Portfolio Optimization using Conditional Sharpe Ratio

In this paper we propose a portfolio optimization model that selects the portfolio with the largest worse-case-scenario sharpe ratio with a given confidence level. We highlight the relationship between conditional value-atrisk based sharpe ratio and standard deviation based sharpe ratio proposed in literature. By utilizing the results of Rockafellar and Uryasev [5],we evaluate conditional value- at- risk for each portfolio. Our model is expected to enlarge the application area of practical investment problems for which the original sharpe ratio is not suitable, however should device effective computational methods to solve optimal portfolio selection problems with large number of investment opportunities. Here conditional sharpe ratio is defined as the ratio of expected excess return to the expected shortfall.This optimization considers both risk and return, of which changes will effect the sharpe ratio. That is the fitness function for dynamic portfolio is the objective function of the model.


INTRODUCTION
In portfolio construction process, optimization techniques can provide significant assistance to an investor. They help to accommodate each investor's specific objective. According to Mortowitz's (1952 [2]) seminal work, investors should choose a portfolio having either minimum risk for a given level of return or maximum return for a given level of risk. These tools offer investors a trade-off between mean return and risk. Risk and return are always a couple of conflict. In order to create a powerful risk management tool, investor should face return and risk, simultaneously. An alternative approach is to choose portfolio that, instead of maximizing the riskadjusted return, maximizes the ratio of return to risk. In other words the objective is to find a portfolio that maximizes the sharpe ratio. Sharpe Ratio (SR) presented by W. Sharpe [6] is the ratio of the differential of the return from an investment and return from a benchmark divided by the standard deviation of the return of the portfolio. There the most common traditional risk measurestandard deviation is used which penalizes gains and losses equally Markowitz [2] pointed out that a better risk measure would penalize only losses and he proposed semi-variance as a desired alternative. There is a substantial literature devoted to replacing mean-variance optimization. An alternative risk measure that has gained attention in recent years is Value-at-Risk (VaR). Value-at-Risk (VaR), a widely used performance measure, can solve the problem by finding maximum loss with a given confidence level. Although (VaR) is very popular measure of risk, it has some undesirable mathematical properties such as lack of subadditivity and convexity (Artzner et al. [1]). As a measure of risk, conditional value-atrisk (CVaR) (shortfall) framework can capture all of the nonlinearities and asymmetries of the potential return distribution. An innovation of Rockafellar and Uryasev [4,5] makes portfolio optimization against expected shortfall technically practical by formulating it as a linear programming problem. This paper proposes an optimal portfolio allocation approach using a modified version of the traditional sharpe ratio that we refer to as the 'conditional sharpe ratio' (CVaRSR). The CVaRSR (conditional sharpe ratio) explicitly takes into account the uncertainty involved in estimating the sharpe ratio and takes a more conservative view than the traditional sharpe ratio by including the effects of higher order moments of return distribution. So it is more reliable than traditional portfolio optimization using sharpe ratios.
We test the effectiveness of our CVaRSR approach with a numerical example involving a simple three asset portfolio and simulated returns. In our model, the aim of the investor is to maximize the conditional sharpe ratio.
Due to possible intermediate payments from an asset, we have to deal with re-investment problem which turns original one period model into a multiperiod one. In particular, we provide evidence that this strategy is effective in mitigating market volatility and volatility of the sharpe ratio.
The sections of the paper are arranged as follows. In section 2, we begin with a discussion of downside risk measures -VaR and CVaR, CVaR based sharpe ratio; model formation will be presented in section 3; section 4 discusses the details of our numerical example including simulations and substantial analysis; the final section 5 is reserved for our conclusions.

Downside risk measures
Let R w be the portfolio return associated with portfolio weight vector w. Then loss function R i is the return of ith asset and w = (w 1 ,w 2 ,...,w n ) is corresponding weight vector. Also R i is given by Value-at-Risk with confidence level β is defined as the negative of (1 − β) quantile of R w with β ∈ [0,1] and N possible portfolio outcomes Conditional Value-at-Risk with confidence level β is defined as the expected (average) value of (1 − β)% largest losses.
where R k w are the ordered return scenarios.
Also . here R k w are return scenario and International Letters of Chemistry, Physics and Astronomy Vol. 53 131

2.2.
CVaR based sharpe ratio Conditional sharpe ratio is modified version of traditional sharpe ratio. Traditional sharpe ratio is given by here R is return of the portfolio and R F is risk free rate. σ is standard deviation of the return. In modified sharpe ratio, instead of standard deviation, conditional value-at-risk is used as risk measure. So conditional sharpe ratio is defined as where CVaR β (R) is conditional value-at-risk of the portfolio return R with confidence level β.

MODEL FORMATION
The investor has three investment opportunities on the financial market. Besides stock and bond, the investor invests in option with investment horizon time T.
where r S and r B represent the return of the stock and the bond for the remaining time [T 1 ,T]. In our model, the inner optimization loop optimizes expected return of the option for the optimal choice of re-investment strategy V and the optimal choice of re-investment strategy V and the optimal value of the objective function gives us return of option that is used in outer optimizations loop. The outer optimization loop optimizes conditional sharpe ratio for the optimal choice of initial portfolio weight vector w. So our portfolio optimization model reads as the following 132 ILCPA Volume 53 , subject to the constraints (9) Here the subscripts i and j correspond to the values that occur in simulation run number. Also note that the dimension of the problem is of the order of the number of simulated paths N. However, this also shows that the number of simulation runs determines the size of the problem as considering other investment opportunities would only slightly increase the dimension of the problem. In above model, Eq. (1) is objective function with goal to maximize conditional sharpe ratio. Eq. (2) represents portfolio return at horizon time T for each scenario. Eq. (3) is objective function of inner loop with aim to maximize expected return of option for time [T 1 ,T] for each scenario. Eq. (4) represents option's return of re-investment for each scenario. Eq. (5) and Eq. (7) make sure that portfolio weights add upto 1. Eq. (6) and Eq. (8) guarantee that short selling is not allowed. Eq. (9) gives VaR value. A sequential linear programming approach can be used to find optimal solution that maximize conditional sharpe ratio.

NUMERICAL EXAMPLE AND SUBSTANTIAL ANALYSIS
Real market data of S&P 500 for year 2013 is downloaded from finance.yahoo.com to predict future stock price with the help of model for geometric brownian motion: dS t = µS t dt + σS t dw t S t denotes stock price at time t with volatility σ. µ is the constant drfit of stock price. dw t represents brownion motion with mean zero and variance unity. The stochastic process for short-term market interest rates are assumed to follow vasicek model ∆r = α(b − r)∆t + σε√∆t r is current market rate of interest with volatility σ, b gives long-rum mean of interest rate and α is speed of mean-reversion.
The present value (PV) of bond with yearly coupons using short-term market interest rate is defined as International Letters of Chemistry, Physics and Astronomy Vol. 53 D represents bond's face value with coupon payments C and T is bond's maturity. r t is market interest rate at time t. To calculate price of call option, Black-Schole's model is used: , here S is stock price at time t, T is maturity time and K is strike price. N(d 1 ) gives cumulative normal distribution with volatility σ. We simulate 100 scenarios for the stock price and the interest rate. We consider an investment problem with investment horizon 4 years and assume that call option matures after 1 year. If option ends up in-the-money, then the payoff of the call option is re-invested in the stock and the bond with weight vector V = [V S ,V B ]. The parameter values are given in Table 1. Risk free rate = 4% By using Standard Evolutionary Engine for Analytic Solver Platform of Frontline Systems to simulate paths, the inner loop of the model optimizes re-investment weight vector V .However the outer loop of the model decides optimal value of initial weight vector W. The optimization report is given in Table 2.

134
ILCPA Volume 53  Table 3 shows the superiority of CVaR based sharpe ratio CVaRSR over standard deviation based sharpe ratio SR. From above table, it can easily be seen that CVaR based sharpe ratio, i.e., the ratio of mean excess return to CVaR is useful for improving and evaluation the performance of portfolios. We construct an optimal portfolio consisting of risky assets which maximizes the conditional sharpe ratio and prove the usefulness of the new measure. The results for different strike prices of the call option are given in Table 4. Above results show that by using a call option with high strike price, we can increase conditional Sharpe ratio (CVaRSR) of our portfolio.
But after a certain level of strike prices,the call option leads to total loss as the option ends-up out-of-money. So a call option with very high strike price is the most risky investment. First two columns of Table 4 show that as the strike price of option increases, price of call option decreases. That is one of the reasons for the higher strike price leads to higher sharpe ratio. Moreover, risk of losing money from stock investment and risk of losing money from call option investment go side by side. If the option ends up out-of money, the risk of losing from stock investment has increased substantially.

CONCLUSIONS
This paper consider a new approach to improve performance of the portfolio. In our model, we propose an alternative to the traditional sharpe ratio using the concept of conditional Sharpe ratio for multi-period problem. The numerical example shows the superiority of our approach over the traditional Sharpe ratio. Changing the performance measure can result in a large difference in the optimal-weighted portfolio performance. Thus, it is advisable to study the impact of this by International Letters of Chemistry, Physics and Astronomy Vol. 53 searching for different approaches and analyzing their results. With the acknowledgement of coherence of CVaR, CVaR is widely used to measure risk. Consequently it is important to suggest CVaR based Sharpe ratio. At the same time, this model is a general program, in other words, the model keeps the best answer after each loop. Moreover, this method satisfies common concept that considers both return and risk without emphasizing particularly on each side. Sometimes conservative investors can not accept higher risk, on the other hand, there are some adventurists who seek the largest returns. Using this model it is easy to satisfy both type of investors by changing a few parameters and set some restrictions. The optimization model can handle trading constraints, such as short selling restriction, while still retaining in the class of linear optimization programs. This model has the advantage that the problem is solvable in reasonable time.