A chart element showing the open, high, low, and close over a given period. It is the basic data unit of technical analysis.

An abbreviation for Open, High, Low, Close, and Volume: the standard market-data format retrieved from exchanges by BeeQuant's data-processing and extraction node.

The time span represented by each candlestick. BeeQuant supports intervals such as 1m, 5m, 15m, 30m, 1h, 4h, and 1d, configured in the data-source node.

The pair of assets being traded. BTCUSDT, for example, means trading Bitcoin against USDT. Configure pairs in the data-source node; an asset-selection strategy can use multiple pairs at once.

Buying an asset in anticipation of selling it later at a higher price. In a timing strategy, a long-entry condition triggers it; in an asset-selection strategy, use the Long subtype.

Borrowing and selling an asset in anticipation of buying it back at a lower price and retaining the difference. Shorting requires a contract market; spot markets support long positions only.

Establish a new trading position. A condition expression triggers entry in timing strategies, while the rebalancing mechanism performs it automatically in asset-selection strategies.

Exit an existing position and realize its profit or loss. Exit conditions, take-profit or stop-loss rules, and the liquidation threshold can all trigger a close.

A current market holding, including its direction, quantity, entry price, and unrealized profit or loss.

The difference between the expected and actual execution prices, caused by market movement or insufficient liquidity. It is configured as a percentage in the trader node and affects backtest realism.

A fee charged by an exchange as a percentage of executed notional value, such as 0.04%. Configure it in the trader node; it directly reduces net strategy returns.

The availability of buy and sell orders in a market. Highly liquid pairs such as BTCUSDT generally have lower slippage and faster execution.

The immediate purchase or sale of the actual asset, without leverage. Spot supports long positions but not short positions. Select Spot as the market type in BeeQuant.

A perpetual contract using USDT as margin and settlement currency. Profit and loss are denominated in USDT, simplifying management. Select USDT-margined contract in BeeQuant; this is the most common contract type.

A perpetual contract that uses the underlying cryptocurrency as margin and settlement currency. Profit and loss are settled in that asset. Select Coin-margined contract in BeeQuant.

A contract with no expiry date that uses funding payments to keep its price anchored to the spot market. All contracts supported by BeeQuant are perpetual contracts.

Borrowed capital used to increase exposure. With 10× leverage, 100 USDT can control a 1,000 USDT position. Configure leverage per asset in the trader node.

Collateral required to open a leveraged position. Higher leverage reduces required margin but increases liquidation risk.

An exchange-mandated position close when margin can no longer support the position, preventing a negative balance. Higher leverage makes liquidation more likely.

A fee exchanged between long and short holders in perpetual contracts every eight hours to keep contract prices near spot prices. Longs pay shorts when the rate is positive; the reverse occurs when it is negative.

A strategy that uses condition expressions to decide when to enter and exit. It defines four conditions—open long, close long, open short, and close short—and trades when signals fire. It suits one or a small number of instruments.

A strategy that selects the best instruments from a universe. It ranks instruments by a factor, holds those with the best ranks, and periodically rebalances the portfolio.

An asset-selection subtype that goes long only the highest-ranked instruments. It suits bullish markets and selects from highest to lowest by the sorting field.

An asset-selection subtype that shorts only the lowest-ranked instruments. It suits bearish markets and selects from lowest to highest by the sorting field.

An asset-selection subtype that goes long the highest-ranked and short the lowest-ranked instruments simultaneously. It hedges market risk and targets relative returns. A contract market is required.

A Boolean expression that opens a position in a timing strategy. It supports column references, comparisons, and logical operations only—not calls to time-series operators. Example: rsi < 30 & golden == True.

A Boolean expression that closes a position in a timing strategy. Its syntax matches an entry condition. If omitted, take-profit, stop-loss, or liquidation rules are the only mechanisms that close the position.

The factor-column name used by an asset-selection strategy to rank all instruments. Compute it in the data-processing node first. The highest- or lowest-ranked instruments become holdings.

At a configured interval, an asset-selection strategy recalculates factor ranks, buys newly selected instruments, sells instruments that fall out of the selection, and updates the portfolio.

The number of candlesticks between two asset-selection rebalances. For a 4h interval, a value of 24 means rebalancing every four days (24 × 4h).

An asset-selection allocation mode in which all selected instruments receive the same fraction of capital (1/N).

An asset-selection allocation mode that assigns unequal weights by rank. Higher-ranked instruments receive more capital, with concentration controlled by the temperature.

Controls weight concentration in top-heavy mode. Lower values concentrate more weight in the highest-ranked instruments, while higher values approach equal weighting. The default is 2.0.

A strategy style that trades in the direction of the market trend: long in an uptrend and short in a downtrend. Common indicators include moving-average crosses, ADX, and MACD.

A strategy style based on the expectation that price returns toward its mean: short above the mean and long below it. Common indicators include RSI and Bollinger Bands.

A strategy style that buys assets with strong recent gains and sells those with strong recent losses. A common factor is pct(close, 20).

A calculation expression in a data-processing node. It supports arithmetic, operator-function calls, and assignment, transforming raw OHLCV data into factors usable by a strategy.

An expression defining an AI model's training target in a data-processing node. It commonly uses shift(close, -N) to construct future returns and supports operator-function calls.

A Boolean expression that filters data. Filters in data-processing nodes support operator functions; filters in asset-selection strategy nodes do not support time-series functions and allow only comparisons and logical operations.

A Boolean expression used in strategy-type nodes for timing entries/exits or asset-selection filters. It supports column references, arithmetic, comparisons, and logical operations, but cannot call time-series functions. A True/False result triggers a signal.

A built-in BeeQuant calculation function. More than 80 operators are organized into five groups: basic, time-series, cross-sectional, technical-indicator, and candlestick-pattern operators.

An operator that calculates each element independently and needs no historical window, such as abs(), log(), if(), or clip().

An operator grouped by instrument and calculated over rolling time windows. For example, t_mean(close, 20) calculates a 20-period moving average. Configure enough history periods.

An operator grouped by timestamp and calculated across all instruments. For example, c_rank(momentum) ranks every asset's momentum at the same instant. It is designed for multi-instrument strategies.

An implementation of a classic technical-analysis indicator, calculated by instrument, such as ta_rsi(), ta_macd_hist(), ta_bbands_upper(), or ta_cross_over().

An operator that recognizes a classic candlestick pattern and returns a Boolean, such as ta_hammer(), ta_engulfing_bull(), or ta_morning_star().

A data-processing node parameter specifying the number of historical candlesticks needed for factor calculations. It must be at least as large as the greatest window argument among all operators, or calculations will produce NaN.

A column automatically supplied by the data source and directly usable in expressions: open, high, low, close, and volume.

A variable name emitted by an upstream data-processing or AI-model node and detected automatically by a strategy node. If data processing defines rsi = ta_rsi(close, 14), a strategy can reference rsi directly.

Automatically closes a trade when its profit reaches a configured percentage, locking in gains. Configure it in the position-management node (for example, 0.1 means 10%).

Automatically closes a trade when its loss reaches a configured percentage, limiting losses. Configure it in the position-management node (for example, 0.05 means 5%).

Moves the stop upward for a long position or downward for a short position as price moves favorably, protecting unrealized gains while retaining upside. Available in normal mode only.

Defines several profit targets and closes part of the position at each one, balancing realized gains with the pursuit of higher returns. Available in normal mode only.

The core risk-node parameter. When aggregate unrealized account loss reaches this percentage, all positions are forcibly closed to protect capital. A value of 0.2 triggers at a 20% unrealized loss.

The number of minutes for which trading pauses after a liquidation event, allowing the market to stabilize and helping prevent consecutive losses. Configure it in the risk node.

The method for controlling how much capital each trade uses. BeeQuant supports normal mode with fixed-percentage entries and grid mode with layered entries.

The share of capital used for each entry in normal mode. A value of 0.5 uses 50% of total capital per entry.

The largest allowed allocation to one instrument in an asset-selection strategy, preventing excessive concentration. A value of 0.2 caps one asset at 20% of total capital.

The greatest decline from a historical peak to a subsequent trough. It measures the largest loss a strategy has experienced and is one of the most important risk metrics.

An advanced position-management mode that adds one layer whenever price falls by a configured amount, reducing average entry cost. It includes range and trend variants and applies only to timing strategies.

Total profit or loss over the strategy's run, expressed as a percentage: (ending equity - starting equity) / starting equity × 100%.

Return converted to an annual rate for comparison across different run lengths: (1 + total return)^(365 / days run) - 1.

Excess return earned per unit of risk: (strategy return - risk-free rate) / return volatility. Above 1 is good and above 2 is excellent. It is a common sorting factor in asset-selection strategies.

Similar to the Sharpe Ratio, but its denominator includes only downside volatility. It focuses on loss risk and measures risk-adjusted return more precisely when upside volatility should not be penalized.

Annualized return divided by maximum drawdown. A larger value indicates that the strategy earned more return while experiencing less drawdown.

The percentage of all trades that are profitable. Win rate alone has limited meaning and should be evaluated together with the profit/loss ratio.

Average profit divided by average loss. A value above 1 means winning trades earn more than losing trades lose. High win rate plus a high profit/loss ratio is the ideal combination.

The portion of strategy return above a market benchmark such as holding BTC, representing active-management skill. Alpha above 0 means the strategy outperformed the market.

The relationship between strategy and market returns. Beta = 1 means they move together; Beta = 0 means no market relationship, which is the target of a market-neutral strategy.

The virtual capital at the start of a backtest, configured in the trader node. It does not change strategy logic, but it affects absolute profit/loss and position quantities.

A gradient-boosted decision-tree algorithm developed by Microsoft. It trains quickly, uses memory efficiently, and suits tabular factor data. BeeQuant's built-in node supports ranking, binary-classification, and regression tasks.

Long Short-Term Memory, a recurrent neural network that uses gates to learn long-range dependencies in time series and capture temporal price patterns.

A deep-learning architecture based on self-attention. It supports parallel computation and captures global dependencies; unlike the sequential processing of an LSTM, it attends to every position in a sequence at once.

A large language model such as GPT or Claude. BeeQuant's LLM node processes unstructured data such as news, market sentiment, and announcements, complementing traditional quantitative factors.

Predicts one of two outcomes, such as up or down, and emits signed confidence from -1 to 1. Positive values are bullish, negative values bearish, and larger absolute values indicate greater confidence.

Predicts a continuous value such as future return and emits the predicted numerical value. It suits direct prediction of the magnitude of price movement.

Predicts relative scores among instruments for asset-selection strategies. The strategy ranks predictions and selects the best instruments.

The factor variables supplied to an AI model from data-processing results, such as return, volatility, RSI, and MACD. Feature quality directly affects model performance.

The AI model's training target, defined by Label annotation in the data-processing node. A common label is shift(close, -24) > close: whether price rises over the next 24 hours.

The amount of historical data, in candlesticks, used to train a model. A window that is too short reveals too little structure; one that is too long includes stale information. Typical values are 720–1,440.

The model-retraining interval in candlesticks. At the end of each prediction window, the model retrains on the newest data to adapt to market changes.

The name of the AI model's output column, such as pred. It is added to the data for downstream strategy nodes to reference.

A model performs well on training data but poorly on unseen data because it learned historical noise instead of a real pattern. Remedies include a larger training window, lower model complexity, and stronger regularization.

A simplified AI-model configuration in which the system automatically chooses suitable hyperparameters from the data, avoiding manual tuning and helping new users get started.

BeeQuant's visual strategy builder. Drag, drop, and connect seven core node types to create a complete quantitative strategy without writing code.

A basic functional unit in a workflow. BeeQuant has seven node types: data source → data processing → [AI model] → strategy type → position management → risk control → trader.

A required parameter card for each data branch. It does not retrieve data or emit a DataFrame; it defines the retrieval parameters required by the downstream data-processing and extraction node: date range, exchange, market type, pairs, and candlestick interval.

The feature-engineering center and the only node that actually retrieves base OHLCV fields globally. It reads upstream source parameters, retrieves raw OHLCV, and transforms it into strategy factors. It includes factor configuration, AI-training Label annotations, and filters. Raw OHLCV is not emitted downstream unless passed through explicitly, for example close = close.

An optional node that receives factors from data processing for machine-learning training and prediction. It supports LightGBM, LSTM, Transformer, and LLM models.

The core trading-signal generator. It decides when to trade in a timing strategy or what to trade in an asset-selection strategy, converting upstream data into signals.

Controls capital allocation and entry size. Normal mode supports fixed positions, take profit, stop loss, and trailing stops; grid mode supports layered entries and range or trend variants.

The final safety layer. It has a liquidation threshold that closes all positions at a configured aggregate unrealized loss and a cooldown that pauses trading afterward.

The strategy's execution endpoint, converting signals into orders. It supports backtest, simulation, and live modes.

A simulation of strategy trading on historical data. Bar-level precision uses candlestick OHLC prices; tick-level precision uses a synthetic price path interpolated from minute OHLC. Neither uses actual individual trades or order-book events.

A backtest using the four OHLC prices in each candlestick to simulate execution. It is fast but less precise and is suitable for initial strategy validation.

A backtest using a synthetic price path interpolated from minute OHLC. It has finer granularity but is not actual tick data and contains no real trades, quotes, or order-book events. It supports detailed tuning but is not equivalent to live execution.

Testing a strategy with virtual funds in real market conditions. It receives live data and simulates order placement to validate live-market behavior without financial risk.

Connecting an exchange API and executing the strategy with real funds. It requires API credentials. Validate a strategy with backtests and paper trading before enabling live trading.

An exchange-issued API key and secret used for programmatic order access. Bind them securely in the platform before live trading.