Asset Pricing with Regime-Dependent Preferences and Learning