Full Program »
File
pdf 456KB |
Estimation of risk-neutral distributions is of major importance on security valuation, risk management and asset allocation. Its forecasting ability has often been proved to fail by the literature based on the results of Berkowitz test. But, can we really discard forecasting ability of option implied risk-neutral distributions? This paper compares the extraction of risk-neutral distributions from option prices using two parametric models: mixture of two lognormal distributions and generalized distributions of the second kind; and two non-parametric models: kernel regression and spline methods. Non-parametric techniques are limited within the range of the observed data, therefore we deal with the estimation of the tails by extrapolating outside the available range of data as well as by appending tails drawn from a generalized pareto distribution. In order to test whether the extracted risk-neutral distributions are good forecasters of future movements of the underlying block-bootstrap simulations are run, concluding that Berkowitz test assumptions do not hold, and so forecasting ability of such densities cannot be rejected, as the test would suggest, for any of the methods analyzed.
Author(s):
Antoni Vaello SebastiĆ
Universitat de les Illes Balears
Spain
Maria Magdalena Vich Llompart
Universitat de les Illes Balears
Spain