DEVON WAS APPOINTED TO ANALYZE THE FX BUSINESS OF A GLOBAL BANK

Devon was appointed by a US regulator to analyze (the “Bank”) who had been a middle of the pack FX participant, until in 2005 it developed a corporate strategy to become the world’s leading FX liquidity provider using an electronic platform. Deutsche Bank had the highest FX market share, at times over 50%.

EXAMPLES OF SPREAD DECAY FOR THE BANK CUSTOMERS

  1. The Bank’s electronic FX trading platform contained an automated control called “Last Look”. This gave the Bank an opportunity to decline customer trades that it felt were systematically unfair to the bank as a Liquidity Provider, e.g. trading on stale prices due to latency or trades made on information that was not available generally to the market.
  2. The Bank’s e-FX business placed each client into one of several categories to apply various trade acceptance rules for its automated Last Look process.

GUI USERS*

  • Large spread
  • No decay
  • High spread retention
  • Able to show the best prices

HIGH FREQUENCY FUNDS

  • Mixture of models
  • Very fast turn around models
  • Large positions hedged over aggregators
  • Need constant dialog with client about models and trading methods

 

STATISTICAL ARBITRAGE

  • Very sharp flow
  • Very difficult to monetize
  • PnL transfer
  • Not useful even for hedging

DEVON’S CHALLENGED THE BANK’S CLAIM THAT THEIR USE OF LAST LOOK WAS SOLELY INTENDED TO ELIMINATE TOXIC FLOWS

  • Devon requested the Bank’s internal as well as external trading data and used the combined data and tighter time series to analyze and demonstrate that the Bank’s arguments were shown to be false; didn't stand examination as a justification of Last Look.
  • Devon found clients were classified into different buckets which surprisingly were not consistent with the categories graphed previously.
  • Devon tested client’s classifications to the tolerances chosen in their individual customer Last Look tolerance profile.
  • To do this, Devon used the data parsed into 50 to 100 MB timeseries to track spread decay in different client classes as well as the history of rejected trades vs accepted trades.
  • Devon filtered multiple petabytes of data, including internal communications (e.g. emails) and showed that the Bank was focused on tuning Last Look parameters to maximize P&L rather than focus on Toxic Flow.

 

EXAMPLE OF AN UNFAIR TRADE REJECTION

By reviewing the data in shorter time series Devon is able to examine the market movements within the holding period:

EXPLAINING THE UNFAIR TRADE REJECTION

  1. The client requests to buy EUR/USD and the trade-price is at the current Banks offer rate when the trade arrives. The trade is not an instance of latency arbitrage, neither exactly at the trade event, nor a fraction into the future, nor with respect to the best offer in a time period backward.
  2. In the second half of the holding period, the market moves up about one pip. As a consequence, the mid-price touches the traded price. The trade is caught by the Max Loss test and it is rejected.
  3. The rejection is caused by a movement which begins 250 milliseconds after the trade request. There is no evidence of any causality between this move and a condition existing at the time of the trade request.
  4. Conclusion: Therefore, in our analysis the null hypothesis is that the rejection was caused by a market movement in the holding period, an instance of the Bank capitalizing on the free option in Last Look.

DEVON’S FINDINGS

  • Devon was able to prove through analysing trade data and internal bank communications that the Bank was actually using Last Look, and in particular, the hold time, to maximize PnL rather than to simply protect itself from “toxic flow”
    • Quote from the Banks internal email: A compilation of Last Look reject statistics dated August 15, 2017, compiled by a team member, and sent as email attachment to his superior, reported that “Following the max loss release [in March], rejects decreased on average from 7 to 4 yards per week.” At the end of his email, the team member presents a table with figures representing the additional monthly profit for the Bank from May to August 2015, had the Bank had a default max loss setting of -15 GBP rather than -25 GBP.
  • Furthermore, Devon was also able to prove the Bank could meet the objective to minimize toxic flow using only its maximum loss per trade trigger. Thus Devon demonstrated that;
    • All other aspects being equal, longer holding times and higher thresholds give more rejects and P&L
    • The combination of an asymmetric Max Loss trigger and a hold period allowed the Bank to unfairly take advantage of market movements to increase profits. Simply put the Bank was able to reject negative trades in excess of the Max Loss trigger and keep all positive ones, regardless of the magnitude.

 

CERTAIN OF DEVON’S OTHER FINDINGS RELATED TO E-FX

  • The bulk of The Bank’s arguments for using their chosen Last Look methodology were false; as Devon demonstrated, the bank systematically lied to:
    • Senior management
    • The sales force
    • It’s clients
    • Compliance
    • Regulators
    • Bank of England (FEMR)
  • The Bank built a library of arguments on its need for Last Look: they built models to do scenario analysis on PnL but in reality they tuned Last Look with an emphasis on additional PnL.