An extended regularized Kalman filter based on Genetic Algorithm: Application to dynamic asset pricing models

This study presents an extended version of the regularized Kalman filter for the application of dynamic asset pricing models, which deploy the Genetic Algorithm to solve a convex optimization problem involving both ℓ1− and ℓ2− based regularization terms in the correction equation. Our approach, namely GA-ErgKF, is firstly verified on synthetic data and then employed in a concrete financial case dealing with U.S. industry portfolios’ analysis over the period from July 1963 to December 2018. The results indicate that the GA-ErgKF algorithm is capable of tracking the unobserved time-varying parameters, even under the mixture noise pattern that contains both the Gaussian noise and sparse noise. Besides, this approach delivers superior performance over the alternative methods, thereby leading to better levels of out-of-sample explanatory power and hedging performance, especially when the asset returns fluctuate wildly. We further explore the microstructures of U.S. industries in the circumstance of a dynamic six-factor model calculated by the GA-ErgKF algorithm. It turns out that the microstructures differ noticeably between the volatile and steady industries, in terms of the density (also distribution) of the sparse noise, the volatilities of the abnormal returns and risk exposures, and the impact of risk factors. In addition, our approach captures fluctuations of the estimations under the economic and financial distress and identifies long-term stable profitability and investment patterns in the Food Products and Consumer Goods sectors.