Simulation Glossary

The terms this book uses for the simulation. Exchange terms in general — order book, spread, auction, tick — are in the Reference Manual's glossary and in How an Exchange Works.

Accelerated time — Running the trading calendar faster than the wall clock (EM_SPEED). Only the calendar is faster; the traders act at their usual pace. See Accelerated Time.

ai-sim — The process profile that runs the simulation: the full exchange, the market maker, the market model and the AI swarm.

AI trader — A program that trades as a participant (AI001…), following a preset. Many of them run inside one pm-ai-swarm.

Anchor (--anchor-sim) — How strongly the market maker moves its quotes toward the true value each second: the share of the remaining gap it closes.

Bias — A value trader's own, persistent error about a share's value. Different biases are why value traders disagree and trade with each other.

Budget — The order actions per second (new orders and cancels) the whole swarm aims for.

Composition — The mix of presets in the swarm: how many traders of each kind.

Credibility — How believable a rumour looks, 0 to 1. Informative, not reliable.

Efficiency — How closely prices follow the true values; measured as the gap.

EM_SPEED, EM_START — The container's settings for accelerated time: how many times faster, and from when.

Gap — ln(price / value): how far a price is from its true value, about the same as the percentage difference. Summarised over all shares as its RMS (root mean square).

Headline — A piece of news: scope, kind, sentiment, and the secret impact.

Impact — The log move a confirmed headline causes in the true value; secret. 0.05 is about +5 %.

Informed trader — A trader that knows something about the true values or the news: the value-investor, institutional and news-trader presets.

Jump — A sudden move in a value without any headline: an unannounced event.

Market model — pm-market-sim: keeps the true values, moves them, publishes news.

Noise trader — A trader that trades in random directions, without information: noise-retail, market-taker, iceberg-seller.

Preset — A kind of AI trader: a strategy, an execution style, a tempo and risk limits, in one YAML file.

Price discovery — The process by which trading moves a price toward the value.

Retention — How long a record is kept before it is deleted to bound the disk: 7 days for individual trades in the statistics and in clearing, about an hour for order events, 30 days for logs, the last 60 MB of the audit trail.

Rumour — A headline that may not be true. It moves no value until it is confirmed; if it is retracted, it never does.

Scope — What a headline is about: one symbol, a sector, or the whole market.

Sector — An industry group of shares that move together: eleven in s300-sim.

Sentiment — How good a headline sounds, −1 to +1; public, and a blurred version of the impact.

s300-sim — The bundled example the simulation runs: 300 shares, AI001–AI100, MM01, TRADER01–TRADER10, OPS01.

Swarm — pm-ai-swarm: the process that runs many AI traders, in one or more worker processes.

True value — A share's hidden value in the market model; what price discovery aims at.

Volatility — How much a value (or a price) moves: the standard deviation of its log changes, quoted per year (30 % a year is about 1.9 % a day).

Watchdog — The container's once-a-minute check that restarts any process of the profile that has died (EM_RESTART_DEAD=1).