Scribd
Upload
324 views3 pages

ARDL

A simple transformation of The ARDL model, termed the "bewley transformation," allows asymptotically valid inference using the t-statistics on the long-run coefficients. Incorporating more information in the regression is likely to yield estimators with improved characteristics.

Uploaded by

Mohammed Seid
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
324 views3 pages

ARDL

A simple transformation of The ARDL model, termed the "bewley transformation," allows asymptotically valid inference using the t-statistics on the long-run coefficients. Incorporating more information in the regression is likely to yield estimators with improved characteristics.

Uploaded by

Mohammed Seid
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/ 3

Unpublished appendix to Developing Country Exports of Manufactures: Moving Up the Ladder to Escape the Fallacy of Composition?

A More General ARDL Model

While the Engel-Granger OLS estimation of a static levels regression has the properties of being economical and superconsistent, it carries a nite sample bias, suggesting that superior estimates might be obtained by accounting for the short-run dynamics. Simulations suggest that including dynamic terms has benets in nite samples.1 While the I(0) terms are not important asymptotically, they do assume importance in small samples. The ARDL model can be reparametrized to yield an error correction model. However, the particular form of the ECM yielded by such a reparametrization is not particularly convenient to estimate when the cointegrating parameters are unknown, given the non-linear estimation required due to the interaction between and j .2 A simple transformation of the ARDL model, termed the Bewley transformation, allows asymptotically valid inference using the t-statistics on the long-run coecients. Such a transformation oers an alternative way of estimating the cointegrating relationship which has some nite sample advantages over the Engel-Granger approach. As suggested by the simulations in Inder (1993), incorporating more information in the regression (in this case, via the Bewley transformation) is likely to yield estimators with improved characteristics. Consider the following ARDL(p,q) model:3
q p

Yt = 0 +
j=0

j L Xt +
i=1

i Li Yt +

(1)

where L represents the lag operator.4 This can be reparameterized as follows:


p q

1
i=1

i L

Yt = 0 +
j=0

j Lj Xt +

(2)

Now,
1 2

See Patterson (2000), for example. See, for instance, Wickens and Breusch (1987) for a discussion. 3 This particular derivation is based on Patterson (2000). 4 For example, L0 Xt = Xt , L1 Xt = Xt1 , and so on.

j Lj Xt =
j=0 j=0 q

j Xt
j=1

j Xt +
j=1 q

j Xt1
q

j=2 q

j Xt1 +
j=2 q

j Xt2
j=3 q

j Xt2 + . . . q Xtq+1 j Xt1 . . . q Xtq+1

=
j=0

j Xt
j=1 q

j Xt
j=2

= B0 Xt
j=0

Bj Xtj+1

(3)

where B0 = Also,
p

q j=0

j , B1 =

q j=1

j , . . . , Bq = q .

1
i=1

i L i Y t = Y t
i=1 p

i Li Yt
p p p

= Yt
i=1 p

i Y t +
i=1 p

i Y t
i=1

i Yt1 +
i=2

i Yt1

i=2

i Yt2 +
i=3 p

i Yt2 + . . . + p Ytp
p p

1
i=1

i Y t +
i=1 p

i Yt +
i=2

i Yt1 + . . . + p Ytp+1 (4)

= (1 1 )Yt +
i=1

i Ytj+1

where 1 = p i , 2 = p i , . . . , p = p . i=1 i=2 Thus the ARDL(p,q) model of equation (2) can be rewritten as: ( q Bj ) ( p i ) B0 0 t i=1 i=1 + Xt Xtj+1 Ytj+1 + 1 1 1 1 1 1 1 1 1 1

Yt = or,

(5)

Yt = 0 + B0 Xt Bj Xtj+1 Ytj+1 + j

(6)

Note that the long-run coecients 0 and B0 Xt are now isolated, and the remaining variables are stationary. Note also that the long-run solution to the model in equation (6) can be found by setting Xt = Yt = t = 0. This transformation is particularly useful when Yt and Xt are I(1). Finally, note that when p = q = 1, 1 = 1 , B0 = 0 + 1 , B1 = 1 , and equation (6) reduces to:

Yt =

0 0 + 1 1 1 t + Xt Xt Yt + 1 1 1 1 1 1 1 1 1 1 2

(7)

Due to the presence of contemporaneous links between variables, the Bewley transformation requires the use of instrumental variables.5 Typically Yt1 is used as an instrument for Yt .

References
Inder, B. (1993). Estimating long-run relationships in economics. Journal of Econometrics, 57, 5368. Patterson, K. (2000). An Introduction to Applied Econometrics: A Time Series Approach. St. Martins Press, New York, U.S.A. Wickens, M. R. and Breusch, T. S. (1987). Dynamic specication, the long-run, and the estimation of transformed regression models. Economic Journal, Conference 1988, 189 205.

Notice that the presence of a rst-dierenced instance of the dependent variable on the right hand side of equation (7) means that the level instance of the same variable is present on both sides.

You might also like