
import numpy as np
import pandas as pd
import os
import pandas as pd
import matplotlib.pyplot as plt
from statsmodels.tsa.stattools import acf
%matplotlib inline


closes = pd.read_csv("stockclose.csv", index_col=0, parse_dates=True)
closes.head()


stockIndex = '000001.SZ'
asset = closes[stockIndex].dropna()['1999':]
plt.style.use("bmh")

asset.plot()
plt.legend()
asset
# plt.style.available

1999-01-04    14.55
1999-01-05    14.49
1999-01-06    14.58
1999-01-07    14.90
1999-01-08    15.04
              ...  
2019-10-25    16.88
2019-10-28    16.66
2019-10-29    16.91
2019-10-30    16.43
2019-10-31    16.26
Name: 000001.SZ, Length: 5042, dtype: float64


from math import log
plt.rcParams['font.sans-serif'] = ['SimHei']
plt.rcParams['axes.unicode_minus'] = False


def cal_log_return(ts, name):
    def freq_log_return(lts, ts, freq):
        return lts.resample(freq).last() - lts.resample(freq).last().shift()

    lts = ts.map(log)
    lrweek = freq_log_return(lts, ts, "W")
    lrmonth = freq_log_return(lts, ts, "M")
    lrquarter = freq_log_return(lts, ts, "Q")
    lryear = freq_log_return(lts, ts, "Y")
    lr = [lrweek, lrmonth, lrquarter, lryear]
    fig, axes = plt.subplots(nrows=2, ncols=2, figsize=(10, 6), sharex=True)
    freqs = ["周收益率", "月收益率", "季度收益率", "年收益率"]
    for i in range(2):
        for j in range(2):
            axes[i][j].plot(lr[i*2+j])
            axes[i][j].set_title(freqs[2*i+j])
    fig.suptitle(name, fontsize=20)
    return np.array(lr)


lr = cal_log_return(asset, stockIndex)

<ipython-input-4-8ae3e31df5e7>:23: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray
  return np.array(lr)


def distr(lrlist, name):
    fig, axes = plt.subplots(nrows=2, ncols=2, figsize=(10, 8))
    fig.suptitle(name+"收益率分布", fontsize=20)
    freqs = ["周", "月", "季度", "年"]
    for i in range(2):
        for j in range(2):
            axes[i][j].hist(lrlist[2*i+j], bins=int(110 / (10)**i))
            axes[i][j].set_title(freqs[2*i+j] + "分布")
    plt.show()


distr(lr, stockIndex)


rolling_window = 252  # 一年为滑动窗口
return_series = asset.pct_change()
# 一年滑动夏普比
anualMean = return_series.rolling(
    window=rolling_window, min_periods=1, center=False).mean()
anualStd = return_series.rolling(
    window=rolling_window, min_periods=1, center=False).std()
Sharpratio = np.sqrt(rolling_window)*(anualMean/anualStd)['2000':]
Sharpratio.plot()
plt.legend()
plt.title("Sharp Ratio_One Year RW", fontsize=16)

Text(0.5, 1.0, 'Sharp Ratio_One Year RW')


# Calculate the max drawdown in the past window days for each day
rolling_max = asset.rolling(rolling_window, min_periods=1).max()
daily_drawdown = asset/rolling_max - 1.0

# Calculate the minimum (negative) daily drawdown
max_daily_drawdown = daily_drawdown.rolling(rolling_window, min_periods=1).min()

# Plot the results
daily_drawdown['2000':].plot()
max_daily_drawdown['2000':].plot()

# Show the plot
plt.title("年滑动窗口最大回撤", fontsize=15)
plt.show()


short_window = 20 
long_window = 120

signals = pd.DataFrame(index = asset.index)
signals['signal'] = 0

# Create short simple moving average over the short window
signals['short_mavg'] = asset.rolling(short_window, min_periods=1, center=False).mean()

# Create long simple moving average over the long window
signals['long_mavg'] = asset.rolling(long_window, min_periods=1, center=False).mean()

signals.plot()

<AxesSubplot:>


signals['signal'][short_window:] = np.where(signals['short_mavg'][short_window:] 
                                            > signals['long_mavg'][short_window:], 1.0, 0.0)
plt.scatter(signals.index, signals['signal'].values)

<ipython-input-9-36d604b1025f>:1: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  signals['signal'][short_window:] = np.where(signals['short_mavg'][short_window:]
C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:670: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  iloc._setitem_with_indexer(indexer, value)

<matplotlib.collections.PathCollection at 0x25a4fe02a30>


signals['positions'] = signals['signal'].diff()
signals['positions']

1999-01-04    NaN
1999-01-05    0.0
1999-01-06    0.0
1999-01-07    0.0
1999-01-08    0.0
             ... 
2019-10-25    0.0
2019-10-28    0.0
2019-10-29    0.0
2019-10-30    0.0
2019-10-31    0.0
Name: positions, Length: 5042, dtype: float64


# Initialize the plot figure
fig = plt.figure()

# Add a subplot and label for y-axis
ax1 = fig.add_subplot(111,  ylabel='Price')

# Plot the short and long moving averages
signals[['short_mavg', 'long_mavg']].plot(ax=ax1, lw=2.)

# Plot the buy signals
ax1.plot(signals.loc[signals.positions == 1.0].index, 
         signals.short_mavg[signals.positions == 1.0],
         '^', markersize=8, color='red')
         
# Plot the sell signals
ax1.plot(signals.loc[signals.positions == -1.0].index, 
         signals.short_mavg[signals.positions == -1.0],
         'v', markersize=7, color='green')
         
# Show the plot
plt.show()


# Set the initial capital
initial_capital= float(10000.0)

# Create a DataFram `positions`
positions = pd.DataFrame(index=signals.index).fillna(0.0)

# Buy a 100 shares
positions['Value'] = 100*signals['signal']  

# Initialize the portfolio with value owned   
portfolio = positions.multiply(asset, axis=0)
portfolio.tail()


pos_diff = positions.diff()

# Add `holdings` to portfolio
portfolio['holdings'] = (positions.multiply(asset, axis=0)).sum(axis=1)
portfolio['cash'] = initial_capital - (pos_diff.multiply(asset, axis =0)).sum(axis=1).cumsum()


# Add `total` to portfolio
portfolio['total'] = portfolio['cash'] + portfolio['holdings']

# Add `returns` to portfolio
portfolio['returns'] = portfolio['total'].pct_change()

# Print the first lines of `portfolio`
print(portfolio.tail())

             Value  holdings    cash    total   returns
2019-10-25  1688.0    1688.0  9967.0  11655.0  0.000086
2019-10-28  1666.0    1666.0  9967.0  11633.0 -0.001888
2019-10-29  1691.0    1691.0  9967.0  11658.0  0.002149
2019-10-30  1643.0    1643.0  9967.0  11610.0 -0.004117
2019-10-31  1626.0    1626.0  9967.0  11593.0 -0.001464


# Create a figure
fig = plt.figure()

ax1 = fig.add_subplot(111, ylabel='组合价值（元）')

# Plot the equity curve in dollars
portfolio['total'].plot(ax=ax1, lw=2.)

ax1.plot(portfolio.loc[signals.positions == 1.0].index, 
         portfolio.total[signals.positions == 1.0],
         '^', markersize=10, color='red')
ax1.plot(portfolio.loc[signals.positions == -1.0].index, 
         portfolio.total[signals.positions == -1.0],
         'v', markersize=10, color='green')
plt.title("组合价值变动图", fontsize = 15)
# Show the plot
plt.show()


# Isolate the returns of your strategy
returns = portfolio['returns']

# annualized Sharpe ratio
sharpe_ratio = np.sqrt(252) * (returns.mean() / returns.std())

# Print the Sharpe ratio
print(sharpe_ratio)

0.1752452210952673


# Define a trailing 252 trading day window
window = 252

# Calculate the max drawdown in the past window days for each day
rolling_max = portfolio['total'].rolling(window, min_periods=1).max()
daily_drawdown = portfolio['total']/rolling_max - 1.0

# Calculate the minimum (negative) daily drawdown
max_daily_drawdown = daily_drawdown.rolling(window, min_periods=1).min()

# Plot the results
daily_drawdown.plot()
max_daily_drawdown.plot()

# Show the plot
plt.show()


# Get the number of days in `aapl`
days = (portfolio['total'].index[-1] - portfolio['total'].index[0]).days

# Calculate the CAGR 
cagr = ((((portfolio['total'][-1]) / portfolio['total'][1])) ** (365.0/days)) - 1

# Print CAGR
print("年化收益率:",cagr)

年化收益率: 0.007119633448701146


asset.head()

1999-01-04    14.55
1999-01-05    14.49
1999-01-06    14.58
1999-01-07    14.90
1999-01-08    15.04
Name: 000001.SZ, dtype: float64


asset = pd.read_excel("sh000001.xlsx", index_col=0, parse_dates=True).dropna(axis=1)
asset.head()
asset['close'].plot()
plt.title("资产价值变动趋势", fontsize = 15)

Text(0.5, 1.0, '资产价值变动趋势')


window = 20 #月频
start_index = '1996'
Pc = asset['close']
Pl = asset['low'].rolling(window, min_periods=1, center=False).min()
Ph = asset['high'].rolling(window, min_periods=1, center=False).max()

fastStochastics = (Pc - Pl)/ (Ph - Pl)*100
plt.figure(figsize=(24,4))
fastStochastics['2010':].plot()
plt.title("Fast Stochastics", fontsize=17)

# 加入原DataFrame
asset['FastStochastics'] = fastStochastics[start_index:]


plt.figure(figsize=(24,4))
perc_R = (Ph - Pc.shift(window-1))/(Ph-Pl) *(-100)
perc_R['2010':].plot()
plt.title("William's R", fontsize = 17)
asset["WilliamsR"] = perc_R[start_index:]


slowStochastics = fastStochastics.rolling(3, min_periods=1, center=False).mean()
plt.figure(figsize=(24,4))
slowStochastics['2010':].plot()
plt.title("Slow Stochastics", fontsize=17)
asset['SlowStochastics'] = slowStochastics[start_index: ]


#Set Label
asset['month_return'] = asset['close'].shift(-20)/asset['close'] -1
asset['label'] = 1 * (asset['month_return'] > 0.05) + (-1) * (asset['month_return'] < -0.05)
asset[start_index:].head()


ex_volatility = asset['month_return'].rolling(window, min_periods=1,center=False).std()[start_index:]
asset['Ex-postVolatility'] = ex_volatility[start_index:]
plt.figure(figsize = (12,6))
ex_volatility.plot()

<AxesSubplot:xlabel='date'>


short_window = 5
mid_window = 20
long_window = 250
signals = pd.DataFrame(index = asset.index)


asset['short_signal'] = 0.0
asset['long_signal'] = 0.0
signals['short_mavg'] = asset['close'].rolling(window=short_window, min_periods=1, center=False).mean()
signals['mid_mavg'] = asset['close'].rolling(window=mid_window, min_periods=1, center=False).mean()
signals['long_mavg'] = asset['close'].rolling(window=long_window, min_periods=1, center=False).mean()
signals


asset['short_signal'].iloc[short_window:] = np.where(signals['short_mavg'][short_window:] > signals['mid_mavg'][short_window:], 1.0, 0.0)
asset['long_signal'].iloc[mid_window:] = np.where(signals['mid_mavg'][mid_window:] > signals['long_mavg'][mid_window:], 1.0, 0.0)
asset.head()

C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:670: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  iloc._setitem_with_indexer(indexer, value)
C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:670: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  iloc._setitem_with_indexer(indexer, value)


from IPython.display import Image
from sklearn import tree
import pydotplus
%matplotlib inline
asset = asset[:'2019-09']
end_index = '2018'
train_data = asset[start_index:end_index]
test_data = asset[end_index:]
clf = tree.DecisionTreeClassifier(max_depth=4)
clf = clf.fit(train_data[['FastStochastics', 'SlowStochastics', 'WilliamsR', 
                          'Ex-postVolatility', 'short_signal', 'long_signal']].values,
              train_data['label'].values)

# 注意：
dot_data = tree.export_graphviz(clf, out_file="temptree.dot", feature_names=["FastStochastics", "SlowStochastics", "WilliamsR", 
                                                                             "Ex-postVolatility",'short_signal','long_signal'],
                                class_names=['-1', '0', '1'],
                                filled=True, rounded=True)


dot_file = r'.\temptree.dot'
graph = pydotplus.graph_from_dot_file(dot_file)
Image(graph.create_png())


clf.score(test_data[['FastStochastics', 'SlowStochastics', 'WilliamsR',
                     'Ex-postVolatility','short_signal','long_signal']].values, test_data['label'].values)

0.6666666666666666


from sklearn.model_selection import train_test_split
X = asset[start_index:][['FastStochastics','SlowStochastics','WilliamsR','Ex-postVolatility']]
Y = asset.label[start_index:].values
X_train,X_test, Y_train, Y_test = train_test_split(X, Y,test_size=0.33,random_state=10)


from sklearn.pipeline import Pipeline
pipeline = Pipeline([('clf',tree.DecisionTreeClassifier(criterion='gini'))])


from sklearn.model_selection import GridSearchCV
hyperparameters = { 
    'clf__max_depth': (3, 4, 5, 6, 8, 9) 
}  
clf = GridSearchCV(pipeline, hyperparameters, cv=10, n_jobs=1,verbose=1)
#print(Y_test)
clf.fit(X_train, Y_train)

Fitting 10 folds for each of 6 candidates, totalling 60 fits

GridSearchCV(cv=10,
             estimator=Pipeline(steps=[('clf', DecisionTreeClassifier())]),
             n_jobs=1, param_grid={'clf__max_depth': (3, 4, 5, 6, 8, 9)},
             verbose=1)


from sklearn.metrics import classification_report
print ('最佳效果：%0.3f' %clf.best_score_)  
print('最优参数')  
print(clf.best_params_)
predictions = clf.predict(X_test) 
#print(clf.predict(X_test))
print(classification_report(Y_test, predictions)) 
clf.predict(X_test)

最佳效果：0.589
最优参数
{'clf__max_depth': 4}
              precision    recall  f1-score   support

          -1       0.47      0.22      0.30       372
           0       0.64      0.80      0.71      1037
           1       0.43      0.38      0.40       492

    accuracy                           0.58      1901
   macro avg       0.51      0.47      0.47      1901
weighted avg       0.55      0.58      0.55      1901

array([0, 0, 1, ..., 0, 1, 1])


from datetime import datetime
feature_list = ['FastStochastics','WilliamsR', 'SlowStochastics', 
                'Ex-postVolatility', 'short_signal','long_signal']
all_days = list(X.index)
start_backtest = 4500 #起始回测时间
end_backtest = 4600

#start_day = all_days.index(start_month)
day_length = len(all_days)

print(all_days[start_backtest])

2014-08-04 00:00:00


## 不同滑动窗口长度， 不同最大深度 做CV，
# 设定全局变量
rolling_windows = [40, 60 ,120, 250]
refreshFreq = 1 #可以设定多长时间更新一次模型，这里设定为1天，在运行大样本的时候考虑算力问题可以变成5左右以做算力上的妥协
hyperparameters = { 
    'clf__max_depth': (3, 4, 5, 6) 
}  
def DecissionTreeClf(featureList, prob = False):
    X = asset[feature_list][start_index:]
    Y = asset['label'][start_index:]
    reports = []
    for rw in rolling_windows:
        report = pd.DataFrame(index = X.index[start_backtest:])
        report['Correct'] = 0 #预测正确率
        report['Signal'] = 0 #策略信号
        report['Return'] = 0 #策略信号
        for day in range(start_backtest, end_backtest, refreshFreq):
            pipeline = Pipeline([('clf',tree.DecisionTreeClassifier(criterion='gini'))])
            clf = GridSearchCV(pipeline, hyperparameters, cv=5, n_jobs=1,verbose=1)
            train_days = all_days[day-rw:day]
            test_days = all_days[day:day+refreshFreq]
            X_train, Y_train = X.loc[train_days], Y.loc[train_days] 
            clf.fit(X_train, Y_train)
            prediction = clf.predict(X.loc[test_days])
            report.loc[test_days,'Correct'] = (prediction == Y.loc[test_days])
            report.loc[test_days, 'Signal'] = prediction
            if prob:
                prediction_prob = clf.predict_proba(X.loc[test_days])
                #print(prediction_prob)
                report.loc[test_days, 'Prob'] = prediction_prob[0][0] - prediction_prob[0][2]
        reports.append(report)
        print("===========")
        print("Finish Rolling Window {}".format(rw))
    return reports

reports = DecissionTreeClf(feature_list)

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits


best_report = None
best_pre = 0 
for report in reports:
    corRate = 100*(report['Correct'] == 1).sum()/(end_backtest-start_backtest)
    print("预测正确率: {}%".format(corRate))
    print("信号个数: ", (report['Signal'] != 0).sum() )
    if corRate > best_pre:
        best_report = report
        best_pre = corRate

预测正确率: 80.0%
信号个数:  55
预测正确率: 76.0%
信号个数:  61
预测正确率: 74.0%
信号个数:  53
预测正确率: 61.0%
信号个数:  32


def backtest(report, initial_capital = 5*10**6):
    report = report[:(end_backtest-start_backtest) + 20]
    report.iloc[(end_backtest-start_backtest)+1:,1] = 0
    report.loc[:,'Signal2'] = report['Signal'].shift(20).fillna(0.0)
    report.loc[:, 'Signal'] = report['Signal'] - report['Signal2'] 
    #initial_capital = 10**6
    report.loc[:,'Signal'] = report['Signal'] * 100
    #print(report)
    portfolio = pd.DataFrame(index = report.index).fillna(0.0)
    portfolio.loc[:,'change'] = report['Signal'].multiply(asset['close'], axis=0)
    portfolio.loc[:,'position'] = report['Signal'].cumsum()
    portfolio.loc[:,'holdings'] = portfolio['position'].multiply(asset['close'], axis = 0)
    portfolio.loc[:,'cash'] = initial_capital - portfolio['change'].cumsum()
    portfolio.loc[:,'total'] = portfolio['cash'] + portfolio['holdings']
    portfolio.iloc[:(end_backtest-start_backtest) + 20,:].total.plot()
    plt.title("Capital Line", fontsize = 15)
    portfolio.loc[:,['holdings','cash']][:(end_backtest-start_backtest) + 20].plot()
    plt.title("Holdings and Cash", fontsize = 15)
    return portfolio
portfolio = backtest(reports[0])
#plt.figure()
#asset.loc['2012-07':'2013-02', 'close'].plot()
portfolio.tail()

C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:1765: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  isetter(loc, value)
C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:1596: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  self.obj[key] = _infer_fill_value(value)
C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:1745: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  isetter(ilocs[0], value)


def price_momentum(data, n):
    return data / data.shift(n)
def acceleration_between_times(data, n):
    mo = price_momentum(data, n)
    return mo - mo.shift(n)
asset['Momentum'] = price_momentum(asset['close'],20)
asset['Acc_Between_Times'] = acceleration_between_times(asset['close'], 20)
asset.tail()


asset['ado'] = (asset['high'] - asset['open'] + asset['close'] - asset['low']) * 50 / (
    asset['high'] - asset['low'])


window = 20 #月频
delta = asset['close'].pct_change()

U = delta.copy()
U[delta<=0]=0.0
D = abs(delta.copy())
D[delta>0]=0.0
U_average = U.rolling(window).mean()
D_average = D.rolling(window).mean()
asset['RSI']= 100-100/(1+U_average/D_average)


def positive_volume_index(asset):
    PVI = asset['close']*0
    PVI[0] = 100          #初值为100
    length = len(PVI)
    for i in range(1,length):
        if asset['vol'][i] < asset['vol'][i-1]:
            PVI[i] = PVI[i-1]
        if asset['vol'][i] >= asset['vol'][i-1]:
            PVI[i] = PVI[i-1]*(1+(asset['close'][i]-asset['close'][i-1])/asset['close'][i-1])
    return PVI
asset['PVI'] = positive_volume_index(asset)


def negative_volume_index(asset):
    NVI = asset["close"]*0
    NVI[0] = 100          #初值为100
    length = len(NVI)
    for i in range(1,length):
        if asset['vol'][i] > asset['vol'][i-1]:
            NVI[i] = NVI[i-1]
        if asset['vol'][i] <= asset['vol'][i-1]:
            NVI[i] = NVI[i-1]*(1+(asset['close'][i]-asset['close'][i-1])/asset["close"][i-1])
    return NVI
asset['NVI'] = negative_volume_index(asset)


import numpy as np
import matplotlib.pylab as plt
import pandas as pd
%matplotlib inline


stocks = pd.read_excel("sh000001.xlsx",index_col=0,parse_dates=True).dropna(axis = 1)
stocks.head()#请改为stocks=asset


dPVT=(stocks["close"]-stocks["close"].shift(1))/stocks["close"].shift(1)*stocks["vol"]
dPVT[[0]]=0
PVT=dPVT.copy()
for i in range(len(PVT)-1):
    PVT[[i+1]]=PVT[[i]].values+dPVT.shift(-1)[[i]].values
PVT.plot()

<AxesSubplot:xlabel='date'>


OBV=stocks["vol"]*0
for i in range(len(OBV)-1):
    tmp1=stocks["close"].shift(-1)[[i]]
    tmp2=stocks["close"][[i]]
    if((tmp1>tmp2).bool()):
        OBV[[i+1]]=OBV[[i]].values+stocks["vol"].shift(-1)[[i]].values
    if((tmp1<tmp2).bool()):
        OBV[[i+1]]=OBV[[i]].values-stocks["vol"].shift(-1)[[i]].values
    if((tmp1==tmp2).bool()):
        OBV[[i+1]]=OBV[[i]].values
OBV.plot()

<AxesSubplot:xlabel='date'>


N=70
PROC=100*(stocks["close"].shift(N)/stocks["close"]-1)
PROC=PROC.dropna()

VROC=100*(stocks["vol"].shift(N)/stocks["vol"]-1)
VROC=VROC.dropna()

PROC.plot()

<AxesSubplot:xlabel='date'>


VROC.plot()

<AxesSubplot:xlabel='date'>


ADL=(2*stocks["close"]-stocks["low"]-stocks["high"])/(stocks["high"]-stocks["low"])*stocks["vol"]
ADL.plot()

<AxesSubplot:xlabel='date'>


WADL=stocks["vol"]*0
for i in range(len(WADL)-1):
    if((stocks["close"].shift(-1)[[i]]>stocks["close"][[i]]).bool()):
        WADL[[i+1]]=WADL[[i]].values+stocks["close"].shift(-1)[[i]].values-min(stocks["low"].shift(-1)[[i]].item(),stocks["close"][[i]].item())
    if((stocks["close"].shift(-1)[[i]]<stocks["close"][[i]]).bool()):
        WADL[[i+1]]=WADL[[i]].values+stocks["close"].shift(-1)[[i]].values-max(stocks["high"].shift(-1)[[i]].item(),stocks["close"][[i]].item())
    if((stocks["close"].shift(-1)[[i]]==stocks["close"][[i]]).bool()):
        WADL[[i+1]]=WADL[[i]].values
WADL.plot()

<AxesSubplot:xlabel='date'>


from sklearn import tree
%matplotlib inline
from IPython.display import Image
import pydotplus
from sklearn.datasets import load_iris
import os


stocks['month_return'] = stocks['close'].shift(-20)/stocks['close']-1
stocks['label'] = 1 * (stocks['month_return'] > 0.05) + (-1) * (stocks['month_return'] < - 0.05)

short_window = 5
mid_window = 20
long_window = 250
signals = pd.DataFrame(index = stocks.index)

stocks['short_signal'] = 0
stocks['long_signal'] = 0
signals['short_mavg'] = stocks['close'].rolling(window=short_window, min_periods=1, center=False).mean()
signals['mid_mavg'] = stocks['close'].rolling(window=mid_window, min_periods=1, center=False).mean()
signals['long_mavg'] = stocks['close'].rolling(window=long_window, min_periods=1, center=False).mean()

stocks['short_signal'].iloc[short_window:] = np.where(signals['short_mavg'][short_window:] > signals['mid_mavg'][short_window:], 1.0, 0.0)
stocks['long_signal'].iloc[mid_window:] = np.where(signals['mid_mavg'][mid_window:] > signals['long_mavg'][mid_window:], 1.0, 0.0)
stocks.head()

C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:670: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  iloc._setitem_with_indexer(indexer, value)
C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:670: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  iloc._setitem_with_indexer(indexer, value)


stocks['PVT']=PVT
stocks['OBV']=OBV
stocks['PROC']=PROC
stocks['VROC']=VROC
stocks['ADL']=ADL
stocks['WADL']=WADL

stocks_plus=stocks.dropna()
stocks_plus.head()


asset['PVT']=PVT
asset['OBV']=OBV
asset['PROC']=PROC
asset['VROC']=VROC
asset['ADL']=ADL
asset['WADL']=WADL


print(stocks_plus['label'][:'2019'])
print(stocks_plus[['PVT','OBV','PROC','VROC','ADL','WADL','short_signal','long_signal']][:'2019'])

date
1991-04-01    0
1991-04-02   -1
1991-04-03   -1
1991-04-04   -1
1991-04-05   -1
             ..
2019-09-04    0
2019-09-05    0
2019-09-06    0
2019-09-09    0
2019-09-10    0
Name: label, Length: 6956, dtype: int32
                     PVT           OBV       PROC       VROC           ADL  \
date                                                                         
1991-04-01  1.847010e+04  4.095700e+06 -17.170546 -89.499125  1.199900e+06   
1991-04-02  1.879453e+04  4.177300e+06 -13.876743 -75.857843  5.828571e+04   
1991-04-03  1.964553e+04  4.383600e+06 -10.336045 -98.642753  2.063000e+05   
1991-04-04  1.973194e+04  5.435300e+06  -5.890569 -99.695731  0.000000e+00   
1991-04-05  1.801815e+04  4.276400e+06  -1.061379 -99.870567  5.297829e+05   
...                  ...           ...        ...        ...           ...   
2019-09-04  5.712950e+10  6.541108e+12  -1.606135  -0.969128  2.254959e+10   
2019-09-05  5.742333e+10  6.571652e+12  -2.383233 -34.861292 -1.186953e+10   
2019-09-06  5.752310e+10  6.593334e+12  -3.126750  -5.104213  2.087867e+10   
2019-09-09  5.772708e+10  6.617672e+12  -4.166970 -19.782897  2.078481e+10   
2019-09-10  5.769899e+10  6.593671e+12  -4.339997 -10.028372  1.046992e+10   

                WADL  short_signal  long_signal  
date                                             
1991-04-01     23.70           0.0          0.0  
1991-04-02     24.18           0.0          0.0  
1991-04-03     24.68           0.0          0.0  
1991-04-04     25.16           0.0          0.0  
1991-04-05     24.97           0.0          0.0  
...              ...           ...          ...  
2019-09-04  16008.02           1.0          1.0  
2019-09-05  16036.47           1.0          1.0  
2019-09-06  16054.47           1.0          1.0  
2019-09-09  16079.61           1.0          1.0  
2019-09-10  16073.40           1.0          1.0  

[6956 rows x 8 columns]


clf = tree.DecisionTreeClassifier(max_depth = 6)
clf = clf.fit(stocks_plus[['PVT','OBV','PROC','VROC','ADL','WADL','short_signal','long_signal']][:'2019'], stocks_plus['label'][:'2019'])
dot_data = tree.export_graphviz(clf, out_file=None,
                                feature_names=['PVT','OBV','PROC','VROC','ADL','WADL','short_signal','long_signal'],
                                class_names=['-1','0','1'],
                                filled=True, rounded=True,
                                special_characters=True)


graph = pydotplus.graph_from_dot_data(dot_data)  
graph.write_png("stocks_DecisionTree.png")
Image(graph.create_png())


clf.score(stocks[['PVT','OBV','PROC','VROC','ADL','WADL','short_signal','long_signal']]['2019':], stocks['label']['2019':])

0.7526315789473684


# 设定全局变量
rolling_windows = [40]
refreshFreq = 10 #可以设定多长时间更新一次模型，这里设定为1天，在运行大样本的时候考虑算力问题可以变成5左右
hyperparameters = { 
    'clf__max_depth': (3, 4, 5, 6) 
}


print(asset.columns)

Index(['open', 'high', 'low', 'close', 'amount', 'vol', 'FastStochastics',
       'WilliamsR', 'SlowStochastics', 'month_return', 'label',
       'Ex-postVolatility', 'short_signal', 'long_signal', 'Momentum',
       'Acc_Between_Times', 'ado', 'RSI', 'PVI', 'NVI', 'PVT', 'OBV', 'PROC',
       'VROC', 'ADL', 'WADL'],
      dtype='object')


Mom_featureList = ['Momentum', 'Acc_Between_Times', 'ado']
Sto_featureList = ['FastStochastics','WilliamsR', 'SlowStochastics', 'Ex-postVolatility']
Ind_featureList = ['RSI','PVI','NVI']
Indicators_fearureList=['PVT','OBV','PROC','VROC','ADL','WADL']
Ave_featureList = ['short_signal','long_signal']
feature_list = Mom_featureList + Sto_featureList + Ind_featureList + Ave_featureList + Indicators_fearureList

reportsMom = DecissionTreeClf(Mom_featureList+Ave_featureList)
reportsSto = DecissionTreeClf(Sto_featureList+Ave_featureList)
reportsInd = DecissionTreeClf(Ave_featureList+Ave_featureList)
reportsIndicator = DecissionTreeClf(Indicators_fearureList+Ave_featureList)

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
===========
Finish Rolling Window 40
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
===========
Finish Rolling Window 40
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits


#backtest(reportsMom[0])
#print("预测正确率: {}%".format((reportsMom[0].Correct == True).sum())) 
## 不同滑动窗口长度， 不同最大深度 做CV，
# 设定全局变量
rolling_windows = [40, 60 ,120, 250]
refreshFreq = 1 #可以设定多长时间更新一次模型，这里设定为1天，在运行大样本的时候考虑算力问题可以变成5左右以做算力上的妥协
hyperparameters = { 
    'clf__max_depth': (3, 4, 5, 6) 
} 

def DecissionTreeClf(featureList, prob = False):
    X = asset[feature_list][start_index:]
    Y = asset['label'][start_index:]
    reports = []
    for rw in rolling_windows:
        report = pd.DataFrame(index = X.index[start_backtest:])
        report['Correct'] = 0 #预测正确率
        report['Signal'] = 0 #策略信号
        report['Return'] = 0 #策略信号
        for day in range(start_backtest, end_backtest, refreshFreq):
            pipeline = Pipeline([('clf',tree.DecisionTreeClassifier(criterion='gini'))])
            clf = GridSearchCV(pipeline, hyperparameters, cv=5, n_jobs=1,verbose=1)
            train_days = all_days[day-rw:day]
            test_days = all_days[day:day+refreshFreq]
            X_train, Y_train = X.loc[train_days], Y.loc[train_days] 
            clf.fit(X_train, Y_train)
            prediction = clf.predict(X.loc[test_days])
            report.loc[test_days,'Correct'] = (prediction == Y.loc[test_days])
            report.loc[test_days, 'Signal'] = prediction
            if prob:
                prediction_prob = clf.predict_proba(X.loc[test_days])
                print(prediction_prob)
                report.loc[test_days, 'Prob'] = prediction_prob[0][0] - prediction_prob[0][2]
        reports.append(report)
        print("===========")
        print("Finish Rolling Window {}".format(rw))
    return reports


reportsTotal = DecissionTreeClf(Mom_featureList+Sto_featureList+Ind_featureList+Ave_featureList+Indicators_fearureList)

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits

Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits
Fitting 5 folds for each of 4 candidates, totalling 20 fits


backtest(reportsTotal[0])
print("预测正确率: {}%".format((reportsTotal[0].Correct == True).sum()))

C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:1765: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  isetter(loc, value)
C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:1596: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  self.obj[key] = _infer_fill_value(value)
C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:1745: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  isetter(ilocs[0], value)

预测正确率: 81%


from sklearn.ensemble import AdaBoostClassifier


X = asset[feature_list][start_index:]
Y = asset['label'][start_index:]
hyperAda = {
    'Ada__n_estimators': (10, 30, 60, 100, 200)
}
reports = []
for rw in rolling_windows:
    report = pd.DataFrame(index=X.index[start_backtest:])
    report['Correct'] = 0  # 预测正确率
    report['Signal'] = 0  # 策略信号
    report['Return'] = 0  # 策略信号
    for day in range(start_backtest, end_backtest, refreshFreq):
        pipeline = Pipeline(
            [('Ada', AdaBoostClassifier(tree.DecisionTreeClassifier(max_depth=1)))])
        clf = GridSearchCV(pipeline, hyperAda, cv=5, n_jobs=1, verbose=1)
        train_days = all_days[day-rw:day]
        test_days = all_days[day:day+refreshFreq]
        X_train, Y_train = X.loc[train_days], Y.loc[train_days]
        clf.fit(X_train, Y_train)
        prediction = clf.predict(X.loc[test_days])
        report.loc[test_days, 'Correct'] = (prediction == Y.loc[test_days])
        report.loc[test_days, 'Signal'] = prediction
    reports.append(report)
    print("===========")
    print("Finish Rolling Window {}".format(rw))

Fitting 5 folds for each of 5 candidates, totalling 25 fits

Fitting 5 folds for each of 5 candidates, totalling 25 fits

Fitting 5 folds for each of 5 candidates, totalling 25 fits

Fitting 5 folds for each of 5 candidates, totalling 25 fits

Fitting 5 folds for each of 5 candidates, totalling 25 fits

Fitting 5 folds for each of 5 candidates, totalling 25 fits

Fitting 5 folds for each of 5 candidates, totalling 25 fits

Fitting 5 folds for each of 5 candidates, totalling 25 fits

Fitting 5 folds for each of 5 candidates, totalling 25 fits

Fitting 5 folds for each of 5 candidates, totalling 25 fits


print("AdaBoost 样本外预测正确率: {}%".format((reports[0].Correct == True).sum())) 
portfolioAda = backtest(reports[0])
#plt.figure()
#asset.loc['2012-07':'2013-02', 'close'].plot()
portfolioAda.tail()

AdaBoost 样本外预测正确率: 83%

C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:1765: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  isetter(loc, value)
C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:1596: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  self.obj[key] = _infer_fill_value(value)
C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:1745: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  isetter(ilocs[0], value)


from sklearn.ensemble import RandomForestClassifier

hyperRF = {
    'RF__n_estimators': (100, 50, 30)
}
reports = []
for rw in rolling_windows:
    report = pd.DataFrame(index=X.index[start_backtest:])
    report['Correct'] = 0  # 预测正确率
    report['Signal'] = 0  # 策略信号
    report['Return'] = 0  # 策略信号
    for day in range(start_backtest, end_backtest, refreshFreq):
        pipeline = Pipeline(
            [('RF', RandomForestClassifier)])
        #clf = GridSearchCV(pipeline, hyperRF, cv=5, n_jobs=1, verbose=1)
        clf = RandomForestClassifier()
        train_days = all_days[day-rw:day]
        test_days = all_days[day:day+refreshFreq]
        X_train, Y_train = X.loc[train_days], Y.loc[train_days]
        #print(Y_train)
        clf.fit(X_train.values, Y_train.values)
        prediction = clf.predict(X.loc[test_days])
        report.loc[test_days, 'Correct'] = (prediction == Y.loc[test_days])
        report.loc[test_days, 'Signal'] = prediction
    reports.append(report)
    print("===========")
    print("Finish Rolling Window {}".format(rw))

===========
Finish Rolling Window 40
===========
Finish Rolling Window 60
===========
Finish Rolling Window 120
===========
Finish Rolling Window 250


print("AdaBoost 样本外预测正确率: {}%".format((reports[0].Correct == True).sum())) 
portfolioRF = backtest(reports[0])
#plt.figure()
#asset.loc['2012-07':'2013-02', 'close'].plot()
portfolioRF.tail()

AdaBoost 样本外预测正确率: 82%

C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:1765: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  isetter(loc, value)
C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:1596: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  self.obj[key] = _infer_fill_value(value)
C:\Users\kyjma\anaconda3\lib\site-packages\pandas\core\indexing.py:1745: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  isetter(ilocs[0], value)

Name of technical indicator:	Type of technical indicator:
Momentum	Oscillators
Accumulation Distribution Oscillator	Oscillators
Fast Stochastics %K	Stochastics
Median Price	Indicators
Fast Stochastics %D	Stochastics
Negative Volume Index	Indexes
Highest High	Indicators
Positive Volume Index	Indexes
Lowest Low	Indicators
Slow Stochastics %K	Stochastics
Price and Volume Trend	Indicators
Slow Stochastics %D	Stochastics
Accumulation Distribution Line	Oscillators
Acceleration Between Times	Oscillators
Relative Strength Index	Indexes
Bollinger Bands	Indicators
Volume Rate of Change	Indicators
PercentB	Indicators
Price Rate of Change	Indicators
Bandwidth	Indicators
On-Balance Volume	Indicators
Volatility	Stochastics
Chaikin Volatility	Stochastics
William’s % R	Stochastics
William’s Accumulation Distribution Line	Indicators

	open	high	low	close	amount	vol	FastStochastics	WilliamsR	SlowStochastics	month_return	label
date
1996-01-02	550.26	550.26	537.38	537.87	223.12	48247800	0.462351	-5.793546	1.497209	-0.009854	0
1996-01-03	535.23	542.74	530.79	542.42	252.57	55619200	10.331349	-13.262859	4.500095	-0.009347	0
1996-01-04	541.94	558.94	539.76	558.76	451.49	110591300	24.846762	-10.233632	11.880154	-0.039516	0
1996-01-05	559.03	561.36	536.12	536.37	597.19	138637100	4.956916	-5.845252	13.378342	-0.000615	0
1996-01-08	533.33	540.04	527.91	539.17	287.29	69620700	10.109535	-17.983480	13.304404	-0.030269	0

	short_mavg	mid_mavg	long_mavg
date
1990-12-19	100.000000	100.000000	100.000000
1990-12-20	102.195000	102.195000	102.195000
1990-12-21	104.506667	104.506667	104.506667
1990-12-24	107.017500	107.017500	107.017500
1990-12-25	109.664000	109.664000	109.664000
...	...	...	...
2019-10-10	2924.700000	2977.107500	2835.175720
2019-10-11	2932.998000	2976.497500	2835.945800
2019-10-14	2953.536000	2976.911500	2836.750080
2019-10-15	2969.032000	2975.227000	2837.547160
2019-10-16	2979.802000	2973.102500	2838.176600

	change	position	holdings	cash	total
date
2015-01-22	-334334.0	400.0	1337336.0	5369227.0	6706563.0
2015-01-23	-335176.0	300.0	1005528.0	5704403.0	6709931.0
2015-01-26	-338318.0	200.0	676636.0	6042721.0	6719357.0
2015-01-27	-335296.0	100.0	335296.0	6378017.0	6713313.0
2015-01-28	-330574.0	0.0	0.0	6708591.0	6708591.0

	open	high	low	close	amount	vol	FastStochastics	WilliamsR	SlowStochastics	month_return	label	Ex-postVolatility	short_signal	long_signal	Momentum	Acc_Between_Times
date
2019-09-24	2979.48	3002.90	2973.76	2985.34	206245.81	16390276800	65.888764	-83.361962	71.018973	NaN	0	0.010400	1.0	1.0	1.042524	0.068855
2019-09-25	2977.67	2977.67	2955.43	2955.43	203782.75	16854313600	48.172718	-88.355150	60.021286	NaN	0	0.009240	1.0	1.0	1.018345	0.035331
2019-09-26	2964.48	2970.04	2928.26	2929.09	216277.57	18893254400	32.571225	-90.037316	48.877569	NaN	0	0.008638	0.0	1.0	1.012209	0.025423
2019-09-27	2929.49	2939.08	2920.93	2932.17	153311.01	13290577600	34.395546	-92.809335	38.379830	NaN	0	0.006694	0.0	1.0	1.014269	0.020405
2019-09-30	2927.92	2936.48	2905.19	2905.19	142591.71	11664680800	13.507064	-74.612245	26.824612	NaN	0	0.005715	0.0	1.0	1.006566	0.000150

Machine Learning Strategy [CHI]

金融数据分析大作业 금융데이터분석 수업 프로젝트

Machine Learning And Strategy¶

资产价格走势和收益率分布¶

夏普比和最大回撤¶

移动平均线策略及其评估¶

Back-testing¶

分析¶

Stochastics¶

Strategy¶

训练流程¶

BackTesting¶

增加新指标¶

Indicators¶

Price and Volume Trend¶

On-Balance Volume¶

Volume and Price Rate of Change¶

Accumulation Distribution Line¶

William’s Accumulation Distribution Line¶

Decision Tree¶

分别决策并投票¶

所有指标集合在一起建立决策树¶

AdaBoost¶

Random Forest¶

	000001.SZ	000002.SZ	000004.SZ	000005.SZ	000006.SZ	000007.SZ	000008.SZ	000009.SZ	000010.SZ	000011.SZ	...	688020.SH	688022.SH	688028.SH	688029.SH	688033.SH	688066.SH	688088.SH	688122.SH	688333.SH	688388.SH
1990-12-19	NaN	NaN	NaN	3.83	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
1990-12-20	NaN	NaN	NaN	3.83	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
1990-12-21	NaN	NaN	NaN	3.83	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
1990-12-24	NaN	NaN	NaN	3.83	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN
1990-12-25	NaN	NaN	NaN	3.83	NaN	NaN	NaN	NaN	NaN	NaN	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN

	Value
2019-10-25	1688.0
2019-10-28	1666.0
2019-10-29	1691.0
2019-10-30	1643.0
2019-10-31	1626.0

	open	high	low	close	amount	vol	FastStochastics	WilliamsR	SlowStochastics	month_return	label	Ex-postVolatility	short_signal	long_signal
date
1990-12-19	96.05	100.00	95.79	100.00	0.49	126000	NaN	NaN	NaN	0.342500	1	NaN	0.0	0.0
1990-12-20	104.30	104.39	99.98	104.39	0.09	19700	NaN	NaN	NaN	0.285947	1	NaN	0.0	0.0
1990-12-21	109.07	109.13	103.73	109.13	0.02	2800	NaN	NaN	NaN	0.230093	1	NaN	0.0	0.0
1990-12-24	113.57	114.55	109.13	114.55	0.03	3200	NaN	NaN	NaN	0.167351	1	NaN	0.0	0.0
1990-12-25	120.09	120.25	114.55	120.25	0.01	1500	NaN	NaN	NaN	0.107443	1	NaN	0.0	0.0

	open	high	low	close	amount	vol	month_return	label	short_signal	long_signal	PVT	OBV	PROC	VROC	ADL	WADL
date
1991-04-01	120.69	120.73	119.68	120.73	5.67	1199900	-0.049532	0	0.0	0.0	18470.101449	4095700.0	-17.170546	-89.499125	1.199900e+06	23.70
1991-04-02	121.21	121.29	120.73	121.21	0.39	81600	-0.059979	-1	0.0	0.0	18794.527855	4177300.0	-13.876743	-75.857843	5.828571e+04	24.18
1991-04-03	121.71	121.71	121.21	121.71	0.98	206300	-0.070249	-1	0.0	0.0	19645.530247	4383600.0	-10.336045	-98.642753	2.063000e+05	24.68
1991-04-04	122.20	122.20	121.24	121.72	5.05	1051700	-0.076487	-1	0.0	0.0	19731.940567	5435300.0	-5.890569	-99.695731	0.000000e+00	25.16
1991-04-05	121.07	121.73	121.03	121.54	6.04	1158900	-0.081701	-1	0.0	0.0	18018.154829	4276400.0	-1.061379	-99.870567	5.297829e+05	24.97