Event Study Regression Results NOT Robust (effect of pollution on housing prices)
Hello!
I'm trying to run an event study regression on my data to find the correlation between pollution levels before & after a fire on housing prices in each zipcode, by month. Run across multiple zipcodes, 25 months total, 2 fires in the time period, t1=1 is treated by the fire in 2018-08-15, t2=1 is treated by the fire in 2018-11-15...(data attached)
I ran simple a regression without controls (ln price = alpha + beta * poll + epsilon) and then one controlling for treated and after dummy var (including event month) for both t1=1 & t2=1 (ln price = alpha + beta*poll + theta *after + delta * treated + epsilon )
Both seemed to have robust results (do files attached).
Without controls: Pooled beta (effect of poll on ln_price): 0.0027
With controls for t1: beta_poll = 0.0025, theta_after = 0.0690, delta_treated1 = -0.5472
With controls for t2: beta_poll = 0.0027, theta_after = 0.0762, delta_treated2 = 0.1533
MY MAIN QUESTION:
I'm having trouble running the data as an event study regression.
OG event study regression - event study regression NOV 6.2.25.do - (effect of pollution on housing prices from NOV fire) was not robust from p values.
The coefficients results are the closest to what I want to see though: pre fire very close to 0 effect, directly during/after fire a negative impact then a positive coefficient due to scarcity.
Then tried to lag the effect of pollution on housing prices by 2 months - event study regression LAGGED NOV 6.2.25 - and still the results are not robust. Also omits the 2 months before the fire event due to multicollinearity but I thought that shouldn't be happening?
Did the same lagged regression instead for august event and the same thing happens... non-robust results and omits the same months.
Any advice / tweaks would be appreciated!!
Thanks in advance!
I'm trying to run an event study regression on my data to find the correlation between pollution levels before & after a fire on housing prices in each zipcode, by month. Run across multiple zipcodes, 25 months total, 2 fires in the time period, t1=1 is treated by the fire in 2018-08-15, t2=1 is treated by the fire in 2018-11-15...(data attached)
I ran simple a regression without controls (ln price = alpha + beta * poll + epsilon) and then one controlling for treated and after dummy var (including event month) for both t1=1 & t2=1 (ln price = alpha + beta*poll + theta *after + delta * treated + epsilon )
Both seemed to have robust results (do files attached).
Without controls: Pooled beta (effect of poll on ln_price): 0.0027
With controls for t1: beta_poll = 0.0025, theta_after = 0.0690, delta_treated1 = -0.5472
With controls for t2: beta_poll = 0.0027, theta_after = 0.0762, delta_treated2 = 0.1533
MY MAIN QUESTION:
I'm having trouble running the data as an event study regression.
OG event study regression - event study regression NOV 6.2.25.do - (effect of pollution on housing prices from NOV fire) was not robust from p values.
The coefficients results are the closest to what I want to see though: pre fire very close to 0 effect, directly during/after fire a negative impact then a positive coefficient due to scarcity.
Then tried to lag the effect of pollution on housing prices by 2 months - event study regression LAGGED NOV 6.2.25 - and still the results are not robust. Also omits the 2 months before the fire event due to multicollinearity but I thought that shouldn't be happening?
Did the same lagged regression instead for august event and the same thing happens... non-robust results and omits the same months.
Any advice / tweaks would be appreciated!!
Thanks in advance!
Comment