I am using the Hull-White model (extended version of Vasicek) to predict Canadian zero-coupon bond yield curves. Most of the time the model does a pretty good job of fitting the real curve in terms of shape. Here, we can see some of the predictions of the models (15 days after the calibration window):
In green circles, I have annotated how accurately the model can fit the shape of the real curves even after 15 steps.
However, a frequent problem can be observed in the predictions. Sounds like the model does not do a perfect job in predicting flat curves and adds an overshoot (to up or down) on the short end of the curves. In the following plots you can see some instances (Blue lines are predictions):
Do you have any idea why this happens and what can I do to fix it?