Economic Decoupling as a Readiness Signaling Device

Abstract

This paper presents a game-theoretic analysis of economic decoupling as a deterrence mechanism amid the renewed risk of international conflict between two economically integrated superpowers. In our model, a potential aggressor considers initiating military conflict, while the defender, anticipating this risk, can preemptively decouple to signal its readiness to resist aggression. We examine both a full-information game and a more realistic setting with incomplete information about the defender’s priorities. Under full information, we identify the optimal degree of decoupling that ensures effective deterrence. With incomplete information, deterrence becomes more fragile: although a peace equilibrium remains possible within a narrower parameter range, hybrid equilibria, where conflict occurs with positive probability, also emerge. The results highlight how even minor informational imperfections can generate intrinsic geopolitical instability, emphasizing the fragility of peace under conditions of economic interdependence.

Introduction

After the Second World War, the global perspective of Western countries on international trade was grounded in Ricardian theory, which emphasized the mutual economic benefits of countries freely exchanging goods and services. While it was acknowledged that redistributive tensions could arise from time to time, World Bank data shows that trade as a share of global GDP increased from roughly 40 percent in 1980 to around 60 percent in 2010, before stabilizing. ¹

As a positive by-product of this wealth-enhancing trade, many experts have argued that bilateral economic interdependence reduces incentives for war by raising its opportunity costs, a view often referred to as the liberal peace hypothesis (Huth and Russett 1984; McDonald 2004; Polachek 1980; Polachek and Xiang 2010; WTO 2023). This idea was foundational to the creation of the European Union and its single market in the aftermath of the Second World War, and it likely contributed to the unprecedented period of peace Europe has experienced since 1945, following centuries of recurrent conflict (Martin et al. 2008). The same logic also underpinned a key argument for the Western world’s decision to open to communist China in the 1970s (Kissinger 2014).

Not everybody shared this point of view. The idea that economic considerations are often subordinated to political goals in international relations can be traced back to the early writings of Hirschman (1945). He showed how, in the 1930s, Germany used international trade to increase the economic dependence of South and East European countries as part of its preparation for war, with the implicit aim of compelling them to collaborate with the Nazi regime once conflict erupted. His analysis anticipated the more skeptical position later developed by authors such as Waltz (1979) and Barbieri (1996, 2009), who emphasize that economic interdependence can create vulnerabilities and power asymmetries rather than reliably fostering peace.

After 2017, the voices of trade skeptics grew louder, and protectionism became the new norm (Fajgelbaum et al. 2020), reflecting what appears to be a “global backlash” against globalization (Colantone et al. 2022; Goldberg and Reed 2023). This shift has taken place amid rising geopolitical tensions, as several authoritarian regimes have openly challenged the existing world order and, in particular, the previously uncontested U.S. hegemony. Academic research increasingly confirms that, in the face of growing geopolitical tensions, states are nowadays seeking greater autonomy in key sectors such as defense, energy, and food supply. To achieve political and security objectives, governments are deploying a broad range of economic policies to reshape international trade and supply chains (Clayton et al., 2025; Goldberg and Reed, 2023; Mohr and Trebesch, 2024; WTO, 2023).

Furthermore, many governments appear willing to forgo some of the efficiency gains from trade in order to pursue geopolitical objectives, particularly alignment with strategic allies (Antràs 2024; Baldwin 2025). In support of this argument, several scholars analyzed granular bilateral trade data and showed that, after 2017, trade flows tend to reorganize along a reconfiguration of the main supply chains, with declining flows between blocks of countries with a similar geopolitical preference, and stronger flows within these blocks (Airaudo et al., 2025; Aiyar et al., 2023; Attinasi et al., 2025; Gopinath et al., 2025; Qiu et al., 2025).

This paper develops game-theoretic foundations for understanding the ongoing process of economic decoupling at a time when the risk of conflict between major powers is once again rising. Our analysis sheds light on why, after decades of deepening integration, the United States and China may now find it optimal to reverse course.

Economic decoupling is defined here as a national strategy intended to reduce a state’s economic dependence on another country (Witt et al. 2023), particularly with respect to flows of goods and services, critical inputs, capital, research partnerships, and the integration of supply chains. It constitutes a reversal of trade specialization and economic cooperation, requiring both the initiating and the target countries to redesign their networks of partners or to produce goods and services domestically. To pursue this strategy, a state may impose tariffs, introduce targeted trade restrictions, limit capital movements, restrict access to domestic markets, subsidize key sectors and firms, and regulate the exchange of technology and information with the foreign country.

This policy is generally expected to result in welfare losses for both the country initiating the process and the target country. ² Why, then, would a country willingly bear such substantial separation costs? We suggest that by shifting the burden of separation before a conflict arises, a state pre-commits to military actions should it become necessary. This revealed readiness serves as a deterrent, ensuring that the other state does not engage in aggression.

To formalize this argument, we model the problem as a sequential game between two states with highly integrated economies that may come into conflict over control of an unarmed territory. Because the two states’ economies are deeply interconnected, any armed conflict would entail substantial economic losses. The attacker would exploit this vulnerability, expecting such costs to deter the defender from engaging in military action, and would therefore attempt to seize the territory. The defender, however, can adopt a preemptive strategy to reduce economic dependency, thereby lowering prospective wartime separation costs. By incurring these costs in advance, the defender renders its threat of military retaliation credible.

In the first part of the paper, we analyze the complete-information game and identify the general conditions under which a “peace equilibrium” (where the attacker is deterred) exists. We also characterize the optimal level of decoupling that ensures effective deterrence at the lowest possible cost for the defender. This strategic analysis builds on the rationale first advanced by Schelling (1960) who explained that a threat is credible only if the player manages to pre-commit; this can be done only if the cost of the treat is charged before its would-be implementation. This sunk-cost logic was formalized and received a microeconomic foundation in traditional entry-deterrence models (Dixit 1979, 1980; Spence 1977). In these models, incumbent firms in an oligopoly market may strategically invest in building some barrier to entry (excess productive capacity, advertising, reputation, patents or product variety), since this would allow them to quickly expand output and lower prices if a rival decides to enter the market. In turn, such pre-committed expenses should dissuade potential new competitors to enter the market, and appropriate part of the profit. The decoupling cost in our analysis is similar to cost of excess capacity or any other barrier to entry.

In the second part, we introduce a small degree of uncertainty by assuming that the attacker does not know the defender’s true valuation of the territory, which remains private information. As in the complete-information setting, the defender may adopt a preemptive decoupling strategy to signal its readiness to defend while simultaneously revealing information about its unobserved preferences. In this way, the entry-deterrence game encompasses a signaling game à la Spence (1973). In a traditional signaling game, players in the sender role have private information about their productive characteristics; high-type agents strive to reveal their type to the receiver by sending an appropriate signal, knowing that low-type agents can imitate their actions. In general, the solution is a Perfect Bayesian Equilibrium, where sender's actions and the receiver’s beliefs are mutually consistent (see the surveys by Riley (2001) and Sobel (2020)).

Our analysis shows that, under certain conditions, a hybrid Perfect Bayesian Equilibrium may arise in which the attacker randomizes over whether to initiate an attack. The conclusion of this section tempers optimism about the existence of a peace equilibrium, which can no longer be taken for granted. A key contribution of our analysis is to show that, in the specific context studied here, even a small amount of uncertainty can generate substantial geopolitical instability.

To sum up, in this paper the decoupling strategy of the defender is a genuine sunk cost, which is used strategically by the defender, either to reverse the balance of gains and deter the attacker in a full information setting, or, in an incomplete information setting, to convey information about its type and therefore prompt the potential attacker to adapt its own strategy in a direction favorable to the defender.

Related Literature

Our paper can be seen as a theoretical contribution to the empirical literature on the relationship between economic integration and conflict deterrence. Empirical work by authors such as Huth and Russett (1984, 1988) and Oneal and Russet (1997) generally supports a positive link, although this view has been met with skepticism by Lebow and Stein (1990). Our findings suggests, however, that the effects of economic integration can differ depending on whether it occurs between political allies or strategic rivals.

A substantial literature in the study of international conflict acknowledges the role of incomplete information as a source of conflict. The game-theory foundations were laid out by Harsanyi (1967, 1968a, 1968b), who extended the concept of Nash equilibrium to games in which players possess only limited knowledge of their opponents’ payoffs. As Harsanyi (1995) emphasized in his Nobel Prize Lecture, in arms-control negotiations “each side is well informed about its own policy objectives, its peaceful or bellicose attitude toward the other side, its military strength, its own ability to introduce new military technologies, and so on – but may be rather poorly informed about the other side’s position in terms of such variables.”

Our paper is also related to the literature that analyses the roles of signals in military strategy, international conflict, and national security. Application of the traditional signaling model helped to explain phenomena as diverse as ancient Chinese military strategies (Cotton and Liu 2011), persuasion and disinformation tactics (Arce 2024), forced entry into the nuclear club (Jelnov et al. 2018), contemporary terrorist threats (Arce and Sandler 2007; Lapan and Sandler 1993; Zhuang et al. 2010), and even ammunition shortages at the outbreak of the war in Ukraine (Besancenot and Vranceanu 2026). In many of these applications, costly signals (sunk costs) such as audience costs (Fearon 1994), military mobilization (Fearon 1997; Slantchev 2005), or inclusion of an opposition-party in the negotiation (Schultz 1998), can credibly reveal information about players’ preferences or capabilities and thus help prevent conflict. At the same time, Powell (1999) emphasizes that costs are not merely signal of type, but strategic investments in credibility within an evolving power structure. Audience costs may create “lock-in,” mobilization can increase chances to prevail in conflict, and broad domestic endorsements may harden bargaining positions.

Many scholars have developed bargaining models to explain why parties sometimes fail to reach the efficient outcome of peace (see surveys by Garfinkel and Skaperdas (2007), Baliga and Sjöström (2013), and Ramsay (2017)). In an early contribution, Brito and Intriligator (1985) showed that private information can act as a transaction cost that prevents surplus-maximizing bargains: a potential aggressor may bluff, and uncertainty about his type can rationally induce resistance and conflict. Fearon (1995) formalized this logic by casting war as a bargaining failure driven by incentives to misrepresent capabilities or resolve and by commitment problems, demonstrating that even within a bargaining range, asymmetric information and non-transferable utility can still yield inefficient conflict. These insights connect to Powell’s (1999) dynamic “in-the-shadow-of-war” model, where war occurs when a state opts out because further bargaining is worse than fighting; refinements by Leventoğlu and Tarar (2008) show how greater bargaining symmetry can generate equilibria with negotiated settlements rather than delay. Polachek and Xiang (2010) and Gartzke et al. (2001) use the bargaining framework to study the pacifying effects of economic integration. Other work highlights how uncertainty may be strategically produced: Meirowitz and Sartoriet al. (2008) show that pre-crisis military investments can induce states to obfuscate capabilities, endogenously generating uncertainty that precipitates war. Research on coordination and cheap talk further illustrates the role of private information: Baliga and Sjöström (2004) demonstrate that in an arms-race game with type uncertainty, threshold equilibria can lead to arms buildup, while cheap talk may mitigate conflict; Baliga and Sjöström (2012) analyze how extremists manipulate public messages; and Baliga and Sjöström (2020) examine how private cost types shape outcomes in a bargaining model with irreversible commitments. Martin et al. (2008) build a negotiation model with imperfect information in which that wars can occur because disputes on how to share the surplus under peace may escalate into a military conflict; their empirical evidence reveals a negative effect of bilateral integration on the likelihood of war, but a positive correlation between multilateral integration an the same likelihood. Acemoglu and Wolitzky (2024) develop an inspection game in which the defender receives only a noisy signal of the attacker’s first move and establish conditions for a hybrid sequential equilibrium with both sides mixing between aggression and restraint. A sequential game of potential conflict between two states, characterized by asymmetric information about the probability of prevailing in war and by heterogeneous domestic political costs of conflict, was developed by Bueno De Mesquita and Lalman (1990, 1992). Their analysis thoroughly examines the consequences of alternative information structures and shows that initial beliefs about the extent of a state’s domestic opposition play a crucial role in shaping crisis outcomes, even though such beliefs are not easily manipulated.

Several studies reveal that conflict might arise even with perfect information. In Fernandez and Glazer (1991), the mere possibility that the union could strike, even in situations where no strike occurs on the equilibrium path, only to strengthen its bargaining position. Consequently, the union’s threat to strike may remain credible despite the costs it would incur by following through. As the authors note, this logic also applies to tariff wars. Slantchev (2003) demonstrates that war can arise as an outcome of an imperfect bargaining process when fighting unfolds dynamically and states use costly conflict to enforce minimax threats, revealing how inefficiency can stem from strategic structure rather than informational asymmetries.

In contrast to these studies that focus primarily on the outbreak or persistence of war, our analysis applies the entry-deterrence game introduced in the first part of the paper to examine strategic choices aimed at avoiding war between two economically integrated states. While our approach differs in structure from the canonical bargaining and crisis interaction models surveyed above, it shares with them a central concern for how imperfect information shapes equilibrium outcomes, and it arrives at a closely related conclusion: under incomplete information, the commitment mechanism of decoupling becomes less efficient, thereby increasing the likelihood of conflict even when peace would be mutually beneficial.

The paper is organized as follows. The next section presents our basic model and derives the subgame perfect equilibrium of the sequential game with complete information. Another section introduces the incomplete information game and analyzes its equilibria. The following section presents a policy discussion that illustrates our deterrence mechanism. The last section concludes.

The Complete Information Framework

Main Assumptions

In the first part of the paper, we model the problem as a full-information game between State A (the potential attacker) and State D (the potential defender). States A and D share an integrated economy, the result of years of mutual trust and the pursuit of gains from specialization.

At a certain point, State A reveals its intention to invade an unarmed territory that was initially a trading partner of State D. In response, State D may choose to intervene and use force to defend the territory. ³

The probability that State D prevails over State A in an armed conflict is denoted by p. For the sake of parsimony, we follow Fearon (1997) and assume that this probability is exogenous.

Let W be the gain for the attacker from acquiring control over the territory.

Let L be the loss to the defender from losing access to the territory.

Let F be the cost of military conflict, assumed equal for both sides for simplicity.

Let E denote the total cost that each state would bear if they were to fully separate or “dis-integrate.” Such a situation may arise, for example, if a sudden war breaks out between them (Polachek and Xiang 2010). This would lead to the abrupt suspension of bilateral trade in goods and services, the disruption of FDI and financial flows, supply-chain breakdowns leading to shortages, the forced withdrawal of firms with associated fire-sale losses, and other economic disruptions. By definition, the more economically integrated the two countries are, the larger E is.

A strategic defender, anticipating the possibility of war, may choose to reduce economic integration in advance. It can do so by raising tariffs and subsidizing domestic producers, restricting capital and intellectual-property flows, and erecting barriers to bilateral trade and investment. All these strategic barriers generate welfare losses, which are subsumed into a single variable.

Let C denote the decoupling cost, with 0 < C ≤ E. It represents the portion of the full economic cost of separation that is unilaterally chosen by the defender and incurred by both states, irrespective of whether conflict arises. If conflict does occur, the remaining (wartime) separation cost is E − C.

From the defender’s perspective, the decoupling cost C is the key control variable and plays an essential strategic role. Although C is sunk by the time conflict arises – and therefore does not directly affect the defender’s behavior during an attack – it indirectly shapes the defender’s incentives because it raises the opportunity cost of not defending. Specifically, without decoupling (C = 0) the defender might prefer “no defense” to “defense”, but prefers “defense to “no defense” once C is sufficiently large.

In case of war (defined as a situation in which the attacker attacks and the defender defends), the expected gain, U _A for the attacker and U _D for the defender, is given by: ⁴

$\begin{array}{c}\hfill U_A=(1-p)W-F-E\end{array}$

(1)

$\begin{array}{c}\hfill U_D=(1-p)(-L)-F-E.\end{array}$

(2)

The goal of each player is to maximize their payoff (or minimize losses).

The basic sequence of decisions is as follows:

Stage 1. State D chooses whether to decouple from State A. It can either refrain from decoupling by choosing C = 0, or initiate decoupling by choosing a positive value of C > 0.

Stage 2. State A decides whether to invade the territory. If it does not invade, the game ends. If it does, the other player, State D, must make another decision.

Stage 3. If, at stage 2, State A chooses to invade, State D decides whether to defend the territory. If State D does not defend, State A gains control of the territory. If State D defends, a military conflict ensues. The probability that State D prevails in the conflict is given by p.

Figure 1 presents the decision tree. The upper branches represent outcomes where the defender does not decouple (C = 0), while the lower branches correspond to situations in which the defender chooses to decouple (C > 0). The attacker can choose to attack (A) or refrain from attacking (NA, for no attack). The defender, in turn, can choose to defend (Df) or not defend (NDf, for no defense).

Figure 1.

Decision tree: Full information.

At each point in the game, decision makers choose the action that maximizes their expected utility from that point onward. They cannot commit to future actions but anticipate their opponents’ choices; hence, outcomes satisfy subgame perfection.

Equilibrium if Strategic Decoupling is Not an Option

Let us consider first a simplified context in which the defender has not the option to decouple. If the defender cannot implement a preemptive decoupling strategy (C > 0), the game tree is much simpler, since the attacker becomes the first mover (in Figure 1, it is represented by the upper branches, within the dotted-line area). If the attacker invades, the defender must chose whether to defend or not.

We first can show that:

Proposition 1

If the degree of integration of the two economies is large enough (E ≥ pL − F), peace is not an outcome. The Nash equilibrium is: State A: invades; State D: does not defend.

Proof 1

The proof proceeds by backward induction. The condition of no military response by State D is:

$\begin{array}{c}\hfill -L\ge (1-p)(-L)-F-E\end{array}$

(3)

$\begin{array}{c}\hfill \iff E\ge pL-F\end{array}$

(4)

Obviously, this condition holds irrespective of the integration level (E ≥ 0) if the cost of military action (F) is large (F > pL). To rule out a trivial outcome, in the following we assume that this is not the case, i.e., the threshold $\left(pL-F\right)$ is positive.

When condition (4) holds, State D does not defend the territory. Then, the optimal strategy for State A is to invade, because the payoff from invading is larger than the payoff from not invading:

$W>0.$

(5)

In the opposite case, if the two economies are not very integrated, i.e., E < pL − F, then State D should defend the territory. In turn, the aggressor would invade if (1 − p)W − F − E > 0 and would not invade if (1 − p)W − F − E < 0. A relatively trivial outcome occurs for (1 − p)W − F < E < pL − F: because it anticipates a military response, the potential aggressor does not invade the territory.

In the following, we focus on the non-trivial situation (high pre-conflict integration), and assume that condition (4) holds: E ≥ pL − F > 0.

“Peace Equilibrium” With Strategic Decoupling

We analyze now a situation in which the defender can implement pre-war decoupling and ask the question whether decoupling can prevent aggression by State A.

In the previous sub-section we have shown that if State D does not defend the territory, State A should invade it. So a necessary condition for A to hold-back is to have “defend” as the preferred strategy of State D. For State D, defense is the dominant strategy against invasion if:

$(1-p)(-L)-F-E>-L-C⟺C>\tilde{C}$

(6)

with:

$\tilde{C}=F+E-pL.$

(7)

Under condition (4), $\tilde{C}>0$ .

Moving backwards, in the context where the State D responds to aggression, it is optimal for State A not to invade if:

$(1-p)W-F-E<-C⟺C<\widehat{C}$

(8)

with:

$\widehat{C}=F+E-(1-p)W.$

(9)

Decoupling might be used by a defender to deter a potential aggressor if conditions (6) and (8) are fulfilled simultaneously. Any decoupling cost C such as: $\tilde{C}<C<\widehat{C}$ can be used as a deterrence device. A non-empty interval for C exists only if $\tilde{C}<\widehat{C}$ , or:

$p>\frac{W}{L+W}.$

(10)

Furthermore, under perfect information and no payoff uncertainty, the defender would pick the lowest cost C that ensures the emergence of the no-conflict equilibrium:

$\tilde{C}=F+E-pL$

(11)

Finally, we must check that for the defender it is worth paying the (lowest) cost of decoupling $\tilde{C}$ to avoid war, rather than not decoupling (C = 0), holding back against aggression and “simply” bearing the loss L (see Figure 1). Formally, this condition writes as:

$\tilde{C}=(F+E-pL)<L.$

(12)

To rule out trivial situations, for the remainder of this paper we assume that Conditions (10) and (12) both hold.

We can state that:

Proposition 2

If the necessary conditions

$p>\frac{W}{W+L}\text{and}\tilde{C}<L$

are satisfied, then costly preemptive decoupling by the defender constitutes part of a subgame perfect equilibrium strategy profile. In this equilibrium, the attacker is deterred, and the game results in a “peace equilibrium.” Moreover, when these conditions hold, the cost associated with the optimal level of decoupling is given by

$\tilde{C}=F+E-pL,$

as defined in equation (11).

Figure 2 summarizes the optimal strategies of the attacker at stage 2 and of the defender at stage 3 of the decoupling game, for any possible values of C as chosen by the defender at stage 1, with an upper limit at E. We represent the critical thresholds $\tilde{C}$ and $\widehat{C}$ , in a configuration where $\tilde{C}<L<\widehat{C}$ .

Figure 2.

Optimal decisions at stages 2 and 3 depending on the decoupling level C at stage 1.

We remind that at stage 1, a rational defender will not chose C bigger than L (thus the vertical dotted line). Indeed, if at stage 1 it chooses not to decouple (C = 0), his/her loss is L; yet, in any other configuration, the smallest loss is C.

The main result of this section is that, by voluntarily choosing to bear the irreversible cost of decoupling (sunk cost), the defender credibly commits to fight if conflict arises. In doing so, it can successfully deter the attacker from invading the territory, incurring a smaller loss than it would by abandoning the territory.

This is undoubtedly a strong result, but it depends on certain necessary conditions and initial assumptions. A key assumption underlying this result concerns the information each player has about the other’s goals. In the next section, we analyze the consequences of a minimal alteration to this assumption.

The Incomplete Information Framework

Main Assumptions

The context of our analysis in this section is similar to that of the previous section. As before, two states, A and D, share an integrated economy. At a certain point, State A adopts a bellicose stance and signals its willingness to challenge State D’s control over an unarmed territory. As already mentioned, the analysis focuses on situations where, in full information, there is a decoupling strategy that allows the defender to signal its readiness to fight (formally, both Conditions (10) and (12) are fulfilled).

In contrast to the previous model, we now assume that the strategic value of the unarmed territory for State D is private information to this player, thus it is unknown to the potential aggressor, State A. ⁵

To model this payoff uncertainty in a simple way, we assume that the defender can either attribute a high strategic importance to the territory, corresponding to a substantial loss L ^H (H-type defender) or a low strategic importance corresponding to a loss L ^L (L-type defender), with L ^H > L ^L . We also assume that the gap between L ^H and L ^L is relatively narrow. ⁶

Based on past interactions, State A assigns a probability μ to the event that it is facing a defender characterized by a loss L ^H , and a probability (1 − μ) to the event that it is facing a defender characterized by a loss L ^L . While the defender knows its true type, the attacker only knows the distribution of types, but not the actual type of defender it faces; this corresponds to a typical situation of “mistrust” in the classification of Acemoglu and Wolitzky (2024).

The basic sequence of decisions is the following:

Stage 1. At the outset of the game Nature decides about the type of the defender;

Stage 2. The defender decides on the optimal level of decoupling (it can be zero);

Stage 3. The attacker decides whether to invade or not the territory. If it does not invade, the game stops;

Stage 4. If at Stage 3 the attacker invaded the territory, the defender decides whether to engage the military conflict or not.

Figure 3 presents the decision tree of this game.

Figure 3.

Decision tree: Incomplete information.

At the outset of the game, Nature chooses randomly the type of the defender, which can be of the H-type with probability μ and of the L-type with probability (1 − μ).

Given its priorities, the defender must decide whether it implements economic decoupling to preempt conflict. As shown in the previous section, the optimal level of decoupling is defined in equation (11) as a decreasing function of the loss L incurred in case of invasion, $\tilde{C}(L^i)=E+F-p{L}^i$ , with i ∈ (H, L). In a perfect information setting, a high loss L ^H implies a low deterrence cost $\tilde{C}(L^H)$ and a low loss L ^L implies a high decoupling cost $\tilde{C}(L^L)$ , with $\tilde{C}(L^L)>\tilde{C}(L^H)$ . Because we assumed that L ^L and L ^H are close to each other, we can further consider that $\tilde{C}(L^L)\in [\tilde{C}(L^H),\widehat{C}]$ .

In the complete information framework in previous section we focused on a situation in which $L>\tilde{C}(L)$ , therefore, for the defender, the no-decoupling strategy (C = 0) is dominated by the decoupling strategy ( $\tilde{C}(L)>0$ ). To adopt a similar context of analysis, we now assume that the condition $L^L>\tilde{C}(L^L)$ is fulfilled, thus the L-type defender should never use the no-decoupling strategy. Furthermore, this condition also ensures that $L^H>\tilde{C}(L^H)$ , therefore the H-type defender neither will implement the no-decoupling strategy. ⁷

Subject to the threat of conflict, defenders must choose a strategy in the set $C\in \left\{0,\tilde{C}(L^H),\tilde{C}(L^L)\right\}$ . At the next step, given the level of decoupling implemented by State D, State A must choose a strategy in the set {Attack, No Attack}. Should State A invade the territory, the State D decides whether to defend or not the territory.

Recall that if the defender decides not to decouple (C = 0), it will not defend at the last stage (we assumed that condition (4) holds), thus the attacker should invade the territory. Therefore, in Figure 3, the no decoupling case entails the pair of (attacker; defender) payoffs (U _A = W; U _D = −L ⁱ ), with i ∈ (L, H).

A Perfect Bayesian Equilibrium (PBE) of the decoupling game is defined as a situation where each type of defender implements its optimal strategy given the attacker’s beliefs, and the attacker implements its optimal strategy based on correct beliefs about the type of defender.

Following the traditional method of analyzing games with incomplete information, let us first study first the existence of pure strategy equilibria (separating, pooling) and then hybrid (or mixed-strategy) equilibria.

Pure Strategy Equilibria

Separating Equilibrium

In a separating equilibrium, each type of defender adopts a distinct decoupling strategy which, once implemented, reveals its type to the attacker without uncertainty. Given this information, the attacker plays its optimal strategy (attack or not).

We analyze an equilibrium in which a defender who assigns a high value to the territory (L ^H ) implements a (low) decoupling level $\tilde{C}(L^H)$ , while a defender who assigns a low value (L ^L ) to the same territory, adopts a (high) decoupling level $\tilde{C}(L^L)$ . In turn, in a separating equilibrium these choices reveal the type of defender, thus the attacker should back-off since condition (8) holds.

Let us denote by s (L ⁱ ) the strategy played by a i − type defender.

Let us consider first the situation where, in equilibrium, each type of defender plays its perfect information strategy:

$s(L^i)=\left\{\begin{array}{ccc}\tilde{C}(L^L)\hfill & if\hfill & i=L\hfill \\ \tilde{C}(L^H)\hfill & if\hfill & i=H\hfill \end{array}\right..$

(13)

The on-path beliefs of the attacker can be represented as conditional probabilities in a standard way: $\mathrm{Pr}\left[H|\tilde{C}(L^L)\right]=0$ and $\mathrm{Pr}\left[H|\tilde{C}(L^H)\right]=1$ .

In this case, the level of decoupling perfectly reflects the type of the defender and, because the defender implements the (lowest) decoupling level that ensures deterrence, $\tilde{C}(L^i)=(F+E-p{L}^i)$ , the attacker finds it optimal to hold-back. The final (equilibrium) set of payoffs for the attacker and defender (U _A ; U _D ) would be $(-\tilde{C}(L^i);-\tilde{C}(L^i))$ , with i ∈ (L, H).

However, such an equilibrium can exists only if no defender has the incentive to deviate from the equilibrium strategy. It is obvious that a L-type defender has no incentive to play $\tilde{C}(L^L)$ . In line with the equilibrium beliefs, if such a defender chooses the (less expensive) decoupling level $\tilde{C}(L^H)$ , the attacker should infer that this defender is of the H-type and refrain from invading, leading to the defender’s payoffs $-\tilde{C}(L^H)$ . Because $\tilde{C}(L^H)<\tilde{C}(L^L)$ , the strategy $\tilde{C}(L^L)$ is suboptimal for a L-type state. This separating equilibrium does not exist.

It can also be shown that a less intuitive separating equilibrium in which a L-type defender plays $\tilde{C}(L^H)$ and a H-type defender plays $\tilde{C}(L^L)$ cannot exist either. In such a context, the L-type defender would be attacked systematically. He would be better-off if choosing $\tilde{C}(L^L)$ .

As a conclusion, the decoupling game with incomplete information presents no separating equilibrium. This outcome comes with no surprise: because there is no interaction between C and L, the single-crossing condition (Spence 1973; Mirrlees 1971) is not satisfied.

Pooling at the Low Level of Decoupling

Let us now consider the feasibility of a pooling equilibrium in which – regardless of its type – the defender implements the low level of decoupling, $\tilde{C}(L^H)$ , while the attacker plays its optimal strategy.

$s(L^i)=\left\{\begin{array}{ccc}\tilde{C}(L^H)\hfill & if\hfill & i=L\hfill \\ \tilde{C}(L^H)\hfill & if\hfill & i=H\hfill \end{array}\right..$

(14)

In such an equilibrium, decoupling is not an informative signal, thus the attacker’s posterior beliefs correspond to its priors. Specifically, after observing the decoupling strategy $\tilde{C}(L^H)$ , it assigns the probability μ to facing the H-type defender and the probability (1 − μ) to facing the L-type defender. The posterior beliefs can be written as conditional probabilities: $\mathrm{Pr}\left[H|\tilde{C}(L^H)\right]=\mu$ .

To fully characterize the structure of beliefs, we must define the off-path beliefs.

Case 1

Let us first assume that if any state deviates and chooses the large decoupling level $\tilde{C}(L^L)$ it would be perceived as L-type state, i.e.,: $\mathrm{Pr}\left[H|\tilde{C}(L^L)\right]=0$ .

At the low level of decoupling $C=\tilde{C}(L^H)$ , an attacker who considers invading the territory expects that the H-type defender will fight for it, and that the L-type will hold-back because the existing decoupling level is below its “defend threshold”, $\tilde{C}(L^H)<\tilde{C}(L^L)$ . The attacker’s expected payoffs associated to each strategy (Attack and No Attack) are:

$\begin{array}{c}\hfill \mathbf{E}[U_A|\text{Attack},\tilde{C}(L^H)]=\mu \left[-E-F+(1-p)W\right]+(1-\mu )\left(W-\tilde{C}(L^H)\right)\end{array}$

(15)

$\begin{array}{c}\hfill \mathbf{E}[U_A|\text{NoAttack,}\tilde{C}(L^H)]=-\tilde{C}(L^H).\end{array}$

(16)

Therefore the attacker will not invade if:

$\mathbf{E}[U_A|\text{NoAttack},\tilde{C}(L^H)]>\mathbf{E}[U_A|\text{\hspace{0.17em}Attack,}\tilde{C}(L^H)]\iff \mu p>\frac{W}{W+L^H},$

(17)

and invade in the opposite case.

When condition (17) is fulfilled, the attacker holds-back and, regardless of the type of defender, the set of payoffs (U _A ; U _D ) is $(-\tilde{C}(L^H);-\tilde{C}(L^H))$ .

In this case, it is straightforward to demonstrate that no defender has an incentive to deviate from the equilibrium strategy. Indeed, given the assumption regarding out-of-equilibrium beliefs, a defender who deviates by choosing the higher decoupling level $\tilde{C}(L^L)$ would be perceived as a type-L state. Since $\tilde{C}(L^L)>\tilde{C}(L^H)$ this choice would indeed deter aggression. However, the same outcome (no attack) is achieved in the pooling equilibrium at a lower cost, since $\tilde{C}(L^H)<\tilde{C}(L^L)$ . Therefore, no defender has an incentive to deviate from the pooling equilibrium.

Case 2

It can be shown that this pooling equilibrium also exists under the alternative off-path beliefs. If a defender who implements a high level of decoupling $\tilde{C}(L^L)$ is perceived as a H-type state, a defender that deviates and implements the decoupling level $\tilde{C}(L^L)$ does deter invasion. The final payoff of this state would be $-\tilde{C}(L^L)$ which is nonetheless lower than the equilibrium payoff $-\tilde{C}(H)$ .

Because the pooling outcome at $\tilde{C}(L^H)$ is sequentially rational for every defender type for a broad class of off-path beliefs, we describe this PBE as robust to off-path belief specifications.

We summarize our findings as:

Proposition 3

A PBE with pooling on the low decoupling level $\tilde{C}(L^H)$ and no invasion exists under the necessary condition μp > W/(W + L ^H ).

In the opposite case where μp < W/(W + L ^H ), the attacker should invade the territory; in turn, the H-type defender would engage in conflict while the L-type would hold back. However, in this case a L-type defender which would have chosen the no-decoupling strategy would obtain the payoff −L ^L , which is less expensive than the decoupling strategy $\tilde{C}(L^H)$ entailing the final payoff $-L^L-\tilde{C}(L^H)$ ; the decoupling cost is wasted when invasion is inevitable. Because deviation from equilibrium is optimal for this state, we conclude that this pooling equilibrium is impossible for a relatively low product μp.

Pooling at the High Level of Decoupling

We analyze the feasibility of a pooling equilibrium where, regardless of its type, the defender implements the high level of decoupling $\tilde{C}(L^L)$ . In this pooling situation any of these defenders would fight in case of invasion and therefore invasion is suboptimal for the attacker.

Similar to the former analysis, the decoupling strategy is not informative, thus the attacker’s posterior beliefs match its priors, $\mathrm{Pr}\left[H|\tilde{C}(L^L)\right]=\mu$ .

The question of the off-paths beliefs is essential when analyzing this equilibrium.

Deviation Signals a H-Type Defender

Let us first consider that if a defender deviates and chooses the low level of decoupling $\tilde{C}(L^H)$ , the attacker assumes that it is of the H-type, $\mathrm{Pr}\left[H|\tilde{C}(L^H)\right]=1$ .

If a defender which deviates to $\tilde{C}(L^H)$ is recognized as a H-type state, this low level of decoupling is sufficient to deter aggression. Because this strategy is less expensive, $\tilde{C}(L^H)<\tilde{C}(L^L)$ , this defender has an incentive to deviate from the equilibrium strategy; the pooling equilibrium at the high decoupling level does not exist for these off-paths beliefs.

Deviation Signals a L-Type Defender

Let us consider the alternative off-path beliefs: if a defender deviates to the low level of decoupling the attacker assumes that it is of the L-type: $\mathrm{Pr}\left[H|\tilde{C}(L^H)\right]=0$ .

In this case, the attacker will systematically aggress a defender which deviates from the equilibrium level, since the decoupling level $\tilde{C}(L^H)$ is below the level that prompts military action by a L-type defender $\tilde{C}(L^L)$ . We analyze whether defenders, depending on their type, have an incentive to deviate from the equilibrium strategy:

A H-type defender who deviates is subject to aggression, yet it would fight: it obtains the payoff −(F + E + (1 − p)L ^H ) which is lower than the equilibrium payoff $\tilde{C}(L^L)$ , thus it has no incentive to deviate.

A L-type defender implementing $\tilde{C}(L^H)$ is attacked, and would not fight, thus obtaining payoff $-L^L-\tilde{C}(L^H)=-(F+E+L^L-p{L}^H)$ . Deviation is suboptimal iff

$-(E+F+L^L-p{L}^H)\le -(E+F-p{L}^L)⟺(1+p)L^L\ge p{L}^H.$

This condition is fulfilled for L ^L close to L ^H as assumed. In the opposite case, this equilibrium does not exist.

Comparing the above-mentioned “deviation payoffs” we observe that the H-type defender which deviates from the equilibrium strategy has more to lose than the L-type defender: (F + E + (1 − p)L ^H ) > (F + E + L ^L − pL ^H ). Divinity/D1 criterion (Banks and Sobel 1987) prescribes shifting beliefs toward the L-type after this deviation (the deviation to $\tilde{C}(L^H)$ is more attractive for the L-type than for the H-type), which is exactly the off-path belief used above.

We summarize our findings in:

Proposition 4

A PBE with pooling on the high decoupling level $\tilde{C}(L^L)$ and no invasion exists if the gap between L ^H and L ^L is small (if (1 + p)L ^L ≥ pL ^H ).

Hybrid PBE

In a hybrid equilibrium, at least one player finds it optimal to implement a mixed strategy. In Online Appendix A we show that a hybrid equilibrium in which the attacker has a pure strategy (either Attack or No Attack) while the defender implements a mixed strategy is impossible. We therefore consider hybrid PBE in which:

(1) At least on type of defender randomly chooses between the two possible levels of decoupling;

(2) The attacker implements a mixed strategy; more specifically, the attacker which observes the low level of decoupling $\tilde{C}(L^H)$ chooses at random between attacking (A) or not (NA). On the other hand, if the attacker observes the high level of decoupling, $\tilde{C}(L^L)$ , it can only hold back because the high level of decoupling ensures the military response of the defender in case of the attack.

Hybrid PBE 1: Type-L Defender Randomizes

Let us first study a situation in which:

(1) A defender of type H plays $\tilde{C}(L^H)$ with probability 1 and a defender of type L randomizes between $\tilde{C}(L^H)$ and $\tilde{C}(L^L)$ , playing $\tilde{C}(L^H)$ with probability (1 − α) ∈ (0, 1) and $\tilde{C}(L^L)$ with probability α.

$s(L^i)=\left\{\begin{array}{ccc}\alpha \tilde{C}(L^L)+(1-\alpha )\tilde{C}(L^H)\hfill & if\hfill & i=L\hfill \\ \tilde{C}(L^H)\hfill & if\hfill & i=H\hfill \end{array}\right..$

(18)

• (2) The attacker attacks after observing $\tilde{C}(L^L)$ . After observing $\tilde{C}(L^H)$ it attacks with probability v ∈ (0, 1).

Beliefs

By Bayes’ rule, after observing $\tilde{C}(L^H)$ the posterior probability of facing a type H defender, $\mathrm{Pr}\left[H|\tilde{C}(L^H)\right]={\mu }^{\prime }$ , is simply:

${\mu }^{\prime }=\frac{\mu \cdot 1}{\mu \cdot 1+(1-\mu )(1-\alpha )}.$

(19)

After observing C (L ^L ), the posterior belief is $\mathrm{Pr}(H\mid \tilde{C}(L^L))=0$ , consistent with the equilibrium actions, as only the type L can choose $\tilde{C}(L^L)$ .

Defender of L-Type’s Indifference Condition

If the L-type plays $\tilde{C}(L^H)$ and faces attack with probability v, it will not fight, yielding expected payoff:

$U_{DL}(\tilde{C}(L^H))=-\tilde{C}(L^H)-v{L}^L.$

(20)

If it plays $\tilde{C}(L^L)$ , the attacker does not attack, yielding payoff:

$U_{DL}(\tilde{C}(L^L))=-\tilde{C}(L^L).$

(21)

Indifference requires $U_{DL}(\tilde{C}(L^H))=U_{DL}(\tilde{C}(L^L))$ , which gives:

$v_1=\frac{\tilde{C}(L^L)-\tilde{C}(L^H)}{L^L}=\frac{p(L^H-L^L)}{L^L}>0.$

(22)

The existence of the equilibrium requires that v₁ < 1 ⇔ p < L ^L /(L ^H − L ^L ). This condition is fulfilled if L ^L is close to L ^H , as formerly assumed.

Attacker’s Indifference Condition

After observing $\tilde{C}(L^H)$ , the attacker assigns the probability μ′ to face a type-H defender which fights, and a probability (1 − μ′) to the event of facing a L − type defender that does not. The expected payoff from attacking is:

$\mathbf{E}[U_A\mid \text{Attack},\tilde{C}(L^H)]={\mu }^{\prime }[-E-F+(1-p)W]+(1-{\mu }^{\prime })(W-\tilde{C}(L^H)).$

(23)

The payoff from not attacking is:

$\mathbf{E}[U_A\mid \text{NoAttack},\tilde{C}(L^H)]=-\tilde{C}(L^H).$

(24)

Indifference requires:

$\begin{array}{c}\hfill {\mu }^{\prime }[-E-F+(1-p)W]+(1-{\mu }^{\prime })(W-\tilde{C}(L^H))=-\tilde{C}(L^H)\\ \hfill {\mu }^{\prime }[E+F+pW-\tilde{C}(L^H)]=W\\ \hfill p{\mu }^{\prime }\left(W+L^H\right)=W\end{array}$

${\mu }^{\prime }p=\frac{W}{W+L^H}.$

(25)

Substituting the Bayes posterior (equation (19)) into (25) yields:

$\frac{\mu }{\mu +(1-\mu )(1-\alpha )}=\frac{W}{p(W+L^H)},$

(26)

which gives the randomization probability:

$(1-\alpha )=\frac{\mu }{(1-\mu )}\frac{p(W+L^H)-W}{W}.$

(27)

If (1 − α) ∈ (0, 1), the above equation defines a valid hybrid PBE.

We check that (1 − α) > 0 is equivalent to $p\ge \frac{W}{W+L^H}$ , which is fulfilled by our Condition (10).

Condition (1 − α) < 1 is fulfilled for μp < W/(W + L ^H ).

We conclude this subsection by:

Proposition 5

For $\mu p<\frac{W}{W+L^H}$ , the decoupling game admits a hybrid PBE in which a type-H defender always plays $\tilde{C}(L^H)$ , type L randomizes between $\tilde{C}(L^H)$ and $\tilde{C}(L^L)$ , and the attacker mixes after $\tilde{C}(L^H)$ . The mixing probabilities are given by:

$v_1=\frac{p(L^H-L^L)}{L^L},\text{and}(1-\alpha )=\frac{\mu }{(1-\mu )}\frac{p(W+L^H)-W}{W},$

(28)

provided that W/(W + L ^H ) < p < L ^L /(L ^H − L ^L ).

Hybrid PBE 2: Type-H Defender Randomizes

We study now an equilibrium in which:

(1) The defender of type H randomizes between $\tilde{C}(L^L)$ and $\tilde{C}(L^H)$ , playing $\tilde{C}(L^H)$ with probability (1 − β) ∈ (0, 1) and $\tilde{C}(L^L)$ with probability β. The defender of type L plays $\tilde{C}(L^H)$ with probability 1.

$s(L^i)=\left\{\begin{array}{ccc}\tilde{C}(L^L)\hfill & if\hfill & i=L\hfill \\ \beta \tilde{C}(L^L)+(1-\beta )\tilde{C}(L^H)\hfill & if\hfill & i=H\hfill \end{array}\right..$

(29)

• (2) The attacker never attacks after observing $\tilde{C}(L^L)$ . After observing $\tilde{C}(L^H)$ , it attacks with probability v ∈ (0, 1).

Beliefs

In this context, by Bayes’ rule, after observing $\tilde{C}(L^H)$ the posterior probability of facing type H, $\mathrm{Pr}\left[H\mid \tilde{C}(L^H)\right]={\mu }^{\prime }$ is:

${\mu }^{\prime }=\frac{\mu \cdot (1-\beta )}{\mu \cdot (1-\beta )+(1-\mu )}.$

(30)

After observing $\tilde{C}(L^L)$ , the posterior belief is $\mathrm{Pr}(H\mid \tilde{C}(L^L))=1$ , consistent with the assumption that in this equilibrium only an H-type defender can send the costly signal $\tilde{C}(L^L)$ .

Defender of H-Type’s Indifference Condition

If the H-type plays $\tilde{C}(L^H)$ , in case attack (with probability v) it will fight; its expected payoff is:

$U_{DH}(\tilde{C}(L^H))=(1-v)[-\tilde{C}(L^H)]+v[(1-p)(-L^H)-F-E].$

(31)

If it plays $\tilde{C}(L^L)$ , the attacker does not attack, yielding payoff:

$U_{DH}(\tilde{C}(L^L))=-\tilde{C}(L^L).$

(32)

Indifference requires $U_{DH}(\tilde{C}(L^H))=U_{DH}(\tilde{C}(L^L))$ , which gives the equilibrium attack probability:

$v_2=\frac{p(L^H-L^L)}{L^H}.$

(33)

The existence of the equilibrium requires that v₂ < 1 ⇔ p < L ^H /(L ^H − L ^L ), a condition that is obviously fulfilled since L ^H /(L ^H − L ^L ) > 1.

Attacker’s Indifference Condition

The attacker’s decision is similar to the problem analyzed in the previous subsection. As already mentioned, after observing $\tilde{C}(L^H)$ , the attacker assigns the probability μ′ to face a type-H defender which fights, and a probability (1 − μ′) to the event of facing a L − type defender that does not. The expected payoff from attacking is:

$\mathbf{E}[U_A\mid \text{Attack},\tilde{C}(L^H)]={\mu }^{\prime }[-E-F+(1-p)W]+(1-{\mu }^{\prime })(W-\tilde{C}(L^H)).$

(34)

The payoff from not attacking is:

$\mathbf{E}[U_A\mid \text{NoAttack},\tilde{C}(L^H)]=-\tilde{C}(L^H).$

(35)

The indifference equation is identical to equation (25), of course, with a different posterior probability;

${\mu }^{\prime }p=\frac{W}{W+L^H}.$

(36)

Substituting the Bayes posterior equation (30) in this equation, yields:

$\frac{\mu (1-\beta )}{\mu (1-\beta )+(1-\mu )}=\frac{W}{p(W+L^H)},$

(37)

which gives the probability of the H-type defender to play $\tilde{C}(L^H)$ :

$(1-\beta )=\frac{(1-\mu )}{\mu }\frac{W}{p(W+L^H)-W}.$

(38)

If (1 − β) ∈ (0, 1), the above equation defines a valid hybrid PBE.

We check that (1 − β) > 0 is equivalent to $p\ge \frac{W}{W+L^H}$ , which is fulfilled by our Condition (10).

Condition (1 − β) < 1 holds for $\mu p>\frac{W}{W+L^H}$ .

To sum up,

Proposition 6

For $\mu p>\frac{W}{W+L^H}$ , the decoupling game admits a hybrid PBE in which the type-L defender always plays $\tilde{C}(L^H)$ , type H randomizes between $\tilde{C}(L^H)$ and $\tilde{C}(L^L)$ , and the attacker randomizes after $\tilde{C}(L^H)$ . The mixing probabilities are given by:

$v_2=\frac{p(L^H-L^L)}{L^H},\text{and}(1-\beta )=\frac{(1-\mu )}{\mu }\frac{W}{p(W+L^H)-W},$

(39)

provided that $p\ge \frac{W}{W+L^H}$ .

To conclude the section on hybrid equilibria, we remark that an equilibrium in which both type of defenders simultaneously randomize between the low decoupling level $\tilde{C}(L^H)$ and the high decoupling level $\tilde{C}(L^L)$ , and the attacker attacks with a probability v does not exists. Indeed, defenders indifference conditions cannot be fulfilled simultaneously, since v₁ ≠ v₂.

Regions of Equilibria

To summarize the former findings:

For $\mu p<\frac{W}{W+L^H}$ , the game presents a Hybrid PBE 1 in which the L-type defender randomizes between $\tilde{C}(L^H)$ and $\tilde{C}(L^L)$ , while the H-type defender always plays $\tilde{C}(L^H)$ . The attacker responds by randomizing after observing $\tilde{C}(L^H)$ , with probabilities characterized in Proposition 5.

For $\mu p>\frac{W}{W+L^H}$ the game presents:

 ◦ A Pooling equilibrium at the low decoupling level in which both types play $\tilde{C}(L^H)$ , and the attacker refrains from invading. This equilibrium is robust, since deviation is unprofitable for both types under a broad set of off-path beliefs. It guarantees peace at the lowest deterrence cost (Proposition 3)

 ◦ A Hybrid PBE 2 in which the L-type defender always plays $\tilde{C}(L^H)$ , while the H-type defender randomizes between $\tilde{C}(L^H)$ and $\tilde{C}(L^L)$ . The attacker then randomizes after observing $\tilde{C}(L^H)$ , with probabilities given in Proposition 6.

A Pooling equilibrium at the high decoupling level $\tilde{C}(L^L)$ exists only under special (punitive) off-path beliefs and provided that losses are sufficiently close, to ensure that (1 + p)L ^L ≥ pL ^H .

Figure 4 illustrates these regions in the (p, μ) space. In line with Condition (10), economic decoupling is a meaningful strategy for the defender only for $p>\frac{W}{W+L^H}$ , which is represented by the vertical line.

Figure 4.

Regions of equilibria of the game with incomplete information.

Multiple equilibria arise for the same parameter values. Relative to the complete information case, peace is no longer guaranteed: even when the “robust” pooling equilibrium exists (for μp > W/(W + L ^H )), it coexists with a hybrid equilibrium in which aggression occurs with positive probability. This highlights the inherent instability of the game, whereby a shift from a “good” equilibrium (pooling) to a “bad” equilibrium (hybrid with conflict) may be triggered by geopolitical shocks that deteriorate the attacker’s beliefs.

The likelihood that the most attractive peace equilibrium of “Pooling at the low decoupling level” prevails increases with the probability p that the defender wins in a conflict. Conversely, a shift in the balance of power in favor of the attacker (declining p) may move the game’s outcome toward Hybrid PBE 1 or the less stable “Pooling at the high decoupling level”. In practice, this implies that states should continuously monitor the relative military strength of their opponents to anticipate potential shifts in equilibrium.

The theoretical analysis provides a rationale for a defender to pursue economic decoupling, even when such a strategy entails significant welfare costs. Yet under uncertainty about the defender’s objectives, decoupling may not reliably secure deterrence. These conclusions rest on simplifying assumptions and must therefore be extrapolated to real-world contexts with caution. With this caveat, it is worth asking whether the mechanism identified in our game-theoretic framework has relevance for contemporary geopolitical tensions.

A Policy Discussion: Strategic Decoupling and U.S.–China Relations

Belief in the virtues of free trade, both as a driver of prosperity and as a guarantor of peace, underpinned the West’s open stance toward China after the mid-1970s. ⁸ Following its accession to the WTO in 2001, China’s exports surged as the country integrated into global supply chains, initially leveraging low wages. While China did reduce many trade barriers, it retained restrictions on imports and foreign direct investment and provided state support to targeted sectors. Despite these frictions, Western firms deepened their investments in China, both to access its domestic market and to use it as a manufacturing base for global exports.

With its large and relatively low-wage labor force, flexible labor market, and improving infrastructure, China soon emerged as the “world’s factory,” producing more goods than the next nine largest manufacturing nations combined (Baldwin 2024). Its share of global manufacturing exports rose from just 3 percent in 1995 to 20 percent by 2020.

Recent data suggest, however, that this integration created asymmetric dependencies. According to evidence from The Economist based on the MERICS Trade Dependency Database, U.S. (EU) dependence on Chinese imports rose from less than 2 percent in 2000 to 7.5 percent (5.5 percent) in 2022, while China’s dependence on imports from the U.S. and the EU declined to below 2 percent. ⁹ Moreover, China now dominates the supply of key inputs, from electric-vehicle batteries to critical raw materials such as gallium, graphite, germanium, and manganese, which are resources with few, if any, substitutes in the production of advanced technologies, including weapons systems.

Since 2018, new geopolitical tensions have further altered the calculus. Breaking with the longstanding doctrine of “Peaceful Rise” promoted by his predecessors, Xi Jinping consolidated power as President for life of the People’s Republic of China (PRC). He has also systematically signaled Beijing’s intention to reunify with Taiwan, including through the possible use of force. The PRC has accelerated the modernization of its armed forces, and in several areas it now appears to outpace U.S. capabilities in the Pacific. ¹⁰ Chinese military drills around Taiwan have multiplied, increasing in scale and sophistication. ¹¹ In response, the 2022 U.S. National Defense Strategy identified dissuading the PRC from using aggression to threaten vital U.S. interests as a central objective (US-DoD 2022), an objective reaffirmed in the U.S. National Security Strategy of 2025 (White House 2025).

In light of our analysis, the economic decoupling strategy initiated by the United States is no surprise. Elected as the 45th President of the U.S. in 2017, Donald Trump appears as the most prominent spokesperson for the renewed anti-globalization trend. Furthermore, the second Trump administration has downplayed traditional economic arguments for protectionism, instead emphasizing the strategic goals of re-shoring, re-industrialization, and fair trade as vital to the national interest. ¹²

During his first term (2017–2021), President Donald Trump imposed tariffs on China exports that amounted to 16.2 percentage point increase on average. His successor, President Joe Biden, maintained these tariffs and reinforced this protectionist policy with domestic subsidies and export controls on strategic goods, most notably advanced microprocessors (Goldberg and Reed 2023). These measures proved effective: the share of Chinese imports in total U.S. imports fell from 22 percent in 2018 to less than 14 percent in 2023 (Freund et al. 2024; Irwin 2025). Over time, additional export bans targeted Chinese firms in sensitive sectors (Chorzempa et al. 2024). Re-elected in 2024, President Trump intensified this decoupling strategy, raising tariffs on Chinese exports by a further 26.8 percent points above their 2024 level (by December 2025) and imposing administrative restrictions on exports of strategic technologies to China. By December 2025, average US tariffs on Chinese exports stand at 47.5 percent, and cover all goods. ¹³ When in 2025 China imposed restrictions on the supply of rare earths, the United States federal government responded with a set of policies aimed at supporting domestic production, expanding national stockpiles, and diversifying and securing imports.

Taken together, these protectionist policies have triggered a substantial, largely unobserved restructuring process, with firms in both countries attempting to bypass official restrictions (for instance, through trade via “straw-man” intermediaries), replace China with trade partners from allied countries, and engage in substantial reshoring (Witt et al. 2023). These adjustments allow firms to offset some, but not all, of the costs associated with decoupling.

Our theoretical framework provides one possible strategic rationale for the recent U.S. policy shift: by pursuing rapid economic decoupling, the United States may seek to render credible the threat of a military response should China use force against U.S. national interests, thereby sustaining effective deterrence. As is often the case in practice, the concept of national interest is not defined with utmost precision in official statements and policy documents. For example, successive U.S. administrations have deliberately cultivated a degree of “strategic ambiguity” regarding their position with respect to Taiwan. Our analysis of the incomplete-information game shows that, as long as the true strategic value of Taiwan to the United States remains private information, the no-conflict pooling equilibrium is no longer unique. In particular, if geopolitical shocks deteriorate the aggressor’s beliefs, the game may spontaneously shift from a “good” equilibrium (pooling with no aggression) to a hybrid equilibrium in which conflict arises with positive probability.

That said, the model is likely too simple to provide a fully satisfactory characterization of the strategic environment surrounding Taiwan. Taiwan is not an unarmed territory, as assumed in our framework; on the contrary, it holds significant defense capabilities and has now undergone an extensive military modernization program with U.S. support. Other countries in the region, such as Japan, might also provide assistance in the event of conflict. On the other hand, China’s acceleration in building offensive capabilities, as well as its alliances with states hostile to the existing world order, represents an additional source of uncertainty not captured by the model. This evolving balance of power may move the system away from a stable pooling equilibrium at low levels of decoupling toward less stable configurations in which peace can no longer be taken for granted.

Conclusion

Recent geopolitical tensions among major economic powers have challenged the efficiency-based rationale for free trade on grounds of national security (Antràs 2024; Baldwin 2025). On the one hand, highly integrated economies may be less prone to conflict, as the costs of war increase with interdependence. On the other hand, the logic may invert: a revisionist state planning for war can deliberately deepen integration to “tie the hands” of its opponent (Fearon 1997; Hirschman 1945). In such circumstances, discretionary economic decoupling may serve as a strategic signal of readiness to confront aggression.

Contemporary real-world developments appear to be aligned with this mechanism. Since 2017, the United States has pursued a resolute decoupling strategy from China, moving away from the traditional efficiency-based rationale of gains from trade. This policy shift coincides with China’s increasingly assertive posture toward Taiwan, alongside efforts widely interpreted as aiming to expand its strategic influence in the Indo-Pacific. Similarly, the European Union’s heavy dependence on Russian gas and oil before 2022 became a liability that President Vladimir Putin likely expected to deter support for Ukraine.

This paper develops an Attacker–Defender game to formalize how economic decoupling can serve as a deterrence mechanism. The logic follows the “credible threat” paradigm of Schelling (1960): a threat becomes credible if its costs are incurred in advance. By absorbing the costs of decoupling before conflict materializes, the defender signals its readiness for war, potentially deterring aggression altogether. We refer to this outcome as a “peace equilibrium.”

Analysis of the complete information game highlights the necessary conditions for this mechanism to operate. At the outset, integration must be neither too shallow nor too deep. Within this range, we identify an optimal level of decoupling that ensures effective deterrence – and show that it depends on key parameters such as the cost of military conflict.

The model is then extended to incomplete information by allowing uncertainty about the defender’s valuation of the contested territory. Depending on this valuation, the defender may be of two types. In this setting, the game admits various Perfect Bayesian Equilibria: two pooling equilibria, in which deterrence is sustained, and two hybrid equilibria, in which the attacker randomizes between attacking and holding back, while one type of defender randomizes between resisting and refraining.

This suggests a cautionary lesson for policymakers: even carefully designed economic decoupling cannot fully eliminate the risk of conflict if key strategic values remain private or uncertain. Maintaining credible deterrence requires not only policy actions but also transparency aiming at shaping the aggressor’s beliefs, since shifts in those beliefs can trigger endogenous transitions from “safe” to riskier equilibria.

Several avenues for future research remain. Our analysis relies on a stylized setting, and extensions could incorporate active resistance by the targeted territory or strategic interaction between rival superpowers simultaneously seeking control of a resource while deterring one another. Another extension would allow for reciprocal incomplete information, with both attacker and defender holding private information about preferences and capabilities, or about their respective chances of prevailing in conflict. Finally, economic decoupling is modeled as a pure sunk cost used by the defender as a commitment device; allowing such an investment to also shift the balance of power in favor of the defender may restore separation under incomplete information.

Despite the simplicity of the model, our results shed light on why economic decoupling can be a strategically meaningful response in today’s international environment. At the same time, they underscore the difficulty policymakers face in identifying the optimal degree of decoupling and the risks of miscalibration in either direction.

Supplemental Material

Supplemental Material - Economic Decoupling as a Readiness Signaling Device

Supplemental Material for “Economic Decoupling as a Readiness Signaling Device” by Damien Besancenot and Radu Vranceanu inThe Journal of Conflict Resolution