Game theory says China won't invade Taiwan. The markets say probably not. Now what?

A working executive at a global electronics company in Denver in May 2026 has the same problem every CFO in the supply-chain side of the Fortune 500 has. The board has been asking about Taiwan for three years. The CEO has been telling the board, "We have it covered," for three years. Neither statement is true in any operational sense. Have it covered means the planning team has read a McKinsey memo, drafted a one-page contingency for an invasion scenario, and filed it next to the contingency for a global pandemic.

The standard analyst report on Taiwan, in May 2026, says nothing the executive does not already know. The standard war game — the one that gets quoted at every CSO summit — says a Chinese invasion of Taiwan in 2026 would almost certainly fail, but would devastate the world economy in the attempt. The standard prediction market — Polymarket's flagship Taiwan contract — says a 7% probability of invasion before the end of the year, with more than $23 million traded on that single question as of this week.

What does the executive do with these three inputs? The honest answer in most executive teams is not much. The reports get read. The war game gets quoted in slide 14 of the board deck. The prediction market price gets mentioned, by the one COO in the room who reads Polymarket on weekends, and dismissed by the other ten leaders as gambling. The plan that comes out the other side of that conversation is the same plan that went in. This is the failure mode the discipline of predictive planning is engineered to close.

This article is about how.

What game theory actually says about Taiwan

The intellectual ancestor of predictive planning is game theory. The work of Thomas Schelling on coercive bargaining, of John Nash and Robert Aumann on equilibrium and common knowledge, of James Fearon's bargaining model of war, and of Graham Allison's Thucydides Trap — all of it forms the backbone of how serious analysts model what rational actors do under uncertainty. Predictive planning inherits that backbone. The Loop — Scan, Story, Stake, Steer — is a working analyst's game theory, made operational for an executive team that does not have a defense-PhD bench to lean on.

Applied to Taiwan, the game-theoretic answer is the one most analysts have settled into by 2026. A January 2026 piece, The Taiwan "Idiocy", made the strong-form version of the case: in the cold light of the payoff matrix, a Chinese invasion of Taiwan is the surest way for Xi Jinping to lose everything he is trying to build. The CCP's legitimacy rests on economic delivery to the Chinese middle class. A failed invasion would shatter that delivery for a generation. A successful one would do almost as much damage, by closing every market the Chinese economy depends on. A blockade, an extended gray-zone campaign, an economic squeeze on the semiconductor lifeline — all of these are higher-expected-value moves for Beijing than an amphibious assault, by a wide margin.

The CSIS war game — Cancian, Cancian, and Heginbotham's The First Battle of the Next War — confirms the math from the other direction. CSIS ran 24 iterations of a 2026 invasion. In most of them, the US, Taiwan, and Japan defeated the invasion. In all of them, the cost was catastrophic. The US typically lost two aircraft carriers, dozens of surface combatants, hundreds of aircraft, and thousands of service members in the first three weeks. The Chinese navy ended the war "in shambles." Taiwan's economy was destroyed. There is no version of the war in which any party wins in any sense an executive would recognize as winning.

So game theory says don't invade. The wargames say don't invade. The prediction markets — Polymarket's end-of-2026 contract at 7%, the June 30, 2026 contract at 2%, the September 30 contract at 5% — say the crowd agrees. If your strategy team is running game theory correctly, they have already told the board: the rational expectation is no invasion.

Why rational answers are not plans

Here is where most executive conversations about Taiwan stop, and where the discipline of predictive planning is built to keep going. A rational expectation is a useful input. It is not a plan.

Pearl Harbor was an irrational decision by a standard payoff matrix in December 1941. Japan's military planners knew the war was unwinnable in any extended conflict. They attacked anyway. The Russian invasion of Ukraine in February 2022 was, by every Western payoff matrix circulating before the war, an irrational move that would isolate Russia economically, accelerate NATO expansion, and bog Russian forces down in a country they could not occupy. Putin invaded anyway. Both decisions look, in hindsight, like the textbook examples of the failure modes Schelling spent his career describing — actors operating under domestic constraints, distorted intelligence, sunk-cost commitments, and audience-cost dynamics that the rational-actor model treats as exogenous.

The lesson is not that game theory is wrong. The lesson is that the rational-actor calculation tells you what should happen. It does not tell you what will happen, and it does not tell you what to do if the actor whose payoff matrix you are modeling is operating under a different set of constraints than the ones in your model.

This is the gap predictive planning is engineered to close. The CSIS war game and the 7% Polymarket number are useful inputs. They are not a discipline. A discipline is a continuous operating loop that watches the conditions under which the rational answer holds, and adjusts when the conditions shift. That loop is what an executive team is supposed to install, and almost never does.

The signal layer for Taiwan, May 2026

If you were running the Scan phase of Woodring's Loop on the Taiwan question this week, here is what your signal layer would carry. Not predictions. Inputs to a continuously updating view of which scenario — invasion, blockade, sustained gray-zone coercion, status quo — is converging or collapsing.

Prediction market prices. Polymarket's "by end of 2026" contract sits at 7% YES, with over $23 million in cumulative volume. The "by June 30" contract sits at 2%. The "by September 30" contract sits at 5%. The shape of the curve matters as much as the numbers. A rising end-of-year price with stable short-dated prices implies the market is moving probability mass into Q4. A falling short-dated price with a stable annual price means the market is taking imminent risk off the table without changing its longer view. Both of those signals are different from a flat curve, and the executive team should know which one is in front of them on any given Monday.

PLA aerial and maritime activity. The PLA carried out 169 ADIZ incursions into Taiwan's airspace in April 2026 — significantly lower than the same period in 2025. That number can be read two ways. The optimistic read: coercion is dialing back. The pessimistic read, which is the read most experienced PLA analysts have settled on, is that Beijing is adjusting its coercion strategy precisely because a high frequency of incursions was desensitizing Taiwan's threat awareness. A lower number, in that read, is a worse signal than a higher one. The Scan does not resolve the read. It surfaces both and forces the Story phase to choose.

China Coast Guard incursions. April 2026: three CCG incursions into Taiwan-administered waters near Kinmen, one near Pratas. These are the gray-zone signals that most executives never see, because they do not show up in mainstream coverage. They are the precise category of weak signal the Loop is built to catch.

Undersea cable activity. Taiwan has logged 36 undersea cable damage incidents in five years, with multiple in 2025 connected to China-linked vessels. The Taiwan Coast Guard now runs continuous patrols on TP3 and the major submarine cables. A cluster of cable incidents inside a 14-day window is a signal that, historically, has preceded major PLA exercises. A working signal layer flags cable incidents as a discrete category and tracks the clustering rate.

PLA exercise tempo. The Justice Mission-2025 and Joint Sword series have continued an encirclement-and-blockade theme through early 2026. CSIS's Lights Out war game on a Chinese blockade scenario published in 2025 is the right intellectual companion piece to Cancian's invasion war game. A blockade is the modal scenario most serious analysts now see as more likely than invasion. Your signal layer should weight exercise patterns against the blockade scenario, not just against the amphibious-assault scenario.

Political signals. Lai Ching-te has labeled China a foreign hostile force and accelerated counter-infiltration measures. The KMT is increasing its engagement with Beijing ahead of the 2028 election. The CCP is doing outreach to Taiwanese opposition parties at a tempo that suggests Beijing prefers a political path over a military one. These are not noise. They are the political instruments of the same coercion campaign the PLA exercises represent, and a working Scan reads them together.

Economic and semiconductor signals. TSMC produces roughly 90% of the world's most advanced semiconductors and 99% of the chips used to train frontier AI models. Taiwan is preparing to revise export controls and add 18 items to its strategic high-tech list. The US is moving on semiconductor export policy on its own track. Each of these moves changes the cost-benefit calculation in Beijing. A blockade that interrupts semiconductor flow has different consequences in a world where TSMC's Arizona fabs are at scale than in a world where they are not. Your signal layer tracks the calendar of those fab milestones the way an old-school analyst would have tracked the calendar of OPEC meetings.

Ambient public attention. Google Trends data on search queries like Taiwan invasion and China Taiwan war, segmented by country, is a lagging but useful proxy for how seriously the policy-attentive public in the US, Japan, and Korea is taking the question on any given week. Spikes correlate, with a lag, with policymaker statements and major exercises. The Scan does not predict policy from public attention. It does watch the gap between the two — when the public has tuned out and policymakers are signaling concern, that gap is itself a signal.

What a Stake looks like

The Loop does not stop at Scan. The next phase is Story — translating the signals into a small set of pressure-tested scenarios. For Taiwan, in May 2026, the live scenarios are roughly four:

Status quo with managed coercion — the modal scenario, priced at roughly 70–80% across most serious analyst books. PLA exercises, cable incidents, ADIZ incursions, economic squeeze, but no kinetic conflict.
Blockade or quarantine — the "war below the threshold of war" scenario. Beijing tests US and Taiwanese resolve by restricting maritime traffic. CSIS's Lights Out puts this at the higher end of plausibility for the second half of the decade.
Limited kinetic action — strike on a Taiwanese outlying island (Pratas, Kinmen), cyber operations against critical infrastructure, missile demonstrations. Coercion that crosses a line without crossing the main line.
Full amphibious invasion — the scenario the CSIS war game models. Priced at 7% by end of 2026 in the markets, with most professional analysts agreeing the number should rise modestly as we get closer to the 2027 PLA centennial.

The Stake phase is where most executive teams collapse, because Stake requires committing real resources against scenarios none of which is certain. The right move, almost always, is to size the primary stake against the modal scenario (status quo with managed coercion) and keep tranched optionality against the higher-impact tail scenarios. For an electronics company, that looks like running primary sourcing against the modal world, with pre-negotiated secondary sourcing relationships and inventory hedges sized to switch over inside thirty days if the blockade scenario starts to converge. The Stake Sizing Matrix — one of the diagnostic instruments inside the Loop — gives an executive team a working language for that allocation conversation.

Steer is where most plans actually live or die

The most underweighted phase of the Loop, in my experience, is Steer. Most executive teams can be talked into running a half-decent Scan and a half-decent Story — they have analyst capacity, they have a strategy team, they can spend a quarter on this. What they cannot do, almost universally, is adjust their stake when reality moves.

The Taiwan question is the cleanest possible illustration of why this matters. If the prediction markets begin to drift — if the end-of-year contract climbs from 7% to 12% to 18% over three months, and the September contract climbs from 5% to 9% to 14% in parallel — that is the signal layer telling you the modal scenario is shifting. The CCG incursion frequency, the cable incidents, the exercise calendar, and the political tempo will be moving with it. The question is not whether your team will notice. The question is whether your team has the operating rhythm — the Convergence Review on the calendar, the pre-authorized stake adjustments, the political work of reallocation already mapped — to actually move resources before the scenario fully converges.

The answer in most companies is no. The plan stays the plan. The board meeting cadence does not bend to the signal. Resources move after the Wall Street Journal puts the scenario on the front page, which is the moment the optionality is already gone. Steer is the discipline of moving before the front page. It does not require predicting the future. It requires running a Loop fast enough that the future does not catch you sitting in the wrong scenario.

The unipolar tail

One last thing worth saying directly, because most analyses of Taiwan stop one level shallower than they should.

The reason this scenario matters beyond a supply-chain question is that the modal outcome — sustained gray-zone coercion, US fatigue, semiconductor leverage, political infiltration — is not a de facto status quo. It is a slow-motion path to Chinese unipolarity in Asia, accomplished without an invasion. A Taiwan that falls into a Beijing political orbit over five to ten years, without a shot fired, is the same strategic outcome for China as a Taiwan that falls after a war — just cheaper, faster, and harder for an American president to mobilize against. The first island chain breaches without a kinetic event. Japan and Korea read the room and hedge. AUKUS gets quietly hollowed. The US security architecture in the Pacific stops being the guarantor it has been since 1945. The world that comes out the other side of that decade is one where Beijing sets the operating rules in Asia, the semiconductor monopoly has migrated, and US economic and security guarantees become a question rather than an assumption.

This is the scenario the rational-actor calculation underweights, because it does not require any single irrational move by Beijing. It only requires sustained, patient, coercive pressure that the West does not learn how to counter at the speed the pressure moves. Predictive planning is the discipline engineered to catch that convergence early — when the signals are still weak, the stakes are still small, and the reallocation is still cheap. By the time the convergence is in the headlines, the discipline has already done its work or already failed.

What this article is actually arguing

Game theory tells you what should happen. The wargames tell you what would happen if it did. The prediction markets tell you what the crowd thinks will happen. None of them tell you what to do. The discipline does.

If your executive team has read the war games and quoted the Polymarket price and filed the contingency memo, you have done the analytical work. You have not done the planning work. The two are not the same, and the gap between them is where every "Taiwan in 2027" briefing turns into a slide nobody acts on.

The Loop is built to close that gap. Scan the signals, every week. Build the scenarios, refresh them every quarter. Size the stakes, with reversibility tagged. Steer when reality moves, on a calendar your board agrees to in advance, before the convergence is in front of you.

This is the discipline. It is small enough to install in a quarter. It is large enough to change which decisions your organization makes when the world surprises you, which it will.