Estimating hidden liquidity from granular order-book data

[tradingexpert]

Visible queue depth is only part of executable liquidity in many markets.

Icebergs, reserve replenishment, dark pools and participant reaction functions can all create situations where executed size materially exceeds displayed liquidity.

Interesting questions for replay/simulation systems:

• What intellectually sound and reliable models can be used to estimate these?
• What observable signatures are useful?
• Can replenishment behavior be modeled probabilistically?
• What are the venue-specific assumptions that can be perused?
• Should hidden liquidity estimation be deterministic, heuristic, or adaptive?
• How to deal with overfitting on current data coverage? Is that even a problem?