Using a local adaptive Forward Intensities Approach (FIA) we investigate multiperiod corporate defaults and other delisting schemes. The proposed approach is fully datadriven and is based on local adaptive estimation and the selection of optimal estimation windows. Time-dependent model parameters are derived by a sequential testing procedure that yields adapted predictions at every time point. Applying the proposed method to monthly data on 2000 U.S. public rms over a sample period from 1991 to 2011, we estimate default probabilities over various prediction horizons. The prediction performance is evaluated against the global FIA that employs all past observations. For the six months prediction horizon, the local adaptive FIA performs with the same accuracy as the benchmark. The default prediction power is improved for the longer horizon (one to three years). Our local adaptive method can be applied to any other speci cations of forward intensities.
Accuracy ratio Forward default intensity Local adaptive Mutiperiod prediction