|
Guest
|
Hello,
I am working with a possion regression model as follows:
- Data: y_1,....,y_100
- x_i ~~ log (y_i) %% ~~ = approximatelly
- x_i = mu + phi*(x_(i-1) - mu) + eps %% phi is known
- y_i ~ Poiss ( psi ) %% psi=exp(x_i)
So, x_i is AR(1) and im finding mu by MLE. My problem is that the data contains 0's, hence I cannot take the log. However, Im using the follow approx for x_i ~~ log(y_i + C), where C=0.00001, for example. The result for mu seems to dont vary that much for small C values, but Im wondering if someone can suggest some other trick to overcome the 0's in the data.
Thanks for any suggestion! |
|
|
