This paper presents the first detailed empirical study of quasi-centralized limit order books (QCLOBs), a market structure where institutions can only trade with counterparties with whom they have sufficient bilateral credit. Using high-quality data from Hotspot FX for three currency pairs (EUR/USD, GBP/USD, EUR/GBP) over 30 trading days, the authors reconstruct the global limit order book and trade-price series. They argue that the traditional quote-relative coordinate frame is unsuitable for QCLOBs because institutions cannot see global quotes; instead, they propose a trade-relative frame based on the most recent trade prices, which are visible to all participants.

The empirical analysis reveals several key features: market orders are rare (less than 0.05% of order flow), most limit orders are cancelled, and institutions exhibit selective liquidity-taking, often matching only a fraction of the available queue depth. The authors also observe that global bid–ask spreads can become negative, though briefly. A striking finding is that the distributions of order flow and LOB state vary considerably across days but can be collapsed onto a single curve via a simple linear rescaling. Motivated by this universality, they propose a semi-parametric model that provides similar performance to parametric curve-fitting but is simpler and faster to implement. This work highlights important differences between QCLOBs and centralized LOBs, and offers new tools for analyzing such markets.