This dissertation is concerned with econometric theory and its applications. More specifically, the specific research issue of interest is the classic topic of estimation and inference in econometric models where researchers wish to investigate causal relationships in the absence of a randomized control trial. To overcome the problem of confounding effects we propose extensions of the recent Granular Instrumental Variables (GIV) methodology. In chapter 1 we provide a high-level introduction and motivate the topic in greater detail. In chapter 2, we extend the GIV methodology by relaxing several strong assumptions imposed on the error term and factor loadings and we further allow for asymptotic regimes where both the cross-sectional dimension and time series dimension diverge jointly to infinity. Additionally, we also fully exploit the structure of the model and overidentify the parameters of interest. We illustrate our contributions with an empirical application to the global crude oil market. In the 3rd chapter, the GIV methodology is further developed to accommodate large dynamic panels with unit specific endogenous variates, which require unit-specific GIVs. We develop a split-panel jackknife (SPJ) GMM-PCA iterative procedure to estimate the structural parameters of interest. Overidentification tests can be carried out to test model validity. We illustrate the SPJ GMM-PCA procedure in two applications: (1) estimation of demand for new automobiles and (2) estimation of the determinants of banks' capital adequacy ratios.