Method for forecasting instable policy enforcement, is described, wherein a behavior dynamic Bayesian network (DBN) model and a policy finite state transducers extended with tautness functions and identities (TFFST) model is analytically composed to derive predictions of the consequences of enforcing a given policy, in particular to detect flip-flop configuration changes in a system. The method comprises the steps of—translating (1) the Bayesian network that holds the Behavior Model (BM) into a finite state transducers extended with tautness functions and identities (TFFST); —computing (2) the union of the Bayesian network (BM) and Policy Model (PM) finite state transducers extended with tautness functions and identities (TFFSTs); —composing (3) the finite state transducers extended with tautness functions and identities (TFFST) produced in the previous step with itself; and—detecting (4, 5, 6, 7) repetitions of events in the input and the output of every possible path; —if at least one repetition is found, detecting a possible instability (9).