MIT OpenCourseWareSession 1: Stocks: Horse Market, Oil Market, and Transportation System
MIT OpenCourseWare
Course setup 0:00
Hazhir Rahmandad opens the 15.879 System Dynamics Research Seminar and explains that this version is meant to bridge the gap between a standard system dynamics course and the kind of serious modeling needed in consulting and research. The class is recorded for MIT OpenCourseWare, uses email and Dropbox for material sharing, and runs for three hours with a break. Grading, where it applies, depends on weekly modeling work, assignments, and active participation.
Modeling uses 5:31
The class is built around weekly modeling tasks. Students submit work, a few models are chosen at random, and the group critiques them together before turning to next week’s problem and its main design choices. He then separates major uses of system dynamics: process consultation and group model building, theory building for general insight, theory testing with data from one or more cases, prediction for longer-term futures, and policy design. Each use asks for a different level of detail, data, and calibration.
Model uses and scale 18:32
Detailed system dynamics models work well in dispute cases, legal work, and big projects. They also support repeated decisions in supply chain and production management. These models are usually grounded in data and take more time to build, maintain, and update. Smaller exploratory models are used to understand dynamics, test simple policies, and build intuition for hard organizational problems.
Process and problem framing 24:02
The class focuses on the smaller end of this range and on iterative modeling, where you add one piece at a time, understand it, and then expand. The main steps are problem definition, qualitative modeling, simulation, calibration, and analysis, but the course spends less time on early framing. Clear problem definition matters because it sets the model boundary, the detail level, and which policy levers belong in the work.
Defining the problem 37:03
A model begins by naming the mechanism you care about. It may involve two actors competing, word of mouth, or a Bass diffusion pattern in which more adopters appear as the product gets better, cheaper, and more viable. Once those pieces are clear, you have your marching orders: build a generic model that captures them and test what they imply for policy or theory. Canonical building blocks such as the NK model, the multi-armed bandit, and the prisoner’s dilemma often serve this role, and people make new contributions by adding one more piece to a known core. Another route is to start from the policy question itself, ask what model can answer it with confidence, and then decide what must be inside the boundary, how long the horizon is, and how much aggregation or detail you need.
Analysis process 55:30
Start from equilibrium if you can, then add a shock and watch what the model does. Pick one mode of behavior, form a hypothesis by tracing causes backward, and test it by removing the suspected cause and seeing whether the behavior changes. If the test fails, go back and trace again. After that, change all parameters by 10 percent and check whether your predictions about the outcomes are right.
Aggregation and stocks 1:01:37
Bad aggregation choices are hard to reverse, so you need to think about them early and stay consistent across the model. Clear definitions matter, and overlapping ideas like morale, job satisfaction, and motivation can blur the logic. For stocks, start from sources of inertia such as what you can observe, how people adjust their perceptions, and physical accumulations. Use MECE so stocks do not overlap and together cover what you need.
Use known formulations 1:13:01
Use existing stock and flow formulations instead of inventing new ones. Jim Hines’s taxonomy, available in Vensim and at sdmolecules.org, gives a hierarchy from simple stocks to patterns like proportional split, closing the gap, and Hines experience coflow. Learn them by writing the equations, naming inputs and outputs, checking parameters, and testing pulse, step, ramp, and oscillatory responses.
Build with discipline 1:18:30
Keep each chunk focused on its job. Limit equations to three inputs or fewer, prefer simple analytic forms over table functions, avoid if-then and other discrete breaks when you can, and keep units consistent. Do not chase a desired behavior by blind parameter search or by piling on structure. Understand the model first, then adjust parameters with purpose.
Audience shapes the model 1:33:03
A model for a sociology journal and a model for a high net worth client need different levels of detail. The journal version should stay simple, with a clear qualitative mechanism and high-level aggregation, so the idea can be understood without a modeling background. The investment version needs tighter calibration to data, more external inputs, and more detail on things outside the model, such as the stock market, interest rates, breeding horses, and the number of active breeders. The time horizon also changes. The theory model should run long enough to show feedback effects and possible collapse. The investment model needs a one-year projection, with some longer look ahead on whether to stay in the market or leave it. Those choices shape what you model first and how deep you go.
Horse stocks and delays 1:53:00
Keep the horse model spare. A separate breeding stock is useful, but a fraction of prime-age horses can stand in for it if the breeding delay is only a few months against a twenty-year horizon. The same logic applies to price trends. Use standard stock-and-flow pieces for expected price, but do not add extra detail just to show it.
Status demand and transport 1:59:01
Treat status as one slow stock, but leave its link to demand a little loose for now. Money is not a binding limit in this market, so leave it out. Word of mouth and Bass diffusion are possible, but not a priority. Then shift to congestion. Travel time, road building, traffic volume, public transit revenue, and driving attractiveness form the main loops in the urban transport diagram.
Choosing useful stocks 2:13:03
The discussion stays on how to choose stocks that match the time horizon. Cultural preference may change too slowly to count as a stock in a one- to five-year plan, so it can be treated as a parameter instead. Highway capacity and public transit capacity are clear physical stocks. Urban and suburban residents also matter, because long-run residence choice responds to highway capacity, even if that feedback is weak over only a few years.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
