As I said in your other thread where you asked this question, there is no such thing as an "estimated coefficient of each observation".
Observations don't have coefficients, they have predicted values. Is that what you want? Or do you want the model coefficients (which do not change from observation to observation)?