This dissertation, consisting of 2 chapters, explores how institutional investors play an important role in financial markets. In Chapter 1, I first present a new asset pricing anomaly: a simple dividend-based currency strategy, which shorts a currency on the date its country's recent aggregate dividend payment by listed companies is large, exhibits a significant Sharpe ratio and alpha not explained by standard factors in the currency market. To understand this anomaly, I identify the significant price impact of predetermined dividend payments on exchange rates around payment dates. I propose a dividend repatriation channel where benchmark investors (ETFs and mutual funds) predictably repatriate a certain proportion of dividends received in local currency. I build a model in which heterogeneous financial intermediaries with limited risk-bearing capacity accommodate benchmark investors' currency demands stemming from dividend repatriation flows. In line with the model's implications, I find that the price impact of dividend flows on FX around the payment date is large when the intermediary capital ratio is low, CIP deviations are large, and FX implied volatilities are high. My findings have implications for currency-market elasticity, capital regulations, and FX regimes. In Chapter 2, I develop a machine learning procedure to estimate investors' demand system in high dimension, which accommodates a large universe of stock characteristics, including price-based characteristics (e.g., momentum, valuation ratio, etc.). I propose an identification strategy based on the inter-temporal structure of latent demand to address the endogeneity of price-based characteristics, in addition to the instrumental variables. Using the U.S. stock market data, I illustrate how we can use the estimated high-dimensional demand system to analyze the time variations in the importance of stock characteristics in investors' holdings, each stock characteristic's impact on cross-sectional stock returns, and identify which investors are significant for characteristic pricing.