Sie sind auf Seite 1von 9

Laboratory Experiment 4

Empirical Test of CAPM



The CAPM describes that the relation between expected return and risk is linear and beta is the only
measure of risk. That is, for any asset i
] ) ( [ ) (
f mt i f it
R R E R R E + = | (1)
Where,
it
R denotes the return on asset i and
mt
R denotes the return on market portfolio (broad
index) for the month t.
i
| measures sensitivity of stock i to market movements of a security and
defined as

) (
) , cov(
mt
mt it
i
R Var
R R
= | (2)

Testing the validity of CAPM implies testing the following hypotheses:
a) The expected return is linearly related to beta
b) The intercept is equal to the risk-free rate
c) The risk premium is significantly different from zero
The two main approaches for testing asset pricing models are time-series regressions and cross-
sectional regression. We use Fama-Macbeth cross-section regression which is an extension of the
cross-section regression with added advantages. The Fama-Macbeth regression is a two stage
procedure. In the first stage, we use market-model regression on excess return to estimate the betas
as


t i
e
t m i i
e
t i
R a R
, , ,
c | + + = i each for T t , , 2 , 1 = (3)

e
t i
R
.
is the excess returns over the risk free rate
f
R on stock i in period t ,
e
t m
R
,
is the excess
market return at time t ,
i
a is the intercept,
i
| is the slope of the regression and
t i,
c is the
disturbance term.

In the second stage, the cross-sectional regression on the estimated betas are performed at each
period as

t each for N i R
t i t i t
e
t i
, , 2 , 1

, , 0 .
= + + = o | (4)

0
is the intercept,
i
o is the pricing error, is the slope and
i
|

is the estimated value from the


time series regression. The validity of CAPM implies that
0
=0 and is positive and significant. The
t-statistic can be calculated as

T
t
/ )

o

=

where,

is the average cross-sectional regression estimates and standard error )

( o is given as

=
=
T
t
t
T
1

=
=
T
t
t
T
1
2
2
2
)

(
1
)

( o


In the recent past, various alternatives to the CAPM have been suggested. One such alternative is
the conditional asset pricing model. A conditional CAPM is defined when either market risk premium
or beta risk of an asset or both are sensitive to information that affects the markets conditional
expected returns. Due to the influence of microeconomic factors (such as operational changes in the
company, or changes in the internal business environment) and/or macroeconomic factors (such as
the rate of inflation, general business conditions, and expectations about relevant future events),
return distribution, expected returns, and betas vary over time unlike those assumed in the CAPM.


Summary of Work Done
Dataset contains monthly stock prices of 30 companies from June 2008 to 14
th
August 2014.
We calculated the monthly returns of each company and fitted a linear regression between
companys return and markets return to estimate the of each company. Further, we again
fitted a linear regression between of each company and the monthly return of each
company.
After that we assumed varies with time. We estimated conditional for each company
using Kalman filter.

Methodology

We calculated the monthly returns from the stock prices using formula-



We fitted linear regression between company return and market return as follows-


t i
e
t m i i
e
t i
R a R
, , ,
c | + + = i each for T t , , 2 , 1 =

From this, we estimated the values of for each company.

Further we fitted a linear regression between and company returns to estimate lambda.

t each for N i R
t i t i t
e
t i
, , 2 , 1

, , 0 .
= + + = o |

Now, we assumed varies with time, so we fitted a Kalman filter to estimate the conditional month-
wise.

Data:
Dataset contains monthly stock prices of 30 companies from June 2008 to 14
th
August 2014.
Source : Bloomberg

Computational Details:
We estimated the static of 30 companies using a linear regression technique. All the
coefficients are significant for 95 % confidence interval. Then we estimated the coefficients
of linear regression between of each company and its return.

Then we compared the dynamic (calculated using kalman filter) with the static .
The following is a Kalman Code used for our analysis:
Sub beta()
Dim ri(1 To 60, 3 To 32), rm(1 To 60), k, h(1 To 1, 1 To 2), p(1 To 2, 1 To 2), mult1, x(1 To 2, 1 To 1),
alpha, beta, k_new(1 To 2, 1), x_new(1 To 2, 1)
Dim identity(1 To 2, 1 To 2), subt2(1 To 2, 1 To 2), sum, q(1 To 2, 1 To 2), p_final(1 To 2, 1 To 2),
p_rec1, p_rec2, p_rec3, p_rec4

For t = 3 To 32
a = t + 14
al = 2 * a
Cells(1, al) = "Alpha" & t - 2
be = 2 * a + 1
Cells(1, be) = "Beta" & t - 2

alpha = Cells(t + 62, 2)
beta = Cells(t + 62, 3)
p_rec1 = 1
p_rec2 = 0
p_rec3 = 0
p_rec4 = 1

For i = 1 To 60
rm(i) = Cells(i + 1, 2)
ri(i, t) = Cells(i + 1, t)
Next

'Variables are defined
For i = 1 To 60
identity(1, 1) = 1
identity(1, 2) = 0
identity(2, 1) = 0
identity(2, 2) = 1
p(1, 1) = p_rec1
p(1, 2) = p_rec2
p(2, 1) = p_rec3
p(2, 2) = p_rec4
r = 0.2
q(1, 1) = r
q(1, 2) = 0
q(2, 1) = 0
q(2, 2) = r

'Equation 1 is solved
h(1, 1) = 1
h(1, 2) = rm(i)
mult1 = Application.WorksheetFunction.MMult(h, p)
transh = Application.WorksheetFunction.Transpose(h)
mult2 = Application.WorksheetFunction.MMult(mult1, transh)
add1 = Application.WorksheetFunction.sum(mult2, r)
inverse1 = Application.WorksheetFunction.MInverse(add1)
mult3 = Application.WorksheetFunction.MMult(p, transh)
k = Application.WorksheetFunction.MMult(mult3, inverse1)
'Range("i2:i3") = k
'Cells(12, 13) = k(2, 1)

'Equation 2 is solved
x(1, 1) = alpha
x(2, 1) = beta
mult4 = Application.WorksheetFunction.MMult(h, x)
Cells(10, 11) = mult4
eq2 = Cells(10, 11)
subt1 = ri(i, t) - eq2
For j = 1 To 2
k_new(j, 1) = subt1 * k(j, 1)
x_new(j, 1) = x(j, 1) + k_new(j, 1)
Next
'Cells(7, 14) = x_new(1, 1)
'Cells(7, 15) = x_new(2, 1)

'Equation 3 is solved
mult5 = Application.WorksheetFunction.MMult(k, h)
For j = 1 To 2
For k = 1 To 2
subt2(j, k) = identity(j, k) - mult5(j, k)
Next
Next
p_new = Application.WorksheetFunction.MMult(subt2, p)
Range("at7:au8") = p_new

'Equation 4 is solved
For j = 1 To 2
sum = 0
For k = 1 To 2
sum = sum + identity(j, k) * x_new(k, 1)
x(j, 1) = sum
Next
Next
Cells(i + 1, al) = x(1, 1)
alpha = x(1, 1)
Cells(i + 1, be) = x(2, 1)
beta = x(2, 1)

'Equation 5 is solved
For j = 1 To 2
For k = 1 To 2
p_final(j, k) = q(j, k) + p_new(j, k)
Next
Next
'Range("at15:au16") = p_final
p_rec1 = p_final(1, 1)
p_rec2 = p_final(1, 2)
p_rec3 = p_final(2, 1)
p_rec4 = p_final(2, 2)
Next
Next
End Sub

Results
Table 1: Values of
i
a and for each company (snapshot for few companies)
Company No alpha beta
AXSB IN Equity (R2) 0.007135 1.528113
YES IN Equity (R2) 0.013883 1.876925
ICICIBC IN Equity (R3) 0.002304 1.676748
KMB IN Equity (R3) 0.008001 1.608265
HDFCB/F IN Equity
(R3) 0.010923 0.998456
IDBI IN Equity (R1) -0.0056 1.625363
DBNK IN Equity (R1) -0.00041 1.464762
FB IN Equity (R1) 0.010247 1.286325
BOI IN Equity (R2) -0.00247 1.286309
SNDB IN Equity (R1) 0.006126 1.350434
PNB IN Equity (R3) 0.004704 1.267452
UNBK IN Equity (R2) 0.004161 1.111027
SBIN IN Equity (R4) 0.002011 1.313236
RCAPT IN Equity (R3) -0.01196 2.342177
BAF IN Equity (R4) 0.02708 1.512055
LICHF IN Equity (R1) 0.015989 1.627103
INFO IN Equity (R4) 0.007786 0.503071
WPRO IN Equity 0.006966 0.81317
HCLT IN Equity 0.020421 0.786988
TECHM IN Equity 0.011348 1.190646
TCS IN Equity 0.019643 0.592328
OFSS IN Equity 0.011679 0.795136
CRISIL IN Equity 0.020826 0.50617
BIOS IN Equity 0.007827 0.987608

Table 2: Values of lambda 1 and lambda 2 (snapshot)
Date
lambda
1
lambda
2
8/14/2014 0.042574 -0.04428
7/31/2014 0.063242 -0.07611
6/30/2014 0.096675 -0.0086
5/31/2014 -0.1486 0.254116
4/30/2014 -0.01371 0.027768
3/31/2014 -0.08625 0.178681
2/28/2014 0.032593 -0.0165
1/31/2014 0.062118 -0.11165
12/31/2013 0.130689 -0.05142
11/30/2013 -0.014 0.011602
10/31/2013 -0.02381 0.136209
9/30/2013 -0.01322 0.069231
8/31/2013 0.062175 -0.12287
7/31/2013 0.2305 -0.24398
6/30/2013 0.055311 -0.07916
5/31/2013 0.044485 -0.03759
4/30/2013 -0.08999 0.114551
3/31/2013 0.04349 -0.05623
2/28/2013 0.074243 -0.12195
1/31/2013 0.058795 -0.03189
12/31/2012 -0.00178 0.035156
11/30/2012 0.009154 0.046171
10/31/2012 0.012425 -0.02556
9/30/2012 -0.03392 0.148611



Table 3: Summary of parameters lambda 1 and lamda2.

lambda 1
(intercept)
lambda 2
(slope)
mean 0.022751 -0.00426
sd 0.061205 0.086831
t-dist 2.879327 -0.37989

t critical 0.063252 -0.06325

Table 4: Values of lambda1 and lambda 2 using kalman filter
Conditional Case(Using Kalman Filter)
Date Lambda 1 Lambda 2
8/14/2014 0.0063 0.0285
7/31/2014 0.0197 0.0462
6/30/2014 0.0161 0.1118
5/31/2014 0.0127 0.0485
4/30/2014 0.0241 0.1298
3/31/2014 0.0055 0.0247
2/28/2014 0.0107 0.0929
1/31/2014 0.0012 0.0300
12/31/2013 0.0073 0.0285
11/30/2013 0.0196 0.0950
10/31/2013 0.0027 0.0740
9/30/2013 0.0192 0.1316
8/31/2013 0.0222 0.0747
7/31/2013 0.0086 0.0379
6/30/2013 0.0071 0.0485
5/31/2013 0.0096 -0.0370
4/30/2013 0.0167 0.0541
3/31/2013 0.0091 0.0526
2/28/2013 0.0012 0.0252
1/31/2013 0.0029 0.0160
12/31/2012 0.0226 0.0284
11/30/2012 0.0125 0.0371
10/31/2012 0.0063 0.0731
9/30/2012 0.0033 0.0210
8/31/2012 0.0135 0.0066
7/31/2012 0.0090 -0.0412
6/30/2012 0.0154 0.0225
5/31/2012 0.0123 0.0441
4/30/2012 0.0055 0.0261
3/31/2012 0.0088 0.0418
2/29/2012 0.0079 0.0542
1/31/2012 0.0204 0.0561
12/31/2011 -0.0160 0.0657
11/30/2011 -0.0129 0.0210
10/31/2011 0.0104 0.0064
9/30/2011 0.0009 0.0466
8/31/2011 0.0017 0.0084
7/31/2011 0.0082 0.0030
6/30/2011 0.0126 0.0043
5/31/2011 -0.0182 0.0222
4/30/2011 0.0013 0.0528
3/31/2011 0.0016 0.0201
2/28/2011 0.0015 0.0531
1/31/2011 0.0045 0.0075
12/31/2010 0.0166 0.0254
11/30/2010 0.0010 0.0049
10/31/2010 0.0019 0.0408
9/30/2010 0.0084 -0.0109
8/31/2010 0.0138 0.0078
7/31/2010 0.0182 0.0094
6/30/2010 0.0076 0.0212
5/31/2010 0.0110 0.0054
4/30/2010 0.0286 0.0101
3/31/2010 0.0141 0.0037
2/28/2010 0.0186 0.0040
1/31/2010 0.0057 0.0161
12/31/2009 0.0164 0.0171
11/30/2009 0.0009 0.0436
10/31/2009 0.0212 0.0167
9/30/2009 0.0071 0.0010

Mean 0.0042 0.0012
Stand Dev 0.0121 0.0478
Stand Dev/sqrt(T) 0.0016 0.0062
T Statistic -2.6937 0.1866

Analysis of results
Table 1 shows the values of coefficients of linear regression between company return and market
returns. All the coefficients are significant. These are values for individual company for a given
period of time. Tables 2 shows the values of coefficients of linear regression between company
and company return. Table 3 shows the summary of parameter lambda 1 and lambda 2. T- Statistics
shows that all the coefficients are significant for a 95 % confidence level. Table 4 shows the values
of parameters lambda1 and lambda 2 using kalman filter.
Conclusion
We conclude that the returns are more dynamics in nature, hence Kalman filter method to
calculate conditional betas are more efficient to determine the risk return relationship.
We could calculate these conditional betas by any of the methods. In actual practice also,
beta varies with time. This shows that our assumption of constant beta was wrong.

Das könnte Ihnen auch gefallen