There are three components to this project: Theory; Empirical Testing; and Dissemination. All components are linked to the current policy question of how taxes influence debt and systemic risk, and all use novel dynamic structural models. I am unique in explicitly linking such models to empirical testing.Theory: “Learning, Capital Structure and Systemic Risk.” Standard dynamic structural models of financing assume firms know the stochastic process governing cash flow. I will first consider a partial equilibrium model. Here firms are exposed to rare event risk, with the true probability being unknown. Firms learn and update beliefs regarding risk. Relative to standard models, firms are debt conservative and there is leverage persistence. In many cases, firms increase leverage only if they have avoided a negative shock long enough. In order to analyze asset pricing implications, I plan to embed such firms in a general equilibrium setting with a common catastrophic risk having unknown probability. Firms rationally respond to “Great Moderations” by increasing leverage. Recessions are more severe after long tranquil periods due to high debt overhang. A third paper, Re-Examining the Link Between Leverage and Systematic Risk, considers cross sectional asset pricing implications of credit shocks. The standard levered beta formula is erroneous, and the pre-tax cost of capital increases with debt. Together, the models show privately optimal debt is lower than recognized, and that tax breaks for debt reduce welfare.Empirical Testing: “Natural Experiment Policy Evaluation—A Structural Critique.” A common approach to testing whether taxes influence corporate financing and investment decisions is to compare leverage and investment before/after tax changes. I use a structural model as a laboratory to show that lack of a statistically significant change is not sufficient to reject the null that “taxes matter.” I will first consider an economy where the tax rate is a Markov process. Flotation costs on debt and real irreversibility limit the response of financing and investment to changes in shadow prices. More importantly, responses to tax changes are attenuated whenever they are partially anticipated and not permanent. Standard tests violate rational expectations by implicitly assuming tax changes come as surprises, with each new change being viewed as permanent, until the next surprise. My argument implies that standard tax experiments cannot falsify the null that taxes affect behaviour. Further, one cannot generalize elasticities if the policy transition matrix differs. I will propose an alternative Bayesian approach to hypothesis testing. My argument casts doubt on standard interpretations of historical evidence of tax change effects, suggesting true elasticities may be much higher. I will consider extending this argument to settings with endogenous policy choices.“Dissemination. The objective of this phase is to lower entry barriers by making the methodology accessible via a non-technical primer, and by making the models readily available using a user-friendly online platform.