研究目的：
在上文《自定义指数系列1 - 构建价值100指数（10年10倍）》中，我们构造了自定义价值100指数，并且取得了还不错的效果。
但是，自定义价值100指数本身还是随着市场有较大的波动，那么我们能否通过技术择时，来达到更高的收益、更低的回撤呢？
研究内容：
上文中我们构造了月度调仓的自定义价值100指数，计算出了日频的open close high low。分别尝试均线择时、RSRS择时，观察技术择时在自定义指数上的效果。
研究结论：
RSRS择时在自定义指数上，也可以取得不错的效果。在自定义指数上，N=18 M=1100 阈值=0.7也是一个较好的参数。
首先我们来回顾一下RSRS的计算方式：
在每个交易日，取最近N个时间序列的high low价格进行ols回归，得到beta斜率和R2(rsquared)；
在每个交易日，取最近M个beta斜率，进行标准化，得到标准分zscore；
对标准分zscore，乘以rsquared，得到修正标准分；
对修正标准分，再乘以斜率beta，得到右偏修正标准分。
在上文中，我们构造了自定义价值100指数。下面，我们对自定义价值100指数，进行RSRS的计算。
import pickle
# 读取自定义指数数据文件
pkl_file = open('zs_df.pkl', 'rb')
zs_df = pickle.load(pkl_file)
pkl_file.close()
# OLS线性回归
def calc_OLS_list(x, y):
X = sm.add_constant(x)
model = sm.OLS(y, X)
results = model.fit()
beta = results.params[1]
r2 = results.rsquared
return beta, r2
import statsmodels.api as sm
import pandas as pd
# 计算RSRS
N = 18
M = 1100
factor_index = []
factor_list = []
for day_index in range(N, len(zs_df)):
df = zs_df[day_index - N: day_index]
high_list = df['high'].tolist()
low_list = df['low'].tolist()
# print(len(high_list), len(low_list))
beta, r2 = calc_OLS_list(low_list, high_list)
date = zs_df.index[day_index]
# print(date, beta, r2)
factor_index.append(date)
factor_list.append([beta, r2])
# 生成dataframe，行索引为日期，列名为'beta', 'r2'
factor_df = pd.DataFrame(data=factor_list, index=factor_index, columns=['beta', 'r2'])
factor_df = factor_df.reindex(columns=['beta', 'r2', 'score', 'revise_score', 'rightdev_score'])
for day_index in range(M, len(factor_df)):
date = factor_df.index[day_index]
# print(date)
beta_list = factor_df['beta'][day_index - M: day_index]
r2_list = factor_df['r2'][day_index - M: day_index]
# score
mean = beta_list.mean()
std = beta_list.std()
score = (beta_list[-1] - mean) / std
revise_score = score * r2_list[-1]
rightdev_score = score * r2_list[-1] * beta_list[-1]
# 标准分
factor_df.loc[date]['score'] = score
#修正标准分
factor_df.loc[date]['revise_score'] = revise_score
#右偏标准分
factor_df.loc[date]['rightdev_score'] = rightdev_score
print(factor_df)
beta r2 ... revise_score rightdev_score 2005-02-01 0.721068 0.796776 ... NaN NaN 2005-02-02 0.712813 0.796776 ... NaN NaN 2005-02-03 0.723315 0.614239 ... NaN NaN 2005-02-04 0.806353 0.684180 ... NaN NaN 2005-02-16 0.909182 0.737740 ... NaN NaN 2005-02-17 0.939005 0.827477 ... NaN NaN 2005-02-18 0.923678 0.859396 ... NaN NaN 2005-02-21 0.915748 0.875366 ... NaN NaN 2005-02-22 0.914922 0.883110 ... NaN NaN 2005-02-23 0.925498 0.898250 ... NaN NaN 2005-02-24 0.919584 0.918033 ... NaN NaN 2005-02-25 0.899269 0.921517 ... NaN NaN 2005-02-28 0.967134 0.942492 ... NaN NaN 2005-03-01 0.966841 0.948925 ... NaN NaN 2005-03-02 0.965966 0.951675 ... NaN NaN 2005-03-03 0.969137 0.953157 ... NaN NaN 2005-03-04 0.953566 0.950340 ... NaN NaN 2005-03-07 0.920248 0.945716 ... NaN NaN 2005-03-08 0.873502 0.930443 ... NaN NaN 2005-03-09 0.810561 0.908395 ... NaN NaN 2005-03-10 0.936448 0.943958 ... NaN NaN 2005-03-11 0.975239 0.915703 ... NaN NaN 2005-03-14 1.092860 0.932706 ... NaN NaN 2005-03-15 1.102901 0.881529 ... NaN NaN 2005-03-16 1.065494 0.861151 ... NaN NaN 2005-03-17 0.985226 0.814816 ... NaN NaN 2005-03-18 0.925480 0.784785 ... NaN NaN 2005-03-21 0.912233 0.858611 ... NaN NaN 2005-03-22 0.950629 0.907123 ... NaN NaN 2005-03-23 0.907662 0.925752 ... NaN NaN ... ... ... ... ... ... 2019-04-24 1.032696 0.961064 ... 0.571160 0.589831 2019-04-25 1.012395 0.938994 ... 0.563454 0.581877 2019-04-26 1.086277 0.917585 ... 0.434201 0.439583 2019-04-29 1.121149 0.944821 ... 0.834293 0.906273 2019-04-30 1.067524 0.957335 ... 1.056917 1.184961 2019-05-06 1.064483 0.968816 ... 0.758259 0.809459 2019-05-07 0.976013 0.968607 ... 0.748148 0.796391 2019-05-08 0.961438 0.978698 ... 0.227659 0.222198 2019-05-09 0.962896 0.982678 ... 0.142977 0.137463 2019-05-10 0.972397 0.984691 ... 0.152259 0.146610 2019-05-13 0.948904 0.987986 ... 0.209443 0.203661 2019-05-14 0.961378 0.987683 ... 0.069786 0.066221 2019-05-15 0.963548 0.987122 ... 0.144936 0.139339 2019-05-16 0.962013 0.986692 ... 0.158230 0.152462 2019-05-17 0.967529 0.984934 ... 0.149444 0.143767 2019-05-20 0.975740 0.982703 ... 0.182881 0.176942 2019-05-21 0.971554 0.979539 ... 0.232748 0.227102 2019-05-22 0.949886 0.975775 ... 0.208441 0.202512 2019-05-23 0.959606 0.963275 ... 0.080278 0.076255 2019-05-24 0.980501 0.948110 ... 0.137394 0.131844 2019-05-27 0.975072 0.916422 ... 0.257264 0.252248 2019-05-28 0.999516 0.886433 ... 0.219605 0.214130 2019-05-29 0.963649 0.813485 ... 0.345561 0.345393 2019-05-30 0.940352 0.658666 ... 0.139911 0.134825 2019-05-31 0.946206 0.744973 ... 0.020308 0.019097 2019-06-03 0.933361 0.729153 ... 0.050708 0.047980 2019-06-04 0.947285 0.756437 ... -0.007189 -0.006710 2019-06-05 0.989779 0.791843 ... 0.057725 0.054682 2019-06-06 1.091828 0.885225 ... 0.267119 0.264388 2019-06-10 1.114550 0.910277 ... 0.852680 0.930980 [3486 rows x 5 columns]
import matplotlib.pyplot as plt
# 设置字体 用来正常显示中文标签
plt.rcParams['font.sans-serif'] = ['Arial Unicode MS']
# 用来正常显示负号
plt.rcParams['axes.unicode_minus'] = False
"""
绘制直方图
data:必选参数，绘图数据
bins:直方图的长条形数目，可选项，默认为10
normed:是否将得到的直方图向量归一化，可选项，默认为0，代表不归一化，显示频数。normed=1，表示归一化，显示频率。
facecolor:长条形的颜色
edgecolor:长条形边框的颜色
alpha:透明度
"""
plt.hist(factor_df['beta'], bins=50, density=0, facecolor="#1F77B4", edgecolor=None, alpha=1)
# 显示横轴标签
plt.xlabel("区间")
# 显示纵轴标签
plt.ylabel("频数/频率")
# 显示图标题
plt.title("beta分布直方图")
plt.show()
plt.hist(factor_df['score'], bins=50, density=0, facecolor="#1F77B4", edgecolor=None, alpha=1)
# 显示横轴标签
plt.xlabel("区间")
# 显示纵轴标签
plt.ylabel("频数/频率")
# 显示图标题
plt.title("标准分-分布直方图")
plt.show()
plt.hist(factor_df['revise_score'], bins=50, density=0, facecolor="#1F77B4", edgecolor=None, alpha=1)
# 显示横轴标签
plt.xlabel("区间")
# 显示纵轴标签
plt.ylabel("频数/频率")
# 显示图标题
plt.title("修正标准分-分布直方图")
plt.show()
plt.hist(factor_df['rightdev_score'], bins=50, density=0, facecolor="#1F77B4", edgecolor=None, alpha=1)
# 显示横轴标签
plt.xlabel("区间")
# 显示纵轴标签
plt.ylabel("频数/频率")
# 显示图标题
plt.title("右偏修正标准分-分布直方图")
plt.show()
从以上的分布图看，标准分的分布有尖峰厚尾的现象，且均值基本为0；经过标准分乘以R2之后，修正标准分的分布更符合正态分布；而右偏修正标准分，呈现出右偏态。
可以看出，对自定义指数的RSRS计算，计算出的数值分布与在沪深300指数上的分布一致。
下面，我们把beta、标准分、修正标准分、右偏标准分、自定义价值100指数收盘价以及上下阈值$^+_-0.7$的时间序列绘制成折线图。
# 展示折线图
def show_factor_value_line(factor_df):
# 颜色 蓝 橙 绿 红 紫
color_list = ['#5698c6', '#ff9e4a', '#60b760', '#e05c5d', '#ae8ccd', 'black', 'black']
label_list = ['beta', 'score', 'revise_score', 'rightdev_score', 'close', 'up', 'low']
# 这里需要用每天的因子数据分档，计算出5个折线
y_list = [[], [], [], [], [], [], []]
for i in range(0, len(factor_df)):
date = factor_df.index[i]
factor_data = factor_df.iloc[i]
for l in range(0, 4):
y_list[l].append(factor_data[label_list[l]])
close = zs_df.loc[date]['close']
y_list[4].append(close)
y_list[5].append(0.7)
y_list[6].append(-0.7)
fig = plt.figure(figsize=(12, 8))
ax = fig.add_subplot(1, 1, 1)
ax.set_title("")
for i in range(0, len(y_list)):
yi = y_list[i]
ax.plot(yi, color_list[i], label=label_list[i])
x = np.arange(0, len(factor_df), 300)
x_label = []
for i in range(0, len(factor_df)):
if i in x:
date = factor_df.index[i]
x_label.append(date)
plt.xticks(x, x_label, rotation='vertical')
plt.xticks(rotation=360) # 旋转45度显示
legend = ax.legend(loc='upper left', shadow=False)
plt.show()
# 展示折线图
show_factor_value_line(factor_df)
可以看出，beta在时间序列上有波动，稳定大于0；beta的标准分、修正标准分、右偏标准分，在均值0左右上下波动，标准分的最低值较低，右偏标准分的峰度较高。但是右偏标准分的正值高峰，似乎和自定义价值100指数的最佳买点不匹配。
在沪深300指数上，我本人更倾向于使用修正标准分，在光大的择时周报中，沪深300指数的RSRS指标也是用修正标准分计算的。
那么，RSRS修正标准分在自定义价值100指数上择时，能否取得良好的效果呢？标准分、修正标准分、右偏标准分在自定义价值100指数上择时，到底谁更厉害？RSRS能否跑赢双均线？
接下来，我会对比3个标准分、MA60均线以及无择时的效果，我们拭目以待。
通过上文中的代码，已经获取到了样本内的RSRS指标数据。
以下分别对比MA择时收益、RSRS择时收益，最后对最好的MA均线择时和最好的RSRS标准分择时做一个对比。
# 求每天的收益率
ret_df_index = []
ret_df_data = []
for i in range(0, len(zs_df)):
date = zs_df.index[i]
ret_df_index.append(date)
factor_data = zs_df.loc[date]
ret_df_data.append(factor_data['close'])
ret_df = pd.DataFrame(data=ret_df_data, index=ret_df_index, columns=['close'])
# 收盘价的变化率为收益率
ret_df['rt'] = ret_df.pct_change()
ret_df = ret_df.reindex(columns=['close', 'rt', 'ma120', 'ma60', 'ma30', 'ma20', 'ma10','ma5'])
for i in range(120, len(ret_df)):
date = ret_df.index[i]
ret_df.loc[date]['ma120'] = ret_df['close'][i - 120:i].mean()
for i in range(60, len(ret_df)):
date = ret_df.index[i]
ret_df.loc[date]['ma60'] = ret_df['close'][i - 60:i].mean()
for i in range(30, len(ret_df)):
date = ret_df.index[i]
ret_df.loc[date]['ma30'] = ret_df['close'][i - 30:i].mean()
for i in range(20, len(ret_df)):
date = ret_df.index[i]
ret_df.loc[date]['ma20'] = ret_df['close'][i - 20:i].mean()
for i in range(10, len(ret_df)):
date = ret_df.index[i]
ret_df.loc[date]['ma10'] = ret_df['close'][i - 10:i].mean()
for i in range(5, len(ret_df)):
date = ret_df.index[i]
ret_df.loc[date]['ma5'] = ret_df['close'][i - 5:i].mean()
# 上移，转为当期收益率
ret_df['rt'] = ret_df['rt'].shift(-1)
ret_df['rt'] = ret_df['rt'].fillna(0)
print("ret_df")
print(ret_df)
ret_df
close rt ... ma10 ma5
2005-01-06 0.984399 -0.004273 ... NaN NaN
2005-01-07 0.980193 0.007802 ... NaN NaN
2005-01-10 0.987840 0.001123 ... NaN NaN
2005-01-11 0.988949 0.005529 ... NaN NaN
2005-01-12 0.994417 0.001192 ... NaN NaN
2005-01-13 0.995603 -0.009026 ... NaN 0.987160
2005-01-14 0.986617 -0.023293 ... NaN 0.989400
2005-01-17 0.963636 0.006111 ... NaN 0.990685
2005-01-18 0.969524 -0.008519 ... NaN 0.985844
2005-01-19 0.961265 -0.017065 ... NaN 0.981959
2005-01-20 0.944861 0.030554 ... 0.981244 0.975329
2005-01-21 0.973730 0.019164 ... 0.977290 0.965180
2005-01-24 0.992391 0.007783 ... 0.976644 0.962603
2005-01-25 1.000115 -0.005659 ... 0.977099 0.968354
2005-01-26 0.994455 -0.014611 ... 0.978216 0.974472
2005-01-27 0.979925 -0.005268 ... 0.978220 0.981110
2005-01-28 0.974763 -0.007963 ... 0.976652 0.988123
2005-01-31 0.967001 0.000989 ... 0.975467 0.988330
2005-02-01 0.967957 0.054081 ... 0.975803 0.983252
2005-02-02 1.020304 -0.013117 ... 0.975646 0.976820
2005-02-03 1.006921 0.022975 ... 0.981550 0.981990
2005-02-04 1.030055 0.009578 ... 0.987756 0.987389
2005-02-16 1.039920 -0.006288 ... 0.993389 0.998448
2005-02-17 1.033382 -0.011582 ... 0.998142 1.013032
2005-02-18 1.021413 0.017595 ... 1.001468 1.026117
2005-02-21 1.039385 0.022405 ... 1.004164 1.026338
2005-02-22 1.062672 0.003274 ... 1.010110 1.032831
2005-02-23 1.066152 0.005165 ... 1.018901 1.039355
2005-02-24 1.071658 0.005566 ... 1.028816 1.044601
2005-02-25 1.077623 0.000103 ... 1.039186 1.052256
... ... ... ... ... ...
2019-04-24 12.016030 -0.028016 ... 12.258106 12.219701
2019-04-25 11.679392 -0.021139 ... 12.197510 12.156202
2019-04-26 11.432507 -0.007791 ... 12.140640 12.035861
2019-04-29 11.343441 0.000871 ... 12.066889 11.841165
2019-04-30 11.353322 -0.062390 ... 11.991801 11.691963
2019-05-06 10.644983 0.014747 ... 11.892320 11.564938
2019-05-07 10.801965 -0.017862 ... 11.723466 11.290729
2019-05-08 10.609015 -0.017850 ... 11.575552 11.115244
2019-05-09 10.419648 0.035000 ... 11.395855 10.950545
2019-05-10 10.784340 -0.007813 ... 11.228875 10.765787
2019-05-13 10.700081 -0.002176 ... 11.108464 10.651990
2019-05-14 10.676795 0.022112 ... 10.976869 10.663010
2019-05-15 10.912877 0.016998 ... 10.876610 10.637976
2019-05-16 11.098376 -0.031375 ... 10.824647 10.698748
2019-05-17 10.750163 -0.013776 ... 10.800140 10.834494
2019-05-20 10.602065 0.017821 ... 10.739824 10.827658
2019-05-21 10.791003 -0.013027 ... 10.735533 10.808055
2019-05-22 10.650428 -0.016051 ... 10.734436 10.830897
2019-05-23 10.479475 -0.005572 ... 10.738578 10.778407
2019-05-24 10.421081 0.017422 ... 10.744560 10.654627
2019-05-27 10.602639 0.004183 ... 10.708234 10.588810
2019-05-28 10.646989 0.001608 ... 10.698490 10.588925
2019-05-29 10.664113 -0.004381 ... 10.695510 10.560122
2019-05-30 10.617398 -0.005729 ... 10.670633 10.562859
2019-05-31 10.556569 -0.015678 ... 10.622535 10.590444
2019-06-03 10.391060 -0.012315 ... 10.603176 10.617542
2019-06-04 10.263096 -0.002946 ... 10.582075 10.575226
2019-06-05 10.232860 -0.009529 ... 10.529285 10.498447
2019-06-06 10.135346 0.011077 ... 10.487528 10.412197
2019-06-10 10.247620 0.000000 ... 10.453115 10.315786
[3504 rows x 8 columns]
# MA均线择时收益
def show_ma_factor_return_rate(ori_factor_df):
factor_df = ori_factor_df.dropna()
color_list = ['black','#5698c6', '#ff9e4a', '#60b760', '#e05c5d', '#ae8ccd','grey','gold']
label_list = ['不择时','ma120', 'ma60', 'ma30','ma20','ma10','ma5','ma5-ma60']
# 这里需要用每天的因子数据分档，计算出5个折线
y_list = []
for i in range(0, len(label_list)):
y_list.append([])
# 收益率折线图从0开始
for i in range(0, len(label_list)):
yi = y_list[i]
yi.append(0)
pre_date = None
flag_list = [0, 0, 0, 0]
for date in factor_df.index:
if type(pre_date) == type(None):
pre_date = date
continue
# pre_date为date的上一个调仓日，这里获取上一个调仓日的因子值
pre_factor_data = factor_df.loc[pre_date]
# 获取上一个调仓日到本调仓日的收益率
ret = ret_df.loc[pre_date]['rt']
# 累计
for i in range(1, 7):
yi = y_list[i]
score_name = label_list[i]
ma = ret_df.loc[pre_date][score_name]
if ret_df.loc[pre_date]['close'] > ma:
sum_ret = yi[-1] + ret
else:
sum_ret = yi[-1]
yi.append(sum_ret)
if ret_df.loc[pre_date]['ma5'] > ret_df.loc[pre_date]['ma60']:
y_list[7].append(y_list[7][-1] + ret)
else:
y_list[7].append(y_list[7][-1])
# 不择时
y_list[0].append(y_list[0][-1] + ret)
pre_date = date
fig = plt.figure(figsize=(12, 8))
ax = fig.add_subplot(1, 1, 1)
ax.set_title("")
for i in range(0, len(y_list)):
yi = y_list[i]
ax.plot(yi, color_list[i], label=label_list[i])
x = np.arange(0, len(factor_df), 300)
x_label = []
for i in range(0, len(factor_df)):
if i in x:
date = factor_df.index[i]
x_label.append(date)
plt.xticks(x, x_label, rotation='vertical')
plt.xticks(rotation=360) # 旋转45度显示
legend = ax.legend(loc='upper left', shadow=False)
plt.show()
# ma均线择时收益率
show_ma_factor_return_rate(factor_df)
由上图可以看出，在自定义价值100指数上，20日均线择时效果最好，5日均线择时效果最差。只有20日均线择时跑赢了指数本身。
# RSRS择时收益
def show_score_factor_return_rate(ori_factor_df):
factor_df = ori_factor_df.dropna()
color_list = ['black','#5698c6', '#ff9e4a', '#60b760', '#e05c5d', '#ae8ccd']
label_list = ['不择时','score', 'revise_score', 'rightdev_score']
# 这里需要用每天的因子数据分档，计算出5个折线
y_list = [[], [], [], []]
# 收益率折线图从0开始
for i in range(0, len(label_list)):
yi = y_list[i]
yi.append(0)
pre_date = None
flag_list = [0, 0, 0, 0]
for date in factor_df.index:
if type(pre_date) == type(None):
pre_date = date
continue
# pre_date为date的上一个调仓日，这里获取上一个调仓日的因子值
pre_factor_data = factor_df.loc[pre_date]
# 获取上一个调仓日到本调仓日的收益率
ret = ret_df.loc[pre_date]['rt']
# 累计
for i in range(1, 4):
yi = y_list[i]
score_name = label_list[i]
if pre_factor_data[score_name] > 0.7:
flag_list[i] = 1
elif pre_factor_data[score_name] < -0.7:
flag_list[i] = 0
# 累计收益率
if flag_list[i] == 1:
sum_ret = yi[len(yi) - 1] + ret
else:
sum_ret = yi[len(yi) - 1]
yi.append(sum_ret)
# 不择时
y_list[0].append(y_list[0][-1] + ret)
pre_date = date
fig = plt.figure(figsize=(12, 8))
ax = fig.add_subplot(1, 1, 1)
ax.set_title("")
for i in range(0, len(y_list)):
yi = y_list[i]
ax.plot(yi, color_list[i], label=label_list[i])
x = np.arange(0, len(factor_df), 300)
x_label = []
for i in range(0, len(factor_df)):
if i in x:
date = factor_df.index[i]
x_label.append(date)
plt.xticks(x, x_label, rotation='vertical')
plt.xticks(rotation=360) # 旋转45度显示
legend = ax.legend(loc='upper left', shadow=False)
plt.show()
# 收益率
show_score_factor_return_rate(factor_df)
由上图看出，在自定义价值100指数上择时，RSRS标准分、修正标准分、右偏标准分择时效果分别由好到差，与在沪深300指数上表现出了完全相反的结果。但是三个指标都跑赢了指数本身。
# 20日均线与RSRS标准分
def show_ma_score_factor_return_rate(ori_factor_df):
factor_df = ori_factor_df.dropna()
color_list = ['black','#5698c6', '#e05c5d']
label_list = ['不择时', 'score','ma20']
# 这里需要用每天的因子数据分档，计算出5个折线
y_list = []
for i in range(0, len(label_list)):
y_list.append([])
# 收益率折线图从0开始
for i in range(0, len(label_list)):
yi = y_list[i]
yi.append(0)
pre_date = None
flag_list = [0, 0, 0, 0]
for date in factor_df.index:
if type(pre_date) == type(None):
pre_date = date
continue
# pre_date为date的上一个调仓日，这里获取上一个调仓日的因子值
pre_factor_data = factor_df.loc[pre_date]
# 获取上一个调仓日到本调仓日的收益率
ret = ret_df.loc[pre_date]['rt']
# 累计
if pre_factor_data['score'] > 0.7:
flag_list[1] = 1
elif pre_factor_data['score'] < -0.7:
flag_list[1] = 0
# 累计收益率
if flag_list[1] == 1:
sum_ret = y_list[1][-1] + ret
else:
sum_ret = y_list[1][-1]
y_list[1].append(sum_ret)
ma = ret_df.loc[pre_date]['ma20']
if ret_df.loc[pre_date]['close'] > ma:
sum_ret = y_list[2][-1] + ret
else:
sum_ret = y_list[2][-1]
y_list[2].append(sum_ret)
# 不择时
y_list[0].append(y_list[0][-1] + ret)
pre_date = date
fig = plt.figure(figsize=(12, 8))
ax = fig.add_subplot(1, 1, 1)
ax.set_title("")
for i in range(0, len(y_list)):
yi = y_list[i]
ax.plot(yi, color_list[i], label=label_list[i])
x = np.arange(0, len(factor_df), 300)
x_label = []
for i in range(0, len(factor_df)):
if i in x:
date = factor_df.index[i]
x_label.append(date)
plt.xticks(x, x_label, rotation='vertical')
plt.xticks(rotation=360) # 旋转45度显示
legend = ax.legend(loc='upper left', shadow=False)
plt.show()
show_ma_score_factor_return_rate(factor_df)
可以看出，rsrs标准分的择时效果胜过ma20择时。
另外我写回测使用多回测框架，发现在自定义价值100指数上，N=18 M=1100 阈值=0.7也是一个较好的参数，回撤较低，年化较高。还有比较好的几组参数是N=22 M=800 阈值=0.9;N=15 M=600 阈值=0.9;。
在上文系列1，我们构造了一个自定义价值100指数；RSRS择时在自定义价值100指数上，也可以取得不错的择时效果。值得一提的是，N=18 M=1100 阈值=0.7在自定义价值100也是一个较好的参数。
有兴趣的朋友可以试试每次选市值最小的100个股票，构造下证100指数，看看效果。
本社区仅针对特定人员开放
查看需注册登录并通过风险意识测评
5秒后跳转登录页面...
移动端课程
