PRA - Consultation Paper | CP5/18 , "Algorithmic trading". February 2018
PRA - Supervisory Statement 5/18 , Algorithmic trading
MiFID II - Algorithmic trading - AFS summary
FCA : Algorithmic and High Frequency Trading (HFT) Requirements
Acronyms and terminology:
MTF - Multilateral Trading Facility
OTF - Organized Trading Facility
HFT - High Frequency Trading
DMA/DEA - Direct Market Access / Direct Electronic Access
PnL - Profit and Loss
Trader - owner of trading system
Algorithm - model interpreting the market and own performance
System - embeds market, own portfolio views, algorithm and order execution module
Validation of algorithmic trading is an exercise to prove integrity of the following items:
- Stress resiliency of the algorithm (protection against singularities or internal self-generating order flow) with respect to different scenarios:
- simulation of quote flooding
- physical disconnection
- develop sequence of actions
- check risk reports
- focus on biggest risks
- close position or
- hedge
- turn to alternative connections, brokers
- Compensating controls
- Pre-trade control includes estimations of:
- profit margin (alternatively distribution of)
- fee
- risk margin charged by exchange
- Kill-switch (Red button, ) is a combination of
- hardware based blocks (circuit breaker disconnects from the market)
- software based blocks (control over single order gate, like order flow internalizer)
- foreseen implementation of post-disaster scenarios, like
- shut down all orders outflow
- recall/cancel all orders outstanding
- neutralize unnecessary exposure (to the best knowledge of the system)
- develop and test the system against such post-disaster scenarios:
- What-if all market goes all UP/DOWN,
- All orders move away from the mid-price
- etc.
- Backtesting algorithm:
- portfolio (PnL) performance
- price prediction
- order flow prediction vs realized LOB dynamics
- order execution vs market reaction
- system performance due to:
- latency
- memory capacity
- internal responsiveness
- algorithm performance during periods of stress
- requires development of time/flow-pressure simulator
- algorithm may work differently under sequential tick injection and under time pressure, if/when speed has priority over smart result
- Feed integrity (completeness and reliability of available information)
- how many messages are lost
- how does it impact the decision flow
- Risk controls:
- realtime control of own view vs exchange view over
- order mismatch
- position mismatch
- same for end-of-day settlements (alignment with Settlements)
- impact of variability of latency (if not co-located, but still even with collocation there might be a problem)
- when all above is working, check Market risk
- Resilience of trading systems:
- reliable connectivity
- presence of counter-balancing mechanisms embedded into the system allowing for compensation of the damage - so-called feedback mechanisms
Validator has to check algorithm performance keeping in mind that the following is avoided:
- disorderly market - voluntary or accidental destabilizing impact is equally damaging for the reputation of the Trader
- market abuse:
- huge inventory or huge competitive technical advantage may destabilize the market to the profit of the owner of the algorithm. In case of such event it would be difficult to prove the innocence.
- Intentional abuse of price formation versus behaviour of market players
- business risk arises if Trader is a member of exchange with obligations to maintain the market and if system performance is non perfect:
- How must such trader react to failure of his system? Can he fulfill his/her obligations
Ensure that the responsible Trader knows and understands the behavior of the system. All above items requires reflection in policies.
No comments:
Post a Comment