Prof. R C Manocha Autocorrelation: What Happens If The Error Terms Are Correlated?

Prof.
r c manocha
AUTOCORRELATION
WHAT HAPPENS IF THE
ERROR TERMS ARE
CORRELATED?
TYPES OF DATA AVAILABLE
FOR EMPIRICAL ANALYSIS
• Generally, three types of data are
available for empirical analysis:
1. Time series data
2. Cross-section data
3. Pooled-data : combination of time series
& cross-section data.
TIME SERIES DATA
• A time series data is a set of observation values that a
variable take at different times.
State Y 1 Y2 X1 X2
Delhi 200 210 1200 1300
Punjab 1500 1700 800 950
Haryana 1100 1050 930 1100
Y 1 =Potatoes produced in year 2007(tonnes)
Y2 =Potatoes produced in year 2008(tonnes)
X1= Price of potatoes per tonnee in year 2007
X 2 =Price of potatoes per tonnee in year 2008

Note: money supply is increasing each year.
CROSS-SECTION DATA
• Cross-section data are data on one or more
variables collected in the same point in time.
• Example: census data collected by Census
Bureau every 10 years.
• In above example, we have two cross sectional
tables for three states-one for the year 1990 &
the other for the year 1991
• Cross sectional data create their own problems.
Specifically, “heterogeneity”.
SPATIAL AUTO CORRELATION
• EXAMPLE : cross section data:
In cross section studies, data are often collected
on the basis of a random sample of cross
sectional units, such as “households”
(consumption function analysis) or “firms”( in
investment study analysis). If by chance, the
error pertaining to one household or firm is
correlated with the error term of another house
or firm, then such a correlation shall be called
“spatial autocorrelation”.
SERIAL CORRELATION
• If you observe stock prices (say in BSE) ,
it is not unusual to find ups & downs in
shares for several days in succession
( bulls or beers). Obviously, in such a
situation, data follows a natural ordering
over time so that successive observations
are likely to exhibit inter-correlations.
This type of autocorrelation is called Serial
Correlation.
POOLED DATA
• In pooled data, data are of both types-
time series as well as cross section data.
• Example: for each year, we have 3 cross
sectional observations and for each state,
we have two time series data-one for the
year 1990 & the other for the year 1991.
AUTOCORRELATION DEFINED
• If there exists correlation between
members of the series –ordered in time
(time series data) or space( cross section
data), it is called autocorrelation.
• In first case, it is called serial
autocorrelation
• In second case it is called space
autocorrelation.
contd
• Most economic data consists of time
series and there is very often a correlation
in the errors corresponding to successive
time-periods. This is the problem of
autocorrelation or serial correlation.
• The error term μ t at time period t is
correlated with error terms μt+1, μt+2 and
μt-1 , μ t-2 and so on.
contd
• Such correlation in the error term arises due to omitted
variables that the term captures.
• Correlation between μt and μt – k is called an auto
correlation of the order k.
• Correlation between μt and μt – 1 is called an auto
correlation of the first order and is denoted by ρ1 and so on.
•
•
•
Does autocorrelation exist in
classical regression model?
• No.
There is no autocorrelation in this case.
Hence there is no error in disturbances;
Therefore E ( ui uj ) =0 i ≠ j
This means that disturbance term relating to
any one term is not influenced by the
disturbance of any other term.
What happens when we have
autocorrelation?
• The disturbance of any one term can
effect the disturbance of another term.
• E ( ui uj ) ≠ 0 i ≠ j
PATTERNS OF
AUTOCORRELATION
There are :
• CYCLIC PATTERN
• UPWARD PATTERN
• DOWNWARD PATTERN
• BOTH-LINEAR & QUADRATIC
PATTERNS
These do not support the classical model as
the error terms are correlated.
REASONS OF
AUTOCORRELATION
1. Inertia or sluggishness of economic time
series.
2. Specification bias resulting from
excluding important variables from the
model or using incorrect functional form.
3. The cobweb phenomenon
4. Data massaging; &
5. Data transformation
EFFECT OF CORRELATION ON OLS
(ORDINARY LEAST SQUARES)
• OLS estimators remain unbiased.
• OLS estimators remain consistent; &
• Asymptotically normally distributed.
However, they are no longer efficient
(minimum variance is not there.)
2
Hence the usual t, F and X (chi square)
test cannot be legitimately applied.
REMEDIAL MEASURES
• The remedy depends upon the nature of
interdependence among the disturbances ui. But as the
disturbances are unobservable, the common practice is
to assume that they are generated by some mechanism.
• The mechanism most commonly used is “Markov first
order autoregressive scheme, which assumes that the
disturbance in the current time period is linearly related
to the disturbance term in the previous time period &
ρ(rho)-the coefficient of autocorrelation provides the
extent of interdependence.
• This mechanism is known as AR(I) scheme.
DETECTION OF
AUTOCORRELATION
• The Durbin-Watson (DW) test is the most
often used to test for the presence of
autocorrelation.
• This test is applicable only for small
samples and is appropriate only for the
first order autogressive scheme
( ut = ρ(rho)ut-1 +vt ) where rho is the
coefficient of first order serial correlation.
DW TEST PROCEDURE
• H0 : ρ(rho)=0 i.t u’s are not autocorrelated with first
order scheme.
• H1 : ρ ≠ 0
To test the null hypothesis, we use the Durbin-Watson

statistic:
t=n 2 t=n 2
d=∑(μt – μt-1) / ∑ μt
WHERE μt is the estimated residual for period t
t=2 t=1
Contd.
2 2
Since ∑μt and ∑ μ t-1 are approximately

equal , if the sample is large.
We have d is apprx equal to 2(1-ρ)
If ρ = +1 d=0
If ρ = -1 d=4
From graph, we can see that
When d is close to 0 or 4, the residuals are
contd
Highly correlated.
d=0 d=2 d=4
If d is less than dl we reject the null hyp of
no autocorrelation
If d is greater than du we donot reject the
null hyp of no autocorrelation
If d is greater than dl but less than du we say
that the test in inconclusive.

Prof. R C Manocha Autocorrelation: What Happens If The Error Terms Are Correlated?

Uploaded by

Copyright:

Available Formats

Prof. R C Manocha Autocorrelation: What Happens If The Error Terms Are Correlated?

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Prof. R C Manocha Autocorrelation: What Happens If The Error Terms Are Correlated?

Uploaded by

Copyright:

Available Formats

Prof.

X 2 =Price of potatoes per tonnee in year 2008

To test the null hypothesis, we use the Durbin-Watson

Since ∑μt and ∑ μ t-1 are approximately

You might also like