Strategy Pipeline API

The pipeline’s callable surface is the zrb Python package — shared data, backtest, beta, and alpha layers used by every strategy replication.

zrb is currently an internal library. B2B customers can get access scoped through an engagement — contact support@zentradings.com.

Setup

uv venv && uv pip install -e ".[dev]"
source .venv/bin/activate
pytest

`zrb.data` — price data

zrb.data.prices(tickers, start=None, end=None, field="Close", refresh=False) -> pd.DataFrame

Adjusted price history for a list of tickers, backed by a local cache. Pass refresh=True to re-download.

`zrb.backtest` — portfolio simulation

zrb.backtest.run(weights, returns, cost_bps=0.0) -> pd.DataFrame
zrb.backtest.long_short(signal, frac=0.2, min_names=6) -> pd.DataFrame
zrb.backtest.summary(returns, periods=252) -> pd.Series
zrb.backtest.drawdown(returns) -> pd.Series

run turns a weight matrix into portfolio returns net of transaction costs (cost_bps).
long_short builds a dollar-neutral portfolio from a cross-sectional signal: top/bottom frac of names, requiring at least min_names per side.
summary annualizes (Sharpe, vol, etc.); drawdown gives the drawdown series.

`zrb.beta` — is there alpha left?

zrb.beta.spanning_regression(returns, factors) -> ...
zrb.beta.report(returns, factors=None, periods=252) -> None

spanning_regression regresses strategy returns on known factor returns. A statistically insignificant intercept means the “strategy” is repackaged beta — you could have held the factors directly.

`zrb.alpha` — does the ranking predict?

zrb.alpha.ic(signal, fwd_returns, method="spearman") -> pd.Series
zrb.alpha.ic_summary(signal, fwd_returns, **kw) -> pd.Series
zrb.alpha.quantiles(signal, fwd_returns, n=5) -> pd.Series
zrb.alpha.report(signal, fwd_returns, ...) -> ...

ic — per-period rank information coefficient between signal and forward returns.
quantiles — forward returns by signal quantile; a real signal shows a monotone ladder.
report — the full verdict: mean IC, the quantile ladder, and gross-vs-net after turnover costs, side by side.

Worked example

from zrb import data, backtest, alpha

px = data.prices(["NVDA", "TSM", "ASML.AS", "005930.KS"], start="2020-01-01")
fwd = px.pct_change().shift(-1)
signal = ...  # your cross-sectional signal, same index/columns as px

alpha.report(signal, fwd)                    # does the ranking predict?
w = backtest.long_short(signal)
res = backtest.run(w, px.pct_change(), cost_bps=5)
backtest.summary(res["net"])

Get started

Research Agent