Three Essays on Financial Econometrics and Patent

The dissertation contains three chapters. First two chapters are about financial econometrics. The last chapter is on patents.;Chapter 1: The nonparametric theoretical work of Li et al. (2013) shows that the realized volatility estimator is asymptotically biased under the assumption that observation times are endogenous. Motivated by this finding, we develop a model for irregularly spaced returns where durations, the time span between two consecutive transactions, are endogenous. The model contains a bivariate Ornstein-Uhlenbeck (OU) process that jointly models equity latent volatility and trading intensity. Together with two other processes modeling trading prices and durations, the time endogeneity is captured by our model. The mode has a linear state space representation. We obtain an asymptotically unbiased volatility estimator via the Kalman filter. Estimates from Microsoft (MSFT) high-frequency trading data reveal a positive time endogenous effect between durations and logarithmic prices.;Chapter 2: Implied volatility derived from the Black-Scholes model often performs poorly in practice due to the assumptions of underlying asset volatility being constant over the option lifetime and normality of returns. To circumvent this issue, Dumas, Fleming, and Whaley (1998) developed a Deterministic Volatility Function (DVF) model. They found that by incorporating implied volatility with option moneyness and maturity significantly reduces option pricing errors. On the other hand, in Chapter 1 we developed the so-called endogenous time volatility estimator based on high-frequency equity trading data. Inspired by their work, we investigate the benefits of integrating the underlying asset endogenous time volatility to the DVF framework. By doing so, we find that DVF with endogenous time volatility factor helps a little regarding in-sample fit. However, the original DVF model still beats our proposed model in terms of out-of-sample forecast performance.;Chapter 3: Decision makers aiming to promote innovation often care about the connections between industrial innovation and economic development. Analysis of industry-level patent statistics can shed light on these connections. Concordance systems for assigning patents to industries are necessary for such analysis. This paper uses patent data matched at the firm level to investigate the accuracy and precision of prevailing technology-industry concordances and proposes new concordances. In contrast to previous concordances, which link patents to specific industries using textual analysis of patent titles and abstracts and official industry descriptions, we use firm-level 'microdata' from the OECD to match patents to industries via the identified economic sectors of patent applicants. Using this applicant sector matching, we examine previous concordances via binary probability regression models. We propose two new concordances using applicant sector matching, namely the Raw OECD Concordance (ROC) and the OECD Bayesian Averaged Concordance (OBAC). The ROC concordance is based on simple conditional frequencies, whereas the OBAC uses Bayesian model averaging to combine multiple regression models using technology classifications to predict industry membership. We find that text-based concordances poorly predict applicant sector matches, and that our OBAC concentrates more of its probability mass among the most likely industries, as compared to ROC. This research therefore yields a new concordance for policy analysis, as well as introduces a new method (BMA) that can improve existing concordances.