agent-based-sim

Define agent classes for informed traders, noise traders, and market makers, then let randomly selected agents trade each step against a limit order book. The provided NumPy implementation runs thousands of steps and outputs price history, volume, and per-group P&L.

Kyle lambda is the price impact parameter from Kyle (1985), computed as sigma_v divided by two times sigma_u. It determines how much each trade moves the price, and it decreases as noise trader volume provides camouflage for informed trades.

Yes. Gode and Sunder (1993) showed zero-intelligence agents achieve near-full allocative efficiency, and Farmer, Patelli and Zovko (2005) explained 96% of spread variation on the London Stock Exchange with a single parameter.

Noise traders trade randomly without information, so they systematically transact at prices away from the true probability. Their losses fund the positive P&L of informed traders who trade toward the true value.

Use agent-based models when dynamics emerge from heterogeneous agent interactions, such as spread formation, information-driven convergence, or regime switches, which no closed-form stochastic differential equation can capture.