form of rp = n X ri xi = i=1 Pn n X bi xi , i=1 Pn n X αi xi , i=1 ! n X (7) βi xi , i=1 where i=1 αi xi ≤ i=1 βi xi . Then the crisp expression for the credibilistic expected value of rp is Ec " n X i=1 ri xi # = n 1X 4 i=1 (4bi − αi + βi )xi , (8) and the crisp form of the credibilistic variance value of rp is Vc " X n i=1 ri xi # 33 =  n P βi xi i=1 3 + 21  n P i=1 βi xi 2   n  n 2  n 3 P P P α xi + 11 βi xi αi xi − αi xi i=1 i=1 i=1 n  i=1 P 384 βi xi n P i=1 = 11 128 n X i=1 βi xi !2 + 7 128 n X i=1 β i xi ! n X i=1 ! α xi + 11 384 n X i=1 αi xi !2 −  n P αi xi i=1 384  n P i=1 3 βi xi . (9) 1462 X. Zhang et al. / Journal of Computational and Applied Mathematics 234 (2010) 1458–1465 Therefore, the model (5) can be converted into the following crisp form: min x s.t . n X 11 128 i=1 n 1X 4 i=1 n X βi xi !2 + n X 7 128 i=1 n X (4bi − αi + βi )xi − xi + i=1 n X βi xi i=1 − (pi x+ i + s i xi ) = i=1 ! n X i=1 + ! α xi + 11 384 n X α i xi i=1 !2 −  n P 384  − (pi xi + si xi ) ≥ µ0 , n X αi xi i=1 n P 3 β i xi i=1  (10) x0i , i=1 − xi = x0i + x+ i = 1, . . . , n, i − xi , 0 ≤ xi ≤ ui , i = 1, . . . , n, − x+ i = 1, . . . , n, i xi = 0, + xi ≥ 0, i = 1, . . . , n, x− i = 1 , . . . , n. i ≥ 0, The model (6) in the special case of pi = si ≥ 0 can be transformed as min x s.t . n X 11 128 i=1 n 1X 4 i=1 n X i=1 βi xi !2 + n X 7 128 i=1 n X (4bi − αi + βi )xi − xi + n X pi |xi − n X i=1 x0i ! α xi + 11 384 n X i=1 α i xi !2 | ≥ µ0 , −  n P αi xi i=1 384  n P i=1 3 β i xi  (11) i=1 pi |xi − x0i n X |= i=1 0 ≤ x i ≤ ui , βi xi ! x0i , i=1 i = 1, . . . , n. 4. The SQP method for solving the credibilistic portfolio adjusting problem In order to obtain the optimal investment strategy, we need an efficient algorithm to solve the proposed portfolio adjusting problem (10). + − − ′ Let y be a 3n-dimensional vector and y′ = [x1 , . . . , xn , x+ 1 , . . . , xn , x1 , . . . , xn ]. We use a prime ( ) to denote matrix e transposition and adopt the convention that all non-primed vectors are column vectors. Let e α , β,e u be 3n-dimensional vectors and e α ′ = [α1 , . . . , αn , 0, . . . , 0], e β ′ = [β1 , . . . , βn , 0, . . . , 0], e u′ = [u1 , . . . , un , ∞, . . . , ∞]. Let e l be a 3ndimensional vector of zeros. The model (10) can be ultimately written in matrix notation as min y s.t . (y′ C1 y)2 f (y) = y′ C0 y − y′ C2 y y′ Qi y = 0, i = 1, . . . , n, p′ · y + µ0 ≤ 0, A · y − q = 0, el ≤ y ≤ e u, (12) where: 7 e ′ 11 11 ee′ β β + 128 βe α + 384 e αe α ′ , C1 = e αe α ′ , C2 = 384e βe α ′ .Qi = li+n l′i+2n (i = 1, . . . , n) C0 , C1 , C2 are 3n × 3n matrices and C0 = 128 are 3n × 3n matrices where lm is a 3n-dimensional vector of zeros except for component m which has value unity. 4b −α +β 4b −α +β p′ = [− 1 41 1 , . . . , − n 4n n , p1 , . . . , pn , s1 , . . . , sn ] is a 3n-dimensional vector. A is an (n + 1) × 3n matrix and A = h I e′ −I g′ I s′ i ,in which I is an n × n identity matrix, e′ = [1, . . . , 1], g ′ = [p1 , . . . , pn ] and s′ = [s1 , . . . ,P sn ] are n-dimensional vectors. n q′ = [x01 , . . . , x0n , i=1 x0i ] is an (n + 1)-dimensional vector. From the model (12), we can tell that the proposed credibilistic portfolio adjusting problem is a kind of nonlinear minimization problem of finding a 3n-dimensional vector y, and the constraints are a mixture of linear and nonlinear equality constraints, lower and upper bounds on the variables and inequality constraints. 1463 X. Zhang et al. / Journal of Computational and Applied Mathematics 234 (2010) 1458–1465 Table 1 Fuzzy returns and upper bounds of eight assets. Assets 1 2 3 4 5 6 7 8 bi 0.051 0.1426 0.5264 0.20 0.1432 0.3629 0.4662 0.20 −0.0945 −0.1451 0.2953 0.4088 0.20 0.3587 0.6298 0.20 0.1076 0.2588 0.3923 0.20 0.1088 0.418 0.4865 0.20 0.1065 0.2445 0.4158 0.20 0.2779 0.3855 0.3855 0.20 αi βi ui As is known, in constrained optimization, the general aim is to transform the problem into an easier subproblem that can then be solved and used as the basis of an iterative process. The Sequential Quadratic Programming (SQP) method represents the state of the art in nonlinear programming methods and is one of the most popular and robust algorithms for nonlinear continuous optimization. Schittkowski [18] has implemented and tested a version that outperforms every other tested method in terms of efficiency, accuracy, and percentage of successful solutions, over a large number of test problems. SQP algorithms are viewed by many as the best approach to NLP solution (see [19]). An overview of SQP is found in [20,21]. Therefore, we employ the SQP methods to solve the credibilistic portfolio adjusting problem. The Lagrangian function of the model (12) is L(y, λ) = f (y) + n X i=1 λi y′ Qi y + λn+1 (p′ · y + µ0 ) + λn+2 (Ay − q) + λn+3 (y − e u) + λn+4 (e l − y) where λi , i = 1, . . . , n + 4, are Lagrange multipliers. At an iterate yk , a basic sequential quadratic programming algorithm defines an appropriate search direction dk as a solution to the quadratic programming (QP) subproblem: min d∈R3n s.t . 1 ′ d Hk d + ∇ f (yk )′ d 2 y′k Qi d + y′k Qi yk = 0, i = 1, . . . , n, p′ d + (p′ yk + µ0 ) ≤ 0, Ad + (Ayk − q) = 0, d + (yk − e u) ≤ 0, −d + (el − yk ) ≤ 0, (13) where Hk is the Hessian of L(y, λ). The general idea of the SQP method for solving the credibilistic portfolio adjusting problem (12) (or (10)) is stated here: Step 1. Set the investor’s current existing portfolio as a starting point for the solution, that is, y0 = [x01 , . . . , x0n , 0, . . . , 0]. Step 2. Solve the QP subproblem (13) to find dk using an active set strategy. If kdk k ≤ ε , stop and output the optimal results: yk , f (yk ), total transaction cost and the expected return obtained. Otherwise determine the step length parameter δk by means of an appropriate line search procedure. Step 3. Set yk+1 = yk + δk dk . Update Hk+1 by using the BFGS method: hk h′ H ′ θ ′ θk Hk Hk+1 = Hk + ′ k − k ′ k , hk θk θk Hk θk where hk = ∇ L(yk+1 , λ) − ∇ L(yk , λ) and θk = yk+1 − yk . Step 4. k = k + 1, go to Step 2. Moreover, the model (11) can be solved with the same method by setting pi = si (i = 1, . . . , n). 5. Numerical examples In the real world of portfolio management, we need to estimate possibility distributions of the returns of assets. Assume that there are eight risky assets for investment and the investor’s current existing portfolio before adjusting is (x01 , x02 , x03 , x04 , x05 , x06 , x07 , x08 ) = (0.1988, 0.1176, 0.1258, 0.0798, 0.0844, 0.1490, 0.1107, 0.1339). The returns of the assets in the next investment period are regarded to be triangular fuzzy variables ri = (bi , αi , βi ) with the right width βi greater than left width αi , i = 1, . . . , 8. Applying the simple method for determining the triangular type possibility distribution introduced in [22], we use the historical data to estimate the return rates of the eight risky assets. The parameters of membership functions for the asset returns and the upper bounds of holdings are given in Table 1. The unit purchase and sale costs for all the assets are set as pi = 0.003 and si = 0.006 (i = 1, . . . , 8), respectively. Suppose that the investor seeks an optimal portfolio consisting exclusively of these eight risky assets and does not invest the additional capital during the portfolio adjusting process. Thus, we can apply the model (10) to solve this kind of portfolio adjusting problem. In MATLAB language, we implement the SQP method for the credibilistic portfolio adjusting problem by invoking the constrained optimization routine ‘fmincon’. Varying the preset expected return value µ0 = 0.1, 0.15 and 0.17, we have 1464 X. Zhang et al. / Journal of Computational and Applied Mathematics 234 (2010) 1458–1465 Table 2 The optimal adjusting strategies for eight risky assets for different values of µ0 . Assets µ0 = 0.1 1 2 3 4 5 6 7 8 µ0 = 0.15 µ0 = 0.17 x+ i x− i xi x+ i x− i xi x+ i x− i xi 0.0000 0.0000 0.0305 0.0000 0.0395 0.0000 0.0265 0.0421 0.0435 0.0000 0.0000 0.0798 0.0000 0.0166 0.0000 0.0000 0.1553 0.1176 0.1563 0.0000 0.1239 0.1324 0.1372 0.1760 0.0000 0.0000 0.0000 0.0000 0.0573 0.0000 0.0232 0.0661 0.000 0.0223 0.0448 0.0798 0.0000 0.0000 0.0000 0.0000 0.1988 0.0953 0.0800 0.0000 0.1417 0.1490 0.1339 0.2000 0.0000 0.0201 0.0000 0.0000 0.0647 0.0000 0.0519 0.0661 0.0000 0.0000 0.1236 0.0798 0.0000 0.0011 0.0000 0.0000 0.1988 0.1377 0.0022 0.0000 0.1491 0.1479 0.1626 0.2000 Vc 0.0320 0.0327 0.0330 C 0.0013 0.0013 0.0018 Ec 0.1347 0.1530 0.17 Table 3 The optimal strategies for different values of µ0 with pi = 0.001 (i = 1, . . . , 8). µ0 x1 x2 x3 x4 x5 x6 x7 x8 Vc C Ec 0.1 0.15 0.17 0.1249 0.1821 0.1988 0.1176 0.0872 0.1506 0.1655 0.0824 0.0062 0.0000 0.0000 0.0000 0.1383 0.1168 0.1375 0.1157 0.1449 0.1466 0.1442 0.1854 0.1589 0.1925 0.2000 0.2000 0.0312 0.0324 0.0331 0.0013 0.0012 0.0014 0.1351 0.1528 0.17 Table 4 The optimal strategies for different values of µ0 with si = 0.003 (i = 1, . . . , 8). µ0 x1 x2 x3 x4 x5 x6 x7 x8 Vc C Ec 0.1 0.15 0.17 0.1253 0.1988 0.1988 0.1176 0.1058 0.1139 0.1654 0.0846 0.0020 0.0000 0.0000 0.0000 0.1382 0.1248 0.1690 0.1160 0.1490 0.1490 0.1440 0.1363 0.1662 0.1924 0.2000 0.2000 0.0313 0.0328 0.0328 0.0011 0.0007 0.0012 0.1353 0.1530 0.1701 the result displayed in Table 2, in which Ec means the credibilistic net expected return on the optimal portfolio after paying transaction costs, Vc means the credibilistic variance on the optimal portfolio after paying transaction costs and C represents the total transaction costs for adjusting (purchasing and selling). From Table 2, it can be seen that when the preset return value becomes bigger, the variance becomes larger, which reflects the relationship between the risk and return. When µ0 = 0.15, the investor should hold the assets 1 and 6 without a transaction, add to the holding of asset 7 by 0.0232 and increase the holding of asset 8 up to its upper bound, meanwhile decreasing the holding of asset 3 by 0.0448 and selling out the whole holding of asset 4. Then the holding of the portfolio after adjusting is obtained as x = (0.1988, 0.0953, 0.0800, 0, 0.1417, 0.149, 0.1339, 0.2). The total transaction costs are 0.0013. The minimal credibilistic variance is 0.0327 and the portfolio obtained can acquire 0.153 of net expected return. Furthermore, we illustrate the influence of the unit purchase and sale costs on portfolio selection decision making. Firstly, we decrease the unit purchase cost of each asset from pi = 0.003 to pi = 0.001 (i = 1, . . . , 8) and repeat the solution process. The results are listed in Table 3. We can see that, with the same preset return level µ0 = 0.15, the optimal portfolio after adjusting is x = (0.1821, 0.0872, 0.0824, 0, 0.1168, 0.1449, 0.1854, 0.2) with credibilistic variance Vc = 0.0324 and the total transaction costs C = 0.0012. Compared to the corresponding results in Table 2, the total transaction costs are generally less under the same preset expected return and the related optimal strategies are quite different, which implies that the unit purchase cost has a great impact on the portfolio selection. On the other hand, we reduce the unit sale cost of each asset from si = 0.006 to si = 0.003 (i = 1, . . . , 8) and repeat the solution process. The results are given in Table 4. It is obvious that the total transaction costs are generally less than and the optimal strategies are different from the original results in Table 2. It is also shown that the unit sale cost of assets has an immediate effect on optimal strategy making. 6. Conclusion By taking the returns of risky assets as fuzzy variables, this paper makes use of the credibilistic expected value and credibilistic variance to measure the return and risk of assets. A class of credibilistic mean–variance adjusting models with transaction costs are proposed. Under the assumption that the returns of risky assets are taken as triangular fuzzy variables, we convert the optimization models into crisp forms which is a kind of nonlinear minimization problem of finding a 3n-dimensional vector with both nonlinear equality constraints and linear equality and inequality constraints. We apply the sequential quadratic programming method for the credibilistic portfolio adjusting problems to make the investment strategy derivation an easy implementation task. Numerical examples are given to illustrate the ideas and application of the X. Zhang et al. / Journal of Computational and Applied Mathematics 234 (2010) 1458–1465 1465 proposed credibilistic portfolio adjusting models. The computation results show that the unit purchase and sale cost have great influence on the optimal strategy making. Acknowledgement This research was supported by the National Natural Science Foundation of China (70825005), NCET(06-0749), the Humanities and Social Science Youth Foundation of the Ministry of Education of China (No. 07JC630059) and the Natural Science Foundation of Guangdong Province of China (No. 8451064101000358). References [1] H.M. Markowitz, Portfolio Selection: Efficient Diversification of Investments, Wiley, New York, 1959. [2] W.G. Zhang, Z.K. Nie, On admissible efficient portfolio selection problem, Applied Mathematics and Computation 159 (2004) 357–371. [3] W.G. Zhang, Y.L. Wang, An analytic derivation of admissible efficient frontier with borrowing, European Journal of Operational Research 184 (2008) 229–243. [4] L.A. Zadeh, Fuzzy sets as a basis for a theory of possibility, Fuzzy Sets and Systems 1 (1978) 3–28. [5] D. Dubois, H. Prade, Possibility Theory: An Approach to Computerized Processing of Uncertainty, Plenum Press, New York, 1988. [6] C. Carlsson, R. Fullér, P. Majlender, A possibilistic approach to selecting portfolios with highest utility score, Fuzzy Sets and Systems 131 (2002) 13–21. [7] T. Hasuike, H. Katagiri, H. Ishii, Portfolio selection problems with random fuzzy variable returns, Fuzzy Sets and Systems (2008) doi:10.1016/j.fss.2008.11.010. [8] M. Inuiguchi, T. Tanino, Portfolio selection under independent possibilistic information, Fuzzy Sets and Systems 115 (2000) 83–92. [9] Y. Yoshida, An estimation model of value-at-risk portfolio under uncertainty, Fuzzy Sets and Systems (2009) doi:10.1016/j.fss.2009.02.007. [10] W.G. Zhang, Y.L. Wang, Z.P. Chen, Z.K. Nie, Possibilistic mean–variance models and efficient frontiers for portfolio selection problem, Information Sciences 177 (2007) 2787–2801. [11] B. Liu, Theory and Practice of Uncertain Programming, Physica-Verlag, Heidelberg, 2002. [12] B. Liu, Uncertainty Theory, 2nd ed., Springer-Verlag, Berlin, 2007. [13] B. Liu, Y.K. Liu, Expected value of fuzzy variable and fuzzy expected value models, IEEE Transactions on Fuzzy Systems 10 (2002) 445–450. [14] X.X. Huang, Portfolio selection with fuzzy returns, Journal of Intelligent and Fuzzy systems 18 (2007) 383–390. [15] X.X. Huang, Mean–semivariance models for fuzzy portfolio selection, Journal of Computational and Applied Mathematics 217 (2008) 1–8. [16] R.D. Arnott, W.H. Wanger, The measurement and control of trading costs, Financial Analysts Journal 46 (1990) 73–80. [17] J.B. Schattman, Portfolio Selection Under Non-Convex Transaction Costs and Capital Gains Taxes, Ph.D. Thesis. Rutgers Center for Operations Research, Rutgers University, USA, 2000. [18] W. Hock, K. Schittkowski, A comparative performance evaluation of 27 nonlinear programming codes, Computing 30 (1983) 335–345. [19] P.E. Gill, W. Murray, M.A. Saunders, M.H. Wright, Constrained nonlinear programming, in: G.L. Nemhauser, A.H.G. Rinnooy Kan (Eds.), in: Optimization, Handbooks in Operations Research and Management Science, vol. 1, Elsevier, 1989, pp. 171–210. [20] K. Schittkowski, NLQPL: A Fortran-subroutine solving constrained nonlinear programming problems, Annals of Operations Research 5 (1985) 485–500. [21] R. Fletcher, Practical Methods of Optimization, John Wiley and Sons, 1987. [22] W.G. Zhang, W.L. Xiao, On weighted lower and upper possibilistic means and variances of fuzzy numbers and its application in decision, Knowledge and Information Systems 18 (2009) 311–330.