System dynamics is a method for understanding how complex systems behave over time by modeling the feedback loops, delays, and accumulations that drive change. Rather than looking at isolated events or snapshots, it maps the circular cause-and-effect relationships within a system to reveal why problems persist, why growth stalls, or why well-intentioned policies backfire. The approach originated at MIT in the late 1950s and has since been applied to everything from supply chain management to global sustainability modeling.
Where System Dynamics Came From
System dynamics grew out of a practical puzzle. Jay Forrester, an engineer at MIT, found himself in conversation with managers at General Electric who couldn’t explain why their appliance plants in Kentucky would sometimes run three or four shifts, only to lay off half their workforce a few years later. The boom-and-bust cycle seemed to have a life of its own, disconnected from actual consumer demand.
Forrester suspected the problem wasn’t bad luck or poor management. It was the structure of the system itself: the way orders, inventory, production decisions, and shipping delays interacted to amplify small changes into large swings. He built a simple model of that inventory system with pencil and paper, simulating how decisions rippled through the chain. That first model, completed in the late 1950s, was the beginning of system dynamics. By 1958, Forrester had published “Industrial Dynamics” in the Harvard Business Review, laying out the core idea that the structure of a system, not just individual decisions within it, determines its behavior.
Feedback Loops: The Central Idea
The building block of every system dynamics model is the feedback loop: a closed chain of cause and effect where the output of one variable eventually circles back to influence that same variable. There are two types, and grasping the difference between them is the single most important step in understanding the field.
A reinforcing loop amplifies change. Any situation where an action produces a result that promotes more of the same action is a reinforcing loop. A classic example: word of mouth for a popular product. More customers lead to more recommendations, which lead to more customers. Reinforcing loops drive exponential growth or exponential decline, depending on the direction.
A balancing loop resists change and pushes a system toward a target or equilibrium. A thermostat is the textbook example: when the room gets too cold, the heater kicks on, which raises the temperature, which eventually tells the heater to shut off. Balancing loops are the reason many systems stabilize, but they’re also the reason some problems stubbornly resist intervention.
Most real systems contain both types of loops interacting simultaneously, which is what makes their behavior so counterintuitive. A city’s population might grow through a reinforcing loop (more people attract more jobs, which attract more people) while simultaneously triggering balancing loops through pollution, disease, and strain on sanitation infrastructure. A system dynamics model of New York City’s garbage system, for instance, traces how modernization increases sanitation capacity, which reduces bacteria, which reduces disease, which allows the population to grow further, which generates more garbage. The loops work at cross-purposes, and the system’s behavior over time depends on which loops dominate at any given moment.
Stocks, Flows, and Delays
Feedback loops explain the circular logic of a system, but to simulate how it actually behaves over time, you need two more concepts: stocks and flows.
A stock is anything that accumulates. Think of water in a bathtub, money in a bank account, carbon dioxide in the atmosphere, or patients in a hospital. It’s a quantity you could measure at a single point in time. A flow is the rate at which a stock changes: water pouring in through the faucet (an inflow) or draining out (an outflow). The stock at any moment is the result of all past inflows minus all past outflows.
This distinction matters because stocks create delays. Even if you shut off the faucet completely, the tub still holds water. Even if a country reduces emissions to zero tomorrow, the carbon already in the atmosphere persists for decades. These delays are one of the main reasons people misjudge how systems will respond to intervention. We tend to think in terms of flows (“we cut emissions by 20%”) when the outcome depends on the stock (“atmospheric concentration is still rising”).
Two Types of Diagrams
System dynamics practitioners use two visual tools, each suited to a different stage of thinking.
A causal loop diagram maps the feedback structure of a system at a high level. It shows which variables influence which, whether each relationship is positive or negative, and where the reinforcing and balancing loops are. These diagrams are excellent for communication: they give a holistic overview of a system’s complexity and make it easy to spot feedback loops at a glance. But they don’t distinguish between stocks and flows, which means they can’t reveal the delays that often dominate a system’s behavior.
A stock and flow diagram adds that layer. It explicitly marks which variables are accumulations (stocks) and which are rates of change (flows), making it possible to write the mathematical equations that drive a simulation. Stock and flow diagrams point out where lags exist in the system, which is critical for understanding why policies that seem logical can take years to show results, or why a problem can keep getting worse even after the root cause has been addressed.
Common Patterns That Repeat Across Systems
One of the most useful discoveries in system dynamics is that very different systems often share the same underlying feedback structures. These recurring structures are called archetypes, and learning to recognize them helps you diagnose problems faster.
- Limits to growth: An effort initially generates strong performance, but over time it hits a constraint that slows everything down, no matter how much additional energy is applied. A startup that grows rapidly until it can’t hire fast enough to maintain quality is a common example.
- Fixes that fail: A quick fix addresses the symptoms of an urgent problem but sets unintended consequences in motion that eventually make the original problem worse. Prescribing antibiotics for every minor infection reduces symptoms today but breeds resistant bacteria over time.
- Shifting the burden: A short-term solution relieves pressure in a way that undermines a more fundamental solution. The classic case is relying on overtime instead of hiring, which solves the immediate workload problem but burns out existing staff, making the underlying capacity gap worse.
- Drifting goals: When a gap appears between a goal and actual performance, the response is to lower the goal rather than improve performance. Over repeated cycles, this leads to a gradual decline in standards.
- Success to the successful: Two or more efforts compete for the same finite resources. Whichever one gets slightly ahead attracts more resources, which pulls it further ahead, starving the other. Think of how a university department that publishes well gets more funding, enabling more publications, while a struggling department falls further behind.
- Growth and underinvestment: Growth approaches a limit that could be avoided by investing in additional capacity, but the decision is made not to invest. Performance drops, demand declines, and the drop in demand appears to justify the decision not to invest in the first place.
The Modeling Process
Building a system dynamics model follows a structured sequence, developed at MIT, that moves from qualitative understanding to quantitative simulation. It begins with problem definition: identifying the key variables, sketching reference modes (graphs of how those variables have behaved over time), and writing a clear problem statement. The modeler then identifies the “momentum policies,” meaning the default decisions and rules currently operating in the system.
From there, the work becomes iterative. You develop a dynamic hypothesis, a causal loop diagram that proposes which feedback structures are responsible for the behavior you’ve observed. Then you build the first feedback loop as a formal stock-and-flow model, simulate it, and analyze whether it reproduces the patterns you see in reality. You add the next loop, simulate again, analyze again, and continue. Each iteration tests whether your structural theory of the problem actually generates the problematic behavior. If it doesn’t, you revise your hypothesis.
The Limits to Growth Model
The most famous system dynamics model is World3, built at MIT in the early 1970s for the Club of Rome. It simulated five interrelated global sectors: population, capital, agriculture, nonrenewable resources, and pollution. The resulting book, “The Limits to Growth,” presented 12 scenarios covering the period from 1900 to 2100. Its authors were careful to emphasize that the scenarios were not predictions.
The best-known scenario, called the “standard run” or business as usual, showed exponential growth leading to overshoot and collapse, triggered primarily by the depletion of nonrenewable resources. A second scenario assumed aggressive technological solutions: greatly reduced pollution, increased crop yields, and resource efficiency above all historical levels. Even in that optimistic case, collapse was postponed rather than avoided, with steep declines appearing near the end of the simulation period. Only one scenario, called “stabilized world,” avoided overshoot entirely. It modeled a future in which population, industrial production, and resource consumption reached a steady state.
A 2023 recalibration of the World3 model updated its parameters to better match real-world data on global development over the past half-century. The recalibrated model still showed the same overshoot-and-collapse pattern as the original business-as-usual run, with the main difference being that the peaks of most variables shifted slightly higher and a few years further into the future. The parameters that changed most were those related to industrial capital lifetime, pollution transmission delay, and the time required for urban-industrial land development.
Where System Dynamics Is Used Today
Public health has been one of the most active application areas since the 1970s. System dynamics models have been used to study the epidemiology of heart disease, diabetes, HIV/AIDS, cervical cancer, dengue fever, and drug-resistant infections. In substance abuse, models have tracked heroin addiction dynamics, cocaine prevalence, and the effects of tobacco reduction policies. Healthcare delivery models have simulated patient flows through emergency departments and extended care facilities, capacity planning for health maintenance organizations, dental and mental health systems, and the strain on health infrastructure during natural disasters or terrorist attacks.
The approach is particularly valuable for problems where the delivery system and the disease interact. A policy that reduces risk factors for diabetes, for example, might also overwhelm primary care capacity if it drives a surge in screenings. System dynamics models can simulate that kind of interaction and help policymakers sequence their interventions to avoid creating new bottlenecks.
Beyond health, system dynamics is used in supply chain management, urban planning, energy policy, climate modeling, and corporate strategy. The common thread is any problem where delays, feedback, and accumulations make intuitive reasoning unreliable. Recent work has begun integrating system dynamics with machine learning, combining the feedback-driven structure of traditional models with the predictive power of AI. Applications in waste management, water systems, and energy planning have shown promise, though adoption of this hybrid approach is still in its early stages.