API Reference

Lightweight library for backtesting factor strategies.

pqr.picking

Implemented classic factor strategies and pretransforms for them.

pqr.picking.filter(factor, *, universe)

Filters factor values based on given universe.

Actually, replaces factor values with nans, where universe is equal to False.

Parameters

factor (pd.DataFrame) – Matrix with factor values.
universe (pd.DataFrame) – Matrix with True/False, indicating to include factor values or not.

Returns

Filtered factor values.

Return type

pd.DataFrame

pqr.picking.look_back(factor, *, period, agg)

Aggregates factor values column-wise each period by agg.

If agg is not predefined can work very slow. In this case the best decision is to write your own effective realisation of transformation function.

Parameters

factor (pd.DataFrame) – Matrix with factor values.
period (int) – Period to look back on the data.
agg ({"pct", "mean", "median", "min", "max"} or callable) – Aggregation func to apply on factor values. If callable is given function must be appliable on pd.Series and return float.

Returns

Aggregated factor values.

Return type

pd.DataFrame

pqr.picking.lag(factor, *, period)

Lags factor values for period.

Can be used both forward and backward.

Parameters

factor (pd.DataFrame) – Matrix with factor values.
period (int) – Period to look back on the data.

Returns

Lagged factor values.

Return type

pd.DataFrame

pqr.picking.hold(factor, *, period)

Spread factor values for period.

Can be used to react on new information every period timestamps.

Parameters

factor (pd.DataFrame) – Matrix with factor values.
period (int) – Period to look back on the data.

Returns

Rolling mean of factor values.

Return type

pd.DataFrame

pqr.picking.quantiles(factor, *, min_q, max_q)

Picks when factor values are between min_q and max_q quantiles in a period.

Parameters

factor (pd.DataFrame) – Matrix with factor values.
min_q (float) – Quantile to estimate lower boarder of factor values to pick.
max_q (float) – Quantile to estimate upper boarder of factor values to pick.

Returns

Matrix of True/False, indicating whether factor values are between quantile boarders or not.

Return type

pd.DataFrame

pqr.picking.top(factor, *, k)

Picks when factor values are at the top k in a period.

Parameters

factor (pd.DataFrame) – Matrix with factor values.
k (int) – Place to estimate lower boarder of factor values to pick.

Returns

Matrix of True/False, indicating whether factor values are in the top or not.

Return type

pd.DataFrame

pqr.picking.bottom(factor, *, k)

Picks when factor values are in the bottom k in a period.

Parameters

factor (pd.DataFrame) – Matrix with factor values.
k (int) – Place to estimate upper boarder of factor values to pick.

Returns

Matrix of True/False, indicating whether factor values are in the bottom or not.

Return type

pd.DataFrame

pqr.picking.thresholds(factor, *, min_t, max_t)

Picks factor values between min_t and max_t thresholds.

Parameters

factor (pd.DataFrame) – Matrix with factor values.
min_t (float) – Lower boarder of factor values to pick.
max_t (float) – Upper boarder of factor values to pick.

Returns

Matrix of True/False, indicating whether factor values are between thresholds or not.

Return type

pd.DataFrame

pqr.allocation

Basic operations to get weighted positions (holdings) from strategy signals.

pqr.allocation.allocate(signals, *, weights)

Calculates portfolio holdings.

Weighted signals are normalized to 1.

Parameters

signals (pd.DataFrame) – Matrix, consists of True/False, indicating presence of an asset in a portfolio.
weights (pd.DataFrame) – Matrix with weights (e.g. market capitalization).

Returns

Matrix of holdings, each row sum equals to 1.

Return type

pd.DataFrame

pqr.allocation.ew(signals)

Calculates equally-weighted holdings.

Parameters: signals (pd.DataFrame) – Matrix, consists of True/False, indicating presence of an asset in a portfolio.
Returns: Matrix of holdings, each row sum equals to 1 and all non-zero row values are the same.
Return type: pd.DataFrame

pqr.allocation.scale(holdings, *, leverage)

Calculates leveraged portfolio holdings.

Parameters

holdings (pd.DataFrame) – Matrix of weighted positions.
leverage (pd.DataFrame) – Series with leverage value in each period.

Returns

Leveraged holdings.

Return type

pd.DataFrame

pqr.allocation.limit(holdings, *, min_leverage, max_leverage)

Clips portfolio leverage by min and max allowed leverage in a period.

Parameters

holdings (pd.DataFrame) – Matrix of weighted positions.
min_leverage (float) – Minimum allowed total leverage in a period.
max_leverage (float) – Maximum allowed total leverage in a period.

Returns

Matrix with scaled weights.

Return type

pd.DataFrame

pqr.evaluation

Functionality to estimate performance of a strategy.

pqr.evaluation.evaluate(holdings, *, universe_returns)

Calculates portfolio returns.

Parameters

holdings (pd.DataFrame) – Weights of a portfolio.
universe_returns (pd.DataFrame) – Returns of universe, available to trade for a strategy.

Returns

Periodic returns of a portfolio.

Return type

pd.Series

pqr.evaluation.to_returns(prices)

Calculates universe returns from given prices of assets universe.

1st period returns are set to zero. All incorrect values (nans and infs) are converted to zero.

Parameters: prices (pd.DataFrame) – Matrix with close prices of traded assets and nans for missing assets (e.g. already delisted)
Returns: Returns of assets universe.
Return type: pd.DataFrame

pqr.utils

Some additional stuff and “sugar” for creating pipelines.

pqr.utils.align(*args)

Aligns dataframes and series to ake them having the same index and columns.

Parameters: args (sequence of pd.DataFrame or pd.Series) – Dataframes and series to be aligned.
Returns: Aligned dataframes and series.
Return type: tuple of pd.DataFrame or pd.Series

pqr.utils.compose(*steps)

Combines functions to pipeline.

Parameters: steps (sequence of callable) – Steps to be composed into pipeline.
Returns: Function, realizing full pipeline.
Return type: callable

pqr.utils.freeze: alias of functools.partial