【公开学术报告】Online Video Advertising: A Dynamic Bayesian Model of Wearout, Spillover, and Sales Impact across Website and Amazon Channels

Date & Time: 9:30-11:00am, Fri. 18^th, October 2019

Venue: Room 2101, Tongji Building A

Language: English

Speaker：Dr. Linyi LI (University of Utah, US)

ABSTRACT

Despite the growing popularity of online video advertising, very little is known about its performance, especially in comparison to search and (non-video) display. We examine the effectiveness of online video advertising in driving website and Amazon sales using data from a maker of personal care products which engaged in significant video advertising in addition to search and display advertising. Using a dynamic linear model estimated in a Bayesian framework with Hamiltonian Markov chain Monte Carlo (MCMC), we compare the effects of online video, search, and display advertising in terms of sales impact, wear out, and spill over onto other online advertising channels. We find that video advertising contributes more to the build-ups of goodwill than either search or display advertising, with views having a stronger effect than impressions. As a result, video advertising has a stronger long-run impact on sales than either search or display. Video advertising also exhibits substantially slower wear out than search or display and greater spill over to other advertising channels, increasing the effectiveness of subsequent search and display impressions as measured by click-through rate. Finally, we find that video ads drive substantial cross-sales-channel purchases. Search advertising produces clicks (and short-term sales) at a higher rate, but contributes less to the accumulation of goodwill than video advertising and has less spill over to other advertising channels. Traditional (non-video) display advertising exhibits the lowest performance on all dimensions except the impact of clicks on cross-channel sales.