Multiple linear regression

1
 Generalising the Simple Model to
Multiple Linear Regression

• Before, we have used the model

yt    xt t=u1,2,...,T
t
• But what if our dependent (y) variable depends on more than one
independent variable?
For example the number of cars sold might plausibly depend on
1. the price of cars
2. the price of public transport
3. the price of petrol
4. the extent of the public’s concern about global warming
• Similarly, stock returns might depend on several factors.
• Having just one independent variable is no good in this case - we want to
have more than one x variable. It is very easy to generalise the simple
model to one with k-1 regressors (independent variables).
 Multiple Regression and the Constant Term

• Now we write
yt  1   2 x2t   3 x3t  ...   k xkt  ,ut=1,2,...,T
t

• Where is x1? It is the constant term. In fact the constant term is usually represented by a
column of ones of length T:

1
1
x1   


1
1 is the coefficient attached to the constant term (which we called  before).
 Different Ways of Expressing
the Multiple Linear Regression Model

• We could write out a separate equation for every value of t:

y1  1   2 x21   3 x31  ...   k xk1  u1
y2  1   2 x22   3 x32  ...   k xk 2  u2
  
yT  1   2 x2T   3 x3T  ...   k xkT  uT
• We can write this in matrix form
y = X +u

where y is T  1
X is T  k
 is k  1
u is T  1
 Inside the Matrices of the
Multiple Linear Regression Model

• e.g. if k is 2, we have 2 regressors, one of which is a column of ones:

 y1  1 x21   u1 
 y  1 x   u 
 2   22   1   2

       2    
     
 yT  1 x2T  uT 
T 1 T2 21 T1

• Notice that the matrices written in this way are conformable.

 How Do We Calculate the Parameters (the  )
in this Generalised Case?
• Previously, we took the residual sum of squares, and minimised it
w.r.t.  and .
• In the matrix notation, we have
 uˆ 1 
 uˆ 
uˆ   2 
 
 
uˆ T 

• The RSS would be given by

 uˆ1 
uˆ 
uˆ ' uˆ  uˆ1 uˆ2  uˆT  2   uˆ12  uˆ22  ...  uˆT2   uˆt2
 
 
uˆT 
 The OLS Estimator for the
Multiple Regression Model

• In order to obtain the parameter estimates, 1, 2,..., k, we would

minimise the RSS with respect to all the s.

• It can be shown that

 ˆ1 
 
ˆ  ˆ 2 
     ( X X ) 1 X  y

 
 ˆ k 
 Calculating the Standard Errors for the
Multiple Regression Model

• Check the dimensions:  is k  1 as required.

• But how do we calculate the standard errors of the coefficient estimates?

• Previously, to estimate the variance of the errors, 2, we used s 2 

 uˆ 2
t
.
T 2
2 u' u
• Now using the matrix notation, we use s  T  k

• where k = number of regressors. It can be proved that the OLS estimator of

the variance of  is given by the diagonal elements of s2 ( X ' X ) 1 , so that
the variance of 1 is the first element, the variance of 2 is the second
element, and …, and the variance of k is the kth diagonal element.
 Calculating Parameter and Standard Error Estimates
for Multiple Regression Models: An Example
• Example: The following model with k=3 is estimated over 15 observations:
y  1   2 x2   3 x3  u
and the following data have been calculated from the original X’s.
 2.0 35 . 
. 10 30 . 
( X ' X ) 1   35
. 10 .  ,( X ' y)   2.2  , u' u  10.96
. 65
10 . 4.3 
. 65  0.6 
Calculate the coefficient estimates and their standard errors.
 Calculating Parameter and Standard Error Estimates
for Multiple Regression Models: An Example
• Example: The following model with k=3 is estimated over 15 observations:
y  1   2 x2   3 x3  u
and the following data have been calculated from the original X’s.
 2.0 35 . 
. 10 30 . 
( X ' X ) 1   35
. 10 .  ,( X ' y)   2.2  , u' u  10.96
. 65
10 . 4.3 
. 65  0.6 
Calculate the coefficient estimates and their standard errors.
 Calculating Parameter and Standard Error Estimates
for Multiple Regression Models: An Example
• Example: The following model with k=3 is estimated over 15 observations:
y  1   2 x2   3 x3  u
and the following data have been calculated from the original X’s.
 2.0 35 . 
. 10 30 . 
( X ' X ) 1   35
. 10 .  ,( X ' y)   2.2  , u' u  10.96
. 65
10 . 4.3 
. 65  0.6 
Calculate the coefficient estimates and their standard errors.

• To calculate the coefficients, just multiply the matrix by the vector to obtain
X ' X 
1
X'y
• To calculate the standard errors, we need an estimate of 2.

2 RSS 10.96
s    0.91
Tk 15  3
 Calculating Parameter and Standard Error Estimates
for Multiple Regression Models: An Example (cont’d)

• The variance-covariance matrix of is given by

 183
. 320
. 0.91
s2 ( X ' X ) 1  0.91( X ' X ) 1   320
. . 
0.91 594
0.91 594
. . 
393
• The variances are on the leading diagonal:

Var ( 1 )  183

. SE ( 1 )  1.35
Var ( 2 )  0.91  SE ( 2 )  0.96
Var (  )  3.93
3 SE (  )  1.98
3

• We write: yˆ  1.10  4.40 x2t  19.88 x3t

1.35 0.96 1.98
 Testing Multiple Hypotheses: The F-test

• We used the t-test to test single hypotheses, i.e. hypotheses involving

only one coefficient. But what if we want to test more than one
coefficient simultaneously?

• We do this using the F-test. The F-test involves estimating 2 regressions.

• The unrestricted regression is the one in which the coefficients are freely
determined by the data, as we have done before.

• The restricted regression is the one in which the coefficients are

restricted, i.e. the restrictions are imposed on some s.
 The F-test:
Restricted and Unrestricted Regressions

• Example
The general regression is
yt = 1 + 2x2t + 3x3t + 4x4t + ut (1)

• We want to test the restriction that 3+4 = 1 (we have some hypothesis
from theory which suggests that this would be an interesting hypothesis to
study). The unrestricted regression is (1) above, but what is the restricted
regression?
yt = 1 + 2x2t + 3x3t + 4x4t + ut s.t. 3+4 = 1

• We substitute the restriction (3+4 = 1) into the regression so that it is

automatically imposed on the data.
3+4 = 1  4 = 1- 3
 The F-test: Forming the Restricted Regression

yt = 1 + 2x2t + 3x3t + (1-3)x4t + ut

yt = 1 + 2x2t + 3x3t + x4t - 3x4t + ut

• Gather terms in ’s together and rearrange

(yt - x4t) = 1 + 2x2t + 3(x3t - x4t) + ut

• This is the restricted regression. We actually estimate it by creating two new

variables, call them, say, Pt and Qt.
Pt = yt - x4t
Qt = x3t - x4t
so

Pt = 1 + 2x2t + 3Qt + ut is the restricted regression we actually estimate.

 Calculating the F-Test Statistic

• The test statistic is given by

RRSS  URSS T  k
test statistic  
URSS m
where URSS = RSS from unrestricted regression
RRSS = RSS from restricted regression
m = number of restrictions
T = number of observations
k = number of regressors in unrestricted regression
including a constant in the unrestricted regression (or the total number
of parameters to be estimated).
 The F-Distribution

• The test statistic follows the F-distribution, which has 2 d.f.

parameters.

• The value of the degrees of freedom parameters are m and (T-k)

respectively (the order of the d.f. parameters is important).

• The appropriate critical value will be in column m, row (T-k).

• The F-distribution has only positive values and is not symmetrical. We

therefore only reject the null if the test statistic > critical F-value.
 Determining the Number of Restrictions in an F-test

• Examples :
H0: hypothesis No. of restrictions, m
1 + 2 = 2 1
2 = 1 and 3 = -1 2
2 = 0, 3 = 0 and 4 = 0 3

• If the model is yt = 1 + 2x2t + 3x3t + 4x4t + ut,

then the null hypothesis
H0: 2 = 0, and 3 = 0 and 4 = 0 is tested by the regression F-statistic. It tests
the null hypothesis that all of the coefficients except the intercept coefficient
are zero.

• Note the form of the alternative hypothesis for all tests when more than one
restriction is involved: H 1: 2  0, or 3  0 or 4  0
 What we Cannot Test with Either an F or a t-test

• We cannot test using this framework hypotheses which are not linear
or which are multiplicative.
e.g. H0: 2 3 = 2 or H0: 2 2 = 1 cannot be tested.
 The Relationship between the t and the F-
Distributions

• Any hypothesis which could be tested with a t-test could have been
tested using an F-test, but not the other way around.

For example, consider the hypothesis

H0: 2 = 0.5
H1: 2  0.5 2  0.5
We could have tested this using the usual t-test: test stat 
SE ( 2 )
or it could be tested in the framework above for the F-test.
• Note that the two tests always give the same result since the t-
distribution is just a special case of the F-distribution.
• For example, if we have some random variable Z, and Z  t (T-k) then
also Z2  F(1,T-k)
 F-test Example

• Question: Suppose a researcher wants to test whether the returns on

a company stock (y) show unit sensitivity to two factors (factor x2
and factor x3) among three considered. The regression is carried out
on 144 monthly observations. The regression is yt = 1 + 2x2t +
3x3t + 4x4t+ ut
- What are the restricted and unrestricted regressions?
- If the two RSS are 436.1 and 397.2 respectively, perform the test.
Critical value is an F(2,140) = 3.07 (5%) and 4.79 (1%).
 F-test Example

• Question: Suppose a researcher wants to test whether the returns on a company

stock (y) show unit sensitivity to two factors (factor x2 and factor x3) among three
considered. The regression is carried out on 144 monthly observations. The
regression is yt = 1 + 2x2t + 3x3t + 4x4t+ ut
- What are the restricted and unrestricted regressions?
- If the two RSS are 436.1 and 397.2 respectively, perform the test.

• Solution:
Unit sensitivity implies H0:2=1 and 3=1. The unrestricted regression is the one
in the question. The restricted regression is (yt-x2t-x3t)= 1+ 4x4t+ut or letting
zt=yt-x2t-x3t, the restricted regression is zt= 1+ 4x4t+ut
In the F-test formula, T=144, k=4, m=2, RRSS=436.1, URSS=397.2
F-test statistic = 6.68.
Conclusion: Reject H0.
 Data Mining

• Data mining is searching many series for statistical relationships

without theoretical justification.

• For example, suppose we generate one dependent variable and twenty

explanatory variables completely randomly and independently of each
other.

• If we regress the dependent variable separately on each independent

variable, on average one slope coefficient will be significant at 5%.

• If data mining occurs, the true significance level will be greater than
the nominal significance level.
 Goodness of Fit Statistics

• We would like some measure of how well our regression model actually fits
the data.
• We have goodness of fit statistics to test this: i.e. how well the sample
regression function (srf) fits the data.
• The most common goodness of fit statistic is known as R2. One way to define
R2 is to say that it is the square of the correlation coefficient between y and y .
• For another explanation, recall that what we are interested in doing is
explaining the variability of y about its mean value, y , i.e. the total sum of
squares, TSS:
TSS    yt  y 
2

• We can split the TSS into two parts, the part which we have explained (known
as the explained sum of squares, ESS) and the part which we did not explain
using the model (the RSS).
 Defining R2

• That is, TSS = ESS + RSS


 ty  y 2
 
 tˆ
y  y 2
  t
ˆ
u 2

t t t
• Our goodness of fit statistic is
ESS
R2 
TSS
• But since TSS = ESS + RSS, we can also write

ESS TSS  RSS RSS

R2    1
TSS TSS TSS
• R2 must always lie between zero and one. To understand this, consider two
extremes
RSS = TSS i.e. ESS = 0 so R2 = ESS/TSS = 0
ESS = TSS i.e. RSS = 0 so R2 = ESS/TSS = 1
 The Limit Cases: R2 = 0 and R2 = 1

yt
y
t

x x
t
t
 Problems with R2 as a Goodness of Fit Measure

• There are a number of them:

1. R2 is defined in terms of variation about the mean of y so that if a model

is reparametrized (rearranged) and the dependent variable changes, R2 will
change.

2. R2 never falls if more regressors are added to the regression, e.g.

consider:
Regression 1: yt = 1 + 2x2t + 3x3t + ut
Regression 2: y = 1 + 2x2t + 3x3t + 4x4t + ut
R2 will always be at least as high for regression 2 relative to regression 1.

3. R2 quite often takes on values of 0.9 or higher for time series regressions.
 Adjusted R2

• In order to get around these problems, a modification is often made

which takes into account the loss of degrees of freedom associated
2
with adding extra variables. This is known as R , or adjusted R2:

 T 1 
R 2 1  (1  R 2 )
T  k 
• So if we add an extra regressor, k increases and unless R2 increases by
a more than offsetting amount,R 2 will actually fall.

• There are still problems with the criterion- it’s a “soft” rule
 A Regression Example:
Hedonic House Pricing Models

• Hedonic models are used to value real assets, especially housing, and view the asset as
representing a bundle of characteristics.
• Des Rosiers and Thérialt (1996) consider the effect of various amenities on rental values
for buildings and apartments 5 sub-markets in the Quebec area of Canada.
• The rental value in Canadian Dollars per month (the dependent variable) is a function of
9 to 14 variables (depending on the area under consideration). The paper employs 1990
data, and for the Quebec City region, there are 13,378 observations, and the 12
explanatory variables are:
LnAGE - log of the apparent age of the property
NBROOMS - number of bedrooms
AREABYRM - area per room (in square metres)
ELEVATOR - a dummy variable = 1 if the building has an elevator; 0 otherwise
BASEMENT - a dummy variable = 1 if the unit is located in a basement; 0 otherwise
 Hedonic House Pricing Models:
Variable Definitions

OUTPARK - number of outdoor parking spaces

INDPARK - number of indoor parking spaces
NOLEASE - a dummy variable = 1 if the unit has no lease attached to it; 0
otherwise
LnDISTCBD - log of the distance in kilometres to the central business district
SINGLPAR - percentage of single parent families in the area where the
building stands
DSHOPCNTR- distance in kilometres to the nearest shopping centre
VACDIFF1 - vacancy difference between the building and the census figure

• Examine the signs and sizes of the coefficients.

– The coefficient estimates themselves show the Canadian dollar rental price per
month of each feature of the dwelling.
 Hedonic House Price Results
Dependent Variable: Canadian Dollars per Month

Variable Coefficient t-ratio A priori sign expected

Intercept 282.21 56.09 +
LnAGE -53.10 -59.71 -
NBROOMS 48.47 104.81 +
AREABYRM 3.97 29.99 +
ELEVATOR 88.51 45.04 +
BASEMENT -15.90 -11.32 -
OUTPARK 7.17 7.07 +
INDPARK 73.76 31.25 +
NOLEASE -16.99 -7.62 -
LnDISTCBD 5.84 4.60 -
SINGLPAR -4.27 -38.88 -
DSHOPCNTR -10.04 -5.97 -
VACDIFF1 0.29 5.98 -
Notes: Adjusted R2 = 0.65l; regression F-statistic = 2082.27. Source: Des Rosiers and
Thérialt
(1996). Reprinted with permission of the American Real Estate Society.
 Tests of Non-nested Hypotheses

• All of the hypothesis tests concluded thus far have been in the context
of “nested” models.

• But what if we wanted to compare between the following models?

Model 1 : yt  1   2 x2t  ut
Model 2 : yt  1   2 x3t  vt

• We could use R2 or adjusted R2, but what if the number of explanatory

variables were different across the 2 models?

• An alternative approach is an encompassing test, based on

Model
examination of the hybrid model:3 : yt   1   2 x2t   3 x3t  wt
 Tests of Non-nested Hypotheses (cont’d)

• There are 4 possible outcomes when Model 3 is estimated:

 2 is significant but 3 is not
 3 is significant but 2 is not
 2 and 3 are both statistically significant
– Neither 2 nor 3 are significant

• Problems with encompassing approach

– Hybrid model may be meaningless
– Possible high correlation between x2 and x3.