Cointegration with Multiple Structural Breaks

[Kahn]

Hello,

I am very new to RATS so please accept my apologies in advance if this question is too simplistic.

I am studying whether a group of emerging stock-markets are cointegrated over the period 2012-2016 using daily data. Having performed the Bai-Perron multiple break tests I found that there are apparently 4 or 5 breaks spread out quite evenly. Given this, is there anyway I can test for cointegration as the literature I have read suggests using the Gregory Hansen test for one break or using small sub-samples. However, as there are so many breaks in 4 years of data, wouldn't the sub-samples be too small to draw any reliable inference?

Kahn.

[TomDoan]

Bai-Perron (at least the critical values) doesn't apply to non-stationary regressors and besides that, a cointegrating regression will generally be quite misspecified dynamically (really substantial serial correlation) so you can't really rely upon the results of that. If it's a "group" of series, wouldn't there likely be multiple cointegrating vectors? If that's the case, then running a single cointegrating regression is likely to give very misleading results. Any linear combination of cointegrating vectors is also cointegrating so what you might be detecting is changes to the dominant CI but not a change to the space of CI's themselves.

[Kahn]

Hello Tom,

Thanks for your reply.

Yes, my data includes a group of 12 indices from the GCC economies. The problem is that when I test for cointegration using the Johansen test the results suggest no cointegration whatsoever. Following this result I conducted tests for structural breaks as breaks can distort the findings from standard cointegration tests. Therefore, I currently have results suggesting no cointegration between the series and the presence of structural breaks. Note that my sample is for 2012-2016 which was quite volatile in terms of political and economic turmoil in the region. I am unsure as to how I should proceed further with this analysis. What tests etc could I run to account for all these factors? Any advice is greatly appreciated.

Kahn.

[TomDoan]

If they're indexes in local currency, wouldn't they be knocked around by exchange rates?

[Kahn]

Hello Tom,

They are all expressed in USD but I see the essence of what you have said.
Please see the attached picture which is unmarked but each diagram shows two indices from one GCC country. Note that I have indexed all of the series over the sample period and although this may not be strictly correct I just wanted a more normalised comparison of the movements.
From pure eye-balling, do these seem cointegrated? My supervisor is unwilling to accept a result with no cointegration between stock indices but is also not giving much advice on how to proceed.. Please provide any guidance possible.

[TomDoan]

First off, all those should be analyzed in log form, so scaling to start at 100 should have no effect. The first five are (to the eye certainly) not even close to being cointegrated. The bottom right one looks very odd as the orange line seems to be running almost exactly three months ahead of the other one. Is it possible that one of those is misdated?