Industrial Organization
Applied Econometrics
Microeconomic Theory

Industrial Organization
Econometrics
Applied Microeconomics
Microeconomic Theory

May 1998 (Oral) Industrial Organization, Econometrics
May 1997 (Written) Microeconomic Theory and Macroeconomic Theory

Demand for Soft Drinks: Characteristics, Corners and Continuous Choice

Professor Steven Berry
Professor John Rust
Professor Donald Andrews
Professor Donald Brown

Summer 2001

M. Phil., Yale University, May 1999
M. A., Yale University, May 1998
Master in Economics, University of Hong Kong, May 1994
Bachelor in Economics, University of Hong Kong, May 1992

Yale University Fellowship, 1996-2000
Yale Dissertation Fellowship, Spring 2001

Teaching Fellow, Graduate Microeconomic Theory, Yale, Fall 2000
Teaching Assistant, Introductory Microeconomics, Yale, Spring 2000
Teaching Assistant, Mathematical Economics I, Yale, Fall 1999
Teaching Assistant, Introductory Microeconomics, Yale, Spring 1999
Teaching Assistant, Mathematical Economics II, Yale, Fall 1998

• "Demand for Soft Drinks: A Discrete Choice Model," manuscript, Yale University, 1999.
• "Identification in Continuous Demand Models," manuscript, Yale University, 1999.
• "Estimating the Demand for Soft Drinks: A Continuous Choice, Differentiated Products Model with Corner Solutions," manuscript, Yale University, 2000.
• "Estimating the Demand for Soft Drinks: A Dynamic Continuous Choice Model," work in progress.
• "Art for Art’s Sake? Marketing Decisions at a Non-Profit Arts Organization" (with Christopher Makler), work in progress.

In this research I consider demand models where (1) consumers care about product characteristics, (2) consumers can choose multiple brands and a continuous quantity of products, and (3) corner solutions exist, that is, consumption demand for some products is zero. These models are empirically applied to purchasing decisions of soft drinks.

I propose to use a continuous choice model (CCM) embedded in a hedonic approach. This model has the advantage of incorporating complicated relationships among decision variables without resorting to restrictive assumptions such as those in multi-stage decision-making approaches. However, the existence of corner solutions gives rise to a typical censored data problem. When the number of alternatives is large, conventional approaches used to solve the problem are impractical because of computational complexity. To solve the censored data problem when numerous alternatives are available, the simulated method of moments (SMM) developed by McFadden (1989) and Pakes and Pollard (1989) is used. I show that under general conditions the SMM estimators in this model are consistent and asymptotically normal.

I apply this methodology to estimate the demand for soft drinks using micro level data. The CCM allows each consumer to have different degrees of substitutability, and even complementarity, among different brands, depending on the characteristics of each product. By estimating parameters in the utility function of individual consumers, I am able to generate the market demand for each brand. Then I compute numerically the corresponding own-price and cross-price elasticities of demand and also the effects of marketing activities. The results show that consumers are variety seeking over product characteristics such as flavor, packaging, and container size, and yield complementarity when two brands are very different in product characteristics, a situation which cannot be explained in discrete brand choice models.

The second part of the dissertation extends this methodology to dynamic models. Consumers generally have an expectation of the price of each brand, and tend to buy more for a longer period of consumption when there is a price cut. Ignoring this inventory-holding behavior would result in overestimating the price elasticity of demand. One problem for empirical research is that, though economists observe quantity purchased and length of time between purchases, endogenous variables of inventory holdings are latent. I construct a dynamic CCM in which inventory holdings are functions of prices and other observable variables. This necessarily implies the need to simulate the whole path of decision variables to form the dynamic moment condition in order to use the SMM. When the cost of inventory holdings is positive, optimal decisions in an infinite horizon model are equivalent to optimal decisions in some finite horizon models. It turns out that a finite number of successive policy iterations will obtain the exact optimal control variables, not just approximations. Finally I estimate the demand for each brand, and compare the results with static models.