Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


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

Re: st: event study


From   Austin Nichols <[email protected]>
To   "[email protected]" <[email protected]>
Subject   Re: st: event study
Date   Mon, 9 Sep 2013 12:06:47 -0400

Andrew Reed <[email protected]>
The stated desideratum makes no sense:
"the change in the natural logarithm of entity i's exchange rate in
the previous three windows, surrounding an event, where the event
occurs at t=0"
--what are the "previous three windows" there?

Are you assuming the event has a one-time impact during the window,
and does not have an impact on future changes, after the window?  That
is what your equations seem to imply. If so, why would you not measure
the change from t = -1 to t = +1, +3, or +7 instead? I.e. exclude
other changes over periods of that length that overlap your window.

For that, you would define your downgrade dummy as 1 on the end date
of the window and missing elsewhere in the window (to use data before
and after to establish the counterfactual), then define your depvar as
y=EX-L2.EX or EX-L4.EX or EX-L8.EX as appropriate.

clear
range date -5 12 18
g ex=exp(rnormal()+(date>=3)/2)
tsset date
g event1=date==4 if !inrange(date,3,3)
g y1=ex-L2.ex
g event3=date==6 if !inrange(date,3,5)
g y3=ex-L4.ex
g event7=date==10 if !inrange(date,3,9)
g y7=ex-L8.ex
reg y1 event1, nohe
reg y3 event3, nohe
reg y7 event7, nohe
list, noo

You should also worry about correlations across observations; exchange
rates are not independent across entities!  You should think carefully
about clustering of errors.

You will get some very bad estimates if your event in fact affects
changes in all future periods instead of just having a one-time impact
during a short window, of course.


On Mon, Sep 9, 2013 at 11:00 AM, Andrew Reed <[email protected]> wrote:
> Dear Statalisters,
>
> I am implementing an event study methodology so as to judge whether credit rating news announcements affect 9 exchange rates for 72 entities. I am currently wondering how to best go about create event windows for the following event windows:
>
> [-1,+1]
> [-1,+3]
> [-1,+7]
>
> I want to measure the change in the dependent variable of my specification, i.e. the change in the natural logarithm of entity i's exchange rate in the previous three windows, surrounding an event, where the event occurs at t=0.
>
> My specification is the following.
>
> d.EX_i,s = a0 + a1_upgrade_i,t + a2_downgrade_i,t + a3_credit rating_i,t + a4_bond spread_t + error
>
> Where I get confused is how to do this. At first I thought I would just have a full measure of the change in exchange rates and then, for example with the [-1,+1] window, I would just augment the actual event in order to account for this. Let's use the following example, using completely arbitrary numbers, but looking at a downgrade at date 3.
>
> date  d.ex    downgrade
> 1        +0.5           0
> 2         +0.5          1
> 3        -0.4           1
> 4        -0.3           1
> 5        +0.2           0
> 6        +0.4           0
>
> So the dummy for downgrade establishes the window. After talking with a friend I'm not sure this works. We have thus contemplated using a dummy variable, much like the downgrade variable just seen, in order to establish the window with the exchange rate variable itself. In multiplying this binary variable with the d.ex variable and just counting a downgrade as happening at date 3, the observations would look like the following:
>
> date   event[-1,+1]   (d.ex)*(event[-1,+1])             downgrade
> 1               0                               0                                       0
> 2               1                               +0.5                            0
> 3               1                               -0.4                                    1
> 4               1                               -0.3                                    0
> 5               0                               0                                       0
> 6               0                               0                                       0
>
> I hope this makes sense. This is the last hurdle I need to clear before running regressions for a final writeup of results. Any input is helpful and I thank you beforehand for your thoughts.
>
> Best,
>
> Drew

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


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index