Source-linked AI summary
COVID-ABS: An Agent-Based Model of COVID-19 Epidemic to Simulate Health and Economic Effects of Social Distancing Interventions
Petrônio C. L. Silva, Paulo V. C. Batista, Hélder S. Lima, Marcos A. Alves, Frederico G. Guimarães, Rodrigo C. P. Silva
TL;DR
The paper addresses the difficulty of predicting the epidemiological and economic implications of COVID-19 social-distancing policies. It develops COVID-ABS, a SEIR agent-based model of a society and its economy, and finds lockdown-based scenarios best preserve lives, while masks with partial isolation offer a more realistic alternative when lockdown is infeasible.
Problem
COVID-19 social-distancing interventions affect public health and the economy, but their short- and medium-term implications are difficult to predict in complex societies.
Method
The paper develops a SEIR agent-based model representing people, houses, businesses, government, and healthcare, and simulates seven intervention scenarios.
Results
Lockdown and conditional lockdown best preserve lives, while masks with 50% social isolation are the best alternative when lockdown cannot be implemented; vertical isolation is ineffective.
Takeaways & Limitations
COVID-ABS can support policy planning by simulating epidemiological and economic effects of interventions and can be adapted to other contexts with corresponding data.
Takeaways & Limitations
The model hard-codes social constructs and periodic routines, although users can adapt them and adjust parameters across scales from communities to countries.
Abstract
from arXiv · showhide
The COVID-19 pandemic due to the SARS-CoV-2 coronavirus has directly impacted the public health and economy worldwide. To overcome this problem, countries have adopted different policies and non-pharmaceutical interventions for controlling the spread of the virus. This paper proposes the COVID-ABS, a new SEIR (Susceptible-Exposed-Infected-Recovered) agent-based model that aims to simulate the pandemic dynamics using a society of agents emulating people, business and government. Seven different scenarios of social distancing interventions were analyzed, with varying epidemiological and economic effects: (1) do nothing, (2) lockdown, (3) conditional lockdown, (4) vertical isolation, (5) partial isolation, (6) use of face masks, and (7) use of face masks together with 50% of adhesion to social isolation. In the impossibility of implementing scenarios with lockdown, which present the lowest number of deaths and highest impact on the economy, scenarios combining the use of face masks and partial isolation can be the more realistic for implementation in terms of social cooperation. The COVID-ABS model was implemented in Python programming language, with source code publicly available. The model can be easily extended to other societies by changing the input parameters, as well as allowing the creation of a multitude of other scenarios. Therefore, it is a useful tool to assist politicians and health authorities to plan their actions against the COVID-19 epidemic.
1. Introduction
The paper introduces COVID-ABS, a SEIR agent-based model for simulating COVID-19’s epidemiological and economic effects under social-distancing interventions. It evaluates seven scenarios, finding lockdown-based policies best for preserving lives, while masks combined with partial isolation offer a more realistic compromise.
- COVID-ABS models a closed society through agents representing people, houses, businesses, government, and healthcare.
- Agent-based simulation is used because social-distancing policies interact with complex, nonlinear, stochastic societal behavior that is difficult to describe mathematically.
- The model and experiments are publicly available, and the paper presents COVID-ABS as a tool for politicians and health authorities planning epidemic responses.
- The study assesses seven intervention scenarios, including no intervention, lockdown, conditional lockdown, vertical isolation, partial isolation, masks, and masks with 50% social-isolation adhesion.
- Lockdown and conditional lockdown best control infections and deaths, but produce the greatest economic losses and require government enforcement of isolation.
- Masks combined with partial isolation flatten the infection curve, reduce economic effects relative to lockdown, and may be more realistic because of social cooperation.
2. Related Work
Related work is dominated by equation-based SIR and SEIR models, while fewer studies use agent-based models to examine COVID-19’s economic effects. The paper positions its parameterized SEIR agent-based model as transferable across contexts when corresponding data are available.
- Most epidemic simulations in the literature use equation-based SIR or SEIR models, while agent-based COVID-19 studies remain less common.
- Prior agent-based studies examined healthcare demand, urban transmission under isolation, and combinations of isolation, case isolation, and travel restrictions.
- Few studies examined COVID-19’s economic impacts with agent-based models; one estimate projected a 5.3% annual-GDP production loss from a one-month Tokyo lockdown.
- Other work proposed tools combining health, social, and economic impacts of interventions such as school closures, working from home, and subsidies.
- This paper uses Brazilian data to simulate seven health and economic intervention scenarios, while its fully parameterized model can be transferred when corresponding data are provided.
3. COVID-ABS: Proposed Agent-based System Modeling
COVID-ABS models a closed society as interacting agents whose iterative behaviors generate epidemiological and economic response variables. Its contact-based SEIR dynamics, social routines, and configurable parameters support intervention simulation while introducing computational and modeling constraints.
- System representation: COVID-ABS represents humans, families, businesses, government, and healthcare within a shared finite environment.Agents have distinct attributes and actions, and their interactions cover social, epidemiological, and economic relationships.
- System representation: The iterative model takes parameter set P and produces observable epidemiological or economic response variables Y_t from agent states Θ_t.Each iteration advances the internal states of n agents according to agent type, parameters, and discrete time.
- Simulation dynamics: The model simulates hourly, daily, weekly, and monthly agent actions, including movement, contagion, commerce, taxes, labor, and household expenses.A1 movement can create proximity to A1, A2, or A3 agents, while A1-A3 contact supports commercial transactions.
- Simulation dynamics: O(n^2) complexity results from checking the distance between every pair of agents at each iteration.Distances determine epidemiological or economic contacts when agents are within threshold δ.
- Epidemiological modeling: Contagion depends on proximity threshold β1 and contact probability β2, while medical progression uses stochastic age-grouped rates and limited healthcare capacity.Some parameters were empirically estimated so epidemiological responses correspond to those of a SEIR model.
- Model scope and calibration: Social constructs and periodic routines improved economic response variables compared with earlier model versions, while remaining hard-coded but user-adaptable.The parameter set can represent settings ranging from a single community to a complete country.
4. Experimental Methodology
The study evaluates seven social-distancing scenarios with repeated two-month simulations, comparing epidemiological and economic response variables against defined metrics and a no-pandemic baseline.
- Scenario design: Seven scenarios reflecting adopted or hypothetical social-distancing interventions were formulated and implemented in the COVID-ABS Python package.The source code is publicly available.
- Simulation protocol: Each scenario was executed 35 times for T = 1,440 hourly iterations, covering exactly two months and one accounting cycle.Salary and tax payments occur on the 30th day of the month.
- Evaluation metrics: The epidemiological comparison uses infection and death variables condensed into the infection peak IP and peak-time metric TIP.The intervention goal is to reduce deaths and keep hospitalizations below healthcare capacity.
- Economic evaluation: Economic effects are assessed through wealth evolution and changes in wealth relative to a no-pandemic baseline scenario B.The baseline isolates economic dynamics for comparing intervention impacts.
5. Results and Analysis
Results are presented for seven Brazil-based intervention scenarios using a transferable parameterized model, with the no-pandemic baseline providing an artificial economic reference.
- Study scope: The study evaluates seven intervention scenarios and allows the model to be transferred to other societies by changing social, demographic, and economic parameters.Additional scenarios can also be created for new regions.
- Baseline: The no-pandemic baseline uses β9 ← 0 and β10 ← 1 to represent economic behavior without a pandemic.It serves as the reference for comparing the other scenarios.
- Baseline: The baseline economic output is artificial and does not represent the reality of any country, although it is based on pre-pandemic projections.The projections are tied to the parameters listed in Table 2.
- Baseline: In the baseline, GDP indicates recession-like dynamics: people and government lose wealth while businesses remain near equilibrium after accounting for labor and tax expenses.The result is consistent with predictions of stagnation in Brazil.
- Scenario 1: Do Nothing: Without intervention, economic curves resemble the baseline, but infection surpasses healthcare capacity and drives deaths upward, making this the most catastrophic scenario.The economic resemblance does not prevent a high number of lost lives.
5.3. Scenario 2: Lockdown
Lockdown restricts population mobility throughout the simulation to minimize viral spread, while conditional lockdown activates restrictions at an infection threshold and releases them after the infection level falls.
- Scenario 2: Lockdown: Unconditional lockdown keeps all people in their houses from t = 0 to T by suppressing walking and work routines.The policy is designed to minimize viral spreading and save as many lives as possible.
- Scenario 2: Lockdown: Under lockdown, deaths are set to Dt ← 0 and infections remain below healthcare capacity in the described simulation.The infected broadly contact only housemates.
- Scenario 2: Lockdown: After two months of lockdown, businesses lose 20% of their GDP share because people cannot generate wealth while businesses continue paying labor expenses and taxes.The scenario is described as the worst for industry.
- Scenario 3: Conditional Lockdown: Conditional lockdown activates when It ≥ 0.05 and releases when It ≤ 0.05, applying the same mobility restrictions as Scenario 2.The policy therefore responds to the infection curve rather than remaining continuously active.
- Scenario 3: Conditional Lockdown: Conditional lockdown controls viral spreading, but recession occurs during restrictions and business performance recovers after release.Its business wealth remains below the no-pandemic baseline but above the complete-lockdown curve.
- Scenario 3: Conditional Lockdown: Conditional lockdown is less efficient than complete lockdown and depends on effective testing and governmental enforcement of isolation.The healthcare system must reliably estimate infection levels to trigger the policy.
5.5. Scenario 4: Vertical Isolation
Vertical isolation keeps known infected people and risk groups at home while allowing younger, lower-risk people to work, but its underlying assumption is fragile and the simulated outcomes resemble doing nothing.
- Policy definition: Vertical isolation confines known infected people and risk groups, including elderly people and those with pre-existing diseases, while others work normally.In the model, people over 65, under 18, and symptomatic people stay home.
- Policy assessment: The policy assumes that people outside risk groups would not develop severe disease, an assumption described as fragile.The cited literature found the policy ineffective.
- Policy assessment: Vertical isolation produces nearly the same epidemiological and economic results as Scenario 1, doing nothing.The results are reported as consistent with the cited literature.
5.6. Scenario 5: Partial Isolation
Scenario 5 tests intermediate partial-isolation levels and finds that 50% adherence does not provide effective epidemiological control, while higher isolation flattens infections but increases economic losses.
- Implementation: Scenario 5 randomly selected agents to stay home with probability IL = 0.5.The scenario assesses an intermediate isolation level rather than full lockdown.
- Epidemiological effects: At IL = 0.5, infections flatten relative to the baseline and vertical-isolation scenarios but remain less controlled than under lockdown scenarios.Deaths still grow exponentially before reaching their peak.
- Epidemiological effects: IL = 0.5 behaves economically similarly to the baseline, but the results recommend an isolation level greater than 0.5 for effective epidemiological control.The recommendation follows the observed infection and death trajectories.
- Isolation-level trade-off: As IL increases from 0.3 to 0.9, the infection curve flattens while wealth loss for A3 agents increases, highlighting mobility’s economic importance.The wealth-loss response is represented by the W_A3,S,t curve.
5.7. Scenario 6: Use of Face Masks
Scenario 6 models mandatory face masks and physical distancing without restricting mobility by reducing transmission parameters, producing flatter infections while preserving economic performance near baseline.
- Policy: Scenario 6 requires face masks and physical distancing but imposes no mobility restrictions on agents.The policy represents mask use as a non-mobility-based intervention.
- Implementation: The model represents masks and distancing by setting β1 = 0.5 and β2 = 0.3.These parameter reductions encode lower contagion distance and contagion rate.
5.8. Scenario 7: Use of Face Masks and 50% of Social Isolation
Scenario 7 combines face masks with 50% partial isolation, flattening infections and reducing economic damage relative to lockdown while remaining more acceptable to the general population.
- Policy design: Scenario 7 combines mandatory face masks with partial isolation at IL = 0.5.Its transmission parameters are β1 = 0.5 and β2 = 0.3.
- Economic and social effects: Despite an economic downturn, Scenario 7 suffers less economic damage than lockdown scenarios and faces less resistance from the general population.The scenario is therefore presented as a more socially cooperative alternative to lockdown.
5.9. Comparisons Among the Scenarios
The scenario comparisons show that lockdown policies provide the strongest epidemiological control but impose the greatest business losses, while Scenario 7 offers a more balanced alternative when lockdown cannot be enforced.
- Comparison framework: The comparison evaluates infection curves, infection peaks, time to peak, maximum deaths, and economic outcomes across intervention scenarios.Figures 14–17 contrast epidemiological and economic response variables.
- Epidemiological comparison: Scenarios 2 and 3 achieve the best epidemiological values, followed by Scenario 7.The comparison includes infection and mortality outcomes.
- Economic comparison: For citizens, Scenarios 2 and 3 are not economically damaging under the assumption that businesses do not fire workers, whereas they are worst for businesses.Healthcare costs increase public expenditure in high-death scenarios such as Scenarios 1 and 4.
- Pareto comparison: Scenario 2 is the best policy from a life-preservation perspective and Pareto dominates the simulated alternatives for people and government.It is nevertheless financially the worst case for businesses.
- Practical choice: When lockdown cannot be enforced, Scenario 7 has the fewest deaths among the remaining business-Pareto-optimal solutions and is best for government and people on wealth and deaths.This conclusion applies after Scenarios 2 and 3 are excluded.
6. Conclusion
The paper presents COVID-ABS as an agent-based model for evaluating COVID-19 interventions across epidemiological, social, and economic effects. Its scenario results favor lockdown for preserving lives, while masks combined with partial isolation offer a more feasible alternative, and the open model supports policy planning and extension.
- Conclusion: Few prior studies considered COVID-19 simultaneously as an epidemiological, social, economic, and governmental policy problem.The paper argues that social and economic aspects cannot be neglected alongside epidemiological priorities.
- Conclusion: COVID-ABS simulates epidemiological and economic effects in a closed society and supports policy assessment through configurable scenarios and response variables.The software contains 29 input parameters and 10 output response variables, and can be extended to other scenarios.
- Conclusion: The authors conclude that COVID-ABS can support governments and health authorities in planning social interventions and assessing their potential effectiveness.The model is open, customizable, and intended for designing region-specific scenarios.
- Conclusion: Lockdown and conditional lockdown best preserved lives by producing slower epidemic growth, fewer infections, and fewer deaths.Seven intervention scenarios were evaluated, with these two scenarios receiving the strongest preservation-of-life results.
- Conclusion: When lockdown is infeasible, 50% social isolation combined with masks and physical distancing was the best evaluated alternative for preserving lives.The paper presents this combination as the practical alternative among the evaluated scenarios when lockdown cannot be implemented.