Week 10: Agent-Based Economic Modeling

Guest lecture — thousands of simple trading agents, none of them smart, still manage to reproduce the bubbles and crashes a real market produces.

3 May 2027

Guest lecture by Dr. Lior Tanaka-Bramwell.

You don’t need a smart agent to get a realistic-looking market. You need enough of them, each running a simple rule — buy under a threshold, sell above it, adjust the threshold slowly based on recent trades — and the interaction between thousands of dumb agents produces price bubbles, herd selling, and crashes that look uncannily like the real thing, without any single agent intending any of it.

Three ingredients are enough, and none of them require anything you’d call intelligence: a strategy (the buy/sell rule above), a way to observe price (what “recent trades” the threshold adjusts against), and a budget — each agent starts with a fixed stash of currency and stock, so its rule can only ever act on what it actually holds, not a hypothetical unlimited wallet. That budget constraint is what turns a simple rule into something that can run out of room to buy or sell, which is exactly the kind of pressure that produces a crash instead of a smooth line.

Why this matters for a game economy specifically#

A live game economy already has thousands of real, if unpredictable, human agents running exactly this kind of simple heuristic — “buy low, sell high”, “panic if the price drops fast”, “follow whatever the streamer just recommended”. An agent-based model built from those same simple rules is cheap to run and can be stress-tested in ways a live economy never can: what happens if a new item floods the market overnight, or a sink gets removed without warning. It’s how a designer gets an early answer to “will this break the game” before it ships.

What the model can’t tell you#

An agent-based model is only as honest as its rules. If the model’s agents don’t include the deliberately weird behaviour real players show — speculative hoarding ahead of an announced patch, or a coordinated guild deciding to crash a market out of spite — the model will confidently predict stability right up until a real, ungoverned crowd of players proves it wrong.

No studio this week#

There’s no studio session this week — this is a single guest lecture, kept deliberately self-contained so it doesn’t create dependencies on a session Lior isn’t teaching.