Abstract
This dissertation studies pricing dynamics using evidence from the US brewing industry and from the Amazon marketplace. In the first chapter, I analyze the relationship between market structure and inter-temporal price discounts in the U.S. brewing industry. Most studies assume that consumers face constant product prices within a month or a quarter. However, consumers can respond to price discounts and strategically adjust their shopping behavior. Firms exploit consumers' responses to temporary price discounts to inter-temporally price discriminate across consumers. A change of market structure may affect firms' price-discount strategies. I use the case of 2008 Miller/Coors joint venture to investigate how the change in market structure affects the dynamics of price-discount strategies of firms and quantify its welfare effects. I begin by documenting an empirical pattern that competing firms provide simultaneous promotions at stores in the pre-merger periods, while the merged firm alternates promotions after the merger. I then use autoregressive regressions to verify this empirical pattern statistically. To quantify the welfare effects of the change in price-discount strategies, I develop a structural model to characterize heterogeneous demand functions of consumers who stockpile (storers) and consumers who lack storage capacities (non-storers). I infer that a substantial number of consumers stockpile at the promotional prices. The percentages of sales to storers differ by brands and range from 13 percent to 26 percent. Storers are more price-sensitive and more likely to switch between brands. On the supply side, I model firms’ price-discount strategies using a two-stage game: in the first stage, firms consider whether to use an inter-temporal price discrimination strategy; in the second stage, firms simultaneously determine the product prices. If competitors used constant price strategies, firms can increase their profits by at least 8 percent when switching to an inter-temporal price discrimination strategy. Likewise, if competitors used inter-temporal price discrimination strategies, firms can increase their profits by at least 6 percent when switching to inter-temporal price discrimination. In equilibrium, firms, therefore, choose inter-temporal price discrimination. After the market-structure change, the merged firm (with two close-substitute products) can and does increase its profit by 9 percent by staggering products on sale. I simulate the post-merger product prices and determine the difference of welfare effects with/without considering the promotion-strategy adjustment. Static models of competition ignore this effect which leads to a substantial under-estimation of the welfare impact of market mergers.
The second chapter, which is joint work with Denis Nekipelov, studies the pricing dynamics of algorithmic agents in the Amazon marketplace. Availability of algorithmic tools allowed many small retailers manage multi-product inventory and dynamically price their products in online marketplaces.
At the same time, when the price is determined by an automated algorithm rather than retailer's own decision, the link between the product's marginal cost and the price traditionally studied in Industrial Organization is lost. The information regarding the marginal cost is implicitly communicated through the automated price updates generated by the automated algorithm. In this chapter, we use the ideas from the online learning literature in Computer Science to restore the link between the observed price changes and the marginal costs of retailers. The methodology developed in Nekipelov et al. (2015) uses the notion of regret to evaluate the automated algorithm. Regret measures the relative performance of the algorithmic dynamic strategy relative to the benchmark which corresponds to the best-fixed price in hindsight. This idea allows us to recover the identified set that contains the retailer's marginal cost as well as the expected regret of her automated price strategy. We apply this methodology to study dynamic pricing on Amazon's marketplace. We find that expected regret for most retailers is close to zero. As a result, despite the simplicity of their algorithmic tools, they have good dynamic performance. At the same time, the estimated markups of retailers imply demand elasticities that are compatible with traditional retail markets. This may indicate that online marketplaces where small retailers use algorithmic tools may have good performance while achieving similar outcomes for consumers as the traditional retail.
