OPTIMAL CHOICE UNDER SHORT SELL LIMIT WITH SHARPE RATIO AS CRITERION AMONG MULTIPLE ASSETS

Transcription

1 OPTIMAL CHOICE UNDER SHORT SELL LIMIT WITH SHARPE RATIO AS CRITERION AMONG MULTIPLE ASSETS Ruoun HUANG *, Yiran SHENG ** Abstract: This article is the term paper of the course Investments. We mainly focus on modeling long-term investment decisions of a typical utility-maximizing individual, with features of Chinese stoc maret in perspective. We adopt an OR based methodology with maret information as input parameters to carry out the solution. Two main features of this article are: first, we tae the no short-sell constraint in Chinese stoc maret into consideration and use an approach otherwise identical to Marowitz to wor out the optimal portfolio choice; this method has critical and practical implication to Chinese investors. Second, we incorporate the benefits of multiple assets into one single well-defined utility function and use a MIQP procedure to derive the optimal allocation of funds upon each of them along the time-line. * Ruoun HUANG, , huangr.06@sem.tsinghua.edu.cn, ** Yiran SHENG, , shengyr.06@sem.tsinghua.edu.cn,

2 Contents I. Bacground... 3 II. Introduction... 3 III. Our Model-Overall View... 4 IV. Determination of Insurance value via Monte-Carlo Stimulation... 7 V. Our Model-Stoc Maret Optimal Choice under Short Sell Limit... 8 VI. Computing Process and Numerical Result... 9 VII. Conclusion Appendix A - Stoc to be Selected Appendix B - R code Appendix C R result Reference / 17

3 I. Bacground This article is for a problem described as follows: Suppose you are an investor Just graduated 200,000 annual salary 500,000 saving How to allocate your assets into: Ban deposit Treasures Stocs Mutual funds Real estate Others The optimal choice of investment has long been a difficult problem for investors. The structure of people s utility with respect to time and ris influences behaviors of various people. Along with the mainstream method used by scholars, we use well defined utility function to describe investor s attitude toward ris and return. As we now, there are some existing theories describing the optimal investment choice, i.e. Marowitz method. However, most existing theories are based on relatively complete/perfect maret, where short selling of assets is allowed. In our opinion, we must consider this short sell limit in order to avoid discrepancy, which may be large in a maret with short sell limit. As a result, we are going to put forward our analysis under short sell limit on stoc maret. II. Introduction In section III, we show the basic structure of our model and integrate strategy for different investment categories into one universal maximization problem. In section 3 / 17

4 IV, we further discuss the Monte-Carlo simulation procedure of parameters used in section III. In section V, we give a detailed analysis on stoc maret with short sell constraint, using an approach otherwise identical to Marowitz. In section VI, we apply some data into our model and compute the results. Section VII concludes. III. Our Model-Overall View To adequately characterize the long-run investment decision of a utility maximizing individual, we need to set up a discrete, multi-period model. We assume that the individual lives for M years, and that his lifetime well-being is determined by the yearend consumption of each period. Furthermore, we assume that he or she has time-additive and state-independent utilities. Thus, the maximization problem can be stated as follows: M max{ E( U ( D ))} 1 subject to feasible set of Where D is his yearend consumption at th period, U is his corresponding utility in th year, and vector is our investment strategy. For simplicity, we assume U has uniform structure and a discount factor r along the timeline, that is: r U e u( D ) Next, we will specify the investment strategy. We classify our investments into four categories: risy assets, risless asset (borrow), risless asset (lend/save), house purchase and insurance product. Denote the money invested on each category in the th year is j. Notice insurance is invested only at time 0, we will simplify its purchased amount as 4. Here we denote housing expense every year to be H. However, since the strie event for insurance happens on an unpredictable date in some future time and may not be exactly the same yearend consumption realization 4 / 17

5 date. Therefore we need to mae a little modification in the model, introducing a continuous random variable T called survive time which is the interval starts from time 0 until the strie event happens. We assume T follows some distribution F(t). Once the strie event happens, the person s annual income drops from I H to I L. The expected utility gain from insurance is, L is the lump sum payment once the strie event happens: M ru rt V E( e u( L)) e u( L) Pr[ T du] With this factor included, the full model is: M M r ru max (, ) ( ) 4 Pr H E e u D e u L T du u House D I ( I I ) T (1 R ) s T H L H L j j, 1 j 4 j1 j1 where T s.t. 1, T 0, T D D 0,j=1,2,3 jm Where I H and I L are the given annual income, R 1 is the random return on risy portfolio, R2 r is the ris free (borrow) rate, R3 r is the ris free (lend/save) rate, s is the yearly spread payment of insurance. D is the survival level annual consumption. Now suppose E( u( D)) ae( D) b var( D), the above equation becomes, assuming different category of assets are independent with each other: 5 / 17

6 M r max (, ) ( ) var( ) H e ae D b D V u House E( D ) I ( I I )(1 F( )) (1 E( R )) s (1 F( )) L H L j j, 1 j 4 j1 j1 var( D ) ( I I ) F( )(1 F( )) var( R ) s F( )(1 F( )) H L 1, The value of V can be derived from Monte-Carlo stimulation, we will discuss this later. Next, we are going to discuss the term H. In order to mae our planning applicable for computer, we set up a binary vector B, where B 0, not buy house at this year 1, buy house (pay initial) this year We can easily reach the result that: Ip Ap Ip H Ap Ap B, Ip=Initial payment, Ap=Annual payment 0 Ap Ip We denote the transformation matrix as P Note that the optimization problem is just a quadratic problem lie: 1 T T max x Σx c x 2 st.. Ax b with x 1,1, 1,2 1, M, 2,1 2, M, 3,1 3, M, H1 HM, 4 This problem is equivalent to: 1 T T max y Σy d y 2 st.. Dx e with y 1,1, 1,2 1, M, 2,1 2, M, 3,1 3, M, B1 BM, 4 T T 6 / 17

7 and d T c T P This is always applicable since P 0. IV. Determination of Insurance value via Monte-Carlo Stimulation In order to specify the mechanism of insurance product in a sense of its contribution to life-time utilities, we introduce a latent variable model. Suppose Y is a latent variable follows standard normal distribution. When Y falls below a critical value, the strie event in insurance contract happens, id est: Pr[ T t] F( t) Pr[ Y y] ( y) Therefore there is a one-to-one mapping between Y and T: T F 1 ( ( Y)) Since T is a rv in nature, we need to carry out a Monte-Carlo Stimulation to determine the expected utility gain from insurance products. Throughout this session we assume F(t)=1-e -ht ; where h is a constant hazard rate, its mathematical meaning is the strie event conditional probability density at any time. log(1 ( Y)) T h The following table gives input parameters for the stimulation: Parameters Input value Hazard rate: h 0.06 Lump sum payments: L Spread payment: s 500 For h =0.06, the probability that the strie event will not happen within 30 years 7 / 17

8 is about 17%. After stimulation runs over Y, we have V = u( 4 L) V. Our Model-Stoc Maret Optimal Choice under Short Sell Limit Firstly, in classical model, where short selling of assets is allowed, the optimal problem has simple solution as follows: w 1 M f Σ e E r H Where Er A 1 T Σ -1 e B e T Σ -1 e C 1 T Σ -1 1 M 2 D BC A r A D C 2 C r f C, A H B 2Arf Crf 2 Actually under Marowitz s world, it is the solution of the maximizing Sharpe Ratio. Here in our analysis, we also use Sharpe Ratio as a criterion of portfolio evaluation. However, we assume that short sell is limited, as in Chinese stoc maret. The mathematical description of our analysis is as follows: max w E rw r f w subject to T 1w1, wi 0, i There is no simple solution to this non-linear planning problem as that of Marowitz. Hereby we use software R (R Development Core Team, 2008), pacage 8 / 17

9 Rdonlp2 to solve this problem. Secondly, there are too many stocs in the maret, specifically more than 1600 in A share maret. From a general view these stocs are to some extent homogeneous, since they are often highly related. In our analysis, for the convenience of coumputing, we choose several stocs from the same industry, id est 深 发 展 A (000001) from Financial Institution industry, et cetera. Hereby we choose 143 stocs from the stocs in A share maret, listed in Appendix A (page 12). VI. Computing Process and Numerical Result The data we choose are as follows: Stoc: 143 A stocs as in Appendix A (page 12). Utility discount rate: 3% Ris free rate (Borrow): 6.5% Ris free rate (Lend/save):2.5% House initial payment: $1,800,000 House annual payment: $150,000 (other input data are in Appendix B-R code) The numerical result is: Choose these stocs as a mutual fund. 变 量 No. 股 票 号 码 股 票 名 称 比 例 万 科 A 许 继 电 气 攀 渝 钛 业 云 南 白 药 长 春 高 新 盐 湖 钾 肥 江 淮 动 力 双 汇 发 展 哈 飞 股 份 云 天 化 / 17

10 贵 州 茅 台 东 方 电 气 Invest in stoc maret by buying following pieces of mutual fund: Year Amount Year Amount The borrowing behavior is: Borrow pieces= $25, at year 6, zero otherwise. The saving behavior is: (zero otherwise) Year Amount 1 $ 689, $ 895, $1,107,349 4 $1,324,284 5 $1,546,643 Buy house at year 6. Besides, the mutual fund has the frontier lie: 10 / 17

11 pltreturn Optimal Choice under Short Sell Limit with Sharpe Ratio as Criterion Among Multiple Assets pltsigma2 The straight line in the graph above is the original frontier without short sell limit, which appears to be much larger than our frontier. Note that that frontier is a parabolic curve at a proper scale. VII. Conclusion a) Short sell limit plays an important role in Chinese stoc maret From the graph we showed above, we can see that with limit of short sell, the feasible set shrins much compared to situation without short sell constraint. So, we must consider short sell limit in Chinese stoc maret. b) Stoc maret as an asset of investment should be careful considered In our optimal solution, mutual fund is invested at relatively a very low level. This implies one should better choose other investment assets other than stoc. c) House purchase is the core of life due to large amount of utility it brings All the financing method in our analysis, including borrowing and saving, contribute to the final target of buying a house. The optimal choice indicates one 11 / 17

12 should save money for 5 years, and as long as he has enough or near to enough money, he purchases a house with small amount of loan. This solution is very close to modern Chinese young people. d) Further wor to be done The overall utility assumption also has its limitations. The state independent assumption tends to be invalid since the insurance is introduced in this model. The utility of house can be decided by more factors. Besides, the period of house payment and/or payment structure should be flexible. Appendix A - Stoc to be Selected No 股 票 号 码 股 票 名 称 No 股 票 号 码 股 票 名 称 No 股 票 号 码 股 票 名 称 深 发 展 A 顺 鑫 农 业 蓝 星 新 材 万 科 A 云 南 铜 业 维 维 股 份 中 国 宝 安 双 汇 发 展 恒 顺 醋 业 南 玻 A 南 宁 糖 业 ST 中 农 中 粮 地 产 中 关 村 上 海 家 化 ST 深 泰 神 火 股 份 振 华 港 机 深 鸿 基 华 西 村 长 春 燃 气 中 航 地 产 紫 光 股 份 山 东 高 速 德 赛 电 池 中 国 重 汽 恒 丰 纸 业 中 金 岭 南 东 方 钽 业 广 东 明 珠 农 产 品 煤 气 化 湘 电 股 份 深 圳 机 场 新 中 基 江 淮 汽 车 许 继 电 气 华 工 科 技 贵 州 茅 台 冀 东 水 泥 诚 志 股 份 中 铁 二 局 金 融 街 隆 平 高 科 交 大 昂 立 沈 阳 机 床 三 九 医 药 老 白 干 酒 吉 林 化 纤 邯 郸 钢 铁 海 螺 水 泥 东 阿 阿 胶 中 国 国 贸 新 华 医 疗 徐 工 科 技 首 创 股 份 光 明 乳 业 富 龙 热 电 华 能 国 际 北 大 荒 SST 华 塑 中 海 发 展 大 众 交 通 攀 渝 钛 业 中 国 石 化 百 联 股 份 四 环 生 物 招 商 银 行 *ST 白 猫 美 的 电 器 歌 华 有 线 豫 园 商 城 云 南 白 药 哈 飞 股 份 交 大 南 洋 12 / 17

13 靖 远 煤 电 双 鹤 药 业 陆 家 嘴 泰 山 石 油 南 京 高 科 工 大 高 新 东 北 电 气 宇 通 客 车 常 林 股 份 桐 君 阁 上 海 梅 林 西 单 商 场 东 北 制 药 同 仁 堂 中 粮 屯 河 蓝 星 清 洗 长 航 油 运 锦 江 股 份 海 螺 型 材 啤 酒 花 浪 潮 软 件 攀 钢 钢 钒 云 天 化 南 京 熊 猫 铜 陵 有 色 同 方 股 份 宁 波 海 运 长 春 高 新 重 庆 路 桥 昆 明 机 床 东 方 电 子 国 金 证 券 上 海 机 电 燕 京 啤 酒 包 钢 稀 土 上 海 医 药 中 色 股 份 东 方 航 空 通 化 东 宝 西 飞 国 际 乐 凯 胶 片 东 方 电 气 盐 湖 钾 肥 中 青 旅 火 箭 股 份 中 水 渔 业 中 国 船 舶 伊 利 股 份 酒 鬼 酒 香 江 控 股 南 京 化 纤 北 京 旅 游 太 原 重 工 张 江 高 科 云 铝 股 份 莲 花 味 精 烟 台 冰 轮 兖 州 煤 业 江 淮 动 力 长 城 电 工 鲁 西 化 工 中 牧 股 份 贵 糖 股 份 伊 力 特 唐 山 陶 瓷 青 岛 碱 业 五 粮 液 钱 江 水 利 Appendix B - R code library("rcplex") library("matrix") library("xlsreadwrite") library("rdonlp2") #Read libraries #Parameters, 1 =$1000 MonteCarloNumber=10000 InitialSaving=500 years=30 HouseInitial=1800 HouseAnnual=150 HouseY=10 r=0.03 #Times of experiment in Monte-Carlo #Initial saving #Total Years #House Initial payment #House annual payment #House payment years #Utility Discount rate 13 / 17

14 rbottom=0.025 #Ris free (lend,save) rate rtop=0.065 #Ris free (Borrow) rate L=30 #Insurance lamp-sum payment HouseUtil=3500 #utility of House, in money sense h=0.06 #hazard rate s=.5 #spread of Insurance Dbottom=10 #lowest living expense Ih=200 #Annual Salary Il=10 #Annual salary after losing job B=3 #ris averse rhouse=0.0 #House price growth rate yy=rnorm(montecarlonumber,0,1) tt=-log(1-pnorm(yy))/h vv=exp(-r*tt) V=mean(vv) #Calculate V using Monte Carlo start=ceiling(mean(tt)) F<-function(x){ tst=1-exp(-x*h) tst }#Define F #start to optimize stoc fn <- function(x){ -(expectret%*%x - rbottom/12)/ sqrt(+t(x)%*%sigma%*%x) #Min fn.} inmat=read.xls("s2o.xls",sheet=1,type="double") n=ncol(inmat) sigma=cov(inmat);expectret=apply(inmat,margin=2,fun=mean) p <- rep((1/n),n) #initial value par.l <- rep(0,n); par.u <- rep(1,n) #Parameter range lin.u <- 1; lin.l <- 1 A <- t(rep(1,n)) ret <- donlp2(p, fn, par.lower=par.l, par.upper=par.u, A=A, lin.u=lin.u, lin.l=lin.l, name="stocopl") rstoc<-expectret%*%(ret$par) sigma2stoc=t(ret$par)%*%sigma%*%(ret$par)*12 sigma2stoc=as.numeric(sigma2stoc) rstoc=as.numeric(rstoc) #vector=(stoc1,stoc2...stoc30,borrow1,borrow2..borrow30,save1...sa ve30,b1...b30,insurance) tempm=diag(houseinitial,years) for (i in 1:years){ for (j in (i+1):(i+housey)){ if (j<=years ) tempm[j,i]=houseannual } tempm[i,i]=houseinitial*exp(i*rhouse)} transform_mat_q=as.matrix(bdiag(diag(1,3*years),tempm,1)) 14 / 17

15 term1_v1=matrix(0,1,4*years+1) for (i in 1:years){ if (i<years) term1_v1[i]=-exp(-i*r)+exp(-(i+1)*r)*(1+rstoc) else term1_v1[i]=-exp(-i*r) }#coefficient of stoc for (i in 1:years){ if (i<years) term1_v1[i+years]=+exp(-i*r)-exp(-(i+1)*r)*(1+rtop) else term1_v1[i+years]=-exp(-i*r) }#coefficient of borrow for (i in 1:years){ if (i<years) term1_v1[i+2*years]=-exp(-i*r)+exp(-(i+1)*r)*(1+rbottom) else term1_v1[i+2*years]=-exp(-i*r) }#coefficient of save for (i in 1:years){ term1_v1[i+3*years]=-exp(-i*r) }#coefficient of house mysum=0 for (i in 1:years){mysum=mysum+1-F(i)} term1_v1[1+4*years]=v*l-s*mysum term1_v2=matrix(0,1,4*years+1) for (i in 1:years){ term1_v2[i+3*years]=exp(-i*r)*houseutil} term1=term1_v1%*%transform_mat_q+term1_v2 myc=t(term1) #first order o. mysum=0 for (i in 1:years){mysum=mysum+(1-F(i))*F(i)} Q=as.matrix(bdiag(diag(sigma2stoc,years),diag(0,3*years),s*s*mysum)) #second order o. a1=diag(-1,years) for (i in 2:years){ a1[i,i-1]=1+rstoc} a3=diag(-1,years) for (i in 2:years){ a3[i,i-1]=1+rbottom} a2=diag(1,years) for (i in 2:years){ a2[i,i-1]=-1-rtop} a4=diag(-1,years); a4=a4%*%tempm a5=rep(-s,years) for (i in start:years){a5[i]=0} a=cbind(a1,a2,a3,a4,a5) al=t(rep(0,4*years+1)) for (i in (3*years+1):(4*years)){al[i]=-1} a=rbind(a,al) #Amat o 15 / 17

16 b=rep(dbottom-ih,years+1) for (i in start:years){b[i]=dbottom-il} b[years+1]=-1 b[1]=b[1]-initialsaving #bvec o. Q=Q*(-2)*B myvtype=rep("c",4*years+1) for (i in (3*years+1):(4*years)){myvtype[i]="B"} ub=rep(inf,4*years+1) for (i in (4*years-HouseY+1):(4*years)){ub[i]=0} myresult<- Rcplex(cvec=myc,Amat=a,bvec=b,Qmat=Q,lb=0,ub=ub,objsense="max", sense="g", vtype=myvtype) print(myresult) Appendix C R result The optimal choice of stocs under short sell limit, whose parameter is not zero: 变 量 No. 股 票 号 码 股 票 名 称 比 例 万 科 A 许 继 电 气 攀 渝 钛 业 云 南 白 药 长 春 高 新 盐 湖 钾 肥 江 淮 动 力 双 汇 发 展 哈 飞 股 份 云 天 化 贵 州 茅 台 东 方 电 气 Final MIQP result is: (total years=30) [1] e e e e e-02 [6] e e e e e-02 [11] e e e e e-03 [16] e e e e e-03 [21] e e e e e-03 [26] e e e e e+00 [31] e e e e e+00 [36] e e e e e+00 [41] e e e e e+00 [46] e e e e e+00 [51] e e e e e / 17

17 [56] e e e e e+00 [61] e e e e e+03 [66] e e e e e+00 [71] e e e e e+00 [76] e e e e e+00 [81] e e e e e+00 [86] e e e e e+00 [91] e e e e e+00 [96] e e e e e+00 [101] e e e e e+00 [106] e e e e e+00 [111] e e e e e+00 [116] e e e e e+00 [121] e+00 Reference Huang, Chi-fu and Litzenberger, Robert H Valuation of Comlex Securities with Preference Restrictions. Foundations for Financial Economics. s.l. : Prentice Hall, 1989, 6. Hull, John Options, Futures and Other Derivatives. s.l. : Pearson Education Ltd., ISBN P. Spellucci, TAMURA Ryuichi. Rdonlp2. R pacage. [Online] R Development Core Team R: A Language and Environment for Statistical Computing. [Online] R Foundation for Statistical Computing, Vienna, Austria Schmidt, Wolfgang Monte-Carlo Variance Reduction-A new universal method for multivariate problems woring paper. Sharpe, W Capital asset prices: A theory of maret equilibrium under conditions of ris. Journal of Finance. 1964, 19, pp / 17