Quantitative measurement of loyalty under principal-agent relationship

This research explores the situation when a firm uses agents to sell its products. The empirical setting is financial firms selling annuity products. Developing a measurement of agent loyalty helps minimize the moral hazard and adverse selection problem.; The study's contribution is in taking a modest empirical look at measures that can be used when the firm has information about product sales, but less information about the agent. I developed an integrated approach extending frameworks from frailty survival analysis, quality control and machine learning that are new to marketing research.; My approach to the agency problem differs fundamentally from the existing literature. Previous research focuses exclusively on the normative aspects of the agency relationship: how to structure the contractual relation, including compensation incentives, between the principal and agent to provide appropriate incentives for the agent to make choices that will maximize the principal's welfare given uncertainty and imperfect information. I focus on the agent's performance measure through the statistical models and I develop measures of loyalty measurement.; Three characteristics are examined here. In the non-linear principal-agent structure, the agent sells multiple principals' products to customers. The risk factor is shared among customers buying products from the same agent. This model is similar to the biological survival problem in the same family. That common risk sharing, the frailty, is assumed to have an effect on the product retention time. The research first examines the common frailty characteristic and shows strong statistical evidence of the frailty effect on the survival time. Moreover, the analysis shows that policies sold by the same agent have dependencies among their survival times.; Next, the analysis assesses the agent's steadiness of performance. The steadiness is the positive indicator of sustainability for any domain. Investigating the structure of steadiness gave a successful loyalty analysis.; Finally, the third analysis uses probabilistic Markovian state transitions to describe the agent's observable performance: a Hidden Markov Model. Although this model does not show a desirable fit, the research has high potential.; Thus this dissertation successfully measures the performance of the agent to allow the principal to monitor behavior.