def vix_momentum_bias(timeseries, date, votality_calculation_range,
observation_range):
vix, vix_id = download_stock("^vix", date, 1, False)
for i in range(len(vix) - 1 - votality_calculation_range, 0, -1):
del vix[i]
votality_rate = (vix[len(vix) - 1] - vix[0]) / len(vix)
bias = 0.00 # smaller the better
for i in range(timeseries.raw_size() - 1,
timeseries.raw_size() - observation_range, -1):
if timeseries.raw_datapoint(timeseries.raw_size() -
1) < timeseries.raw_datapoint(i):
bias += 1
else:
bias -= 1
# always be negative (positive momentum * negative = negative bias)
# (negative momentum * negative = positive bias)
bias *= -votality_rate
return bias
def __init__(self, name, id):
self.id = id
self.name = name
self.start_date = ""
self.data = None
self.close_price = 0
self.percentage = 0.00
self.shares = 0
self.required_purchase_sales = 0.00
self.percentage_diff = 0
today = date.today()
today = str(today)
date_info = today.split("-")
date_info[0] = str(int(date_info[0]) - 1)
for i in range(len(date_info)):
if i == len(date_info) - 1:
self.start_date += date_info[i]
else:
self.start_date += date_info[i] + "-"
self.data, _ = download_stock(self.id, self.start_date, 1, True)
self.close_price = self.data[len(self.data) - 1]
bh_change_matrix.append(bh_change)
for i in range(_range, _range - evaluate_range, -1):
if stock[_range] > stock[i]:
momentum += 1
if momentum >= evaluate_range * 8 / 12:
momentum = evaluate_range
else:
momentum = evaluate_range / 2
stock_asset = int(total_asset * (momentum / evaluate_range))
shares = int(stock_asset / stock[_range])
stock_asset = shares * stock[_range]
cash = total_asset - shares * stock[_range]
previous_asset = total_asset
print("Total = ", total_asset, " [ Stock = ", stock_asset, ", Cash = ",
cash, "] --> Shares: ", shares, ", < CHANGE=", change, ">",
"<BH Change = ", bh_change, ">", stock[_range])
print("Final Profit = ", (total_asset - principal) * 100 / principal)
print(stock[0])
graph(change_matrix, 'red', 'momentum.jpg', False)
graph(bh_change_matrix, 'blue', 'momentum.jpg', False)
# momentum 비중 높으면 --> 채권 비중을 주식 비중으로 이동
# momentum 비중 낮으면 --> 주식 비중을 채권 비중으로 이동
if __name__ == "__main__":
csv, id = download_stock()
stock = upload(csv, 5, True)
momentum_investing(stock, 100000000, 8, 1)
git_update()
#!/usr/bin/env python3
from service import download_stock
while True:
stock, id = download_stock()
# get date YYYY-MM-DD string 1 year before
start_date = ""
today = str(date.today())
date_info = today.split("-")
date_info[0] = str(int(date_info[0]) - 1)
for i in range(len(date_info)):
if i == len(date_info) - 1:
start_date += date_info[i]
else:
start_date += date_info[i] + "-"
print("Downloading/uploading historcial stock data from finance.yahoo.com....")
# download historical data, get one time series from stock, get close price
for id in stock_id_list:
historical, id = download_stock(id, start_date, 1, False)
timeseries, close_price = fetch_last_time_series(historical, QUARTER)
dataset.append(timeseries)
price.append(int(close_price))
cash_balance = 100.0 - total_stock_balance
adjustment_value = 0.3 # THIS VALUE COULD BE TUNED
# run sequential prediction before rebalancing
predictions = []
print("Running prediction models... ")
for i in range(len(dataset)):
stock, prediction = sequential_prediction(model=stock_id_list[i].lower(), stock_id=stock_id_list[i].lower(), date=start_date, log=False, bias_type='REGRESSION', itr=10)
if prediction == []: # meaning that no model exists
predictions.append('NO MODEL')
else:
def sequential_prediction(model=None,
stock_id=None,
date=None,
graphing=True,
log=True,
add_bias=True,
bias_type='DEFAULT',
itr=5):
if model == None:
model = select_model()
print("Model = [", model, "]\n")
stock = None
if stock_id == None or date == None:
stock, id = download_stock()
else:
stock, id = download_stock(stock_id, date, 1, True)
predictor = None
model_exists = None
prediction_matrix = []
try:
predictor = Model(model)
model_exists = True
except:
model_exists = False
if model_exists == True:
timeseries, final_close = fetch_last_time_series(stock, QUARTER)
bias = 0.00
if add_bias == True:
if bias_type == 'VIX':
bias = vix_momentum_bias(timeseries, date, WEEK, MONTH)
elif bias_type == 'REGRESSION':
bias = regression_momentum_bias(timeseries, WEEK)
else: # no specific BIAS type inputted
bias_mode = int(
input("Bias Type [0: Votality, 1: Regression] :: "))
# compute bias using momentum calculations with VIX index
if bias_mode == 1:
bias = regression_momentum_bias(timeseries, WEEK)
else:
bias = vix_momentum_bias(timeseries, date, WEEK, MONTH)
#print("BIAS = ", bias)
for count in range(itr):
trend = moving_average(timeseries, MONTH)
matrix = sampling(trend, 0, 2, STANDARD_SAMPLING_RANGE)
timeseries.set_sampled_matrix(matrix)
timeseries.normalize_timeseries()
prediction = 0.00
result = predictor.predict(timeseries)
for i in range(result.shape[0]):
for j in range(result.shape[1]):
prediction = rescale(result[i][j], timeseries.minimum(),
timeseries.maximum())
# apply the pre-calculated bias to the prediction results
prediction += bias
prediction_matrix.append(prediction)
raw = timeseries.raw_matrix()
for i in range(0, len(raw)):
raw[i] = rescale(raw[i], timeseries.minimum(),
timeseries.maximum())
del raw[0]
raw.append(prediction)
timeseries.set_raw_matrix(raw)
count += 1
if graphing == True:
graph_title = "../Images/" + model + "_sequential_prediction_demo.png"
graph(prediction_matrix, 'red', graph_title, False)
return stock, prediction_matrix
models = all_models()
download_date = ""
today = str(date.today())
date_info = today.split("-")
date_info[0] = str(int(date_info[0]) - 1)
for i in range(len(date_info)):
if i == len(date_info) - 1:
download_date += date_info[i]
else:
download_date += date_info[i] + "-"
if __name__ == "__main__":
print("UPDATING ALL EXISTING MODELS:\n")
for m in models:
stock, id = download_stock(m, download_date, 1, False)
if len(stock) <= QUARTER:
print("Failed! [NOT ENOUGH DATA]")
else:
dataset = partition_time_series(stock, QUARTER, 35, False)
for timeseries in dataset:
matrix = moving_average(timeseries, MONTH)
matrix = sampling(matrix, 0, 2, STANDARD_SAMPLING_RANGE)
timeseries.set_sampled_matrix(matrix)
timeseries.normalize_timeseries()
predictor = model.Model(m)
predictor.update(dataset, True, 10, 32, 0)
print("Updated model for {0}!".format(m))
git_update()