Dear S.J. Balaji,

my general suggestion is to test for structural break before imposing it in your estimate.

In order to select a brek-point I suggest to look carefully at the data and the phenomenon under investigation.

In particular, in time series break point will be evident in that if a break point exist, it would be easier to "see" two (or more) trends in a series, rather than a single trend.

Moreover, an investigation of the phenomenon would suggest if and when a break point should be tested. For instamce, if a change in policy or economic conditions occur at a specific date, you may test for a structural break at that point.

A third way, is to impose the structural break in time series, and then test if the process differ under the two process (pre and post-break). The break may occurr several times, and in this case you will be in the case of regimes.

The third case pretian to a couple of work I have done recently. You will find references and methdological detail in the working papers and in the articles. Hope will be of interest.

Santeramo, Fabio Gaetano & Cioffi, Antonio, 2012. "The entry price threshold in EU agriculture: Deterrent or barrier?," Journal of Policy Modeling, Elsevier, vol. 34(5), pages 691-704.

Available at: http://ideas.repec.org/p/pra/mprapa/36431.html

Antonio Cioffi & Fabio Gaetano Santeramo & Cosimo Damiano Vitale, 2011. "The price stabilization effects of the EU entry price scheme for fruit and vegetables," Agricultural Economics, International Association of Agricultural Economists, vol. 42(3), pages 405-418, 05.

Available at: http://ideas.repec.org/p/pra/mprapa/24828.html

Hi Fabio,

Thanks for the suggestion. My ideal actually is one your second and third points. I have data series for few crops. While some exhibit breaks in trend, some don't. I think i could employ appropriate structural break tests to test their presence and number of breaks, and then try to trace some evidences. I will get back to you once i read your papers.

Thanks

Sounds a good idea.

Please feel free to ask any information or clarification.

Regards,

Fabio

One useful way to look for a potential breakpoint is to use recursive residuals. The equation is estimated multiple times over an expanding time window and the recursive residual is the residual for the period immediately following the equation. When equations go off track, often the recursive residuals will jump. Easy to do in Eviews.

Hi John,

Thanks. I tried recursive method and my doubt in that case is on "window size". I find different break years when the window size is different. Also, when the window size is say 10, it would mean that the break year is in between 10 point gap. Averaging those points would be good to obtain the break-point?

Fabio & John,

Today i was trying with Quandt-Andrews unknown breakpoint test in Eviews. It allows one to shift either the intercept alone or trend alone or both. What would it mean in "real sense" that an "intercept shift" or a "trend shift" or "both". Im little confused and your suggestions would improve my understanding...............

Hi,

say you are considering a time series of a proces, say rainfall seriess, or price series, or any other phenomenon.

If the intercept shift you can think as if the process does not change, but we have a sudden shock that affect the starting point. For istance, aftrer time t the rainfall, or the prices, have a suddenshock, but their processes over time do not change.

If the trend shift, it is as if the process change, say speeding up or slowing down.

If trend and intercept change, both situation occur.

Thanks Fabio,

In my case, break-points are same in both the cases (2 breaks with 1% significant) when i allow either to shift intercept or slope. But when i try to allow both, break point what i obtain is new at 1% significance. Is it possible ? or i'm using a wrong method?

Thanks....

It is possible to have 1 percent in all cases.

I suggest to consider when the break points occur in all cases and try to understand what is the most reasonable case, looking at the phenomenon.

See my article, it may help somehow...

SJ Recursive residuals do not depend on choice of a window. They are a sequence of one-period-ahead residuals that result from reestimating the equation. Since the equation is not forced to 'fit' the data point the residual is being calculated from, they can be very useful. E-views will calculate them for OLS.

Thanks John....... Will try....... What did the term "not forced to fit" exactly mean? I can't get......