Professional Documents
Culture Documents
CHAPTER 10
10-#.
10-).
0 is the .-intercept of the regression line- and 1 is the slope of the line.
The conditional mean of Y- ((Y 0 X)- is the pop lation regression line. The regression model is sed for nderstanding the relationship !et"een the t"o *aria!les- X and Y' for prediction of Y for gi*en *al es of X' and for possi!le control of the *aria!le Y- sing the *aria!le X. The error term capt res the randomness in the process. Since X is ass med nonrandom- the addition of ma3es the res lt (Y) a random *aria!le. The error term capt res the effects on Y of a host of n3no"n random components not acco nted for !% the simple linear regression model. The e5 ation represents a simple linear regression model "itho t an intercept (constant) term. The least-s5 ares proced re prod ces the !est estimated regression line in the sense that the line lies 7inside8 the data set. The line is the !est n!iased linear estimator of the tr e regression line as the estimators 0 and 1 ha*e smallest *ariance of all linear n!iased estimators of the line parameters. Least-s5 ares line is o!tained !% minimi&ing the s m of the s5 ared de*iations of the data points from the line.
10-2.
10-4. 10-6.
10.10. Least s5 ares is less sef l "hen o tliers exist. 9 tliers tend to ha*e a greater infl ence on the determination of the estimators of the line parameters !eca se the proced re is !ased on minimi&ing the s5 ared distances from the line. Since o tliers ha*e large s5 ared distances the% exert nd e infl ence on the line. A more ro! st proced re ma% !e appropriate "hen o tliers exist.
10-1
0.667 0.431
-0.967 95% 10.12 + or - 2.77974 0.967 -0.431 Con'i(ence In er)al 'or In ercep 1 *1+, C-I- 'or 0 0.000 95% 6.38 + or - 9.21937
10.1#. b1 < SSXY @SSX < 6)+.+6@21/.64 < 1.## 10-1). (Template: Simple Regression.xls- sheet: Regression) Th s- b0 < ).0/2 b1 < 0.142
r2 0.9217 Coefficient of Determination r 0.9601 Coefficient of Correlation s(b1) 0.0164Standard Error of Slo e
Con'i(ence In er)al 'or Slope 1 *1+, C-I- 'or 1 95% 0.18663 + or - 0.03609 Con'i(ence In er)al 'or In ercep 1 *1+, C-I- 'or 0 95% -3.05658 + or - 2.1372 Pre(ic ion In er)al 'or # 1 *1+, P-I- 'or Y gi)en X X 95% 10 -1.19025 + or - 2.8317 Pre(ic ion In er)al 'or E.#/"0 1 *1+, P-I- 'or E.Y / X0 " + or ANO1A Ta2le So%rce SS "e#n. 128.332 Error 10.8987 $otal 139.231 (' 1 11 12
10-#
10-1+. b1 < SSXY @SSX < #.11 b0 < y b1 x < 11/.) (#.11)(44.6) < ##.#26 10-1/. Simple Regression
In'la ion " 1 2 3 4 5 6 7 8 9 1 2 12.6 -10.3 0.51 2.03 -1.8 5.79 5.87 Re %rn # -3 36 12 -8 53 -2 18 32 24
Error -20.0642 17.9677 -16.294 -14.1247 36.4102 -20.0613 3.64648 10.2987 2.22121
!nflation % ret&rn on 'toc(' r2 0.0873 Coefficient of Determination r 0.2955 Coefficient of Correlation s(b1) 1.18294Standard Error of Slo e
Con'i(ence In er)al 'or Slope 1 95% *1+, C-I- 'or 1 0.96809 + or - 2.7972
Con'i(ence In er)al 'or In ercep 1 *1+, C-I- 'or 0 95% 16.0961 + or - 17.3299
(' 1 7 8
10-)
There is a "ea3 linear relationship (r) and the regression is not significant (r #- A- p-*al e) 10-11. Simple Regression
#ear " 1 2 3 4 5 1960 1970 1980 1990 2000 1al%e # 180000 40000 60000 160000 200000
)*era#e *al&e of )'ton +artin r2 0.1203 Coefficient of Determination r 0.3468 Coefficient of Correlation s(b1) 2498Standard Error of Slo e
Con'i(ence In er)al 'or Slope 1 95% *1+, C-I- 'or 1 1600 + or - 7949.76
Con'i(ence In er)al 'or In ercep 1 *1+, C-I- 'or 0 95% -3040000 + or - 1.6E+07
(' 1 3 4
10-+
3 4 16008 + :E906
1960
1970
1980 "
1990
2000
2010
There is a "ea3 linear relationship (r) and the regression is not significant (r #- A- p-*al e). Limitations: sample si&e is *er% small. $idden *aria!les: the 20s and 40s models ha*e a different *al ation than other decades possi!l% d e to a different model or st%le. 10-12. Regression e5 ation is: Credit Card Transactions < )6.1212 = 0.011#6 Be!it Card Transactions
r2 0.9624 Coefficient of Determination r 0.9810 Coefficient of Correlation s(b1) 0.06129Standard Error of Slo e
Con'i(ence In er)al 'or Slope 1 95% *1+, C-I- 'or 1 0.6202 + or 0.17018
Con'i(ence In er)al 'or In ercep 1 *1+, C-I- 'or 0 95% 177.641 + or - 110.147 Pre(ic ion In er)al 'or # 1 *1+, P-I- 'or Y gi)en X X + or Pre(ic ion In er)al 'or E.#/"0 1 *1+, P-I- 'or E.Y / X0 " + or ANO1A Ta2le So%rce SS "e#n. 332366 Error 12984.5 $otal 345351
(' 1 4 5
There is no implication for ca salit%. A third *aria!le infl ence co ld !e 7increases in per capital income8 or 7;BC ;ro"th8.
10-/
( y b
b1 x )
0
( y b @ b E ( y b @ b0 E
1
b1 x ) # D < #
b1 x ) #
( y b b x ) D < # x( y b b x )
0 1
0 1 0 1
Setting the t"o partial deri*ati*es to &ero and simplif%ing- "e get:
x%
Sol*ing the a!o*e t"o e5 ations sim ltaneo sl% for b0 and b1 gi*es the re5 ired res lts. 10-16. 66F C.>. for 1 : 1.#//)) #.402(0.0+62#) < E1.11/4- 1.)6+6D. The confidence inter*al does not contain &ero. 10-#0. GS( < 2.1#6 Arom the AH9IA ta!le for Cro!lem 10-11:
ANO1A Ta2le So%rce SS "e#n. 1024.14 Error 22.888 $otal 1047.03 (' 1 3 4 MS 1024.14 7.62933
10-#1. Arom the regression res lts for pro!lem 10-11 s(b0) < #.462 s(b1) < 0.42)
s(b1) 0.87346Standard Error of Slo e s(b0) 2.89694Standard Error of !nterce t
Con'i(ence In er)al 'or In ercep 1 *1+, C-I- 'or 0 95% 6.38 + or - 9.21937
6/F C.>. for the slope: 10.1# J #.2262+ < E2.)+0#1- 1#.4662+D 6/F C.>. for the intercept: 1.)4 J 6.#16)2 < E-#.4)6)2- 1/./66)2D
10-1
10-#). s(b0) < 0.621 s(b1) < 0.011' estimate of the error *ariance is GS( < 0.661. 6/F C.>. for 1: 0.142 = #.#01(0.011) < E0.1/14- 0.####D. Kero is not a pla si!le *al e at < 0.0/.
Con'i(ence In er)al 'or Slope 1 *1+, C-I- 'or 1 95% 0.18663 + or - 0.03609 s(b1) 0.0164Standard Error of Slo e
Con'i(ence In er)al 'or In ercep 1 *1+, C-I- 'or 0 95% -3.05658 + or - 2.1372
10-#+. s(b0) < 4/.++ s(b1) < 0.1/)+ (stimate of the regression *ariance is GS( < 41## 6/F C.>. for b1: 1.//14 #.221 (0.1/)+) < E1.1#1- 1.624D Kero is not in the range.
Con'i(ence In er)al 'or Slope 1 *1+, C-I- 'or 1 95% 1.55176 + or - 0.42578 s(b1) 0.15336Standard Error of Slo e
Con'i(ence In er)al 'or In ercep 1 *1+, C-I- 'or 0 95% -255.943 + or - 237.219
10-#/. s # gi*es s information a!o t the *ariation of the data points a!o t the comp ted regression line. 10-#1. >n correlation anal%sis- the t"o *aria!les- X and Y- are *ie"ed in a s%mmetric "a%- "here no one of them is 7dependent8 and the other 7independent-8 as the case in regression anal%sis. >n correlation anal%sis "e are interested in the relation !et"een t"o random *aria!les- !oth ass med normall% distri! ted. 10-#2. Arom the regression res lts for pro!lem 10-11:
r 0.9890 Coefficient of Correlation
10-2
0.)+14 (1 .1#0)) @ )
< 0.1+0
Accept $0. The t"o *aria!les are not linearl% correlated. 10.)0. .es. Aor example s ppose n < / and r < ./1' then: t<
r (1 r ) @( n #)
#
< 1.0# and "e do not reLect $0. M t if "e ta3e n < 10-000 and
r < 0.0+- gi*ing t < 1+.#4- this leads to strong reLection of $0. 10-)1. :e ha*e: r < 0.42/ and n < 10. Cond cting the test: t (4) <
r (1 r ) @( n #)
#
<
.42/ (1 .42/ # ) @ 4
< /.11
There is statistical e*idence of a correlation !et"een the prices of gold and of copper. Limitations: data are time-series data- hence not dependent random samples. Also- data set contains onl% 10 points.
.)2 (1 .)2 # ) @ 1)
t (1)) <
< ).11
z <
<
1 1 ln E(1 = r)@(1 N /)D < ln (1.)2@0.1)) < 0.)44+ # # 1 1 ln E(1 = )@(1 N )D < ln (1.##@0.24) < 0.##)2 # #
< 1@ n ) < 1@ 1# < 0.1#2 z < ( z )@ < (0.)44+ N 0.##)2)@0.1#2 < 1.#62. Cannot reLect $0.
10.)1. ,sing 7TINV(,df)8 f nction in (xcel- "here df < n-# < /#: <T>HI(0.0/-/#) < #.0011+/ And T>HI(0.01- /#) < #.12)2 ReLect $0 at 0.0/ ! t not at 0.01. There is e*idence of a linear relationship at < 0.0/ onl%. 10-)2. t (11) < b1@s(b1) < ).1@#.46 < 1.02#2. Bo not reLect $0. There is no e*idence of a linear relationship sing any 10-)4. ,sing the regression res lts for pro!lem 10-11: critical *al e of t is: t( 0.0/- )) < ).14# comp ted *al e of t is: t < !1@s(!1) < 10.1# @ 0.42)+1 < 11./41 ReLect $0. There is strong e*idence of a linear relationship.
10-4
10-)6. t (11) < b1@s(b1) < 0.142@0.011 < 11.16 ReLect $0. There is strong e*idence of a linear relationship !et"een the t"o *aria!les. 10-+0. b1@ s(b1) < 1100@#+64 < 0.1+1 Bo not reLect $0. There is no e*idence of a linear relationship. 10-+1. t (/4) < b1@s(b1) < 1.#+@0.#1 < /.60 .es- there is e*idence of a linear relationship. 10.+#. ,sing the (xcel f nction- TDIST(x,df,#tails) to estimate the p-*al e for the t-test res lts- "here x < 1./1- df < /4/16# N # < /4/160- Otails < # for a #-tail test: TB>ST(1./1- /4/160-#) < 0.1)1. The corresponding p-*al e for the res lts is 0.1)1. The resgression is not significant e*en at the 0.10 le*el of significance. 10-+). t (#11) < z < b1@s(b1) < 0.14@1#.0) < 0.0/1/ Bo not reLect $0. There is no e*idence of a linear relationship sing any res ltsP) 10-++. b1 < /.+6 s(b1) < 1.#1 t (#1) < +./)2 .es- there is e*idence of a linear relationship. 10.+/. The coefficient of determination indicates that 6F of the *ariation in c stomer satisfaction can !e explained !% the changes in a c stomerQs materialism meas rement. 10-+1 a. The model sho ld not !e sed for prediction p rposes !eca se onl% #.0F of the *ariation in pension f nding is explained !% its relationship "ith firm profita!ilit%. !. The model explains *irt all% nothing. c. Cro!a!l% not. The model explains too little. 10-+2. >n Cro!lem 10-11 regression res lts- r # < 0.6241. Th s- 62.4F of the *ariation in "ealth gro"th is explained !% the income 5 antile.
r2 0.9781 Coefficient of Determination
. (:h% report s ch
10-+4. >n Cro!lem 10-13- r # < 0.6##. Th s- 6#.#F of the *ariation in the dependent *aria!le is explained !% the regression relationship. 10-+6. r # in Cro!lem 10-16: r # < 0.1#0) 10-/0. Reading directl% from the G>H>TAM o tp t: r # < 0.61#
10-6
10-/1. Mased on the coefficient of determination *al es for the fi*e co ntries- the ,R model explains )1.2F of the *ariation in long-term !ond %ields relati*e to the %ield spread. This is the !est predicti*e model of the fi*e. The next !est model is the one for ;erman%- "hich explains 1).)F of the *ariation. The regression models for Canada- Sapan- and the ,S do not predict long-term %ields *er% "ell. 10-/#. Arom the information pro*ided- the slope coefficient of the e5 ation is e5 al to -1+.1. Since its *al e is not close to &ero ("hich "o ld indicate that a change in !ond ratings has no impact on %ields)- it "o ld indicate that a linear relationship exists !et"een !ond ratings and !ond %ields. This is in line "ith the reported coefficient of determination of 11./1F. 10-/). r # in Cro!lem 10-15: r # < 0.42)
r2 0.8348 Coefficient of Determination
10-/+.
!eca se the first term on the right is the s m of the "eighted regression resid als- "hich s m to &ero. The second term is the s m of the resid als- "hich is also &ero. This esta!lishes the res lt:
( y y )
T y) ( y
T) ( y y
10-//. Arom (5 ation (10-10): b1 < SSXY@SSX. Arom (5 ation (10-)1): SSR < b1SSXY. $ence- SSR < (SSXY @SSX)SSXY < (SSXY) #@SSX 10-/1. ,sing the res lts for pro!lem 10-11: A < 1)+.#)4 F(1-)) < 10.1#4 ReLect $0.
F 134.238 Fcritical p+)al%e 10.128 0.0014
10-10
F 102.389
Fcritical 7.70865
p+)al%e 0.0005
10-/6. F (1-2) < 0.1162+ Bo not reLect $0. 10.10. F (1-10#) < GSR@GS( <
42-161 @ 1 < 201.4 1#-2+/ @ 10#
There is extremel% strong e*idence of a linear relationship !et"een the t"o *aria!les. 10-11. t (#k ) < F (1-3) . Th s- F(1-#0) < Eb1@s(b1)D# < (#.//1@+.1##)# < 0.)4+/ Bo not reLect $0. There is no e*idence of a linear relationship. SS @ SS X < Eb1@s(b1)D < XY s @ SS X E sing (5 ations (10-10) and (10-1/) for b1 and s(b1)- respecti*el%D
t (#k )
#
10.1#
SS @ SS XY X <
MSE @ SS X
<
SS# XY @ SS X < SSR@1 < GSR < F (1-k) GS( GS( GS(
E!eca se SS# XY @ SS X < SSR !% (5 ations (10-)1) and (10-10)D 10-1). a. $eteroscedasticit%. !. Ho apparent inade5 ac%. c. Bata displa% c r*at re- not a straight-line relationship. 10-1+. a. Ho apparent inade5 ac%. !. A pattern of increase "ith time. 10-1/. a. Ho serio s inade5 ac%. !. .es. A de*iation from the normal-distri! tion ass mption is apparent.
10-11
10-11. ,sing the res lts for pro!lem 10-11: Resi(%al Anal3sis
d D%r2in+Wa son s a is ic 3.39862
Resi(%al Plo
4 3 2 1 Error 0 -1 -2 -3 -4 "
Resid al *ariance fl ct ates' "ith onl% / data points the resid als appear to !e normall% distri! ted.
1 0 -10 -5 -1 -2 -3 0 5 10
Resi(%als
10-1#
10-12. Resid als plotted against the independent *aria!le of Cro!lem 10-14:
U resids
1.#= U U U
0.0=
U U
U U U U U U
-1.#=
? alit% )0 +0 /0 10 20 40
10-14.
10-1)
Resi(%al Anal3sis
10.16. >n the American (xpress example- gi*e a 6/F prediction inter*al for x < /-000: T < #2+.4/ = 1.#//)(/-000) < 1-//1.)/. y C.>. < 1-//1.)/
(#.016)()14.11)
1+
< E/-4/+.+- 2-#+4.)D 10-20. ;i*en that the slope of the e5 ation for 10-/# is N1+.1- if the rating falls !% ) the %ield sho ld increase !% +).4 !asis points. 10-21. Aor 66F C.>.: 1-//1.)/ t .00/(#)) < #.402
1+ (/-000 )-122.6#) # 1 + #/ +0-6+2-//2.4+
(#.402)()14.11)
10-1+
10-22. a) simple regression e5 ation: . < #.226))2 V N 0.#4+1/2 "hen V < 10- . < #2./06#
In ercep Slope b0 b1 +0-;7<1=> ;->>5::>
Y 13
10-1/
:hen V < 10- . < #+.#)1 10.24. ,sing (xcel f nction- TINV(x, df)- "here x < the p-*al e of 0.0)+ and df < #0/4 N #: T>HI(0.0)+- #0/1) < #.1#1+42. Since the slope coefficient < -0.0/1- t-*al e !ecomes negati*e- t < -#.1#1+42.
b1 0.0/1 = = 0.0#+0+ t #.1#+42 !) ,sing an < 0.0/- "e "o ld reLect the n ll h%pothesis of no relationship !et"een the response *aria!le and the predictor !ased on the reported p-*al e of 0.0)+.
a) standard error of the slope: s b1 = 10-26. ;i*en the reported p-*al e- "e "o ld reLect the n ll h%pothesis of no relationship !et"een ne roticism and Lo! performance. ;i*en the reported coefficient of determination- 16F of the *ariation in Lo! performance can !e explained !% ne roticism. 10-40. The t-statistic for the reported information is:
t=
,sing (xcel f nction- TDIST(t,df,#tails)- "e get a p-*al e of 0.000014: TB>ST(+.#)1- 20- #) < 1.411#(-0/. There is a linear relationship !et"een fre5 enc% of online shopping and the le*el of percei*ed ris3.
10-41 (Arom Ginita!) The regression e5 ation is Stoc3 Close < 12.1 = 0.+02 9per >ncome
10-11
p 0.000 0.000
Anal%sis of Iariance S9,RC( BA SS GS F p Regression 1 1#011 1#011 1#6.+6 0.000 (rror 11 1+4/ 6) Total 12 1)/00 T < W/1.#+. Stoc3 close !ased on an operating income of W)0/G is %
(Ginita! res lts for Log Y) The regression e5 ation is LogXStoc3 Close < #.)# = 0.00//# 9per >nc Credictor Constant 9per >nc s < 0.04+## Coef #.)1/) 0.00//#01 R-s5 < 6/.1F Stde* 0.1022 0.000)1#6 t-ratio #1./0 12.1+ p 0.000 0.000
Anal%sis of Iariance S9,RC( BA Regression 1 (rror 11 Total 12 ,n s al 9!ser*ations 9!s. x % 1 #+0 ).4012
GS #.#022 0.0021
F )11.#/
p 0.000
Ait ).1+01
Stde*.Ait 0.0)11
Resid al 0.1111
St.Resid #.#0R
R denotes an o!s. "ith a large st. resid. T < W/+.40 Stoc3 close !ased on an operating income of W)0/G is %
10-12
The regression sing the Log of monthl% stoc3 closings is a !etter fit. 9perating >ncome explains o*er 6/F of the *ariation in the log of monthl% stoc3 closings *ers s 46F for non-transformed Y. 10-4#. a) The calc lated t-*al e for the slope coefficient is:
t=
,sing (xcel f nction- TDIST(t,df,#tails)- "e get a p-*al e of 0.0 TB>ST(6#.0- /64- #) < 0. There is a linear relationship. !) The excess ret rn "o ld !e 0.6/6#: A(R < 0.6/ = 0.6#(0.01) < 0.6/6# 10-4) a) adding # to all V *al es: ne" regression: . < / V = 12 since the intercept is b0 = Y b1 X - the onl% thing that has changed is that the *al e for V!ar has increased !% #. Therefore- ta3e the change in V-!ar times the slope and add it to the original regression intercept. !) adding # to all . *al es: ne" regression: . < /V = 6 sing the form la for the intercept- onl% the *al e for .-!ar changes !% #. Therefore- the intercept changes !% # c) m ltipl%ing all V *al es !% #: ne" regression: . < #./ V = 2 d) m ltipl%ing all . Ial es !% #: ne" regression: . < 10 V = 2 10-4+ .o are minimi&ing the s5 ared de*iations from the former x-*al es instead of the former %*al es. 10-4/ a) . < ).4#01)) V = /#.#2)0)1
In ercep Slope b0 b1 =;-;>:0:6 :-7;01::
!)
c)
r# < 0.6++6- *er% high' A < ###.6)1 (p-*al e < 0.000): !oth indicate that V affects .
10-14
d)
since the 66F C> does not contain the *al e 0- the slope is not 0
Con'i(ence In er)al 'or Slope 1 99% *1+, C-I- 'or 1 3.82013 + or - 0.77071
e)
f)
V < 1#.+6)/+
h)
10-16
Case 1): Le*el of le*erage a) Le*erage < -0.114 N 0.0+0 (Rights) !) ,sing (xcel f nction- TDIST(t,df,#tails)- "e get a p-*al e of 0.0 TB>ST(#.1#- 1)02- #) < 0.0046 There is a linear relationship. c) The reported coefficient of determination indicates that shareholdersQ rights explain 11./F of the *ariation in a firmQs le*erage. Case 1+: Ris3 and Ret rn 1) . < 1.1116/2 V N 1.0102#+
In ercep Slope b0 b1 +1-050>;< 1-1665=>
#) ))
+)
/)
10-#0
1)
1 0 -10 -5 -1 -2 -3 0 5 10
Resi(%als
2)
10-#1