🔢 mathematics

The Trust-Building Game: Dynamic Signaling, Markov Equilibrium, and Threshold Cooperation

This paper develops a discounted dynamic Trust-Building Game to characterize a pure stationary Markov equilibrium, demonstrating that the trust builder's optimal signaling policy follows a unique activation-band structure over intermediate public states rather than a simple monotone threshold, while providing exact analytical bounds and numerical validations for cooperation dynamics.

Original authors: Madjid Eshaghi Gordji, Mohamadali Berahman, Zeinab Shahini

Published 2026-09-30

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Madjid Eshaghi Gordji, Mohamadali Berahman, Zeinab Shahini

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Trust is not a single moment of faith; it is a story told over time. In the real world, we decide whether to rely on someone based on what we have seen them do before. If a friend has kept their promises in the past, we are more likely to trust them with a new secret. If they have let us down, we hesitate. This simple idea—that our history shapes our future choices—is the foundation of a new mathematical model developed by researchers at Semnan University in Iran. They wanted to understand the precise mechanics of how trust builds up, how it can be lost, and, most surprisingly, when it is actually smart to stop trying to prove you are trustworthy.

The researchers focused on a specific type of interaction: a repeated game between two people. One person, the "builder," wants to earn the trust of the other, the "trustee." The builder can choose to send a signal, a visible action that says, "I am being cooperative." This action costs the builder something, like time or effort. The trustee watches these actions and keeps a running score of the builder's behavior. This score is not a perfect memory of every single event, but rather a weighted summary where recent actions matter more than old ones, though past actions still linger. The trustee has a clear rule: if the score is high enough, they will cooperate; if it is too low, they will not.

For a long time, economists and psychologists have assumed that the path to trust is straightforward: the more you try to prove yourself, the more trust you get. The natural assumption is that a person should keep signaling their good intentions as long as they want the other person to cooperate. The new study challenges this intuition directly. By using a rigorous mathematical framework to simulate thousands of interactions, the researchers discovered that the optimal strategy for the trust builder is often not to signal constantly, nor to give up entirely. Instead, the best strategy is to signal only during a specific, middle range of trust scores.

The study reveals a counterintuitive pattern called an "activation band." Imagine the public trust score as a dial that can be turned from zero to one. The researchers found that if the dial is very low, it is often too expensive for the builder to try to turn it up; the cost of the signal is too high compared to the small chance of immediate reward. If the dial is already very high, the builder does not need to signal at all because the trustee is already cooperating, and the builder can save their effort. The sweet spot, where signaling is most valuable, is in the middle. Here, the builder's action is just enough to push the score over the trustee's threshold, triggering cooperation immediately while also building a buffer for the future.

This finding overturns the idea that trust-building is a simple, one-way street where more effort always equals more trust. The researchers proved mathematically that under certain conditions, the smartest move is to signal only when the trust score is in that intermediate zone. If the score is too low, the builder waits. If the score is too high, the builder rests. They only act when the situation is just right. This behavior is not a mistake or a lack of effort; it is a calculated response to the costs of signaling and the persistence of the trust score.

The study also clarified how trust decays. When the builder stops sending positive signals, the trust score does not vanish instantly. It fades away slowly, like a sound fading in a large room, because the model gives weight to past actions. This means that once trust is established, it can survive for a while even if the builder stops trying to prove themselves. The researchers calculated exactly how long this survival lasts and how many periods of cooperation a builder can expect before the score drops too low again. They also showed that the moment a builder decides to signal is different from the moment the trustee decides to cooperate, a subtle timing detail that changes the entire strategy.

To ensure their mathematical conclusions were correct, the team ran detailed computer simulations. They tested their theory against a wide range of scenarios, changing the cost of signaling and how much the players valued future rewards. In every case where their specific mathematical conditions were met, the computer simulations matched their theory perfectly. The builder's strategy always settled into that specific band of activity. When they looked at scenarios outside those strict conditions, the strategy changed, but it never returned to the simple idea of "always signal when you want trust."

The work is a purely theoretical exploration, meaning it uses mathematical logic and computer calculations rather than experiments with human volunteers. The researchers were careful to state that their model is a simplified version of reality. They did not try to measure actual human psychology or use real-world data to calibrate their numbers. Instead, they built a clean, logical machine to see what happens when you strip away the noise of human emotion and focus only on the incentives. The result is a clear picture of how incentives shape behavior in a world where trust is built one step at a time.

The implications of this work go beyond simple games. It suggests that in any situation where reputation matters, from business partnerships to international diplomacy, the most effective strategy might not be constant effort. Sometimes, the best way to maintain trust is to know when to stop trying to prove yourself and let your past actions do the work. Conversely, when trust is low, it might be wiser to wait for the right moment to act rather than wasting resources on futile attempts to convince a skeptical partner. The study provides a precise map for these decisions, showing that the path to trust is not a straight line, but a carefully chosen path through a landscape of costs and rewards.

Technical Summary: The Trust-Building Game: Dynamic Signaling, Markov Equilibrium, and Threshold Cooperation

Problem Formulation
This paper addresses the strategic dynamics of trust in repeated interactions, specifically focusing on how a "trust builder" (Player 1) can influence a "trustee" (Player 2) through observable signaling. The authors formulate a discounted infinite-horizon dynamic game where the interaction is governed by a public state variable, $T_t \in [0, 1]$ , which summarizes the history of observable signals. Unlike standard reputation models relying on Bayesian learning about hidden types, this model treats the public state as a reduced-form, deterministic score updated via a persistence-weighted recursion: $T_{t+1} = \lambda T_t + (1-\lambda)s_t$ , where $s_t \in \{0, 1\}$ is the builder's binary signal and $\lambda \in [0, 1)$ is a persistence parameter.

The trustee observes the updated state $\tilde{T}_t$ (post-signal) and decides whether to cooperate ( $a_t=1$ ) or not ( $a_t=0$ ). The trustee's payoff is affine in the state, creating a scenario where cooperation is only beneficial if the state exceeds a specific threshold. The builder incurs a cost $c$ for sending a positive signal ( $s_t=1$ ) and receives a benefit $g$ only if the trustee cooperates. The core problem is to characterize the optimal stationary Markov strategies for both players, particularly determining whether the builder's optimal signaling policy follows a simple monotone threshold rule or a more complex structure.

Methodology
The analysis employs dynamic programming within a stationary Markov perfect equilibrium framework.

Trustee's Optimization: The authors first derive the trustee's best response. Because the trustee's current action does not influence the state transition (one-sided state control), the continuation value cancels out of the decision comparison. This allows the derivation of an exact, stationary cooperation threshold $\theta = \ell / (b + \ell)$ as a dynamic best response, independent of the trustee's discount factor.
Builder's Optimization: The builder faces a standard Bellman equation where the value function $V^*(T)$ depends on the immediate payoff (benefit of cooperation minus signaling cost) and the discounted continuation value. The authors establish the existence of a pure stationary Markov equilibrium using the contraction mapping theorem on the space of bounded functions.
Analytical Bounds and Regimes: By analyzing the difference in Bellman values between signaling and not signaling ( $\Delta(T)$ ), the authors derive exact bounds on the signaling gain. They identify specific parameter regimes where the optimal policy can be characterized in closed form without solving the full Bellman equation explicitly.
State Dynamics and Timing: The paper separately analyzes the deterministic evolution of the public state under persistent signaling, deriving exact formulas for the weighted-history representation, the earliest threshold-crossing time, and the geometric decay of the state when signaling stops.
Numerical Validation: The analytical results are complemented by deterministic value iteration on a discrete state grid. The authors use these computations not as empirical evidence but to verify the analytical activation-band policy and to explore parameter regimes outside the analytically derived sufficient conditions.

Key Contributions and Results

Existence and Monotonicity: The paper proves the existence of a pure stationary Markov equilibrium and establishes that the builder's optimal value function is bounded and non-decreasing in the public state.
The Activation-Band Policy: The central theoretical contribution is the identification of a parameter regime (specifically where the discount factor $\delta_1 < 1/2$ $δ_{1} < 1/2$ and signaling cost $c$ $c$ satisfies $\frac{\delta_1 g}{1-\delta_1} < c < g$ $\frac{δ _{1} g}{1 - δ _{1}} < c < g$ ) in which the unique optimal stationary policy is an activation band. In this regime, the builder signals only when the public state lies in an intermediate interval $A = [L, U)$ $A = [L, U)$ .
- For low states ( $T < L$ ), signaling is too costly relative to the immediate and future benefits.
- For intermediate states ( $T \in A$ ), signaling is optimal because it pushes the post-signal state above the trustee's cooperation threshold.
- For high states ( $T \ge U$ ), signaling is suboptimal because the state is already high enough to induce cooperation without incurring the current cost.
- Significance: This result demonstrates that a monotone one-sided threshold policy (where one signals if and only if $T$ is below a certain level) cannot be imposed in general. The optimal policy can exhibit a "no-signal–signal–no-signal" structure.
State Dynamics and Timing: The authors derive exact formulas for the state trajectory under persistent positive signaling, showing that the time to reach the cooperation threshold depends logarithmically on the persistence parameter $\lambda$ . Crucially, they distinguish between the first state index where the threshold is crossed ( $\tau_T$ ) and the first period of cooperation ( $\tau_C$ ), establishing the relation $\tau_C = \tau_T - 1$ due to the within-period timing of the trustee's decision.
Numerical Insights: Outside the analytically characterized regime, numerical scans reveal that the signaling region shifts with the discount factor and cost. Higher patience (higher $\delta_1$ ) expands the signaling region to lower states, while higher costs shrink it. The numerical results also suggest that the strict-signaling set remains connected (a single interval) across the scanned parameters, though this is not proven analytically for the continuous state space.

Significance and Claims
The paper claims to provide a tractable, reduced-form mathematical framework that disentangles three often-conflated elements of trust: state evolution, the trustee's response, and the builder's intertemporal signaling.

Theoretical Distinction: The primary significance is the demonstration that the trustee's threshold-based cooperation rule does not imply a corresponding threshold-based signaling rule for the builder. The interaction between current cooperation gains, signaling costs, and continuation values can generate non-monotonic optimal policies.
Reduced-Form Nature: The authors explicitly state that the public state is not a Bayesian posterior nor an empirically calibrated measure of psychological trust. The model is a "reduced-form mathematical mechanism" designed to analyze the strategic consequences of a specific state recursion structure.
Modesty of Claims: The paper does not claim to replace Bayesian reputation theory or to provide empirical evidence. The numerical results are presented as model evaluations and consistency checks rather than global theorems. The authors acknowledge limitations, including the binary signal space, exogenous signal credibility, and the restriction to stationary Markov strategies, noting that these define the boundaries of the current results and suggest directions for future extensions (e.g., continuous signals, noisy observation, or history-dependent strategies).

In summary, the paper establishes that under specific conditions, optimal trust-building behavior is concentrated in an intermediate range of public trust scores, challenging the assumption that trust-building efforts should simply increase as trust levels rise or fall monotonically.

More like this