Factor Engine

Quantitative factor mining and evaluation for systematic strategies

The factor engine computes quantitative signals (factors) from market data and evaluates their predictive power. Factors provide systematic, backtestable features for strategy development.

The factor engine is opt-in. Set FACTOR_ENGINE_ENABLED=true in your environment to activate it. It increases CPU and memory usage due to continuous computation across all tracked symbols.

What are factors?

A factor is a numeric value derived from market data that aims to predict future returns. Factors capture properties like momentum, mean reversion, volatility clustering, and volume patterns.

Unlike ad-hoc technical indicators, factors are:

• Normalized — comparable across different symbols and time periods
• Evaluated — effectiveness is measured through statistical metrics
• Composable — combined into multi-factor strategies

Built-in factors

The engine includes 64 expressions organized across six categories:

Trend

Factor	Expression	Description
trend_ema_cross	ema(20) / ema(50) - 1	EMA crossover momentum
trend_strength	adx(14)	Trend strength via ADX
trend_direction	sign(close - ema(200))	Binary trend direction
price_momentum	roc(close, 20)	20-period rate of change
trend_persistence	hurst(close, 100)	Hurst exponent for trend persistence

Momentum

Factor	Expression	Description
rsi_factor	normalize(rsi(14))	Normalized RSI
macd_momentum	macd_hist(12,26,9)	MACD histogram
stoch_momentum	stoch_k(14,3)	Stochastic oscillator
roc_5	roc(close, 5)	Short-term rate of change
williams_r	williams_r(14)	Williams %R

Volatility

Factor	Expression	Description
realized_vol	std(returns, 20)	20-period realized volatility
atr_factor	atr(14) / close	Normalized ATR
bb_width	bbwidth(20, 2)	Bollinger Band width
vol_regime	vol(20) / vol(60)	Short vs long-term volatility ratio
garman_klass	gk_vol(20)	Garman-Klass volatility estimator

Volume

Factor	Expression	Description
volume_momentum	volume / sma(volume, 20)	Volume relative to average
obv_slope	slope(obv, 20)	On-balance volume trend
vwap_deviation	(close - vwap) / atr(14)	Deviation from VWAP
volume_price_trend	vpt(20)	Volume-price trend
mfi_factor	mfi(14)	Money flow index

Statistical

Factor	Expression	Description
skewness	skew(returns, 60)	Return distribution skewness
kurtosis	kurt(returns, 60)	Return distribution kurtosis
z_score	zscore(close, 50)	Z-score vs rolling mean
autocorrelation	autocorr(returns, 1, 20)	Lag-1 return autocorrelation
mean_reversion	-1 * zscore(close, 20)	Mean reversion score

Composite

Factor	Expression	Description
quality	sharpe(20) * (1 - max_dd(20))	Risk-adjusted quality
momentum_quality	roc(20) * (1 / vol(20))	Volatility-adjusted momentum
trend_vol	adx(14) * atr_factor	Trend strength × volatility
composite_score	rank_avg(momentum, quality, trend)	Multi-factor composite

Custom factor creation

Define custom factors using the expression engine:

curl -X POST https://api.production.hyperoru.com/api/factors \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "my_momentum",
    "expression": "roc(close, 10) * (volume / sma(volume, 20))",
    "description": "Volume-weighted 10-period momentum"
  }'

Expression engine syntax

The expression engine supports arithmetic operations, built-in functions, and nesting:

Syntax	Example	Description
Arithmetic	a + b, a * b, a / b	Standard math operators
Functions	ema(close, 20)	Technical indicator functions
Nesting	ema(rsi(14), 5)	Compose functions
Constants	100, 2.5	Literal numbers
References	close, volume, high, low	Price/volume fields

Factor effectiveness metrics

Each factor is continuously evaluated for predictive power:

Metric	Description	Good Value
IC (Information Coefficient)	Correlation between factor values and forward returns	> 0.03
ICIR (IC Information Ratio)	Mean IC / Std(IC), measures consistency	> 0.5
Win Rate	Fraction of periods where factor direction matches return direction	> 55%
Turnover	How frequently factor rankings change	Depends on strategy

curl "https://api.production.hyperoru.com/api/factors/{factor_id}/effectiveness?lookback=90d" \
  -H "Authorization: Bearer $TOKEN"

Factor computation pipeline

The pipeline runs on a configurable schedule (typically every 1–5 minutes) and caches results in the FactorCache for instant access by strategies and the API.