Serial Correlation in Panel Data
Posted: Fri Oct 10, 2014 2:46 pm
Hi,
I'm new to eviews and econometrics in general. I'm having problems eliminating serial correlation in a house price model for the City of Nottingham (average real price paid at relatively small spatial disaggregations).
I have 3 years' data (2001,2004,2007) and large N. (I can also stretch back to 1998 and 1995, but only on the dependent variable). My intention was to estimate a fixed/random effects model on this using a range of explanatory variables.
The problem is with the dependent variable.
ADF test gives a probability of 1 indicating a unit root. Correlograms at level and 1st differences give significant autocorrelations (and if I include 1998 and 1995, 2nd differences too).
I don't think the data is explosive (regressing y on y-1 c gives a coefficient of 0.74).
Do I need to detrend/deflate the data somehow? Including lags/AR terms in a model still produces residuals that are serially correlated (across all lags in a correlogram).
Can anyone help me correct this??
Thanks in advance!
I'm new to eviews and econometrics in general. I'm having problems eliminating serial correlation in a house price model for the City of Nottingham (average real price paid at relatively small spatial disaggregations).
I have 3 years' data (2001,2004,2007) and large N. (I can also stretch back to 1998 and 1995, but only on the dependent variable). My intention was to estimate a fixed/random effects model on this using a range of explanatory variables.
The problem is with the dependent variable.
ADF test gives a probability of 1 indicating a unit root. Correlograms at level and 1st differences give significant autocorrelations (and if I include 1998 and 1995, 2nd differences too).
I don't think the data is explosive (regressing y on y-1 c gives a coefficient of 0.74).
Do I need to detrend/deflate the data somehow? Including lags/AR terms in a model still produces residuals that are serially correlated (across all lags in a correlogram).
Can anyone help me correct this??
Thanks in advance!