Current Research

We investigate the effects of online word of mouth on the demand for new products using Twitter data. Twitter can both generate buzz & awareness as well as provide information on product quality that can readily diffuse through the population. Leveraging comprehensive data from the US movie industry and Twitter, we estimate a structural model of consumer demand for attending theatrical releases in 2014-2015 that incorporates both information channels. The results show that both channels are important, but differ across types of movies. We find pre-release tweet volume is the most important channel for large franchise movies, generating buzz that influences box office earnings on the opening weekend. In contrast, the sentiment expressed in online WoM after a movie’s release influences box office demand in subsequent weekends for smaller movies. Demand for mid tier movies responds to increasing awareness driven by the volume of tweets posted after a movie is released.

This paper investigates whether firms’ advertising influences online consumer word of mouth in new product markets. The increasing importance of online WoM in consumer decision making paired with the rise of ‘second screening’ by consumers means that firms are increasingly seeking to understanding how TV advertising impacts online behaviour. We utilize a large, granular data set on minute-by-minute tweeting behaviour for all wide release movies in the United States over 20142015 and match it with data on individual advertisements on national television to study whether TV ads lead to short-run spikes (i.e in the following 5 minutes) in online WoM. Our econometric approach utilizes a high dimensional fixed effects strategy to control for the quantity of consumer tweets that are not driven by advertising. Preliminary results show that TV ads cause short term increases in online WoM with an extra TV ad leading to 3.5 extra tweets, the equivalent of 1.5 additional minutes of tweeting. We also show that the effect of ads on tweeting only exists pre-release, and is larger the further away an ad is from the release date.

We analyze binary choice models in communication networks, in which both the formation of links in the network as well as the action choices are endogenous. In particular we study the situation where an individual’s payoffs are influenced by their own ideology, decisions of peers and beliefs about the actions of the entire population. We provide a complete characterization of the equilibrium action choices and networks, where agents choose their strategies - actions and links - according to a perturbed best response update rule. We show that a threshold exists in the cost of forming a new connection and an individual’s relative preference to coordinate an action with their peers, giving rise to either a sparse or a densely connected communication network. Moreover, we show how the theoretical model can be efficiently estimated using data on agents’ choices and their network of interactions. We then structurally estimate the model using datasets of Twitter users that allows us to construct their social networks during the popular uprisings in Latin America (2011 and 2012) and the Arab Spring (2011). Preliminary results emphasize the importance of the actions made by both peers (local effects) and the wider network (global effects).


More Publications

(2016). Commitment and Cooperation in Partnerships. Games, 7(1) Art. No 4.

Preprint PDF

Working Papers

More Publications

(2019). Online WoM and the Performance of New Products (Job Market Paper).


(2018). Gender Effects and Loss Aversion: Yes, No, Maybe?. Working Paper (R&R at J. of Risk and Uncertainty).


(2017). The Regulation of Public Service Broadcasters: Should there be more advertising on television?. CEPR Discussion Paper 12428 (R&R at IJIO).



Teaching @ UZH

As an Instructor:

As a Teaching Assistant:

  • Cross Section & Panel Data Econometrics, 2nd year Masters level, 2014-15 & 2015-16
  • Empirical Analysis of Asymmetric Information in Banking and Insurance Markets, 2014-2015

Software Carpentry & Data Carpentry

I am an instructor & lesson maintainer for Software Carpentry and Data Carpenty, who offer short courses on programming and reproducible research practices in professional and academic settings.


Lesson Maintainence (with many others):

Contributions to lessons:

Thesis Supervision

I actively supervise Bachelor’s and Master’s theses in Industrial Organization, Applied Econometrics and Media Economics. Contact me via email if you are interested in knowing more.


R Packages

Empirical IO

demandr: Nested Logit Demand Estimation with Aggregate Market Shares

  • Estimate Nested Logit Demand models on aggregate market share data
  • Coauthored with Christian Oertel (U Zurich)
  • Status: Under development
  • Code: available on Github


hayshir: R Companion to “Econometrics” by Fumio Hayashi

  • Provides all data sets used in textbook, and companion code to each chapter in text
  • Website
  • Code: available on Github


Installation Guide for Software and Packages

While teaching programming skills I have assembled a set of instructions for setting up a computing environment that I find useful for economics / marketing research. Instructions are provided for Linux (Ubuntu flavored), Mac and Windows environments.

  • Installation Guide
    • Borrowed and updated from the programming course I co-taught at UZH in 2017.

Reproducible Research Workflows

The Snakemake workflow management system helps to create reproducible and scalable data analyses. It was originally designed for Bioinformatics workflows, but I have found it suits research in empirical economics and marketing just as well. Official documentation for Snakemake is here. Below are some templates that I use for my research and an extended tutorial.

Some of my older projects use Waf to implement a reproducible workflow. Waf was introduced to me by Hans-Martin von Gaudecker. He provides great templates here in multiple programming languages along with an extended tutorial. It’s a good alternative, but I think the Waf learning curve is a tad steeper.

Markdown templates

Most of my written work is in markdown. Here are some templates I regularly use: