Forecasting Second Lockdowns and their Macro-Economic Impact Using RavenPack

RavenPack | August 11, 2020

This article looks at how alternative data can forewarn investors of the reintroduction of economically damaging lockdown measures and further to estimate the impact of such measures on the economy

Many countries have had second waves of coronavirus cases and as a result have had to reimpose lockdown measures, with damaging consequences for their economies.

RavenPack news analytics can help investors forecast the reimposition of such measures, and also, once imposed, the economic fallout.

After conducting research on 10 different APAC countries, some of which have had a second lockdown, and some not, our researchers found that an acute rise in certain predictive headlines provided a confirmatory aid for anticipating the reintroduction of second wave lockdown measures.

Whilst far from providing a ‘holy grail’, the model was, nevertheless, 70% accurate at indicating second lockdowns across the 10 APAC countries studied, as can be seen from the table at the bottom of the next section.

We recommend using the model as a factor in a mosaic style analysis approach, rather than a stand-alone indicator.

In this article, we will first explain how the lockdown model was developed and then go onto discuss briefly how analysts and investors might use the platform to go a stage further and predict the likely economic impact of reimposed measures.

Headlines Warning of Second Waves

From our analysis we found that the two headlines which most accurately seem to provide an early warning of the onset of second waves and the reimposition of lockdown measures are those that contain the phrases “highest cases (for) weeks” and “first case (in) weeks”.

In the case of countries that have experienced second lockdowns, headlines containing one or other of these phrases showed a tendency to produce a spike, often in the first half of June which predicted another lockdown. The chart below illustrates a classic example of the pre-lockdown pattern. It shows the tell-tale spike in the number of headlines mentioning “first case (in) weeks”, in this case in Hong Kong. The spike in early June preceded the reintroduction of measures on July 27.

alt text

Below is another example, this time from Australia, where a more severe second wave led to a lockdown in Melbourne. Here again, we see the tell-tale spike in headlines of “first, case, (in) weeks” in early June prior to the reimposition of measures.

alt text

In our third example, we look at headline counts in China. Here too we see the same spike occurring for “first, case (in) weeks” before an area of Beijing was brought under lockdown following an outbreak at the Xinfadi wholesale food market.

alt text

A similar spike can be observed for the other key phrase that is predictive of second lockdowns “highest cases (for) weeks”, only with a shorter lag.

alt text

Countries Where There Has Been No Second Wave Lockdown Yet

In contrast to countries that have experienced second waves and reimposed lockdowns, the predictive headline spike above does not appear as frequently in the headline count profiles for countries where there has been no second lockdown.

Take, for example, the case of India which, whilst suffering an extremely acute pandemic, nevertheless, has so far not reintroduced lockdown measures (it is still in fact technically in the last stage of withdrawing from the first set of measures).

For India, the same headline count profile shows a different schema, with no clear crescendo spike as in the examples above.

alt text

The exact same goes for the headline count for the other phrase “highest cases (for) weeks”.

alt text

Malaysia is another country which has so far evaded a second wave and an analysis of associated headline counts reinforces this as we see no predictive spike.

alt text

For the other key phrases in the study “first case (for) weeks”, in the case of Malaysia, only one headline was found, making it unnecessary to draw a chart.

Singapore provides our third example: it too is an Asian country where there has been no second wave lockdown, and once again, here too we see a lack of any spike to forewarn of the reimposition of measures.

alt text

For the other headline, in the case of Singapore, the platform came back with a zero count, so there was no necessity to draw a chart.

Japan: A Warning of What is to Come?

Japan appears to be a special case as although it has avoided the necessity of reintroducing lockdown measures the model and recent news flow may be signaling the country is at risk of introducing a lockdown in the future.

In the case of Japan, we see a wider spike occurring for “first case (in) weeks” in mid-July, which could be an early warning sign of lockdown measures to come.

alt text

We also see a similar spike occurring, although earlier in late June for the count of the other key headline “highest cases (in) weeks”

alt text

Although the patterns for Japan have a wider shape than those associated with previous examples, it may still be possible they could be warning of measures that could be introduced down the road.

At the same time, Japan suffered a relatively benign first wave due to a combination of unique cultural, demographic, and geographic factors, so this may raise the bar in terms of expecting a return to lockdown, especially punitive measures.

This is why the data should be used in conjunction with other sources and a big picture approach.

Signal Accuracy

The table below shows whether countries have had a second lockdown or not and whether the model using the two different phrases was successful in forecasting it.

alt text

Using News Analytics to Forecast Economic Growth

Assuming second lockdowns are reintroduced, how can news analytics be used to predict the economic impact of the resulting caesura?

Previous studies have shown that news analytics - in particular RavenPack news sentiment - can be used to gauge the likely outcome of future macroeconomic releases such as GDP, and also incorporated in nowcasting macroeconomic models.

To explain how this might work, let us take a very simple example using GDP, and let us assume forecasts based on economist’s expectations for next quarter’s GDP are positive but aggregated news sentiment, generated from scanning thousands of articles online about economic growth, is suggesting that growth will be lackluster - in such instances, there is usually a strong possibility the official forecasts will be off.

One example of how RavenPack news analytics data has been used to improve macroeconomic forecasts can be found in the case of Chinese GDP data. Here a study conducted by Danish Economist Asger Lunde, found RavenPack news sentiment could be used as a stand-alone model for forecasting future economic growth well ahead of the release of official figures.

Lunde concluded, “The results suggest that including news sentiment significantly improves forecasts.”

The data cannot only be used to forecast future changes in GDP it can also be applied to forecasting more frequent macroeconomic releases such as weekly and monthly jobs data, housing data, retail sales, and monthly PMI releases.

Since many of these releases are themselves used by economists to forecast quarterly GDP, even earlier estimates of them based on news analytics data can enable even earlier estimation of themselves and GDP.

Finally, news analytics has been successfully employed in many bottom-up models. One such model used news about company earnings to assess growth in individual U.S. states as an input into assessing municipal bond valuations.

The same methodology could also be applied at a national level both for government bond yields and gross economic growth.

Subscribe to COVID-19 Updates

Request a Trial

Fill out the form below and see RavenPack in action.