Trading Dev AcademyFree quant education

Quantitative finance & development glossary

182 definitions with examples. Study the related calculations in the curriculum or explore Python tools.

Ablation

A comparison that removes or changes one component while holding the rest fixed.

Example: Remove candle features to test their contribution under the same costs and risk limits.

Adverse selection

A tendency to trade when the counterparty’s information or timing makes the fill unfavorable.

Example: A passive buy can fill just before the market moves lower.

Antiderivative

A function whose derivative equals the given function.

Example: x² is an antiderivative of 2x; adding a constant gives another.

ATR

Average true range: a smoothed measure of bar range including gaps from the prior close.

Example: Divide a price distance by prior ATR to express it in comparable range units.

Autocorrelation

Correlation of a series with a lagged version of itself.

Example: Lag-one autocorrelation compares observations one step apart.

Availability timestamp

The time at which a value could first have been used by the modeled decision process.

Example: A report about the previous quarter becomes usable after its release and processing delay.

Backpressure

A mechanism that limits incoming work when a consumer cannot keep up.

Example: A bounded event queue forces a defined overload policy.

Backtest

A historical simulation of a fully specified decision and execution process.

Example: A signal at today’s close must use a subsequently eligible execution price.

Baseline

A simple, predeclared comparator for evaluating incremental value.

Example: Compare a neural forecast with a trailing mean on the same periods.

Basis point

One hundredth of a percentage point; 0.0001 as a decimal.

Example: 10 basis points is 0.1%, or 0.001.

Bayesian updating

Combining a prior belief with the likelihood of observed evidence.

Example: Evidence that is more likely under one hypothesis shifts its posterior weight upward.

Bellman equation

A recursive relation connecting value now to immediate reward and future value.

Example: The continuation value must use the next-state distribution and chosen policy or optimization.

Bootstrap

A resampling procedure used to estimate uncertainty under a chosen dependence model.

Example: Block resampling preserves some local time dependence; it does not create an untouched test set.

BOS

Break of structure: a specified break in the established swing direction.

Example: In the stated up-state convention, a close above the confirmed high is bullish BOS.

Breakout

A specified observation beyond a boundary fixed before the observation.

Example: A close above a prior high plus a one-unit buffer is a reproducible trigger.

Broadcasting

Rules that align array dimensions for elementwise operations without explicit repetition.

Example: A column vector times a row vector forms a matrix, which can be an accidental shape bug.

Brownian motion

A continuous-time process with independent Gaussian increments and variance proportional to elapsed time.

Example: A time increment .04 has Brownian standard deviation .2.

Calibration

Agreement between predicted probabilities and observed frequencies across comparable cases.

Example: Among many events assigned probability .7, roughly 70% should occur if predictions are well calibrated.

Candlestick

A display of a bar’s open, high, low and close.

Example: The body joins open and close; wicks extend to the extremes.

Capacity

The scale at which a strategy remains viable under liquidity, impact and operating constraints.

Example: An edge on a small order may disappear when participation increases.

Cash flow

An amount paid or received at a specified time.

Example: An investor deposit increases cash but is not trading profit.

Chain rule

Multiply the sensitivities along a composition of functions.

Example: For y=(2x+1)², the derivative is 4(2x+1).

CHOCH

Change of character: a specified break against the established structure.

Example: In the stated up-state convention, a close below the protected low is bearish CHOCH.

CNN

A convolutional neural network that applies shared filters across positions.

Example: A one-dimensional CNN can process a sequence of candle features.

Cointegration

A stationary linear combination of individually nonstationary series under a specified model.

Example: High correlation alone does not establish cointegration.

Conditional probability

An event probability given specified information.

Example: The chance of a loss conditional on a volatility regime may differ from the unconditional chance.

Confidence interval

An interval from a procedure with a stated repeated-sampling coverage under its assumptions.

Example: A 95% procedure aims to cover the fixed parameter in 95% of repeated samples.

Constant rate

A change per unit time that remains fixed over the modeled interval.

Example: Adding 2 units each hour to an initial 5 gives 11 units after 3 hours.

Constraint

A condition limiting allowed choices.

Example: Portfolio weights may be required to sum to one.

Continuity

Agreement between a function’s value and its limit at a point.

Example: A removable hole can have a limit without a defined function value.

Contrastive learning

Learning representations by drawing selected examples together and separating others.

Example: The choice of positive and negative pairs determines what similarity means.

Correlation

A standardized measure of linear co-movement.

Example: Assets can be highly correlated without forming a stationary tradable spread.

Covariance

Expected joint movement of two variables relative to their means.

Example: Two returns moving together contribute positive covariance under the model.

Cross-validation

Repeated separation of fitting and evaluation samples under a defined sampling scheme.

Example: Trading labels require chronological and overlap-aware splits.

Curvature

How a curve bends; the second derivative describes changes in slope.

Example: For x², the second derivative is 2 everywhere.

Data leakage

Information crossing a boundary that should separate fitting, selection or evaluation.

Example: Scaling all observations before splitting lets test data influence training.

Day-count convention

A rule for converting dates into accrual fractions for a financial contract.

Example: Actual/360 and Actual/365 can produce different interest accruals.

Delta

A derivative price’s local sensitivity to its underlying price, holding other inputs fixed.

Example: Delta .6 approximates a .60 price change for a small one-unit underlying move.

Denominator

The expression below a fraction bar; it cannot be zero in ordinary division.

Example: In three quarters, 4 specifies the number of equal parts.

Dependency lockfile

A record of resolved software versions used to reproduce an environment.

Example: A library update should trigger numerical regression checks before adoption.

Derivative

The local rate of output change with respect to an input, when the limit exists.

Example: For f(x)=x² at x=3, the derivative is 6 output units per input unit.

Derivative contract

A contract whose cash flows depend on a specified underlying asset, rate or event.

Example: An option and a futures contract can reference the same underlying but create different obligations.

Differential equation

An equation relating a function to its derivatives.

Example: A constant growth rate and a starting balance determine a linear path.

Diffusion

A continuous-time stochastic model driven by a Brownian component.

Example: Drift controls modeled local average change; diffusion scale controls local uncertainty.

Discounting

Converting future cash flows into present values with a specified rate model.

Example: At 10% for one period, 110 due later has present value 100.

Distribution

A model of possible outcomes and their probabilities.

Example: A fair die has six possible values with equal probability.

Divergence

A specified disagreement between confirmed price and indicator swings.

Example: A lower price low paired with a higher indicator low is one bullish convention.

Diversification

Combining exposures whose risks are not perfectly aligned.

Example: Ten highly similar strategies may provide little additional diversification.

Domain

The inputs for which an expression is defined.

Example: The real-valued function log(x) requires x>0.

Dot product

The sum of pairwise products of corresponding vector entries.

Example: [2,3] dotted with [4,5] gives 23.

Drawdown

Decline from a previous peak in the chosen wealth series.

Example: Falling from 100 to 80 is a 20% drawdown.

Drift

The modeled instantaneous conditional mean rate of change.

Example: A positive drift does not ensure every realized price increment is positive.

DTW

Dynamic time warping: an alignment minimizing accumulated local mismatch subject to allowed moves.

Example: A band limits how far observation positions can shift during matching.

Duration

A local measure of bond-price sensitivity to yield, under a stated convention.

Example: Modified duration 5 implies about a 5% price decline for a one-percentage-point yield rise, locally.

Eigenvalue

A scaling factor for a direction preserved by a matrix transformation.

Example: An eigenvector changes length, but not its direction, under that transformation.

Elliott Wave

A framework for classifying price swings into specified impulse and corrective patterns.

Example: A finished five-wave count cannot be assumed known during the third wave.

Embargo

Excluding a specified interval near an evaluation boundary to limit information overlap.

Example: A gap of several observations is justified by the actual information timeline, not by convention alone.

Embedding

A learned numerical representation of an input.

Example: Nearby coordinates indicate similarity under the learning objective, not guaranteed similar returns.

Equation

A statement that two expressions have equal values.

Example: 2x+1=7 gives x=3.

Euler method

A numerical update using the current rate over a finite time step.

Example: Starting at 5 with rate 2 per hour, a half-hour Euler step reaches 6.

Event-driven backtest

A simulation that updates state in an explicit sequence of market, order and fill events.

Example: A signal submits an order; a later fill changes cash and inventory.

EWMA

An exponentially weighted moving average giving declining weight to older observations.

Example: A larger new-observation weight reacts faster and smooths less.

Excess return

Return less a stated reference return over the same interval.

Example: A 1% portfolio return minus a 0.2% benchmark return gives 0.8 percentage points.

Expected shortfall

Average loss in a specified worst-probability tail, with a stated convention for boundary mass.

Example: A 95% measure concerns the worst 5% of losses under the model.

Expected value

The probability-weighted average of possible outcomes.

Example: An equal chance of +10 or −4 gives expected payoff 3 before costs.

Exponential

A function whose variable appears in an exponent.

Example: exp(0)=1; multiplying exp(a) by exp(b) gives exp(a+b).

Feature

An input supplied to a statistical or machine-learning model.

Example: A completed candle’s signed body fraction is one numerical feature.

Feature pipeline

A reproducible sequence of transformations from raw observations to model inputs.

Example: Fit missing-value handling and scaling only on the training window.

Function

A rule assigning one output to each allowed input.

Example: For f(x)=x², input 3 produces output 9.

Funding rate

A periodic transfer rate under a perpetual-derivative contract’s venue rules.

Example: Funding cash flow depends on side, notional, rate and settlement time.

Futures

Contracts creating specified settlement obligations on a stated underlying and expiry.

Example: The quoted price change must be multiplied by the contract multiplier and quantity.

FVG

Fair value gap: here, a three-bar non-overlap convention between the first and third bar.

Example: First high 102 and third low 104 define a bullish zone [102,104].

Gamma

The sensitivity of delta to the underlying price.

Example: Gamma describes how the local hedge ratio changes as spot moves.

Gann angle

A chart line defined by specified price and time units.

Example: A 1×1 line is one chosen price unit per time unit; its screen angle changes with scaling.

GARCH

A model updating conditional variance from earlier shocks and variance.

Example: A large squared residual can raise the next variance forecast.

GASF

Gramian angular summation field: a matrix encoding pairwise angular relationships of scaled values.

Example: It rearranges information from a sequence; it does not add new market observations.

Gradient

The vector of first partial derivatives.

Example: For f(x,y)=x+2y, the gradient is [1,2].

Gradient descent

An iterative update moving parameters opposite an estimated objective gradient.

Example: The learning rate controls the step size; a large step can overshoot.

Gross exposure

Sum of absolute position notionals divided by capital under a stated valuation convention.

Example: A 60% long and 40% short book has 100% gross exposure.

Harmonic pattern

An ordered pivot sequence constrained by specified leg ratios and tolerances.

Example: One matching AB-to-XA ratio does not complete a Gartley detector.

Hedge ratio

The relative quantities chosen to offset a specified exposure.

Example: A regression coefficient needs conversion into quantities and contract units.

Heikin-Ashi

An OHLC transformation that uses price averages and a recursive opening value.

Example: Its synthetic close need not be an executable market price.

Hessian

The matrix of second partial derivatives.

Example: It describes local curvature in several directions, rather than just the slope.

Heteroskedasticity

Variation in conditional error variance across observations or time.

Example: Return uncertainty can rise sharply during a volatility episode.

Hidden state

An unobserved model state inferred from observations.

Example: A hidden regime may be inferred from observed returns rather than directly recorded.

Hyperparameter

A choice controlling a model or training procedure rather than an ordinary fitted coefficient.

Example: Tree depth and learning rate are hyperparameters.

Idempotency

A property that repeating the same operation has the same effect as performing it once.

Example: Replaying a fill identifier should not add the quantity twice.

Implementation shortfall

Difference between an actual execution outcome and a stated decision-price benchmark, including the chosen cost components.

Example: Delay, fills and unexecuted quantity can each contribute.

Information ratio

Mean active return divided by its standard deviation under a stated frequency and annualization convention.

Example: Use active returns against the same benchmark, not total portfolio volatility.

Initial condition

The value that specifies where a dynamic model starts.

Example: y(0)=5 fixes the starting amount at time zero.

Integral

An accumulation of contributions over an interval or region.

Example: A flow of 3 litres per minute for 4 minutes adds 12 litres.

Integration by parts

An integration identity derived from the product rule.

Example: Choose one factor to differentiate and another to integrate, then check the boundary term.

Itô correction

The extra second-order term caused by Brownian quadratic variation.

Example: An ordinary chain rule omits this term when transforming a diffusion.

Jacobian

The matrix of first derivatives of several outputs with respect to several inputs.

Example: Each row tracks how one output responds to the input coordinates.

Kelly

Position sizing that maximizes expected logarithmic wealth under a specified return model.

Example: Estimation error can make a model’s full Kelly allocation excessively aggressive.

Kill switch

A controlled mechanism for stopping new activity and handling outstanding risk according to policy.

Example: Stopping order submission alone does not cancel resting orders.

Lag

A previous observation measured in the series’ time steps.

Example: At daily bar 10, lag one is bar 9.

Latency

Elapsed time between specified events in a data or execution process.

Example: Measure decision-to-acknowledgment separately from decision-to-fill latency.

Leverage

Exposure relative to the capital supporting it, under a stated definition.

Example: Exposure 300 against equity 100 is three times equity.

Limit

The value approached as the input approaches a point.

Example: As x approaches 2, x+2 approaches 4.

Liquidity

The ability to trade a quantity promptly at a stated cost.

Example: A narrow spread for one share does not imply deep liquidity for a large order.

Log return

The natural logarithm of the ratio of ending to starting positive wealth.

Example: A wealth ratio of 1.02 gives log(1.02), approximately 0.019803.

Logarithm

The exponent needed to produce a positive value from a specified base.

Example: Because 2³=8, the base-two logarithm of 8 is 3.

Look-ahead bias

Using information before it was actually available at the simulated decision time.

Example: Fitting a scaler on the complete dataset leaks future information.

Loss function

The objective measuring model error or a training preference.

Example: Squared prediction error penalizes larger misses more heavily.

Margin

Collateral required under a trading or contract arrangement, not a general measure of maximum loss.

Example: Futures initial margin supports an exposure that can be much larger than the collateral.

Market impact

A trade’s effect on available prices and subsequent execution costs.

Example: A larger order may consume several levels of quoted liquidity.

Markov

A model where the specified current state contains the information needed for the next-state distribution.

Example: A two-state regime model uses transition probabilities out of the current state.

Matrix

A rectangular array representing relationships or transformations.

Example: A two-asset covariance matrix has two rows and two columns.

Mean

The sum of values divided by their count.

Example: The mean of 2, 5 and 8 is 5.

Mean reversion

A modeled tendency for a deviation to move toward a reference level.

Example: Reversion speed and the stability of the reference are separate assumptions.

Model drift

A change in the relationship between model inputs and outcomes or in their distributions.

Example: A model calibrated in one liquidity regime can deteriorate in another.

Momentum

Past directional movement or relative strength measured over a specified lookback.

Example: A trailing return is an input hypothesis, not a future return.

Monte Carlo

Numerical estimation by simulating draws from a specified model.

Example: Average many simulated payoffs to estimate the model’s expected payoff.

Multiple testing

Evaluating many hypotheses or variants, increasing opportunities for apparently favorable chance results.

Example: The best of 1,000 noisy strategies needs a selection-aware evaluation.

Net exposure

Signed sum of position notionals divided by capital under a stated valuation convention.

Example: A 60% long and 40% short book has 20% net exposure.

Notional

A reference measure of the economic exposure controlled by a position.

Example: Contract count × multiplier × reference price is one common futures convention.

Numerator

The expression above a fraction bar.

Example: In three quarters, 3 counts the selected parts.

OHLCV

Open, high, low, close and volume recorded for a bar.

Example: A daily bar does not reveal the exact sequence of all intraday trades.

Optimization

Choosing a feasible input to minimize or maximize a stated objective.

Example: Minimize modeled risk subject to position limits; the result depends on the model.

Option

A contract granting a specified exercise right under agreed terms.

Example: A call grants a right to buy at its strike; its premium is not its notional exposure.

Order book

A venue-specific collection of resting orders or aggregated price levels.

Example: Best bid and ask show only part of the liquidity available across venues.

Oscillator

An indicator describing recent movement on a chosen scale, often bounded.

Example: Stochastic %K measures a close’s position inside a recent range.

Out-of-sample

Data withheld from the relevant fitting and selection procedure.

Example: Repeatedly tuning to a test set makes it part of model selection.

Overfitting

Adapting a model or rule to sample-specific noise that does not generalize.

Example: Selecting the best of thousands of noisy backtests can produce an apparent winner.

Paper trading

Simulated order operation using a mock or broker-provided paper environment.

Example: A paper fill can be easier to obtain than a real fill at the same quote.

Partial derivative

Sensitivity to one independent input while holding the others fixed.

Example: For f(x,y)=x+2y, changing x by 1 with y fixed changes f by 1.

Partial fill

Execution of only part of an order quantity.

Example: An order for 100 shares can receive fills of 30 and 20 while 50 remain.

PDE

A partial differential equation involving derivatives in several independent variables.

Example: An option-pricing model may involve both time and underlying price.

Pivot

A turning-point candidate under a specified detection and confirmation rule.

Example: A pivot needing two later bars becomes available two bars after the extremum.

Point-in-time data

Data reconstructed according to what was actually available at a historical decision time.

Example: A revised earnings value must not replace the original value in an earlier backtest.

Policy

A rule or distribution for choosing an action from the available state.

Example: A policy may choose how much inventory to trade at the next decision time.

Portfolio

A collection of positions considered jointly.

Example: Two positions need a joint risk assessment, not just separate volatility estimates.

Power

Repeated multiplication for a positive integer exponent; other exponents require their domain conventions.

Example: 2³ means 2×2×2=8.

Power rule

Differentiate a power by multiplying by its exponent and reducing the exponent by one, on its differentiable domain.

Example: The derivative of x³ is 3x².

Probability

A number from zero to one assigned to an event under a specified model.

Example: A fair die assigns probability one sixth to any one face.

Property-based test

A test that checks general invariants over many generated inputs.

Example: Doubling fixed quantities should double a linear cash-flow calculation.

Purging

Removing training observations whose label-information intervals overlap a held-out evaluation period.

Example: A five-day target may cross a split even when its feature row precedes it.

Quantile

A cutoff at a specified cumulative probability.

Example: A loss quantile is not a cap on losses beyond that cutoff.

Random seed

An initialization value for a pseudorandom generator.

Example: A fixed seed repeats a simulation but does not establish that its assumptions are correct.

Rate limit

A service constraint on request frequency or volume over a specified interval.

Example: Retrying too quickly after throttling can worsen a data outage.

Ratio

One quantity divided by another; the units determine its meaning.

Example: 6 dollars shared across 3 items gives 2 dollars per item.

Reconciliation

Checking that independent records explain the same balances and changes.

Example: Starting equity plus investment profit plus external flows should match ending equity.

Regression

A model relating an outcome to one or more explanatory inputs.

Example: A linear regression estimates a slope and intercept; it does not establish causation.

Regularization

A penalty or restriction that discourages overly flexible fitted models.

Example: Ridge regression shrinks coefficients toward zero using a chosen penalty.

Reinforcement learning

Learning a policy from sequential actions and rewards in an environment.

Example: Execution research can model remaining inventory as part of the state.

Renko

A price-movement chart constructed from bricks under a chosen size and reversal convention.

Example: Multiple bricks may be emitted by one source-price jump at one timestamp.

Reproducibility

Ability to reconstruct a result from recorded inputs, code, configuration and environment.

Example: Save data hashes, versions, split boundaries and random seeds with a run.

Residual

Observed value minus the model’s prediction.

Example: An observation of 12 and forecast of 10 give residual 2.

Resistance

A candidate price area where an advance may meet a reaction.

Example: A level can fail or change its role after a confirmed break.

Retracement

The portion of a completed move subsequently given back.

Example: A move from 100 to 120 followed by 110 retraces half the upswing.

Reward

The immediate outcome signal specified for an RL agent.

Example: A reward can include marked profit minus trading costs and an inventory penalty.

Riemann sum

An approximation to accumulated area using rectangles.

Example: Four rectangles of width .5 sample a curve over an interval of length 2.

Risk contribution

An allocation of portfolio risk to holdings under a specified decomposition.

Example: A small capital weight can dominate risk when its volatility and correlations are large.

Sample

The observations available for analysis.

Example: Twenty recorded trades are a sample, not every possible future trade.

Secant

A line through two points on a curve.

Example: For x² between x=2 and x=3, the slope is 5.

Shapelet

A short sequence used as a candidate local pattern or matching feature.

Example: Compare a fixed training template with windows in the available history.

Sharpe ratio

Mean excess return divided by its standard deviation using consistent horizons.

Example: Annualization requires additional assumptions about return dependence.

Simple return

Price change plus distributions, divided by initial price, for a specified holding interval.

Example: Buying at 100 and selling at 103 without distributions gives a 3% holding-period return.

Slippage

The difference between a chosen reference price and actual execution price.

Example: Paying 100.03 against a buy reference of 100 adds .03 per unit.

Slope

Output change per unit of input change.

Example: A rise of 6 over an input change of 2 gives slope 3.

Spread

A difference between two prices or rates, whose meaning depends on context.

Example: Bid–ask spread measures quoted transaction friction; a pair spread compares two assets.

Square root

The nonnegative number whose square equals the given nonnegative value.

Example: The square root of 9 is 3; solving x²=9 separately allows both 3 and −3.

Standard deviation

The square root of variance, expressed in the original measurement units.

Example: Variance 9 dollars squared corresponds to standard deviation 3 dollars.

Standard error

Estimated variability of an estimator across repeated samples.

Example: For independent observations, a mean’s standard error typically shrinks with the square root of sample size.

Stationarity

Stability of specified distributional properties over time.

Example: Weak stationarity requires a constant mean and lag-dependent covariance, not constant prices.

Substitution

A change of variable that rewrites a calculation in simpler terms.

Example: Set u=2x+1 and account for du=2 dx in an integral.

Support

A candidate price area where a decline may meet a reaction.

Example: A support touch is an observation; it does not guarantee a bounce.

Survivorship bias

Distortion from selecting only entities that remain in the dataset at the end.

Example: Backtesting current index members excludes some companies that failed or were removed.

Tangent

A line matching a differentiable curve’s local slope at a point.

Example: The tangent to x² at x=3 has slope 6.

Target

The outcome a model is trained to predict, with a defined availability time.

Example: A five-day forward return is known only after those five days have elapsed.

Taylor approximation

A local polynomial built from a function’s value and derivatives.

Example: Near zero, exp(x) is approximately 1+x; accuracy depends on the distance and omitted terms.

Tick

The smallest permitted quoted price increment for a stated instrument.

Example: A .25-point tick with a 50-per-point multiplier changes contract value by 12.50.

Tracking error

Standard deviation of portfolio returns minus benchmark returns at a specified horizon.

Example: Two portfolios with equal volatility can have very different tracking errors.

Transformer

A neural architecture using attention to combine information across input positions.

Example: A causal attention mask prevents attending to later tokens in the current sequence.

Unit test

A small automated check of a component against an independently specified expectation.

Example: A hand-calculated fee fixture checks the sign and unit conversion.

Variable

A named quantity whose value may change.

Example: In y=2x+1, changing x changes y.

Variance

Average squared deviation from the mean; the sample estimator commonly uses n−1.

Example: Squaring deviations prevents positive and negative deviations from canceling.

Vector

An ordered list of quantities.

Example: [.25,.75] can represent weights in two assets.

Vectorization

Expressing operations on arrays so numerical kernels process many elements together.

Example: Multiply a vector of held exposures by a matching vector of returns.

Volatility

The scale of return or price fluctuations under a stated horizon and measure.

Example: Daily and annual volatility use different time conventions.

Volume

Quantity traded under a specified instrument and venue convention.

Example: Total volume includes both sides of every executed trade.