Topic · Research Practice & Backtesting
← All topicsSimulation Environments
32 articles · 6 checkpoints · 20 deeper reads · 6 reference notes
A standalone topic: it is on no roadmap, so read it on its own terms.
Every article, in reading order
plant a flag as you finish eachRead these first
A backtest that just checks whether the last price crossed your limit isn't simulating a market, it's simulating a coin flip. A matching engine simulator replays the actual rules an exchange uses to decide who trades with whom.
A simulator built from first principles and never checked against reality is a guess with good production values. Calibration is comparing what the simulator predicted to what actually filled, and adjusting the model until the gap closes.
An agent-based simulator can produce price paths that look completely realistic and still be worthless for testing a strategy, because looking realistic and behaving realistically under stress are two different things to calibrate for.
A full order-book simulator tells the truth slowly; a simplified fill model lies quickly. Research runs on the fast one because it has to, and that choice quietly decides which strategies get found.
An order isn't a single event, it's a sequence of states connected by rules about which transitions are even allowed. Skip the state machine and you get double-counted fills and phantom positions that only show up during a race condition.
Two resting orders at the identical price can have completely different odds of filling, depending on nothing but who arrived first. A simulator that ignores queue position gives every passive strategy a fill rate it will never see live.
Then the rest