quantlab.factor.kunquant_cs#

Custom KunQuant cross-sectional operators: each symbol against the others on one bar.

KunQuant is the library this project uses to compute most factors. A factor formula is written as a graph of operators (WindowedAvg, Rank, …), and KunQuant compiles that graph to native C++ code that runs over a whole (timestamp, symbol) array at once. This module adds operators that work on one bar at a time; quantlab.factor.kunquant_ts has the ones that look along time.

CrossSectionalZScore is a cross-sectional normalization: at each timestamp, each symbol is compared with all the other symbols on the same bar. Its time-series counterpart, kunquant_ts.WindowedZScore, compares each symbol with its own recent past; which one is right depends on the strategy that consumes the factor.

CrossSectionalWinsorize (winsorizing, 缩尾) and CrossSectionalTrim (trimming, 截尾) handle outliers on each bar: the first clips values to that bar’s lower and upper quantiles across symbols, the second replaces values outside them with NaN. They are typically applied before a cross-sectional z-score so that a few extreme symbols do not dominate its mean and spread.

CrossSectionalWeightedMean, CrossSectionalTopN, CrossSectionalSigmaClip, CapWeightedStandardize, the weighted least-squares residual CrossSectionalWLSResidual and CrossSectionalIndustrySizeFill are the estimation-universe tools of a Barra-style risk factor (quantlab.factor.predefined.barra), usable by any factor. CrossSectionalNeutralize fits the same industry and size regression as CrossSectionalIndustrySizeFill and keeps the residual: it neutralizes a factor against industry, size or both. RenormalizedCombine is elementwise, with no time or symbol window; it is here because it is a per-bar step of that same pipeline (combining descriptors).

Classes

`CapWeightedStandardize`	Standardize to a weighted mean of 0 and an equal-weighted std of 1 in a universe.
`CrossSectionalIndustrySizeFill`	Fill a missing value from a per-bar weighted regression on industry and size.
`CrossSectionalNeutralize`	Neutralize `y` against industry, size or both: its residual from a per-bar regression.
`CrossSectionalSigmaClip`	Drop data errors and clip outliers at multiples of a universe's standard deviation.
`CrossSectionalTopN`	1 for the `n` symbols with the largest `v` on each bar, 0 elsewhere.
`CrossSectionalTrim`	Cross-sectional trimming (截尾): drop values outside each bar's quantile bounds.
`CrossSectionalWLSResidual`	Residual of a per-bar weighted least-squares fit of `y` on `x`, with an intercept.
`CrossSectionalWeightedMean`	Weighted mean of `v` across symbols per bar, broadcast to every symbol.
`CrossSectionalWinsorize`	Cross-sectional winsorization (缩尾): clip each bar to its quantile bounds.
`CrossSectionalZScore`	Cross-sectional z-score, `(x - mean_t) / std_t` over symbols per bar.
`RenormalizedCombine`	Fixed-weight sum of several values, renormalized over the ones present.