RCI - Rank Correlation Index
RCI(順位相関指数)がTA-Libに無いのでPythonで書いてみた。

rci.py

    import numpy as np

    def rci(close: np.ndarray,
            timeperiod: int = 9) -> np.ndarray:
        rci = np.full_like(close, np.nan)
        rank_period = np.arange(1, timeperiod + 1)
        for i in range(timeperiod - 1, len(close)):
            rank_price = close[i - timeperiod + 1:i + 1]
            rank_price = np.argsort(np.argsort(rank_price)) + 1
            aa = 6 * sum((rank_period - rank_price)**2)
            bb = timeperiod * (timeperiod**2 - 1)
            rci[i] = (1 - aa / bb) * 100
        return rci

ちょっと改良版
forループをやめて高速化しました。
ただしtimeperiodが大きすぎるとメモリ消費が激しく
Forループ版との速度差が縮まっていきます。

rci_np.py

def rci(close: np.ndarray,
        timeperiod: int = 9) -> np.ndarray:
    rank_target = [np.roll(close, i, axis=-1) for i in range(timeperiod)]
    rank_target = np.vstack(rank_target)[:, timeperiod - 1:]
    price_rank = np.argsort(np.argsort(rank_target[::-1], axis=0), axis=0) + 1
    time_rank = np.arange(1, timeperiod + 1).reshape(timeperiod, -1)
    aa = np.sum((time_rank - price_rank)**2, axis=0, dtype=float) * 6
    bb = float(timeperiod * (timeperiod**2 - 1))
    cc = np.divide(aa, bb, out=np.zeros_like(aa), where=bb != 0)
    rci = (1 - cc) * 100
    rci = np.concatenate([np.full(timeperiod - 1, np.nan), rci], axis=0)
    return rci