Decomposing firm productivity has been challenging for some time. We propose a flexible functional form for total factor productivity that treats misspecification and endogeneity. The model also treats heteroscedasticity. Our productivity measure nests both the input distance function and output distance function. We provide details of a novel Local Likelihood estimation, a non-parametric technique, to estimate productivity which also has excellent finite-sample properties. In an empirical application, we measure bank productivity at the global level. Results show that productivity is correctly identified, and the flexibility of our methodology allows to estimate the impact of equity and nonperforming loans on bank productivity. Technology has positively contributed to productivity. However, nonperforming loans, bank risk-taking and raising capital have had the opposite effect.