I have daily closing values of multiple stock indices and i have lists of public holidays for the whole period. I want to inspect the existence of a holiday effect. My regression equation is as follows:
R = c + beta1*DBt + beta2*DAt + e
R = log daily return
DBt = equals one if t is the day before a holiday. i.e. pre-holiday, zero otherwise.
DAt = equals one if t is the day after a holiday, i.e. post-holiday, zero otherwise.
How can I create such dummy variables in EViews or Stata?
It will be good idea to add a third variable without the holidays, which will have 1 throughout. This will be a better specification. The interpretation of the coefficients is important. It is a good question that is researchable.
This can be done very simply in Stata. I assume you have the necessary variables (index, time, holiday) but change them to the correct naming conventions as needed:
*begin stata code
bysort index (time) : gen DB_t = holiday[_n+1]
bysort index (time) : gen DA_t = holiday[_n-1]
*end state code
Another way to do this is to set your data into a panel dataset format in Stata.:
*begin stata code
encode index, gen(panel) // index has to be numeric, skip this if it already is
xtset panel time
gen DB_t = f1.holiday
gen DA_t = l1.holiday
*end stata code
This second method is a efficient way if you then want to do robustness checks with two-period leads/lags of holidays.
Dear Maged Nada
If you build your model in this way, you fall into the dummy variable trap.
you need to add a single dummy variable.
You can create dummy with the GenDummy add-ins in Eviews.
If I understand the problem Maged Nada is posing, I don't think there is a dummy variable trap so long as there are some observations that are neither pre- nor post-holiday.
There are roughly 250 trading days and 10 holidays per year. Therefore you should have 4% of observations be treated with pre-holiday and 4% treated with post-holiday, with zero observations as both.
You're right, Jeffery! The c in the model incorporates the non-holidays (trading days).
I assume you are using STATA
I assume you have annual data 1970 to 2019
I assume the data is stored in a variable called year
I assume 1990 is the event date
To generate a dummy variable for pre and post event, you type in one line the following:
recode year (1970/1989=0 "Before") (1990/2019=1 "After") , gen(dummy)
#Maged Nada
In my opinion variables should be taken into account not only by year but also by countries
Hi Maged,
You can do in this way in Stata
1. Keep two separate data files for stock data and holidays.
2. Specify in a same way date for both datasets
3. Replace holidays = 0
4. Sort by date (sort date) both datasets and then merge two datasets by date: merge m:1 date using "holidays" if you use stock data as a master file, because you have multiple stock indices and the date will be repeated in the data.
5. Replace all missing values for holiday variable = 1. So, you will have 0 - holidays, and the rest - 1.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression