Scribd
Upload
235 views36 pages

Running A Proper Regression Analysis: V G R Chandran Govindaraju Uitm Email: Website

This document provides an overview of running a proper regression analysis. It discusses exploring the data, developing regression models, checking assumptions like linearity, normality, autocorrelation and heteroskedasticity. It also covers using dummy variables, interactions, and time series techniques like unit root tests and cointegration. Key steps outlined are exploring the data, developing one or more regression models, identifying the most suitable model, and making inferences. Recommended books on regression analysis and econometrics are also listed.

Uploaded by

Arafatul Alam Patwary
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
235 views36 pages

Running A Proper Regression Analysis: V G R Chandran Govindaraju Uitm Email: Website

This document provides an overview of running a proper regression analysis. It discusses exploring the data, developing regression models, checking assumptions like linearity, normality, autocorrelation and heteroskedasticity. It also covers using dummy variables, interactions, and time series techniques like unit root tests and cointegration. Key steps outlined are exploring the data, developing one or more regression models, identifying the most suitable model, and making inferences. Recommended books on regression analysis and econometrics are also listed.

Uploaded by

Arafatul Alam Patwary
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 36

RUNNING A PROPER REGRESSION ANALYSIS

V G R CHANDRAN GOVINDARAJU UITM Email: [email protected] Website: www.vgrchandran.com/default.html

Topics
Running a proper regression analysis First half of the day:
1. What is regression? 2. How to estimate? (Simple and Multiple Regression) 3. Checking the assumptions of regression

Second half of the day


1. Regression with dummy variables 2. Recap: Time Series Econometrics

Types of data
Cross sectional Time series Panel data Where to get the data, DOS and BNM Lets download some data Data transformation level data, growth rate, index numbers, nominal to real values, exponential to linear models, etc

What to do after obtaining your data?


START

EXPLORE DATA

DEVELOP ONE OR MORE REGRESSION MODELS

IS ONE OR MORE REG. MODELS SUITABLE FOR DATA

no

REVISE THE MODEL /NEW MODEL

IDENTIFY MOST SUITABLE MODEL

MAKE INFERENCES & REPORT

Explore the data


Data cleaning Feel your data
Descriptive Statistics Correlations and Plots

What is regression?
Relationship between two variables (simple) or more than two variables (multiple) Models: Y = + X + is the intercept is the coefficient is the error term

Regression (Simple Example)


n 1 2 3 4 5 6 7 8 9 10 11 12 13 14 y 23 29 49 64 74 87 96 97 109 119 149 145 154 166 x 1 2 3 4 4 5 6 6 7 8 9 9 10 10

What is regression? (continue) SIMPLE EXAMPLE


Lets plot the data scatter plots Fitting a regression line. Findings the error term (residuals) Residuals are the most important part of regression (will let you known why later)

Estimating the alpha and beta.


Using Excel, SPSS, Eviews, Microfit, STATA, SPLUS Will teach how to use Excel, EVIEWS and SPSS (just an overview) Interpreting the outputs

Things to evaluate (output)


Economic Criteria signs and size of the effects (coefficient) follows economic theory demand for food (price variable) Coefficient of determinants Significance test on parameters (also joint test) Model selection criteria Functional form Econometric criteria - assumptions (do not violate)

Assumptions of linear regression


Linearity Normality Autocorrelation/Serial Correlation Heterogeneity Multicollinearity

Linearity
Straight enough condition (scatter plots) SPSS: Graphs: Scatter: Matrix: enter the dependent (outcome) variable first and then each of the independent variables (categorical/nominal variables dont need to be entered, but do it anyway to see what it looks like). SPSS: Analyze: Regression: Linear: Ramsey RESET test.

Normality
We do not need to test each series Just test the residuals Jarque-Berra statistics or the QQ and PP plots We use JB stat. Null Hypo: Normal

What to do if data is not normal?


Increase sample size Transform data e.g. log values Data may not be normal bcause of specification problems or functional form. Remember linearity

Serial Correlation/AutoCorrelation
Likely a problem in time series data especially data with short frequency What cause autocorrelation?
Omitted variables Misspecification

Consequences of autocorrelation
OLS estimators will be inefficient Variance of the coefficient will be biased and inconsistent

How to detect autocorrelation


Graphical methods: plot the residual and also draw a scatter plot of residual against residual (1) Durbin-Watson test Eviews (Null Hypo: no autocorrelation) Application when model includes constant, only first order and no lagged dependent variables We have a table to compare (DW stat) to the critical values (but rule of thumb if the value nears 2 than it is ok

How to detect autocorrelation


Breusch-Godfrey test for serial correlation It can test higher orders Eviews View/Residual tests/serial correlation LM test

How to solve autocorrelation?


Cochrane-Orcutt iterative procedure (beyond our scope remember I said regression in plain English AR (1)

Test for specification


Ramseys RESET test We include the predicted dependent variable as one of the regressors Lets do it in eviews

Heteroskedasticity
The opposite of homoskedasticity Hetero means unequal; Homo means equal Second part of the word skedasticity means spread (variance) Example: Consumption rich and poor rich have better spread (save and consumption) poor have lower spread There are many ways to test hetero : Graphically plot residual squared against dependent or independent variable there must not be a systematic pattern However graphical methods can be used for multiple regression

Heteroskedasticity
The following test can be used:
Breusch-Pagan LM test, Glesjer LM Test, HarveyGodfrey LM test, Park test, Goldfeld-Quabdt test, and White test Lets use the White test Null Hypo: No hetero or homo

Consequence of hetero and ways to correct it


No change in estimated parameter but standard error is effected (so does the significant) Generalized (or weighted ) least squares (beyond or discussion) Run a heterogeneity corrected regression (lets do a simple (White corrected standard error estimates) Alternatively, we can also use dummy variables to account for hetero

Multicollinearity
Whether there is any relationship between the regressors Consequences parameter is indetermine if perfect multicollinearity (However, real data do not have perfect multicollinearity) Imperfect multicollinearity when regressors are correlated but less than perfect How to detect?
Correlation matrics Check the significance of individual coefficient (t-test) and the joint significance (F-test) Run the regression by separating the regressors VIF Eviews or in SPSS (VIF value of less than 10 is ok)

Structural break and parameter stability test


Aim is to see whether parameters of the models have been constant over the periods Chow test we have to know the point of the break CUSUM and CUSUM Q Test parameter stability

24

Regression Analysis with Dummy Variables y = b0 + b1x1 + b2D2 + . . . bkxk + u

25

Dummy Variables
A dummy variable is a variable that takes on the value 1 or 0 Examples: male (= 1 if are male, 0 otherwise), south (= 1 if in the south, 0 otherwise), etc. Dummy variables are also called binary variables, for obvious reasons

26

A Dummy Independent Variable


Consider a simple model with one continuous variable (x) and one dummy (d) y = b0 + d0d + b1x + u This can be interpreted as an intercept shift If d = 0, then y = b0 + b1x + u If d = 1, then y = (b0 + d0) + b1x + u The case of d = 0 is the base group

27 Example of d0 > 0
Economics 20 - Prof. Anderson

d=1
d0

y = (b0 + d0) + b1x


slope = b1 d=0

b0

y = b0 + b1x
x

28

Dummies for Multiple Categories


We can use dummy variables to control for something with multiple categories Suppose everyone in your data is either a HS dropout, HS grad only, or college grad To compare HS and college grads to HS dropouts, include 2 dummy variables hsgrad = 1 if HS grad only, 0 otherwise; and colgrad = 1 if college grad, 0 otherwise

29

Multiple Categories (cont)


Any categorical variable can be turned into a set of dummy variables Because the base group is represented by the intercept, if there are n categories there should be n 1 dummy variables If there are a lot of categories, it may make sense to group some together Example: top 10 ranking, 11 25, etc.

30

Interactions Among Dummies


Interacting dummy variables is like subdividing the group Example: have dummies for male, as well as hsgrad and colgrad Add male*hsgrad and male*colgrad, for a total of 5 dummy variables > 6 categories Base group is female HS dropouts hsgrad is for female HS grads, colgrad is for female college grads The interactions reflect male HS grads and male college grads

31

More on Dummy Interactions


Formally, the model is y = b0 + d1male + d2hsgrad + d3colgrad + d4male*hsgrad + d5male*colgrad + b1x + u, then, for example: If male = 0 and hsgrad = 0 and colgrad = 0 y = b0 + b1x + u If male = 0 and hsgrad = 1 and colgrad = 0 y = b0 + d2hsgrad + b1x + u If male = 1 and hsgrad = 0 and colgrad = 1 y = b0 + d1male + d3colgrad + d5male*colgrad + b1x + u

32

Other Interactions with Dummies


Can also consider interacting a dummy variable, d, with a continuous variable, x y = b0 + d1d + b1x + d2d*x + u If d = 0, then y = b0 + b1x + u If d = 1, then y = (b0 + d1) + (b1+ d2) x + u This is interpreted as a change in the slope

33

Other use of dummy variables


Seasonal dummy Structural breaks Shocks etc

Lets Recap Our Time Series Analysis


Unit Root Test Cointegration Vector Error Correction Model Granger Causality

Must Have Books (for new researchers)


Pratical Data Analysis Gary Koop (2004) Analysis of Economic Data, John Wiley. Basic Econometrics Gary Koop (2008) Introduction to Econometrics, John Wiley Samprit Chatterjee, Ali S. Hadi, Bertam Price (2000) Regression Analysis by Example, John Wiley. Dimitrios Asteriou and Stephen G. Hall (2007) Applied Econometrics: A Modern Approach Using Eviews and Microfit, Palgrave Basic Statistics and Regression Models De Veaux, Paul Velleman and David Bock, Stats: Data and Models, Pearson. (for basic statistics)

Thank you
QUESTIONS PLEASE
More materials will soon be available (by end of the month) through my website:

www.vgrchandran.com/default.html

You might also like