Black Litterman Model: Indian Experience
In this article: Introduction Literature Review Theoretical Framework Methodology of model implementation Implementation in Indian Stock market Conclusion
INTRODUCTION
Portfolio is the group of stocks taken together by an investor. Investors create portfolios to diversify their risk as well as to beat the market and generate excess return. Portfolio management is art for some and for others it is a science. Portfolio management essentially consists of deciding which securities should be there on basis of risk and return and also how much amount to invest in individual securities. Harry Markowitzs article Portfolio Selection in (1952) laid the foundation of what is known as modern Portfolio theory or the Mean Variance theory. In this article, Markowitz proposed that there is tradeoff between risk (calculated in form of variance) and return (calculated in form of mean) and in general investors are risk averse in nature. Markowitzs Mean variance framework has since then been criticized and improved upon by various scholars. One of the recent enhancements was the work done by Robert Black and Bob Litterman (1991). They proposed a model known as Black-Litterman (BL) model, which takes into consideration not just the quantitative aspects of portfolio management but also takes into account the qualitative aspect. In this model the optimal weights of the securities is combination of the expected returns and the views that investor has with regards to certain (or all) securities. In other words, the Black Litterman model gives more weight to those securities for which investor has positive views and less weight for which investor has negative views. This method thus avoids the extreme allocations which securities have when mean
variance method is followed. Since Black and Litterman gave the model in their original paper in 1991 various modifications have been proposed by various scholars. But still few empirical studies have been done to implement this model in equity markets and fewer in the Indian equity market. This is partly because of the qualitative nature of the model and also in the original paper no clear framework was given.
Problem
Practically all the financial institutions such as asset management firms encounter the problem of managing the portfolio of different asset classes so as to maximize the return for a given level of risk. They also have to deal with problem of optimal capital allocation to each asset class. Most of the times, such decisions are either based on mathematical models or pure intuitions or combination of both. Black Litterman model combines both quantitative as well as qualitative aspects. It allows investors to specify views on one or more assets in the portfolio. But the views are highly subjective in nature and vary from investor to investor.
LITERATURE REVIEW
The Black-Litterman model was presented in a Goldman Sachs fixed income research paper in 1991 by Fisher Black and Robert Litterman. This paper was then published in the Journal of Fixed Income in 1991. The revised paper was published in Financial Analysts Journal (FAJ) in 1992 but this time the paper was inclusive of the rationale for the methodology, and some derivation of the formulas was given, but did not describe all the formulae or gave a full derivation. This resulted in implementation problems and reproducing results became very difficult.
Bevan and Winkelmann (1998) gave a detailed description of how they implemented Black Litterman model & re-iterated that an important feature of the Black-Litterman framework is that investors should take risk where they have views, and the most risk should be taken where they have the strongest views. Christadoulakis (2002) studied the aspects of Bayesian inferences, and applied concepts of Bayesian theory to the assumptions of the model and developed key formulae for posterior returns. Meucci has done commendable work (2005; 2006; 2008; 2010) in this regard and has made it lot easier to understand and implement the model. Meucci through his paper in 2005 extended the method of non-normal views in Black-Litterman model to any kind of the parameter, thus permitting scenario analysis. Izdorek in 2005 simplified the Black Litterman model so that even the nonquantitative investors can use it. Becker and Gutler (2009) explored the estimation of confidence in views by using the analysts forecasts with the dividend discount model and by Monte Carlo simulation. Mankert, C. (2006), incorporated behavioral finance in her discussion on Black Litterman model. In her paper she introduces the factor of home bias, which increases the riskiness of the foreign asset and influences the portfolio weights through the levels of confidence. All the literature review that has been done above is in context of developed markets and only similar study which exists in Indian scenario is one given by Alok Kumar Mishra1, Subramanyam Pisipati and Iti Vyas in 2011.
problem by using both qualitative as well as quantitative aspects of finance. Qualitative aspects are the views which investor has for any security. Black and Litterman in their paper used the market equilibrium portfolio as a starting point and then incorporated the views into the model which tilted the weights towards those securities where the investor was bullish and gave lesser weights to those securities for which investor was bearish. But Meucci in 2009 showed that one can achieve the same results if the starting point is current portfolio of investor or any market index such as BSE-30, NSE-50 etc., thus removing the role of equilibrium portfolio. The views can be of two types:
i. Absolute view:
The views are absolute if investor believes that the return from a certain security will increase or decrease by a specific percentage.
ii. Relative view:
A relative view is one in which investor believes that the return from a certain security will increase or decrease in comparison to some other security in the portfolio. Now the confidence in the views is applied to the model. By confidence it mean the percentage with which investor has given the absolute or relative review. The views and the confidence in them help investor to take more risk where they have stronger views and less risk where they have weaker view. Black Litterman model then uses the Bayesian approach to combine the subjective views of the investor with the equilibrium expected returns to yield expected return vector.
BAYESIAN THEORY
THEORETICAL FRAMEWORK
The Black Litterman is one on its kind which tries to solve the classical portfolio selection
The Black Litterman model eliminates the problem of corner solutions and thus creates more stable portfolio. This happens as the model combines the subjective views of the investor
with the empirical data. This is done using Bayesian probability. The model thus updates the expected excess returns with the views and is able to generate the equilibrium returns. The Black Litterman model uses Bayesian Inference for this purpose. One of the core assumptions of the model is that the security returns are normally distributed. In the Black Litterman model the prior equilibrium distributions are clubbed with the views and uncertainty in views and results in new combined return distribution1.
E[R] = combined return vector (Nx1 matrix) Tau () = weight on views scalar = implied excess equilibrium returns (Nx1 matrix column vector) = variance-covariance matrix (NxN matrix). P = matrix of the assets involved in views (KxN matrix) = uncertainty in the views and is a diagonal covariance matrix of error terms
View-Matrix and Expected Return-Matrix
The View-matrix is a KxN matric and represents the assets involved in the views. It is denoted as P in the Black Litterman model. In this matrix each row represents one view and each column represents a company. As mentioned earlier there are two kinds of views absolute and relative, therefore the sum of the rows will be one or zero respectively. Lets understand the P and Q matrix with help of an example. [ ]
Figure 1: Graphic illustration of the BlackLitterman Return Distribution Black-Litterman Formula
The combined return vector equation which is also sometimes dubbed as master formula of the Black Litterman model.
The first row sums to one which means that this is an absolute view. The second row sums to zero and is a relative view where the investor believes that the asset 1 will outperform asset 4 over the next period. The third row is again a relative views and gives the investor view that asset 3 will outperform asset 6. The Expected return-matrix (Q) is a column vector which consists the estimated returns for each view. This matrix is helpful in a sense that
Where,
1
Idzorek, (A STEP-BY-STEP GUIDE TO THE BLACKLITTERMAN MODEL: Incorporating user-specified confidence levels, 2004)
the view matrix P shown above just specifies that the investor has some views for one or more of the securities but it doesnt quantify these views. The size of the views is shown by the Q matrix. The first element of Q matrix is -6% which depicts that the investor is confident that the security 2 will underperform by giving -6% returns. Thus the model will give less weight to the security 2. The third element is 5% and shows that investor believes the security 6 will outperform security 3 by 5 %.
Variance of Views- Matrix
of and some of them even ignore it. It is also known as weights on views. In most of the research papers available today its value is being taken between 0 and 1. In their original paper from 1992, Black and Litterman took the value of tau near to zero2.
Implied Excess Equilibrium Returns
The implied excess equilibrium returns are derived from the equilibrium returns that clear the market. These are calculated using equation:
is a KxK matrix and it represents the uncertainty in the views. The basic assumption of Black Litterman model is that the views are uncorrelated. This results in which is diagonal in nature. It only consists of variances and the covariance is zero. The variance of views matrix is a square matrix with number of rows and columns equal to number of views that investor has used. [ ]
According to Meucci
Where = Implied Excess Equilibrium Return Vector (Nx1 column vector); Risk aversion coefficient; Covariance matrix of excess returns (NxN matrix); and, Market capitalization weight (Nx1 column vector) of the assets The risk-aversion coefficient () is a parameter that showcases the tradeoff between return and risk. The low value of A means that the investor takes more aggressive bets, resulting in a high risk-high return portfolio.
Where, = variance-covariance matrix (NxN matrix). P = matrix of the assets involved in views (KxN matrix)
Weight-On-Views, Tau
is a scalar quantity and is one of the most confusing components in the Black Litterman model. Every research paper uses its own value
(Asset Allocation: Combining Investors views with Market equilibrium, 1990)
METHODOLOGY
The general structure 3 that is taken while implementing this model is as shown below:
5. Next the risk aversion coefficient and implied equilibrium excess return vector are computed. 6. Now with all the inputs available the expected return vector is calculated using the Black Litterman formula. 7. New covariance matrix is formed by summing the original variance covariance matrix with the matrix obtained from first part of Black Litterman formula as shown below: 8. New weights are determined and performance is determined by performing the sum product of the new weight vector with the return of the securities in month of February.
To implement this model following steps were followed: 1. Initial weights are calculated using the market capitalization of the securities. 2. Next the variance covariance matrix is created. And also the beta, alpha, historical returns and variance are determined. 3. Next the views are calculated. This is done by first determining the number of analyst ratings in different categories such as buy, hold and sell. Also the share price forecast data was taken which was also in three categories, i.e. high, medium and low. Sum product was done for analyst rating categories and share price forecast. This sum product was divided by total number of analyst rating thus resulting in weighted average analyst view about the change in return of the security. 4. With help of these views view matrix is created. From this the omega matrix which shows variance in the views is determined.
3
Implementation in Indian Stock Market
To test the model in Indian environment the initial daily closing price data for 26 stocks (from BSE-30) was taken for the period 1-January-2004 to 31-January-2013. The risk aversion parameter was calculated and the value came out to be -216.039.The low value of A means that the investor takes more aggressive bets, resulting in a high risk-high return portfolio. Although speaking practically it cannot be so much negative as it implies that investors took very aggressive position in the market during the periods when the markets and economy are not performing well for a prolonged period and also they dont even seem to perform well in coming future. One plausible explanation for so low risk aversion coefficient is that the investors are ready to take huge risk because they believe that the risky positions are going to yield higher returns and their portfolio is going to outperform the market. The initial weights of each of these 26 securities were determined using market capitalization.
Mishra et al. (An equilibrium approach for tactical asset allocation: Assessing Black-Litterman model to Indian stock market, 2011)
Company Name Market Capitalisation on 31-Jan-13 Weights based on Market Capitalization Bharti Airtel 1289261.47 0.0496 BHEL 557440.90 0.0214 Cipla 326889.36 0.0126 Dr Reddy's Lab 325589.26 0.0125 GAIL 434263.24 0.0167 HDF Co 1212518.80 0.0466 HDFC Bank 1525902.61 0.0587 Hero Moto 363980.39 0.0140 Hindalco 222043.77 0.0085 HUL 1023580.63 0.0394 ICICI 1373685.71 0.0528 Infosys 1601616.45 0.0616 ITC 2423930.86 0.0932 Jindal Steel 392887.28 0.0151 L&T 949155.95 0.0365 M&M 546289.38 0.0210 Maruti Suzuki 457091.83 0.0176 ONGC 2906086.11 0.1117 RIL 2862367.26 0.1100 SBI 1635538.65 0.0629 Sterlite 382757.51 0.0147 Sun Pharma 743677.29 0.0286 Tata Steel 393196.21 0.0151 Tata Power 240036.27 0.0092 Tata Motors 806938.68 0.0310 Wipro 1013069.97 0.0389 Total 26009795.84 1.0000
As is evident from table 2 that the black Litterman returns vary by huge amount from the CAPM returns. This is due to the views which we have incorporated while implementing this model. Lets take one example from above table. For BHEL the Black Litterman return came out to be -0.0765 which is opposite of CAPM return of 0.1577. This is because our view regarding BHEL was negative of value -0.091. Same reasons exist for all other variations in return. New weights were determined with help of Black Litterman returns by using the formula
Table 1: Initial weights based on market capitalization
The Black Litterman excess returns obtained are compared with the CAPM returns and this has been tabulated below:
Bharti Airtel BHEL Cipla Dr Reddy's Lab GAIL HDF Co HDFC Bank Hero Moto Hindalco HUL ICICI Infosys ITC Jindal Steel L&T M&M Maruti Suzuki ONGC RIL SBI Sterlite Sun Pharma Tata Steel Tata Power Tata Motors Wipro BL Returns 0.0891 -0.0765 0.0839 0.1182 0.1021 0.0146 0.0367 -0.0142 0.1630 0.0223 0.1210 -0.0127 0.0854 0.2764 0.0634 0.0922 0.1822 0.0894 0.0186 0.0498 0.2045 0.0406 0.1083 0.0332 0.0550 -0.0306 CAPM Returns 0.1441 0.1577 0.1230 0.1166 0.1432 0.1558 0.1444 0.1223 0.1677 0.1225 0.1781 0.1388 0.1289 0.1699 0.1606 0.1520 0.1422 0.1431 0.1630 0.1617 0.1771 0.1127 0.1772 0.1533 0.1657 0.1509
These weights were also different from those based on market capitalization. A tabular comparison of old and improved weights is being shown below:
Bharti Airtel BHEL Cipla Dr Reddy's Lab GAIL HDF Co HDFC Bank Hero Moto Hindalco HUL ICICI Infosys ITC Jindal Steel L&T M&M Maruti Suzuki ONGC RIL SBI Sterlite Sun Pharma Tata Steel Tata Power Tata Motors Wipro New Weights Weights based on Market Cap -0.2593 0.0496 1.4990 0.0214 -0.4545 0.0126 -0.6896 0.0125 -0.3712 0.0167 0.3722 0.0466 0.2677 0.0587 0.5584 0.0140 -0.4308 0.0085 0.3545 0.0394 -0.6016 0.0528 0.2166 0.0616 -0.4787 0.0932 -1.0233 0.0151 -0.1730 0.0365 -0.1816 0.0210 -1.2959 0.0176 -0.4771 0.1117 0.5893 0.1100 0.2704 0.0629 -0.5114 0.0147 -0.1562 0.0286 0.3254 0.0151 0.4020 0.0092 0.5222 0.0310 0.7635 0.0389
Table 2: Comparison between Black Litterman returns and CAPM returns
Table 3: Comparison of new weights and old weights
The negative weights show that short selling is permitted in those stocks. A portfolio is created from these new weights. The comparison between original portfolio, Black Litterman optimized portfolio and the BSE-30 return for one month is being shown below:
extreme corner solution that one encounters when he/she implements mean variance model.
Contributed By:
Amandeep Singh Kabli Management Trainee Morarka Foundation Email Id: [email protected] Amandeep Singh Kabli has done his MBA from FMS, Delhi with majors in finance and operations. He is currently into operations profile in Morarka Foundation.
Figure 2: comparison of performance of Black Litterman portfolio and Old portfolio
As is evident from figure 3 portfolio created after implementation of Black Litterman model outperformed both index and original portfolio by huge values. As a matter of fact both the market and original portfolio underperformed and gave negative returns whereas the Black Litterman optimized portfolio gave huge positive returns. This is majorly because of the views that were specified for each security.
CONCLUSION
Black Litterman model can be effectively utilized for portfolio creation. The proposed approach for incorporating views increases the applicability of the Black Litterman model and makes the subjective parts slightly more consistent. Also by taking into consideration views from large number of analysts helped to develop consistency in the views. Black Litterman model avoids the
