MA线

• 收盘 > MA，看多
• 收盘 < MA，看空

data_df.loc[:, "X_SMA"] = ta.sma(close=data_df.Close, length=10)
data_df.loc[:, "X_WMA"] = ta.wma(close=data_df.Close, length=10)
data_df.loc[:, "X_EMA"] = ta.ema(close=data_df.Close, length=10)

data_df.loc[:, "T_SMA"] = data_df.X_SMA < data_df.Close
data_df.loc[:, "T_WMA"] = data_df.X_WMA < data_df.Close
data_df.loc[:, "T_EMA"] = data_df.X_EMA < data_df.Close

随机指标

• 指标成长，看多
• 指标降低，看空

kdj = ta.kdj(high=data_df.High, low=data_df.Low, close=data_df.Close, length=14)

data_df.loc[:, "X_WILLR"] = ta.willr(
    high=data_df.High, low=data_df.Low, close=data_df.Close, length=14
)
data_df.loc[:, "X_STCK"] = kdj[kdj.columns[0]]
data_df.loc[:, "X_STCD"] = kdj[kdj.columns[1]]

data_df.loc[:, "T_WILLR"] = data_df.X_WILLR > data_df.X_WILLR.shift(1)
data_df.loc[:, "T_STCK"] = data_df.X_STCK > data_df.X_STCK.shift(1)
data_df.loc[:, "T_STCD"] = data_df.X_STCD > data_df.X_STCD.shift(1)

RSI、CCI

• 指标过高，看空
• 指标过低，看多
• 指标位於中间值
  • 指标成长，看多
  • 指标降低，看空

data_df.loc[:, "X_RSI"] = ta.rsi(close=data_df.Close, length=6)

def rsi_trend(x):
    if x[0] > 70:
        return False
    elif x[0] < 30:
        return True
    else:
        return x[0] > x[1]


data_df.loc[:, "T_RSI"] = pd.concat(
    [data_df.X_RSI, data_df.X_RSI.shift(1)], axis=1
).apply(rsi_trend, axis=1)

data_df.loc[:, "X_CCI"] = ta.cci(
    high=data_df.High, low=data_df.Low, close=data_df.Close, length=14
)


def cci_trend(x):
    if x[0] > 200:
        return False
    elif x[0] < -200:
        return True
    else:
        return x[0] > x[1]

data_df.loc[:, "T_CCI"] = pd.concat(
    [data_df.X_CCI, data_df.X_CCI.shift(1)], axis=1
).apply(cci_trend, axis=1)

Momentum

动量 > 0 ，看多
动量 <= 0，看空

data_df.loc[:, "X_MOM"] = ta.mom(close=data_df.Close, length=10)

data_df.loc[:, "T_MOM"] = data_df.X_MOM > 0

A/D oscillator与MACD论文的公式很怪暂时先不考虑

结果

使用原始指标准确率：

使用趋势指标准确率：

看起来是有增长，趋势化资料明显减轻了Overfitting状况，明天考虑修改些指标的参数看是否会有所提升。