An article entitled "Machine learning Vasicek model calibration with Gaussian processes" by Raquel Gaspar, assistant professor at ISEG, has been published in the Journal of Statistical Planning and Inference. The article was co-authored with João Beleza Sousa (Instituto Politécnico de Lisboa) and Manuel Esquível (FCT-Universidade Nova de Lisboa).
This article proposes the use of computer learning techniques - machine learning regressions - to estimate the parameters of Gaussian-type stochastic processes widely used in finance. The method is explained by calibrating Vasicek's stochastic interest rate model to the logarithm of zero-coupon bond prices. The proposed method has the advantage of making it possible to estimate the model's parameters directly in terms of the neutral risk measure, and these parameters can be used directly in the evaluation of derivative products, which is an advantage over alternative estimation methods, which only allow estimates to be made in terms of the objective measure.
