Forecasting with LDV, AR(1) and Seasonal Dummies

[canboardrp]

I've built a model that has really strong explanatory power, a good fit and very reasonable (and significant) coefficients. All of this has been developed using pooled least squares.

Included among the explanatory variables are an AR(1) term, 3 quarterly dummies, the constant and a lagged dependent variable which covered the same quarter from the previous year.

When I forecast off of the LS results, I get a very good fit historically. However, when I forecast into the future the model becomes really wonky, with the seasonal changes really seeming to be the cause. I've attached an example chart to show what I mean.

I'm wondering what might be causing this and how might I correct for this?

Example

cbrp.jpg (24.17 KiB) Viewed 2152 times