Algorithmic trading

Algorithmic trading (or “algo trading”) is the use of computer programs to automatically execute trades based on pre-set rules. An algorithm might slice a large order into small pieces and execute them throughout the day, or search for price patterns and execute when conditions are met. Algorithms range from simple (execute 10,000 shares evenly over the next hour) to complex (adapt to real-time volume, market microstructure, and estimated price impact). The vast majority of institutional trading is algorithmic.

Why algorithmic trading?

Market impact reduction: A 100,000-share order placed at once would move the price sharply against you. An algorithm breaks it into 20 pieces of 5,000 shares each, spread throughout the day. The market does not see a huge buyer; prices move less.

Average price improvement: By spreading execution across the day or week, you often achieve better average prices than a single large trade.

Emotion removal: A human trader might panic and sell everything if the price dips slightly. An algorithm sticks to the plan, executing methodically.

Scale: A single algorithm can manage dozens of simultaneous orders across multiple securities and venues.

Common algorithmic strategies

VWAP (volume-weighted average price): Slices the order to target executing at the day’s VWAP. See VWAP order.

TWAP (time-weighted average price): Slices evenly over time. See TWAP order.

POV (participation of volume): Execute at a percentage of the observed volume. If 1 million shares trade throughout the day, and you set POV to 10%, the algorithm buys or sells roughly 100,000 shares as the 1 million move through.

IS (Implementation shortfall): Execute to minimize the “shortfall” between your target price and the actual execution price, balancing speed and market impact.

Execution inline with volume: Detect where large volumes are trading and execute near those prices and times.

Algorithmic trading at different scales

Institutional (typical):

• Large fund wants to buy 1 million shares over one week.
• Submits an algorithm (e.g., VWAP) to their broker or a vendor.
• The algorithm automatically breaks it into pieces, executes across venues, adapts to volume, and achieves average execution near VWAP.
• Result: better average price with minimal market impact.

Retail (limited):

• Most retail brokers offer limited algorithmic options (usually just time-based slicing).
• Interactive Brokers and some others offer more sophisticated algorithms.

Proprietary (HFT):

• HFT firms use extremely sophisticated proprietary algorithms that detect patterns, exploit arbitrage, and execute in microseconds.

Smart order routing and algorithms

A smart order router is a type of algorithm that:

• Receives your order.
• Checks prices on all available venues (lit exchanges, dark pools, etc.).
• Routes your order to the venue(s) offering the best price.
• Adapts as prices change.

Smart order routing is a form of algorithmic trading that optimizes for best execution.

Risks and challenges of algorithmic trading

Overfitting: An algorithm optimized for historical data might fail on new data. For example, an algorithm tuned to execute in quiet markets might fail badly during volatile periods.

Flash crashes: If many algorithms unwind simultaneously (all trying to sell at once), it can cause rapid price declines and market instability. The 2010 Flash Crash involved algorithmic feedback loops.

Latency sensitivity: If a market moves faster than the algorithm can react, results can be poor.

Slippage during stress: In volatile markets (earnings, crises), even sophisticated algorithms can underperform.

Regulation of algorithmic trading

U.S. and international regulators require:

• Disclosure: Brokers must disclose that orders are routed algorithmically.
• Kill switches: Algorithms must have automated safeguards to stop trading if something goes wrong.
• Backtesting: Brokers and algo vendors must test their algorithms on historical data.
• Monitoring: Regulators monitor for suspicious patterns (e.g., too many small orders that seem designed to mislead).
• Best execution: Algorithms must be designed to achieve best execution, not to benefit the broker.

Algorithmic trading fees

Algo services are typically priced:

• Flat fee: e.g., $50–$500 per algorithm execution.
• Basis point fee: e.g., 0.1–1 basis point (0.001–0.01%) of the order value.
• Included with brokerage: Many brokers include basic algorithms for free; advanced algorithms charge extra.

The future of algorithmic trading

Machine learning: Newer algorithms use AI and machine learning to adapt to changing market conditions.

Multi-asset: Algorithms increasingly span stocks, options, futures, and other instruments simultaneously.

Reinforcement learning: Some firms experiment with algorithms that learn and improve over time.

Regulation: Expect tighter rules on algorithm transparency, testing, and safeguards.

See also