This dissertation comprises three independent essays in the domain of macroeconomic development and applied macroeconomics. In the first essay, I study India’s state-led heavy industry drive from the 1950s. Influenced by Soviet industrialization and the post-World War II emphasis on state-led development, India sought to shift production towards capital goods sectors. To what extent did the Nehru-Mahalanobis focus on heavy industries during India’s Second Five-Year Plan (1956–61) foster their growth? To answer this question, I compile a unique dataset tracking large-scale industries from 1951–1965, roughly spanning the first three Five-Year Plans. This dataset harmonizes industrial production, prices, and input-output data from pre-digital data books. Previous studies showed positive effects, but my analysis, which adjusts for an industrial classification change using detailed annual data, does not find a statistically significant impact of the policy on the development of targeted heavy industries. I discuss this episode within the context of development theory from the 1950s and 1960s. I find no learning-by-doing effects emanating from this heavy industry push to targeted industries, highlighting the importance of the quality of state intervention. Weak production linkages between targeted and non-targeted sectors restricted spillover opportunities for broader industrial development. In the second essay, I explore the challenges of using linear estimation for endogenous business cycles when the underlying data generating process is nonlinear. Recent literature highlights that in a univariate context, linear models, due to omitted variable bias, fail to identify the local instability present in a correctly specified nonlinear model. This study extends these findings to a multivariate setting. Although linear models fail to track instability, they have been found to be effective in detecting cyclical mechanisms. Adding simple nonlinearities to a macroeconomic model in the multiplier-accelerator tradition, this study demonstrates that a linear model can predict locally stable, cyclical behavior when the underlying data generating process is unstable and does not exhibit any local oscillatory dynamics. In the third essay, co-authored with Deepankar Basu, we provide necessary and sufficient conditions for diagonalizability of any singular matrix using its rank and eigenvalues. Recent developments in the theory of production networks offer interesting applications and revival of input-output analysis. Some recent papers have studied the propagation of a temporary, negative shock through an input-output network. Such analyses of shock propagation rely on the eigendecomposition of relevant input-output matrices. Only diagonalizable matrices can be eigendecomposed. We apply our results to 5 historical (Germany, India, and Japan) and 670 contemporary IO matrices (15 annual IO matrices for 43 countries from the 2016 release of the World Input Output Database and 25 annual IO matrices for the U.S. economy) and find that some IO matrices are not diagonalizable.