Repository navigation
Sal's Thesis
An Agent-Based Simulation Model of a Post-Capitalist Labor-Time Economy: Decentralized Planning and Benefits Distribution
We present an agent-based simulation model of a post-capitalist economy based on labor-time accounting, in which goods are priced according to the socially necessary labor time needed — directly and indirectly — to produce them, and workers are compensated with non-exchangeable labor-time credits, with no surplus diverted to profit, no property or commodity rents, and no exploitation. The simulation was designed and implemented to explore two interconnected questions: how such an economy can be organized in a realistic, decentralized manner and how it can equitably distribute the benefits of technological productivity improvements.
The simulation models a cybernetically-designed supply chain, in which consumers, distributors, and producers interact through feedback signals. Firms manage inventory using demand signals, negotiate orders through accept-or-reject feedback loops, and facilitate the reallocation of workers to where they are most needed.
Simulating a flexible, decentralized economy of variable scale, we observe that the supply chain achieves stable production without central planning. Within this framework, we plan to investigate how feedback loops affect labor-time pricing under localized productivity improvements. In a separate simulation of 8 sectors over 10,000 cycles, we tested immediate versus delayed price recomputation triggered by firm-level efficiency thresholds. Delayed recomputation, triggered only when collective firm-level efficiency thresholds are met, preserves stable purchasing power across all sectors, validating a more equitable distribution mechanism.
Agent-based simulation (ABS) is a computational modeling paradigm in which a system is represented as a collection of autonomous, interacting entities known as agents. Rather than describing system-level behavior through aggregate equations, ABS models the local rules governing each individual agent and allows macro-level phenomena to emerge organically from their interactions. Each agent possesses a set of attributes — such as state variables, resource levels, or behavioral parameters — and a set of decision rules that govern how it perceives its environment and responds to stimuli from neighboring agents or the broader system. Agents are typically heterogeneous, meaning they may differ in their attributes and rules, and they act autonomously, meaning no central authority dictates their behavior. The simulation advances in discrete time steps, during which every agent evaluates its current state, applies its rules, and updates accordingly. A key strength of an ABS is its ability to capture emergent behavior: system-wide patterns that arise from local interactions and cannot be predicted from the properties of any single agent in isolation. Classic examples include urban traffic flow, the epidemiological spread, and the collapse of supply chains from localized inventory decisions. Agent-based simulation contrasts with two dominant alternatives in computational social science. Equation-based models like system dynamics or differential equation models, describe aggregate stocks and flows but obscure individual heterogeneity and discrete interactions. Game-theoretic models assume rational, optimizing agents with known constraints on information, which limits their realism in settings where agents have bounded rationality and incomplete knowledge. ABS relaxes both assumptions, allowing agents to follow simple heuristics, learn from experience, and operate under unconstrained uncertainty. In the context of this thesis, ABS provides the natural framework for modeling a decentralized labor-time economy. The economy is composed of heterogeneous agents: workers (who are also consumers), producers, and distributors, who coordinate through local feedback signals rather than a central planner. The macro-level properties of interest, including price stability, purchasing power equity, and the distribution of productivity gains, are emergent outcomes of these micro-level interactions. ABS therefore allows us to study whether a viable, equitable labor-time economy can arise from decentralized organization, without neglecting the coordination challenges that such a system must overcome in practice.
The theoretical foundation of this thesis draws on the labor theory of value, which holds that the value of a commodity is determined by the socially necessary labor time required to produce it. This idea, developed extensively in classical political economy (Smith, 1776; Marx, 1867), forms the basis of labor-time economics, a system in which prices reflect production costs in terms of labor rather than market-determined exchange values. The Group of International Communists (GIC 1930, 1936) provides one of the most concrete descriptions of this framework, outlining a system of production and distribution in which workers receive credits corresponding to their labor contributions and spend them on goods and services priced at their labor cost. This model eliminates profit and rent as categories entirely, with all productive output returning to workers as consumption. Unlike market socialist or centrally planned alternatives, a labor-time economy resolves the problem of ensuring that supply meets demand, not through price signals or top-down directives, but through a direct and transparent link between contribution and compensation.
[ Say something here about plans:
- In the GIC scheme, which we largely follow, firms draft and execute plans.
- These plans are created by the firms themselves, individually, rather than by a central authority.
- Firms are assumed to be run democratically by their workers.
- A plan consists of a product name, a quantity to be produced or distributed, and a budget, denominated in hours of labor.
- The budget contains line items for fixed capital (plant, machinery, and tools), input goods (such as raw materials), and direct living labor.
- These input costs are estimates based upon society-wide averages.
- Thus the average socially necessary direct and indirect labor time to produce a good or service is its price.
- Prices are periodically updated based on observed average costs in labor time.
- The budget also includes a line that the GIC calls prd, which is the additive inverse of the sum of the three input lines, and which represents the debt to society implied by the plan — the firm's promise to perform the work that the plan envisions.
- In this sense, labor-time credits are a kind of fiat credit currency: the credits are produced whenever needed, but always accompanied by an equal amount of debt.
- When the firm's output is consumed, the consuming party transfers labor-time credits to the firm, which then adds that amount to prd, bringing it closer to 0. The prd line thus records the degree to which the firm's output has indeed been consumed, showing that the work done was indeed needed to be done.
- If all goes well, when a firm finishes working on a plan, the three input lines are reduced to 0. When the output from the plan is taken up by those who will use it, prd is brought back up to 0. ]
The organizational design of the simulation is grounded in cybernetics — the study of regulatory systems and feedback-driven control. Stafford Beer (1995) developed the Viable System Model (VSM), which describes how complex organizations can maintain stability and adaptability through nested hierarchies of feedback loops. In a VSM-inspired design, no central authority needs complete information; instead, each subsystem monitors its own state and communicates deviations to adjacent levels. This principle directly informs the supply chain architecture of our simulation, where producers, distributors, and workers regulate themselves through demand signals, busyness measures, and order accept/reject cycles, rather than receiving instructions from a central planner. As it happens, our choice of ABS as our modeling method fits this theoretical approach strikingl well. Law (2014) shows how ABS is especially well suitded to modeling such feedback-driven systems, enabling the observation of emergent behavior.
Agent-based simulation has become an increasingly prominent tool for modeling economic systems, particularly those involving heterogeneous actors, decentralized coordination, and emergent phenomena that cannot be captured by representative-agent [PLEASE CLARIFY] or equilibrium [PLEASE CLARIFY] models. Tesfatsion and Judd (2006) establish the foundations of agent-based computational economics, arguing that bottom-up simulation [Missing step: ABS goes from the bottom up — and as opposed to what?] is uniquely suited to studying how macro-level economic patterns arise from micro-level interactions. This approach has been applied to model financial markets, supply chains, labor markets, and institutional change. Notably, agent-based models do not require assumptions of perfect rationality or market clearing [What is "market clearing"?], making them well-suited to the kind of cybernetically organized, non-market economy modeled in this thesis [It's problematic to call a labor-time economy a "non-market" economy. It has a market, it just does not have competitive prices. The transactions are organized locally and not co-ordinated on a global level, just as they are in a market. Part of the idea is to save the efficiency in the transfer of information that we get in a market while avoiding the exploitation we get in a capitalist, money-based market]. The decentralized pull-based supply chain we simulate — in which firms order only what they need in response to downstream demand — mirrors the kanban and lean manufacturing systems studied extensively in the operations research literature. [Citation needed. But, whereas I can see the relevance of lean (just-in-time) manufacturing, you need an argument here for the relevance of the kanban. The "mirroring" is far from obvious.]
A central concern of this thesis is how productivity gains at the firm level should be distributed across an economy. Toivanen and Väänänen (2023) find that innovation and labor-market outcomes are deeply intertwined, with technological improvement generating both gains in output and disruptions in employment and wage structures. Martinsson (2022) provides micro-level evidence from Swedish manufacturing linking patent quality and productivity improvements to firm-level outcomes, highlighting that the benefits of innovation are unevenly distributed in capitalist settings. The problem of worker incentives for innovation is well-documented: Kesting and Kappe (2020) show that employee-driven innovation is strongly correlated with workplace conditions that give workers genuine agency over production decisions. IdeaWake (2023) [I don't understand this citation. IdeaWake is not the author, is it?] documents Toyota's suggestion program as a real-world example of workers voluntarily contributing productivity improvements when they are not penalized for doing so — a finding that motivates the core design question of this thesis: how to structure a labor-time economy so that firm-level productivity gains benefit all workers rather than concentrating advantages in the most efficient sectors.
[I think that you need to say a little more in the above subsection. The comments start with an understanding that supporters of capitalism argue that you need the profit motive to drive innovation. This raises two questions: (1) Is it true that profit drives innovation? (2) Is it possible that innovations will arise without profits? Under the first question, you have three points: (1) profits go to owners, not workers, and innovations come mostly from the workers who know the production details best (for which I think you can also cite the Toyota experience); (2) even assuming that the profits go to the workers, and not the owners, profits are distributed unevenly, providing limited overall social benefit; and (3) innovations that increase productivity can cause the very workers who create or adopt them to lose their jobs, which provides a disincentive against productivity-increasing innovations. Under this third point, you have the empirical observations you mention — to the effect that, if there are no disincentives, and if workers feel appreciated for their innovations, they will innovate. Then come back to the ABS model: we investigated how to remove disincentives from productivity-increasing innovations within a labor-time economy.]
The methodology of this thesis comprises two agent-based simulations implemented in C++, both modeling aspects of a labor-time economy. The first, referred to as the Supply-Demand Signals Simulation (SDSS), models a cybernetically designed supply chain of variable scale, investigating how stable production and pricing can emerge from local feedback interactions between firms and consumers without central planning. The second, referred to as the Productivity-Incentives Simulation (PIS), [fixes the economy at 8 sectors over 10,000 cycles and — this part does not belong here. You need to find a place for it further down, where you get into fine-grained details, if at all] investigates how firm-level productivity gains should be redistributed to preserve purchasing power equity across all workers. Together, the SDSS and PIS address the two core questions of this thesis: how a labor-time economy can be organized, and how it can equitably distribute the benefits of technological improvement.
Both simulations share a common architectural foundation. The economy is modeled as a collection of heterogeneous agents organized into a cybernetic hierarchy, communicating through positive and negative feedback signals rather than receiving instructions from a central authority. Prices are denominated in labor-time units, and workers are compensated in non-exchangeable [non-circulating, not non-exchangeable: they are exchanged for goods and services — or rather, the economy, through the firm, exchanges the worker's labor time for labor-time credits, and then the worker exchanges those credits for the results of other people's labor time.] labor-time credits redeemable solely for consumption. Both simulations advance in discrete time steps, during which each agent evaluates its current state, processes incoming signals, and updates its behavior accordingly.
People represent the worker-consumer agents at the base of the supply chain. Each person [Lower case for the entities; upper case just for their programmatic representations, i.e., classes.] is characterized by a set of attributes, including labor ability [ --> particular skills], health [status], and a consumption basket representing their [--> that person's] needs across sectors. At each time step, a person shops from distributors according to their needs, exchanging labor-time credits for goods. Their labor ability [--> skills] and health [--> health status] jointly determine the rate at which they can execute production or distribution plans, and firms collect data on these attributes to predict how long a given plan will take to complete. People also participate in the societal busyness measure — a running average of how long workers are actively engaged versus on standby [a running average proportion of time that workers are actively engaged in work rather than standing by] — which feeds back into the length of the working day [--> work day].
Distributors sit between producers and people in the supply chain hierarchy. They receive demand signals from people as goods are depleted from their inventory, and use these signals to place orders with producers. [You need to say somewhere much earlier that we are currently only modeling the production, distribution, and consumption of goods, and plan to add services to the model in future work.] A Distributor's order to a Producer takes the form of a quantity and deadline pair calibrated to exactly match the rate at which its inventory is being depleted, implementing a pull-based replenishment system that prevents overproduction and waste. [Like all firms,] distributors also [<-- delete] participate in [--> follow] plans, employing workers to execute distribution tasks, and report labor-time expenditure data to the Price Controller.
Producers convert inputs, machinery, and labor into outputs, forming the productive core of the supply chain. Like distributors, they receive orders from downstream agents and place orders with upstream producers for their own inputs. Upon receiving an order, a producer computes a predicted turnaround time based on its current labor capacity and inventory levels, and sends either an accept signal — along with the predicted completion time — or a reject signal to the ordering firm. If a producer cannot find any accepting upstream supplier at its requested order size, it scales down its order iteratively in a feedback loop until a supplier can be found. Producers also report the precise labor time expended per unit of output to the Price Controller, enabling continuous price updating.
[Insert a section here on the beginning of simulated time. This is where you would introduce the Leontief matrix, among other things. Be sure to explain what it is and why you need it (i.e., because any output product may require any number of any input product, and you need to know how to set initial prices in order for the system to get going.]
The Price Controller is the sole centralized component of the simulation, but operates in a limited and reactive capacity rather than as a top-down planner. It continuously receives discrete reports from firms on the labor time expended to produce each good, and uses these reports to update the socially necessary labor time — and therefore the price — of each product. In theory, the Price Controller could keep all prices in the society up to date at all times by recomputing the full Leontief input-output equation every time a firm completes a plan and reports its actual expenditures. Instead, the Price Controller updates only the relevant Rather than recomputing the full Leontief input-output system at every time step, the Price Controller updates only the relevant entries of the input-output matrix, trading off efficiency for small delays in the propagation of updates through the economy. This trade-off makes the Price Controller both computationally efficient and responsive to real-time changes in production.]** column of the solution when a new report arrives, making it computationally efficient and responsive to real-time changes in production. In the PIS, the Price Controller also implements the recomputation threshold mechanism, triggering economy-wide price and wage adjustments only when at least 4 [some determinate number of] firms in a sector achieve at least 4% [some determinate quantity in] efficiency gains.
Demand signals are the primary coordination mechanism in the SDSS. When a downstream agent — either a person or a distributor — consumes goods from a Firm's inventory, it generates a demand signal proportional to the rate of depletion. Firms use this signal to calibrate their outgoing orders to producers [Is this true? I don't think so. The demand signals are used only to determine the inventory re-order thrsholds. They are one of two essential signals. The other one is simply the orders or purchases.], matching order size and deadline to the observed depletion rate. [That part is true, but that's on the downstream entity, not the upstream producer. The distributor or downstream producer uses demand signals to determine (indirectly) the timing and (directly) the quantity of the order to replenish an inventory stockpile. The upstream firm uses the state of its inventory and the state of its workers to determine its proposed turnaround time.] This creates an efficient pull force that propagates upstream through the supply chain, ensuring that each level of production responds to actual downstream consumption rather than forecasted demand. The result is a kanban-style system that naturally tends toward equilibrium [^state approaching the mean] between overproduction and underproduction. [(An equilibrium is not between A and B; it is a state where the forces driving it in different directions all equal each other, so that the state does not change (much).] [So ... rewrite this section to make sure that what you say is true. Be mindful of what demand signals do. Their name can be a bit misleading. The most important pull signal is simply the purchase (by a consumer) or order (placed by a firm) itself.]
When a firm initiates a production plan but cannot source sufficient labor from the unemployed population alone, it broadcasts a labor request signal specifying the quantity of workers needed. Other firms evaluate their own busyness — a measure of the portion of overall time during which their workers are actively engaged in work — and respond with a list of transferable workers, if those workers sufficiently underutilized. If the requesting firm approves the proposed transfer, the workers are reassigned for the duration of the plan. This mechanism distributes labor across the economy in response to local shortages without requiring a central labor allocation authority. [I think you'd better say something about further work to ensure the individual worker's autonomy in deciding to move from one firm to another. Your description makes it sound as if workers have no individual choice, which replicates capitalist-style alienation.]
The length of the working day is itself a feedback-driven variable in the SDSS. The simulation tracks societal busyness — the population-wide average of worker activity relative to idle time — and adjusts the working day length accordingly. If societal busyness is too low, indicating widespread underutilization, the working day is shortened to concentrate demand for labor. If busyness is too high, risking labor shortages where every worker is already occupied, the working day is extended. An inefficiency-of-work parameter is used to tune the system toward a target busyness level that maintains a sufficiently large standby population to absorb demand spikes.
The PIS isolates the question of how firm-level productivity gains should propagate through the price and wage system. Eight sectors are simulated over 10,000 discrete cycles, representing approximately 27 years of economic activity. [^Productivity-increasing] Innovation within each sector follows a Poisson distribution with a base probability of 15% and a variance of 1%, generating productivity improvements of between 1% and 20% per innovation event, with parameters calibrated to [calibrated to --> drawn from (or based on or derived from)] Toyota production data to reflect realistic rates of incremental improvement.
Two redistribution mechanisms are tested. Under Solution A, prices and wages [Please make this more precise. What changes is the work day within the sector (only): if, in your sector, you get your work done for an order in half the time, you just work half a day, but you're also just paid for half a day.] are adjusted immediately whenever any firm achieves a productivity improvement, so that the labor-time price of a good and the credits earned by workers in that sector fall in real time. [Explain the possible consequences of this: since you earn credits for work done, you wind up with less purchasing power if you work more efficiently, which is a disincentive against productivity increases.] Under Solution B, prices and wages are held constant until a collective threshold is met: at least 4 firms within a sector must achieve efficiency gains of at least 4% before an economy-wide price recomputation is triggered. [No, this cannot be just within one sector. The idea is that you have to have improvements across several different sectors for everybody to be able to agree to a shorter work day over all. Also, explain the implication: before the overall work day has been reduced, because enough different sectors have increased their productivity, the workers in the extra-productive sectors get to go home early — but are paid for the whole day. This protects them from the disincentive and even incentivizes innovation by giving them an immediate boost in free time without any loss in purchasing power.] The key metric evaluated across both conditions [be explicit — the two states, before and after the accumulation of a critical mass of innovations in different sectors] is purchasing power stability — whether workers in all sectors maintain proportional access to a standard basket of essential goods as productivity improvements accu
mulate over the simulation horizon.
The PIS yields clear and interpretable results across both redistribution mechanisms tested. Simulating 8 sectors over 10,000 cycles representing approximately 27 years of economic activity, the simulation reveals critical differences between immediate and delayed price and wage adjustments in response to firm-level productivity gains.
Under Solution A, in which prices and wages are adjusted immediately whenever any firm achieves a productivity improvement, significant purchasing power disparities emerge across sectors. Workers in high-productivity sectors see their working hours drop from 8.0 to under 5.0, reflecting the reduced labor time required to produce their sector's goods. However, their budget allocation becomes increasingly skewed: goods in less productive sectors — most notably healthcare — do not decrease in price at the same rate, becoming disproportionately expensive relative to the reduced earnings of high-productivity workers. While the cost of a standard basket of essential goods falls to 4.777 labor-time hours under this mechanism, the distribution of that cost is highly uneven, creating a two-tier system in which workers in the most productive sectors are paradoxically least able to afford the same living standards as their less productive counterparts. This outcome is particularly perverse from the standpoint of a labor-time economy's core promise: rather than productivity gains benefiting all workers equally, they concentrate disadvantages in the sectors that innovate fastest. [NO. You have just shown the opposite, and the opposite is true. The most productive workers, who innovate the fastest, are the ones who now have to struggle to make ends meet, because they earn less, whereas those who innovate the most slowly keep the same longer work hours but benefit because the goods that they do not produce have become cheaper. Please make sure that this is clear for you and for your readers.]
Under Solution B, in which economy-wide price and wage recomputation is triggered only when at least 4 firms in a sector achieve efficiency gains of at least 4% [See above. This cannot be right. The threshold for recomputing prices and the standard work day would have to be when enough different sectors enjoy productivity increases, not firms within a single sector. Please let me know if you still have trouble with this point.], workers maintain stable purchasing power as hours decline gradually from 8.0 to 7.0 over the full 10,000 cycle horizon. [The important point is not the time horizon, it's that everybody has a shorter work day across the whole society.] Budget allocation remains balanced across sectors throughout the simulation, with no sector becoming disproportionately expensive relative to others. The cost of a standard basket of essentials remains manageable at 6.953 labor-time hours by cycle 14, compared to an initial cost of 8.000 hours, maintaining proportional access to goods across all sectors regardless of which sectors innovate first or fastest. [It's regardless of that, but, as I say, enough different sectors have to have increased their respective productivity levels for everybody's work day as a whole to change at the same time that all prices drop. This gives you a virtuous deflationary tendency.]
The simulation results demonstrate that the mechanism by which productivity gains are redistributed has fundamental consequences for purchasing power equity in a labor-time economy. Immediate price and wage adjustment, while intuitive in its simplicity, produces a paradoxical outcome: workers in the most productive sectors are penalized for their sector's innovation, while workers in stagnant sectors maintain their purchasing power by virtue of their sector's lack of improvement. [They actually increase their purchasing power, because now at least one necessary good has a lower price — one of the goods that they do not produce.] Delayed, threshold-triggered recomputation resolves this paradox by ensuring that no single sector bears the cost of its own productivity gains in isolation. Instead, price adjustments occur only once enough collective improvement has accumulated across the economy, distributing the benefits of innovation broadly rather than concentrating its costs narrowly. This validates Solution B as the more equitable mechanism and suggests that, in any practical implementation of a labor-time economy, the timing and coordination of price recomputation must be treated as a core policy design decision rather than a secondary technical consideration.
[You need a results section about the SDSS!]
[Move these citations up to the appropriate place — that's done already, actually — and, instead, have a Works Cited list at the end.]
[^classical-econ]: Adam Smith, The Wealth of Nations (Edinburgh, 1776). Karl Marx, Capital, vol. I (...).
Edits to Make:
- Change Firms to Producers