CSET Foretell's latest research and events
Future Indices report · Using Crowd Forecasts to Inform Policy (recording) ·
Upcoming event December 15th, 2:00pm ET: Forecasting U.S.-China [De]Coupling in the Semiconductor Industry
« CSET Foretell Blog

Forecasting Conflict in the South China Sea

Oct 30, 2020 01:26PM UTC
Michael Page and Alex Barker
Foretell is CSET's crowd forecasting pilot project focused on technology and security policy. It connects historical and forecast data on near-term events with the big-picture questions that are most relevant to policymakers. This post is part of our Forecasts Analysis series, where we analyze recent crowd forecasting trends on Foretell.

The crowd currently forecasts a 17 percent chance of a violent conflict1 in the South China Sea during the first half of 2021, or about one chance in a thousand per day. The risk of conflict might be declining, however. Forecasts made in October suggest a lower chance of conflict than forecasts made during any other month since June. While changes in how the question is framed might account for the October drop, a similar trend in the probability China invades Taiwan suggests the October drop might reflect more than methodological changes. We expect November data to be informative on this point.

Source: Foretell. Data pulled October 29, 2020.

Figure 1 shows the distribution of forecasts for the chance of violent conflict in the South China Sea between January 1 and June 30, 2021. Among the 91 people who’ve forecasted since the question was published on October 9, 2020, the average forecast is 17 percent.  

We previously asked the same question covering the period from June through September 2020. Because the operative time period shrank as the deadline approached, the crowds’ forecast approached zero during the life of the question. To allow for comparisons between the two questions, we divided the forecasts by the amount of time remaining during the operative time period, thereby generating risk-per-day forecasts. For example, for the previous question, a 10 percent forecast made on August 15 assessed the risk of conflict between August 15 and September 30, a 47-day period. Dividing 10 percent by 47 days, we get a risk-per-day forecast of 0.2 percent or two in 1,000.

As shown in Figure 2, the risk-per-day forecasts increased between June and September before decreasing significantly in October.2 One interpretation of the data is that the crowd believes the risk of conflict in the South China Sea decreased. As shown in Figure 2, the fact that we see a similar October drop in a related question regarding the probability China invades Taiwan by May 31, 2021 provides some support for this interpretation.

Source: Foretell. Data pulled October 29, 2020. For the South China Sea question, the June-September data comes from an earlier question that asked about the risk of conflict through September 30, 2020; the October data comes from a new question that asks about the risk in the first half of 2021.

An alternative explanation for the October drop is that it’s an artifact of our methodology. The October data is from the new question covering a longer and further-in-the-future time period: January 1 through June 30, 2021. If forecasters overestimate the concentration of risk for small time periods—or underestimate the concentration of risk for large timer periods—that would explain why the risk-per-day forecasts increased through September before decreasing in October, when we switched to the new timeframe. With a few more months of data, we should be able to more confidently determine whether this trend reflects actual changes in the world or whether it’s instead an artifact of our methodology.

The events discussed in this post—conflict in the South China Sea and a Chinese invasion of Taiwan—are black swans: events that are difficult to forecast and unlikely in the course of any given year, yet would have a transformative impact on the geopolitical landscape. Knowing whether the per-day risk of these events is one in 1,000 or one in 100, and whether that risk is trending in one direction or another, matters a great deal. Forecasting such events on Foretell can help policymakers more confidently anticipate destabilizing risks. 


1 Violent conflict is defined in the question to mean the discharge of a weapon with lethal intent and does not include methods such as water cannons, rubber bullets, or ramming.
2 We omitted forecasts made during the final week in September as the probability approached zero. Because forecasts can only be made in whole percentage points, Foretell is not designed to capture very small probabilities. 

To stay updated on what we’re doing, follow us on Twitter @CSETForetell.


Michael Pagee

Michael Page

CSET Research Fellow

Alex Barker

Alex Barker

Student, Georgetown School of Foreign Service

Tip: Mention someone by typing @username