Español – Home / Capital; an experiment
Simulating Pure Capitalism
(Under construction)
The paper Simulating Pure Capitalism: Experimental Evidence with Human Agents was presented with 25 co-authors at the 32nd International Input-Output Association Conference, https://www.iioa.org/conferences/32nd , in June 2026. This is a rushed and provisional text, and a longer version will be published.
Its presentation (in PPT) are available. We summarize its content here.
Abstract
Table of contents and acknowledgments.
Theory.
We analyze three theories: Leontief input-output matrices, maximum growth or the Von Neumann (VN) model, and Maximum Benefit in the Long-term Perspective (MBLP). Input-output matrices are a special case of VN, and VN is a special case of MBLP. We analyze the relationship between MBLP-VN and general equilibrium theory. We also explain our vision of essential capitalism. We will see that our simulations resemble MBLP, and their long-term behavior resembles VN, but that there are also differences.
Description of our simulation.
We simulated capitalism in its purest and most essential form: agents who exchange and produce to maximize profit.
Each agent begins the simulation with a set of goods. First, they exchange with other agents; then they choose a production process, in which they consume the goods they have available and obtain new goods as products; and the procedure is repeated. The objective of each agent is to achieve the highest production at the end of the simulation.
Three simulations were carried out, one in each semester of the 2023-2024 academic year and another in the second semester of the 2024-2025 academic year, with 210 UAB students as agents.
Results and comparison with the theory.
The behavior in the three simulations was similar, which suggests that capitalism behaves regularly and that laws describing its dynamics can be found.
Production: Total production followed patterns parallel to MBLP, and also to VN in the long run, but with some differences.
Prices: Delivery ratios were similar to MBLP values, and to VN values in the long run.
Inefficiency: There were imbalances and waste, and the production level was lower than MBLP; the long-run growth rate reached 20% compared to 25% for VN.
Instability: Decreasing oscillations of a cobweb-like nature developed, whereas MBLP and VN do not exhibit cycles.
Inequality: There was a high concentration of wealth, with a tendency towards monopoly.
Networks: Specialization in production and the formation of networks of regular customers were common.
Regulation: The operation of the agents acted as a real algorithm, achieving regulation of the economy, despite the uncertainty and complexity of the decisions.
A fundamental question arises: how does the market manage to resemble MBLP even in the first session and VN in the long term?
Appendix A1. Methodology.
A theory can only be considered scientific if it is tested against empirical data and evidence. While the natural sciences use controlled experiments to verify their laws, in economics this process is much more difficult due to the complexity of economic systems and the lack of precise data.
We propose using simulations of artificial economies as a tool parallel to laboratory experiments, since they allow us to study the behavior of capitalism under controlled conditions and compare the results with theoretical predictions.
We focus on the theory of Maximum Benefit in the Long-term Perspective (MBLP) and the John von Neumann model (VN), trying to verify whether the simulations reproduce the dynamics predicted by these theories. Validation is based on verifying the similarity between different simulations, their agreement with theoretical predictions, and the prior relationship of these theories with phenomena observed in real economies. Finally, we argue that the artificiality of simulations does not invalidate their scientific utility, just as it does with experiments in other disciplines, and that they can be a valuable tool for studying certain aspects of capitalism.
Appendix A2: Optimal allocations.
An optimal allocation is one that maximizes an objective. We use tools such as linear programming and Lagrange multipliers, which can be interpreted as prices or values of resources. We also analyze the concept of duality, according to which the problem of allocating resources and the problem of valuing those resources are equivalent and closely related. We show that, when there are external exchanges, these values correspond to exchange rates or market prices.
Appendix A3: MBLP and GET.
The relationship between Long-Run Profit Maximization (MBLP) and General Equilibrium Theory (GET) is analyzed. MBLP can be interpreted as a special case of the General Economic Theory (GET), when the turnpike theorems apply or when competition favors those who maximize long-term outcomes. But also as an alternative theory to GET, because the Long-Term Production Theory (MBLP), like Von Neumann model (VN), is production for production's sake and not a consumption-oriented economy.
Furthermore, a distinction is made between two ideal forms of capitalism: PM1, in which agents pursue diverse objectives, and PM2, in which profit is the dominant objective of economic activity. The simulations performed correspond to this second case, as they aim to represent capitalism in its purest and most essential form.
Finally, the possibility is raised that competition favors the evolution from PM1 systems to PM2 systems, as well as the interest in studying this process experimentally in future research.
Appendix A4: Comparison with other experiments and simulations.
We compare our simulation with other classic economic experiments, such as those of Chamberlin, Smith, and Barceló. We emphasize that its main innovation is simulating a complete capitalist system with repeated production and exchanges and quantitatively comparing the results with economic theories such as MBLP and Von Neumann. Unlike traditional experiments, which focus on individual decisions or specific markets, our simulation studies the behavior of the economic system as a whole and in the long term. Furthermore, we compare simulations with human and computational agents, arguing that the former allow us to observe spontaneous behaviors and social relationships that are difficult to reproduce using computer models.
Appendix A5: Some final considerations.
Our capitalisms work: they manage to provide a solution to the extremely difficult allocation problem by acting as a real algorithm, something that some proposed alternative systems have failed to achieve.
But with fundamental problems: this solution exhibits inefficiency, instability, inequality, particularism, and dehumanization, among other difficulties.
Bibliography.