Incentive Mechanisms for Information Collaboration in Agri-Food Supply Chains: An Evolutionary Game and System Dynamics Approach

Abstract

Information collaboration is a core driver of digital transformation and efficiency improvement in agri-food supply chains. This study constructs a quadripartite evolutionary game model involving the government, an information service platform, farmers, and agri-food enterprises. By integrating system dynamics, it analyzes stakeholders’ strategic interactions and evolutionary pathways while exploring the regulatory effects of key parameters in reward and penalty mechanisms on system convergence. The key findings are as follows: (1) The system reaches a stable equilibrium regardless of initial strategy combinations. (2) The reward–penalty mechanism is essential for equilibrium stability, but the reward amount and allocation ratios must meet threshold constraints. (3) Given the significant path-dependent lock-in effect in agri-food enterprises’ strategy convergence under static parameters, a dynamic parameter configuration scheme is proposed to reshape convergence and optimize equilibrium. The simulation results indicate that dynamic parameter regulation sacrifices the regulatory efficiency of the information service platform to enhance the overall collaboration. A joint dynamic reward–penalty strategy improves efficiency but delays platform convergence, whereas a single dynamic incentive offers a balanced trade-off. Based on this, an incentive framework is developed to guide government incentive design. This study expands the theoretical framework of information collaboration in AFSCs and provides practical guidance for policymakers.

1. Introduction

In recent years, intensified cross-border agricultural trade and rising consumer expectations regarding food safety and quality have highlighted the importance of agri-food supply chains (AFSCs) [1]. Developing AFSCs requires integrating diverse stakeholders and ensuring timely information exchange and collaboration, which demand considerable time and resources [2,3]. The advancements in digital technology offer new opportunities for collaboration in AFSCs [4], enabling closer coordination across supply chain segments, reducing operational errors through precise supply–demand matching, and enhancing the efficiency of production decision-making [5].

Supply chain collaboration refers to the relationship in which supply chain partners systematically integrate resources, coordinate business processes, and align organizational structures to achieve shared objectives [6]. The existing research shows that such collaboration among various stakeholders boosts competitive advantage and improves both commercial and social outcomes. It helps members to expand market reach and boost profits [7] while also reducing conflicts, increasing participation, and encouraging sustainable cooperation [8]. Information serves as a core element of this collaboration, facilitating product data exchange, market intelligence sharing, and policy communication. Bi-directional, comprehensive, accurate, and real-time information flows significantly reduce communication barriers and information distortion risks [9].

In AFSCs, information collaboration drives efficient operations by enhancing transparency and interdependence among members [2]. Disadvantaged farmers and small- and medium-sized enterprises gain access to technical resources, market information, and support, thereby improving the quality and stability of end-products. It also reduces uncertainty in the production and distribution processes [10]. When stakeholders share data on production, costs, inventory, and order planning, upstream and downstream members can generate accurate forecasts and develop informed plans. This collaborative model shortens supply chain cycles, lowers transaction costs and food losses, and improves both profitability and service levels. Critically, information collaboration can mitigate the bullwhip effect, optimize the overall system costs, and foster value creation across the chain, providing pathways for AFSC improvement [11,12].

In the existing AFSCs, particularly in developing countries, complex and persistent challenges remain prevalent, with information collaboration offering a critical pathway for breakthrough. First, AFSC markets are highly vulnerable to a range of disruptions, including climate fluctuations, pest infestations, supply–demand imbalances, and unexpected events. In such contexts, timely and effective information collaboration can significantly enhance supply chain resilience, helping to ensure the stable supply of agri-food [13]. By enabling full-chain data monitoring and early risk warnings, it supports enterprises in dynamically adjusting pricing and balancing market supply and demand. Second, the absence of robust information collaboration mechanisms often results in limited transparency among AFSC members and can trigger trust crises, leading to “prisoner’s dilemma” scenarios. Information asymmetry not only impedes fair negotiation but also contributes to inefficiencies and profit distribution imbalances [14]. Establishing effective information collaboration systems allows members to share market data equitably and to build stable, contract-based relationships grounded in cooperative decision-making, enhancing both the stability and sustainability of AFSCs.

Moreover, traditional AFSCs typically suffer from low levels of digitalization, making information collaboration heavily dependent on the development of digital infrastructure. Government incentive mechanisms can effectively encourage stakeholders to invest in digital technologies and actively participate in collaborative efforts. More importantly, the establishment of well-designed incentive-constraint frameworks enables governments to guide standardized and institutionalized collaboration across the supply chain, driving the transformation and upgrading of AFSCs. Governments worldwide, particularly in China, have increasingly recognized the strategic role of agricultural informatization in advancing agricultural modernization. Since 2020, the Chinese government has introduced a series of policies to promote agricultural digitalization. Notably, the 2024 policy agenda highlights key priorities, such as “advancing industry chain digitalization”, “building agricultural information service platforms”, and “encouraging joint investment by farmers and social actors through subsidy mechanisms”.

Although the value of information collaboration in AFSCs is gaining recognition, there remains an urgent need for systematic investigation into the information collaboration behavior of AFSC members, the key factors influencing their collaborative decisions, and the mechanisms driving strategy evolution in multi-agent interactions. These aspects remain underexplored in the current literature and represent the focus of this study. From the perspective of game theory, this study focuses on information collaboration within the AFSC. It proposes an innovative quadripartite evolutionary game model involving the government, farmers, agri-food enterprises (AFEs), and information service platform (ISPs). In the model, the government designs reward and penalty mechanisms to incentivize the participation of different stakeholders in information collaboration. Each actor independently decides whether to actively collaborate, and their strategies are interdependent with dynamic feedback effects. Building on this framework, the study incorporates system dynamics modeling to simulate the evolution of strategies and explore the impact of key parameters on information collaboration. Furthermore, it proposes optimized incentive strategies under practical constraints.

The main contributions of this paper are as follows: (1) It breaks through the traditional binary or tripartite game structures by developing a quadripartite evolutionary game model for information collaboration in the AFSC, placing particular emphasis on government incentives as a core driving factor in the system. (2) It applies system dynamics methods to visualize the strategic evolution of multiple stakeholders and assess the impact of different parameters on overall collaboration efficiency. (3) Based on sensitivity analysis, the study designs a context-responsive dynamic incentive strategy for the government, in which the reward or penalty allocation ratio is adjusted over time to enhance the responsiveness and convergence of AFEs’ strategies. (4) For long-term policy design, the study proposes a diversified incentive system that combines fiscal subsidies, institutional incentives, and market-based mechanisms to replace a single-subsidy model. This approach helps to alleviate fiscal pressure while enhancing the resilience and sustainability of policy implementation.

The remainder of the paper is organized as follows: Section 2 provides a literature review, systematically examining the research progress in agricultural product supply chains, information collaboration, evolutionary game theory, and its applications in agricultural product supply chains. Section 3 constructs a quadripartite evolutionary game model for information collaboration in AFSCs, defining the participants, their interaction relationships, and calculating their payoff matrices. In Section 4, a system dynamics model is constructed on the basis of the quadripartite evolutionary game model, and its effectiveness is verified, which lays a foundation for subsequent simulation and optimization. Section 5 focuses on system dynamics simulation and strategy optimization. It begins with a sensitivity analysis of the key system parameters, followed by the design of a dynamic reward–penalty mechanism to enhance the convergence performance of the system. Finally, a government diversified incentive strategy is proposed to ensure the sustainability of the policy implementation. Section 6 discusses the simulation results and provides corresponding recommendations. Finally, Section 7 summarizes the main findings and limitations of the paper and proposes directions for future research.

2. Literature Review

2.1. Agri-Food Supply Chain

The AFSC includes key stages, such as cultivation or livestock farming, preliminary processing, storage and transportation, advanced processing, and distribution, along with the essential linkages between them [15]. Typically, the AFSC integrates all the value-added activities involved in producing and delivering agri-food to end-users or consumers [16]. Its development requires coordination among a wide range of stakeholders. The AFSC aims to achieve two core objectives: ensuring AFSC safety and stability while achieving Pareto-optimal outcomes through collaboration, balancing profit allocation, cost control, and risk mitigation [17].

Based on stakeholder theory, AFSC participants can be categorized into two groups: primary stakeholders, who directly participate in agri-food circulation and derive economic benefits (such as farmers, production bases, processing enterprises, distributors, and retailers), and secondary stakeholders, who are indirectly involved (such as government agencies, logistics service providers, ISPs, and financial institutions). The latter influence the AFSC’s core operations by regulating material, information, and financial flows. Figure 1 presents a general structural diagram of the AFSC, clearly distinguishing between primary and secondary stakeholders, and using arrows to indicate collaborative relationships and resource flow pathways among them.

As illustrated in Figure 1, farmers are positioned at the initial stage of the AFSC, responsible for primary agri-food activities such as crop cultivation, livestock farming, and raw material harvesting. However, in developing countries, farmers often operate in a fragmented and small-scale manner, placing them in a weak bargaining position within the AFSC. This disadvantage primarily stems from several factors: the lack of market forecasting capabilities, making it difficult to anticipate agricultural trends and consumer demand shifts; delayed adoption of advanced agricultural technologies due to financial and knowledge constraints, hindering the timely use of smart farming equipment and digital tools; and limited capacity for data collection, analysis, and decision-making [18]. To address these challenges, AFSC collaboration frameworks must be designed with farmer accessibility in mind, lower information acquisition costs, and helping them to overcome barriers to digitalized production. Such support is typically funded through government funding programs, which enhance farmers’ connections with downstream entities, foster strategic complementarity, and ultimately improve the overall robustness and sustainability of the AFSC [19].

The existing literature predominantly classifies intermediate actors in AFSCs, such as suppliers, processors, distributors, and retailers, as agri-food enterprises (AFEs). This classification allows researchers to analyze AFEs as a unified entity when analyzing their competitive and cooperative relationships with farmers, consumers, and other secondary stakeholders. For instance, Hamidoğlu developed a government-supported AFSC collaboration model where financial subsidies are used to incentivize AFE participation and Nash equilibrium is applied to analyze joint decision-making between farmers and AFEs [14]. Building on this conceptualization, this study defines AFEs as specialized entities engaged in agri-food processing and distribution. Their operations encompass raw material procurement, processing and packaging, warehousing and logistics, and retail distribution, forming an intermediary network that connects farmers and consumers within the AFSC.

Advancements in agricultural information technology are transforming both agri-food systems and management frameworks. Digital agriculture technologies, including the Internet of Things, big data, cloud computing, and blockchain, enable real-time monitoring and intelligent decision-making across all agri-food elements, significantly enhancing farmers’ precise farming capabilities [20,21]. Consequently, information sharing targeting farmers has become a critical driver of technology adoption, as demonstrated by practical implementations in various countries. For example, India launched the Kisaan SMS Portal; Bangladesh implemented a National Livestock Extension Strategy; and China established a big data platform under its Smart Agriculture initiatives [22]. However, farmer-centric information services are prone to forming data silos, limiting their ability to support supply-chain-level collaboration. Thus, establishing a unified AFSC ISP has become an essential solution. This ISP should serve as a public information hub, aggregating and disseminating agricultural policies, market intelligence, technical guidance, and disaster alerts. It must also function as distributed data middleware, integrating, cleansing, storing, and distributing heterogeneous data across supply chain members. According to Zhao and Yang, agricultural data sharing represents a continuous multi-stakeholder dynamic game in which the participants may adopt opportunistic behaviors, such as withholding information to gain free-rider advantages, thereby compromising data quality and spreading misinformation, ultimately reducing supply chain efficiency [23]. This underscores the need for robust data validation mechanisms and the importance of designing reward–penalty mechanisms to ensure effective stakeholder participation.

To ensure the reliability and sustainability of the food supply while enhancing agricultural value creation, governments play a dual role as regulators and facilitators within AFSCs [24,25]. Government interventions typically encompass a multi-dimensional policy framework, including supply-side measures, such as fiscal subsidies, technical training, research consortiums, and infrastructure investments, alongside demand-side regulations involving food safety standards, production certifications, and punitive legislation aimed at mitigating systemic risks [26,27]. Simultaneously, government participation in AFSC governance fosters trust among AFSC stakeholders, promoting cooperative investments and collaborative innovation while mitigating opportunistic behaviors arising form distrust [28]. Prior studies further indicate that government incentives, particularly reward–penalty mechanisms, are vital in reshaping AFE decision-making by strengthening credible commitment mechanisms among stakeholders, thereby improving the overall supply chain operational efficiency [29]. Nevertheless, regulatory effectiveness is often constrained by information asymmetry, which may result in government failure, highlighting the critical importance of real-time data transparency in policy implementation [30]. Accordingly, incorporating governments as endogenous decision-makers into the AFSC information collaboration model is not only theoretically justified but also practically imperative.

2.2. Information Collaboration in AFSCs

Supply chain collaboration is defined as “two or more independent enterprises jointly designing and executing supply chain operations to achieve performance advantages beyond individual operations” [31]. Such a collaborative mechanism fosters strategic synergy among partners through joint planning and real-time information exchange [6]. In the absence of collaboration, participants tend to make decentralized decisions based on downstream orders and individual inventory strategies. This often leads to the amplification of order and inventory fluctuations upstream, a phenomenon known as the bullwhip effect, caused by order transmission delays, logistics lags, and information silos. Fragmented decision-making of this nature results in multiple cost inefficiencies, including surging inventory holding costs, frequent delivery delays, and resource mis-allocations. Prior studies have demonstrated that information collaboration significantly mitigates the bullwhip effect by enabling member enterprises to optimize ordering strategies, inventory allocation, and production scheduling based on shared information.

Supply chain information collaboration specifically refers to the use of advanced information technologies to integrate, coordinate, and develop resources, business processes, and inter-organizational relationships. It achieves this by facilitating the sharing of operational data and the exchange of market intelligence among partners, thereby allowing for more agile responses to end-customer demands [32]. Empirical evidence shows that the adoption of information technologies for collaborative information sharing effectively reduces overall supply chain costs. Moreover, the value added by information technology lies not only in accelerating information flow but also in enhancing the efficiency of physical goods’ movement [33]. However, information collaboration is constrained by dual cost factors: explicit costs such as infrastructure investment and human resource development, and implicit risks including information leakage and free-riding behaviors among members. Research by the US National Research Council indicates that small- and medium-sized enterprises encounter more severe challenges in this domain, such as high technology adoption thresholds, limited access to training resources, weak bargaining power, and difficulties in building trust-based relationships [34]. These barriers are similarly prevalent in AFSCs, exacerbating the difficulty in implementing effective information collaboration.

In AFSCs, shareable information typically includes (1) agri-food production data, such as appearance, variety, yield, quality, and cost; (2) market information, such as demand forecasts and price trends; and (3) logistics data, such as transportation costs, shelf life, and logistics expenses [35]. Access to such information enables supply chain members to formulate more informed production and marketing plans. However, the benefits of information collaboration are not uniformly distributed. In multi-tiered supply chains, the advantages of information collaboration are often asymmetric; downstream data sharing does not necessarily benefit upstream actors [36]. Key network nodes, such as warehousing centers, can improve inventory efficiency through data sharing and thereby capture disproportionate benefits [37].

Even limited collaboration has been shown to yield measurable improvements in supply chain performance [38]. Nevertheless, information sharing can also entail significant risks [39]. For example, the risk of information leakage may discourage members from disclosing demand fluctuations, or may even prompt them to deliberately provide misleading information to protect their own interests [40,41]. Consequently, to foster participation while mitigating opportunistic behaviors, it is essential to design and implement robust reward–penalty mechanisms that are both reasonable and effective [42,43].

2.3. Application of Evolutionary Game Theory in AFSCs

Evolutionary game theory integrates game-theoretic analytical frameworks with dynamic evolutionary processes, providing an effective quantitative method to investigate strategic interactions among two or more decision-makers under conflicting interests [44]. By simulating bounded rationality and strategy-learning behaviors, evolutionary game theory effectively captures conflict-coordination dynamics in real-world scenarios, making it widely applicable across sociology, economics, and interdisciplinary fields like supply chain management [45,46,47].

In recent years, with the accelerating process of agricultural informatization, evolutionary game models have been increasing applied to characterize information collaboration behaviors in AFSCs. Song and Liu et al., as well as Zheng and Xu, employed evolutionary game models to examine the application and diffusion of blockchain technology within AFSCs [48,49]. Zhang and Xue developed a two-party evolutionary game model between suppliers and manufacturers, analyzing the influence of key parameters on strategy evolution paths and emphasizing the critical role of information collaboration in enhancing supply chain efficiency [32]. Li and He focused on information-sharing behaviors between suppliers and retailers in a fresh AFSC. They established a two-level evolutionary game model and found that information sharing helps to mitigate the bullwhip effect. Their results also indicated that, when the benefits of information sharing are high and the losses from information leakage are low, both parties demonstrate a stronger willingness to share [50]. Zhao and Yang constructed a tripartite evolutionary game model involving a data-sharing platform, data providers, and data consumers, highlighting the pivotal role of reward–penalty mechanisms and information feedback in shaping participant decision-making behaviors [21]. Wang and Peng analyzed information sharing among three stakeholders, farmer cooperatives, processing enterprises, and sales enterprises, in AFSCs. They proposed that farmer cooperatives, as the leading enterprises, have the authority to supervise other participants. Their supervision mechanism imposes penalties on parties that refuse to share information, with the fines being redistributed among the parties engaged in sharing. This study thoroughly examined the influence of seven parameters, including the compensation coefficient, information leakage risk coefficient, synergy coefficient, degree of information sufficiency, level of information utilization, cost of information sharing, and cost of information supervision, on the evolutionary dynamics of the system [51].

The government actively participates in AFSCs through multiple approaches, including regulation, subsidies, or policy guidance and so on. Government involvement plays a significant role in promoting the evolution of AFSCs and is recognized as one of the key driving forces behind their development. Song and Liu et al. constructed a tripartite evolutionary game model involving the government, enterprises, and farmers to examine how blockchain subsidies from the government can facilitate collaboration among agricultural enterprises, highlighting the guiding role of regulators in maintaining supply chain stability [48]. Liu and Zhu, from the perspective of carbon tax policies and green subsidies, revealed the synergistic effect of fiscal interventions on the diffusion of low-carbon agricultural technologies. They further emphasized that subsidy amounts should be linked to the level of green investment by enterprises to improve policy efficiency [52]. Teng and Chen focused on farmers’ pesticide use behaviors and established a tripartite model to explore how green subsidies and penalty policies influence behavioral evolution. Their study confirmed the government’s pivotal role in enhancing the safety of agri-food quality [53]. Zheng and Xu developed a tripartite evolutionary game model involving producers, processors, and the government to explore blockchain-based strategies for traceability in AFSCs. The study showed that the strategic decisions of game participants were mainly influenced by the trade-off between expected benefits and costs, traceability subsidies, and free-rider effects. To ensure the effectiveness of the traceability system, the government was modeled as a regulator, implementing fixed subsidies and penalties to enhance participants’ compliance and willingness to engage [49]. Yang and Dai, using the development of organic agriculture as a background, analyzed how different government subsidy strategies influence farmers’ production behaviors. By designing three policy scenarios, the study found that government subsidies significantly promote the development of organic agriculture. Moreover, subsidizing retailers proved more effective than subsidizing farmers in encouraging organic production [54]. Chen and Sun incorporated the adoption of green agricultural technologies into an evolutionary game framework involving smallholder farmers, family farms, and agricultural technology service organizations. By comparing the strategy evolution under government intervention versus non-intervention scenarios, the study demonstrated that government intervention plays a crucial role in addressing market failures. In particular, the adjustment of subsidy intensity was found to be a key factor in guiding and incentivizing producers to adopt green technologies [55].

The aforementioned studies provide a solid theoretical foundation for understanding the information collaboration in AFSCs. However, a more in-depth depiction of the collaborative behaviors between these stakeholders remains limited. First, due to the complexity of model construction, most of the existing studies are confined to two-party evolutionary games, with relatively few addressing interactions among three parties. In practice, however, AFSCs often involve multiple stakeholders, and each player’s strategy is influenced by the collective actions of others rather than being based on individual decision-making. Second, many information collaborations pay little attention to the government’s role at the production end. They often treat government incentives as fixed or static values, such as constant subsidies or penalties. This does not reflect real-world situations where policies are dynamic, implemented in stages, and limited by fiscal sustainability. Third, with the rise of agricultural digitalization and increasing division of labor, information services are often provided by third-party platforms, not by producers themselves. However, this important trend has not been fully considered in the existing models.

To address these issues, this paper builds a quadripartite evolutionary game model that includes the government, ISP, farmers, and AFEs. The model treats the ISP as an independent player and includes the government’s incentive strategies in the evolution process. Furthermore, building upon conventional static incentive approaches, this study proposes dynamic incentive strategies tailored to different behavioral scenarios. Moreover, considering the constraint of fiscal sustainability, a diversified incentive strategy framework is developed to replace single fiscal subsidies. The results provide a more realistic reflection of the strategic interactions and game characteristics among multiple stakeholders in information collaboration contexts, enriching the game-theoretic dimensions and behavioral insights of government intervention research in AFSCs.

3. Development of Information Collaboration Model in the AFSC

3.1. Model Construction

In this section, we design a quadripartite evolutionary game model for information collaboration in the AFSC. This study focuses on the strategy choices, interactive mechanisms, and evolutionary trajectories of various stakeholders. Based on the previous literature review, we direct participants in information collaboration, farmers, agri-food enterprises (AFEs), an information service platform (ISP), and the government as game agents of the model. The term “agri-food enterprises” refers to a collective designation for intermediary enterprises positioned between the farmers and consumers within AFSCs, including processing enterprise, distributor, and retailer. The structure of the information collaboration in AFSCs is illustrated in Figure 2.

As illustrated in Figure 2, within the AFSC information collaboration framework, the ISP functions as a multi-source data hub, integrating data aggregation, data accuracy monitoring, data storage and distribution, and monitoring result feedback [50]. These core functionalities establish the ISP as the digital backbone essential for AFSC-wide information collaboration. The ISP continuously provides farmers and AFEs with public information (e.g., government policies, climatic conditions, pest/disease alerts, market fluctuations, and reference prices) alongside technical solutions (e.g., precision irrigation algorithms and inventory optimization models). Yu and Chen et al. empirically demonstrated that such real-time advisory services enhance the operational efficiency of supply chain members [19]. Farmers and AFEs gain access to the AFSC collaboration network by adhering to access protocols, which mandate compliance with governance frameworks and shared responsibility clauses in exchange for ISP services. Participants retain data upload autonomy, and information synergy effects are activated only when both parties voluntarily share proprietary data. The information synergy effects significantly reduce the bullwhip effect while enhancing supply chain efficiency and stability [51]. To mitigate data quality risks, the ISP verifies and validates all received shared information. The government employs a dual-leverage incentive mechanism to drive collaboration within the model, consisting of reward incentives and penalty constraints. The reward mechanism establishes a fixed subsidy pool, which is distributed among members adopting proactive strategies based on a predefined ratio. If non-compliant members exist, their allocated shares are redistributed among compliant members. The penalty mechanism imposes fines on entities that withhold information, with the collected fines reinvested into an incentive pool for redistribution among the collaborative government and ISP.

3.2. Model Assumptions

To construct an effective quadripartite evolutionary game model, this study formulates key assumptions based on the characteristics of information collaboration in the AFSC.

Assumption 1.

Game Agents. This model incorporates four interacting participants: the government, the ISP, farmers, and AFEs. The government plays an incentive role in AFSC information collaboration, primarily by implementing reward-and-penalty mechanisms to encourage information collaboration. Under proactive strategies, the government further provides targeted financial subsidies to enhance collaborative engagement. As the dominant force in AFSC information collaboration, the ISP functions as a digital hub responsible for continuously delivering public information and technical services to farmers and AFEs. To enhance the accuracy of received information, the ISP autonomously initiates data verification mechanisms and reports anomalies to the government. Farmers and AFEs are considered subordinate entities in the information collaboration model: farmers, as primary producers, supply raw agricultural products, while AFEs, comprising processors, distributors, and other intermediaries, bridge farmers and consumers by facilitating product transformation and market integration. Their participation in AFSC information collaboration entails both receiving services from the ISP and adhering to its regulatory framework. Specifically, farmers and AFEs retain autonomy over private data uploads, but shared data usage rights must be transferred to the ISP for verification. Additionally, both entities are subject to the government’s reward–penalty framework.

Assumption 2.

Rationality. All game participants are rational decision-makers. Within the bounded rationality framework, all agents dynamically adjust their behavioral decisions based on evolving economic payoffs [56].

Assumption 3.

Strategy. In the quadripartite evolutionary game model, the strategic choices of the four participants, including the government, the ISP, farmers, and AFEs, are represented by probability values $x, y, z\mathrm{and}w\in [0, 1]$, respectively. The government’s strategy set includes “Positive Incentive” and “Common Incentive”; the former refer to allocating additional subsidies from a reward pool to participants engaging in information collaboration to offset collaboration costs, while the latter represent the benchmark penalty and redistribution mechanism, wherein penalties are imposed on non-participants in information sharing and the collected penalties redistributed. The ISP chooses between “Active Monitoring” and “Passive Response”. The former involves proactively verifying shared information from other participants to enhance data accuracy, while the latter solely performs fundamental data storage and distribution functions. Farmers and AFEs decide between “information sharing” and “information concealment”, where the former involves uploading complete production or distribution datasets and the latter entails either refusing to share data or submitting low-quality information.

Table 1. Role positioning and strategy space of game agents.

Game Agents	Role Positioning	Strategy Space	Strategy Definition
Government	Institutional Designer and Resource Allocator	Positive Incentive	Provides subsidies to encourage proactive strategies
		Common Incentive	Penalizes information concealment and redistributes fines
Information Service Platform	Digital Hub and Information Center	Active Monitoring	Initiates data verification mechanisms
		Passive Response	Performs basic data integration
Farmers	Primary Agricultural Producers	Information Sharing	Shares production data with ISP
		Information Concealment	Refuses to share private data
Agri-food Enterprises	Agri-food Processing and Distribution	Information Sharing	Shares distribution data with ISP
		Information Concealment	Refuses to share private data

outlines the core functions and strategy space of each game agent.

Assumption 4.

In the AFSC information collaboration game model, six key types of information interaction relationships exist among the stakeholders (as illustrated in Figure 3). Together, these relationships form the structure of multi-agent information exchange and strategic interaction. First, data service contracts are established between the ISP and both farmers and AFEs. Under these contracts, farmers and AFEs voluntarily share part of their data sovereignty, such as production data, inventory records, and logistics information, in exchange for access to public data, digital services, and value-added insights from other members’ data provided by the ISP. This data exchange is enabled by emerging digital technologies such as blockchain and the Internet of Things (IoT), which improve the efficiency and transparency of data collection, transmission, and verification [54,57]. Second, product circulation contracts are formed between farmers and AFEs to support the delivery and traceability of physical goods. These contracts are based on standard commercial agreements that define each party’s responsibilities in the production and sales process. Through this arrangement, data from the production and processing stages are effectively linked, forming a reliable traceability system that supports downstream information tracking [58]. Third, the government enters into policy response contracts with the ISP, farmers, and AFEs by designing a reward-and-penalty mechanism. Prior studies have shown that governmental incentives significantly enhance both the efficiency and stability of AFSCs [59]. In 2024, China’s Ministry of Agriculture and Rural Affairs issued two major policy documents, Guidelines for the Development of Smart Agriculture and the National Smart Agriculture Action Plan (2024–2028), which clearly advocate for financial subsidies, tax relief, and insurance mechanisms to encourage joint participation from farmers and AFEs. These policies also emphasize the role of ISPs in enabling collaborative governance and data integration [60,61].

3.3. Model Parameters

Based on the characteristics of the main game agents in the quadripartite evolutionary game model of information collaboration and referring to previous assumptions, we analyze the potential information collaboration phenomena, such as reward–penalty mechanism, information leakage risk, information synergy, and rent-seeking behavior, in the model.

Government incentive behavior is one of the important driving forces of the evolutionary game of AFSCs. Referencing the literature [48,53,62], we designed the government’s reward and penalty mechanisms. The reward mechanism under the government’s “positive incentive” strategy: The government sets a fixed reward amount S, first allocating a ratio $\alpha (0<\alpha <1)$ to the ISP that chooses the “active monitoring” strategy, with the surplus allocated between farmers and AFEs that choose “information sharing” in the ratio $\alpha 1:\alpha 2(\alpha 1+\alpha 2=1)$. When any game agency among the ISP, farmers, and AFEs chooses a passive strategy, this participant will not receive rewards; instead, rewards are distributed among those choosing proactive strategies according to the previously established logic. Additionally, under both “positive incentive” and “common incentive” strategies, the government implements penalty mechanisms against participants choosing passive strategies: imposing a fine $P_1$ on farmers who conceal information, AFEs $P_2$, and ISPs that fail to fulfill monitoring responsibilities F. When the ISP implements an “active monitoring” strategy and there are participants concealing information, the government redistributes fines, subsidizing the ISP with a penalty ratio $\beta (0<\beta <1)$ as compensation for fulfilling its responsibilities.

The parameter setting for the information collaboration benefits part refers to the study [51], with benefits measured by the volume of information obtained by each agency. The ISP remains in an information sharing state, with its published public information volume recorded as I, while the information volumes shared by farmers and AFEs are recorded as $I_1$ and $I_2$, respectively. When both farmers and AFEs choose information sharing strategies, they trigger an “information synergy effect”, with a synergy coefficient $\rho (\rho >1)$, allowing each agency to obtain additional synergistic benefits: the ISP gains $\rho (I1+I2)$, farmers gain $\rho (I+I2)$, AFEs gain $\rho (I+I1)$, and the government receives additional social benefits E from the synergy.

When the ISP chooses an “active monitoring” strategy, it enhances information accuracy proportionally to the received information volume. When only farmers share information, the ISP benefit is $\gamma I_1$; when only AFEs share information, the benefit is $\gamma I_2$; when both share information, considering both information synergy and accuracy improvement, the benefit becomes $\gamma \rho (I_1+I_2)$. When the ISP chooses a passive response strategy, it cannot verify information accuracy, resulting in no accuracy improvement.

Any participant sharing information faces potential leakage risks [21,49]: farmers risk $R_1$ (where $R_1<I_1$), AFEs risk $R_2$ (where $R_2<I_2$), and the ISP, which continuously share information, face a constant risk R (where $R<I$). Meanwhile, when the ISP conducts “passive response”, a free-riding effect occurs, where participants who conceal information seek access to public information usage through rent-seeking from the ISP. The rent-seeking costs are $N_1$ by farmers and $N_2$ by AFEs. While tolerating data deficiencies, the ISP faces data distortion losses alongside rental income, with an expected accuracy loss coefficient of $(\gamma -1)$. When farmers withhold information, the loss is $(\gamma -1)I_1$; when AFEs withhold information, the loss is $(\gamma -1)I_2$; when both withhold information, the combined loss is $(\gamma -1)(I_1+I_2)$. Additionally, the government suffers social benefit losses of $L_1$ when farmers conceal information and $L_2$ when AFEs conceal information.

Finally, the government’s cost for “positive incentive” is O, generating additional social benefits U; the cost for “common incentive” is $O’$; the cost for ISP to implement “active monitoring” is C; the cost for farmers to implement “information sharing” is $D_1$; and the cost for AFEs to implement “information sharing” is $D_2$.

Based on the above analysis, the model parameters are shown in

Table 2. Variables in the quadripartite evolutionary game.

Variables	Definitions	Notation
x	Probability of government’s positive incentive	$0\le x\le 1$
y	Probability of ISP’s active monitoring	$0\le y\le 1$
z	Probability of farmers’ information sharing	$0\le z\le 1$
w	Probability of AFEs’ information sharing	$0\le w\le 1$
O	Cost of positive incentive	$O>0$
$O^{\prime }$	Cost of common incentive	$0<O^{\prime }<O$
U	Social benefits of positive incentive	$U>0$
S	Government additional rewards	$S>0$
$\alpha$	ISP’s reward allocation ratio	$0<\alpha <1$
${\alpha }_1$	Farmers’ surplus reward allocation ratio	${\alpha }_1+{\alpha }_2=1$
${\alpha }_2$	AFEs’ surplus reward allocation ratio	${\alpha }_1+{\alpha }_2=1$
E	Government benefits of Information Synergistic Effects	$E>0$
$L_1$	Loss from information concealment by farmers	$L_1>0$
$P_1$	Penalty for farmers’ information concealment	$P_1>0$
$L_2$	Loss from information concealment by AFEs	$L_2>0$
$P_2$	Penalty for AFEs’ information concealment	$P_2>0$
$\beta$	Penalty allocation ratio of ISP	$0<\beta <1$
C	Cost of active monitoring	$C>0$
R	Risk of information leakage on the ISP	$R>0$
F	Penalty for ISP’s passive response	$F>0$
I	Information volume from ISP	$I>0$
$I_1$	Information volume from farmer sharing	$I_1>0$
$I_2$	Information volume from AFE sharing	$I_2>0$
$\rho$	Information synergy coefficient	$\rho >1$
$\gamma$	Information accuracy improvement coefficient	$\gamma >1$
$D_1$	Cost of farmers’ information sharing	$D_1>0$
$R_1$	Risk of farmers’ information leakage	$R_1>0$
$D_2$	Cost of AFEs’ information sharing	$D_2>0$
$R_2$	Risk of AFEs’ information leakage	$R_2>0$
$N_1$	Cost of rent-seeking by farmers	$N_1>0$
$N_2$	Cost of rent-seeking by AFEs	$N_2>0$

. Under bounded rationality, any game agency faces costs and benefits when implementing strategies, with profit being the driving force behind their ultimate strategy selection. The payoff matrices of the four game agencies are shown in

Table 3. Payoff matrix: information collaboration in AFSC under government’s positive incentive strategy.

Strategies		Agri-Food Enterprises	Information Sharing z	Information Concealment 1 − z
Information Services Platform	Active Monitoring y	Information Sharing w	$-O+U-S+E$	$-O+U+(1-\beta )P_2-L_2-S$
			$-R-C+\rho \gamma (I_1+I_2)+\alpha S$	$-R-C+\beta P_2+\gamma I_1+\alpha S$
			$\rho (I+\gamma I_2)-D_1-R_1+(1-\alpha )S$	$I-D_1-R_1+(1-\alpha )S$
			$\rho (I+\gamma I_1)-D_2-R_2+(1-\alpha )S$	$\rho (I+\gamma I_1)-P_2$
		Information Concealment 1 − w	$-O+U+(1-\beta )P_1-L_1-S$	$-O+U+(1-\beta )(P_1+P_2)$ $-L_1-L_2-S$
			$-R-C+\beta P_1+\gamma I_2+\alpha S$	$-R-C+\beta (P_1+P_2)+S$
			$\rho (I+\gamma I_2)-P_1$	$I-P_1$
			$I-D_2-R_2+(1-\alpha )S$	$I-P_2$
	Passive Response 1 − y	Information Sharing w	$-O+U+F-S+E$	$-O+U+F+P_2-S$
			$R-F+\rho (I_1+I_2)$	$-R-F+N_2-(\gamma -1)I_2+I_1$
			$\rho (I+I_2)-D_1-R_1+S$	$I-D_1-R_1+S$
			$\rho (I+I_1)-D_2-R_2+S$	$\rho (I+I_1)-P_2-N_2$
		Information Concealment 1 − w	$-O+U+F+P_1-S$	$-O+U+F+P_1+P_2-L_1-L_2$
			$-R-F+N_1-(\gamma -1)I_1+I_2$	$-R-F+N_1+N_2-(\gamma -1)(I_1+I_2)$
			$\rho (I+I_2)-P_1-N_1$	$I-P_1-N_1$
			$I-D_2-R_2+S$	$I-P_2-N_2$

and

Table 4. Payoff matrix: information collaboration in AFSC under government’s common incentive strategy.

Strategies		Agri-Food Enterprises	Information Sharing z	Information Concealment 1 − z
Information Services Platform	Active Monitoring y	Information Sharing w	$-O^{\prime }+E$	$-O^{\prime }-L_1+(1-\beta )P_1$
			$-R-C+\rho \gamma (I_1+I_2)$	$-R-C+\beta P_1+\gamma I_2$
			$\rho (I+\gamma I_2)-D_1-R_1$	$\rho (I+\gamma I_2)-P_1$
			$\rho (I+\gamma I_1)-D_2-R_2$	$I-D_2-R_2$
		Information Concealment 1 − w	$-O^{\prime }-L_2+(1-\beta )P_2$	$-O^{\prime }+(1-\beta )(P_1+P_2)$
			$-R-C+\beta P_2+\gamma I_1$	$-R-C+\beta (P_1+P_2)$
			$I-D_1-R_1$	$I-P_1$
			$\rho (I+\gamma I_1)-P_2$	$I-P_2$
	Passive Response 1 − y	Information Sharing w	$-O^{\prime }+F+E$	$-O^{\prime }+F+P_1-L_1$
			$-R-F+\rho (I_1+I_2)$	$-R-F+N_1-(\gamma -1)I_1+I_2$
			$\rho (I+I_2)-D_1-R_1$	$\rho (I+I_2)-P_1-N_1$
			$\rho (I+I_1)-D_2-R_2$	$I-D_2-R_2$
		Information Concealment 1 − w	$-O^{\prime }+F+P_2-L_2$	$-O^{\prime }+F+P_1+P_2-L_1-L_2$
			$-R-F+N_2-(\gamma -1)I_2+I_1$	$-R-F+N_1+N_2-(\gamma -1)(I_1+I_2)$
			$I-D_1-R_1$	$I-P_1-N_1$
			$\rho (I+I_1)-P_2-N_2$	$I-P_2-N_2$

, corresponding to the government’s positive and common incentive strategies, respectively.

3.4. Game Model Development

Drawing on the practical operations of information collaboration in AFSCs, this study analyzes the long-term dynamic evolutionary game among the government, the ISP, farmers, and AFEs under bounded rationality. To simplify the institutional setting and focus on the evolution of strategies among game agents, this study introduces several assumptions: (1) the ISP functions as a neutral public service provider. While free-riding may occur, no explicit collusion or information manipulation exists between the ISP and other agents; (2) the government’s reward–penalty mechanisms are assumed to be fully enforceable, particularly that participants who adopt passive strategies cannot evade punishment; and (3) the model does not consider the baseline costs and benefits associated with non-participation in information collaboration.

Based on the payoff matrix in Table 3 and Table 4, the expected payoffs for government under the positive incentive and common incentive strategies are denoted as $U_x$ and $U_{1-x}$, respectively. Their calculation formulas are presented in Equations (1) and (2).

$$\begin{array}{c}\hfill U_x=yzw(-O+U-S+E)+y(1-z)w[-O+U+(1-\beta )P_1-L_1-S]+yz(1-w)[-O+U+(1-\beta )P_2-\\ \hfill L_2-S]+y(1-z)(1-w)[-O+U+(1-\beta )(P_1+P_2)-L_1-L_2-S]+(1-y)zw(-O+\\ \hfill U+F-S+E)+(1-y)(1-z)w(-O+U+F+P_1-S)+(1-y)z(1-w)(-O\\ \hfill +U+F+P_2-S)+(1-y)(1-z)(1-w)(-O+U+F+P_1+P_2-L_1-L_2)\end{array}$$

(1)

$$\begin{array}{c}\hfill U_{1-x}=yzw(-O^{\prime }+E)+y(1-z)w[-O^{\prime }-L_1+(1-\beta )P_1]+yz(1-w)[-O^{\prime }-L_2+(1-\beta )P_2]+y(1-z)\\ \hfill (1-w)[-O^{\prime }+(1-\beta )(P_1+P_2)]+(1-y)zw(-O^{\prime }+F+E)+(1-y)(1-z)w(-O^{\prime }+F+P_1-L_1)+\\ \hfill (1-y)z(1-w)(-O^{\prime }+F+P_2-L_2)+(1-y)(1-z)(1-w)(-O^{\prime }+F+P_1+P_2-L_1-L_2)\end{array}$$

(2)

The calculation formula for the government’s average expected payoff is presented in Equation (3).

$${\overline{U}}_{x,1-x}=x{U}_x+(1-x)U_{1-x}$$

(3)

The change rate of the government’s probability of adopting the positive incentive strategy is presented in Equation (4).

$${\displaystyle \frac{dx}{dt}}=x\left(U_x-{\overline{U}}_{x,1-x}\right)=x(1-x)\left(U_x-U_{1-x}\right)$$

(4)

The government’s replicator dynamic equation can be derived from Equations (1)–(4), with the result presented in Equation (5).

$$\begin{array}{c}\hfill f\left(x\right)=x(1-x)[yzw(-O+U+S+O^{\prime })+yz(1-w)(-O+U+S+O^{\prime })+yz(1-w)(-O+U+S+O^{\prime })+\\ \hfill y(1-z)w(-O+U+S+O^{\prime })+(1-y)zw(-O+U+S+O^{\prime })+\\ \hfill (1-y)z(1-w)(-O+U+S+O^{\prime })+(1-y)(1-z)w(-O+U+S+O^{\prime })]\end{array}$$

(5)

Similarly, the expected payoffs for ISP under the active monitoring and passive response strategies are denoted as $U_y$ and $U_{1-y}$, respectively. Their calculation formulas are presented in Equations (6) and (7).

$$\begin{array}{c}\hfill U_y=xzw(-R-C+\rho \gamma (I_1+I_2)+\alpha S)+x(1-z)w[-R-C+\beta P_1+\gamma I_2+\alpha S]+xz(1-w)[-R-C+\beta P_2\\ \hfill +\gamma I_1+\alpha S]+x(1-z)(1-w)[-R-C+\beta (P_1+P_2)+S]+(1-x)zw(-R-C+\rho \gamma (I_1+I_2))\\ \hfill +(1-x)(1-z)w(-R-C+\beta P_1+\gamma I_2)+(1-x)z(1-w)(-R-C+\beta P_2+\gamma I_1)\\ \hfill +(1-x)(1-z)(1-w)(-R-C+\beta (P_1+P_2))\end{array}$$

(6)

$$\begin{array}{c}\hfill U_{1-y}=xzw[-R-F+\rho (I_1+I_2)]+x(1-z)w[-R-F+N_1-(\gamma -1)I_1+I_2]+xz(1-w)[-R-F+N_2\\ \hfill -(\gamma -1)I_2+I_1]+x(1-z)(1-w)[-R-F+N_1+N_2-(\gamma -1)(I_1+I_2)]+(1-x)zw[-R-F+\rho (I_1+I_2)]\\ \hfill +(1-x)(1-z)w[-R-F+N_1-(\gamma -1)I_1+I_2]+(1-x)z(1-w)[-R-F+N_2-(\gamma -1)I_2+I_1]\\ \hfill +(1-x)(1-z)1-w)[-R-F+N_1+N_2-(\gamma -1)(I_1+I_2)]\end{array}$$

(7)

The calculation formula for the ISP’s average expected payoff is presented in Equation (8).

$${\overline{U}}_{y,1-y}=y{U}_y+(1-y)U_{1-y}$$

(8)

The change rate of the ISP’s probability of adopting the active monitoring strategy is presented in Equation (9).

$${\displaystyle \frac{dy}{dt}}=y\left(U_y-{\overline{U}}_{y,1-y}\right)=y(1-y)\left(U_y-U_{1-y}\right)$$

(9)

The ISP’s replicator dynamic equation can be derived from Equations (6)–(9), with the result presented in Equation (10).

$$\begin{array}{c}\hfill f\left(y\right)=y(1-y)\{xzw[-C+F+(\gamma -1)\rho (I_1+I_2)+\alpha S]+x(1-z)w[-C+F+\beta P_1-N_1+\\ \hfill (\gamma -1)(I_1+I_2)+\alpha S]+x(1-w)[-C+F+\beta P_2-N_2+(\gamma -1)(I_1+I_2)+S]+\\ \hfill (1-x)zw[-C+F+(\gamma -1)(I_1+I_2)]+(1-x)z(1-w)[-C+F+\beta P_1-N_1+(\gamma -1)(I_1+I_2)]\\ \hfill +(1-x)(1-z)w[-C+F+\beta P_2-N_2+(\gamma -1)(I_1+I_2)]\}\end{array}$$

(10)

Similarly, the expected payoffs for farmers under the information sharing and information concealment strategies are denoted as $U_z$ and $U_{1-z}$, respectively. Their calculation formulas are presented in Equations (11) and (12).

$$\begin{array}{c}\hfill U_y=xyw[\rho (I+\gamma I_2)-D_1-R_1+(1-\alpha )S]+x(1-y)w[\rho (I+I_2)-D_1-R_1+S]+xy(1-w)\\ \hfill [I-D_1-R_1+(1-\alpha )S]+x(1-y)(1-w)(I-D_1-R_1+S)+(1-x)yw[\rho (I+\gamma I_2)-D_1-R_1]+\\ \hfill (1-x)(1-y)w[\rho (I+I_2)-D_1-R_1]+(1-x)y(1-w)(I-D_1-R_1)+(1-x)(1-y)(1-w)(I-D_1-R_1)\end{array}$$

(11)

$$\begin{array}{c}\hfill U_{1-z}=xyw[\rho (I+\gamma I_2)-P_1]+x(1-y)w[\rho (I+I_2)-P_1-N_1]+xy(1-w)(I-P_1)+x(1-y)(1-w)\\ \hfill (I-P_1-N_1)+(1-x)yw[\rho (I+\gamma I_2)-P_1]+(1-x)(1-y)w[\rho (I+I_2)-P_1-N_1]+\\ \hfill (1-x)y(1-w)(I-P_1)+(1-x)(1-y)1-w)(I-P_1-N_1)\end{array}$$

(12)

The calculation formula for farmers’ average expected payoff is presented in Equation (13).

$${\overline{U}}_{z,1-z}=z{U}_z+(1-z)U_{1-z}$$

(13)

The change rate of the farmers’ probability of adopting the information sharing strategy is presented in Equation (14).

$${\displaystyle \frac{dz}{dt}}=z\left(U_z-{\overline{U}}_{z,1-z}\right)=z(1-z)\left(U_z-U_{1-z}\right)$$

(14)

The farmers’ replicator dynamic equation can be derived from Equations (11)–(14), with the result presented in Equation (15).

$$\begin{array}{c}\hfill f\left(z\right)=z(1-z)\{xyw[-D_1+(1-\alpha )S-R_1+P_1+N_1]+x(1-y)w[-D_1+(1-\alpha )S-R_1+P_1+N_1]+\\ \hfill x(1-w)[-D_1+(1-\alpha )S-R_1+P_1+N_1]+(1-x)yw(-D_1-R_1+P_1)+(1-x)y(1-w)(-D_1-R_1+P_1)\\ \hfill +(1-x)(1-y)w(-D_1-R_1+P_1)+(1-x)(1-y)(1-w)(-D_1-R_1+P_1+N_1)\}\end{array}$$

(15)

Similarly, the expected payoffs for AFEs under the information sharing and information concealment strategies are denoted as $U_w$ and $U_{1-w}$, respectively. Their calculation formulas are presented in Equations (16) and (17).

$$\begin{array}{c}\hfill U_w=xyz[\rho (I+\gamma I_1)-D_2-R_2+(1-\alpha )S]+x(1-y)z[\rho (I+I_1)-D_2-R_2+S]+xy(1-z)[I-D_2-R_2\\ \hfill +(1-\alpha )S]+x(1-y)(1-z)(I-D_2-R_2+S)+(1-x)yz[\rho (I+\gamma I_1)-D_2-R_2]+(1-x)(1-y)z\\ \hfill [\rho (I+I_1)-D_2-R_2]+(1-x)y(1-z)(I-D_2-R_2)+(1-x)(1-y)(1-z)(I-D_2-R_2)\end{array}$$

(16)

$$\begin{array}{c}\hfill U_{1-w}=xyz[\rho (I+\gamma I_1)-P_2]+x(1-y)z[\rho (I+I_1)-P_2-N_2]+xy(1-z)(I-P_2)+x(1-y)(1-z)\\ \hfill (I-P_2-N_2)+(1-x)yz[\rho (I+\gamma I_1)-P_2]+(1-x)(1-y)z[\rho (I+I_1)-P_2-N_2]+\\ \hfill (1-x)y(1-z)(I-P_2)+(1-x)(1-y)1-z)(I-P_2-N_2)\end{array}$$

(17)

The calculation formula for AFEs’ average expected payoff is presented in Equation (18).

$${\overline{U}}_{w,1-w}=w{U}_w+(1-w)U_{1-w}$$

(18)

The change rate in AFEs’ probability of adopting the information sharing strategy is presented in Equation (19).

$${\displaystyle \frac{dw}{dt}}=w\left(U_w-{\overline{U}}_{w,1-w}\right)=w(1-w)\left(U_w-U_{1-w}\right)$$

(19)

The AFEs’ replicator dynamic equation can be derived from Equations (11)–(14), with the result presented in Equation (15).

$$\begin{array}{c}\hfill f\left(w\right)=w(1-w)\{xyz[-D_2+(1-\alpha )S-R_2+P_2+N_2]+x(1-y)z[-D_2+(1-\alpha )S-R_2+P_2+N_2]\\ \hfill +x(1-z)y[-D_2+(1-\alpha )S-R_2+P_2+N_2]+(1-x)yz(-D_2-R_2+P_2)+(1-x)y(1-z)(-D_2-R_2+P_2)\\ \hfill +(1-x)(1-y)z(-D_2-R_2+P_2+N_2)+(1-x)(1-y)(1-z)(-D_2-R_2+P_2+N_2)\}\end{array}$$

(20)

In summary, Equations (5), (10), (15) and (20) collectively form the evolutionary game replicator dynamic equations for the AFSC information collaboration model.

Based on the multi-agent game relationships and parameter settings, the proposed AFSC information collaboration model comprises four key stakeholders: the government, the ISP, farmers, and AFEs. The strategic behaviors of these entities are influenced by multi-agent, multi-variable interactions. Given the complex nonlinear interactions and time-delay characteristics inherent in evolutionary game models, this study employs system dynamics modeling to simulate and analyze the evolutionary processes and stability of the information collaboration model.

4. System Dynamics Modeling Construction

System dynamics (SD) is a quantitative modeling and simulation method designed for complex socio-economic systems. By constructing causal feedback mechanisms and stock-flow topologies, SD effectively analyzes the multi-dimensional dynamic coupling relationships and strategy diffusion paths among multiple agents [63,64]. Leveraging the Vensim simulation environment, this method enables parameter sensitivity analysis, strategy evolution tracking, and system stability testing, providing a methodological foundation for the dynamic simulation of multi-agent collaboration systems [65,66]. In this study, an AFSC information collaboration quadripartite evolutionary game system is developed based on SD modeling. This framework systematically illustrates the interactions and non-linear dependencies among the four game agents, facilitating quantitative analysis of their strategic decision-making pathways through simulation. Additionally, the SD model allows for an in-depth examination of the impact of various variables on strategic behaviors. This section focuses on analyzing the equilibrium convergence characteristics of the information collaboration system under different government incentive strategies, offering decision-support solutions for digitally empowered AFSC governance.

4.1. System Structure and Boundary Definition

Based on the quadripartite evolutionary game framework encompassing government, ISP, farmers, and AFEs, this study defines the system boundary of the system dynamics model through the core stakeholders of the AFSC. It should be noted that the current system only considers the influence of internal strategy dynamics without incorporating external factors such as policy fluctuations or technological advancements.

By analyzing cross-causal relationships among key components—incentive–penalty mechanisms, information monitoring protocols, information sharing, and information collaboration effect—the model clarifies strategic decision-making pathways across stakeholder groups. The proposed system dynamics architecture consists of four discrete yet interconnected subsystems: government subsystem, information service platform subsystem, farmers subsystem, and agri-food enterprises subsystem, each containing multiple variables with bi-directional causal relationships. These subsystems interact synergistically, collectively generating emergent behavioral patterns within the quadripartite evolutionary game system.

4.2. Causal Loop Diagram

By mapping the causal relationships and feedback mechanisms among the four stakeholders in the evolutionary game, we developed a causal loop diagram of the information collaboration system within the AFSC, as illustrated in Figure 4. To intuitively depict the internal feedback structure and inter-variable relationships, arrows are used to indicate the direction of causality between variables, while plus (+) and minus (−) signs represent positive and negative feedback loops, respectively.

As shown in Figure 4, the main components of the causal loop diagram include positive incentive by government, active monitoring by ISP, and information sharing by farmers and AFEs. Among these, government’s positive incentive serve as a core driving force within the system. By leveraging both positive (rewards) and negative (penalties) incentives, the government encourages ISP to enhance monitoring behavior and motivates farmers and AFEs to engage in information sharing. Acting as an intermediary, the ISP monitors the flow and quality of information, and its monitoring behavior is essential for improving the overall quality of collaboration. When both farmers and AFEs adopt information-sharing strategies, they create positive feedback loops for both the government and the ISP. If both parties simultaneously engage in information sharing, information collaboration is generated, forming a virtuous cycle that may lead the system toward a trajectory of sustainable development.

More specifically, the key causal relationships within the system are as follows: (1) Government’s positive incentive positively influence active monitoring by the ISP through a reward distribution mechanism (+) and further promote information sharing by farmers and AFEs through a reward redistribution mechanism (+). Penalties imposed on farmers and AFEs for concealing information also encourage their willingness to share information (+). (2) The ISP’s active monitoring enhances social benefits by improving data quality, which in turn reinforces the government’s incentives (+). It also positively influences information-sharing behavior among farmers and AFEs (+). (3) Information sharing by farmers or AFEs increases the ISP’s information stock, thereby enhancing its monitoring behavior (+). It also contributes to overall social benefits, further reinforcing government’s positive incentive (+). Additionally, mutual promotion exists between the information-sharing behaviors of farmers and AFEs (+). The resulting collaboration effect further strengthens their willingness to share information, creating a positive feedback loop. Conversely, rent-seeking costs negatively affect the willingness to share information among both farmers and AFEs (−).

4.3. Modeling and Behavioral Logic of Quadripartite Subsystems

4.3.1. Government Subsystem

Under the quadripartite evolutionary game framework, the government plays a central driving role in guiding participants within the AFSC toward information collaboration through incentive mechanisms. The structure of the government subsystem is illustrated in Figure 5.

The core of the government’s strategy lies in whether to adopt a positive incentive strategy. This decision-making behavior is represented by the variable $x\in [0,1]$, where $x=1$ indicates the government chooses a positive incentive strategy and $x=0$ indicates the adoption of a common incentive strategy. The expected returns under the two scenarios are represented by $U_x$ and $U_{1-x}$ in Equations (1) and (2), respectively. The evolutionary path of the government’s strategy is modeled using the replicator dynamic equation. Equation (4) describes the adjustment trajectory of the government’s incentive behavior driven by differences in returns, while the average expected payoff is given in Equation (3). Substituting into the replicator dynamic yields the evolutionary trajectory function of the government’s strategy, as shown in Equation (5). Definitions of the key parameters involved in this model are listed in Table 2.

The construction of the government subsystem is primarily based on evolutionary game theory, which posits that agents adjust their strategies over time in response to differences in payoffs. The replicator dynamic equation precisely describes this mechanism of behavioral adaptation. The government’s incentive behavior is further grounded in institutional economics and regulatory incentive theory, employing a combination of positive rewards and punitive measures to steer participants toward decisions that enhance information collaboration.

4.3.2. Information Service Platform Subsystem

Under the quadripartite evolutionary game framework, the information service platform, as the core of the information collaboration mechanism, undertakes key functions such as information collection, processing, storage, and distribution. Its strategic behavior directly affects the overall quality of information and efficiency of collaboration within the system. The subsystem model of the ISP is illustrated in Figure 6.

The core decision of the ISP lies in whether to adopt an active monitoring strategy. Its behavior is represented by the variable $y\in [0,1]$, where $y=1$ indicates the adoption of active monitoring, and $y=0$ indicates a passive response. The expected returns of the ISP under the two strategies are denoted by $U_y$ in Equation (6) and $U_{1-y}$ in Equation (7), respectively. The evolution of the ISP’s strategy is modeled through a replicator dynamic equation. Equation (9) depicts the adjustment path of monitoring behavior under different return drivers, and the average expected return is expressed in Equation (8). By substituting into the replicator dynamic equation, the evolutionary trajectory function of the ISP is obtained, as shown in Equation (10). Definitions of the key parameters in the model are listed in Table 2.

The modeling of the ISP subsystem is based on evolutionary game theory. The basic assumption is that the ISP continuously compares the payoffs of different strategies across repeated games and updates its behavior accordingly. The replicator dynamic equation reflects the trend in behavioral choices during the strategy evolution process. In addition, the ISP’s behavioral decisions are informed by the incentive and constraint theory in institutional economics. When facing governmental incentives and penalties, the ISP makes strategic decisions based on a comprehensive assessment of monitoring costs, risk control, and incentive returns. In the system, the ISP’s payoffs are also influenced by multiple factors, such as information synergy effects, data leakage risks, and rent-seeking behavior. Therefore, the model integrates asymmetric information theory and rent-seeking game theory to characterize and analyze potential incentive failure phenomena that may occur on the ISP.

4.3.3. Farmer Subsystem

Under the quadripartite evolutionary game framework, farmers are among the most fundamental actors in the AFSC in terms of both information production and utilization. Their information behavior directly influences the efficiency of information collaboration and the quality of data. The subsystem model of farmers is illustrated in Figure 7.

The core decision for farmers lies in whether to adopt an information sharing strategy. Their behavior is represented by the variable $z\in [0,1]$, where $z=1$ indicates the choice to share information and $z=0$ represents the choice to conceal it. The expected payoffs under these two strategies are denoted by $U_z$ in Equation (11) and $U_{1-z}$ in Equation (12), respectively. The evolution of farmers’ strategies is modeled using a replicator dynamic equation. Equation (14) describes the adjustment path of farmers’ information behavior under different payoff drivers, while the average expected return is presented in Equation (13). Substituting this into the replicator dynamics yields the evolutionary trajectory function of the farmers, as shown in Equation (15). Definitions of the key parameters in the model are listed in Table 2.

The modeling of the farmer subsystem is likewise based on the fundamental assumptions of evolutionary game theory: under conditions of bounded rationality and limited information, agents adjust their strategies by imitating those who achieve higher payoffs. When facing government incentives, ISP monitoring, and the strategic behaviors of other stakeholders, farmers make decisions about information sharing based on a comprehensive assessment of factors such as potential collaboration benefits, government rewards, information value, and risk costs. The model also incorporates external incentive and constraint mechanisms from institutional economics, as well as rent-seeking behavior assumptions from game theory, to characterize the strategic game between information sharing and free-riding among farmers.

4.3.4. Agri-Food Enterprises Subsystem

Under the quadripartite evolutionary game framework, agri-food enterprises, as key midstream actors in the AFSC, have a significant impact on the efficiency of the circulation process and the construction of collaborative mechanisms through their information-related behaviors. The subsystem model of AFEs is illustrated in Figure 8.

The core decision for AFEs lies in whether to adopt an information sharing strategy. Their behavior is represented by the variable $w\in [0,1]$, where $w=1$ indicates the choice to share information and $w=0$ indicates the choice to conceal it. The expected payoffs under the two strategies are denoted by $U_w$ in Equation (16) and $U_{1-w}$ in Equation (17), respectively. The evolution of the AFEs’ strategy is likewise modeled using a replicator dynamic equation. Equation (19) describes the adjustment path of AFEs’ information behavior under different payoff drivers, while the average expected return is presented in Equation (18). Substituting into the replicator dynamic equation yields the evolutionary trajectory function of AFEs, as shown in Equation (20). Definitions of the key parameters are provided in Table 2.

The construction of the AFEs subsystem also follows the principles of evolutionary game theory, emphasizing agents’ imitation-based learning behavior under bounded rationality. When facing varying degrees of ISP monitoring, government incentives, and the influence of farmers’ strategies, AFEs make strategic adjustments based on a comprehensive evaluation of factors such as information synergy effects, information value gains, data leakage risk costs, and the benefits from incentive or penalty mechanisms. Theories from institutional economics—particularly incentive compatibility constraints and information asymmetry—provide a theoretical foundation for modeling this subsystem. These theories complement the dynamic decision-making framework of evolutionary game theory and further reveal the behavioral evolution logic of AFEs in the context of information games.

4.4. Integrated System and Structural Overview

Building on the modeling analysis of the four subsystems, government, ISP, farmers, and AFEs, this section presents the integration of all subsystems into a unified system structure. The integrated model combines modular interactions and shared variables to form a closed-loop system featuring dynamic feedback and strategy evolution. As shown in Figure 9, the system structure illustrates the behavioral logic and interaction pathways among stakeholders in the process of information collaboration, providing a solid structural foundation for the subsequent simulation and policy optimization analysis.

4.5. System Dynamics Model Validation

4.5.1. Boundary Adequacy Test

To assess the boundary adequacy of the system dynamics model developed in this study, a behavior reproduction test was conducted by simulating the dynamic evolution of key system variables. A model validation variable $V\left(t\right)$ was constructed, defined as $V\left(t\right)={\displaystyle \frac{1}{3}}[y\left(t\right)+z\left(t\right)+w\left(t\right)]$. This variable represents the overall level of information sharing within the system and is influenced by multiple feedback mechanisms, such as government positive incentive and ISP active monitoring. As it reflects the behavioral evolution of multiple agents in the game, it serves as an integrated indicator of system-wide dynamics.

As shown in Figure 10, the variable $V\left(t\right)$ exhibits a typical S-shaped growth curve over time—characterized by rapid growth in the early stage, gradual deceleration in the middle phase, and eventual stabilization. This pattern indicates the presence of interacting reinforcing and balancing feedback loops within the current model boundary, thereby validating the model’s boundary adequacy.

4.5.2. Structural Validity Test

To ensure the structural rationality and logical consistency of the proposed system dynamics model, this study introduces extreme condition testing as a key method for verifying structural validity. This approach, rooted in the classical validation framework of system dynamics, involves assigning extreme initial values to key model variables and observing whether the system behaviors align with theoretical expectations. Such testing helps to determine whether the model structure maintains logical closure and stability.

Specifically, four representative scenarios were designed: (i) initial government incentive set to zero ($x=0$), (ii) initial ISP monitoring set to zero ($y=0$), (iii) full information sharing by farmers ($z=1$), and (iv) complete information concealment by AFEs ($w=0$). The evolution of strategy choices by each game participant under these conditions was then examined; the evolutionary trajectories are illustrated in Figure 11.

In the case of government (Figure 11a), with the exception of the scenario where $x=0$, all other settings led to a rapid increase and eventual stabilization of government positive incentive levels, indicating the government’s strong sensitivity to the behaviors of farmers and AFEs and its self-reinforcing feedback characteristics. Similarly, the ISP monitoring behavior (Figure 11b) exhibited clear positive feedback loops. Under the scenarios of full farmer information sharing ($z=1$) and full AFEs concealment ($w=0$), ISP monitoring intensified quickly and reached stability, suggesting a highly responsive mechanism to system-level behavioral changes.

From the perspective of farmers information-sharing behavior (Figure 11c), the probability of sharing information rapidly declined to nearly zero in the absence of government incentives ($x=0$), highlighting the critical role of incentives in promoting information collaboration. Conversely, under the no-monitoring scenario ($y=0$), farmers’ willingness to share still increased sharply to near one, demonstrating the dominant influence of government incentives and synergy gains. Interestingly, when AFEs concealed information ($w=0$), farmers showed a stronger willingness to share, reflecting a typical strategic stress response in the game.

Regarding AFEs’ information-sharing behavior (Figure 11d), the absence of government incentives ($x=0$) led to a significant drop in the probability of sharing, ultimately nearing zero. In contrast, under other scenarios (such as full farmer sharing or absence of monitoring), AFEs’ behavior exhibited an upward trend, eventually stabilizing at a sharing strategy. This trend indicates that AFEs decision-making is highly dependent on external incentives and the strategic choices of other stakeholders, confirming the presence of collaborative feedback mechanisms within the game structure.

In summary, the extreme condition tests confirm several key features of structural validity in the model: (1) Logical consistency: System responses under extreme conditions align with the expected evolutionary dynamics of game theory. (2) Interdependence of variables: Strategic behaviors among participants are mutually influential and form dynamic feedback loops. (3) Structural closure: The system can transition smoothly from extreme initial conditions to a stable distribution of strategies. (4) Stability: Most variables converge to equilibrium within a short time, reflecting the model’s dynamic controllability. Government and ISP adjustments occur significantly faster than those of AFEs and farmers, which is consistent with real-world game behavior patterns.

4.5.3. Parameter Consistency Verification

To evaluate the rationality and consistency of key parameter settings in the proposed system dynamics model, this study adopts a combined approach of trend consistency analysis and sensitivity testing. Specifically, model parameters were set and calibrated based on relevant policy documents, typical case studies, and the existing literature. The value ranges and initial settings are consistent with the realistic game-theoretic context of information collaboration in the AFSC.

On this basis, several representative state variables—including government incentive level $x\left(t\right)$, ISP monitoring behavior $y\left(t\right)$, farmers’ information sharing level $z\left(t\right)$, and AFEs’ information sharing level $w\left(t\right)$, were selected for simulation analysis. The results, as shown in Figure 12, demonstrate that, under multiple parameter configurations, the model outputs exhibit strong trend consistency. That is, the evolution trajectories of key variables over time align well with patterns observed in policy practice and previous empirical studies. For example, farmers’ willingness to share information increases in response to enhanced external incentives, and AFEs’ cooperation tendency rises with improved expected payoffs, indicating that the model has strong explanatory power in reflecting real-world behaviors.

Moreover, sensitivity analysis indicates that the model is highly responsive to key parameters such as the reward amount S, the reward allocation ratio $\alpha$, and the penalty allocation ratio $\beta$. Despite variations in these parameters, the system consistently exhibits dynamic stability and reasonable evolutionary trends, demonstrating the model’s structural robustness and predictive reliability.

In summary, this study systematically validated the proposed system dynamics model of the quadripartite evolutionary game through three levels: boundary adequacy testing, structural validity testing, and parameter consistency verification. The boundary behavior test confirmed that the model can reproduce typical S-shaped evolutionary trajectories within the defined system scope, reflecting reasonable feedback structures and behavioral dynamics. The extreme condition analysis verified the model’s logical structure and strategic responsiveness, demonstrating theoretical consistency, interdependence among variables, and structural closure. Parameter consistency analysis further showed that the model responds significantly to changes in key parameters, with output trends largely aligned with real-world expectations. These three layers of validation collectively confirm the scientific soundness of the model structure and the reliability of simulation results, laying a solid foundation for subsequent policy simulation and mechanism optimization.

5. System Dynamics Simulation and Strategy Optimization

Based on the in-depth analysis of the system dynamics model in Section 4, this study employs Vensim PLE 10.2.1 to construct a multi-agent system dynamics model for AFSC information collaboration. In this study, the simulation period is set to 10 years. The basic time parameters are defined as follows: initial time = 0, final time = 10, and time step = 0.0625 year. The parameter settings in this model are comprehensively determined based on policy context, relevant literature, and simulation requirements of the model. The specific parameter values and their sources are shown in the

Table 5. Initial values of the system dynamics model.

Variables	Initial Values	Reference Source	Variables	Initial Values	Reference Source
x	0.5	Simulation setting	y	0.5	Simulation setting
z	0.5	Simulation setting	w	0.5	Simulation setting
O	30	[67]	$O^{\prime }$	15	[67]
U	50	[67]	S	25	Simulation setting
$\alpha$	0.4	Simulation setting	${\alpha }_1$	0.5	Simulation setting
${\alpha }_2$	0.5	Simulation setting	E	10	[51]
$L_1$	5	[21]	$P_1$	12	[53]
$L_2$	6	[21]	$P_2$	15	[53]
$\beta$	0.7	Simulation setting	C	15	[21]
R	4	[21]	F	12	[53]
I	12	[51]	$I_1$	9	[51]
$I_2$	12	[51]	$\rho$	1.5	[51]
$\gamma$	1.3	[51]	$D_1$	15	[51]
$R_1$	3	[21]	$D_2$	18	[51]
$R_2$	4	[21]	$N_1$	5.5	[50]
$N_2$	6.5	[50]

.

5.1. Impact Analysis of Initial Strategy Combinations

To investigate the impact of initial strategies on the evolutionary process and system stability, this study simulates the dynamic evolution pathways of three initial strategy combinations: low (0.2, 0.2, 0.2, and 0.2), medium (0.5, 0.5, 0.5, and 0.5), and high (0.8, 0.8, 0.8, and 0.8) for the four game agents: the government, the ISP, farmers, and AFEs, as illustrated in Figure 12. The simulation results indicate that the behavioral evolution of each participant exhibits similar yet slightly distinct characteristics.

As shown in Figure 12a, the government ultimately converges to the positive incentive equilibrium, although the convergence speed varies. When the initial strategy combination is high (0.8), the system reaches stable convergence within 0.4375 years. In contrast, for medium (0.5) and low (0.2) initial strategies, the convergence times extend to 0.5625 years and 0.625 years, respectively. Additionally, the low initial strategy combination exhibits a faster growth rate in the early simulation phase, which is closely linked to the penalty feedback mechanism triggered by the default risks of farmers and agri-food enterprises.

As shown in Figure 12b, the ISP consistently converges to the active monitoring strategy across all initial conditions. However, the low initial strategy combination (0.2) demonstrates a stronger convergence advantage. Specifically, when the initial cooperation probability is 0.2, the system surpasses the probability trajectories of the medium and high combinations within 0.4375 years and achieves stable convergence first from the initially lower cooperation willingness of farmers and AFEs, which, under the dual reinforcement effects of the penalty redistribution mechanism and the government’s positive incentive, accelerates the ISP’s strategy adjustment process.

As illustrated in Figure 12c,d, both farmers and AFEs eventually converge to the information-sharing strategy, and the convergence speed positively correlated with the initial strategy combination probability; that is, the higher the initial strategy combination, the shorter the time required for stable convergence. A comparison of their convergence paths reveals significant temporal heterogeneity: while farmers achieved stable convergence in 1.8125 years, AFEs required up to 7.625 years to complete strategy adjustments (using a low initial strategy combination as an example). This discrepancy may stem from the higher cost sensitivity and more complex decision-making environment of AFEs, leading to a longer strategy adjustment period.

Overall, the impact of initial strategy selection on the evolutionary paths of the government, ISP, farmers, and AFEs is relatively minor, primarily reflected in differences in convergence speed, which influence the shape of the convergence curves. Given that trust deficits are common in the early stages of information collaboration system construction, this study adopts the low initial probability combination (0.2, 0.2, 0.2, and 0.2) as the benchmark scenario to systematically analyze the regulatory effects of government incentives on the stability of information collaboration.

5.2. Impact Analysis of Government’s Reward Amount

The reward amount S provided by the government under proactive incentives is a key factor influencing participation in information collaboration. When determining the reward amount, the government must consider both the social benefits generated by agricultural supply chain information collaboration and whether the management costs and reward expenditures can be balanced. Additionally, the reward allocation should partially offset participants’ costs and risks to effectively encourage broader participation in information collaboration. This section employs system dynamics simulation to reveal the regulatory mechanism of S on the stability of information collaboration.

As shown in Figure 13, when the benchmark reward S = 25, the system converges to the ideal equilibrium state (1, 1, 1, and 1). However, when S deviates from this threshold (S = 5 or S = 50), the equilibrium state is disrupted. When S = 5 (blue solid line in Figure 13), the information-sharing probabilities of farmers and AFEs decline to zero at 1.3125 years and 1.125 years, respectively, resulting in the failure of the information collaboration system. Although the ISP maintains an active monitoring strategy through the penalty redistribution mechanism and rent-seeking revenues, the government’s evolutionary trajectory initially rises briefly before rapidly declining, approaching zero at 1.875 years. This occurs because, in the early stages, the government initially favors positive incentive due to reduced expenditures; however, as farmers and AFEs withdraw from information sharing, the substantial loss of social benefits drives the government to shift toward a passive common incentive strategy. Eventually, the system stabilizes at (0, 1, 0, and 0), leading to the collapse of the information collaboration model.

When S = 15 (red dashed line in Figure 13), the system exhibits non-monotonic dynamic characteristics. To further analyze the long-term evolution, the simulation period is extended to 20 years. The results indicate that the government’s strategy eventually stabilizes at positive incentive, with a shorter convergence time than the benchmark scenario. This suggests that moderately reducing the reward amount can optimize policy execution efficiency by constraining costs. However, the ISP’s monitoring strategy declines after 2.935 years, reaching nearly zero by year 14. This decline occurs because the withdrawal of AFEs from information sharing reduces the ISP’s information value-added revenue. Combined with the reduction in government incentive intensity, the ISP’s regulatory cost coverage becomes insufficient, prompting a shift to a passive response strategy. Farmers experience an extended convergence cycle due to reduced rewards, yet they ultimately maintain a stable information-sharing state. In contrast, AFEs exhibit strategy oscillation: in the early stage, the weakened incentives and heightened monitoring pressure cause their information-sharing probability to first increase and then decline, leading to a shift to information concealment at 3.25 years. However, as the ISP ceases active monitoring, the government’s reward and penalty mechanisms, with increased reward distribution and penalty pressure, prompt AFEs to resume information sharing, ultimately stabilizing at 1. The system reaches a metastable state (1, 0, 1, and 1), where information synergy still exists, but the prolonged strategic evolution cycle and the absence of necessary verification mechanisms adversely affect both supply chain efficiency and sustainability.

When S = 50 (gray long-dashed line in Figure 13), the system experiences rapid destabilization due to the unsustainability of government finances. Specifically, the government’s strategy peaks at 0.125 years but collapses quickly thereafter. Farmers and AFEs initially show a brief increase in collaboration probability due to high rewards, but they rapidly withdraw as subsidies are exhausted. The ISP maintains its monitoring strategy through penalty distributions, but its evolutionary trajectory exhibits a pronounced plateau influenced by the strategies of other participants. Ultimately, the system stabilizes at (0, 1, 0, and 0), leading to the collapse of the information collaboration model.

In summary, the government must strike a precise balance between incentive effectiveness and fiscal sustainability when setting reward amounts to ensure the long-term viability of information collaboration. Excessive rewards (high S) trigger a fiscal deficit spiral, rendering policy implementation unsustainable; conversely, insufficient rewards (low S) lead to participant withdrawal, causing the collapse of the collaboration network. While suboptimal reward levels may enable the system to achieve limited stability through self-regulation, they inevitably reduce overall system efficiency. Therefore, determining an optimal reward range is crucial for designing an effective government incentive mechanism.

5.3. Impact Analysis of Static Reward and Penalty Allocation Ratios

This section, based on the previously constructed information collaboration game system for the AFSC, focuses on analyzing the nonlinear impact mechanisms of key regulatory parameters in government incentive strategies on the decision-making behaviors of heterogeneous agents. Using a system dynamics approach, the study concentrates on two dimensions: reward allocation under positive incentive and penalty redistribution under common incentive.

5.3.1. The Impact Analysis of the Reward Allocation Ratios

When the government adopts a positive incentive strategy, this study constructs a two-tiered reward allocation mechanism: the government-provided reward S is first allocated to the ISP that adopts an active monitoring strategy in proportion $\alpha$, and the remaining portion is then redistributed between farmers and AFEs according to the ratio ${\alpha }_1:{\alpha }_2$ (with ${\alpha }_1+{\alpha }_2=1$). Based on fixed parameters, with a penalty allocation ratio $\beta =0.7$ and a benchmark reward amount S = 25, this section analyzes how the reward allocation ratio $\alpha$ regulates the strategies of multiple agents, as well as how ${\alpha }_1$ and ${\alpha }_2$ influence the strategic adjustments of farmers and AFEs.

By setting $\alpha$ to four scenarios, including low (0.1), benchmark (0.4), high (0.7), and extremely high (0.9), as shown in Figure 14, the system explores the regulatory effects of the government’s reward allocation ratio on the evolutionary dynamics of multi-agent strategies. Under the benchmark scenario ($\alpha =0.4$, depicted by the red dotted line), the system converges to the ideal equilibrium state (1, 1, 1, and 1), indicating that the government, ISP, farmers, and AFEs all adopt proactive strategies. However, when $\alpha$ deviates from the benchmark value, the system exhibits significant strategy fluctuations and shifts in equilibrium.

When $\alpha =0.7$ (represented by the green dashed line in Figure 14), the ISP converges more rapidly due to the increased rewards, while the rewards for farmers and AFEs decrease. Their strategic evolution, however, exhibits an asymmetric response: for farmers, insufficient subsidy coverage of costs prolongs the convergence cycle, although they still maintain an information sharing strategy; for AFEs, the reduced subsidy fails to offset costs and risk premiums, causing their strategy to gradually shift away from sharing after peaking at 0.3125 years. Notably, the withdrawal of AFEs triggers reward secondary allocation, which causes a secondary increase in the strategy probability for farmers, thereby incentivizing them to achieve a stable information sharing state. This evolutionary process reveals that AFEs are more sensitive to changes in incentive strength. Although the exit of AFEs eliminates the information synergy effect, the government continues to maintain a positive incentive strategy through the social benefits generated by farmer collaboration; however, due to the decay of these social effects, the government’s policy convergence period is prolonged, as evidenced by the green dashed line in Figure 14a falling below the benchmark curve in the later stages of evolution. Ultimately, the system achieves a metastable equilibrium (1, 1, 1, and 0).

Under the extreme scenario where $\alpha =0.9$ (depicted by the gray long-dashed line in Figure 14), the system exhibits a polarized characteristic: the ISP, benefiting from high rewards, converges at an accelerated rate; farmers and AFEs completely withdraw from collaboration due to an incentive vacuum; and the government’s strategy peaks at 0.4375 years ($p=0.98<1$) but then rapidly declines as social benefits collapse, shifting towards a conservative common incentive strategy. Consequently, the ISP further reinforces its monitoring strategy through the penalty redistribution mechanism, leading to a “monitoring island” equilibrium (0, 1, 0, and 0), at which point the information collaboration system completely fails.

The study indicates that an optimal regulatory range exists for $\alpha$; deviations from this range trigger systemic risks: a low $\alpha$ results in monitoring failure, while a high $\alpha$ induces system collapse. This provides a reference for the government to further optimize the reward allocation ratio.

Furthermore, the study examines the regulatory effects of the reward allocation ratio ${\alpha }_1$ and ${\alpha }_2$ on the strategic choices of farmers and AFEs. As previously defined, ${\alpha }_1$ and ${\alpha }_2$ exclusively regulate the secondary allocation of government rewards between farmers and AFEs, ensuring that ${\alpha }_1+{\alpha }_2=1$. Given that a shift in equilibrium strategy does not affect the stability of the government and the ISP, this section focuses solely on analyzing the evolutionary trajectories of farmers and AFEs.

As shown in Figure 15, when ${\alpha }_1$ decreases from 0.5 to 0.33, the incentive intensity for farmers declines, making it insufficient to cover the cost of information sharing, thereby triggering a strategy reversal (illustrated by the green dashed line in Figure 15a). Meanwhile, since ${\alpha }_2$ increases to 0.67, AFEs achieve excessive cost compensation, leading to a faster convergence toward the information sharing strategy. Conversely, when ${\alpha }_2$ decreases from 0.5 to 0.33 while ${\alpha }_1$ rises to 0.67, AFEs withdraw from information sharing due to insufficient incentive coverage, whereas farmers accelerate convergence through residual reward secondary allocation (illustrated by the red dotted line in Figure 15b). The findings indicate that the effective regulatory range for ${\alpha }_1$ and ${\alpha }_2$ is relatively narrow as exceeding this range introduces systemic risks: a low ${\alpha }_1$ causes farmers to exit information sharing, while a low ${\alpha }_2$ leads to information concealment by AFEs. When formulating secondary reward allocation strategies, the government must pay close attention to the subsidy–cost coverage ratio for farmers and AFEs and adjust the allocation ratio in a timely manner to maintain the effective operation of the collaboration system. Due to its narrow regulatory range, the subsequent dynamic analysis in this study does not discuss this further, and ${\alpha }_1$ and ${\alpha }_1$ are set as fixed values with ${\alpha }_1={\alpha }_2=0.5$.

5.3.2. The Impact Analysis of the Penalty Allocation Ratio

When the government adopts a common incentive strategy, this study establishes a penalty redistribution mechanism to regulate information concealment behavior: farmers or AFEs that engage in information concealment are subject to fines, which are then distributed between the government and the ISP in proportion of $\beta :(1-\beta )$. Based on previous analyses, farmers and AFEs exhibit a strategy-following behavior, meaning that their strategy adjustments are influenced by the incentive strategies of the government, with a time lag relative to their evolutionary trajectories. This section systematically analyzes the response patterns of the four entities under the penalty redistribution mechanism.

The analysis is conducted using a fixed reward allocation ratio $\alpha =0.4$ and a benchmark reward amount S = 25. By setting $\beta$ to 0.4 (low), 0.7 (benchmark), and 0.9 (high), the system explores how the government’s penalty redistribution ratio affects multi-agent strategy regulation, as shown in Figure 16. Under the benchmark scenario ($\beta =0.7$, red dashed line), the system converges to the ideal equilibrium state (1, 1, 1, and 1), indicating that the government, ISP, farmers, and AFEs all adopt proactive collaboration strategies. However, when $\beta$ deviates from the benchmark, while no significant equilibrium shift occurs, the evolutionary trajectories exhibit certain fluctuations.

When $\beta =0.4$ (blue solid line in Figure 16), the government receives a larger share of penalty revenue, improving its incentive cost coverage and enabling a faster convergence to positive incentive. Farmers and AFEs respond to the government’s positive incentive by accelerating their convergence toward information sharing. However, the ISP demonstrates regulatory inertia, with its strategy convergence cycle extending to 2.25 years. This phenomenon is driven by a dual suppression mechanism: first, a reduced penalty share leads to insufficient cost coverage for active monitoring; second, the rapid convergence of farmers and AFEs prevents the ISP from extracting additional rent-seeking revenue from information verification.

When $\beta =0.9$ (green dashed line in Figure 16), the government’s penalty revenue decreases, causing insufficient cost coverage, yet the increase in social benefits still allows it to maintain a positive incentive strategy, albeit with a slightly longer convergence time. The government’s delayed strategy adjustment further extends the convergence cycles of farmers and AFEs. Meanwhile, the ISP, benefiting from increased penalty allocations, ensures sufficient cost coverage, leading to rapid strategic convergence.

The study reveals that the ISP is highly sensitive to changes in the penalty allocation ratio $\beta$: a low $\beta$ significantly delays convergence, whereas a high $\beta$ enhances strategy convergence. However, adjustments to $\beta$ affect the evolutionary trajectories of farmers and AFEs through the lag effect of government incentive policies. This suggests that optimizing ISP monitoring efficiency alone may weaken system resilience. Therefore, when determining the penalty allocation ratio $\beta$, the government must strike a balance between monitoring efficiency and the strategic execution efficiency of farmers and AFEs.

5.4. Design and Optimization of Dynamic Incentive Mechanisms

5.4.1. Benchmark Simulation Under Static Parameter Configuration

A comparative analysis of the simulation results in Figure 14 and Figure 16 indicates that, under the static parameter setting of $\alpha =0.4$ and $\beta =0.7$, the government, ISP, farmers, and AFEs collectively converge to an optimal stable equilibrium state. This outcome, as illustrated in Figure 17, confirms the relative optimality of this benchmark parameter configuration within a static framework. These results serve as a reference baseline for the subsequent design and comparative analysis of dynamic incentive strategies.

However, the simulation results indicate that, although all agents eventually converge to the proactive strategy combination (1, 1, 1, and 1), the convergence rates among heterogeneous agents vary significantly. The strategy convergence of AFEs lags considerably, leading to delays in the realization of information synergy effects and a reduction in system stability. As analyzed earlier, this phenomenon likely stems from the inability of static parameter settings to accommodate the dynamic responses within the evolutionary game system. On one hand, AFEs face higher costs associated with information sharing and greater risks of information leakage compared to farmers. On the other hand, they exhibit greater sensitivity to reward allocation, resulting in substantial differences in strategy adjustment rates under the same static parameters. Therefore, to optimize system-wide synergy and achieve a more stable and sustainable evolutionary state, this study introduces a dynamic parameter design and analyzes its necessity for enhancing the efficiency of information collaboration.

5.4.2. Simulation Analysis of Dynamic Reward Allocation Ratio

Based on the previous parameter analysis, the ISP serves as a dual-path regulatory hub within the government’s incentive mechanism. On the one hand, rewards under the positive incentive strategy are prioritized for allocation to the ISP. On the other hand, fines imposed on farmers and AFEs for information concealment are redistributed between the ISP and the government. To optimize the efficiency of strategy coordination among multiple entities, this study introduces the shadow variable $y\in [0,1]$ to dynamically adjust the reward or penalty allocation ratio. Additionally, to prevent insufficient incentives in the initial phase under dynamic parameters, the initial strategy combination is calibrated to (0.5, 0.5, 0.5, and 0.5) to ensure the stability of the evolutionary trajectory. The dynamic reward allocation ratio $\alpha$ is defined as follows:

$$\alpha =k_1{\alpha }_{0}y$$

(21)

In the above equation, $k_1$ represents the scaling adjustment coefficient for the reward allocation ratio, assumed to be 1. ${\alpha }_0$ is the static benchmark value of the reward allocation ratio, set at ${\alpha }_0=0.4$. $\tau$ denotes According to Equation (9), the system dynamics model was modified, and the simulation results are shown in Figure 18.

The simulation results based on the dynamic reward allocation ratio strategy indicate that all four game participants in the information collaboration model ultimately converge to the ideal equilibrium state (1, 1, 1, and 1). By comparing the evolutionary trajectories of the static strategy (Figure 17) and the dynamic reward allocation strategy (Figure 18), it is evident that the dynamic mechanism significantly enhances system convergence efficiency. Specifically, the convergence periods for farmers and AFEs to adopt information sharing shorten to 1.4375 years and 4.875 years, respectively (compared to 1.75 years and 6.875 years in the static model), representing efficiency improvements of 18% and 29%. Meanwhile, the convergence period for the ISP’s active monitoring extends to 1.0625 years (from 0.9375 years in the static model), reflecting a 12% efficiency reduction. The government’s positive incentive strategy reaches convergence in 0.5625 years (compared to 0.625 years in the static model), marking a 10% improvement in efficiency. These findings suggest that the design of dynamic $\alpha$ achieves a trade-off between incentives and efficiency, significantly improving the execution efficiency of information sharing by farmers and AFEs while maintaining the fundamental stability of the ISP. This, to some extent, facilitates information synergy and enhances overall system resilience. The next section will discuss the performance impact of the dynamic penalty allocation ratio $\beta$.

5.4.3. Simulation Analysis of Dynamic Penalty Allocation Ratio

This section further explores the regulatory mechanism of the dynamic penalty allocation ratio $\beta$ within the information collaboration system. Similar to the dynamic reward allocation ratio $\alpha$ in Section 4.3.2, the adjustment of $\beta$ is also based on the real-time feedback from the ISP’s active monitoring. Accordingly, the dynamic penalty allocation ratio $\beta$ can be expressed as

$$\beta =k_2{\beta }_{0}y$$

(22)

In the above equation, $k_2$ represents the scaling adjustment coefficient for the reward allocation ratio, assumed to be 1. ${\beta }_0$ is the static benchmark value of the reward allocation ratio, set at ${\beta }_0=0.7$. According to Equation (10), the system dynamics model was modified, and the simulation results are shown in Figure 19.

The simulation results based on the dynamic penalty allocation ratio strategy indicate that all four participants in the information collaboration model ultimately converge to the ideal equilibrium state (1, 1, 1, and 1). By comparing the evolutionary trajectories of the static model (Figure 17) and the dynamic penalty allocation model (Figure 19), it is evident that the dynamic mechanism significantly enhances system convergence efficiency. Specifically, the convergence periods for farmers and AFEs to adopt information sharing are reduced to 1.625 years and 5.5625 years, respectively (compared to 1.75 years and 6.875 years in the static model), reflecting efficiency improvements of 12.5% and 19%. The convergence period for the ISP’s active monitoring extends slightly to 1 year (compared to 0.9375 years in the static model), indicating a 6.7% efficiency reduction. Meanwhile, the government’s positive incentive convergence accelerates to 0.5625 years (from 0.625 years in the static model), yielding a 10% efficiency gain. These findings suggest that the design of dynamic $\beta$ achieves a trade-off between incentives and efficiency. While maintaining the stability of the ISP, it enhances the execution efficiency of farmers’ and AFEs’ information-sharing strategies, thereby fostering information synergy and increasing overall system resilience.

A comparative analysis of the simulation results for dynamic $\alpha$ (Figure 18) and dynamic $\beta$ (Figure 19) reveals similar optimization effects on the system. However, dynamic $\alpha$ yields greater efficiency improvements for farmers and AFEs, whereas dynamic $\beta$ results in lower efficiency losses for the ISP. Both strategies enhance government efficiency to a similar extent. This indicates that dynamic $\alpha$ is more effective in providing targeted incentives for farmers and AFEs, making it more suitable for the early stages of collaboration network development. Conversely, dynamic $\beta$ prioritizes improvements in ISP data quality, making it better suited for stable, mature, and high-quality managed AFSCs. Therefore, when formulating incentive policies, the government can adopt dynamic $\alpha$ during the initial phase to stabilize participation in information collaboration and introduce dynamic $\beta$ in the maturity phase to enhance the quality of information sharing. In Section 5.4.4, we further examine the performance impact of simultaneously implementing both dynamic $\alpha$ and dynamic $\beta$.

5.4.4. Simulation Analysis of Joint Dynamic Reward–Penalty Allocation Ratio

Based on the independent regulatory effects of dynamic $\alpha$ and dynamic $\beta$, this section examines the impact of their combined regulation on the information collaboration system. To achieve this, we simultaneously set $\alpha =k_1{\alpha }_{0}y$ and $\beta =k_2{\beta }_{0}y$. Under these conditions, the system dynamics model was modified, and the simulation results are shown in Figure 20.

The simulation results of the information collaboration model indicate that all four game agents ultimately converge to the ideal equilibrium state (1, 1, 1, and 1), albeit with some fluctuations in the evolutionary trajectory. By comparing the evolutionary trajectories of the static strategy (Figure 17) and the joint dynamic strategy (Figure 20), the following analysis is derived. For the ISP, the evolutionary trajectory exhibits an initial decline followed by a rebound, ultimately achieving stable convergence at 1.5 years (compared to 0.9375 years in the static model), representing a 60% delay. This phenomenon stems from a short-term conflict induced by the simultaneous adjustment of both parameters: in the early stages of collaboration, the dynamic reward–penalty allocation ratios remain below their benchmark values, significantly weakening the incentive effect and causing a temporary decline in the probability of active monitoring by the ISP. However, as the system adapts, the ISP gradually readjusts toward an active monitoring strategy. The convergence periods for farmers and AFEs further shorten to 0.5 years and 0.5625 years (compared to 1.4375 years and 5.6875 years in the static model), reflecting efficiency improvements of 93.8% and 90.3%, respectively, which demonstrate the marginal incentive amplification effect of the dynamic mechanism. Similarly, the convergence time for the government’s positive incentive strategy shortens to 0.5 years (from 0.625 years in the static model), representing a 20% improvement in efficiency. These findings suggest that the joint dynamic reward–penalty strategy re-configures system efficiency through asymmetric regulatory effects: while the monitoring efficiency of the ISP is significantly suppressed, it is offset by substantial improvements in the information sharing efficiency of farmers and AFEs. Notably, the response time of AFEs is reduced by approximately 90%, indicating that the joint dynamic strategy effectively overcomes organizational inertia. Meanwhile, although the efficiency of the ISP is severely compromised, the overall efficiency of information collaboration reaches its peak, aligning with the Kaldor–Hicks improvement criterion for optimization.

By comparing the evolutionary trajectories of the joint dynamic reward–penalty strategy (Figure 20) with the single dynamic strategies (Figure 18 and Figure 19) and the static strategy (Figure 17), it is evident that dynamic strategies significantly impact overall system efficiency. However, their specific applicability requires a trade-off between efficiency and fairness. Specifically, under the static strategy, the system maintains a low-efficiency equilibrium, with the convergence period for AFEs extending up to 5.6875 years. Under single dynamic strategies: the dynamic $\alpha$ strategy achieves a balanced trade-off between efficiency and fairness, significantly improving the efficiency of farmers and AFEs while minimizing losses in ISP efficiency; the dynamic $\beta$ strategy, while also enhancing the benefits for farmers and AFEs, is more oriented toward minimizing efficiency losses in the ISP’s active monitoring. The joint dynamic strategy, however, triggers a polarization effect, where farmers and AFEs achieve ultra-fast convergence at the cost of a substantial increase in the ISP’s vulnerability.

Based on the above analysis, governments can select different dynamic strategies depending on specific application scenarios. If ensuring data quality is the primary objective (e.g., agri-food traceability systems), the single dynamic $\beta$ strategy is recommended. If the priority is to promote supply chain information collaboration (e.g., during the early stages of supply chain digital transformation), the joint dynamic strategy is preferable. If a gradual approach to industrial upgrading is needed (e.g., pilot transitions in regions with weak digital infrastructure), the single dynamic $\alpha$ strategy or even a static strategy may be the most suitable choice.

5.5. Analysis of Policy Optimization Performance and Sustainability

5.5.1. Performance Analysis of Dynamic Incentive Mechanisms

The simulation results in Figure 17 reveal that, although all four quadripartite agents ultimately converge to the ideal cooperative strategy combination (1, 1, 1, and 1) under static incentives, there exists a significant discrepancy in convergence speeds among heterogeneous participants. In particular, AFEs exhibit substantially delayed strategy convergence, which impedes the realization of information synergy and undermines system stability. This inefficiency is primarily attributed to the limitations of static parameter settings, which fail to adapt to the dynamic responsiveness inherent in evolutionary game systems. On one hand, AFEs bear higher information sharing costs and face greater risks of information leakage compared to farmers; on the other, they display stronger sensitivity to the distribution of rewards, resulting in considerable variations in adjustment speeds under uniform incentive schemes.

To address this issue and enhance systemic coordination and long-term sustainability, dynamic incentive mechanisms were introduced and tested. The results are summarized as follows:

(1) Dynamic reward allocation ratio ($\alpha$) (as shown in Figure 18). This mechanism notably accelerates convergence among key information-sharing agents. The convergence times for farmers and AFEs are reduced to 1.4375 years and 4.875 years, respectively, representing 17.9% and 29.1% improvements over the static model. The convergence of the government’s incentive behavior also improves by 10%, while the ISP experiences a moderate delay in monitoring response (−13.3%). Overall, the dynamic $\alpha$ mechanism facilitates a trade-off between ISP stability and front-end cooperation efficiency, thereby enhancing the resilience of the collaborative system.

(2) Dynamic penalty allocation ratio ($\beta$) (as shown in Figure 19). This strategy yields similar improvements, with convergence times for farmers and AFEs improving by 12.5% and 19.1%, respectively. The delay in ISP responsiveness is less pronounced (−6.7%), while the government maintains a 10% improvement in convergence efficiency. Compared with dynamic $\alpha$, this mechanism prioritizes ISP stability over front-end acceleration, indicating a better fit for mature systems with higher quality requirements.

(3) Joint dynamic reward–penalty allocation ratio (as shown in Figure 20). While the combined strategy maximizes the acceleration of convergence, particularly for farmers and AFEs, it also introduces polarization. The over-acceleration of front-end actors compromises the ISP’s monitoring stability, revealing potential fragility under extreme incentive imbalances. This shows that the joint dynamic strategy must be carefully implemented under certain conditions and is more suitable for the more radical reform environment.

Based on the above analysis, dynamic incentive mechanisms exhibit distinct strengths under different application scenarios and thus should be adopted contextually to balance efficiency, stability, and fairness. Specifically, the dynamic $\alpha$ strategy proves more effective in mobilizing information-sharing behavior among farmers and AFEs, making it particularly suitable for the early stages of supply chain digital transformation, where active participation is critical. In contrast, the dynamic $\beta$ strategy places greater emphasis on maintaining the regulatory stability of ISP and is thus better suited for mature systems where data quality assurance, such as in agri-food traceability applications, is the primary objective. The joint dynamic $(\alpha +\beta )$ strategy achieves the highest overall system efficiency by accelerating front-end convergence but also increases the risk of polarization among agents, which may compromise system robustness if not properly managed. Consequently, governments should adopt differentiated strategies according to specific policy objectives and infrastructure conditions:

• For scenarios prioritizing data quality (e.g., product traceability), the dynamic $\beta$ strategy is recommended;
• For early-stage reform or rapid mobilization, the joint dynamic $\alpha$ + $\beta$ strategy is preferable;
• For gradual pilot projects in underdeveloped regions, the dynamic $\alpha$ or even static strategies remain viable.

5.5.2. Policy Sustainablity Analysis

Based on the previous sensitivity analysis of the reward amount, it is evident that the reward level S is a key variable influencing the stability of the system. A reduction in the reward amount would weaken the willingness of farmers and AFEs to engage in information collaboration, thereby undermining the overall stability of the system. However, the long-term provision of high subsidies is often unsustainable in practice as it is constrained by fiscal budgets and may lead to path dependency and resource mis-allocation. Therefore, it is necessary to incorporate long-term sustainability into the design of incentive mechanisms.

As the marginal utility of fiscal subsidies declines and budgetary constraints intensify, the information collaboration system faces the challenge of maintaining stable operations in the context of fiscal tapering. Therefore, it is essential to design incentive mechanisms with endogenous driving capabilities to ensure the sustained stability of strategic choices and the long-term resilience of the system. Based on practical feasibility and following the triadic policy logic of “fiscal + institutional + market” instruments, this study constructs a diversified incentive pathway. The core idea is as follows: in the initial stage of building the information collaboration mechanism, the government provides relatively high fiscal rewards to stimulate participation. Once the system reaches a preliminary level of stability, fiscal spending is gradually reduced while institutional incentives, such as tax exemptions, credit-based evaluations, and preferential government procurement, are introduced. Simultaneously, the development of a market-based revenue-sharing mechanism within the ISP is encouraged. This allows the system to gradually reduce its dependence on fiscal subsidies, thereby achieving policy sustainability and long-term systemic stability.

Accordingly, we simulate a realistic scenario within the existing system dynamics model to test whether the system possesses sufficient adaptability and strategic stability under conditions where fiscal subsidies are gradually phased out and institutional and market mechanisms are progressively introduced. The main adjustments are as follows: (1) Phased reduction in fiscal incentives. In the simulation, the government maintains a high level of fiscal subsidies ($S=25$) in the initial stage (year = 1). Thereafter, the subsidy amount S decreases linearly year by year, reaching a low level ($S=5$) by year = 10, thereby simulating a fiscal tapering process. (2) Embedding of institutional incentives. The costs of information sharing for farmers and AFEs, as well as the cost of active monitoring for the ISP, are set to decrease linearly starting from year 1. By year 10, they decline to 5, 8, and 5, respectively, simulating the effects of institutional interventions such as tax exemptions, credit access, and process optimization. (3) Introduction of market mechanisms. From year 1, a revenue-sharing mechanism is established for ISP, introducing a total profit-sharing quota B, which is distributed to farmers and AFEs according to parameters ${\alpha }_1$ and ${\alpha }_2$. This simulates the cultivation of market-based profitability and the shaping of actors’ return expectations. The specific simulation results are presented in Figure 21.

As shown in Figure 21, under the guidance of the multi-dimensional incentive pathway, the strategic evolution trajectories of all four game participants exhibit strong convergence, ultimately reaching the ideal equilibrium state (1, 1, 1, and 1). This indicates that the system has achieved both strategic coordination and long-term stability. From the trajectory patterns, although subsidies decrease year by year, the continuous decline in information-sharing and monitoring costs, coupled with the gradual establishment of the ISP’s revenue-sharing mechanism, ensure that the proactive strategies of farmers and AFEs continue to rise without any reversal. This demonstrates that institutional and market incentives effectively compensate for the withdrawal of fiscal support. The ISP maintains its monitoring responsibilities while also gaining a certain degree of self-motivation through the profit-sharing mechanism, thereby preserving its stability. Meanwhile, although the government’s incentive intensity diminishes over time, its strategy remains unchanged. This suggests that, with well-designed mechanisms, fiscal subsidies are not the only viable approach; strategic sustainability relies more on the coordinated support of institutional and market-based incentives.

The above simulation results indicate that, while standalone fiscal subsidies can rapidly initiate the coordination mechanism, their incentive effects are difficult to sustain over time. In contrast, building a diversified incentive system that integrates fiscal subsidies, institutional incentives, and market mechanisms can effectively enhance the stability and adaptability of the AFSC information collaboration system. Based on these findings, the following policy recommendations are proposed:

• Initiation Phase: It is recommended that the government provide high-intensity fiscal subsidies to stimulate the initial participation of farmers and AFEs, while also supporting the ISP in establishing its monitoring and benefit-sharing mechanisms.
• Transition Phase: Gradually reduce fiscal subsidies and shift the policy focus toward institutional incentives (such as tax reductions, green certifications, and government procurement) to lower collaboration costs and strengthen the market-oriented functions of the ISP.
• Mature Phase: Encourage the ISP to generate revenue based on the value of information and redistribute it to farmers and AFEs, thereby promoting the internalization and normalization of the coordination mechanism and ultimately achieving the self-replacement of policy incentives.

By dynamically adjusting the incentive structure and introducing diversified policy instruments, the government can not only enhance the efficiency of coordination but also ensure the sustainable evolution of the collaborative system under fiscal constraints, thereby providing a stable institutional foundation for digital agriculture and rural revitalization.

6. Discussion

This study constructs a quadripartite evolutionary game framework for AFSC information collaboration, integrating the strategic behaviors of governments, ISPs, farmers, and AFEs. Unlike the existing literature that predominantly focuses on the scenario-based application of digital technologies such as blockchain or IoT [21,51,54], this research emphasizes the dynamic interaction mechanisms and parameter sensitivities that shape the stability and evolution of collaborative systems in the AFSC. By incorporating system dynamics simulations, we provide a deeper understanding of how heterogeneous stakeholders respond to various incentive structures, and how these responses influence the overall trajectory of information sharing within the system.

The simulation findings reveal that, although all agents eventually reach a collaborative equilibrium under appropriate conditions, significant temporal heterogeneity exists in their convergence patterns. Governments and ISPs typically adjust more rapidly, especially under positive incentive strategies, while AFEs exhibit slower convergence due to higher cost sensitivity, information sharing risk premiums, and organizational inertia. Farmers, in contrast, display a relatively moderate convergence speed. Their behavioral response is largely shaped by two factors: (1) the government policy signals and (2) the relative benefit of sharing versus withholding information. When the reward mechanisms are clear and stable, farmers tend to follow ISPs and governments toward cooperative strategies.

The analysis further demonstrates that the effectiveness of government-led incentives hinges not only on the reward amount but also on how rewards and penalties are allocated across different agents. From the perspective of reward amounts, either inadequate incentives or excessive incentives may disrupt system equilibrium: the former reduce members’ enthusiasm for participation, while the latter lead to system collapse due to fiscal unsustainability. From the perspective of reward allocation mechanisms, particularly the reward allocation ratio between ISPs and information providers (farmers and AFEs), there is a significant influence on strategy selection and convergence trajectories. An imbalanced incentive structure may trigger an incentive crowding-out effect, whereby excessive rewards to ISPs lead to overly rapid convergence, undermining the system’s regulatory balance, while under-incentivized actors may withdraw from information-sharing activities. In contrast, regarding penalty allocation mechanisms, although the penalty allocation ratio does not alter the system’s equilibrium state, it does affect the convergence dynamics and behavioral stability of the participating agents.

To address the persistent lag in AFE strategy convergence, we introduce dynamic reward and penalty mechanisms that adapt to system evolution. The simulation results show that dynamic allocation strategies, especially joint dynamic reward–penalty adjustments, can substantially improve convergence speed and system coordination, particularly for AFEs. However, this efficiency gain comes at a cost: the ISP’s monitoring stability is compromised, making the ISP more susceptible to volatility. These trade-offs suggest that dynamic strategies, while promising, require mature institutional environments and robust digital infrastructures to mitigate potential regulatory fragility. The choice of static versus dynamic incentive design should therefore be context-dependent, balancing efficiency with governance capacity.

Importantly, this study also contributes to the broader discussion on the sustainability of incentive mechanisms in digital agriculture. While fiscal subsidies play a critical role in initiating collaboration and reducing early-stage participation costs, over-reliance on direct government subsidies may be unsustainable in the long term. To address this, our results suggest that a more resilient pathway involves transitioning from single fiscal incentives to a diversified incentive framework that combines fiscal tools with an institutional design (e.g., taxes or credit loans) and market-based mechanisms (e.g., data trading or a benefit distribution mechanism). Such a shift not only enhances the long-term stability of information collaboration but also aligns with the broader policy goals of agricultural modernization and digital transformation.

In summary, this study provides empirical and theoretical insights into the incentive design influence in shaping the dynamics of AFSC information collaboration. It offers practical implications for policymakers aiming to construct robust, scalable, and adaptive governance systems in the context of agricultural digitalization.

7. Conclusions

This study investigates the dynamic evolution of information collaboration in AFSCs through a quadripartite evolutionary game model and system dynamics simulation, focusing on the design and optimization of government-led incentive mechanisms. The key findings are as follows:

(1) The proposed model confirms the effectiveness of evolutionary game theory in capturing multi-agent strategic interactions in AFSCs. Under reasonable parameters, the system converges to a stable collaborative equilibrium, validating the incentive compatibility of the designed mechanism.

(2) The initial strategy combinations significantly affect convergence. In the early policy stages, low enthusiasm among non-government agents weakens system stability. The government should enhance participation confidence and adjust the incentive strategies to accelerate convergence—especially for AFEs, whose strategic evolution is notably slow.

(3) The simulation results show that the reward amount S is a sensitive parameter: moderate levels sustain collaboration, while overly high or low values destabilize the system. The government must carefully calibrate the reward levels to offset sharing risks without creating fiscal overdependence.

(4) The reward and penalty allocation scheme is the core of the incentive mechanism. A two-tiered reward system and penalty redistribution improve coordination. However, the effectiveness of allocation ratios ($\alpha$, $\beta$, $\alpha 1$, $\alpha 2$) varies among agents, requiring precision tuning to balance regulatory efficiency and participant responsiveness.

(5) Dynamic incentive strategies enhance policy adaptability. Adjusting $\alpha$ and $\beta$ based on ISP monitoring behavior improves the information-sharing efficiency for farmers and AFEs. While joint dynamic strategies maximize collaboration gains, they may reduce ISP efficiency, necessitating compensatory mechanisms aligned with the Kaldor–Hicks principle. Based on this, this study constructs a scenario-based incentive configuration frame for information collaboration in AFSCs. A three-tiered decision-making framework of “scenario–strategy–parameter” is proposed to assist the government in precisely aligning industry needs and setting appropriate parameters. Specifically: If the priority is to ensure data quality (e.g., in agri-food quality traceability scenarios), adopting a single dynamic $\beta$ strategy is recommended. If the goal is to actively promote AFSC information collaboration (e.g., in the early stages of supply chain informatization reform), a dynamic joint strategy is recommended. If the aim is to progressively upgrade the industry (e.g., during the pilot transformation stage in regions with weak informatization foundations), adopting a single dynamic $\alpha$ strategy or even a static strategy is advisable.

(6) To ensure long-term policy sustainability under fiscal constraints, this study proposes a diversified incentive pathway that evolves over time. In the early stage, fiscal subsidies are employed to stimulate the initial participation and activate the information collaboration mechanism. As the system begins to stabilize, the policy focus gradually shifts toward institutional instruments—such as tax relief, credit access, and preferential procurement—which help to reduce collaboration costs and maintain engagement. In the mature phase, market-oriented mechanisms are introduced, particularly platform-based revenue-sharing schemes that enable participants to derive value from information exchange and coordination.

This study presents a quantitative regulatory framework based on evolutionary game theory for information collaboration in AFSCs, identifying the critical conditions for multi-agent incentive compatibility. It also provides a theoretical foundation and operational guidelines for the government to develop differentiated and adaptive incentive policies. Additionally, the evolutionary game model and reward–punishment strategies used in this study are applicable to various research objects (such as other companies or industries), particularly in scenarios where the government plays a key role.

Nevertheless, this paper still has several limitations that need to be addressed. First, the current model primarily focuses on the internal strategy evolution process of the system, assuming that policy incentives are stable and predictable, without fully considering the impact of external complexities and structural uncertainties (such as market fluctuations, policy changes, technological innovations, or systemic shocks). These factors may disrupt the strategic choices of multiple agents and interfere with the operational trajectory and policy effectiveness of the collaborative system. Future research could incorporate uncertainty modeling approaches to enhance the adaptability of policy recommendations and the stability and resilience of the system. Second, to focus on strategy evolution paths and simplify the model-solving process, this paper adopts several institutional assumptions, such as “government reward and penalty mechanisms are fully enforceable” and “strategies are adjusted and feedback is received immediately”. While these assumptions help to maintain clarity in the model structure and the controllability of system stability, in reality, institutional implementation often experiences delays, and participants may adopt irrational response strategies such as evasion or postponement. Future research could consider introducing feedback delay mechanisms, institutional friction parameters, or incomplete execution scenarios to better align the simulation results with reality and to enhance the model’s capacity to evaluate policy feasibility. Third, although the evolutionary game model reflects bounded rationality, the current version does not deeply characterize behavioral differences among game agents in terms of risk perception, benefit evaluation, and strategy trade-offs. In practice, farmers, AFEs, and governments are often influenced by psychological expectations, experiential biases, or fairness preferences, leading to decisions that are “not optimal but acceptable”. Future studies could incorporate behavioral game theory to introduce cognitive mechanisms such as prospect theory, mental accounting theory, and social preferences, thereby constructing strategy evolution models with greater behavioral heterogeneity and explanatory power—enhancing both the realism and applicability of the model in real-world scenarios.