Taiwan Invasion Probability vs Forum Panic: 15 Months of Polymarket Prices Against PTT

In one line: We compared 15 months of the Taiwan invasion probability on Polymarket with 1,816 PTT Gossiping posts about an attack on Taiwan, and the two are largely separate lines — forum spikes around visible events barely moved the market, and the market’s biggest move of 2026 came a week after PTT’s peak, once the board had calmed down.

Two very different measures of “Taiwan risk”

If you follow cross-strait risk, two public signals are easy to reach for. One is a prediction market: Polymarket’s “Will China invade Taiwan by end of 2026?”, opened on 24 July 2025 and, by 2 November 2026, carrying about $45.5 million in volume. Its price is read as the market’s probability that the event happens before the deadline. The other is what people in Taiwan are saying. On PTT, news and general chatter collect on the Gossiping board (八卦板), and when something happens in the Taiwan Strait it shows up there within hours.

It is tempting to treat forum alarm as an early read on risk. We wanted to see whether the two signals actually move together, so we put them on the same calendar.

What we collected

Each post’s time comes from its PTT id, which embeds the unix timestamp. Each market point is stamped at about 00:00 UTC, which is 08:00 in Taiwan, so a market price “on 30 December” is where it stood at 8am Taiwan time that morning.

Finding 1: the two lines barely move together

We summed PTT posts by week and measured each week’s market swing (highest minus lowest daily price), then correlated them over 65 weeks:

A correlation of 0.30 is weak, and neither lagged version is stronger. PTT does not run ahead of the market, and the market does not run ahead of PTT. We left out the market’s first week, when its opening price of 43.5% fell to 21.5% a day later as the market found its level; with that week included, the same-week figure drops to 0.08.

Finding 2: PTT’s biggest days moved the market a point or two

The three busiest PTT days in the window, plus the busiest day of 2026, set against the market:

On 29 December, PTT carried about 73 times its median daily volume, and every one of the top five posts that day passed 100 pushes (shown as 爆 on PTT). The market was 1.5 points higher the next morning and had given all of it back, and more, three days later. The August 2026 war-game story drew 54 posts on one day, and the market stayed at roughly 4%.

Finding 3: the market’s biggest moves came when PTT was calmer

The largest five-day rise of 2026 ran from 2 to 7 January: 9.5% → 15.5%, six points. From 3 to 7 January, PTT carried between 7 and 11 matching posts a day — above its 2-a-day median, but a fraction of the 147 a week earlier. The titles that week were a mix of follow-ups on the December drills; news from outside Taiwan, such as headlines about the US arrest of Venezuela’s president Maduro, posts asking whether the PLA could attempt something similar, and a Chinese export restriction on rare earths; foreign think-tank reports; and hypothetical questions. No post from 2 to 7 January reached 100 pushes. We mention these topics because they were on the board, not because they explain the market. We did not find a Taiwan-side trigger for the move in PTT’s titles, and this data cannot tell you what the market was reacting to.

A one-day whipsaw in September 2025 looks the same. From 5 to 7 September 2025 the price went 17.0% → 12.5% → 17.5% while PTT had 6, 1 and 0 matching posts. Whatever moved the market there did not show up on Gossiping.

Finding 4: the long slide is mostly the clock

¹ Includes the 43.5% opening price. ² Only 10 daily prices that month.

PTT attention peaked in November and December 2025; the market fell steadily the whole time. That steady fall is mostly what a deadline does. This contract only resolves Yes if an offensive starts by 31 December 2026, so every month that passes without one is a month taken off the table. Even if nobody’s view of the danger changed, the price should fall as the date gets closer.

A back-of-envelope check: the August 2025 average of 21.5%, with about 16.6 months left, works out to roughly 1.45% a month. Hold that monthly rate fixed and the same contract, with 2.6 months left in mid-October 2026, would be worth about 3.7%. It actually averaged 2.3%. So the clock explains most of the drop from 21.5%, and the remaining gap is where the market’s view of the per-month risk itself came down. That is arithmetic under one stated assumption, not a forecast.

What the market means by “invade”

Polymarket’s resolution text is narrower than the everyday word. It resolves Yes only if China commences a military offensive intended to establish control over any portion of Taiwan by 31 December 2026, 11:59 PM ET. Inhabited islands administered by Taiwan count, uninhabited ones do not, and confirmation must come from China, Taiwan, the United Nations or a permanent Security Council member, or a consensus of credible reporting. Large drills, behind the two busiest PTT days in our data, do not meet that bar on their own — one reason a busy day on Gossiping and a flat market price are not in conflict.

What to take from this

PTT heat measures attention to visible events: drills you can see on a map, a president’s remarks, a foreign war-game report. The market price is a different thing: a probability for one narrowly defined outcome before one fixed date. In our 15 months, the biggest PTT days were not the biggest market days, and the market’s biggest moves came when PTT was calmer.

If you track Taiwan risk for work — a supply-chain team, an analyst desk, a newsroom — watch both, for different questions. PTT tells you what the public in Taiwan is reacting to and how strongly; it is not a risk signal by itself, and forum panic is not evidence that the probability has changed.

Limits of this comparison

• Titles only, one board. We searched titles, not post bodies or push comments, and only on Gossiping. Other boards and other platforms would add posts.
• Search caps. PTT returns at most 1,000 results per search. 台海 hit that cap, but its oldest result still dates from March 2025, before the window, so no posts inside the window were cut off.
• Different clocks. PTT is bucketed by Taiwan date and the market by UTC date, eight hours apart.
• Gaps in the price history. Polymarket’s API returned no daily price for 26 days, mostly 30 March to 21 April 2026.
• Correlation is not the whole story. A weekly correlation can miss a relationship that only shows up on a handful of days; that is why the event tables above list the days themselves.

Keeping the PTT side running

The PTT half is the part you can keep running — with one catch we found while writing this up. ptt-scraper’s board mode reads a board’s listing page by page and stops after five pages in a row with nothing new to return. That rule keeps an ordinary run from paging forever, but it does not suit a sparse keyword set on a board as busy as Gossiping: at about 4 matching titles a day, five listing pages in a row without a match are common, so a keyword-filtered board run would stop early and undercount. For a daily count like the one above, use Gossiping’s own title search, as we did, or ask us to run it for you.

Where ptt-scraper fits is the next step: reading what people actually said on the spike days. Every post id in our data maps to a URL, and post mode returns the full post plus every push comment, one row each:

{
  "mode": "post",
  "postUrls": ["https://www.ptt.cc/bbs/Gossiping/M.1766991898.A.A84.html"],
  "includeAuthorIp": false
}

That post is the 29 December 「解放軍『極限抵近』鵝鑾鼻僅9海里」 news thread. A post costs $0.008 and each push comment $0.003, and spike-day threads are long: this one had 1,066 push comments when we checked on 2 November 2026, about $3.21 for the whole thread. Start with the handful of 爆 threads rather than every post of the day. includeAuthorIp: false keeps commenters’ IP addresses out of the output.

Three ways to start

From least effort to most freedom

I run it for you Managed, daily to your sheet You do nothing. From US$299/mo. Use it here Run the free tool Instant, no setup, on this site. Run my actor On Apify, your account Same data, you run it and pay per result. Build your own Sign up for Apify Free account — build the pipeline yourself.

This post contains Apify affiliate links (no extra cost to you). Costs and comparisons are stated honestly, including where Apify is not the best fit.

FAQ

Does PTT forum activity predict the Polymarket Taiwan invasion probability? Not in our data. Across 65 weeks, PTT post volume and the size of the market’s weekly price swings had a correlation of 0.30, which is weak. PTT volume in one week against the market’s swing the next week was 0.19, and the market’s swing against PTT volume the following week was 0.06. Neither side reliably leads the other.

Why did the Polymarket probability fall from about 20% to 2%? Mostly because the deadline got closer. The contract only resolves Yes if an offensive starts by 31 December 2026, so each month that passes without one removes a month in which it could happen. If the per-month chance priced in August 2025 had stayed constant, the contract would have been worth about 3.7% in mid-October 2026; it averaged 2.3% that month.

What counts as an invasion for this Polymarket market? Polymarket’s resolution text requires China to commence a military offensive intended to establish control over any portion of Taiwan by 31 December 2026, 11:59 PM ET. Inhabited islands administered by Taiwan qualify, uninhabited islands do not, and confirmation comes from China, Taiwan, the UN or a permanent Security Council member, or a consensus of credible reporting. By that wording, military drills on their own would not count.

Can I get this PTT count every day without running anything myself? Yes, through the managed monitoring service, which can run the same title searches daily and deliver the series. If you run it yourself, note that ptt-scraper’s board mode is a poor fit for a few narrow keywords on a board as busy as Gossiping: it stops after five listing pages in a row with no match, so it would undercount. Its post mode is the right tool for the next step, pulling full posts and push comments from the spike days.

Further reading

• PTT monitoring: hire someone or let a scraper do most of the job? — if you are deciding how to watch PTT day to day, this compares a person on the boards with a scheduled scraper.
• Two scraping modes and the over18 gate on PTT Gossiping — how board mode and post mode differ, and why the board mode’s keyword filter is a title match rather than a site-wide search.

Chad runs 30+ published Apify actors, including the ptt-scraper shown here. The PTT counts come from Gossiping’s title search and the market prices from Polymarket’s public price-history API, both collected on 2 November 2026; every number above is recomputed from those raw files by one script. PTT titles are quoted as headlines only, without usernames.