The aim of this paper is to justify the use of threshold autoregressive models in
financial time series. A category of potential applications insolves the presence of fixed
costs of adjustment that leads to economic agents adjusting to deviations from
equilibrium only when the benefits from adjustment exceed the costs, this is the case of
the mispricing error.
The new idea is that the error process is not assumed, as usual, as an i.i.d. sequence,
but as a martingale difference sequence.
Finally a numerical analysis on the base of a real data set related to the Italian
derivative financial market has been performed.
