Section 5.1 Section 5.2 Section 5.3 Section 5.

Stock & Watson (2019), Chapter 5

Regression with a Single Regressor:
Hypothesis Tests & Confidence Intervals

Brittany Almquist Lewis

FIN 560A 470A

Research Methods in Finance

1/12
Brittany Almquist Lewis
Chapter 5
Section 5.1 Section 5.2 Section 5.3 Section 5.4

Two-Sided Hypotheses Concerning 1

I Testing hypotheses about the population mean

estimator-hypothesized value
I General form of the t-statistic ) ! standard error of the estimator

I Two-sided hypotheses concerning 1

I H0 : "#1 =
$
%
1,0 vs H& 1 :
' (
1 6= 1,0 (two-sided alternative)
!

ˆ 1-
I t= SE ( ˆ 1 )
1,0
) reject if |t act
| > 1.96 (5% level)
Pn ⇣ ⌘ q
1
(Xi -X̄ )ûi2 ˆ 1 = ˆ2
I ˆ 2ˆ = 1
n ⇥ n-2
h P
n
i=1
2
i2 ) SE ˆ1
i=1 (Xi -X̄ )
1 1
n 0/1/0
)*+,*-*.',./

%),(77879:!
>C?@;%<=
!;<=!>?@
6)
)*,2*,/+3'45'3
1'05/
2/12
B
3 !
)3'3
Brittany Almquist Lewis 2 A:B#
Chapter 5
Section 5.1 Section 5.2 Section 5.3 Section 5.4

Two-Sided Hypotheses Concerning 1

I Two-sided hypotheses concerning 1

I p-value = PrH0 (|t| > |t act |)

!

I In large samples ) p-value = Pr(|Z | > |t act |) = 2 (-|t act |)

%
( !

I Reporting regression equations & application to test scores

\ = 698.9 - 2.28 STR, R 2 = 0.051, SER = 18.6
I TestScore
(10.4) (0.52)
&

6 D 3/12
,
Brittany Almquist Lewis
Chapter 5
Section 5.1 Section 5.2 Section 5.3 Section 5.4

One-Sided Hypotheses Concerning 1

E44
%

&
D @G
F H/')#, 3#
I/0*/-
3'3/4
%

I H0 : 1 = 1,0 vs H1 : 1 < 1,0 (one-sided alternative)

ˆ 1 - 1,0
I t= SE ( ˆ 1 )
) reject if t act < -1.64 (5% level)

J
I p-value = Pr(Z < t act ) = (Z < t act )
! !

I When should a one-sided test be used?

KLM !
!

DN A

4/12
Brittany Almquist Lewis
Chapter 5
Section 5.1 Section 5.2 Section 5.3 Section 5.4

Testing Hypotheses About the Intercept 0

N# )544'H*O/ F PQ "2:#37/)*) N/)3*,+ .'0.50'3/ 37/ 3 F )3'3*)3*. )* )// F *-F *3

A

*)
0'H+/H *, 'I)#053/ 1'05/ 37', 37/ .H*3*.'0 1'05/ '3 37/ )*+,*-*.',./
0/1/0 F P',3
2#5
%

I H0 : 0 = 0,0 vs H1 : 0 6= 0,0 (two-sided alternative)

% F !

ˆ 0-
I t= 0,0
SE ( ˆ 0 )

I In large samples ) p-value = Pr(|Z | > |t act |) = R

2 (-|t act |)
!

5/12
Brittany Almquist Lewis
Chapter 5
Section 5.1 Section 5.2 Section 5.3 Section 5.4

Confidence Interval for 1

6N7/
$S
3/'4)
%*)T*U','IV$$*3+V//,)$*VV7*%W(
⇣ ⌘
I 95% CI for 1 ) ˆ 1 ± 1.96 ⇥ SE ˆ 1
S

h ⇣ ⌘ ⇣ ⌘i
I ˆ 1 - 1.96 ⇥ SE ˆ 1 , ˆ 1 + 1.96 ⇥ SE ˆ 1
!
!

I Application to test scores H',+/

X#/) ,#3 F *,.05X/

O/H# F )# FP/ F .', H/W/.3

9500
I 95% CI for ) -2.28 ± 1.96 ⇥ 0.52 ) [-3.30, -1.26] 37/ %

1
F
! F !
A
!

6 !
(
0*,/)
,# DD$Y-*Z[ 6/12
Brittany Almquist Lewis
Chapter 5 %
Section 5.1 Section 5.2 Section 5.3 Section 5.4

Confidence Interval for 0

⇣ ⌘
I 95% CI for 0 ) ˆ 0 ± 1.96 ⇥ SE ˆ 0
h ⇣ ⌘ ⇣ ⌘i
I ˆ 0 - 1.96 ⇥ SE ˆ 0 , ˆ 0 + 1.96 ⇥ SE ˆ 0

^ @?_%?!>?@>><%G ` A
@?_%?a>%?@Q><%G ` `
\#
6
I+]+

Sbc\V `
!
!
><%G ^@C_!=;AC>?%;_d

7/12
Brittany Almquist Lewis
Chapter 5
Section 5.1 Section 5.2 Section 5.3 Section 5.4

Regression When X is a Binary Variable

I Binary variable ) indicator variable %**%* e

I Di = 1 if condition holds, Di = 0 otherwise

f
I Yi = 0 + 1 Di + ui for i = 1, . . . , n
I Di = 0 ) Y i = 0 + ui ) E (Yi ) = 0

I Di = 1 ) Y i = 0 + 1 + ui ) E (Yi ) = 0 + 1

I 1 ) di↵erence between population means

I Hypothesis tests & confidence intervals

-#
V*0LB#B \#37
-%
D*
Y[D
N*:
!

L*# :#3 -% 3# `
!

g
-#
FB
B
6
%

L*3 g h
!
!

\# 3- a : A
B
- A

D*
A
F
.#,X*3*#,'0
\A
*) 37/ X*--/H/,./ #-
#, L* B
# 1) F L* B
D 8/12
Brittany Almquist Lewis
Chapter 5
Section 5.1 Section 5.2 Section 5.3 Section 5.4

What Are Heteroskedasticity & Homoskedasticity

Heteroskedastic Errors

***%$
%

4i4jf)
% $

% %

6 %
%
%
%

***%
%
%

% %

c'3*,
%

"#4# *,
%

6N7/
!
6

F
%

4/',) )'4/
F
%

N7/ )'4/
%

%
%
%
B

1'H*',./
% %
%

!
6 %
%
%

%
%

9/12
Brittany Almquist Lewis
Chapter 5
Section 5.1 Section 5.2 Section 5.3 Section 5.4

What Are Heteroskedasticity & Homoskedasticity

S'4/ 1'H*',./
.#,X*3*#,'0 #, F k

I var(ui |Xi ) constant for i = 1, . . . , n ) ui is homoskedastic

I Otherwise ) ui is heteroskedastic

10/12
Brittany Almquist Lewis
Chapter 5
Section 5.1 Section 5.2 Section 5.3 Section 5.4

Mathematical Implications of Homoskedasticity

PQ "/35#)8/X')3*# )3',X'HX /HH#H)

l '..5H'3/ /1/, F
'H/F
\# *)
%

lI OLS estimators remain unbiased & asymptotically normal

%

)3',X'HX /HH#H
5,X/H

Y
*) PH#,+
F
I Efficiency of the OLS estimator when the errors are 3H'X*3*#,'0 #8
homoskedastic M#H450') P7/,
'H/ "/3/H#
!

/HH#H)
)8/X')0*.
I Homoskedasticity-only variance formula %

g 4'33/H) M#H
⇣ ⌘ q
s%û2 "2:#37/)*)
6)
I ˜ 2ˆ = Pn 2 ) SE ˆ 1 = ˜ 2ˆ .#,-*X/,./ F *,
0*,/'0
i=1 (Xi -X̄ )
1 1

3/)3*,+ %

I Invalid under heteroskedasticity

11/12
Brittany Almquist Lewis
Chapter 5
Section 5.1 Section 5.2 Section 5.3 Section 5.4

Mathematical Implications of Homoskedasticity

I Heteroskedasticity-robust standard errors

I Eicker-Huber-White standard errors
I Practical implications

12/12
Brittany Almquist Lewis
Chapter 5