Research

Job Market Paper

Modeling Social Learning as Epidemics using Twitter Data

An important debate in macroeconomics is how people form expectations. Rational expectations and adaptive learning approaches assume that all agents in the economy are as good as econometricians. In reality, most people learn by talking to their neighbors, relatives, and friends. However, obtaining data on expectations is extremely difficult. Most of the surveys of expectations are limited to a few economic variables like inflation, unemployment etc. This paper proposes a new way of modeling social learning by using an epidemiological model. Twitter data is used as a proxy for people talking to each other. Disease and expectations both spread infectiously among people before dying out. Predicting the trend of spread of expectations is important for policy formation. Least squares estimation is used to estimate the speed of transmission of both economic and noneconomic news items among people. The estimates obtained by fitting Twitter data on the epidemiological model are compared to the rate of spread of different epidemics. How quickly and effectively does news spread is another question of interest. Fast transmission of news implies that people adjust their expectations quickly vs. slow transmission that implies sticky information. The results suggest that economic news spreads relatively slowly, and there is heterogeneity in person to person learning. Information transmission on twitter spreads like a flu which changes its course over a period of time. However, unlike epidemics each person has a different probability to spread news due to the presence of influencers on Twitter.

JEL Codes : D84, E70

Keywords : Social Learning, Twitter, News Transmission, Person to Person Learning

Dissertation Advisor : Professor Wei Xiao

Dissertation Committee : Professor Barry Jones, Professor Florian Kuhn

* Poster presentation at AEA Conference 2019 held at Atlanta: https://www.aeaweb.org/conference/2019/preliminary/powerpoint/KK8rhKEy

OTHER RESEARCH

Other Papers

• Demand Driven News Shocks, Animal Spirits and Consumer Confidence

In this review, I look at the recent literature on news and animal spirits shock as a source of aggregate macroeconomic fluctuations. The question that the literature has tried to answer is how does the aggregate economy respond to agent's expectations regarding future productivity. I discuss the various approaches that has been adopted to solve the identification issue i.e. separating the news component from the animal spirits component. Both news and sunspots have been incorporated in DSGE models and the theoretical impulse responses are matched with real data to see which shock has more impact on consumption, output etc. So far, only news about future productivity has been considered. I propose the idea that demand driven news shocks can lead to different results and implications. I incorporate news and animal spirits shocks about aggregate demand in a New Keynesian DSGE model to study their relationship with consumer confidence and subsequent macroeconomic activity.

Keywords: consumer confidence, demand driven news shocks, future productivity, animal spirits, identification

JEL Code: D12, D84, E12

• Incorporating Social Learning in Asset Pricing Model using Agent Based Learning

This paper introduces social learning in an asset pricing model with boundedly rational heterogeneous agents who have different expectations about the tipping point of an asset price. Agent based modeling is used to replicate real movements in financial market assuming that each person not only transfers the news to other people but also his/her sentiments regarding the asset price fluctuations.