Bogotá - Ciencias - Maestría en Ciencias - Estadística · 2021
Estimación de cadenas de Markov en tiempo-continuo con observaciones discretas: aplicado a la probabilidad de transición de calificación crediticia
In the present work, the estimation of the credit rating transition probabilities for an arbitrary period of time is carried out through the implementation of a genetic algorithm, assuming that the credit rating transitions follow a Markov process. A continuous-time Markov chain is estimated using discrete data from the maximum likelihood function based on the absolute frequency matrix. The methodology is applied in two cases, the first for global corporate bonds, and the second for Colombian corporate bonds. The results of the application of the developed methodology are evaluated by comparing the estimate of the one-year transition probability matrix obtained by the genetic algorithm with the hope-maximization methods and the quasi-optimization algorithm. It was evidenced that the results obtained through the application of the genetic algorithm proposed in this work are computationally efficient and competitive compared to the methods with which it was compared, likewise, the resulting values in the estimated probability of non-compliance were practically the same, for all three methods.