The COVID-19 pandemic prompted a global response centered around suppressing SARS-CoV-2 transmission through mandated NPIs, including lockdowns of non-essential businesses and remote work policies. These measures disproportionately impacted socioeconomic groups, with low-income workers in essential, in-person jobs facing higher infection risks and job losses. Concurrently, individuals voluntarily altered their behavior, reducing contact and customer-facing service usage due to fear of infection. However, the comparative effectiveness of these self-imposed changes versus mandated NPIs, and their differential impacts across socioeconomic groups, remained unclear. This study addresses this gap by developing a detailed, mechanistic model to simulate the intertwined epidemic and economic dynamics at a granular level, capturing the heterogeneous effects across various socioeconomic strata.

Existing models often offer aggregate perspectives on either the epidemic or economic dimensions, failing to capture the diverse impacts across socioeconomic groups. While some agent-based models (ABMs) simulate epidemic spread and economic decisions at the individual agent level, they often lack the data-driven detail and fine-grained calibration necessary for robust quantitative assessment of policy impacts. This paper builds upon previous epidemic and economic COVID-19 models, incorporating detailed socioeconomic data to create a more comprehensive and granular simulation.

The researchers developed a data-driven, granular ABM of the New York-Newark-Jersey City metropolitan area, using a synthetic population of 416,442 individuals representative of the real population's socioeconomic characteristics (household composition, age, income, occupation, work-from-home capability). The model incorporates detailed socioeconomic attributes derived from census, survey, and mobility data. The epidemic module uses a multi-layered contact network (household, school, workplace, community) informed by anonymized mobility data from Cuebiq, providing daily workplace visitation and community colocation probabilities. A stochastic, discrete-time disease transmission model simulates the spread of SARS-CoV-2. The economic module emphasizes employment and consumption, with hiring/firing decisions driven by industry needs, closures, and remote work possibilities. Consumption dynamically adjusts based on age, income, and fear of infection, particularly affecting customer-facing industries. The input-output network of intermediate goods captures the propagation of economic shocks. The model couples the epidemic and economic modules through reduced consumption due to fear of infection, which is a function of reported daily deaths. The model is calibrated to reproduce key epidemic and economic statistics from the first COVID-19 wave in the New York metro area. Counterfactual scenarios explore variations in fear of infection levels, economic activity closures, and the timing of interventions.

The model accurately replicates key economic statistics (employment, GDP, consumption patterns) from the first COVID-19 wave in New York. It also validates against untargeted empirical properties, such as industry-specific employment changes and the disproportionate impact on low-income individuals (higher unemployment but less reduced consumption compared to high-income individuals). Epidemiologically, the model accurately reproduces death counts, changes in contact patterns post-intervention, and reduction in workplace contacts. Counterfactual analyses reveal that stricter closures and higher fear of infection both lead to increased unemployment but fewer COVID-19 deaths. Low-income workers experience a more substantial impact, with larger increases in unemployment and decreases in deaths compared to high-income workers. Geographical disparities in unemployment are also observed, with low-income areas experiencing higher unemployment rates in the empirical scenario compared to high-income areas. Comparing the impact of closing all non-essential industries versus only customer-facing industries, the study finds that the former leads to a much larger increase in unemployment for a marginal improvement in health outcomes. Delayed implementation of protective measures in high-fear scenarios results in worse health and economic outcomes. Even with age-specific fear of infection, where older individuals exhibit stronger behavioural changes, the overall improvement in health and economic outcomes is only modest compared to uniform fear.

The study highlights a qualitative parallel between mandated NPIs and heightened fear-driven behavioral changes. Both lead to increased unemployment but fewer deaths, particularly impacting low-income workers. The findings challenge the notion that solely relying on voluntary behavioral adaptations would yield optimal health and economic outcomes, particularly given the time lag between infection, death reporting and consequent behavioral adjustment. The effectiveness of each intervention depends on their quantitative effects and timing, and the type of economic activity impacted. The model demonstrates the importance of considering the interplay between supply and demand shocks, emphasizing the complex effects across different socioeconomic groups.

This study demonstrates a similar trade-off between public health and economic outcomes for mandated interventions and spontaneous behavioral changes during the COVID-19 pandemic's first wave. Low-income workers bore the brunt of this trade-off. Future research should explore more nuanced policy interventions, such as targeted support for specific occupations and enhanced surveillance in high-risk industries, to mitigate inequalities and optimize pandemic response strategies. Further research should incorporate more detailed modelling of risk perception and the impact of various information channels on individual behavior during a pandemic.

The model focuses solely on the first wave of COVID-19 in the New York metropolitan area, and does not consider later aspects of the pandemic (masks, testing, tracing, variants, vaccination). The matching of synthetic individuals to mobility traces is probabilistic, and the model assumes a uniform per-contact infection risk across occupational settings. It also simplifies the representation of risk perception and the impact of information channels. Finally, while the model considers industries across the metro area, it doesn’t model individual firms at specific locations. These limitations may affect the generalizability of the findings to other settings or later phases of the pandemic.