def _reg_rolling_APM(reg_: pd.DataFrame,
x1: str,
y1: str,
x2: str,
y2: str,
has_const: bool = False,
use_const: bool = True,
window: int = 20) -> pd.Series:
# print(reg_.index[0])
if len(reg_) <= window:
res = pd.Series(index=reg_.index)
else:
reg_object_am = PandasRollingOLS(x=reg_[x1],
y=reg_[y1],
has_const=has_const,
use_const=use_const,
window=window)
reg_object_pm = PandasRollingOLS(x=reg_[x2],
y=reg_[y2],
has_const=has_const,
use_const=use_const,
window=window)
diff_resids = reg_object_am._resids - reg_object_pm._resids
stat = np.nanmean(diff_resids, axis=1) / np.nanstd(
diff_resids, axis=1, ddof=1) * np.sqrt(window)
res = pd.Series(stat, index=reg_object_am.index[window - 1:])
return res
def velocity(df):
vel_xz = PandasRollingOLS(y=df.lws_xz, x=df.td, window=7)
df['vel_xz'] = ([np.nan] * 6) + list(abs(vel_xz.beta.values))
vel_xy = PandasRollingOLS(y=df.lws_xy, x=df.td, window=7)
df['vel_xy'] = ([np.nan] * 6) + list(abs(vel_xy.beta.values))
return df
def node_inst_vel(df, roll_window_numpts, start):
lr_xz = PandasRollingOLS(y=df.xz, x=df.td, window=roll_window_numpts).beta
lr_xy = PandasRollingOLS(y=df.xy, x=df.td, window=roll_window_numpts).beta
df = df.loc[df.index >= start]
vel_xz = lr_xz[lr_xz.index >= start]['feature1'].values
vel_xy = lr_xy[lr_xy.index >= start]['feature1'].values
df.loc[:, 'vel_xz'] = np.round(vel_xz, 4)
df.loc[:, 'vel_xy'] = np.round(vel_xy, 4)
return df
def Spread_RollingRegression(df,pair1,pair2):
spread=pd.DataFrame()
#calculate hedge ratio using Rolling Regression Function.
rolling_ols=PandasRollingOLS(df[pair1],df[pair2],window=20)
spread['{}_{}'.format(pair1,pair2)]=df[pair1]-rolling_ols.beta['feature1']*df[pair2]
spread.dropna(inplace=True)
return(spread)
def slope_intercept(df, roll_window_numpts, start):
slope = PandasRollingOLS(y=df.magnitude, x=df.td, window=7).beta
intercept = PandasRollingOLS(y=df.magnitude, x=df.td, window=7).alpha
inter = pd.DataFrame(intercept, columns=['intercept'])
df = df.loc[df.index >= start]
m = slope[slope.index >= start]['feature1'].values
df.loc[:, 'slope'] = np.round(m, 4)
b = inter[inter.index >= start]['intercept'].values
df.loc[:, 'intercept'] = np.round(b, 4)
return (df)
def RegBeta(factor, x, y, num, name):
temp = copy.deepcopy(factor[['trade_date', 'ts_code', x, y]])
temp.sort_values(by=['ts_code', 'trade_date'], inplace=True)
res = PandasRollingOLS(temp[x], temp[y], num)
factor[name] = res.beta
index = factor.groupby('ts_code').head(num - 1).index
factor.loc[index, name] = np.nan
return factor
def acceleration(vel_,td):
start_index = len(td) - len(vel_)
accel = PandasRollingOLS(y = pd.Series(vel_),
x = pd.Series(td[start_index:]), window = 7)
accel = accel.beta.values
return accel
def get_spread(data):
spread = pd.DataFrame()
XStock = list(data)[0]
YStock = list(data)[1]
pairs = YStock+'-'+XStock
model = PandasRollingOLS(y=data[YStock], x=data[XStock], window=50)
spread[pairs] = data[YStock] - model.beta['feature1']*data[XStock]
spread.dropna(inplace=True)
return spread
def beta_calculation(self):
self.data = self.import_data()
model0 = PandasRollingOLS(y=self.data[self.y_ticker],
x=self.data[self.x_ticker],
window=self.lookback)
self.data = self.data[self.lookback - 1:]
self.data['betas'] = model0.beta
return self.data
def fit(self, x_train, y_train, standardize=False):
self.x_train = x_train
self.y_train = y_train
self.standardize = standardize
if self.standardize:
self.standardizescaler.fit(x_train)
x_train = self.standardizescaler.transform(x_train)
self.regressor = PandasRollingOLS(y=y_train, x=x_train, window=self.window_size, has_const=self.has_const,
use_const=self.use_const)
def accel(df, roll_window_numpts, start):
accel = PandasRollingOLS(y=df.velocity, x=df.td, window=144).beta
df = df.loc[df.index >= start + timedelta(hours=75.5)]
acceleration = accel[accel.index >= start +
timedelta(hours=75.5)]['feature1'].values
df.loc[:, 'acceleration'] = np.round(acceleration, 4)
return (df)
def node_inst_vel(df, roll_window_numpts, start):
lr_xyz = PandasRollingOLS(y=df.magnitude, x=df.td, window=7).beta
df = df.loc[df.index >= start + timedelta(hours=3.5)]
velocity = lr_xyz[lr_xyz.index >= start +
timedelta(hours=3.5)]['feature1'].values
df.loc[:, 'velocity'] = np.round(velocity, 4)
return (df)
def node_inst_vel(filled_smoothened, roll_window_numpts, start):
print (filled_smoothened)
try:
lr_xz = PandasRollingOLS(y=filled_smoothened.xz, x=filled_smoothened.td,
window=roll_window_numpts)
except:
print('Error')
pass
return lr_xz
def _reg_beta(data_x, data_y, window):
window = window[0]
if type(window) != int:
return np.zeros(len(data_x))
data = pd.DataFrame({'x': data_x.flatten(), 'y': data_y.flatten()})
res = PandasRollingOLS(data['x'], data['y'], window)
data['beta'] = res.beta
value = np.array(data['beta'].tolist())
value = np.nan_to_num(value)
return value
def get_spread(self, close_df):
spread = pd.DataFrame()
pair1 = list(close_df)[0]
pair2 = list(close_df)[1]
pairs = pair1 + '-' + pair2
rolling_ols = PandasRollingOLS(y=close_df[pair1],
x=close_df[pair2],
window=48)
spread[pairs] = close_df[
pair1] - rolling_ols.beta['feature1'] * close_df[pair2]
spread.dropna(inplace=True)
return spread
def _ts_beta(df1, df2, window): # 历史滚动回归系数
window = window[0]
if type(window) != int:
return np.zeros(len(df1))
df = pd.DataFrame({'0': df1, '1': df2})
df['time'] = trade_date
df['code'] = stock_code
res = PandasRollingOLS(df['0'], df['1'], window)
df['res'] = res.beta
index = df.groupby('code').head(window - 1).index
df.loc[index, 'res'] = np.nan
value = df['res']
return np.nan_to_num(value.values)
def test_const_false():
# Case where use_const=False and has_const=False
# See Issue # 6
X = pd.DataFrame(np.arange(5), columns=['X'])
Y = pd.DataFrame(np.arange(0, 10, 2) + 1, columns=['Y'])
window = 2
reg_df = pd.concat([Y, X], axis=1)
rr = PandasRollingOLS(
y=reg_df.iloc[:, 0], # Series
x=reg_df.iloc[:, 1:], # DataFrame
window=window,
has_const=False,
use_const=False)
assert np.allclose(rr.beta.values,
np.array([[3.], [2.6], [2.38461538], [2.28]]))
def _reg_rolling(reg_: pd.DataFrame,
x_name: str,
y_name: str,
has_const: bool = False,
use_const: bool = True,
window: int = 20) -> pd.Series:
if len(reg_) <= window:
alpha = pd.Series(index=reg_.index)
else:
reg_object = PandasRollingOLS(x=reg_[x_name],
y=reg_[y_name],
has_const=has_const,
use_const=use_const,
window=window)
alpha = reg_object.alpha
return alpha
def acceleration(vel, td, thresh):
accel = PandasRollingOLS(y=pd.Series(vel),x=pd.Series(td),window=7)
accel = accel.beta.values
# array = array.reshape(len(array))
# vel = []
# m = 2
# for i in range(len(array) - m):
# vel.append(array[i+m])
#
# vel = pd.Series(vel)
# accel = np.array(vel - vel.shift(1))
# accel = abs(accel)
# actual = [1 if i >= (0.0003) else 0 for i in accel]
# actual = np.array(actual)
return accel
def test_datareader_frame():
import os.path
p = os.path.join(os.path.dirname(os.path.abspath(__file__)), "pdr.csv")
data = pd.read_csv(p)
y = data["usd"]
x = data.drop("usd", axis=1)
window = 12 # months
model = PandasRollingOLS(y=y, x=x, window=window)
assert isinstance(model.beta, pd.DataFrame)
assert model.beta.shape == (219, 2)
tgt = np.array([
[3.28409826e-05, -5.42606172e-02],
[2.77474638e-04, -1.88556396e-01],
[2.43179753e-03, -2.94865331e-01],
[2.79584924e-03, -3.34879522e-01],
[2.44759386e-03, -2.41902450e-01],
])
assert np.allclose(model.beta.head().values, tgt)
def _reg_rolling(reg_: pd.DataFrame,
x_name: str,
y_name: str,
win: int,
has_cons: bool = False):
if len(reg_) <= win:
res = pd.Series(index=reg_.index)
else:
try:
X = reg_[x_name]
Y = reg_[y_name]
reg_object = PandasRollingOLS(x=X,
y=Y,
has_const=False,
use_const=has_cons,
window=win)
res = pd.Series(reg_object._resids[:, -1],
index=reg_.index[win - 1:])
except Exception as e:
print(e)
res = pd.Series(index=reg_.index)
return res
def hr_signal(df, stock1, stock2):
window = 20
entry = 2
exit = 0
temp = pd.DataFrame()
temp[stock1] = df[stock1]
temp[stock2] = df[stock2]
model = PandasRollingOLS(y=temp[stock1], x=temp[stock2], window=window)
temp['beta'] = model.beta
temp = temp.dropna()
temp['spread'] = temp[stock1] - temp.beta*temp[stock2]
temp['zscore'] = (temp['spread'] - np.mean(temp['spread']))/np.std(temp['spread'])
temp['longsignal'] = (temp['zscore'] <= -entry) & (temp['zscore'] < -exit)
temp['shortsignal'] = (temp['zscore'] >= entry) & (temp['zscore'] > exit)
temp['closelong'] = (temp['zscore'] >= -exit)
temp['closeshort'] = (temp['zscore'] <= exit)
return temp
def rolling_ndays_ffmodels(ndays=90):
register_matplotlib_converters()
code_list = [
'SPY', 'XLB', 'XLE', 'XLF', 'XLI', 'XLK', 'XLP', 'XLU', 'XLV', 'XLY'
]
for code in code_list:
etf = ETF(code, '2010-01-01', '2019-09-14')
etf.price_acquire()
etf.data['ETF_Daily_return'] = (
etf.data['Close'] / etf.data['Close'].shift(1) - 1)
etf.data['Date'] = etf.data['Date'].apply(
lambda x: x.strftime("%Y%m%d"))
data = pd.DataFrame.merge(etf.data, ff.data, how='left', on='Date')
data = data.dropna(axis=0, how='any')
model = PandasRollingOLS(y=data.ETF_Daily_return,
x=data[['Mkt_RF', 'SMB', 'HML']],
window=ndays)
X = (data.loc[ndays:, 'Date']).reset_index(drop=True)
X = X.apply(lambda x: datetime.datetime.strptime(x, "%Y%m%d"))
plt.plot(X, model.beta)
plt.title('Beta to the Fama-French factors of ETF:' + code + '')
plt.show()
return
def getSignal(index): # 在此处定义因子
df = copy.deepcopy(index)
df['pre_close'] = df['close'].shift(1)
df['HnLOpen'] = (df['open']-df['pre_close'])/df['pre_close']
df['period'] = df['close'].rolling(window=10).mean()
df['MA'] = (df['close']-df['period'])/df['period']
df['DIF'] = df['close'].ewm(min_periods=12,adjust=False,alpha=2/(1+12)).mean()-df['close'].ewm(min_periods=26,adjust=False,alpha=2/(1+26)).mean()
df['DEA'] = df['DIF'].ewm(min_periods=9,adjust=False,alpha=2/(1+9)).mean()
df['macd'] = 2*(df['DIF']-df['DEA'])
#df['macd'] = ta.MACD(df['close'],12,26,9)[2]
df['max'] = df.apply(lambda x: max(x['close']-x['pre_close'],0),axis=1)
df['abs'] = df.apply(lambda x: abs(x['close']-x['pre_close']),axis=1)
df['RSI'] = 100*df['max'].ewm(min_periods=14,adjust=False,alpha=2/(1+14)).mean()/df['abs'].ewm(min_periods=14,adjust=False,alpha=2/(1+14)).mean()
#df['RSI'] = ta.RSI(df['close'],14)
df['ATR'] = (ta.ATR(df['high'],df['low'],df['close'],10))/df['period']
df['ADX'] = ta.ADX(df['high'],df['low'],df['close'],14)
df['buy'] = df.apply(lambda x: max(x['high']-x['pre_close'],0),axis=1)+df['close']-df['low']
df['sell'] = df['buy']+df.apply(lambda x: max(x['pre_close']-x['low'],0),axis=1)+df['high']-df['close']
df['dmkTD'] = df['buy'].rolling(window=9).sum()/df['sell'].rolling(window=9).sum()
df['pre_low'] = df['low'].shift(1)
df['l_max'] = df.apply(lambda x: max(x['low']-x['pre_low'],0),axis=1)
df['l_abs'] = df.apply(lambda x: abs(x['low']-x['pre_low']),axis=1)
df['sma'] = df['l_abs'].rolling(window=3).mean()/df['l_max'].rolling(window=3).mean()*1000
df['ema'] = df['sma'].ewm(min_periods=3,adjust=False,alpha=2/(1+3)).mean()
df['ll'] = df['low'].rolling(window=13).min()
df['hh'] = df['ema'].rolling(window=13).max()
df['turn'] = df.apply(lambda x: (x['ema']+x['hh']*2)/2 if x['low']<=x['ll'] else 0,axis=1)
df['turn'] = df['turn'].ewm(min_periods=3,adjust=False,alpha=2/(1+3)).mean()/618
df['turn'] = df['turn'].apply(lambda x: 500 if x>500 else x)
df['low_n'] = df['low'].rolling(window=9).min()
df['high_n'] = df['high'].rolling(window=9).max()
df['RSV'] = (df['close']-df['low_n'])/(df['high_n']-df['low_n'])*100
n = len(df)
K,D,J = [],[],[]
for i in range(n):
if df['RSV'].isnull().iloc[i]:
K.append(50)
D.append(50)
J.append(np.nan)
else:
K.append(K[-1]*2/3+df['RSV'].iloc[i]/3)
D.append(D[-1]*2/3+K[-1]/3)
J.append(3*K[-1]-2*D[-1])
df['K'] = K
df['D'] = D
df['KDJ'] = J
df['KDJ_1'] = df['KDJ'].shift(1)
df['KDJ_2'] = df['KDJ'].shift(2)
df['turn'] = df.apply(lambda x: x['turn'] if ((x['KDJ']>x['KDJ_1']) and (x['KDJ_2']>x['KDJ_1'])) else 0, axis=1)
df['DAY'] = df.index
up,dw,mid = PandasRollingOLS(df['high'],df['DAY'],15),PandasRollingOLS(df['low'],df['DAY'],15),PandasRollingOLS(df['close'],df['DAY'],15)
df['upA'],df['upB'],df['dwA'],df['dwB'],df['midA'],df['midB'] = up.alpha,up.beta,dw.alpha,dw.beta,mid.alpha,mid.beta
df['err_up'] = (df['high']-df['DAY']*df['upB']-df['upA'])/df['high']
df['err_dw'] = (df['high']-df['DAY']*df['dwB']-df['dwA'])/df['high']
df['err_mid'] = (df['high']-df['DAY']*df['midB']-df['midA'])/df['high']
df['meanHigh'] = (df['close']-df['high'].rolling(window=15).mean())/df['close']
df['meanLow'] = (df['close']-df['low'].rolling(window=15).mean())/df['close']
df['rtn'] = (df['close']-df['pre_close'])/df['pre_close']
df['dwrtn'] = df['rtn'].apply(lambda x: 0 if x>0 else x)
df['negILLIQ'] = df['dwrtn']/df['volume']
df['negILLIQ'] = 10**5*df['negILLIQ'].rolling(window=20).sum()/(df['dwrtn']<0).rolling(window=20).sum()
df['rng'] = (df['high']-df['low'])/df['pre_close']
df['ILLIQ'] = df['rng'].rolling(window=10).mean()
df['cvILLIQ'] = df['rng'].rolling(window=20).std()/df['rng'].rolling(window=20).mean()
df['vol_quantile'] = df['volume'].rolling(window=40).quantile(.75, interpolation='lower')
df['vol_max'] = df.apply(lambda x: x['volume'] if x['volume']>x['vol_quantile'] else x['vol_quantile'],axis=1)
df['vol_increase'] = (df['volume']-df['vol_quantile'])/df['vol_max']
return df[['trade_date','pre_close','HnLOpen','MA','macd','RSI','ATR','ADX','dmkTD','turn',\
'err_up','err_dw','err_mid','meanHigh','meanLow','negILLIQ','cvILLIQ','ILLIQ','vol_increase']]
timestamp = np.array(node.ts.values)
df_td = np.array(node.td.values)
rol_xz = fitting.rolling_window(xz, window=17)
timestamp = fitting.rolling_window(timestamp, window=17)
df_td = fitting.rolling_window(df_td, window=17)
lws_xz, lst_val, ts, n_td = fitting.rolling_lws(rol_xz, timestamp, df_td)
lst_val = np.array(lst_val)
#
# lws = low_ess(xz,np.arange(len(xz)), fraction = 0.1)
# lws = lws[:,1]
#
# lws = pd.Series(lws)
vel = PandasRollingOLS(y=pd.Series(lst_val),x=pd.Series(lst_val),window=7)
vel = abs(vel.beta.values)
vel = vel[0:2000]
td = df_td[0:2000]
# vel = PandasRollingOLS(y=lws, x=node.td,window=7)
# vel = abs(vel.beta.values)
#
# vel = vel[0:20000]
#
accel = acceleration(vel,td,t)
# pred = current_pred(vel,t)
#
# start_index = len(pred) - len(act)
#
# actual.append(act)
# predicted.append(pred[start_index:])
def roll_slope(data0, data1, per):
return PandasRollingOLS(y=data1, x=data0, window=per).beta
def velocity(df):
vel_ = PandasRollingOLS(y=df.lws,x=df.td, window=7)
vel_ = ([np.nan] * 6) + list(abs(vel_.beta.values))
return vel_
'time': time
})
return img_data
if __name__ == '__main__':
dir_path = 'D:\\Users\\yonat\\Desktop\\HUJI\\HUJI Homework\\Advanced Physics Lab A\\Water Heating - Experiment B\\Camera Pics\\8.6.2020\\main exp pics\\'
results_path = 'D:\\Users\\yonat\\Desktop\\HUJI\\HUJI Homework\\Advanced Physics Lab A\\Water Heating - Experiment B\\Results\\8.6.2020\\bubble_pixels_ex1.csv'
experiment_path = 'D:\\Users\\yonat\\Desktop\\HUJI\\HUJI Homework\\Advanced Physics Lab A\\Water Heating - Experiment B\\Results\\8.6.2020\\main_run_data.csv'
exp_data = read_experiment_data(experiment_path, cooling_measurement=True)
# bubbles_data = analyze_all_images_in_dir(dir_path)
bubbles_data = pd.read_csv(results_path)
window_size = 20
exp_data['temp_change_rate'] = PandasRollingOLS(x=exp_data.time, y=exp_data.temp, window=window_size).beta \
.shift(-int(window_size / 2))
fig, axs = plt.subplots(2, sharex=True)
exp_data.plot(x='time',
y='temp_change_rate',
linestyle='None',
marker='.',
grid=True,
ax=axs[0],
label='Temperature rate of change',
c='k')
# exp_data.plot(x='time', y='temp', linestyle='None', marker='.', grid=True, ax=ax)
scale_const = 2e-7
secondary_axis = axs[0].secondary_yaxis(
'right',
data = pd.read_csv(config.input_data_path + '/' + 'EWA EWC' + '.csv',
index_col='Date')
x = data['EWA']
y = data['EWC']
x_ticker = 'EWA'
y_ticker = 'EWC'
k = np.polyfit(x, y, 1)
xx = np.linspace(min(x), max(x), 1000)
yy = np.polyval(k, xx)
lookback = 100
modelo2 = PandasRollingOLS(y=y, x=x, window=lookback)
data = data[lookback - 1:]
betas = modelo2.beta
data['beta'] = betas
data['numunits'] = data.apply(lambda x: x[x_ticker] - x['beta'] * x[y_ticker],
axis=1)
model = smf.OLS(y, x)
results = model.fit()
def cointegration_test(y, x):
ols_result = smf.OLS(y, x).fit()
return ts.adfuller(ols_result.resid, maxlag=1)
def alpha2_genetic_TFZZ(
cls,
data: pd.DataFrame,
high_name: str = PVN.HIGH.value,
close_name: str = PVN.CLOSE.value,
amount_name: str = PVN.AMOUNT.value,
volume_name: str = PVN.VOLUME.value,
adj_factor_name: str = PVN.ADJ_FACTOR.value) -> pd.Series:
"""
alpha2因子来自: <<20200220-天风证券-基于基因表达式规划的价量因子挖掘>>
alpha2计算公式： 𝐴𝑙𝑝ℎ𝑎2: 𝑡𝑠_𝑟𝑒𝑔𝑏𝑒𝑡𝑎(𝑛𝑒𝑔(𝑠_𝑙𝑜𝑔(𝑠𝑢𝑏(𝑑𝑖𝑣(𝑉𝑊𝐴𝑃,𝑃𝑅𝐸𝐶𝐿𝑂𝑆𝐸),1))),
𝑚𝑖𝑛(𝑠𝑢𝑏(𝑑𝑖𝑣(𝐻𝐼𝐺𝐻,𝑃𝑅𝐸𝐶𝐿𝑂𝑆𝐸),1),𝐴𝑀𝑂𝑈𝑁𝑇),20)
VWAP = Amount / Volume: 计算VWAP后需要用复权因子进行调整，否则VWAP与PRECLOSE计算出来的收益率存在跳空现象
价格序列需要进行复权因子调整：因为后续需要进行滚动回归，不进行复权因子调整会出现价格不连续
标准化过程可能会出现最大值等于最小值情况，分母为零，出现无限大，将值换为空值
进行大小比较时，若存在空值则为空
:param data:
:param high_name: 最高价
:param close_name: 收盘价
:param amount_name: 成交额
:param volume_name: 成交量
:param adj_factor_name: 复权因子
:return:
"""
# 设置双重索引并且排序
data.set_index([KN.TRADE_DATE.value, KN.STOCK_ID.value], inplace=True)
data.sort_index(inplace=True)
data['VWAP'] = data[amount_name] / data[volume_name] * data[
adj_factor_name]
# 生成Y
cal_sub1 = data[[close_name, 'VWAP'
]].groupby(KN.STOCK_ID.value, group_keys=False).apply(
lambda x: x['VWAP'] / x[close_name].shift(1) - 1)
cal_sub1 = cal_sub1.droplevel(0)
data['reg_y'] = -np.sign(cal_sub1) * np.log(abs(cal_sub1))
# 生成X
cal_sub2 = data[[high_name, 'VWAP']].groupby(KN.STOCK_ID.value).apply(
lambda x: x[high_name] / x[close_name].shift(1) - 1)
data['return_sta'] = cal_sub2.groupby(KN.TRADE_DATE.value).apply(
lambda x: (x - x.min()) / (x.max() - x.min()))
# 处理无限大值
data[np.isinf(data['return_sta'])] = np.nan
data['volume_sta'] = data[amount_name].groupby(
KN.TRADE_DATE.value).apply(lambda x: (x - x.min()) /
(x.max() - x.min()))
# 处理无限大值
data[np.isinf(data['volume_sta'])] = np.nan
data['reg_x'] = data[['return_sta', 'volume_sta']].min(axis=1,
skipna=False)
# 滚动回归
result = data[['reg_x', 'reg_y']].groupby(
KN.TRADE_DATE.value, group_keys=False).apply(lambda x: pd.Series(
index=x.index) if len(x) < 20 else PandasRollingOLS(
x=x['reg_x'], y=x['reg_y'], window=20).beta['feature1'])
result.name = sys._getframe().f_code.co_name
return result
