The Evolution of Algorithmic Trading

Algorithmic trading is a relatively modern phenomenon. As recently as the late 1990s, half of the New York Stock Exchange's trading volume was executed manually by floor traders. While the other half was processed electronically, humans still directed the vast majority of this flow; only a tiny fraction was managed by what we now consider "algorithms." It wasn't until the early to mid-2000s that these tools gained widespread traction among institutional traders.

Today, algorithms dominate equity markets. Although slower to adopt, traditionally manual markets—like fixed income and foreign exchange—now heavily rely on them as well. Looking forward, algorithmic trading is poised for even greater expansion. Thanks to substantial economies of scale and continuous advancements in technology and data analytics, algorithm performance is sharper than ever. Many industry professionals predict that algorithms will soon become the default execution method across most asset classes, eventually dropping the "algorithmic" prefix to be known simply as "trading."

What is an Algorithm?

To discuss this topic accurately, we first need a clear definition. At its core, a trading algorithm is a set of coded instructions designed to execute orders based on a predetermined strategy. These algorithms can be categorized in several distinct ways:

Categorization by Order Type

• Single-Order Algorithms: Designed to process one order at a time. If fed a list of orders, the algorithm will trade them simultaneously but completely independently of one another.

• List-Based Algorithms: Designed to execute multiple orders jointly. When trading a specific basket, this algorithm factors in how every other order in that list is being handled.

Categorization by Function

• Smart Routers: These route orders to one or more venues at a specific moment, typically to secure the best available price.

• Algorithms (Slicing): While smart routers are technically algorithms, traders usually reserve the term "algorithm" for programs that slice a large order into smaller pieces, "working" it over time to achieve a specific benchmark. These slicing algorithms frequently rely on smart routers to direct their generated "child" orders.

Categorization by Pacing Approach

Algorithms typically pace orders over time in one of three primary ways:

• Schedule-Based Algorithms: These execute orders according to a predetermined schedule. Common examples include Time-Weighted Average Price (TWAP), Volume-Weighted Average Price (VWAP), and variations of Arrival Price/Implementation Shortfall (IS). Although the overarching pacing is scheduled, the specific execution tactics remain highly dynamic.

• Dynamic Algorithms: These adjust their trading behavior based on real-time market shifts, such as accelerating or decelerating in response to volume changes. Percent of Volume (POV)—also known as Volume Participation (VP)—and IS variants are standard examples.

• Opportunistic Algorithms: These operate in a more episodic, binary fashion, executing trades only when highly favorable conditions arise (e.g., a tight bid-ask spread combined with deep quote depth). If no opportunity presents itself, the algorithm will not trade. This is ideal for large orders or illiquid assets where forced schedules could cause excessive market impact.

• Hybrid Variants: These combine opportunistic trading with backup logic. They prioritize highly favorable conditions but will periodically execute smaller trades to ensure the order doesn't fall behind a minimum threshold, such as a VWAP schedule.

Execution Components

Beyond the high-level strategy used to slice a parent order, algorithms rely on shared sub-components to execute the resulting child pieces. Because different algorithms require similar functionalities, these components are frequently shared but parameterized to meet specific goals:

• Routing Logic (Smart Routers): Selects the optimal venue for the child orders.

• Display Logic: Dictates how much of the order is visible to the market, how much is hidden by the venue, and how much is held "in reserve" until conditions change or the visible portion fills.

• Pricing Logic: Determines how aggressively to price a child order, directly impacting its likelihood of execution.

The Execution Lifecycle: An Example

Imagine a TWAP algorithm designed to trade steadily over a set period. It first slices the parent order into evenly spaced, equal-sized child orders. For each slice, the pricing logic determines the aggression level, and the display logic sets the visible quantity. The smart router then selects the best venue and dispatches the order.

Throughout this process, the algorithm continuously monitors the market. If conditions shift, it is ready to resize, reprice, or reroute the order. This cycle repeats until all child orders are executed, or the order is canceled (either manually by the client or automatically due to time or price constraints).

Course Outline Overview

• Part 2: A broad introduction to algorithmic trading and core conceptual challenges.

• Part 3: Essential practical terminology used throughout the course.

• Part 4: Strategic factors that influence trading decisions and algorithmic design, laying an intuitive theoretical foundation.

Section 2: High-Level Strategies

• Modules 4–7: A detailed exploration of single-order algorithms (schedule-based, dynamic, and opportunistic) and how they navigate real-world market complexities.

• Module 8: A look at multi-order algorithms (e.g., pairs trading, portfolio strategies) and the unique complications of list-based execution.

Section 3: Child Order Execution

• Part 11: A deep dive into pricing and sizing logic for child orders.

• Part 12: Smart routing for marketable and non-marketable orders. This covers the optimal selection of venues, differentiating between "lit" (visible) and "dark" (hidden) liquidity pools.

Section 4: Performance Measurement

• Module 11: An exploration of the most widely used trading benchmarks and evaluation metrics.

• Module 12: Specific empirical issues that frequently complicate the evaluation of trade execution and performance.