Date of Award
Doctor of Philosophy (PhD)
In this dissertation, I study how the existence of consumer learning in a digital goods environment influences the profitability of various firm strategies. I develop a structural model of consumer’s learning-by-using. Adopting a Bayesian learning framework, the model describes how product experience influences willingness to pay, and also allows identification of key factors behind learning and quantification of the trade-offs that firms face. I estimate the model using a novel data set of videogame users’ play record. Using the estimated parameters, I first consider the optimal design of free trials. Digital goods providers often offer a trial version of their product in order to familiarize consumers with the product. The trial configurations considered herein include limiting duration of free usage (i.e. ``time-locked trial'') and limiting access to certain features (i.e. ``feature-limited trial''). I find that time-locked trials outperform feature-limited trials, and the revenue implication depends on the rate of demand depreciation during the trial period.
I then consider the optimal product unbundling strategy. As digital goods can be considered as a bundle of identical services to be consumed at different points in time, the firm can unbundle and sell each component separately over time. Offering the product in an unbundled manner allows consumers to adopt part of the product after the learning takes place, resulting in higher willingness to pay through the option value. I find that pay-per-use, an extreme form of product unbundling, outperforms traditional outright sale when there exists consumer learning, while it does not in the absence of learning. Hence the existence of consumer learning has a substantial impact on the firm’s optimal policy.
In addition to empirically studying the implications of consumer learning, I also examine an econometric problem of identifying state dependence in consumer utility. Identifying state dependence is challenging when there exists consumer heterogeneity unobservable to a researcher. I show that if consumers make two decisions at each decision occasion, one being a discrete choice from multiple alternatives and the other being a consumption intensity of the selected option, then we can nonparametrically separate state dependence and unobserved heterogeneity under mild conditions. Understanding conditions for nonparametric identification helps empirical modelers in choosing their modeling assumptions.
Sunada, Takeaki, "Essays On Consumer Learning And Its Impact On Firm Strategies" (2019). Publicly Accessible Penn Dissertations. 3269.