The talk will be delivered by University of Florida faculty Prof. Bikram Karmakar
13 July, 2022, Bengaluru: The Office of Research and Publications (R&P) at IIM Bangalore will host a seminar titled: ‘Understanding Early Adoption of Hybrid Cars via a New Multinomial Probit Model with Multiple Network Weights’, on July 18 (Monday), 2022, at 2:30 pm, in Classroom P12. The talk will be delivered by Prof. Bikram Karmakar, University of Florida, from the Decision Sciences (DS) area.
The seminar will be in-person, and all standard Covid protocol will be followed.
About the talk: Modeling demand for durable products such as cars is challenging as there are no repeated purchases for most customers. One way to try and overcome this data scarcity is to pool information across similar customers. The researchers implement such a pooling strategy by proposing a new multinomial probit model that simultaneously accommodates different network structures among customers by connecting them through multiple weighted networks. Unlike the traditional multinomial spatial probit, the research model links consumer connectedness to their preference and marketing mix coefficients so that each subset of the parameter vector is correlated in a unique way. The research proposes and implements a novel Monte-Carlo Expectation-Maximization (MCEM) based approach to parameter estimation that significantly increases the number of consumers and choice alternatives that the model can handle. The method modifies the computationally expensive E-step in the classical EM algorithm by a fast Gibbs sampling based evaluation. Further, it implements the M-step using a fast back-fitting method that iteratively fits weighted regressions based on associated similarity matrices for each subset of the coefficients. The research establishes the convergence properties of the proposed MCEM algorithm, presents computational perspectives on the scalability of the proposed method, and provides a distributed computing-based implementation.
The research shows that a multinomial probit model based on two different similarity structures can significantly improve the prediction of customer choices. Specifically, the best fitting model includes spatially contiguous weight structures on the intercepts based on the similarity between the consumers’ previously owned vehicles, while cross-customer correlated coefficients are based on the geographical distance between consumers. The research demonstrates how an automobile manufacturer can leverage the estimated heterogeneous spatial contiguity effects to develop more effective targeted promotions to accelerate the consumer adoption of a hybrid car.
About the speaker: Dr. Bikram Karmakar is an Assistant Professor in the Statistics Department at the University of Florida, USA. He received his PhD in Statistics from the Wharton School, University of Pennsylvania in 2019, with Dylan S. Small as thesis advisor. Before that, he completed his Bachelor’s and Master’s education from Indian Statistical Institute. His research focuses on causal inference, design and analysis of observational studies, and applying statistics in social sciences, public policy, health and marketing. His research is supported by research grants from the National Science Foundation and the National Institute of Health, USA.
The talk will be delivered by University of Florida faculty Prof. Bikram Karmakar
13 July, 2022, Bengaluru: The Office of Research and Publications (R&P) at IIM Bangalore will host a seminar titled: ‘Understanding Early Adoption of Hybrid Cars via a New Multinomial Probit Model with Multiple Network Weights’, on July 18 (Monday), 2022, at 2:30 pm, in Classroom P12. The talk will be delivered by Prof. Bikram Karmakar, University of Florida, from the Decision Sciences (DS) area.
The seminar will be in-person, and all standard Covid protocol will be followed.
About the talk: Modeling demand for durable products such as cars is challenging as there are no repeated purchases for most customers. One way to try and overcome this data scarcity is to pool information across similar customers. The researchers implement such a pooling strategy by proposing a new multinomial probit model that simultaneously accommodates different network structures among customers by connecting them through multiple weighted networks. Unlike the traditional multinomial spatial probit, the research model links consumer connectedness to their preference and marketing mix coefficients so that each subset of the parameter vector is correlated in a unique way. The research proposes and implements a novel Monte-Carlo Expectation-Maximization (MCEM) based approach to parameter estimation that significantly increases the number of consumers and choice alternatives that the model can handle. The method modifies the computationally expensive E-step in the classical EM algorithm by a fast Gibbs sampling based evaluation. Further, it implements the M-step using a fast back-fitting method that iteratively fits weighted regressions based on associated similarity matrices for each subset of the coefficients. The research establishes the convergence properties of the proposed MCEM algorithm, presents computational perspectives on the scalability of the proposed method, and provides a distributed computing-based implementation.
The research shows that a multinomial probit model based on two different similarity structures can significantly improve the prediction of customer choices. Specifically, the best fitting model includes spatially contiguous weight structures on the intercepts based on the similarity between the consumers’ previously owned vehicles, while cross-customer correlated coefficients are based on the geographical distance between consumers. The research demonstrates how an automobile manufacturer can leverage the estimated heterogeneous spatial contiguity effects to develop more effective targeted promotions to accelerate the consumer adoption of a hybrid car.
About the speaker: Dr. Bikram Karmakar is an Assistant Professor in the Statistics Department at the University of Florida, USA. He received his PhD in Statistics from the Wharton School, University of Pennsylvania in 2019, with Dylan S. Small as thesis advisor. Before that, he completed his Bachelor’s and Master’s education from Indian Statistical Institute. His research focuses on causal inference, design and analysis of observational studies, and applying statistics in social sciences, public policy, health and marketing. His research is supported by research grants from the National Science Foundation and the National Institute of Health, USA.