This is the finish line for Agent-Based Models & Market Simulation. You started by asking why you would build a market bottom-up from interacting agents instead of learning a price generator top-down, and you saw that an ABM makes macro behavior emerge as an output rather than baking it in as an assumption. From there you met the zero-intelligence baseline — proof that the trading institution, not the agents’ brains, does most of the work — then stocked the agent zoo with market makers, momentum chasers, value traders, and noise traders, and watched their interaction manufacture volatility clustering and fat tails. You learned that calibrating these models means fighting an intractable likelihood with simulation-based methods, that calibration is never validation, and that the real payoff is counterfactuals: questions about markets that never existed. This exam is graded and irreversible. There are no hints, each answer locks the moment you submit it, and your score stays hidden until the very end. Read every option carefully before you commit.
Course Recap
Big picture
The Agent-Based Market Simulation Arc
- ABM and Market Simulation
- Why Build From Agents
- Bottom-up not top-down
- Macro behavior emerges
- Interaction is the engine
- Slow and many parameters
- Zero-Intelligence Agents
- Random quotes still efficient
- Structure over smarts
- ZI-C respects budget
- Baseline to beat
- The Agent Zoo
- Market makers stabilize
- Momentum destabilizes
- Value traders mean-revert
- Santa Fe and ABIDES
- Emergent Stylized Facts
- Clustering from feedback
- Fat tails from herding
- Non-interacting is Gaussian
- Produce not reproduce
- Calibrating an ABM
- Likelihood is intractable
- Simulated method of moments
- ABC and emulators
- Identifiability is hard
- Validating Honestly
- Calibration is not validation
- Out of sample facts
- Sensitivity analysis
- Overfitting tax
- Counterfactual Payoff
- Re-run with new rules
- Flash crash stress tests
- Reactive training ground
- Handles market impact
- Why Build From Agents
One run, one shot
This is the graded, irreversible final exam. There are 34 questions, shown one at a time. Submitting an answer locks that question for good — there is no Back button, no retry, and no Restart. A wrong answer simply fails that question and the exam moves on. Your running score stays hidden until the final screen, and the pass mark is 70%. Some questions accept more than one correct option and say so in the stem, so read every option before you submit.
What is the defining feature of an agent-based model (ABM) of a market, as opposed to a top-down generative model?
Select an answer to continue.
Key Takeaways
You can now build a market, not just describe one
Finishing this exam means you can reason about markets from the agents up: you know why interaction — not intelligence — is the engine, why the zero-intelligence baseline is the bar to beat, how a well-chosen population of market makers, trend followers, and value traders manufactures clustering and fat tails, and why producing those facts is necessary but never sufficient. You can calibrate against an intractable likelihood with simulated moments and modern simulation-based inference, you can tell calibration from validation and resist a too-good in-sample fit, and you understand the real prize: counterfactuals, flash-crash stress tests, and a reactive training ground that handles market impact where a backtest cannot. On the quant ladder, that puts you at the frontier where simulation, mechanism, and machine learning meet — equipped to use ABMs and generative models as the complementary tools they are.