GAS Modeling with Actuarial and Financial Applications

Cristiano Fernandes (PUC-Rio), Cesar Neves (CPES-Funenseg/UERJ/SUSEP) and Henrique Helfer (PUC-Rio)

Abstract: The generalized autoregressive score models (GAS) of Creal, Koopman and Lucas (2013), also named dynamic conditional score models (DCS) by Harvey (2013), is a recently proposed framework for the development of time series models with time varying parameters. It is particularly useful for the development of feasible non Gaussian time series models, and as such has found many applications in Finance as it can be checked in the GAS website (http://www.gasmodel.com/gaspapers.htm). In this talk we give a brief overview of GAS models and present two applications. In the finance application we use GAS models together with time varying copulas to generate joint scenarios for capacity factors of several wind plants belonging to different places in Northeast Brazil. These scenarios are then used as input to raise the distribution of cash flows associated with a portfolio of contracts attached to these wind plants. Our actuarial application deals with the extension of the well-known model of Lee-Carter, a benchmark used in the actuarial industry for forecasting mortality rates. We propose five different GAS models (Poisson, binomial, negative binomial, Gaussian and beta) to estimate the Lee-Carter parameters and dynamically forecast the mortality rates for the U.S. male population.
References:
Harvey, A. (2013). Dynamic Models for Volatility and Heavy Tails: with Applications to Financial and Economic Time Series. Cambridge University Press.
Creal, D., Koopman, S.J. and A. Lucas (2013). Generalized autoregressive score models with applications. Journal of Applied Econometrics 28, 777–795.