Did Collective Intelligence predict the coronacrisis?

Did Collective Intelligence predict the coronacrisis?

Stock markets can make or break a mood. Or do moods perhaps drive the market? The relationship between sentiment and market performance is the subject of a great number of academic papers. At Cindicator, we’re putting this into practice, tracking sentiment through questions sent to thousands of analysts since 2015.

Last November, we added a special new question type dealing with the economy. In this post, we will share how indicators based on these questions changed during the coronavirus crisis and foreshadowed the S&P 500’s local low.

What is Cindicator’s Collective Intelligence platform

If you’re not familiar with Cindicator, here’s how it works. Users register on the Collective Intelligence app for iOS, Android or web. It’s free, and over 140,000 people from 135+ countries have signed up.

Every day, the Cindicator team posts questions about stocks, crypto, commodities, and more. Users make forecasts by answering these questions. They have ‘skin in the game’: if they are right, they win points and climb the rating, based on which the best analysts receive rewards in crypto at the end of the month.

For most questions, Cindicator applies machine learning models to assign weightings to each forecast based on the individual analyst’s track record and other factors. The result is the ‘Hybrid Intelligence’ indicator. For new questions, we start by using the median and then test different models to see how the accuracy reacts. Since 2017, Cindicator’s accuracy has been above 60%.

Economic and political sentiment questions

Starting in late November, we asked a series of new questions every weekend to find out how analysts view the economic and business environment in their countries and what they expect in the future.

We ask about:

  • Current levels of economic uncertainty;
  • Current and future levels of unemployment;
  • The future attractiveness of the business environment;
  • Current and future political stability.

The analysts give an answer as a percentage between 0% and 100%.

For these experimental sentiment questions, we didn’t apply any ML and instead used the median as the indicator and also calculated the difference between the current and the previous week.

Question examples

  • What’s the level of economic uncertainty people in the country you live in are currently facing? ( 0–50% = reduced economic uncertainty; 50–100% = increased economic uncertainty)
  • What is the current employment situation of people in the country you live in? ( 0–50% = negative, difficult to find employment, job losses; 50–100% = positive, easy to find a new job)
  • How will the employment situation in the country you live in develop over the next 365 days? (0–50% = negative development; 50–100% = positive development)
  • How will the business environment in the country you live in develop over the next 365 days? (0–50% = negative development for business environment; 50–100% = positive development for business environment)
  • What is the current level of political uncertainty people in the country you live in are facing? (0–50% = low uncertainty; 50–100% = high uncertainty)
  • How will the political environment in the country you live in develop over the next 365 days? (0–50% = negative development; 50–100% = positive development)

Now let’s review the results.

Employment sentiment and S&P 500

On the above chart, the turquoise line is the median answer to the question about the current employment situation in the analyst’s country. The magenta bars show the percentage difference between the indicator for the week in question and the previous week.

We can see that the indicators usually hovered between 40% and 70%. As a reminder, an answer of 0–50% suggests that the situation is “negative, difficult to find employment, job losses”. Meanwhile, an answer of 50–100% indicates “positive, easy to find a new job”.

During the weekend of 14–15 March, we saw the lowest indicator since we started asking these questions: 25%. The following weekend (21–22 March), the indicator was neutral at 53.8%.

The difference between the two responses is the highest we’ve ever seen: (0.538 - 0.25) / 0.25 = 115%.

The day after this indicator, the S&P 500 index reached its local low of USD 2,191.9.

On the chart below, we can see a similar trend for the question about the employment situation over the next 365 days. The week-on-week differenceis even more pronounced.

Business environment sentiment and S&P 500

The business environment indicator showed a similar trend. From mid-December to early March, this indicator was more stable than the others.

Political environment sentiment and S&P 500

The difference between indicators for political uncertainty peaked the week after the S&P 500 low. This indicator was more volatile than the others, as evidenced by greater differences in previous weeks.

The indicator for the future political environment is more similar to the other indicators.

Economic uncertainty sentiment

The indicator for economic uncertainty showed increasing volatility. While the peak difference came after the S&P 500 low, the previous peak also occurred on the weekend after the S&P 500 reached its all-time high.

Predict the next crisis – get access to private pilot

It appears that a peak in the differences between sentiment indicators predicted the S&P 500 local low.

Of course, more data is needed given the extraordinary events that have taken place: President Trump announced a national emergency and BTC crashed 40% to name a few.

These events, however, are unpredictable and a sentiment indicator can quantify the overall reaction to the barrage of announcements and market moves.

If you want to catch black swans and be among the first to understand how the crisis will develop – whether it’s U-, W-, V-, L-shaped or something else – subscribe to sentiment indicators.

NB: Because this is an early pilot study, we can only distribute these indicators to a strictly limited number of subscribers. First come, first served.

Appendix: Coronavirus timeline

31 DEC Chinese authorities treat dozens of cases of pneumonia of unknown cause.

11 JAN China reports its first death.

20 JAN Other countries, including the United States, confirm cases.

23 JAN Wuhan, a city of more than 11 million, is cut off by the Chinese authorities.

30 JAN The WHO declares a global health emergency.

31 JAN The Trump administration restricts travel from China

2 FEB The first coronavirus death is reported outside China.

5 FEB A cruise ship in Japan quarantines thousands.

7 FEB A Chinese doctor who tried to raise the alarm dies.

11 FEB The disease the virus causes receives a new name.

13 FEB There are over 14,000 new cases in Hubei province.

14 FEB France announces the first coronavirus death in Europe.

23 FEB Italy sees major surge in coronavirus cases and officials lock down towns.

24 FEB The Trump administration asks Congress for $1.25 billion for coronavirus response.

28 FEB The number of infections in Europe spikes.

29 FEB The United States records its first coronavirus death and announces travel restrictions.

13 MAR President Trump declares a national emergency.

15 MAR The CDC recommends no gatherings of 50 or more people in the US.

17 MAR France imposes a nationwide lockdown.

17 MAR The EU bars most travelers from outside the bloc for 30 days.

19 MAR For the first time, China reports zero local infections.

23 MAR Prime Minister Boris Johnson locks Britain down.

24 MAR India, a country of 1.3 billion, announces a 21-day lockdown.

26 MAR The United States leads the world in confirmed coronavirus cases.

27 MAR Trump signs coronavirus stimulus bill into law.

30 MAR More states issue stay-at-home directives.

2 APR Global cases top 1 million, and millions lose their jobs.