I don't see why not - there are papers which use system identification for modelling stocks.

Maybe if you apply it at a microstructural scale (think seconds) with order book and trade print data from many sources/symbols. Certainly not if you blindly apply it to the time series of the price.