Overview
The QuantFlow Feature Type library provides 133 formula-based financial feature types organized into 6 dimensions. Each FeatureType is a declarative YAML template with a mathematical formula compiled to an IR DAG by FeatureDAG — same definition, batch and streaming backends.
6 Dimensions
FeatureTypes are grouped by their role:
| Dimension | Count | Focus |
|---|---|---|
| Signal | 61 | Directional predictions — OFI, momentum, flow pressure, mean reversion signals |
| Execution | 18 | Execution quality — spreads, market impact, depth elasticity, trade intensity |
| Quality | 13 | Signal quality metrics — SNR, entropy, information coefficient, Sharpe ratio |
| Regime | 27 | Market state classification — liquidity regime, toxicity, stress indices, VaR |
| Stability | 9 | Signal stability diagnostics — autocorrelation, half-life, turnover, term structure |
| Technical | 5 | Classic indicators — RSI, MACD, Bollinger Bands, MA crossover, support/resistance |
Dimension Index
Signal Features
Directional prediction signals that form the core of alpha generation. These FeatureTypes detect buy/sell pressure, momentum, and mean reversion.
Execution Features
Liquidity and cost metrics used for execution quality assessment and as execution proxies.
Quality Features
Post-hoc signal quality evaluation — SNR, entropy, IC, win rate. Used to assess signal reliability.
Regime Features
Market state descriptors that classify the environment — trend strength, liquidity conditions, toxicity, volatility regimes.
Stability Features
Time-series stability diagnostics that measure how a signal behaves over time — autocorrelation decay, half-life, turnover.
Technical Features
Classic technical analysis indicators reimplemented as formula-based FeatureTypes.
FeatureType Anatomy
Each FeatureType is defined as a YAML file with:
name: ofi
description: Order Flow Imbalance - measures net aggressive order flow
category: order_flow
dimension: signal
required_inputs: [best_bid_size, best_ask_size]
output_column: ofi
parameters:
window:
type: integer
default: 1
constraints: {min: 1, max: 100}
formula: "cumsum((diff(best_bid_size, window) - diff(best_ask_size, window)), 0)"
The formula field is compiled by FeatureDAG's IR compiler into a DAG of computation steps, then lowered to batch (Polars) or streaming (DolphinDB) backends.