Estimating univariate ARIMA/ARMA using non-stationary series

[csolleza]

Hello,

I'm trying to estimate an ARIMA/ARMA using a single time series data. Unit root test (ADF) suggests that series is non-stationary. Differencing it once makes the series stationary. A look at the correlogram suggests that lag 1 will remedy autocorrelation. When I enter the command "equation eq.1ls d(series) c ar(1) ma(1)" the resulting equation's R^2 is 0.11. But if I use the non-difference series "equation eq.1ls (series) c ar(1) ma(1)" the R^2 is 0.99. What could I have done/misunderstood wrongly? Greatful for any tips!


C

[startz]

R^2 measures the fraction of the variance explained of the dependent variable. It's much easier to get a prediction of the level of the variable than a prediction of the change.