Practical 2

Question 1

Correlation Matrix

• If data is normal, use Pearson correlation (Pairwise) (Parametric)
• If not, use Spearman correlation (Non-parametric)

Answer 1

Correlation Matrix

• If data is normal, use Pearson correlation (Pairwise) (Parametric)
• If not, use Spearman’s rank correlation (Non-parametric)

Question 2

If coefficients in the model are > 0.8 (80%) => high autocorrelation
(i.e. the 2 variables have more than 80% similarity, 2 almost identical variables)
=> inflate the model.
=> Don’t include both in the model!

Answer 2

If coefficients in the model are > 0.8 (80%) => high autocorrelation
(i.e. the 2 variables have more than 80% similarity, 2 almost identical variables)
e.g. ROA (return on assets) and ROE (return on equity)
=> inflate the model.
=> Don’t include both in the model!

Question 3

• Don’t use log variables for Descriptive analysis as they’re there only to ensure normality when doing modelling in Predictive analysis.

Answer 3

• Don’t use log variables for Descriptive analysis as they’re there only to ensure normality when doing modelling in Predictive analysis.

Question 4

Panel Data Regression:
(Fixed effects/ Random effects)
Test for Hypothesis:
H0: There is a POSITIVE SIGNIFICANT relationship between X and Y
- check if positive or not: look at the sign of coefficient
- check significance:
p-value < 0.01: high significance ***
p-value < 0.5: moderate significance **
p-value < 0.1: weaker significance *

In STATA, p-value is ‘P>|t|’
F value is ‘Prob > F’

Answer 4

Panel Data Regression:
(Fixed effects/ Random effects)
Test for Hypothesis:
H0: There is a POSITIVE SIGNIFICANT relationship between X and Y
- check if positive or not: look at the sign of coefficient
- check significance:
p-value < 0.01: high significance ***
p-value < 0.5: moderate significance **
p-value < 0.1: weaker significance *

In STATA, p-value is ‘P>|t|’
F value is ‘Prob > F’

Question 5

R-squared (‘R-sq’ in STATA, normally use Overall value) shows how many % of changes in dependent variable are explained by change in independent variables.
- check the benchmark of R-sq in literature to see what the acceptable R-sq is.

Answer 5

R-squared (‘R-sq’ in STATA, normally use Overall value) shows how many % of changes in dependent variable are explained by change in independent variables.
- check the benchmark of R-sq in literature to see what the acceptable R-sq is.

Question 6

Fixed effects regression is stricter in linear assumptions.

Random effects regression relax some assumptions.

Answer 6

Fixed effects regression is stricter in linear assumptions.

Random effects regression relax some assumptions.

Question 7

Use the Hausman test:
If ‘Prob>chi2’ (in STATA) <= 0.05 => use Fixed effects regression as main test.
If ‘Prob>chi2’ > 0.05 => use Random effects

Answer 7

Use the Hausman test:
If ‘Prob>chi2’ (in STATA) <= 0.05 => use Fixed effects regression as main test.
If ‘Prob>chi2’ > 0.05 => use Random effects

Question 8

Normally distributed data have:
Skewness: +- ?
AND Kurtosis: +- ?

Answer 8

Normally distributed data have:
Skewness: +-0.96
AND Kurtosis: +-3