Unit 11 Portfolio Selection: Objectives
Unit 11 Portfolio Selection: Objectives
Unit 11 Portfolio Selection: Objectives
11.1 INTRODUCTION
In the previous unit, we noted that an investor's opportunity set of investments or
portfolios will be defined by the `efficient set'. But we left the question of actually
finding the efficient set unanswered. This unit will first provide a logical approach to
delineating efficient set. We will then discuss some of the practical problems of
implementing this approach, and present another model, known as `single-index
model', that simplifies the portfolio selection process to a great extent. Finally, we
will indicate some other portfolio selection techniques.
11.2 FINDING THE EFFICIENT SET
We may recall that an efficient set is a continuous curve (see Figure 10.3 in Unit 10)
which, in turn, means that there are infinite number of efficient portfolios. This poses
a typical problem to the investors. How can one determine the composition (i.e.,
combination of assets and portfolio weights) of each of an infinite number of efficient
portfolios? Markowitz did contemplate this problem and, in a major breakthrough,
presented a solution algorithm based on `quadratic programming' technique. While a
complete description of the algorithm, which is referred to as Markowitz's `critical
line method', is beyond the scope of this unit, we may give you a rough idea of what
is being accomplished by it:
11.2.1 Constrained Minimization Problem
As we know, an efficient set can be determined by minimising portfolio risk (i.e.,
return variance) for any level of expected return. If we specify the return at some level
and minimize
23
Portfolio Theory risk, we have one point (i.e., a portfolio) on the efficient frontier. Thus, we need to
solve the following constrained minimisation problem:
n n
Minimise variance( σ 2 p ) = ∑ ∑x x σ
i=1 j=1
i j ij
24
Following the Lagrange Multipliers method, we now write the objective function as Portfolio Selection
Minimise Z = σ 2 (p) + λ[R p * - R p ]
or, Minimise Z = [0.012 x12] + [0.04 x22]+ [0.014x1]+ [.004 x1 x2]+0.014+ λ [.01x1 -
0.45 x2+.005]
where λ is known as the `Lagrangain multiplier'. The expression within the bracket
ensures that the return constraint will be always satisfied while minimizing the
variance.
Thus values of x1, x2 and λ for which Z will be minimum, can be obtained by setting
the partial derivatives equal to zero, and then solving the equations simultaneously.
This is shown below:
δz/δx1 = .024x1 + .004x1 + .01 λ -.014 = 0
Portfolio No. 6 7 8 9 10
Expected Return (%) 6.7 7.3 7.9 8.4 9.0
Standard Deviation (%) 14.1 15.4 16.9 18.4 20.0
Risk
In this context, it would be interesting to know the concepts of `corner portfolios' as
introduced by Markowitz. Any set of efficient portfolios can be described in terms of
still a smaller sub-set of efficient portfolios, which Markowitz termed as `corner
portfolios'. The distinguishing feature of two adjacent corner portfolios is that: (a) one
portfolio will contain either all the assets which appear in the other, plus one additional
asset, or (b) all but one of the assets which appear in the other. Thus, while moving
along the efficient frontier curve from one corner portfolio to the next, portfolio weights
will vary until either one asset drops out of the portfolio or another enters. The point (or
the portfolio) at which .a change in the composition of assets takes place marks a new
corner portfolio. For instance, portfolios numbered 1 and 4 in Table 11.1, may be
considered as corner portfolios.
An important property of corner portfolios is that any combination of two adjacent
corner portfolios will result in a portfolio that lies on the efficient set between the two
corner portfolios. For example, if an investor puts 30 per cent of his or her available
funds in the portfolio numbered 1 and 70 per cent in the portfolio numbered 4 (see Table
11.1), then the resulting portfolio of the following composition (or portfolio weights)
will be another efficient portfolio lying between the corner portfolios 1 and 4.
Ashok Leyland : .30 x 58.6 +.70 x 37.7 =44.0%
ACC : .30 x 0.0 + .70 x 33.0 = 23.1%
Grasim : .30 x 41.4 + .70 x 29.3 = 32.9
26
Thus, a computer algorithm may be developed which first determines some Portfolio Selection
successive corner portfolios, and proceeds next to delineate a set of efficient
portfolios lying between every two adjacent corner portfolios. Each of these
portfolios will correspond to a dot in the return-risk space, which can be finally
connected to draw the graph of the efficient set.
It is easy to see that the Markowitz's approach to trace efficient set is extremely
demanding i n its input data needs and computation requirements. This has been
probably best expressed by Markowitz himself : "...it is reasonable to ask security
analysts to summarize their researches in 100 carefully considered variances of
returns. It is not reasonable, however, to ask for almost 5000 carefully and
individually considered covariances". Indeed, while analysts and portfolio managers
are accustomed to thinking about expected rates of return, they are much less
comfortable in assessing the possible ranges of variation in their expectations, and are
usually, not at all accustomed to estimating covariance of returns among assets.
The problem is made more complex by the number of estimates of covariance (or
correlation) required. For a set of 200 shares, for example, we need to compute [200
(200-1)/ 2] = 19,900 covariance. It is unlikely that the analysts will be able to directly
estimate such a staggering number of inputs. Obviously, what we need is an alternate
formula for portfolio variance, that lends itself to easy computation even when we are
dealing with a large set of assets. However, an understanding of Markowitz process
would sharpen your understanding on the portfolio theory and management though
you may not use in your day to day life Markowitz method of portfolio construction
for stocks.
Activity 1
i) Efficient set
…………………………………………………………………………………
…………………………………………………………………………………
…………………………………………………………………………………
…………………………………………………………………………………
…………………………………………………………………………………
…………………………………………………………………………………
…………………………………………………………………………………
…………………………………………………………………………………
…………………………………………………………………………………
The line running through the scatter points is the 'line of best fit', or an estimate of what
is known as a share's `characteristic line'. Algebraically, the characteristic line can be
defined as
Ri = a i + βi R m (11.1)
where
Ri = the return of security i
ai = the components of share i's return that is independent of the market's
performance-a random variable;
Rm = the rate of return on market index-a random variable; and
βi (beta) = the slope. of the characteristic line that measures the expected change
in R, given a change in Rm
It is useful to break the term Ri , in two components:
1. α i (alpha), the expected value of ai ; and
2. ei, the random element with a Mean value of zero.
In terms of graphical presentation (see Figure 11.2) e 1 (or residuals, a s they are
frequently referred to) measure vertical deviations from the characteristic line. With
this, equation (11.1) can now be written as
R i = α i + β i R m + ei (11.2)
where Rm and ei , (both random variables) are conveniently assumed to be not correlated
with each other.
28
It is further assumed that the residuals are not correlated across shares of different Portfolio Selection
companies; that is, ei, is independent of ej for all values of i and j. This is an important
assumption; it implies that the only reason shares vary together, systematically, is
because of a common co-movement with the market. Thus, single-index model assumes
away all other possible effects on shares' returns, such as industry effects.
11.3.2 Systematic Risk and Diversifiable (or Residual) Risk and Covariance of
Returns
With some manipulations of equation (11.2), we get the following important results:
(a) the expected return, R i = α j + β i R m
(b) the variance of share's return, σ 2i = β 2iσ 2 m + σ 2ei where σ 2 m and σ 2ei are
variances of the distribution of Rm and ei, respectively; and
(c) covariance of returns between shares i and j, σ ij = β iβ jσ 2 m
It is apparent from (a) above that the expected return has two components: a unique or
non-market part, a, and market related part, βi Rm. Even though shares have many
common characteristics and, as a result, tend to move together, their numerous individual
and distinguishing properties cause shares to co-move with the market at different rates.
Accordingly, how sensitive a share's price is to changes in the overall market i.e. the
value of its `beta' is of great significance in determining the expected return.
Like the expected return, we can always split the variance of share's returns into two
parts, as shown in (b) above. The first component β 2iσ 2 m , is called the `systematic risk' or
`market risk' of the investment. Since σ 2 m is the same for all shares, systematic risks will
differ among different shares accordingly to the magnitudes of their `betas', βi . Simply
stated, beta measures sensitivity of a share's price movements compared with those of the
market index. Shares having betas less than 1 can be said to be `defensive'. One per cent
increase (decrease) in the market return is likely to be accompanied by a less than one per
cent increase (decrease) in the shares' rate of return. The investors are thus defended to
some extent against the occurrence of major down fall in the market return. On the other
hand, shares with individual beta values greater than one are considered to be more
`aggressive' or more risky, as one per cent increase (decrease) in the market return is
likely to be accompanied by an even greater increase (decrease) in the shares' rates of
return. A beta of one implies `average' riskiness; every one per cent return on the market
is associated with one per cent opportunity return on the share. Beta can be negative as
well, reflecting that share prices can rise when the market falls and vice versa; but this is
normally unusual.
The systematic risk is caused by macro events like oil crisis, an unexpected change in rate
of inflation, etc. The macro events are broad and affect nearly all shares to one degree or
another, and they may have an impact on the general level of stock market. Thus, one
cannot reduce systematic risk by diversifying investment across different shares. That is
why the systematic risk is often called `non-diversifiable' risk.
The second component of variance of share's returns, σ 2 ei , is known as `residual
variance' or `unsystematic risk' or `diversifiable risk'. The source of this kind of risk is
`micro' events, which have impact on individual shares but no sweeping impact on other
shares. Examples include the introduction of a new product(s) or the sudden obsolescence
of an old one. They might also include labour strike lockout or the resignation or death of
a key person in the firm, or splitting up of a business family. Since micro events affect
only the individual shares under consideration, their impact can be reduced to a great
extent by holding a diversifiable portfolio. We will explain how diversification of risk
takes place after a while.
At this point, we may recall that under Markowitz model we are required to compute
covariance of returns for every pair of assets comprising the portfolio. We have also
observed that without having estimates of covariance, one cannot compute the variance of
portfolio returns. However, if the single-index model is a valid description of the process
generating shares returns, there is no need for direct estimates of the covariance. All that
we need to know are the values of share betas and variance of returns on market index;
the covariance between any two shares i and j can next be obtained easily by employing
the
29
Portfolio Theory relationship as noted above. Needless to say that the relationship is much less
demanding in terms of estimation procedure and computation time.
What is more amazing to note here is that the single-index model does not require even
the indirect estimates of covariance of returns between shares. The model provides a
still simpler formula for computing variance of portfolio returns. We will now
explain this.
11.3.3 Variance of Portfolio Returns
We begin by restating that the total risk or variance of returns on share `i' is given by
σ 2i = βiσ 2 m + σ 2 ei (11.3)
Total variance = Systematic Risk + Residual variance
This equation holds-for a portfolio of shares as well. Rewriting the equation for a
portfolio, we get
σ 2i = β 2 p σ 2 m + σ 2 ep (11.4)
Total Portfolio variance = Portfolio Systematic Risk + Portfolio residual variance
Where the subscript `p' denotes a portfolio.
It can be further shown that
n
βp = ∑xβ
i=1
i i
Suppose, an investor is planning to put equal amounts of his investible fund in these
two shares. Then we have
βp = 0.54 x.50+1.13 x -.50=0.56
If σ 2 m is equal to 81.0 per cent, the variance of the returns of the portfolio under
consideration will be given by
σ2p = (.56)2x 81.0 +40.2 =65.6
σp = 8.1%
Let us now add one more share to the above portfolio, say, the share of ACC with a
beta of 1.63 and residual variance of 179.6 per cent. Suppose, the investor decides
once again to invest equal amount. This time σ p will work out to be 11.7 percent,
whereas σ 2 ep , will be 37.8 per cent.
It is interesting to note that while the portfolio's systematic risk component ( β p ) has
increased due to the addition of a more risky share, its non-market related risk
component has declined. Given the single-index model's assumption that residuals
(ei's) of different shares are not correlated (we have already explained this assumption),
it is not difficult to
30
appreciate how a portfolio's residual variance begins to diminish as the number of Portfolio Selection
shares (n) in the portfolio is increased. Assume for a moment that an investor forms a
portfolio by placing equal amounts of his funds into each of n shares. Equation (11.6)
then becomes
n
σ 2 ep = [ 1 ] ∑ [ 1 ] σ 2ei
n i=1 n
where the term within the bracket denotes average residual variance of the shares
comprising the portfolio. As the number of shares in the portfolio gets large,
portfolio's average residual variance falls so rapidly that most of it is effectively
eliminated even for moderately sized portfolios.
At this stage, it would be appropriate to contrast the procedure for computing
portfolio variance as outlined above with that of the Markowitz model. We have
mentioned earlier that for a portfolio of 200 shares, Markowitz model requires 19,900
estimates of covariance. Under the single-index model we need, however, only 200
estimates of beta, 200 estimates of residual variance, and one estimate for the variance
of returns on market index. Indeed, this is a dramatic reduction in the input data for
computing portfolio variance.
But how accurate is the portfolio variance estimate as provided by the single-index
model's simplified formula? If it is the Markowitz formula, we know that the
variance number of perfectly accurate, given, of course, the accuracy of the
covariance estimates. Besides, the formula makes no assumptions regarding the
return generating process. On the other hand, the single-index model assumes that the
market factor solely determines the shares' returns and residuals. are not correlated
across different shares. Thus, the accuracy of the single-index model's formula for
portfolio variance is as good as the accuracy of underlying assumptions. Quite
obviously, the assumptions are not strictly accurate. Many researchers have found
that there are influences beyond the market that cause shares to move together. In
addition, empirical evidence suggests that residuals are correlated to some degree,
which is not altogether unexpected. After all, if something (good or bad) happens to a
company, some other companies, such as its suppliers and competitors, would be
affected simultaneously. The residuals that appear for the shares of these other
company would not, therefore, be independent of each other. However, one can
always expect that the degree of correlation would not be large enough to impair the
relative efficiency with which the single-index model estimate the portfolio variance.
11.3.4 Estimating Beta and the Diversifiable Risk Component
The estimation of beta and the diversifiable risk component of a share involves fitting
a `characteristic line' as shown in Figure 11.2, such that the vertical deviations of the
scatter points from the fitted line are minimized. The statistical procedure for
obtaining a line of best fit is known as `simple linear regression' or `ordinary least
squares method (OLS)'. The beta can be computed manually or using computers.
Today, analysts generally use computers to get beta value. For instance if you have
the monthly returns of a Market Index (like BSE Sensex) and an individual stock's
return (say Grasim) in the Microsoft Excel sheet, you can easily compute the beta
using a function called =SLOPE(Range of Stock Return, Range of Market Return).
At the end of the Unit, the computation of beta is illustrated.
Although the above estimation procedure looks quite straight forward, it is fraught
with several practical problems. For instance, what should be the length of beta
estimation period-two years, three years or more? Or, should we base our calculation
on annual return data? There are many shares which are not regularly traded on the
stock exchange; accordingly, their price quotations remain unchanged even the case
of ill-traded shares? No doubt, the literature on the subject provides some answers to
all such questions, but they need be verified empirically in our context. Unfortunately,
there is dearth of empirical studies with the Indian shares' data. Even if we obtain
satisfactory estimates of historical data, we still face the problem of estimating future
(or ex ante) beta. What is of concern to us is betas for future holding period, and not
the historical betas.
Since large-scale expectational data on returns of individual shares as well as of market
index are not available, one cannot directly estimate future betas by fitting regression 31
Portfolio Theory lines. So, the historical beta must be estimated first and then we can make some
adjustments to it for deriving the future beta.
Activity 1
i) List out two major points of difference between Markowitz's approach and
Sharpe's single-Index Model of selecting optimal portfolio.
………………………………………………………………………………
………………………………………………………………………………
………………………………………………………………………………
………………………………………………………………………………
………………………………………………………………………………
ii) List out relevant data for computing beta of an equity share.
………………………………………………………………………………
………………………………………………………………………………
………………………………………………………………………………
………………………………………………………………………………
………………………………………………………………………………
………………………………………………………………………………
………………………………………………………………………………
………………………………………………………………………………
………………………………………………………………………………
……………………………………………………………………………….
We now take a note of some other portfolio selection models that seem to hold great
promises to practical applications. One such model is the `multi-index model'. There
are different variants of this model and each of them is developed to capture some of
the non-market influences that cause shares to move together (recall that single-index
model accounts for only market related influences). The non-market influences, in
essence, include a set of economic factors or industry characteristics that account for
common movement in share prices. While it is easy to find a set of indices that are
associated with non-market effects over any period of time, it is quite another matter
to find a set that is successful in predicting covariance that are not market related.
There is still a great deal of work to be done before multi-index models consistently
outperform the simpler one.
Another model that takes into account a wide spectrum of practical considerations in
portfolio selection is the goal-programming model. In real life, an investor's goals
and desires transcend. the notion of a trade-off between only risk and return. For
example, an investor may prefer to invest some minimum amount in several different
shares, but at the same time he or she may not like individual investment to exceed a
specified limit. Additionally, he or she may prefer dividend income to capital
appreciation. There may also be a desire not to allow the portfolio beta to be either
above or below a predetermined level. Apart from holding such diverse goals and
desires, the investor may even set the order of
32
their priorities. In this kind of investment problem situation, the goal-programming Portfolio Selection
model is ideally suited to provide an optimal solution. Further goal programming
solution can be easily obtained by available computer packages.
11.5 SUMMARY
This unit has provided some insights into Markowitz's approach to trace the efficient
set. The application of Markowitz's model requires estimation of large number of
covariance. And without having estimates of covariance, one cannot compute the
variance of portfolio returns. This makes the task of delineating efficient set
extremely difficult. However, William Sharpe's `single-index model' simplifies the
task to a great extent. Even with a large population of assets from which to select
portfolios, the number of required estimates are amazingly less than what are
required in Markowitz's model. But how accurate is the portfolio variance estimate as
provided by the single-index model's simplified formula? While the Markowitz's
model makes no assumption regarding the source of the covariance, the single-index
model does. Obviously, the accuracy of the latter model's formula for portfolio
variance is as good as the accuracy of its underlying assumption.
In passing, we have also mentioned in this unit other portfolio selection models, such
as `multi-index model' and `goal programming model' which have high intuitive
appeal but would require much more work before they outperform the simple ones.
34
Appendix Portfolio Selection
Table 11.1: Monthly Return of Sensex and Select Stocks and Beta Computation
Month End Sensex Castro! Colgate Infosys
30-Jan-1997 7.65% -0.12% 6.98% 0.03%
27-Feb-1997 -2.34% 0.18% 11.88% 14.83%
27-Mar-1997 8.98% 5.35% 10.53% 15.27%
5-Apr-1997 2.41% -4.64% -1.73% 19.97%
27-May-1997 -3.48% -2.63% -7.21% 10.62%
27-Jun-1997 11.93% 18.46% -1.47% 23 .79%
25-Jul-1997 1.37% 31.21% 7.45% 12.42%
27-Aug-1997 -2.20% 5.11% 5.14% 30.34%
26-Sep-1997 -4.21% -6.44% -8.84% 21.17%
27-Oct-1997 0.23% -1.27% -0.85% -13.59%
27-Nov-1997 -7.78% 12.78% -6.78% 3.81%
26-Dec-1997 0.15% 3.35% -8.47% -17.87%
27-Jan-1998 -9.83% -7.55% -6.04% -2.68%
27-Feb-1998 10.56% 0.65% 7.79% 22.36%
27-Mar-1998 7.91% -0.36% 8.46% 20.00%
27-Apr-1998 4.46% 0.40% 1.83% 26.28%
27-May-1998 -7.49% -7.15% -11.87% 17.17%
26-Jun-1998 -16.11% -11.72% -9.59% -11.16%
27-Jul-1998 -2.64% -9.72% 12.78% 12.21%
27-Aug-1998 -3.58% 7.03% -19.58% 6.69%
25-Sep-1998 8.36% 9.97% -5.18% -5.18%
27-Oct-1998 -10.I7% 3.87% -12.86% -2.76%
27-Nov-1998 -3.89% -1.15% 0.96% -3.66%
24-Dec-1998 6.48% 12.36% 13.60% 24.04%
27-Jan-1999 13.18% 19.18% 7.72% 59.68%
27-Feb-1999 1.36% -1.80% -14.19% 25.65%
26-Mar-1999 5.82% 0.24% -2.61% -7.42%
26-Apr-1999 -9.79% -5.48% 1.17% -4.39%
27-May-1999 19.02% 8.25% 18.86% 23.61%
25-Jun-1999 6.45% -1.19% -1.02% 14.02%
27-Jul-1999 11.71% 4.82% 27.70% 23.33%
27-Aug-1999 6.04% 2.97% 8.43% 24.44%
27-Sep-1999 -3.18% -14.02% -5.01% 25.43%
27-Oct-1999 0.86% -5.09% -2.35% 6.92%
26-Nov-1999 -1.08% -3.83% -14.41% 28.03%
27-Dec-1999 2.41% -10.10% -2.34% 31.66%
27-Jan-2000 11.43% 13.61% 0.00% 15.45%
25-Feb-2000 4.73% -13.93% -27.40% 19.54%
27-Mar-2000 -8.48% 1.70% -5.97% 19.53%
27-Apr-2000 -9.07% -4.66% -3.01% -23.70%
26-May-2000 -12.71% -3.87% 12.41% -24.82%
27-Jun-2000 14.74% 5.92% 7.36% 36.20%
27-Jul-2000 -8.66% -14.00% 10.91% -16.87%
25-Aug-2000 3.17% 0.57% -16.67% 19.13%
27-Sep-2000 -5.72% -8.96% -0.87% -7.70%
27-Oct-2000 -10.45% -19.15% -3.46% -3.44%
27-Nov-2000 6.44% 24.56% 6.36% 5.3 7%
27-Dec-2000 -2.31% 12.42% -0.61% -27.45%
25-Jan-2001 11.67% -5.00% -10.24% 23.12%
27-Feb-2001 -6.02% 1.74% -1.05% -15.97%
27-Mar-2001 -9.21% -8.86% -12.10% -18.29%
27-Apr-2001 -7.36% -6.28% 10.45% -30.45%
25-May-2001 6.93% 4.79% 0.43% 29.07%
27-Jun-2001 -6.78% -6,06% -7.07% -15.31%
27-Jul-2001 -4.69% -3.64% -0.53% 2,35%
27-Aug-2001 2.05% 23 .55% 2.67% 7.93%
27-Sep-200 -18.17% -2.74% -4.21% -39.81%
25-Oct-2001 11.29% -1.64% 2.21% 28.41%
27-Nov-2001 8.78% -22.79% 0.56% 24.69%
27-Dec-2001 -4.74% -4.79% 1.32% 4.02%
25-Jan-2002 6.40% 3.84% -5.16% -2.33%
Beta 1.0000 0.49735 0.4392 1.5338
35