banner icon

Client case study: Detecting FX quote publication latency

Reducing foreign exchange trading losses by detecting latency hotspots

Concerned that latencies within their request for quote (RFQ) and request for streaming quote (RFSQ) processes may be causing delayed quote publication, our client wanted the ability to detect latency sources by examining each and every step taken by loss making trades, tracing them all the way back to the pricing data ticks used to generate the quote.

Operating a low latency environment, our client was consuming pricing updates from multiple foreign exchange (FX) liquidity pools and then using this information to formulate pricing quotes.  These quotes were then being propagated out to a customer base, which included banks and other institutions.

This was achieved using the RFQ model, whereby the bank’s customers would request a specific size/instrument to be quoted for a short duration, or through the RFSQ approach, in which situation they would be sent continuous streams of quotes on set sizes throughout the day.  The pricing quotes provided through both of these models were then tradable for a given duration, guaranteeing the quoted price, within a specific timeframe.

 

The Challenge: Revealing latency sources

Our client was concerned that latencies within their RFQ and RFSQ processes may be causing the late publication of quotes, which when traded against, were generating financial losses.  The client wanted to be able to quickly identify and eliminate latency sources.

Time delays could have been occurring within systems internal to the bank, causing the quotes leaving the bank to be stale before they even exited the institution, or alternatively, within the client gateway, causing delayed quote receipt.

Hop-by-Hop latency measurements

To determine if avoidable latencies were occurring, our client wanted to understand the timings between:

  • The receipt of pricing updates
  • The use of that data to generate a pricing quote
  • The submission of that quote out through their client gateway
  • The sending of the quote and the receipt of a request to trade back from the customer

This information was required for both trades performed using common currency pairs and also trades involving synthetic crosses.

 

The Approach: Combined network and application monitoring

To achieve these measurements and the detection of latency sources, a combination of network and application monitoring techniques were applied.  This solution delivered full visibility across every system and network, providing a highly granular view from pricing data input to quote output.

tracing-diagram

An example of the contributing trade elements Velocimetrics’ tracing capabilities can identify

FX trade reconstruction and latency measurement

As our client was unable to publish the price-quote relationship to Velocimetrics in real-time a heuristic solution was implemented to
determine which price inputs generated tradable quotes.  The solution involved matching the timestamps of input pricing data to emitted quotes.  Using this method, the client was able to reconstruct the complete path taken by any executed trade identifying:

  • The quote that the customer had traded off
  • The RFQ submission had caused the quote to be emitted
  • The pricing tick that drove the generation of that particular quote

In addition to recognising these elements, the client was also able to understand timings at each and every stage.

Supplement monitored data with additional information

Our client was also able to gain full timings and traceability where synthetic crosses had been applied to currency pairs, where no single ID survived throughout the flow.

Within synthetic crossing situations multiple pricing updates can influence a trading decision. However, as Velocimetrics is a flexible and extensible solution the implementation team were able to configure the solution to understand the client’s specific synthetic crossing rules.  In doing so, Velocimetrics was able to apply the relevant prices to the monitored data and determine which pricing ticks drove actual trading decisions.

Identify data relationships to comprehend auto-hedger responsiveness

Our client also chose to extend the Velocimetrics monitoring solution to their auto-hedging system. Using heuristic methods, Velocimetrics was then able to identify the relationships between trade inputs and outputs, and understand how responsive the auto-hedger was proving to be.  In doing so, our client was able to identify opportunities to eliminate time delays.

 

The Results: Swift and efficient identification of latency bottlenecks generating time delays

Instead of trawling through time-consuming log files to determine if loss-making trades may have been caused by network or system issues, this solution enabled our client to quickly search using the trade ID.  For any given trade a highly granular report detailing all of the hops taken and their associated timings, for all elements that contributed to the generation of a quote that led to the creation of a trade, could then be instantly revealed.

In doing so, our client was able to quickly detect latency impacting events across their quoting infrastructure, both within systems and across their networks, and focus engineering efforts on these areas to improve overall performance.

Interested in seeing what Velocimetrics' technology could offer your firm?

Schedule a demo and see how uncompromised visibility accompanied by performance-improving analytics could transform your business insight

Find out More

Client summary

  • Firm type: Investment bank
  • Asset class: Foreign exchange
  • Location implemented: London

Goals:

  • Detect and eliminate sources of latency throughout the quote production process
  • Achieve hop-by-hop latency measurements for common currency pairs and synthetic crosses

Approach:

  • Network and application monitoring
  • FX trade reconstruction
  • Accurate latency measurements for each individual hop and complete processes
  • Supplement monitored data with additional information
  • Identification of input and output data relationships within the auto-hedger system

Results:

  • The ability to quickly detect latency
  • Increased efficiencies
  • Decreased trading losses due to operational performance