Scribd
Upload
78 views

Assignment 3 (Text)

This document discusses serial correlation in time series data and various tests to detect and correct for it. It addresses the following points: (a) Plot of S&P500 returns shows fluctuations but no clear trend, with a positive mean return of 10%. (b) The Breusch-Godfrey test rejects the null hypothesis of no serial correlation at all significance levels. (c) Lagging the regression residuals yields a statistically significant AR(1) coefficient of 0.237, indicating some positive serial correlation. (d) Residuals are serially correlated at 5% but not 99% significance, inconsistent with the assumed AR(1) model if true. (e) The

Uploaded by

lolDevR
Copyright
© © All Rights Reserved
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
78 views

Assignment 3 (Text)

This document discusses serial correlation in time series data and various tests to detect and correct for it. It addresses the following points: (a) Plot of S&P500 returns shows fluctuations but no clear trend, with a positive mean return of 10%. (b) The Breusch-Godfrey test rejects the null hypothesis of no serial correlation at all significance levels. (c) Lagging the regression residuals yields a statistically significant AR(1) coefficient of 0.237, indicating some positive serial correlation. (d) Residuals are serially correlated at 5% but not 99% significance, inconsistent with the assumed AR(1) model if true. (e) The

Uploaded by

lolDevR
Copyright
© © All Rights Reserved
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/ 2

Advanced Econometrics - Assignment 3

Devvrat Raghav

The log file for this assignment has been emailed.

(a.) The returns on S&P500 seem to fluctuate between positive and negative values for successive
months, although no clear temporal trend emerges from the plot. However, the mean return appears
to be positive and lies at approximately 10%.

(b.) We use the Breusch-Gofrey test for AR(1) serial correlation, where the null hypothesis is the
absence of any serial correlation. Post-estimation, we reject the null hypothesis for all levels of
significance, given that the p-value of the Chi-Square statistic is = 0.

(c.) Regressing the residuals on their lagged counterparts yields a coefficient of 0.237. This is the
sample estimate for ρ, and it is statistically significant. Even so, the magnitude of the coefficient
isn’t extremely large, since most of the variation in our estimated residuals appears to be indepedent
of previous residuals. Hence, it would be more prudent to claim that there is some positive serial
correlation, even though correcting for serial correlation is still recommended.

(d.) At a 5% level of significance, we find that the errors (et ) are serially correlated. However, if
we make use of the 99% confidence level, then the et are found to be not serially correlated. If the
true model is indeed AR(1), then et must not be serially correlated. After all, the AR(1) process is
predicated on the assumption that the residuals can be decomposed into two parts - one of which
depends on previous residuals, and one which is independent of them (et ). Hence, if et are serially
correlated, then AR(1) can not be the true model.

(e.) To pick the best AR(p) structure, we look at the partial autocorrelation function between the
residuals and their lagged counterparts, which is done using either the corrgram or pac commands
in Stata. This allows us to identify the statistical significance of each lag (k), such that the lag is
deemed significant only if its partial autocorrelation function is sufficiently distant from zero (in either
positive or negative direction). For e.g, if the last significant lag is the 5th, then the appropriate AR
structure is AR(5). In our analysis, we find that only the first lag is statistically significant at the 99%
confidenve level, which means that an AR(1) structure should be the most appropriate model.

1
(f.) We can correct for serial correlation in residuals by using the Prais-Winston algorithm. First,
we run the initial regression of rsp500 on pcip and i3 and obtain the residuals (ût ). Then, these
residuals are regressed on their lagged counterparts, i.e.

ût = ρût−1 + et ∀ t ∈ [1, n]

The estimated coefficient of ût−1 is ρ̂, which is then used to transform the initial regression model.
Specifically, the transformed model has the form:

yt − ρ̂yt−1 = β0 (1 − ρ̂) + β1 (xt − ρ̂xt−1 ) + (ut − ρ̂ut−1 )

This can be simplified into:


ỹt = α0 + α1 x̃t + et

Under the assumption that AR(1) is the true model for serial correlation, we know that et are not
serially correlated. Hence, the objective has been achieved.

You might also like