Glossary
Key terms in quantitative finance, factor investing, and market microstructure.
A
- Alpha
- Return that cannot be explained by exposure to systematic risk factors. True alpha is persistent out-of-sample; in-sample alpha is easy to manufacture by overfitting.
- Adverse selection
- The risk a market maker faces when trading against a counterparty who has better information. A market maker who is adversely selected loses money to informed traders.
- Annualised return
- The compound annual growth rate implied by a sequence of daily returns. Computed as (∏(1+rₜ))^(252/T) − 1 for T trading days.
B
- Bid-ask spread
- The difference between the lowest price a seller will accept (ask) and the highest price a buyer will pay (bid). Wider spreads indicate lower liquidity or higher adverse-selection risk.
C
- Capacity
- The maximum amount of capital a strategy can deploy before its own trading moves prices enough to erode returns. Small-cap strategies typically have lower capacity than large-cap ones.
- Cross-sectional ranking
- Ranking securities against each other at a single point in time, rather than comparing a security's current value to its own history. Most factor signals are cross-sectional.
F
- Factor exposure
- The sensitivity of a portfolio's returns to a systematic risk factor (e.g. market, size, value). A factor-neutral portfolio has zero net exposure to known factors.
- Factor zoo
- The large and growing collection of published equity factors, many of which fail to replicate out-of-sample. Named after Harvey, Liu, and Zhu (2016).
I
- IC (Information coefficient)
- The Spearman rank correlation between a signal's predicted cross-sectional returns and actual realised returns. A mean IC near zero means the signal has no predictive value.
- In-sample (IS)
- The historical period used to develop or fit a strategy. IS performance is always optimistic because the strategy was designed using the same data on which it is evaluated.
- Inventory risk
- The risk a market maker accumulates by holding a one-sided position while waiting for offsetting flow. Large inventory positions create mark-to-market risk.
L
- Limit-order book
- A record of outstanding buy and sell orders at specific prices. Orders in the book are passive (they provide liquidity); market orders are aggressive (they take liquidity).
- Liquidity
- The ease with which an asset can be bought or sold without moving its price. Illiquid securities have wider spreads and greater market impact.
- Look-ahead bias
- Using information in a backtest that would not have been available at the time of trading. A common source is using financial data before its actual publication date.
M
- Market impact
- The price movement caused by a trade. Large orders move prices against the trader, raising costs. Market impact grows with order size and falls with liquidity.
- Market maker
- A trader who simultaneously quotes buy and sell prices, profiting from the bid-ask spread in exchange for providing liquidity. Market makers bear adverse selection and inventory risk.
- Maximum drawdown
- The largest peak-to-trough decline in cumulative returns. A measure of tail risk; strategies with high Sharpe but very large drawdowns can be difficult to hold in practice.
- Multiple testing
- The increase in false discovery rate when testing many hypotheses on the same dataset. A strategy that was selected from 1000 tested strategies needs a much higher t-statistic than one tested once.
O
- OOS (out-of-sample)
- Data or periods not used in strategy development. OOS performance is the only honest estimate of future performance; IS performance is always optimistic.
- OOS Sharpe
- The Sharpe ratio measured on data the strategy was not designed on. The primary metric on ConvexPi; a positive OOS Sharpe is evidence (not proof) that a signal generalises.
- Overfitting
- Constructing a model that captures noise rather than signal in the training data. An overfit strategy has high IS performance and near-zero or negative OOS performance.
- Overfitting ratio
- OOS Sharpe ÷ IS Sharpe. A ratio near 1.0 means the strategy generalised well; a ratio near 0 means it overfit. Target ≥ 0.7 as a rough guide.
P
- p-value
- The probability of observing a test statistic at least as extreme as the one observed, under the null hypothesis of no effect. A low p-value is necessary but not sufficient evidence — especially after multiple tests.
R
- Rolling IC
- The information coefficient computed over a rolling window of time, showing how predictive power evolves. Decaying rolling IC suggests the signal is weakening.
S
- Signal decay
- The reduction in predictive power as the holding period lengthens. A signal that is strong at 1-day horizons may be worthless at 20-day horizons.
- Slippage
- The difference between the expected price of a trade and the price at which it executes. Caused by market impact and latency. Slippage erodes live performance relative to backtests.
- Survivorship bias
- The error of using only securities that survived to the present day in a historical analysis, excluding those that were delisted, went bankrupt, or were acquired. Leads to overstated backtests.
T
- Turnover
- The fraction of the portfolio replaced per period, typically annualised. High-turnover strategies trade more and pay more in transaction costs; they require a stronger signal to be profitable.
U
- Universe construction
- The rules that define which securities are eligible for trading at each point in time. A poorly constructed universe can introduce look-ahead or survivorship bias.
W
- Walk-forward validation
- A backtesting protocol where the model is trained only on past data and evaluated on the immediately following period, then the window rolls forward. The correct way to avoid look-ahead bias.