This paper examines the effects of innovation on wage dynamics across German manufacturing sectors. We combine matched employer-employee data from the German Institute for Employment Research (IAB) with patent information from PatStat to estimate the causal impact of innovation at the local-industry level. To address endogeneity concerns, we construct an instrumental variable based on international patent citation networks. Our results show that innovation increases wage growth, with stronger effects for workers employed in abstract occupations. These findings are consistent with a complementarity between innovation and abstract tasks. We then examine the role of labor market institutions, focusing on Collective Bargaining Agreements (CBAs), in mediating the effects of innovation on labor market dynamics. We find that firm-level bargaining strengthens the transmission of innovation gains to routine and manual workers, whereas industry-level agreements are less responsive to firm-specific innovation shocks. We present a simple model to explain and support our results. Our findings highlight the central role played by labor market institutions in shaping the distribution of innovation premia.

At the foundation of economic complexity lie two classes of indicators: economic complexity indices and relatedness density indicators, both of which proved useful for predicting growth and diversification. However, whether a causal interpretation could be given to these relationships has been largely overlooked. In this paper, we claim that recent advances in the econometrics of shift-share variables provide a natural avenue to assess the endogeneity of regression designs typically used in the field of economic complexity. We derive precise casual inference conditions for complexity and relatedness indicators and show how to test them. Additionally, we discuss how to proceed in case these requirements are not satisfied, providing the reader with a practical guide to assess causality in economic complexity regressions. Our general conclusion is that typical regression designs in economic complexity and relatedness density are endogenous. However, we show that, at least for the case of relatedness density, it is possible to take relatively simple correction steps, which allow for a causal interpretation of the estimated coefficients.