Appendix A: Glossary

Agentic AI — AI systems that don't just analyze or recommend but take actions: placing trades, switching strategies, sizing positions. The 2026 frontier — and the area needing the most caution.

Algorithmic trading — Executing trades via pre-programmed instructions (price, timing, volume). The broad category; AI trading is a subset.

Alpha — Return above what you'd expect given the risk taken. The thing every strategy claims to have.

Alternative data — Non-traditional datasets used for investing: satellite imagery, credit-card transactions, geolocation, web-scraped prices, social sentiment.

Backtest — Testing a strategy against historical data. Useful for falsification; dangerous as proof. See overfitting.

Black-Scholes — The classic options-pricing formula. Still the benchmark ML pricing models are measured against.

Colocation — Placing your servers inside the exchange's data center to minimize latency. The physical foundation of HFT.

Copy trading — Automatically mirroring another trader's positions, as on eToro's CopyTrader.

Dark pool — A private trading venue where orders aren't displayed publicly. Institutions use them to trade size without moving prices.

Data snooping — Testing many strategy variants and reporting only the winner. With 50 variants, there's a 92% chance one looks good by pure luck.

Drawdown (maximum) — The worst peak-to-trough loss a strategy suffered. More informative than returns alone.

Execution algorithm — A system for working a large order into the market over time: TWAP, VWAP, POV, Implementation Shortfall.

Factor model — An investment model attributing returns to underlying factors (value, momentum, size, quality...). ML versions learn factors from data rather than specifying them.

Fiduciary — A legal standard requiring someone to act in your best interests. Human advisers can owe it; almost no trading bot or signal vendor does.

FinBERT — A language model fine-tuned on financial text for sentiment analysis, especially earnings calls and filings.

Flash crash — A rapid, deep, brief market plunge driven by automated trading interactions, like May 6, 2010.

HFT (high-frequency trading) — Trading strategies operating at microsecond-to-millisecond timescales, requiring colocation and direct feeds. Not available to retail at any price.

Implementation Shortfall — An execution framework optimizing the tradeoff between market impact and timing risk.

Kill switch — A single action that flattens all positions and halts all trading. Non-negotiable for any automated system.

Latency arbitrage — Profiting from momentary price differences across venues by reacting fastest. Requires institutional infrastructure.

Look-ahead bias — Using information in a backtest that wasn't available at the decision time. Invisible, fatal.

Machine learning (ML) — Algorithms that learn patterns from data: regression, random forests, gradient boosting, neural networks, reinforcement learning.

MAR ratio — Annualized return divided by maximum drawdown. Higher is better; under 1.0 means the worst loss exceeded the average annual gain.

Market making — Simultaneously quoting buy and sell prices, profiting from the bid-ask spread. What Jane Street, Citadel Securities, and Virtu do at scale.

MiCA — The EU's Markets in Crypto-Assets regulation, fully effective December 2025.

MiFID II — The EU's markets directive; Article 17 governs algorithmic trading (order records, market-maker agreements, system testing).

NLP (natural language processing) — AI that reads and interprets human language: news, filings, earnings calls, social media.

Overfitting — A model that memorized historical noise instead of learning real patterns. Perfect backtest, failed future.

Paper trading — Running a strategy against live market data with fake money. Tests plumbing, not profitability.

Pattern day trader — A regulatory classification for frequent traders; FINRA replaced the old $25,000 minimum with new intraday margin standards effective June 2026.

POV/PVOL — Participation-of-volume execution: trade a fixed share of whatever volume appears.

Quant (quantitative) fund — An investment fund driven by mathematical models rather than human stock-picking: Renaissance, Two Sigma, D.E. Shaw.

RegTech — Technology for regulatory compliance: surveillance, reporting, audit trails, model governance.

Reinforcement learning (RL) — ML where an agent learns by trial and error against a reward signal. Used in adaptive trade execution.

Robo-advisor — Automated investment management: diversified portfolios, rebalancing, tax-loss harvesting. Betterment, Wealthfront, Schwab, Fidelity Go, SoFi.

Sharpe ratio — Excess return per unit of volatility. Rough guide: 1–2 is good professional territory; above 3, suspect overfitting.

Signal — A measurable input suggesting an expected return — price patterns, sentiment scores, alternative-data readings.

Slippage — The difference between a signal's price and your actual fill. The backtest-to-reality tax.

Smart order router — Software that finds the best available price across exchanges for your order. You already benefit from this.

Survivorship bias — Testing only on assets that survived (today's index members), making the past look better than it was.

Tax-loss harvesting — Selling losers to offset gains, then buying similar securities. A robo-advisor staple.

TCA (transaction-cost analysis) — Measuring what trades actually cost versus benchmarks. Execution quality, quantified.

TWAP / VWAP — Time-weighted and volume-weighted average price execution strategies.

Walk-forward testing — Validating a strategy by repeatedly fitting on a rolling historical window and testing the next window. Mimics real deployment.

Wash sale — Selling at a loss and repurchasing a substantially identical security within 30 days (before or after); the loss is disallowed currently. Applies across all accounts, including a spouse's. Inside IRAs, the loss is permanently destroyed.