Tele2 = ts.Telescope()
Tele2.load_dfs([r"Z:\KaiData\Theo\2019\RTY.ESTheo.mpk"])
START = datetime.datetime(2019, 4, 1, 0, 0)
END = datetime.datetime(2019, 7, 22, 23, 59)
Tele2.choose_range(START, END)
Tele2.parse_datetime()
date_list = Tele2.df['date'].unique()
trade_day_list = []
not_trade_day_list = []
for date in date_list:
if ts.select_date_time(Tele2.df, date,
datetime.time(8, 30, 0))['TheoPrice'].empty:
not_trade_day_list.append(date)
else:
trade_day_list.append(date)
test_day_list = trade_day_list[4:]
print(trade_day_list)
Tele2.df = Tele2.df.resample('5S', fill_method='bfill')
tm.apply_all_metrics(Tele2.df, sma_period=600)
Tele2.parse_datetime()
Tele2.clip_time(datetime.time(8, 30, 0), datetime.time(12, 0, 0))
date_list = []
def rolling_pairwise_theo_diff_sum_profit(corr_df, data_mpk_file_path, start_date, end_date, start_time = datetime.time(8, 0, 0), end_time = datetime.time(15, 15, 0), time_frame_length = 5, rolling_step = '1T', corr_threshold = 3, up_threshold = 100, down_threshold = 100, trade_cost = 30, output_excel_path= r"C:\Users\achen\Desktop\Monocular\rolling_pairwise_theo_diff_corr.xlsx"):
import numpy as np
import pandas as pd
import telescope as ts
import telescope_metrics as tm
import datetime
from scipy.stats.stats import pearsonr
from tqdm import tqdm
#--------------Input Parameters---------------
CORR_DF = corr_df
df_2019 = data_mpk_file_path
START = start_date
END = end_date
STIME = start_time
ETIME = end_time
TIME_FRAME_LENGTH = time_frame_length
ROLLING_STEP = rolling_step #(has to be a factor of TIME_FRAME_LENGTH)
UP_THRESHOLD = up_threshold
DOWN_THRESHOLD = down_threshold
TRADE_COST = trade_cost
CORR_THRESHOLD = corr_threshold
#---------------------------------------------
#--------------Input Parameters---------------
#df_2019 = r"Z:\KaiData\Theo\2019\YM.ESTheo.mpk"
#START = datetime.datetime(2019, 7, 8, 0, 0)
#END = datetime.datetime(2019, 7, 12, 23, 59)
#STIME = datetime.time(8, 0, 0)
#ETIME = datetime.time(15, 15, 0)
#TIME_FRAME_LENGTH = 5
#ROLLING_STEP = '1T' #(has to be a factor of TIME_FRAME_LENGTH)
#---------------------------------------------
df_lst = [df_2019]
Tele2 = ts.Telescope()
Tele2.load_dfs(df_lst)
Tele2.choose_range(START, END)
Tele2.resample(ROLLING_STEP)
tm.apply_all_metrics(Tele2.df, sma_period = 10)
Tele2.parse_datetime()
Tele2.specific_timeframe(STIME, ETIME)
#print(Tele2.df)
#print(Tele2.grouped)
#print(Tele2.num_of_groups)
#print(Tele2.group_names)
date_list = Tele2.df['date'].unique()
time_list = Tele2.df['time'].unique()
ignore_col_num = int(TIME_FRAME_LENGTH/int(ROLLING_STEP[0])+1)
# change_direction_match_ratio_list = []
# correlatation_list = []
print(time_list[:-ignore_col_num])
#print(Tele2.df['TheoPrice'])
column_list = [time_list[:-ignore_col_num]]
for first_time in tqdm(time_list[:-ignore_col_num]):
if first_time < (ts.cal_datetime(ETIME) - datetime.timedelta(minutes = TIME_FRAME_LENGTH)).time():
first_start = first_time;
first_end = (ts.cal_datetime(first_time) + datetime.timedelta(minutes = TIME_FRAME_LENGTH)).time();
print(first_start)
first_temp_df = ts.select_timeframe(Tele2.df, first_start, first_end)
single_column = []
for second_time in time_list[:-ignore_col_num]:
if (second_time >= (ts.cal_datetime(first_time) + datetime.timedelta(minutes = TIME_FRAME_LENGTH)).time()) & (second_time < (ts.cal_datetime(ETIME) - datetime.timedelta(minutes = TIME_FRAME_LENGTH)).time()):
second_start = second_time
second_end = (ts.cal_datetime(second_start) + datetime.timedelta(minutes = TIME_FRAME_LENGTH)).time()
second_temp_df = ts.select_timeframe(Tele2.df, second_start, second_end)
first_time_move_list = []
second_time_move_list = []
for date in date_list:
# print(first_temp_df[(first_temp_df['date'] == date) & (first_temp_df['time'] == first_end)]['TheoPrice'])
# print(ts.select_date_time(first_temp_df, date, first_end)['TheoPrice'][0])
first_move_theo = ts.select_date_time(first_temp_df, date, first_end)['TheoPrice'][0] - ts.select_date_time(first_temp_df, date, first_start)['TheoPrice'][0]
second_move_theo = ts.select_date_time(second_temp_df, date, second_end)['TheoPrice'][0] - ts.select_date_time(second_temp_df, date, second_start)['TheoPrice'][0]
if ((not np.isnan(first_move_theo)) & (not np.isnan(second_move_theo))):
first_time_move_list.append(first_move_theo)
second_time_move_list.append(second_move_theo)
# print(first_time_move_list)
# print(second_time_move_list)
# count = 0
# for i in range(len(first_time_move_list)):
# if ((first_time_move_list[i] > 0) & (second_time_move_list[i] > 0)) or ((first_time_move_list[i] < 0) & (second_time_move_list[i] < 0)):
# count += 1
# match_ratio = count/len(first_time_move_list)
# change_direction_match_ratio_list.append(match_ratio)
score = 0
correlation = corr_df.loc[first_start,second_start]
for i in range(len(first_time_move_list)):
if first_time_move_list[i] >= UP_THRESHOLD:
if correlation >= CORR_THRESHOLD:
print(correlation)
score += second_time_move_list[i]
score = score - TRADE_COST
if correlation <= -CORR_THRESHOLD:
score += -second_time_move_list[i]
score = score - TRADE_COST
if first_time_move_list[i] <= -DOWN_THRESHOLD:
if correlation >= CORR_THRESHOLD:
score += -second_time_move_list[i]
score = score - TRADE_COST
if correlation <= -CORR_THRESHOLD:
score += second_time_move_list[i]
score = score - TRADE_COST
# corr = pearsonr(first_time_move_list,second_time_move_list)[0]
single_column.append(score)
else:
single_column.append(np.nan)
print(single_column)
column_list.append(single_column)
result = pd.DataFrame(column_list[1:], columns=column_list[0])
result['time'] = time_list[:-ignore_col_num]
result.set_index('time', inplace=True)
#output_df = pd.DataFrame(list(zip(time_list, correlatation_list, change_direction_match_ratio_list)), columns =['start time', 'correlation with next 5 min', 'change direction match ratio'])
#output_df.to_csv(r"C:\Users\achen\Desktop\Monocular\5_min_rolling_correlation_direction_match.csv")
result.to_excel(output_excel_path)
return result
print(time)
first_start = time;
first_end = (ts.cal_datetime(time) + datetime.timedelta(minutes = 5)).time();
second_start = (ts.cal_datetime(time) + datetime.timedelta(minutes = 5)).time()
second_end = (ts.cal_datetime(time) + datetime.timedelta(minutes = 10)).time()
first_temp_df = ts.select_timeframe(Tele2.df, first_start, first_end)
second_temp_df = ts.select_timeframe(Tele2.df, second_start, second_end)
first_time_move_list = []
second_time_move_list = []
for date in date_list:
# print(first_temp_df[(first_temp_df['date'] == date) & (first_temp_df['time'] == first_end)]['TheoPrice'])
# print(ts.select_date_time(first_temp_df, date, first_end)['TheoPrice'][0])
first_move_theo = ts.select_date_time(first_temp_df, date, first_end)['TheoPrice'][0] - ts.select_date_time(first_temp_df, date, first_start)['TheoPrice'][0]
second_move_theo = ts.select_date_time(second_temp_df, date, second_end)['TheoPrice'][0] - ts.select_date_time(second_temp_df, date, second_start)['TheoPrice'][0]
if ((not np.isnan(first_move_theo)) & (not np.isnan(second_move_theo))):
first_time_move_list.append(first_move_theo)
second_time_move_list.append(second_move_theo)
# print(first_time_move_list)
# print(second_time_move_list)
count = 0
for i in range(len(first_time_move_list)):
if ((first_time_move_list[i] > 0) & (second_time_move_list[i] > 0)) or ((first_time_move_list[i] < 0) & (second_time_move_list[i] < 0)):
count += 1
first_move_abs_mean = np.mean(np.absolute(first_time_move_list))
second_move_abs_mean = np.mean(np.absolute(second_time_move_list))
#print(Tele2.grouped)
#print(Tele2.num_of_groups)
#print(Tele2.group_names)
date_list = Tele2.df['date'].unique()
time_list = Tele2.df['time'].unique()
change_direction_match_ratio_list = []
correlatation_list = []
time_index_list = []
#print(Tele2.df['TheoPrice'])
trade_day_list = []
for date in date_list:
noon_theo_price = ts.select_date_time(Tele2.df, date, datetime.time(12,0,0))['TheoPrice'][0]
if (not np.isnan(noon_theo_price)):
trade_day_list.append(date)
column_list = [trade_day_list]
for first_time in tqdm(time_list):
if first_time < (ts.cal_datetime(ETIME) - datetime.timedelta(minutes = TIME_FRAME_LENGTH)).time():
time_index_list.append(first_time)
first_start = first_time;
first_end = (ts.cal_datetime(first_time) + datetime.timedelta(minutes = TIME_FRAME_LENGTH)).time();
first_temp_df = ts.select_timeframe(Tele2.df, first_start, first_end)
#
single_column = []
second_start = second_time
second_end = (
ts.cal_datetime(second_start) +
datetime.timedelta(minutes=TIME_FRAME_LENGTH)).time()
second_temp_df = ts.select_timeframe(
Tele2.df, second_start, second_end)
first_time_move_list = []
second_time_move_list = []
for date in date_list:
# print(first_temp_df[(first_temp_df['date'] == date) & (first_temp_df['time'] == first_end)]['TheoPrice'])
# print(ts.select_date_time(first_temp_df, date, first_end)['TheoPrice'][0])
first_move_theo = ts.select_date_time(
first_temp_df, date,
first_end)[METRIC][0] - ts.select_date_time(
first_temp_df, date, first_start)[METRIC][0]
second_move_theo = ts.select_date_time(
second_temp_df, date,
second_end)[METRIC][0] - ts.select_date_time(
second_temp_df, date, second_start)[METRIC][0]
if ((not np.isnan(first_move_theo)) &
(not np.isnan(second_move_theo))):
first_time_move_list.append(first_move_theo)
second_time_move_list.append(second_move_theo)
# print(first_time_move_list)
# print(second_time_move_list)
# count = 0
def dir_match_backtest(df_2019, start_date, end_date):
import numpy as np
import pandas as pd
import telescope as ts
import telescope_metrics as tm
import datetime
from tqdm import tqdm
import rolling_pairwise_theo_diff_dir_match as rptddm
import rolling_pairwise_theo_diff_sum_profit as rptdsp
from datetime import timedelta
#--------------Input Parameters---------------
df_2019 = r"Z:\KaiData\Theo\2019\NQ.ESTheo.mpk"
START = start_date
END = end_date
df_lst = [df_2019]
Tele2 = ts.Telescope()
Tele2.load_dfs(df_lst)
Tele2.choose_range(START, END)
Tele2.parse_datetime()
#print(Tele2.df)
#print(Tele2.grouped)
#print(Tele2.num_of_groups)
#print(Tele2.group_names)
date_list = Tele2.df['date'].unique()
#print(Tele2.df['TheoPrice'])
trade_day_list = []
for date in date_list:
if ts.select_date_time(Tele2.df, date,
datetime.time(12, 0, 0))['TheoPrice'].empty:
continue
else:
trade_day_list.append(date)
time_length = 60
#corr_threshold_list = [0.5,0.6,0.7,0.8,0.9]
#up_threshold_list = [50, 75, 100]
#down_threshold_list = [50, 75, 100]
corr_threshold_list = [3]
up_threshold_list = [60]
down_threshold_list = [60]
trading_cost = 30
count = 0
for corr_thre in corr_threshold_list:
for up_thre in up_threshold_list:
for date in tqdm(trade_day_list):
print(date)
down_thre = up_thre
if date.weekday() == 4:
corr_start_datetime = date - timedelta(days=4)
else:
corr_start_datetime = date - timedelta(days=6)
corr_end_datetime = date - timedelta(minutes=1)
test_start_datetime = date
test_end_datetime = date + timedelta(days=1) - timedelta(
minutes=1)
output_file_path_corr = "C:\\Users\\achen\\Desktop\\Monocular\\Automatic Backtest\\Direction Match\\NQ_" + str(
time_length) + "MIN_" + corr_start_datetime.strftime(
"%Y_%m_%d") + "_" + corr_end_datetime.strftime(
"%Y_%m_%d") + "_ROLL5_PAIR_DIRMATCH_auto.xlsx"
output_file_path_backtest = "C:\\Users\\achen\Desktop\\Monocular\\Automatic Backtest\\Dirmatch PL\\NQ_" + str(
time_length) + "MIN_" + test_start_datetime.strftime(
"%Y_%m_%d") + "_ROLL5_PAIR_DIRMATCH_PROFIT_U" + str(
up_thre) + "_D" + str(down_thre) + "_T" + str(
trading_cost) + "_auto.xlsx"
corr_df = rptddm.rolling_pairwise_theo_diff_dir_match(
"Z:\\KaiData\\Theo\\2019\\NQ.ESTheo.mpk",
corr_start_datetime,
corr_end_datetime,
datetime.time(8, 0, 0),
datetime.time(15, 15, 0),
time_frame_length=time_length,
rolling_step='5T',
output_excel_path=output_file_path_corr)
if corr_df is None:
continue
else:
count += 1
pl_df = rptdsp.rolling_pairwise_theo_diff_sum_profit(
corr_df,
"Z:\\KaiData\\Theo\\2019\\NQ.ESTheo.mpk",
test_start_datetime,
test_end_datetime,
datetime.time(8, 0, 0),
datetime.time(15, 15, 0),
time_frame_length=time_length,
rolling_step='5T',
corr_threshold=corr_thre,
up_threshold=up_thre,
down_threshold=down_thre,
trade_cost=trading_cost,
output_excel_path=output_file_path_backtest)
if count == 1:
result_df = pl_df
else:
result_df = result_df.add(pl_df)