Skip to content
Genesis Ark

Risk

Pre-trade risk controls every crypto broker needs

Pre-trade checks are the last line between a mistake and a loss. This is a practical list of the controls a crypto broker should run before any order reaches a venue, why each one matters, how to set it, and what the system behind them has to get right.

Author
Genesis Ark Engineering
Published
Reading time
8 min read

Most trading losses that make the news did not come from a bad market view. They came from an order that should never have left the building: a misplaced decimal, a strategy stuck in a loop, a client trading beyond their balance, an order in an instrument the firm was not allowed to trade.

Pre-trade risk controls exist to stop those orders. In crypto, where markets run continuously, liquidity can be thin, and venues settle immediately, there is no back office to catch an error the next morning. If the check does not happen before the order is sent, it does not happen.

This article lists the controls a crypto broker should run on every order, how to think about setting each one, and what the system behind them has to get right.

Principles first

Three principles shape everything below.

Checks run before the venue, every time. A control that runs after an order is sent is monitoring, not prevention. Every order path, including manual orders, API orders, algorithm child orders and hedges, must pass through the same checks.

Checks use live state. A credit check against a balance snapshot from five minutes ago is a check against the wrong number. Pre-trade risk must read positions and balances that include every fill received so far, and every order still open.

A rejection must explain itself. A trader or a client whose order is rejected needs to know which rule failed and by how much. Unexplained rejections lead to people looking for ways around the control.

1. Balance and credit checks

The most basic control: does the account have the money or the asset for this order?

  • Buy orders check that the account has enough of the quote currency, or enough credit, to cover the full notional plus fees.
  • Sell orders check that the account holds the asset, unless short selling is permitted.
  • Open orders count. Balance already committed to open orders must be reserved, or two orders can both pass against the same funds.

For brokers acting as principal or offering credit, the check is against a credit limit rather than a cash balance, but the logic is the same: total exposure after this order must stay within the limit.

2. Order size and notional limits

Limits on how large a single order can be, set per account, per client group and per instrument.

  • Maximum quantity per order, in units of the asset
  • Maximum notional per order, in the reporting currency
  • Minimum order size, to reject orders below what the venue will accept and save a round trip

These catch the fat-finger error: 100 instead of 1.00, or a quantity entered in the wrong unit. Set them with reference to normal client behavior. A limit at ten times the largest order a client has ever placed catches errors without blocking real business.

3. Price collars

A price collar rejects limit orders priced too far from the current market. A buy order priced far above the best offer, or a sell order far below the best bid, is almost always an error, and on a thin book it can fill at a terrible price.

  • Set the collar as a percentage or number of basis points from a reference price, such as the mid or the last trade.
  • Make it wider for volatile or illiquid instruments and narrower for liquid ones.
  • Decide what happens when market data is stale. A collar computed from an old price is worse than none. The safe default is to reject orders when the reference price is too old.

For market orders, which have no limit price, the equivalent control is a maximum slippage: the order may not fill beyond a set distance from the reference price. Many brokers convert client market orders into limit orders at the collar for this reason.

Fig. 1. Limit utilization per account, with each check showing the value it used and the rule that would reject. Illustrative composition with sample data.

4. Position and exposure limits

Limits on the total position an account or the firm can build, rather than on a single order.

  • Per account and instrument: the maximum long or short position in an asset
  • Per account, total: the maximum gross or net exposure across all assets
  • Firm-wide: the house position the broker is willing to carry, particularly important for principal and riskless-principal models where hedging may lag client flow
  • Concentration: the maximum share of exposure in a single asset or at a single venue

Venue exposure is a crypto-specific concern. Balances sit at each venue, and venues can halt withdrawals or fail. A limit on how much of the firm's capital can be held at any one venue is a sensible control in its own right.

5. Order rate and message limits

A strategy that loops, or a client integration that retries without waiting, can send thousands of orders in seconds.

  • Orders per second and orders per minute per account and per session
  • Cancel and replace rate limits, which also protect against patterns regulators may view as manipulative
  • Open order count per account and instrument

These limits protect the broker's own venue rate limits too. If one client exhausts the broker's API quota at a venue, every other client's orders at that venue fail.

6. Restricted instruments and lists

  • Restricted list: instruments the account or the firm may not trade at all, for regulatory, compliance or business reasons
  • Watch list: instruments that may be traded but generate an alert for compliance review
  • Client eligibility: for example, derivatives or certain tokens available only to clients who meet suitability or classification requirements in their jurisdiction

These lists change often. They should be maintained by compliance, apply immediately and leave an audit record of every change.

7. Duplicate order detection

Network retries, impatient users and buggy integrations can all send the same order twice.

  • Idempotency: a repeated client order ID or idempotency key returns the original order instead of creating a new one
  • Duplicate pattern check: an optional warning or rejection when an order with the same account, instrument, side, quantity and price arrives within a short window

8. The kill switch

Every broker needs a way to stop trading quickly, for a strategy, a trader, a client, an instrument, a venue or the whole firm.

A useful kill switch does two things: it rejects all new orders in its scope, and it cancels all open orders in its scope across every venue. It should be triggered by an authorized person or automatically, for example when a loss threshold or an order-rate threshold is crossed. It should be tested regularly, because a kill switch that has never been used is a kill switch you do not know works.

Re-enabling trading after a kill switch should require an explicit decision by an authorized person, and both events should be in the audit trail.

9. Loss limits

Some accounts, particularly proprietary strategies and clients trading on credit, should carry a daily or rolling loss limit. When realized plus unrealized losses cross the limit, the account can be restricted to reducing positions only, or stopped entirely.

Loss limits depend on accurate, current P&L, which in turn depends on accurate positions and reliable prices. They are only as good as the ledger behind them.

Setting limits in practice

A few practical rules make limits useful rather than obstructive.

  • Start from data. Use each client's actual order history to set initial limits, with headroom. Limits picked from the air are either too tight to trade or too loose to matter.
  • Tier them. Group clients into tiers with standard limits, and allow individual overrides with approval.
  • Separate soft and hard limits. A soft limit warns the trader or asks for confirmation. A hard limit rejects. Use soft limits where judgment is appropriate and hard limits where it is not.
  • Require a second approver for raising limits, and record who approved what.
  • Review regularly. Clients grow, markets change and limits drift out of date. A quarterly review against actual usage is a reasonable minimum.

What the system behind the checks must get right

Pre-trade controls are only as reliable as the system that runs them.

  • One path for every order. If any route to a venue bypasses the checks, such as a hedging script or an admin tool, the controls have a hole.
  • Live positions. The checks must read a ledger that updates on every fill and accounts for open orders. That ledger must be correct after a venue disconnect and after a restart.
  • Fail closed. If the risk service cannot determine whether an order is safe, because a price is stale or a position is uncertain, the order should be rejected, not waved through.
  • Speed. Checks add time to every order. They need to be fast enough that traders and strategies are not tempted to route around them.
  • A full record. Every check, pass or fail, with the values it used, belongs in the audit trail. When a regulator or a client asks why an order was accepted or rejected, the answer should be one query away.

A checklist to take away

Before your next client goes live, confirm that you can answer yes to each of these.

  1. Every order path, including hedges and manual orders, passes through the same pre-trade checks.
  2. Balance and credit checks include open orders and the latest fills.
  3. Every account has order size, notional and position limits.
  4. Price collars are active, and orders are rejected when the reference price is stale.
  5. Order rate limits protect both the client and your venue quotas.
  6. Restricted lists are maintained by compliance and apply immediately.
  7. Duplicate orders are prevented by idempotent order IDs.
  8. A kill switch exists at every level you need, and you have tested it recently.
  9. Limit changes require a second approver and are recorded.
  10. Every rejection tells the user which rule failed.

None of these controls is exotic. Together they are the difference between a mistake that is caught in milliseconds and one that is discovered in the morning.

See it on your own workflow.

A scoping call with an engineer, then a demo on sample data that mirrors your venues, desks and limits.