Economic Survival of the Fittest: Forecasting with Network Data

Author: Denis Avetisyan

A new agent-based model demonstrates surprisingly accurate macroeconomic forecasts using only input-output tables and principles of Darwinian selection.

The study leverages FIGARO data and network simulation to model shock propagation and generate competitive forecasts without relying on parameter estimation or complex behavioral rules.

Conventional macroeconomic forecasting relies heavily on time series analysis and parameter estimation, often obscuring the underlying structural relationships within an economy. This paper, ‘Macroeconomic Forecasting from Input-Output Tables Alone: A Darwinian Agent-Based Approach with FIGARO Data’, demonstrates that a surprisingly accurate forecasting model can be built solely from inter-industry relationships captured in input-output tables, using a Darwinian agent-based simulation. Specifically, the approach generates competitive GDP forecasts and simulates shock propagation-including a near-accurate prediction of the initial COVID-19 impact-without estimating any parameters or incorporating explicit behavioral rules. Could this structurally-focused approach, portable across 46 countries via the FIGARO dataset, offer a more robust and transparent alternative to traditional forecasting methods?

Mapping the Economy: Unveiling Interconnectedness

Traditional analyses of economic resilience frequently rely on aggregated data, obscuring the intricate web of dependencies that define modern production networks. These methods often treat industries as isolated entities, failing to capture how a disruption in one sector ripples through the entire economy. A lack of granular detail hinders accurate assessments of vulnerability and limits the ability to predict the cascading effects of shocks – such as supply chain bottlenecks or geopolitical events. Consequently, policymakers and businesses struggle to proactively mitigate risks and build truly robust economic systems, necessitating more detailed mapping of inter-industry relationships to understand the full scope of economic interconnectedness.

The core of the DEPLOYERS simulator lies in the FIGARO IOTable, a comprehensive dataset meticulously detailing the interconnectedness of 64 distinct economic sectors. This isn’t a generalized economic model relying on broad averages; instead, it’s a granular, industry-by-industry representation of production relationships. The table maps how each sector’s output serves as input for others, creating a detailed network of dependencies. Consequently, the simulator can trace the ripple effects of disruptions – be they supply chain bottlenecks, geopolitical events, or technological shifts – with a precision unavailable to less detailed models. This interlinkage mapping allows DEPLOYERS to move beyond simply predicting overall economic trends and instead model how specific shocks propagate through the complex web of production, offering a uniquely insightful approach to understanding economic resilience.

The foundation of accurate economic modeling rests on a detailed understanding of inter-industry relationships, and the DEPLOYERS simulator leverages the FIGARO IOTable – a comprehensive dataset sourced from Eurostat – to achieve precisely that. This 64-sector input-output table meticulously maps the flow of goods and services between industries, enabling the simulator to trace how an initial economic shock – such as a supply chain disruption or a change in consumer demand – ripples through the entire system. Critically, this granular approach allows for robust forecasting; the model demonstrates a remarkably low mean absolute error of just 0.42 percentage points when predicting 5-year normal-year GDP, highlighting the power of structurally-informed economic simulation and providing a valuable tool for policymakers and analysts alike.

From Static Structure to Dynamic Evolution

The DEPLOYERS system introduces firm dynamics into the static FIGARO IOTable by implementing a Darwinian selection process. This involves modeling the continuous entry and exit of firms within each industry sector defined by the IOTable’s structure. Firm birth rates are determined by profitability thresholds, while firm death occurs when profitability falls below a specified level, effectively simulating competitive pressures. This process ensures that the industrial composition of the simulated economy evolves over time, reflecting the ongoing reallocation of resources based on economic performance and adherence to the fixed-coefficient production functions inherent in the Leontief Technology underpinning the model.

The simulation utilizes Leontief Technology, which posits that production functions exhibit fixed proportions of inputs. This means that for a given output level, a firm requires a specific, unalterable combination of inputs – such as labor and capital – in the short run. Mathematically, this is often represented as $X = A \cdot V$ , where X is the output vector, A is the matrix of fixed input coefficients, and V is the vector of input quantities. Consequently, firms cannot readily substitute between inputs in response to price changes or shocks; adjustments to output are primarily achieved through changes in the scale of operations, constrained by the availability of inputs and the fixed coefficients defining the production process. This rigid structure is a core assumption influencing how the simulation models firm behavior and economic dynamics.

The simulation environment models economic dynamics by integrating firm-level decision-making – encompassing production, investment, and pricing – with aggregate market forces such as supply, demand, and competition. This interaction generates emergent macroeconomic behavior, allowing for the observation of economy-wide responses to externally applied shocks. These shocks, representing unforeseen events or policy changes, are propagated through the simulated economy via adjustments in firm behavior and subsequent market equilibria. The fidelity of this propagation is ensured by the underlying Leontief technology and Darwinian selection mechanisms, which govern production constraints and firm entry/exit, respectively, creating a robust system for analyzing economic resilience and adaptation.

Predicting Resilience: A Test of Economic Forecasting

DEPLOYERS utilizes a computational general equilibrium (CGE) framework to model macroeconomic responses to exogenous shocks. This simulation capability allows for the analysis of diverse disruptions, including pandemics, supply chain failures, or shifts in global trade. The model incorporates interconnected sectors – households, firms, government, and international trade – and simulates how these sectors adjust to shocks through price and quantity changes. By altering key parameters representing the nature and severity of the shock, DEPLOYERS can generate counterfactual scenarios, enabling the assessment of potential economic consequences and the evaluation of mitigation strategies before they are implemented.

DEPLOYERS incorporates the ability to evaluate the macroeconomic effects of specific policy interventions designed to lessen the severity of economic recessions. This functionality allows for the quantitative assessment of policies such as the Kurzarbeit (short-time work) policy, which aims to prevent layoffs during demand shocks by subsidizing reduced working hours. The model simulates how these policies alter labor market dynamics, household income, and aggregate demand, ultimately providing estimates of their impact on key economic indicators like GDP, employment rates, and government debt. By varying policy parameters within the simulation, DEPLOYERS can be used to compare the effectiveness of different intervention strategies and inform policy decisions regarding optimal responses to economic downturns.

During a simulation of the initial year of the COVID-19 pandemic, the DEPLOYERS model predicted a Gross Domestic Product (GDP) decline of -4.62%. This projection demonstrated a high degree of correlation with the empirically observed GDP decline of -6.6%. The accuracy of this prediction validates the model’s capacity for forecasting macroeconomic impacts in complex, multi-country scenarios and highlights its efficacy in simulating responses to global economic shocks. The relatively small divergence between modeled and observed values suggests the model effectively captures key economic relationships relevant to pandemic-induced recessions.

Revealing Economic Anatomy: The Landscape of Firm Size

The distribution of firm sizes within an economy isn’t random; rather, it follows a discernible pattern, and DEPLOYERS provides a unique lens through which to examine this landscape. This computational tool allows researchers to map the prevalence of micro, small, and medium-sized enterprises, revealing a pronounced skew towards smaller businesses. Specifically, simulations demonstrate that approximately 87.6% of firms fall into the micro category, while small firms constitute 10.8% and medium-sized firms only 1.6%. This detailed mapping isn’t merely descriptive; it highlights the critical role that micro and small enterprises play in economic structure, offering valuable insights for policies designed to foster competition, innovation, and overall economic resilience. Understanding these underlying patterns is crucial for crafting effective interventions and promoting a dynamic business environment.

A key validation of the DEPLOYERS model lies in its accurate replication of the European firm size distribution. The simulation demonstrates a landscape overwhelmingly dominated by micro-enterprises, constituting 87.6% of all firms. Small businesses account for a further 10.8%, while medium-sized firms represent a comparatively small 1.6% of the total. This closely mirrors data collected by the European Commission, confirming the model’s ability to realistically portray the structure of the European economy and providing a solid foundation for policy analysis and research into firm dynamics.

DEPLOYERS distinguishes itself by integrating granular structural data – encompassing firm characteristics, market dynamics, and policy parameters – with a sophisticated dynamic modeling framework. This synergy allows for nuanced simulations of economic phenomena, moving beyond simplified assumptions to capture the complex interplay of factors influencing firm behavior and market outcomes. Consequently, the tool provides policymakers with a robust platform for ex ante policy evaluation, enabling them to assess the potential impacts of interventions on various firm sizes and market segments before implementation. Researchers benefit from DEPLOYERS’ capacity to generate realistic scenarios, test theoretical predictions against simulated data, and explore the effects of counterfactual policies, ultimately advancing understanding of economic anatomy and informing evidence-based decision-making.

Towards Higher Fidelity: Expanding the DEPLOYERS Horizon

The DEPLOYERS framework is poised for a significant enhancement through the incorporation of FIGARO SUTable, a detailed Supply-Use table. This integration moves beyond aggregated sectoral data, allowing the model to dissect economic activity at a much finer, product-level resolution. By mapping the flow of goods and services between industries with greater precision, researchers anticipate a more nuanced understanding of international trade dependencies and the ripple effects of economic shocks. This increased granularity will not only refine the accuracy of projections but also enable the identification of specific bottlenecks or opportunities within global supply chains, ultimately providing policymakers with more targeted and effective intervention strategies.

The DEPLOYERS framework is poised to become significantly more efficient through the incorporation of an AI Coding Assistant. This tool is designed to automate key stages of the modeling process, drastically reducing the time and expertise required for calibration – the refinement of model parameters to accurately reflect real-world data. Beyond calibration, the assistant will accelerate data analysis, identifying trends and anomalies with increased speed and precision. Crucially, it will also empower users to conduct more robust scenario planning, allowing for rapid prototyping and evaluation of various economic interventions and forecasts – ultimately enabling quicker, more informed decision-making regarding global trade and economic policy.

Analysis of the DEPLOYERS framework consistently reveals a positive bias of 3.7 percentage points when modeling Germany’s economic performance over a nine-year period. This persistent overestimation isn’t a flaw, but rather a key insight into the nation’s economic structure; it underscores Germany’s significant reliance on exports as a driver of growth. The model, therefore, doesn’t simply predict economic activity, but actively illuminates underlying characteristics – in this case, a structural dependency on international trade. This ability to expose such ingrained economic traits demonstrates the framework’s potential as a diagnostic tool, going beyond forecasting to offer a nuanced understanding of a nation’s economic profile and vulnerabilities.

The study reveals a fascinating resilience within economic systems, mirroring a natural process of adaptation. It demonstrates how a foundational structure – the input-output table – can, through simulated competition, yield surprisingly accurate forecasts. This echoes Søren Kierkegaard’s observation, “Life can only be understood backwards; but it must be lived forwards.” The model doesn’t predict through complex calculations, but rather reveals emergent behavior from a simple initial state, much like understanding the past informs our present trajectory. The agent-based approach, stripped of intricate parameter estimation, suggests that sometimes observing the system’s evolution – allowing it to age gracefully – is more insightful than attempting to accelerate or overly engineer its outcomes. The propagation of shocks, simulated through this Darwinian framework, showcases an inherent capacity for systems to learn and adjust, even without explicit programming for such contingencies.

What Remains to be Seen

The demonstrated capacity to generate macroeconomic dynamics from a static input-output table is, at first glance, a curiously elegant reduction. It suggests that much of the elaborate machinery of contemporary forecasting – the econometric calibrations, the behavioral assumptions – may be treating symptoms rather than the underlying structure. However, the simplicity is not without cost. The model, operating as it does on the fulcrum of Darwinian selection, inherently lags adaptation. The static table represents a prior state, and the speed with which the simulated economy converges – or fails to – reveals the limits of its responsiveness to genuinely novel shocks.

Future work must address the question of decay. How quickly does the predictive power of this approach erode as the represented economic structure diverges from reality? The absence of parameter estimation, while a virtue in its parsimony, also implies a lack of explicit mechanisms for incorporating learning or adaptation beyond the competitive process itself. Extending the model to incorporate evolving input-output relationships – a dynamic table, if one will – presents a significant, though not insurmountable, challenge.

Ultimately, this approach offers not a final solution, but a different lens. It highlights the enduring network effects within an economy, and the potential for emergent behavior even from remarkably simple rules. Every abstraction carries the weight of the past, and the true test of this framework will lie in its ability to gracefully age, rather than succumb to the inevitable distortions of time.

Original article: https://arxiv.org/pdf/2603.12412.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/