Affine and quadratic models with many factors and few parameters

摘要

'Classic' affine and quadratic term structure models in the literature usually have three or four factors and tens of parameters. However affine and quadratic term structure models with many factors and few parameters (MFFP), i.e. with up to twenty factors and with six to seven parameters, fit and predict U.S. and Euro sovereign yields better than 'classic' affine and quadratic models. MFFP models also fit the volatility of and the correlations between changes in yields of different maturities better than 'classic' models. MFFP models outperform because fewer parameters reduce in sample over-fitting and because more factors give models more flexibility to match yields of different maturities. Among MFFP models, a type of affine model with stochastic volatility is usually preferable to the homoschedastic affine model, but for U.S. yields the quadratic model seems preferable among five factor MFFP models.