SVAR and Stability

[chssb10]

Hi,

I am currently working on a SVAR model to study the impact of monetary policy on asset prices in China.

I have read a paper titled "Inference in Linear Time Series Models with some Unit Roots" by Sims, Stock and Watson, published in Econometrica in 1990. This paper recommend against first differencing even if the variables in the model are non-stationary.

I've followed their idea in estimating the SVAR model in level, but the SVAR model is not stable in the sense that at least one root of the characteristic polynomial lie outside the unit circle.

So, one thing that I would like to ask is that, if I estimate the model in level, is it necessary for me to have all the roots of the characteristic polynomial lie inside the unit circle?

Thanks!