Research

What is Research?
Research is a process to discover new knowledge.
1)Research is a careful and detailed study into a specific problem,
concern, or issue using the scientific method.
2)Research is a systematic inquiry that investigates hypotheses, suggests
new interpretations of data or texts, and posses new questions for future
research to explore.
Types of Research
Research may be very broadly defined as systematic gathering of data
and information and its analysis for advancement of knowledge in any
subject. Research attempts to answer intellectual and practical questions
through application of systematic methods. The types of research
methodologies vary and are often classified into few categories. Specific
academic fields tend to apply certain methodologies more than others:
 Qualitative: Involves describing in details specific situations using
research tools like interviews, surveys, and observation.
Quantitative: Requires quantifiable data involving numerical and statistical
explanations. Quantitative researchers seek to explain the causes of change
primarily through objective measurement and quantitative analysis
(statistics).
Correlation/Regression Analysis: Involves determining the strength of the
relationship between two or more variables. Correlation / regression
researchers determine whether correlations exist between two quantitative
variables.
Experimental: Relies on controlled experiments that compare the outcome
for an experimental and a control group that differ in a defined way.
What is Econometrics?
1) Econometrics may be defined as the social science in which the
tools of economic theory, mathematics, and statistical inference are
applied to the analysis of economic phenomenon.
2) Econometrics is concerned with the empirical determination of
economic laws.
Methodology of Econometrics
The traditional or classical methodology can be explained in the following
steps.
1)Statement of the theory
2)Specification of the mathematical model of the theory
3)Specification of the statistical, or econometric, model.
4) Obtaining the data
5)Estimation of the parameters of econometric model
6)Hypothesis testing
7)Forecasting or prediction
8)Using model for control or policy purposes.
To explain the preceding steps , let us consider the well-known Keynesian
1:Statement of Theory or Hypothesis
Keynesian postulated that the marginal propensity to consume (MPC),
the rate of change of consumption for a unit (say, a dollar) change in
income, is greater than zero but less than 1.
2. Specification of the Mathematical Model of Consumption
Although Keynes postulated a positive relationship between consumption
and income, he did not specify the precise form of the functional
relationship between the two. For simplicity, a mathematical economist
might suggest the following form of the Keynesian consumption
function:
Y=β1+ β2X 0< β2<1………………………………..1)
Where Y= consumption expenditure and X =income, where β1 and β2 known as the
parameters of the model, are, respectively, the intercept and slope coefficients.
The slope coefficient β2 measures the MPC. This equation states that consumption is
linearly related to income, is an example of mathematical model of the relationship between
consumption and income that is called the consumption function in economics. A model is
simply a set of mathematical equations. If the model has only one equation, as indicated, it
is called a single equation model, whereas if it has more than one equation, it is known as a
multiple-equation model.
The variable appearing on the left side of the equality sign is called the dependent variable
and the variable(s) on the right side is called the independent variables. Thus, in Keynesian
function, consumption expenditure is dependent variable and income is the explanatory
variable.
1) Specification of Econometric Model
The above mathematical model of the consumption function is of limited
interest to econometrician as it shows the exact relationship between
consumption and income. But relationships between economic variables
are generally inexact. There are other variables that also affect
consumption expenditures rather than income. For example, size of family,
ages of the members in the family, family religion, etc are likely to exert
some influence on consumption.
To allow for the inexact relationship between economic variables, the
econometrician would modify the deterministic consumption function in
the above equation.
Y= β1+ β2X +µ………………………….2)
Where u, known as the disturbance, or error term is a random (stochastic)
variable that has well defined probabilistic properties. The disturbance term u
may well represent all those factors that affect consumption but are not taken
into account explicitly.
The above equation is an example of econometric model. More technically, it is
an example of a linear regression model. This function hypothesizes that the
dependent variable Y(consumption) is linearly related to the explanatory
variable X(income) but that the relationship between the two is not exact; it is
subject to individual variation.
 4) Obtaining the Data
To estimate the econometric model mention above, there is a need to obtain
numerical values of β1 and β2 we need data.
5) Estimation of the Econometric Model
After obtaining data, next step is to estimate the parameters of the consumption
function. The numerical estimates of the parameters give empirical content to
the consumption function. The statistical technique of regression analysis is the
main tool to obtain the estimates. By using this technique and data, the estimates
of β1 and β2 can be obtained.
6) Hypothesis Testing
The obtained estimates are in accord with the expectation of the theory that is
being tested.
7) Forecasting or Prediction
If the chosen model does not refute the hypothesis or theory under
consideration, it can be used to predict the future value(s) of the dependent or
forecast, variable Yon the basis of the known or expected future (s) of the
explanatory, or, predictor, variable X.
8) Using of Model for Control or Policy Purpose
The estimated model can be used to control the issue or for further policy
suggestion or proposal.
Data
Economic data sets comes in various forms. We describe the most important
data structures encountered in applied econometrics.
1)Time Series Data
In time series data we observe the values of one or more variables over a period of
time (e.g., GDP for several quarters or years.
2) Cross-sectional Data
In cross section data, values of one or more variables are collected for several
sample units, or entities, at the same point in time (e.g., crime rates for 50 states in
the United States for a given year)
3) Panel Data
In panel data the same cross-sectional unit (say a family or a firm or a state) is
surveyd over time. In short panel data has space as well as time dimensions. There
are other names for panel data, such as pooled data (pooling of time-series and
cross-sectional observations) etc.
Research problem and Use of Statistical Soft ware
There are many statistical soft wares. We describe some important
software here.
1) SPPS 2) Eviews
What is SPSS?
It is a Statistical Package for the Social Sciences. SPSS is chosen because
of its popularity within both academic and business circles, making it the
most widely used package of its type. SPSS is also a versatile package
that allows many different types of analyses, data transformations, and
forms of output in short, it will more than adequately serve our purposes.
SPSS and Ordinary Least Square (OLS) Regression Technique
The dependence of variable on other independent variables. Suppose we
have few variables. The regression function is as follows.
Y= 𝛽1 𝑥1 + 𝛽2 𝑥2 + ⋯ 𝐵𝑘 𝑥𝑘
Y = dependent variable
𝑥1 , 𝑥2 , 𝑥𝑘 are independent variables.𝛽1 , 𝛽2 , … . 𝛽𝑘 are the coefficients.
We have to check the effect of these explanatory variables on the
dependent variables.
1 ) Linear Probability Model
We begin by an examination of the simplest possible model, which has a
dichotomous dummy variable as the dependent variable. We assume that
the dummy dependent variable is explained by only one regressor. For
example, we are interested in examining the labour force participation
decision of adult females. The question is : why do some women enter
the labour force while others do not? We assume that the decision to go
out to work or not is affected by only one explanatory variable (X2i) the
level of family income.
The model is:
Y= 𝛽1 + 𝛽2 + 𝑢𝑖
But since Yi is a dummy variable, we can rewrite the model as:
Di= 𝛽1 + 𝛽2 𝑥2 + 𝑢𝑖
Where 𝑥2𝑖 + is the level of family income (a continuous variable); Di is dichotomous
dummy defined as:
Di= {1 if the ith individual is working
0 if the ith individual is not working
And 𝑢𝑖 is the usual disturbance.
One of the assumptions of the CLRM is that E (𝑢𝑖)=0. This, for given 𝑥2𝑖
𝐸 𝐷𝑖 = 𝛽1 + 𝛽2 𝑥2𝑖
However, since Di is of qualitative nature, here the interpretation is different. Let us
define pi as the probability of Di =1 𝑃𝑖= 𝑃𝑟 𝐷𝑖 = 1 ; therefore 1-Pi is the probability of
… … … . . Di =0 1 − 𝑃𝑖= 𝑃𝑟 𝐷𝑖 = 0
To put this mathematically:

𝐸𝑑𝑖 = 1𝑃𝑟 𝐷𝑖 = 1 + 0𝑃𝑟 𝐷𝑖 = 0

𝐸𝑑𝑖 = 1𝑃𝑖 + 0 1 − 𝑃𝑖 𝐸
Edi=𝑃𝑖
This equation simply suggests that the expected value of Di is equal to the probability that th
ith individual is working. For this reason, this model is called the linear probability model
Therefore, the values obtained 𝛽1 𝑎𝑛𝑑 𝛽2 , enable us to estimate the probabilities that
women with a given level of family income will enter the labour force.
However, we observe that for negative level of income, we have negative probability and fo
higher level of income, we shall obtain probability higher than 1. So this is a problem. So a
alternative estimation method is required that will restrict the values of Di hat to lyin
between 0 and 1 is required. So logit and probit methods can solve this issue
2) Logit Model
In the linear probability model, we saw that the dependent variable Di on the left hand
side, which reflects the probability Pi, can take any real value and is not limited to
being in the correct range of probabilities-the (0,1) range. A simple way to resolve
this problem involves the following two steps. First, transform the dependent
variable, Di, as follows, introducing the concept of odds:
Oddsi= pi/1-pi
Here, oddsi is defined as the ratio of the probability of success to its complement (the
probability of failure). Using the labour force participation example, if the probability
for an individual to join the labour force is 0.75 then the odds ratio is 0.75/0.25=3/1,
or the odds are three to one that an individual is working. The second step involves
taking the natural logarithm of the odds ratio, calculating the logit, Li, as:
 𝑃𝑖
Li=ln
1−𝑃𝑖
Using this in a linear regression we obtain the logit model as:
Li=𝛽1 + 𝛽2 𝑥2𝑖 + 𝑢𝑖
It is easy to see that this model (which is linear to both the explanatory variable and
parameters) can extended to more than one explanatory variable, so as to obtain:
Li=𝛽1 + 𝛽2 𝑥2𝑖 + 𝛽3 𝑥3𝑖 + ⋯ … … 𝛽𝐾 𝑥𝐾𝑖 .
Notice that the logit model resolve the o, 1 boundary condition problem because:
as the probability Pi approaches o the odds approach zero and the logit (ln(0))
approaches−∞
As the probability Pi approaches 1 the odds approach +00 and thelogit (ln(0))
approaches + ∞
3) Probit Model
The probit model is an alternative method of resolving the problem faced by
the liner probability model of having values beyond the acceptable (0,1) range
of probabilities. To do this, obtain a sigmoid function similar to that of the logit
model by using the cumulative normal distribution, which by definition has an
S-shape asymptotical to the (0,1) range.
The logit and probit procedures are closely related that they rarely produce
results that are significantly different. However, the idea behind using the
probit model as being more suitable than the logit model is that most economic
variables follow the normal distribution and hence it is better to examine them
through the cumulative normal distribution. For the high degree of similarity of
the two models, compare the two sigmoid function as shown.
SPSS and OLS Application
SPSS and Descriptive Statistics
1: Open the data in spss
2: Go to Analyze
3: You have descriptive box.
4: Select the variables
5: Send the variables into the box
6: Go to options
7: Click on Mean, S.D, Minimum, Maximum, Kurtosis and Skewness.
8: Click on continue
SPSS and Regression Equation
1: Open the data in spss
2: Go to analyze
3: Go to regression (linear)
4: Put or shift the dependent variable in the dependent variable box
5: Put or shift the independent variables in independent variable box
6: Click on ok and you have the results
SPSS and Binary Logistic model
1: Open the data in spss
2: Go to analyze
3: Go to regression
4: Go to binary logistic model
5: Put or send the dependent variable into dependent box
6: Put the independent variables into independent box
7: Click ok
SPSS and Probit model
1: Open the data into spss
2: Go to analyze
3: Go to regression
4: Click the probit
5: Put or shift dependent variable in response frequency
6: Put or shift independent variable in covariates
7: Put or shift any variable in total observed
8: Press ok
Eviews
Eviews is the ideal package for quickly and efficiently managing
data, performing econometric and statistical analysis, generating
forecasts, or model simulations, and producing high quality graphs
and tables for publication or inclusion in other applications. Eviews is
easy to use, windows based statistical analysis package employed
worldwide by economists financial analysts, market researches and
policy analysts. Eviews supports these researches in a world range of
tasks, from analyzing economical financial data building models and
“what if” scenarios, to conducting research and teaching
econometrics, and estimating the impact of new policies or major
Eviews saves the time and effort. Eviews offers the statistical and
econometric tools you need for analyzing time series, cross sectional
and panel data. From basic descriptive statistics, measures association,
tests of equality, and principle components, to specialized feature such
as long-run variable calculation, causality testing, and unit-root and
co-integration diagnostics, Eviews offers a wide range of tools for
exploring the properties of your data.
Eviews is a complete software system that includes a sophisticated
data management system, powerful batch programing, and matrix
languages, and presentation quality graph and table creation tools to
support your entire work process.
Now we start our estimation.
1)Regression Equation
2)ADF Unit Root Tests
3)ARDL
4)Co-integration Diagnostics
5)Causality Tests
6)ARMA and ARIMA
7)ARCH and GARCH.
8)Panel Data Analysis
9)Random and Fixed Effects
Static Models
Suppose that we have time series data available on two variables, say
y and z, where yt and zt are dated contemporaneously. A static model
relating y to z is
Yt= 𝛽0 + 𝛽2 𝑍1 + 𝑢𝑡 t=1,2..n
The name ‘static model” comes from the fact that we are modelling a
contemporaneous relationship between y and z. Usually a static
model is postulated when a change in z in time t is believed to have
an immediate effect on yt=𝛽1 delta zt, when Δ 𝑢𝑡 = 0 these
models are used when we are interested in knowing the tradeoff
between y and z.
The time series process follows a model that is linear in its parameters.
yt=𝛽0 + 𝛽2 𝑥𝑡1 + ⋯ … . +𝛽𝑘 𝑥𝑡𝑘 + 𝑢𝑖
Where ut: t=1, 2…n is the sequence of errors or disturbances. Here, n is the number of
observations (time periods).
Stationary and Non-stationary Time Series
The notion of a stationary process has played an important role in the analysis of time
series.
a. A stationary time series process is one whose probability distribution are stable
over time in the following sense: if we take any collection of random variables in
the sequence and then shift that sequence ahead h time periods, the joint
probability distribution must remain unchanged.
.
Stationarity is important because, if the series is non-stationary, all the
typical results of the classical linear regression analysis are not valid.
Regression with non-stationary series may have no meaning and are
therefore called “suporios”
1)ADF Unit Root
2)Install Eviews 9 and put the data on desktop
3)Double click on eviews
4)Go into recent file and click on the data (data name)
5)Click on Next
6)Click on Next
7)Click on Finish
8)Click on Yes
9)Now you have the work file data
10) Select the variables (Press control and shift)
11) Click on quick
12) Click on series statistics
13) Click on unit root
14) Click on level, intercept, Schwarz criteria and maximum lag 2
15) Click on ok
16) Now you have found ADF Test statistics. If t calculated value is less than the
critical value at 1%, 5% and 10% then the variable is stationary. Otherwise if it is
not stationary at level then you have to check it at first difference and at second
difference.
17) If all the variables are stationary at level then you have to apply OLS regression.
18) If some variables are stationary at level and some are stationary at first
difference then you have to apply ARDL.
19) If all the variables are non stationary at level then you have to apply Johanson
co-integration.
Eviews and Summary Statistics
1: Select the variables
2: Go to quick
3: Go to group statistics
4: Descriptive statistics
5: Common sample
6: We have series of list
7: Click ok
8: We have summary statistics
Eviews and Correlation
1)Select the variables
2)Go to quick
3)Go to group statistics
4)Descriptive statistics
5)Correlation
6)We have series of list
Click ok
We have results
Eviews and OLS Regression
1: Select the dependent and independent variables
2: Right click on the selected variables
3: Go to open
4: Go to as equation
5: We have an equation estimation
6: Select LS in method
7: And press ok
8: We have results
ARDL Diagnostic Tests

1) Select the dependent and independent variables

2) Right click on selected variables
3) Click on equation
4) You have an equation estimation box
5) Click on ARDL method and we have results
6) Go to views, diagnostic test and then click on bound test and press ok
7) If value of F test is greater from all values on upper and lower bound then we reject
the H0 and accept the H1 which shows that there is a long-run relationship between
the variables
8) For long-run relationship again click on view
9) Click on Diagnostic test
10)Click on long-run form
11)You have results
Lag Adjustment
1)Select all variables
2)Click on right side of the variables
3)Click on VAR
4)Unrestricted VAR, Maximum lag 12 and press ok
5)Again click on view on the results
6)Click on lag structure
7)Click on lag length criteria
We have different criteria’s in different columns. As we have adjusted
lag 2. So before lag 2 on a/c criteria, we have a value having star. So our
adjusted lag will be 2. Because star shows exact lag selection.
 Johanson Co-integration Analysis
We can use johanson co-integration technique if all the variables are stationary at first
difference.
1) Click on variables
2) Click on quick
3) Click on group statistics
4) Click on johansen co-integration
5) Click ok
6) Click on last 6th one
7) We have six columns. On column 5, a/c information criteria we have a static value it
shows exact model lag.
8) So again choose model 5
9) We have results of Unrestricted co-integration rank test
10)Star on None shows very next relation. And this is the value
For Causality

1)Further for causality in johansen co-integration, select all the variables

2)Right click on the selected variables
3)Go to quick
4)Go to group statistics
5) Click Granger Causality
6)You have series list
7)Click ok
8)Lag to include 2
9)Click ok and you have results that shows the causality or not
Note: if the value of trace statistics and Eign values are different from zero or
are greater than critical value at 5% then we say that there exists cointegration
which shows long-run relationship among the variables.
Note: In Granger causality result, if the value of F statistics is significant , it
shows causality.
Interpolation of Data
1)Suppose if data of some years is missing then we have to interpolate the
data
2)This option is available in time series data
3)Go to process then click on interpolate
4) Click on linear interpolation and then press on ok.
 Eviews and ARMA &ARIMA Models
1) Autoregressive Integration Moving Average(ARIMA)
Box and Jenkins (1976) first introduced ARIMA models, the term deriving from:
AR=autoregressive; I=integrated; and, MA=moving average
The following sections will present the different versions of ARIMA models and introduce
the concept of stationary, which will be analyzed extensively
A time series yt said to be stationary if:
a) 𝐸 𝑌𝑡 = 0
𝑏) 𝑉𝑎𝑟 𝑌𝑡 = constant for all t; and
𝑐) 𝐶𝑜𝑣 𝑌𝑡 , 𝑌𝑡 + 𝑘 = constant for all t and all k=/ 0,
Or if its mean, variance and covariances remain constant overtime
 Where,

for simplicity, we don’t include a constant and 𝜙 , 1 and ut is a
Y  
t
Gussian (white noise) error term. The assumption behind the AR(1) model is
Y
t 1
t

that the time series behavior of Yt is largely determined by its own value in
the preceeding period. So what will happen in this largely dependent on
what happened in t-1.
Alternatively, what will happen in t+1 will be determined by the behavior
of the series in the current time t.
Condition for stationarity
The above equation introduces the constant 𝜙 <1 in order to guarantee
stationarity as defined in the previous section. If we have 𝜙 >1, then Yt
will tend to get larger in each period, so we would have an explosive
series.

Simply, the ar (1) model can be in the form of as

𝐸 𝑌𝑡 − 𝑒 𝑌𝑡 − 1 = 𝐸 𝑌𝑡 + 1 = 0
The AR (P) Model
The generalization of the AR(1) model id the AR(P) model; the number in the
parenthesis denotes the order of the autoregressive process and therefore the number of
lagged dependent variables the model will have. For example, the AR(2) model will be
autoregressive model of order two, and will have the form

𝑌𝑡 = 𝜙𝑌𝑡−1 + 𝜙2𝑌𝑡−2 + 𝜇𝑡
Similarly, the AR(P) model will be an autoregressive model of order p, and will have p
lagged terms, as in the following:

𝑌𝑡 = 𝜙 𝑌𝑡−1 + 𝜙𝑌𝑡−2 + 𝜙𝑌𝑡−𝑝 + 𝜇𝑡

Moving Average Models
The MA(1) model
The simplest moving average model is that of order one, or the MA (1) model, which has
the form

𝑌𝑡 = 𝜇𝑡 + 𝜃𝑡−1

Thus the implication behind the MA (1) model is that Yt depends on the value of the
immediate past error, which is known at time t.
The MA(q) model
The general form of the MA model is an MA(q) model of the form:

𝑌𝑡 = 𝜇𝑡 + 𝜃𝑡−1 + 𝜃2 𝜇𝑡−2 + ⋯ 𝜃𝑞 𝜇𝑡−𝑞

ARMA Model
ARMA model can only be made with time series 𝑌𝑡 that are stationary. This
means that mean, variance and covariance of the series are all constant over
time. However most economic and financial time series show trends over
time, and so the mean of 𝑌𝑡 during one year will be different from its mean in
another year. Thus the mean of the most economic and financial time series is
not constant over time which indicates that the series are non-stationary. To
avoid this problem, and to induce stationarity, we need to detrend the raw
data through process called differencing. The first difference of a time series
is 𝑌𝑡 are given by the equation:
Autoregressive Integrated Moving Average (ARIMA)
At most economic and financial time series show trend to some
degree, we nearly always take the first differences of the input series.
If, after the differencing, a series is stationary, then the series is also
called integrated to order one, and denoted 1(1) which completes the
abbreviation ARIMA. If the series, even after first differencing, is not
stationary, second differences need to be taken, using the equation:

ΔΔ = Δ2𝑌𝑡 = Δ𝑌𝑡 − Δ𝑌𝑡 -1

If the series becomes stationary after second differencing it is integrated of
order two and denoted by 1(2). In general if a series d times is differenced in
order to induce stationaity, the series is called an ARIMA (p,d,q) with p being
he number of lags of the dependent variable (the AR terms), d being the
number of differences required to take in order to make the series stationary ,
and q being the number of lagged terms of the error term (the MA terms)
Eviews and ARMA and ARIMA

1) We can use it on time series data. As it is a univariate analysis. We have to work on a

single series.
2) We have to check the stationarity. So we can use the ADF unit root.
3) Assumption is that the error term must be white noise. It means that there is no
autocorrelation in time series.
4) In case if there exits autocorrelation then error term is not white noise.
5) To solve this problem, we take change or lag series till the series are white noise.
6) The variable on which we are going to apply ARMA, it should be stationary.
7) For ARMA, apply unit root if variable is non-stationary
8) Select variable and right click on it and go into equation
9) Write the variable name and c and variable name and (-1)
10) Apply OLS and Press ok
1)We have results. Check if the value of DWT is 2 or near to 2 then the series
are stationary. As with lags we have more good results. We also say it AR
(1).
2)For second lag write variable name and c and variable name and (-1) and
variable name and (-2).
3)Apply OLS and press ok
4)Now check the value of DWT
 Moving Average (MA)

1) Here we have to take lags of error term.

2) Select the variable and right click on the selected variables and then go into equation
3) Write variable’s name and c and MA(1)
4) Apply OLS ARMA and press ok
5) We have the results. So we have to check the value of DWT. If it is ok then it shows that
there is no autocorrelation.
6) In fact correlogram tells us either we can use AR or MA
7) To check it click data time series
8) Go into view and click on correlogram and press ok
*if Auto Correlation Function (ACF) has geometric decay and Partial Auto correlation
function (PACF) has single peak then AR model is appropriate.
*if Auto Correlation Function (ACF) has single peak and Partial auto correlation
function (PACF) has geometric decay then MA can be applicable.
Autoregressive Integration Moving Average (ARIMA)
)If data series are nonstationary at level so we move for first difference
and it is called 1(1) integrated level (1)
)We use Jenkins model for forecasting series. And forecasting through
ARMA is best forecasting.
)We use Jenkins box
)Identification
)Estimation
)Diagnostic checking
)Again check the correlogram after the lag
)Go into Resid and click on open and view and on correlogram
) We have ACF and PACF
9) IF shaded areas or peaks are within the dotted lines then series are stationary
and if theses are outside the dotted line the series are nonstationary.
10) Sometimes we cannot get the series stationary at first difference by taking lag
then we apply the different combination of AR and MA and check the results
11) To do this select the variable and right click on the selected variable and go
into equation.
12) Write the variable name and write c and write the AR(1) Space MA(2)
13) Now check the results and value of Durban Watson test
14) We can also check the value of R2. Its value should be increased. And the
value of shawarz or ackaik. It should be low as compared to previous combination.
1) Combinations can be such as ARMA (1, 2), ARMA (1, 0), etc
2) If value of R2 is high and value of ackaik and shawarz is low then the
combination can be good.
 For Forecasting

1)If we want to forecast the results then we have results so on results

page go to for cast window and click ok
2)We have status of forecasting or evaluation of for cast results
i.Bias proportion ( it shows how much the mean of forecast value
is away from the mean of actual value)
ii.Variance( it shows how much the variance of forecast value is
away from the variance of actual value)
iii.Covariance Portfolio(it shows how much the covariance of
forecast value is away from the actual covariance of the series)
If the covariance portion is greater than bias and variance proportion then
the model is good.
*if it is less than both or anyone then we can say that the model is not good
forecasting.

ARCH and GARCH (Univariate Analysis in Time Series)

Autoregressive Conditional HeterosKedasticity (ARCH)
Engle’s model suggests that the variance of the residuals at time t depends
on the squared errors from past periods. Engle simply suggested that it is
better to model simultaneously the mean and the variance of a series when
it is suspected that conditional variance is not constant. Consider the simple
model
´
This is mean equation
Where xt is a k multiply 1 vector of explanatory variables and B is a k
multiply 1 vector of coefficients. Normally we assume that ut is
independently distributed with a zero mean and a constant variance
sigma2, or, in a mathematical notation:
𝜇𝑡 ~ 𝑖𝑖 𝑑𝑛 0, 𝜃 2
⃰ All these models are univariate in time series analysis. These ARCH
GARCH are used for financial series to capture the effect of uncertainty
(how much series are risky or not risky). So ARCH and GARCH
capture the behavior of conditional variance.
*According to Engel, the variance of the residual terms depends on the
squared error term from the past period
So μt is normally distributed with N (0, δ2)
2
*According to Engel 𝜎 2 = 𝑟0 + 𝑟1 + 𝑟1 𝜇𝑡−1

It depends on the squared error term

Eviews and ARCH
1) Write the variable name, write c and again write the variable (-1) its for mean
2) Now for δ2, select the variable and right click on the selected variable
3) Go into equation and write the variable name, write c, write the variable name
and write(-1)
4) Select ARCH
5) There will appear a window that shows our equation
6) Write the variable name and c and variable name (-1)
7) For ARCH set 1 and for GARCH set 0
8) Then press ok
9) We have two equations i.e., mean and variance
10) Check the value of DWT or value of Ackaik
11)One term is included as RESID (-1)2 in the result
Generalized Autoregressive Conditional Heterkscdasticity (GARCH)
Eviews and GARCH
In fact Engle criticized on itself. Our ARCH model looks like moving average model.
Engel asked we have to make it auto regression. He included one other terms in the
previous equation and give it the name of GARCH
𝜎 2 = 𝑟0 + 𝑟1𝑢2𝑡−1 + 𝑟3 δ3t−1 … 𝑡 − 1
include the lag of δ2 and then its name is GARCH
1) For estimation
2) Select the variable
3) Right click on equation
4) Write the variable, c and variable and then select ARCH
5) Set 1 for GARCH and press ok
6) We have results.
Panel DATA
Panel data estimation is often considered to be an efficient analytical method
in handling econometric data. Panel data set is formulated from a sample that
contains N cross-sectional units (for example countries) that are observed at
different T time periods. Consider, for example, a simple linear model with
one explanatory variables, as given by:
Yit=a +BXit+uit
Where the variables Y and x have both I and t subscripts for i=1,2,….N
sections and t=1,2,…T time periods. If the sample set consists of a constant T
for all cross section l units , or in other words if a full set of data both across
countries and across time has been obtained, then the set is called balanced.
We consider a case where in the sample there are different subgroups of
countries (for example high and low income etc and that differences are
expected in their behavior.
Our model becomes 𝑌𝑖𝑡 = 𝛼 + 𝛽𝑥𝑖𝑡 + 𝜇𝑖𝑡

Where ai can now differ for each country in the sample. At this point there
may be a question of whether the B coefficient should also vary across
different countries, but this would require a separate analysis for each one of
the N cross-sectional units and the pooling assumption is the basis of panel
data estimation.
Different Methods of Estimation
In general, simple linear panel data can be estimated using three different
methods: a) with a common constant as in equation
b) Allowing for fixed effects
c) allowing for random effects
a) the model estimates a common constant (pooled OLS method) a for all
cross-sections (common constant for countries). It implies that there are no
differences between the estimated cross-section and it is useful under the
null hypothesis that the data set is a priori homogenous.
Fixed Effects Method
In the fixed effects methods the constant is treated as group (section)-
specific. This means that the model allows for different constants for each
group (section). So the model is similar to that of first equation of the model.
The fixed effect estimator is also known as the least squares dummy variable
(LSDV) estimator because, to allow for different constant for each group, it
includes a dummy variable for each group.

Consider the following model = 𝛼𝑖 + 𝛽1 𝑥1 𝑖𝑡 + 𝛽2 𝑥2 𝑖𝑡 + ⋯ 𝛽𝑅 𝑥𝑅 𝑖𝑡 +

𝑌 𝑖𝑡
𝜇𝑡
Random Effects Methods
An alternative method of estimation a model is the random effects model. The difference
between the fixed effects and the random effects method is that the random effects
method handles the constants for each section not as fixed, but as random parameters.
Hence the variability of the constant for each section comes from:

𝐴𝑖 = 𝑎 + 𝑣𝑖

Where vi is zero mean standard random variable

The random effect model, therefore takes the following form𝑌𝑖𝑡 = 𝑎 + 𝑣𝑖 + 𝛽1 𝑥1 𝑖𝑡 +
𝛽2 𝑥2 𝑖𝑡 + ⋯ 𝛽𝑅 𝑥𝑅 𝑖𝑡 + 𝜇𝑖𝑡

𝑌𝑖𝑡 = 𝛼𝑖 + 𝛽1 𝑥1 𝑖𝑡 + 𝛽2 𝑥2 𝑖𝑡 + ⋯ 𝛽𝑅 𝑥𝑅 𝑖𝑡 + 𝑣𝑖 + 𝑣𝑖
Eviews and Panal Data Analysis
In panel data we will use both time series and cross section data. We have to stake the
data.
1) Short Panel (in short panel n >T) where n is no of cross section
2) Long Panel(T>n) where T is years or time
3) To work on short panel data is good
4) Here again in panel data we check the stationarity.
5) Is there a 1(0) means stationary at level or is there stationary at first difference 1(1)
6) If all the variables are stationary then we are able to apply regression fixed effects
and random effects.
7) Results from fixed and random can be similar but we have to check which model will
be more appropriate.
1)For this purpose we will have to apply Hausman test (Hausman test)
2)H0= random effect is appropriate
3)H1 = fixed effect is appropriate
4)If hausman test is significant then we will use random effect
5)If hausman test is insignificant then we can use fixed effect
6) We have a panel data.
7) Open data sheet
8)Select the data from original sheet and past it in eviews.
9)We have the variables with data.
1)Panel Data and Eviews
2)1)for data, double click on eviews
3)Click on file
4)Click on import
5)Click on data name, next, next and finish.
6)You have the data variable
1)Unit Roots and Panel Data
2)Select the variables
3) Go to quick
4) Go to series statistics
5) Click on Unit root
6)Enter the name and press ok
7)You have unit root tests
8) Click on level, intercept, and press ok
9) We have results of unit roots
10)If the statistics of all(i.e., LLC, IMP, ADF and PP) or few of them are
significant then the series are stationary.
OLS and Panel data

1) If all variables are stationary then select the variables

1) Right click on the selected variable and go into equation and select OLS and press ok
2) We have results. Check the value of DWT,F Statistics and R-Square.
3) If DWT is low from the 2 then it shows the autocorrelation. So we have to remove it.
4) To remove it simple add AR(1) in the end of the equation and press ok
5) If its value is improved then there is no autocorrelation
6) And if its value is not improved or again the value of DWT is low we can include MA
with AR (1) in the end of the equation.
7) There is a possibility that results cannot be improved by including MA with AR in the
end of the equation. So we should prefer to avoid it and we can include more AR (2)
with AR(1) in the end of the equation and then press ok.
8) In this way we can put different combination of like AR(1) AR(2)…
 For Fixed Effects
1:To check either we should apply fixed or random effect, so firstly we should
apply hausman test
2: For this select the variables the right click and click on open as and go to
equation.
3: Click on panel and click on option
4: Click on random in cross section box and non in period box
5: Press ok
6: We have the results and go into view and click on fixed/random testing and
click on hausman test
8: We have the chi-square results
ARDL and Panel Data
1) We can apply ARDL when some variables are stationary at level and some
variables are stationary at first difference.
2) Select the dependent and independent variables
3) Double click on selected variables
4) Go to quick
5) Go to equation
6) Select ARDL in methods
7) Click on ok
8) You have the results of short run and long-run relationship
9) Value of ECM should be negative and significant which shows the long-run
Again repeat the process and you can apply random effects
11: If there is any autocorrelation which the value of DWT shows then we can use AR but
it is not necessary.
12: So we should just focus on value of R2 and Adjusted R2.
13: And if H0 is insignificant then we have to apply fixed effects
14: For this again select the variables and right click on open as and go into equation
15: Click on panel and panel option
16: Click on fixed in cross section and non in period
17: Then press ok
18: We have the results.
Note: if the value of DWT is very low or high then go into estimate, click on panel option
and apply the white period.