Algorithmic trading uses computer rules to decide or manage trading orders: whether to submit them, when, at what price, in what quantity, or how to change them afterward. It makes an instruction repeatable and executable at scale. It also makes a bad instruction repeatable, so the financial result depends on the trading idea, execution conditions and controls surrounding the code.
A portfolio manager can decide to buy a stock manually and delegate its execution to an algorithm. Conversely, a systematic trading model can generate a target position that a person executes manually. Distinguish the rule that chooses the exposure from the machinery that implements it.
What is being automated?
A trading system can automate several different decisions:
| Decision | Example | What still needs to be established |
|---|---|---|
| Investment decision | Increase a position when a forecast crosses a threshold | Whether the forecast contains a useful alpha signal after costs and risk |
| Execution schedule | Divide an authorised purchase across a trading window | How much urgency and incomplete execution the objective permits |
| Order management | Set limits, cancel stale orders or replace remaining quantity | What the venue has actually accepted, filled or cancelled |
These decisions may run together or in separate components. A simple deterministic schedule qualifies as an algorithm; machine learning is optional. Neither complexity nor automation proves profitability.
There are also jurisdiction-specific definitions. In the EU, MiFID II Article 4(1)(39) focuses on automatic determination of order parameters with limited human intervention. It excludes systems that only route or process orders without determining trading parameters. Its separate definition of high-frequency trading adds features concerning latency-minimising infrastructure, automatic order decisions and high intraday message rates. Sending an order electronically does not by itself settle either classification.
One purchase, four child orders
Suppose a human authorises a 100-share purchase over a chosen trading window. A fictional execution rule allocates 25 shares to each of four equal-duration intervals. The overall instruction is the parent order; the smaller submitted instructions are child orders. Equal time slices resemble a basic time-weighted average price execution schedule.
For an arithmetic illustration, stipulate that every child fills completely, at USD 100, 101, 102 and 101 per share, respectively. Ignore fees. These invented fills cost USD 2,500, 2,525, 2,550 and 2,525, giving
The average execution price is the quantity-weighted average of the fills:
Original illustrative ledger. Four equal-duration intervals and complete fills are stipulated; fees are excluded. The schedule sets targets, while actual executions supply the cash ledger.
This is the VWAP of these fills, not the market's VWAP during the window. Equal fill quantities happen to make it equal to the simple average of the four prices. With unequal quantities, use quantity weights.
The USD 101 result establishes neither a trading profit nor execution skill. There is no subsequent sale or final valuation, and no declared benchmark or alternative execution path. We also cannot infer that splitting the order improved the price: an immediate purchase would have faced a different market interaction.
TWAP, VWAP and Participation Algorithms develops the distinction between target allocations, benchmarks and the order policies needed to implement them. The full lesson requires membership; its public page provides a preview.
The missing instructions are where risk enters
“Buy equal slices” leaves important decisions unresolved. A market order seeks execution against available liquidity, but cannot fix all fill prices in advance. A limit order constrains price and may remain unfilled. The algorithm needs rules for partial fills, rejected orders, changing liquidity, price limits and any remaining quantity at the deadline. Trading faster can increase market impact; waiting can expose the parent to adverse price moves or leave its intended position incomplete.
Order state matters as much as the schedule. If a cancellation request has not been resolved, the old child may still fill. Sending its full replacement immediately can buy more than the parent authorised. Likewise, a lost acknowledgement does not prove that an order was rejected. Systems need reconciled order and execution records, limits that account for outstanding orders, and a defined response when state becomes uncertain.
Live Execution Monitoring works through cancellation races, pending-order exposure and controlled recovery. It also requires membership, with a public preview. Stopping new submissions alone does not remove existing positions or guarantee cancellation of working orders.
These are operational concerns as well as economic ones. FINRA's Regulatory Notice 15-09, addressed to its U.S. member firms, discusses development controls, testing and supervision through deployment and monitoring. In the EU, MiFID II Article 17 sets requirements for relevant investment firms' systems, controls, testing, monitoring and continuity arrangements. The applicable obligations depend on the firm, activity and jurisdiction.
For a quant, the useful evaluation is therefore broader than “did the code send orders?” Test the information available when decisions were made, feasible fills and transaction costs, then reconcile intended exposure with executed exposure. A historical backtest cannot establish live queue position or future liquidity. Automation implements a policy; evidence must establish whether that policy remains economically and operationally sound.
Sources
- Rishi K. Narang, Inside the Black Box: A Simple Guide to Quantitative and High-Frequency Trading, 2nd edition (Wiley, 2013), Chapter 7, pp. 119–121 and 129–131 — execution objectives, aggressive and passive choices, execution measurement and trading infrastructure. The ledger and figure above are original.
- ESMA: MiFID II Article 4, definitions (39)–(40) — EU definitions of algorithmic trading and high-frequency algorithmic trading technique; checked 26 October 2026.
- ESMA: MiFID II Article 17, Algorithmic trading — systems and control requirements for relevant investment firms; checked 26 October 2026.
- FINRA: Regulatory Notice 15-09 (26 March 2015) — supervision and control practices for algorithmic strategies, including testing, deployment and monitoring; checked 26 October 2026.