Overview
QuantFlow Market Data Model is QuantFlow's own normalised market-data schema. In code, configuration and table names it is abbreviated cdm (for example cdm_trades). It is unrelated to the FINOS Common Domain Model (CDM), which describes financial instruments, trades and lifecycle events, and QuantFlow does not use it.
The QuantFlow Feature Library provides 133 ready-to-use formula-based FeatureTypes — reusable building blocks compiled from declarative YAML by FeatureDAG. Each FeatureType is a feature blueprint with configurable parameters, normalizations, and horizons.
No code required. Same YAML compiles to batch and streaming execution.
Feature Types
133 FeatureTypes organized into 6 dimensions:
| Dimension | Count | Focus |
|---|---|---|
| Signal | 61 | Directional predictions — OFI, momentum, flow pressure, mean reversion |
| Execution | 18 | Execution quality — spreads, market impact, depth elasticity |
| Quality | 13 | Signal quality — SNR, entropy, IC, Sharpe ratio, win rate |
| Regime | 27 | Market state — liquidity regime, toxicity, stress, VaR, trend |
| Stability | 9 | Signal diagnostics — autocorrelation, half-life, turnover |
| Technical | 5 | Classic indicators — RSI, MACD, Bollinger, MA crossover |
Each FeatureType is a declarative YAML template with a mathematical formula field compiled to an IR DAG by FeatureDAG's AST Compiler.
Define Custom FeatureType
Custom FeatureTypes follow the same YAML schema as built-in types. Place YAML in .definitions/features/. The formula field is compiled by FeatureDAG's AST Compiler into an IR DAG:
name: my_custom_feature
description: Custom signal description
category: order_flow
version: v0.1.0
dimension: signal
status: stable
required_inputs:
- best_bid_size
- best_ask_size
output_column: my_custom_feature
output_description: Description of the computed output
tags:
- custom
- flow
parameters:
window:
type: integer
description: Rolling window size
required: false
default: 10
constraints:
min: 1
max: 100
formula: rolling_mean(best_bid_size - best_ask_size, window)
FeatureDAG parses the formula, resolves required_inputs to Market Data Model columns, binds parameters to typed defaults, and lowers the resulting IR DAG to Polars (batch) or DolphinDB (streaming) — both backends run the same YAML — no duplicate implementations.