Trading Desk Setup & Procedures

Trading Desk Setup

1.1. Our trading setup is made up of 4 servers, located in key locations in order to minimize exchange connection times as much as possible.

1.2. All exchange API interaction scripting and order/position management algorithms have been built in-house and no part of the infrastructure has been outsourced or pulled from github.

1.3. Generally, stop orders are kept by our servers until triggered and only emergency stop-loss orders are placed directly with the exchanges, as our experience has shown better results when doing so.

Automated Trading System

2.1. Description:
Our main strategy started out as basic, almost infantile at its core and combines simple observations about the market with a purely statistical approach to order placement and risk management, while taking advantage of the full range of available instruments to extract the maximum amount of profit from the open markets (tight stop losses, monitored trailing take profit orders, maker-only trade entries and exits, whenever possible).

2.2. Direction:
We trade both sides of the market asymmetrically, over different exchange accounts, so both longs and shorts are possible at the same time, should the opportunity present itself, but without doubling down in the same direction if conditions worsen.

2.3. Affinity:
This system is most at home on exchanges that also allow shorting the market and can be categorised as mechanical, based purely on price action and order book analysis, as it does not automatically factor in any technical or fundamental data. We input any bias resulted from technical analysis or fundamental factors manually, when needed.

2.4. Entry:
Entries are performed based on order book monitoring for specific market depth ratio triggers, on particular weighted average and convex combinations values of existing order volume. We use weighted averages and parametrized convex combinations instead of simple averages to help us differentiate not just between present volume numbers on both sides of the market, but also between the various shapes of market depth, as in our experience this too can greatly influences outcomes.

2.5. Market Type:
We initially designed the strategy for range-bound markets, but it can also be very profitable in its current state in linearly trending markets, due to the tight stops and trailing profit lock-ins.
Risk of deeper drawdown increases on trends that experience large short-term volatility, but so does the possibility of greater profits, so we can say that rough seas generate an increase in result amplitude rather than being less profitable than steadier markets.

2.6. Losses:
Drawdown is mostly kept linear and individual loses cannot spiral out of control under normal market conditions due to the consistent use of tight stops and temperate position sizing. A decline in profitability can always be observed statistically and there is no emotional or mathematical bias within this algorithm.
It can also be easily winded down if markets prove unfavorable.

2.7. Weaknesses:
The shortcomings of this strategy stand with vertical and lateral scaling and in theory has its limitations in terms of how much value can be efficiently attributed to it:
As we increase position size(vertical scaling), only fractions of entries might start getting executed, therefore actually diluting profits instead of generating more gains with more capital at its disposal.
When spreading assets over more and more independent accounts (lateral scaling), detected opportunities in the market could become insufficient and entry orders could start overlapping to such an extent that the strategy is effectively risking a larger than intended percentage of capital on identical triggers.

2.8. Limitations:
Taking into account current and forecasted market liquidity and variety, some scaling problems might start to arise north of the $10M mark. As we grow capital allocation for this algorithm, we expect to encounter performance plateaus along the line, during which we will temporarily stop allocating more funds until we can work at the new volume effectively.

2.9. Roadmap:
We are constantly working on improving our trading methods. Here is what's in the works for this algorithm:

  • effectively using machine learning for order book analysis and pattern definition
  • integrating order book reaction times and generally factoring in timings into entry and exit triggers
  • extracting valuable unitary insight from multi-exchange order book monitoring, such as:
    • reliably identifying large OTC order execution across exchanges
    • generating actionable geographical trading behaviour patterns

2.10. Fun Fact:
We call our system the Monkey Brain, due to its superhuman ability to follow simple rules and actually close losing trades.