When the Markets Get COVID
Study 2

Announcements study

We look at bond and equity returns around COVID-19 medical announcements, from 60 minutes before to 60 minutes after. The data cover 15,839 hand-collected announcements in 20 countries: case reports, live-streamed press conferences, and speeches by presidents and prime ministers.

Why announcements should matter

A simple model of asset demand

Take an investor with Epstein–Zin preferences who chooses bonds \(B\) and stocks \(S\), where equity payouts carry risk \(\theta\) with variance \(\sigma^2\). With unit elasticity of intertemporal substitution, the demand for equities is

$$\frac{p(S)}{p(B)} \;=\; E_0[\theta] \;-\; (\gamma-1)\,V_0[\theta]\cdot S, \qquad\text{so}\qquad \frac{\partial p(S)}{\partial \sigma^2} < 0 \;\;\text{if}\;\; \gamma > 1.$$

An announcement is an unbiased signal about \(\theta\) that lowers the posterior variance \(\sigma^2\). If investors care about when uncertainty is resolved (\(\gamma>1\)), announcements should on average move demand from bonds to equities. Equity prices should rise. Bond prices should stay flat if bond supply is flat, and dip if it slopes upward.

To test this, we compare the minute-by-minute behavior of each asset around announcements with its behavior at the same clock time before the epidemic (\(Z_t\)). We fit a quadratic in time that is allowed to break at the announcement time \(t^*\):

$$Z_t = \big(c_{pre} + c_{t>t^*}\big) + \big(\alpha_{pre} + \alpha_{t>t^*}\big)\,t + \big(\beta_{pre} + \beta_{t>t^*}\big)\,t^2, \qquad t\in[\,t^*\pm 60\,]$$

Standard errors are HAC-adjusted. A country's epidemic period starts on the day it passes 100 confirmed cases.

Result 1

In both advanced (AE) and emerging (EE) economies, announcements help equities during the epidemic.

Interactive

Equity returns around announcements (epidemic minus pre-epidemic)

Dots are the cross-country, cross-announcement average cumulative return from buying equities 60 minutes before an announcement and holding them for 120 minutes, in the epidemic minus the pre-epidemic sample. The line and band are the fitted quadratic. The dashed line marks the announcement.

Result 2

Bonds are a good hedge against announcement news, especially in advanced economies. Less so in emerging economies.

Interactive

Sovereign bond returns around announcements (epidemic minus pre-epidemic)

Same design, using 10-year sovereign bonds.

These results are not a peso problem. They hold when we use only announcements that carry bad news, when we drop days with news in the top or bottom 1%, and in a sample that ends in December 2020.

The dataset

15,839 announcements, time-stamped to the minute

For each country we took the official press statements from the Ministry of Health website. When a statement had no official timestamp, we used the Twitter account of the Ministry or another government body, and if that failed, major local newspapers. Monetary and fiscal policy announcements are excluded. The dataset is public.

–
announcements
–
in the busiest week
–
first recorded week
–
of announcements are case reports

Weekly announcements by type, built from covid19_announcements.csv. Hover over a bar for the breakdown.

Summary by country (Table 1)
CountryAnnouncementsCase reportsLive-streamedPresident / PMRegularly scheduled
Argentina AR60533%64%3%90%
Australia AU67878%4%1%34%
Brazil BR97564%26%2%54%
Canada CA79158%21%18%31%
Switzerland CH62778%9%13%73%
Chile CL89659%29%3%45%
China CN72182%3%1%75%
Hong Kong CN-HK1,37655%2%1%100%
Colombia CO1,00658%34%8%79%
Germany DE28387%1%7%61%
Spain ES57083%1%17%63%
France FR56777%16%6%84%
India IN75989%1%1%63%
Italy IT65474%17%8%81%
Japan JA33259%5%5%32%
South Korea KR64280%1%4%72%
Mexico MX1,80310%45%21%51%
New Zealand NZ45761%29%7%72%
United Kingdom UK71182%11%7%68%
United States US1,38617%54%7%33%
Total15,83964%18%7%62%

Shares are percentages of each country's announcements. The Total row averages the 20 country shares, as in the paper, so its 64% differs from the 57% in the explorer above, which is the share of all 15,839 announcements.

How we pin down the timing: an example
Tweet announcing the first Coronavirus Task Force press briefing, with a summary of topics discussed
Figure 1

The first U.S. Coronavirus Task Force briefing started at 3:40 p.m. EST, earlier than the White House briefings that followed. It was announced on Twitter 2.5 hours ahead, so we treat it as pre-scheduled.