It would be horrendously easy to simply set the analysis extent to a larger size to extend to your desired boundary, but I would suggest that the results would be meaningless or misleading at best unless you know that such an interpolation would generate valid results beyond the bounds of the input data. Also, why have you picked IDW as the interpolator given the other options available to you. Perhaps there is a problem design issue that you could address before settling in on a means to fill in the gaps between stations.