Title: Automatic Discovery and Generation of Visual Product Characteristics: Application to Visual Conjoint
Authors: Vineet Kumar, Ankit Sisodia, Alex Burnap
Abstract: Visual design characteristics of products play an important role in consumer preferences for many categories. However, characterization of quantification of visual design is a challenging problem. We provide a method to automatically discover and quantify visual characteristics (attributes) from image data using a disentanglement-based approach. While the deep learn[1]ing literature has shown that supervision is required to obtain unique disentangled representations, ground truth visual characteristics are typically unknown in real world applications. Our method does not require such supervision, and instead uses readily available structured product characteristics as supervisory signals to enable disentanglement. No prior knowledge on design characteristics is required, yet we are able to discover human interpretable and statistically independent characteristics. We apply this method to automatically discover visual product characteristics of watches, and discover 6 human interpretable visual characteristics providing a disentangled representation. We conduct visual conjoint analysis to obtain consumer preferences over visual characteristics. Our generative method is also able to create novel visual designs that correspond to ideal points of different consumer segments.
Title: Fintech Failure in Emerging Market Retail: Evidence and Mitigation Approaches from an RCT in Mexico
Authors:
Abstract: Across emerging markets, physical cash is the ubiquitous means of transaction for small-scale retailers and the customers they serve. Policymakers, product manufacturers and Fintech solution providers are interested in understanding how to encourage the growth of digital payments in these markets . In this paper, we first show that policy interventions addressing only constraints related to fixed costs, logistical barriers, and benefits information are inadequate. We do so via a random audit of 104 traditional retailers who were beneficiaries of a large-scale “Technology-drop” program of the Ministry of Economy in Mexico intended to promote their digital payment acceptance. Next, in line with the theoretical literature on two-sided platforms (Rochet and Tirole, 2001), we posit that both supply-side technological frictions and demand-side frictions (for example, the perception that consumers prefer cash) account for the observed ‘Fintech Failure’. Subsequently, we design two novel marketing interventions in the B2B and B2C channel, targeting the relevant supply-side and demand-side frictions, respectively. We test the efficacy of both interventions via a rigorous randomized field experiment with 479 cash-only traditional retailers in Guadalajara, Mexico. We find that the B2B channel intervention reduces ‘Fintech Failure’ by 23 percentage points relative to the control group. This intervention raises the likelihood that the retailer has adequate banking infrastructure, functioning hardware and software to accept digital payments present in-store, and adequate employee knowledge to operate Fintech. Furthermore, the B2C channel intervention has an additional positive impact in reducing 'Fintech failure' compared to the B2B group. Using rich transaction-level data from our Fintech solution provider, we find evidence that in the 12 months following Fintech installation, they: i) are more likely to have a live customer transaction each month by 14 percentage points; ii) process approximately 77% more card transactions from customers; and iii) process an additional 37% in pesos in card transactions, relative to the B2B intervention group.
Title: Algorithm failures and consumers’ response: Evidence from Zillow
Authors:
Abstract: In November 2021, Zillow announced the closure of its iBuyer business. Popular media largely attributed this to a failure of its proprietary forecasting algorithm. We study the response of consumers to Zillow's iBuyer business closure. We show that after the iBuyer business closure, home sellers started making list-pricing decisions that deviated more from the Zestimate, Zillow's algorithmically generated estimate of a home's current value, suggesting that the iBuyer forecasting algorithm failure negatively affected consumer trust in the Zestimate algorithm. Moreover, sellers deviated more by increasing rather than decreasing their list price. We next look at the downstream consequences of the Zillow iBuyer closure on sales outcomes, such as sales price premium over the list price and time on the market. We find that properties are sold for more and in less time, both benefitting home sellers.
Title: Reducing Misinformation Sharing on Social Media Using Digital Ads
Authors:
Abstract: Interventions against misinformation have been a major focus in recent years. There is a particular need to develop “content-free” interventions that do not require specific fact-checks or warnings related to individual false claims, as the scale of content posted exceeds what can reasonably be evaluated in a timely manner. Evidence from survey experiments suggests that most social media users do not want to share inaccurate content but often forget to consider accuracy - and thus that shifting attention to accuracy can reduce misinformation sharing. In this talk, I will describe an investigation of whether this approach can be deployed at scale by serving social media users with digital ads that invoke accuracy. Specifically, I will report the results of two large-scale social media field experiments in which users were randomized to a control, or a treatment that received accuracy prompt ads, and the subsequent sharing is misinformation posts is compared.
Title: Does Amazon Have Pricing Power? Evidence from Pricing during the Covid-19 Pandemic
Authors: Sharmistha Sikdar and Vrinda Kadiyali, Prasad Vana
Abstract: As a leading online retailer, Amazon is facing antitrust scrutiny about its market power. Simultaneously, Walmart is rising as an online rival. We examine daily prices of 238 products on Amazon and Walmart.com before, during and after peak Covid cases (June 2020- December 2021) for evidence of Amazon’s pricing power; Walmart prices provide an important benchmark. We choose the Covid time period since this was a time of higher demand and supply variance, allowing cleaner identification of differences between Amazon and Walmart. We find that Amazon’s key prices- the Buy Box price on which majority of transactions occur- are systematically higher than Walmart’s, and a substantial portion (~49.5%) of these higher Buy Box prices are associated with Amazon’s third-party sellers rather than Amazon’s own seller role. However, despite these higher prices, customer ratings at Amazon are higher than at Walmart. Analysis of customer-level basket purchase data from Comscore shows Amazon customers have higher income, and hence likely higher willingness to pay. This evidence of higher prices, higher ratings and higher incomes is more consistent with market segmentation-based pricing power and does not support antitrust mitigation of market power based on adverse consumer welfare effect of Amazon too-high pricing. It is possible Amazon’s prices are too low for other rivals to compete effectively with lower prices, however consumer ratings do not support loss of consumer welfare. With its current strategy, Walmart appears to not successfully serve its lower-income customers. We discuss implications for regulators and managers.