Multiple Regression Models

Dr. William Lau

Tel: 3943 8572
[email protected]
A Case Study in China regarding
Financial Analysts’ Beauty and Performance
Do more physically attractive sell-side financial
analysts have better performance?
(sample size: 2,328 from Securities Association of China in 2005 – 2014)

 Attractiveness is significantly positively associated with their earnings forecast accuracy, and the
informativeness and profitability of their stock recommendations.
 More attractive analysts are more likely to be selected as star analysts and hired by larger brokerages.
 More attractive analysts are more likely to be received by public firms for site visits, especially in a short
period before firms announce their financial results.
 More attractive analysts are more likely to appear on financial television programs than less attractive
analysts.
 Attractive analysts are better connected with institutional investors.
 Attractive analysts receive more internal resources: they are more likely to have their own research teams
when producing research reports, and their research teams are larger than those of less attractive analysts.
 Beauty effect on forecast performance is more pronounced when the analyst is geographically closer to the
covered firms.
Do more physically attractive sell-side financial
analysts have better performance?
(sample size: 2,328 from Securities Association of China in 2005 – 2014)

 The superior forecast performance of attractive analysts disappears after firm management
publishes its own earning forecasts.
 The positive association between analysts’ physical attractiveness and forecast performance
disappears when managers can sell their shares in the open market, or when the firms have
share pledge agreements with banks.
 Although attractive analysts outperform their peers in earnings forecast in the two years
before Xiao Gang ( 肖鋼 ) took office in 2013, this performance difference disappears in the
two years after.
Lesson Outline
1 General Form of a Multiple Regression Model
2 Model Assumptions
3 A First-Order Model with Quantitative Predictors
4 Fitting the Model: The Method of Least Squares
5 Estimator of σ2, the Variance of ε
6 Testing the Utility of a Model: The Analysis of Variance F-Test
7 Inferences About the Individual β Parameters
8 Multiple Coefficients of Determination: R2 and Ra2
9 Using the Model for Estimation and Prediction
10 An Interaction Model with Quantitative Predictor
11 A Quadratic (Second-Order) Model with a Quantitative Predictor
12 More Complex Multiple Regression Models
13 A Test for Comparing Nested Models
14 A Complete Example
6
7
Model assumptions
  
ε

𝛽 0 + 𝛽 1 𝑥1 +…+ 𝛽𝑘 𝑥𝑘
4- 8 A First-Order Model with Quantitative Predictors
9
10 E(y) = 1+ 2x1 + x2 for x2 = 0,1,2
4- 11 Fitting the Model: The Method of Least Squares
12
13 Side-by-side scatterplots for the data of Table 4.1
 Regression Analysis
14
for the auction price model
15
 Testing the Utility of a Model:
4- 16
The Analysis of Variance (Global F-Test)
17
18 Rejection region for the Global F-test
 MINITAB Regression Analysis for
19
grandfather clock model
4- 20 Inferences About the Individual  Parameters
21

≥ ≤

ˆi / sˆ
i
22
23 Regression Analysis for the auction price model
 Multiple Coefficients of Determination
4- 24 2 2
R and Ra
25 R2 and Adjusted R2

 
 A portion of the SPSS regression output for the
26
auction price model

Model
4- 27 Using the Model for Estimation and Prediction
 MINITAB printout with 95% confidence and
28
prediction intervals for grandfather clock model
 An Interaction Model with
4- 29
Quantitative Predictors
 E(y) = 1 + 2x1 – x2 + x1x2
30 for x2 = 0,1,2
31
32 Examples of no-interaction and interaction models
33
MINITAB regression printout for
grandfather clock model with interaction
 A Quadratic (Second-Order) Model with a
4- 34
Quantitative Predictor
35
36 Quadratic Regression Model
2
Y  β 0  β1X  β 2 X  ε
Quadratic models may be considered when the scatter
plot takes on one of the following shapes:
Y Y Y Y

X1 X1 X1 X1
β1 < 0 β1 > 0 β1 < 0 β1 > 0
β2 > 0 β2 > 0 β2 < 0 β2 < 0
β1 = the coefficient of the linear term
β2 = the coefficient of the squared term
37
38 SPSS scatterplot for data of Table 4.2
39 SPSS output for the quadratic model
 MINITAB graph of least squares fit for the
40
quadratic model
41 Potential misuse of quadratic model
4- 42 More Complex Multiple Regression Models
43
44
45
46
 2 2
47 E(y) = 1 + 7x1 - 10x2 + 5x1x2 – +x 3
1 x 2
48
49 Multiple regression output for Example 4.8
50
51
52
53 MINITAB multiple regression output
4- 54 A Test for Comparing Nested Models
55 Definition of Nested Models

Two models are nested if one model contains all the

terms of the second model and at least one
additional term. The more complex of the two
models is called the complete (or full) model.
The simpler of the two models is called the
reduced (or restricted) model.
56

is nonzero.
57
Rejection region for the F-test
H0: b4 = b5 = 0
58 SAS output for a complete model
59 SAS output for a reduced model
60 Test Your Understanding
Using the data in the Phuket worksheet, fit 1) First Order Model and 2)
Interaction Model, then conduct a partial F-test to check which model is
better.

A) The p-value of the partial F-test is 0.0021, so we would choose the First
Order Model.
B) The p-value of the partial F-test is 0.0021, so we would choose the
Interaction Model.
C) The p-value of the partial F-test is 0.4109, so we would choose the First
Order Model.
D) The p-value of the partial F-test is 0.4109, so we would choose the
Interaction Model.
4- 61 A Complete Example

Y: Realized Road Construction Cost

X1: Department of Transportation’s Estimated Cost
X2: Market Status
 Regression analysis for a complete second-
62 order model for road cost
SUMMARY OUTPUT

Regression Statistics
Multiple R 0.988573896
R Square 0.977278349
Adjusted R Square 0.976782243
Standard Error 296.6429404
Observations 235

ANOVA
df SS MS F Significance F
Regression 5 866726151.1 173345230.2 1969.898554 5.8772E-186
Residual 229 20151320.81 87997.0341
Total 234 886877471.9

Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Lower 95.0% Upper 95.0%
Intercept -2.972080062 30.89109213 -0.096211557 0.923436697 -63.83918626 57.89502613 -63.83918626 57.89502613
DOTEST 0.915527578 0.029170504 31.38538799 6.6509E-85 0.85805068 0.973004476 0.85805068 0.973004476
DOTEST^2 7.19139E-07 3.40418E-06 0.211251496 0.832878904 -5.98839E-06 7.42667E-06 -5.98839E-06 7.42667E-06
STATUS -36.72899577 74.77240576 -0.491210566 0.623747959 -184.0588455 110.600854 -184.0588455 110.600854
STA * DOT 0.324227947 0.119173319 2.720642082 0.007015721 0.089411546 0.559044348 0.089411546 0.559044348
STA * DOT^2 -3.5762E-05 2.47845E-05 -1.442914314 0.150411043 -8.45969E-05 1.30729E-05 -8.45969E-05 1.30729E-05
 Regression analysis for a complete second-
63
order model for road cost
64 SAS output for reduced model for road cost
65 Plot of the least squares lines for the reduced model
 SAS output showing prediction intervals for
66
reduced model