Skip to content

Trading system

A trading system connects data, signals, sizing, risk, execution, accounting and monitoring so intended trades become controlled, reconciled positions.

A trading system is the connected set of data, decision rules, controls and operating processes that turns a trading idea into orders, positions and an accountable financial result. In quantitative trading, it connects signals, sizing, risk, execution, accounting and monitoring. A sound forecast can still produce the wrong exposure if any of those connections fails.
A trading strategy describes the intended policy for taking and changing exposure. The system makes that policy operational: which data arrive, which version of the rules runs, what orders remain active, what actually filled, and how exceptions are handled. The boundary varies between organisations; the useful distinction is intended behaviour versus the machinery and procedures needed to carry it out and verify it.

Follow the state, not just the signal

A trading signal is an input to a decision. A target is a desired holding. An order asks for a transaction. A fill changes the actual holding. Keeping these objects separate prevents a system from booking an intended trade as an executed one.
FunctionWhat it must establish
DataInstrument identities, timestamps, information availability and quality checks. A corrected historical series is not automatically what the trader knew at the time.
SignalsThe forecast, score or condition produced by the authorised rule using available information.
SizingThe desired holdings after considering existing positions, cash, expected trading costs and competing portfolio demands.
Risk controlsWhether the proposed exposure and potentially executable orders fit the mandate and limits; what blocks or escalates a decision.
ExecutionSubmission, acknowledgements, partial fills, rejections, cancellations and the remaining order state.
AccountingConfirmed holdings, cash, fees, valuations and reconciliations against authoritative records.
MonitoringWhether data, orders, exposures, economics and infrastructure behave as expected, with a defined response when they do not.

These are functions, not a requirement for separate software services. A small system may combine them, and people may perform some steps. A forecasting notebook or order-entry screen alone does not cover the full workflow.

Data feeds signals, sizing and risk checks, then execution. Fill reports feed accounting; monitoring checks data, orders and the ledger and can block new submissions. Recorded holdings feed the next sizing decision.
Data feeds signals, sizing and risk checks, then execution. Fill reports feed accounting; monitoring checks data, orders and the ledger and can block new submissions. Recorded holdings feed the next sizing decision.
Original conceptual schematic. Solid arrows show decision or state flow; dashed arrows show observation or control. The functions can be combined and do not describe an implemented product.
The return path matters. The next target must be compared with recorded holdings and outstanding orders, rather than the assumption that every earlier instruction completed. Positions, Trades, Fills and P&L develops that distinction and the cash-and-holdings ledger. The full lesson requires membership; the linked page offers a public preview.

The ledger is a test of the connections

Consider a fictional stock quoted in USD per share. Starting with no shares, the system buys 10 shares at USD 100, then sells all 10 at USD 105. Stipulate complete fills and an explicit fee of USD 2 per side. There are no distributions, financing costs, taxes or currency conversions. The entries below are changes in cash; the purchase requires sufficient funded cash, whose amount is not otherwise specified.
Recorded eventChange in sharesTrading cash flowFee cash flowNet cash change
Purchase fill+10−USD 1,000−USD 2−USD 1,002
Sale fill−10+USD 1,050−USD 2+USD 1,048
Round trip0+USD 50−USD 4+USD 46

With no remaining shares or other cash flows, the cash reconciliation gives the completed trade's net profit:

Net P&L=10(105−100)−(2+2)=46 USD.\text{Net P\&L}=10(105-100)-(2+2)=46\text{ USD}.

The system should be able to connect this result to the decision, orders, unique execution records and fees that produced it. Replaying an already applied fill must not add the shares or cash flow again. A target of 10 shares is insufficient evidence for the purchase entry; the stipulated fills are what justify it.

The fill prices already enter the ledger. Measuring slippage against a reference price can explain execution quality, but charging the same price difference again would double-count that effect. Applicable transaction costs outside the fills and stated fees would require additional entries.

USD 46 verifies this invented reconciliation. It does not establish that the strategy has an edge, that its orders were permissible, or that the system will remain reliable. It is also a cash profit, not a return: no investment-capital denominator or holding period has been specified.

A stop is an action, not a position report

Suppose the system loses contact with its execution venue after submitting an order. A timeout establishes that a response was not received; it does not establish rejection. Automatically sending the same instruction again can create duplicate exposure.

A defined response can block further submissions, retain unresolved orders as potentially executable, obtain authoritative order and fill records, and reconcile holdings before resuming. Stopping new orders does not close existing positions. Requesting cancellation does not prove that every working order has been cancelled.

Live Execution Monitoring works through cancellation races, outstanding-quantity checks and controlled recovery. It is also a membership lesson with a public preview.
Testing therefore has two questions to answer. Does the backtest evaluate the intended policy with credible information timing, fills and costs? Does the operating system preserve correct state through stale data, duplicate messages, partial execution and restarts? Attractive simulated performance answers neither question by itself.
For U.S. FINRA member firms using algorithmic strategies, Regulatory Notice 15-09 discusses development controls, testing, post-deployment monitoring and reconciliation. Those practices have a specific regulatory context; they also illustrate why a trading system needs a lifecycle beyond generating buy and sell signals.

Sources

Last updated
  • trading infrastructure
  • execution
  • risk management
  • reconciliation
  • monitoring
Keep connecting the ideas
← Back to the glossaryBack to top ↑