상세정보
Export to Refworks부가기능
Three Essays in Quantitative Marketing
상세 프로파일
| 자료유형 | 학위논문 |
|---|---|
| 서명/저자사항 | Three Essays in Quantitative Marketing. |
| 개인저자 | Vidal-Berastain, Xavi. |
| 단체저자명 | University of Rochester. William E. Simon Graduate School of Business Administration. |
| 발행사항 | [S.l.]: University of Rochester., 2019. |
| 발행사항 | Ann Arbor: ProQuest Dissertations & Theses, 2019. |
| 형태사항 | 162 p. |
| 기본자료 저록 | Dissertations Abstracts International 81-03B. Dissertation Abstract International |
| ISBN | 9781085750325 |
| 학위논문주기 | Thesis (Ph.D.)--University of Rochester, 2019. |
| 일반주기 |
Source: Dissertations Abstracts International, Volume: 81-03, Section: B.
Advisor: Lovett, Mitchell |
| 이용제한사항 | This item must not be sold to any third party vendors.This item must not be added to any third party search indexes. |
| 요약 | This dissertation consists of three essays structured in three chapters. Chapter one studies the effect of social media has on consumer decisions. It is well-known that social media has become a major pillar of most brand strategies. We investigate the impact of social media posts on brand choices as well as how social media posts are generated in the context of political candidates (brands). We center our study at the early stages of the 2012 Republican Presidential primary race. This time period is sometimes referred to as the pre-race and collect daily positive, negative, and neutral posts on social media primary race as well as daily polls, news coverage, and candidate debate performance. We construct a model with consumers taking myopic decisions based a preference relation that is correlated across time. In doing so, we shed light on the underlying mechanisms linking the strategic generation of word of mouth media to voting decisions. Our results show that the tone of the social media around brands or candidates are relevant measures to account for. Specifically, for influencing voters' decisions, increasing the level of positive word of mouth around a candidate is more effective than a similar increase in the level of exposure in the mass media. We also show that the positive social media posts decrease more slowly than negative as the candidate's poll rank worsens. Finally, our results of the multiplicative attention model reveal important differences in the speed at which the general level of social media attention is translated into positive and negative social media. Chapters two and three switch gears to join the wave of big data that is rapidly expanding the set of questions that can be approached. Chapter two of this dissertation uses a very large panel of consumers to disentangle how consumers change the way they shop as more options become available. Grocery retailing has witnessed almost continuous disruption over the past several decades. While the traditional supermarket format was an important post-war innovation, it has gradually been refined and replaced by a diverse collection of sophisticated retail platforms, including super-centers, club stores, limited assortment stores, organic specialists, and a small, but growing online channel. Consumers now face a rich array of shopping options that differ by size, location, assortment, and price, alongside other less quantifiable dimensions such as ambiance and convenience. We exploit machine learning and data fusion techniques to create and combine a rich collection of data on consumer choice and store entry decisions into a unified longitudinal panel. We further employ frontier synthetic control methods (Abadie 010) to address the issues of selection and heterogeneous response inherent to this context, extending them to our dis-aggregated setting. Grocery retailing is surprisingly complicated, as it involves what is essentially an exercise in joint production between consumer and retailer that takes place over a sophisticated two-sided platform linking consumer product manufacturers (brands) to their downstream shopping base. The last chapter of this dissertation augments chapter two. Specifically, It presents a parallelization strategy that exploits recent improvements in general purpose GPU-computing to improve the performance of the synthetic controls estimator when applied to large and unbalanced data sets. When implemented on unbalanced panels, the synthetic controls estimator suffers from interpolation bias, over-fitting and does not have statistical properties that eases the task of performing causal hypothesis testing. Traditional fixes to these problems such as parameter-regularization (for over-fitting) or permutation tests (inference), helps to mitigate these problems but increase the computational cost exponentially. Chapter shows how to frame the quadratic programming program as a tensor optimizing problem. A tensor is a generalization of a matrix to allow for an arbitrary number of dimensions. We create a tensor containing all the data to optimize and to perform permutation tests and side-loaded into a GPU and run on a modified version of the mirror descent algorithm to perform thousands of optimization problems in parallel. I show the performance of my method and illustrate its power by studying how the entry of different retail platforms changes the way consumers purchase organic product. |
| 일반주제명 | Marketing. Artificial intelligence. Information science. |
| 언어 | 영어 |
| 바로가기 | 
: 이 자료의 원문은 한국교육학술정보원에서 제공합니다. |
