I am an applied scientist working at Uber Technologies, in the experimentation data science group. My background is in economics and statistics.
I received my PhD from Stanford GSB in 2021, where I did research on applied microeconomics. My PhD work was mostly focused on strategic interactions within the context of advertising markets and the media. I use techniques from economic theory, statistics, machine learning and natural language processing. You can read more about my publications and working papers below.
During my time at Stanford, I also completed an M.S. in Statistics, and worked at Facebook’s Core Data Science team as a research intern. Prior to Stanford, I worked in the finance industry in London. I received an M.Phil. in Economics from University of Cambridge, and a B.A. in Economics from Bogazici University.
- Measuring Investigative Journalism in Local Newspapers, Proceedings of the National Academy of Sciences, July 2021
Eray Turkel, Anish Saha, Rhett C. Owen, Greg J. Martin, and Shoshana Vasserman
We develop a machine learning algorithm to measure the investigative content of news articles. Our method combines an unsupervised document influence model with supervised classification using text data. We use our method to examine over-time and cross-sectional patterns in news production by local newspapers in the United States between 2010 and 2020. We find surprising stability in the quantity of (predicted) investigative articles produced over most of the time period examined, but a notable decline in the last two years of the decade, corresponding to a recent wave of newsroom layoffs.
- Regulating Online Political Advertising, Forthcoming in Proceedings of the ACM WWW ‘22, Special Track on Web for Good: Fairness, Accountability, Transparency, Ethics, Sustainable Development and Healthy Society
Online advertising constitutes a major part of all political ad spending, but regulation has not been able to keep up with this rapid change in the advertising industry. In online platforms, ads are typically allocated to the highest bidder through an auction. Auction mechanisms provide benefits to platforms in terms of revenue maximization and automation, but they operate very differently to offline advertising, and existing approaches to regulation cannot be easily implemented in auction-based environments. To address this challenge, I first develop a mathematical model of online ad auctions and deliver key insights that can be used to design regulation for online political ads. I characterize optimal auction mechanisms in environments where a regulator takes into account both the ad revenues collected by the platforms, and other societal objectives such as the share of all ads allocated to political advertisers, or the prices paid by them. I then build a statistical model using data from Twitter’s political advertising database to estimate bidding distributions and generate simulated auctions to analyze the implications of implementing some of the proposed regulatory interventions. The results suggest that achieving favorable societal outcomes at a small revenue cost is possible through easily implementable, simple regulatory interventions.
Yunus Can Aybas and Eray Turkel (Under review, last updated September 2021)
We study games of Bayesian persuasion where communication is coarse. This model captures interactions between a sender and a receiver, where the sender is unable to fully describe the state or recommend all possible actions. The sender always weakly benefits from more signals, as it increases their ability to persuade. However, more signals do not always lead to more information being sent, and the receiver might prefer outcomes with coarse communication. As a motivating example, we study advertising where a larger signal space corresponds to better targeting ability for the advertiser, and show that customers may prefer less targeting. In a class of games where the sender’s utility is independent from the state, we show that an additional signal is more valuable to the sender when the receiver is more difficult to persuade. More generally, we characterize optimal ways to send information using limited signals, show that the sender’s optimization problem can be solved by searching within a finite set, and prove an upper bound on the marginal value of a signal. Finally, we show how our approach can be applied to settings with cheap talk and heterogeneous priors.
Avidit Acharya, Edoardo Grillo, Takuo Sugaya, and Eray Turkel (Revise & Resubmit, last updated September 2021)
We build a game theoretic model electoral campaigns as dynamic contests in which two candidates allocate their advertising budgets over time to affect their relative popularity (i.e. odds of winning), which evolves over time as a mean-reverting stochastic process. We show that time-dependent regulations—for example, those that prohibit spending in the final stages of a campaign—can be welfare-enhancing and outperform static regulations—specifically, aggregate spending caps. Finally, we use our characterization of the equilibrium spending path to recover estimates of the rate of decay in the effectiveness of advertising in actual elections, using a hierarchical Bayesian model. We use these estimates to examine the effects of dynamic regulations in races that include incumbents.
Posters and Presentations
- Contamination-Aware Experimentation on Networks
Mine Su Erturk and Eray Turkel
We study a setting where a decision maker is conducting experiments in a network environment. We assume the existence of multiple analysts conducting experiments on the same network, as is the case in many online platforms. An experiment creates negative externalities on other ongoing experiments by contaminating their results. We analyze an experimenter’s decision making problem in this setting, where the goal is learning an optimal treatment regime over the network while limiting the contamination on other experimenters. We provide theoretical regret bounds and study the performance of our suggested policy through simulations.