My dissertation consists of three chapters where the common theme is the use of various econometric techniques to constructing optimal portfolios and empirically exploring portfolio performance. My first chapter proposes new approaches to constructing mimicking portfolios for non-tradable shocks from a large set of base assets. Although the analytical solution to mimicking portfolios is available, it involves estimating the covariance of asset returns and covariance of returns with replicated shocks. The estimates of those moments are noisy and can have a detrimental impact on the quality of the portfolio performance out of sample, especially in the presence of a large number of base assets. To mitigate the overfitting problem, this chapter imposes regularization constraints on portfolio strategies. The first proposed approach solves the portfolio variance minimization problem with target portfolio betas and a constraint on the upper limit on the norm of portfolio weights. The second approach recasts the mimicking portfolio problem to a GMM estimation problem, where the portfolio weights are the estimated parameters, along with constraints on the norm of the portfolio weights. Compared with the first approach, the second approach does not estimate the portfolio problem inputs explicitly and may not satisfy the beta constraints in-sample, leading to additional flexibility to better perform out-of-sample. This chapter uses simulations to study the comparative advantage of the two approaches, applies the proposed techniques to construct mimicking portfolios for nine macroeconomic and uncertainty shocks, and examines the empirical out-of-sample portfolio performance. In all cases, the regularized mimicking portfolios have feasible portfolio weights without sacrificing the portfolio performance. In my second chapter, I propose a new methodology to form conditional mimicking portfolios. Prior research forms mimicking portfolios mostly based on past realizations of returns and shocks without utilizing conditioning information, which would capture the time variation in statistical moments of returns and shocks. My conditional mimicking portfolios track the target shocks period by period, efficiently use available information about future shocks and returns, and have a minimal conditional variance of returns. The key innovation of my methodology is that I transform the traditional conditional portfolio minimization problem into a set of conditional moment restrictions and apply the optimal Generalized Method of Moments (GMM) estimator to find portfolio weights. To obtain the optimal GMM estimator, I build upon the classical GMM framework and construct the optimal instruments, which are non-parametrically functions of asset characteristics and macroeconomic variables. Compared with the traditional approach, my methodology neither imposes assumptions on the dynamics of conditional returns and shocks nor struggles to identify unobservable investors' information sets. To exploit the finite sample property of conditional mimicking portfolios, I use simulations and apply my methodology to create portfolios that mimic six macroeconomic and uncertainty shocks. Results show that the use of conditioning information helps improve the out-of-sample portfolio performance and also highlight the challenges of forming conditional mimicking portfolios. In my third chapter, I study portfolio management, with a focus on mutual fund performance. Prior research primarily examines the performance of equity mutual funds, leaving bond mutual funds relatively understudied. In this chapter, I construct a holding-based measure, weight shift, to capture the intensity of bond mutual funds' trading activity. The traditional measure of funds' trading activity - portfolio turnover - has several limitations, such as infrequent reporting, sensitivity to large changes in average total net assets, and susceptibility to the amount of one-sided aggregate transactions. I investigate the relationship between funds' trading activity and future fund performance, and find that higher fund trading activity predicts lower fund performance controlling for fund-specific characteristics. The negative impact of trading activity on fund performance is stronger among high-yield funds, in line with the notion that more noise trading incurred by high trading in illiquid markets erodes fund performance. I also examine the incentive for managers to engage in intensive trading activity, and find that managers use high trading intensity as a signal to attract investors, especially retail investors. Finally, I find that weight shift captures the inferior managerial skill and reacts to macro uncertainty. This chapter sheds light on understanding the trading behavior of bond mutual funds and questions the value of active management in bond mutual funds.