[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

# Re: st: re: What to do about multiple observations for one individual in one same period in a panel

 From Austin Nichols To statalist@hsphsun2.harvard.edu Subject Re: st: re: What to do about multiple observations for one individual in one same period in a panel Date Fri, 24 Jul 2009 22:07:27 -0400

```Adrian <kokootchke@hotmail.com> et al.--
I think clustering is the answer to the repeated obs per time period
as I have repeated numerous times.  I don't think a selection model
solves the problem of important correlated unobservables--you have to
go for IV or another method that uses additional information rather
than a functional form assumption IMHO.  I.e. you want -xtivreg2- (on
SSC) with the -cluster- option.

On Thu, Jul 23, 2009 at 4:03 PM, kokootchke<kokootchke@hotmail.com> wrote:
> Austin Nichols said:
>
>> I will just reiterate that it is not clear to me what the nature of
>> the selection problem is in your application. Perhaps it would help
>> if you gave us a sentence or two about what is causing what (what
>> variables are on the right hand side and what the outcome variable is,
>> exactly) in your model, and why. Note that the selection problem
>> addressed by -heckman- is about selection on the error term, not just
>> a low wage related to "low" observables (selection on X is okay), and
>> I don't think countries observe the error term and decide not to issue
>> bonds--it is more likely that the decision not to issue bonds in a
>> year is related to observables, but that there is some endogeneity in
>> the right-hand-side variables to worry about, and dynamics to account
>> for (e.g. you issue bonds this year to cover interest payments on last
>> year's debt).
>
> Well, I think the selection issue is both related to observables and unobservables. The regression model I have right now looks like:
>
> spread = b0 + b1*(GDP growth) + b2*(External debt/GDP) + b3*(Debt service/Exports) + b4*(Reserves/imports) + ... + e
>
> so the spread of a bond depends on macroeconomic fundamentals... and an error term. The first problem is that the spread is only observed whenever countries actually decide to issue a bond. This decision to participate or not participate may depend on observables like you suggest (e.g., if they owe a lot or if they have lots of interest payments due soon -- meaning their external debt and/or debt service ratios are high, they may want to issue another bond to raise additional funds to meet those impending obligations) or unobservables (e.g., market sentiment, changes in investors' risk appetite, etc.). I try controlling for some of these "unobservables" by including some "global" variables that proxy for credit and liquidity conditions abroad, but I still think there are other elements in the error term that my variables won't capture... which is why I'd like to implement a selection model a la Heckman.
>
> The other problem is, as I said earlier, that the frequency of the macro variables is not the same as the frequency of the bond issuances. So, Thailand may decide to issue a bond on January 15 and then another one on March 15, and in both cases, investors observe the same macro variables for that particular quarter:
>
> country  date      qtr    spread gdpgwth reserves
> Thailand 15-Jan-90 1990:1 250    1.8     50
> Thailand 15-Mar-90 1990:1 375    1.8     50
>
> This shows 2 bonds issued at different dates and different spreads, but the macro variables for that quarter are the same.
>
> Kit and Martin suggested I aggregate bonds issued in a given quarter... something I had already considered, but I wanted to know if there were other methods that people have used for these sorts of problems.
>
> Thanks again, Austin, and everyone...
>
> Adrian
>
>
>>
>> Also, you can cluster on countryyear to account for multiple obs per
>> country-year cell, but clustering on country makes more sense (with
>> time dummies for fixed year or month effects related to global capital
>> markets), and incorporates that smaller correction already.
>>
>>
>> On Sun, Jul 19, 2009 at 3:27 PM, kokootchke wrote:
>>> Nick, Kit, Martin... thanks a lot for your suggestions. The only one that had occurred to me was the one that averages spreads (yes, Kit, I have yield spreads, not prices) and other variables by country whenever there are multiple bonds issued in a given time period... I was not a fan of this approach because I would "lose" observations... but then again, I don't know exactly how much I'm actually losing, if all my macro variables for each of those "repeated" observations in one same time period have the same value (different spread for each bond... but same GDP growth, debt/GDP, inflation, etc.).
>>>
>>> Anyway, let me try these suggestions. Thanks a lot!
>>>
>>> Adrian
>>
>> *
>> * For searches and help try:
>> * http://www.stata.com/help.cgi?search
>> * http://www.stata.com/support/statalist/faq
>> * http://www.ats.ucla.edu/stat/stata/
>
> _________________________________________________________________
> Bing™ brings you maps, menus, and reviews organized in one place. Try it now.
> http://www.bing.com/search?q=restaurants&form=MLOGEN&publ=WLHMTAG&crea=TXT_MLOGEN_Local_Local_Restaurants_1x1
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```

 © Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index