Skip to content

OHLCV Features

Overview

The OHLCV feature block derives return, volatility, trend, and price-dynamics features from daily underlying price data. These features operate only on canonical underlying daily data and do not depend on option-level information, implied volatility surfaces, or external joins. All computations are deterministic and strictly causal. The OHLCV block is typically the foundational feature set used in downstream modeling.

Required Input Columns

The following canonical columns must be present in the input DataFrame:

Column Name Description
px_open Opening price of the underlying
px_high Daily high price
px_low Daily low price
px_close Closing price
volume Daily traded volume

All prices are assumed to be positive and adjusted consistently.

Produced Feature Columns

Returns & Gaps

Feature Description
ret_1d Close-to-close log return
overnight_ret Simple overnight return (open vs previous close)
gap_size Absolute overnight gap normalized by prior close

Realized Volatility

All volatility measures are annualized assuming 252 trading days.

Feature Description
realized_vol_21d Rolling close-to-close realized volatility
parkinson_vol_21d Parkinson high–low volatility estimator
garman_klass_vol_21d Garman–Klass OHLC volatility estimator
yang_zhang_vol_21d Yang–Zhang volatility estimator

Volatility of Volatility

Feature Description
vol_of_vol_21d Rolling volatility of realized volatility

Trend & Momentum

Feature Description
ema_10 Exponential moving average (fast)
ema_21 Exponential moving average (slow)
bollinger_pct_b Normalized position within Bollinger Bands
trend_up Directional trend indicator (+1, 0, −1)

How to Call This Feature Block

Use the public feature API when working with canonical underlying data:

from ocf.features.api import ohlcv_features

features = ohlcv_features(underlying_daily)

This validates the input schema and guarantees a stable output contract.

Direct Builder Access (Advanced)

For advanced or internal usage, the feature builder can be called directly:

from ocf.features.ohlcv import build_ohlcv_features

features = build_ohlcv_features(
    underlying_daily,
    vol_window=21,
    ema_fast=10,
    ema_slow=21,
    bb_window=20,
)

This bypasses schema validation and is intended for controlled environments or testing.

As Part of the Composite Feature Set

The OHLCV block is included automatically when using the composite feature API:

from ocf.features.api import underlying_features

features = underlying_features(
    underlying_daily,
    include=["ohlcv"],
)

or together with other feature blocks:

features = underlying_features(
    underlying_daily,
    include=["ohlcv", "iv", "liquidity"],
)

Feature Definitions

Returns

  • Log returns are computed using log(px_close_t / px_close_{t-1}).
  • Overnight return measures open-to-previous-close movement.
  • Gap size captures absolute discontinuity risk across sessions.

Volatility Estimators

The block implements multiple realized volatility estimators:

  • Close-to-close volatility (baseline estimator)
  • Parkinson estimator (high–low range based)
  • Garman–Klass estimator (OHLC-based)
  • Yang–Zhang estimator (accounts for overnight jumps)

These estimators capture different microstructure effects and are commonly used in empirical volatility modeling.

Trend Measures

  • EMAs provide smoothed trend signals at different horizons.
  • Bollinger %B normalizes price position within rolling volatility bands.
  • Trend persistence encodes the sign of daily price changes.

Masking & Edge Cases

  • Rolling features are undefined until sufficient history exists.
  • Initial rows may contain null values due to differencing or rolling windows.
  • No forward-looking information is used at any point.

Configuration

The OHLCV feature block supports the following optional parameters:

Parameter Default Description
vol_window 21 Rolling window for volatility calculations
ema_fast 10 Fast EMA span
ema_slow 21 Slow EMA span
bb_window 20 Bollinger band window

Note

  • All features are computed in a single pass without joins.
  • No smoothing or fitting is applied beyond standard rolling operations.