Lin, Zhongjian (2014-05). Semiparametric Methods and Applications. Doctoral Dissertation. Thesis uri icon

abstract

  • This dissertation studies semiparametric methods and their applications to economics and marketing problems. We propose a game theoretic model to analyze interactions among individuals in a social network with hierarchy. We find significant asymmetric peer effects among individuals in social networks. High status individuals deliver stronger peer effects on low status individuals than vice versa. Additionally, we investigate semiparametric panel data truncated regression models with fixed effects. We show the identification of our model with primitive assumption, establish the consistency and asymptotic normality of our proposed sieve estimator. We conclude that we can achieve ?n -convergent rate for parametric parameters. Besides theoretical semiparametric methods, we study the dynamic effectiveness of marketing mix variables and the competition among the pioneer and early followers in pharmaceutical industry. With two pharmaceutical categories data, we find dynamic effectiveness of advertising and detailing inputs. Pioneer firms and follower firms have different effectiveness of advertising and detailing inputs in different stages. Our out-of-sample analyses show that when the data is rich, the semiparametric model outperforms the parametric model while it is better to deploy parametric model when the sample size is small.
  • This dissertation studies semiparametric methods and their applications to
    economics and marketing problems. We propose a game theoretic model to analyze
    interactions among individuals in a social network with hierarchy. We find significant
    asymmetric peer effects among individuals in social networks. High status individuals
    deliver stronger peer effects on low status individuals than vice versa. Additionally, we
    investigate semiparametric panel data truncated regression models with fixed effects. We
    show the identification of our model with primitive assumption, establish the
    consistency and asymptotic normality of our proposed sieve estimator. We conclude that
    we can achieve ?n -convergent rate for parametric parameters. Besides theoretical
    semiparametric methods, we study the dynamic effectiveness of marketing mix variables
    and the competition among the pioneer and early followers in pharmaceutical industry.
    With two pharmaceutical categories data, we find dynamic effectiveness of advertising
    and detailing inputs. Pioneer firms and follower firms have different effectiveness of
    advertising and detailing inputs in different stages. Our out-of-sample analyses show that
    when the data is rich, the semiparametric model outperforms the parametric model while
    it is better to deploy parametric model when the sample size is small.

publication date

  • May 2014