A Market That Runs Itself¶

Learning objectives

After reading this chapter you will be able to:

Say in a few sentences what the simulation is and what it is for
Name its parts and what each one does
Explain the one idea the whole simulation rests on: a price that is trying to find a value nobody can see

An exchange with nobody at the desks¶

An exchange on its own is an empty room. Order books fill only when someone sends orders, and a class of students can keep a few symbols busy for an hour at most. The simulation fills the room: AI traders send orders into every one of 300 order books, all day, every trading day, whether anyone is watching or not. Students who log on find a market already moving — prices, spreads, queues, news — and trade in it next to the bots.

The traders are not scripted. Each one follows a simple rule of its own — buy what looks cheap, follow a trend, fade a move, react to a headline, or just trade at random — and the prices you see are what those rules add up to. Nobody decides that AAPL should trade at 182.40. It does because a value investor thought it was cheap at 182.30 and a noise trader happened to sell.

The idea underneath: a value nobody can see¶

Real shares have a value — what the company is actually worth — that nobody knows exactly. Prices wander around it: traders who have a good idea of the value buy when the price is below it and sell when it is above, and so pull the price toward it, while everyone else pushes it about. When news arrives the value jumps, and the price follows as fast as traders understand the news. That is price discovery.

The simulation makes this concrete. A process called pm-market-sim keeps a true value for every share: a number that moves the way company values move, together with the market, with its sector, on its own, and in sudden jumps when news comes out. The true values are secret. Students never see them; only some of the AI traders, the market maker and the instructor do. The price on the exchange is whatever the traders agree on, and the gap between price and value is what makes the market interesting:

a well-informed market keeps the gap small, a noisy one lets it grow;
a headline moves the value at once and the price a little later;
a rumour moves the price but not the value — until it is confirmed or retracted.

The True Value explains the model; News and Rumours the news.

The parts¶

flowchart LR
    SIM["pm-market-sim\ntrue values, news"]
    MM["pm-mm-bot  MM01\nquotes every symbol"]
    SW["pm-ai-swarm\nAI001–AI100"]
    ALF["pm-alf-gwy"]
    ENG["pm-engine\n300 order books"]
    SCH["pm-scheduler\nopen, auctions, close"]
    GW["API, market-data and\nlog services"]
    GUI["browser GUIs and\ncommand-line tools"]
    YOU["students and\ninstructor"]

    SIM -- "values (secret)" --> MM
    SIM -- "values (secret), news" --> SW
    SIM -- "news (public)" --> GW
    MM --> ENG
    SW --> ALF --> ENG
    SCH --> ENG
    ENG --> GW --> GUI
    YOU --> GUI
    YOU -- "orders" --> GW

Part	Process	What it does
The exchange	`pm-engine` and its gateways	Matches orders in 300 order books; publishes trades and prices
The clock	`pm-scheduler`	Opens and closes the market: pre-open at 09:00, opening auction, continuous trading 09:30–16:00, closing auction, closed at 16:05; weekends and Swedish holidays off
The market model	`pm-market-sim`	Keeps the hidden true values, moves them during trading, publishes news and rumours
The market maker	`pm-mm-bot` as `MM01`	Quotes a buy and a sell price in every symbol, around a price it moves part of the way toward the true value every second
The AI traders	`pm-ai-swarm` as `AI001`–`AI100`	A hundred traders of eleven kinds, each trading 30 of the 300 symbols
The windows	the GUIs and `pm-*` tools	Show the books, the trades, the news, the logs; let people trade

Everything else — the audit trail, clearing, statistics, the API — is the ordinary exchange, running exactly as it does in a classroom without any bots. The AI traders are participants like any other: they log on, send orders and receive fills through the same gateway a student's trading program uses. Their orders are in the books you look at, their trades in the tape.

The eleven kinds of trader¶

Each AI trader follows a preset: a strategy (what to trade), an execution style (how), a tempo (how often, how big) and risk limits. The hundred traders are a mix:

Preset	How many	What it does
`noise-retail`	25	Small random orders resting near the best prices — the background flow
`value-investor`	22	Buys below its view of the true value, sells above it
`trend-follower`	10	Buys what is rising, sells what is falling
`contrarian`	10	The opposite: fades moves, patiently
`institutional`	8	Big orders toward the value, sliced over minutes
`scalper`	8	Fast momentum trades with tight trailing stops
`news-trader`	5	Trades headlines within seconds
`iceberg-seller`	4	Large orders showing only a fifth of their size
`market-taker`	4	Impatient: takes liquidity with market orders
`block-taker`	2	Takes whole blocks with fill-or-kill orders
`auction-player`	2	Trades mostly in the opening and closing auctions

Only the value, institutional and news traders know anything about the true value or the news. Everyone else trades on prices alone — and that is what keeps the market from being perfectly efficient. How the AI Traders Decide has the details.

What it is good for¶

A live market for a class. Students log on as TRADER01–TRADER10 and trade in books that already have depth, a spread and a tape.
Teaching price discovery. Inject a headline in front of the class and watch the price find the new value; inject a rumour and watch it move a price that should not move.
Teaching microstructure. Auctions, queues, icebergs, stops, market orders sweeping a book — all of them happen every day, unprompted.
A long-running test bed. Leave it for a month and use the history: daily statistics, the audit trail, the traders' daily profit and loss.

What it is not¶

The simulation is a teaching market, not a forecast of a real one. Its values follow a textbook model, its traders a handful of simple rules, and its hundred participants are far fewer than a real exchange's. The prices look and behave like prices — they trend, revert, jump on news, and have the fat-tailed returns of real prices — but no real share is behind them.