Tweets-based

We track the news related to the COVID19 pandemic posted by major newspapers on Twitter. This novel dataset features several dimensions that enable us to study both information production and diffusion. Specifically, our ability to track 'retweets' and 'likes' gives us a high-frequency measure of attention.

Some takeaways...

1) There is significant heterogeneity across countries in the timing of the information diffusion. Across several countries, information diffusion becomes more intense after the beginning of the local epidemic period (i.e., +100 confirmed cases).

2) The time-pattern of the *tweets' tone** is consistent with the observed contagion dynamics. Specifically, the tone became very negative by the end of January as the conditions in China started to precipitate. It improved in early February, when there was still no sign of massive contagion in Europe, and it declined again when the epidemic started in Italy.

*The tone of the tweets is based on a computer-based mapping algorithm that reads our text and classifies the words into three degrees of polarity. Then a relative polarity index (i.e. positive words minus negative words over word count) is computed at the country-level (both hourly and daily). Then, we aggregate this measure in order to obtain a global measure.

For the sake of our asset pricing analysis, we focus on the innovations to the tone of our tweets. One simple way to extract these innovations is to consider the difference in the tone at day t and its 5-day backward looking moving average assessed at time t − 1.

We estimate our asset pricing model through GMM and notice that all portfolios* have an untabulated significant exposure to our contagion-based news. In our sample, the portfolio of countries with the highest share of COVID19 cases tends to be more exposed to contagion news. This sign is consistent with our expectations since negative news about global tone of tweets refers to an adverse shock to equity returns.

*Every day, we group countries into three portfolios according to their relative number of COVID19 cases measured the previous day. We do this separately for AEs and EEs. The H (L) portfolio comprises the top (bottom) countries in terms of COVID19 cases. We also consider an investment strategy long in the H portfolio and short in the L portfolio (HML).

Looking at the output of our specifications and accounting for estimation uncertainty, we conclude that 0.3% is a reasonable lower bound on the daily market price of risk of daily contagion news.

Two observations...

1) This is not the MPR of a financial factor and the associated estimated beta are very small.

2) Contagion risk follows waves with a relatively short half-life. Equivalently, the exposure of our assets to this risk are small and relatively quick in reverting to zero.

Our results confirm that sovereign bonds issued by AEs are not sensitive to contagion risk. Equities, instead, experienced a more pronounced increase in their required risk premium among High-COVID countries. In contrast, in EEs both bonds and equities feature a much more pronounced increase in their riskiness. Bonds' exposure, however, has been smaller than that of equities', confirming that also EEs' bonds are safer with respect to contagion risk.

Now, focusing on the HML strategy helps us to highlight the role played by heterogeneous exposure to contagion risk.

We observe that..

1) The riskiness of bonds has increased less in High-COVID countries than in Low-COVID countries. Equivalently, in High-COVID countries, bonds are relatively safer assets. As a result, an HML-COVID strategy on bonds provides an insurance premium.

2) In AEs, this premium is very moderate, consistent with our prior empirical evidence on the muted response of bonds around medical announcements time. In EEs, instead, the insurance premium is quantitatively relevant both in local units and in USD.