def test_char_to_date(self):
d1 = np.arange("2010-01-01", "2011-01-01", dtype='datetime64[D]')
df = pd.DataFrame({'date1': d1, 'date2': d1.astype(str)})
# test for dataframe
df1 = char_to_date(df)
for date_col in ['date1', 'date2']:
np.testing.assert_array_equal(df1[date_col].values, d1)
# series
for date_col in ['date1', 'date2']:
np.testing.assert_array_equal(char_to_date(df[date_col]), d1)
def __init__(self, data: pd.DataFrame, input_directory: str, is_bloomberg: str) -> None:
self.run_date = extract_str_timestamp(dt.datetime.now())
self.wkdir = input_directory
self.set_output_folder()
if is_bloomberg:
input_data = utils_date.char_to_date(data)
self.start_date = extract_str_timestamp(input_data.index.min())
self.end_date = extract_str_timestamp(input_data.index.max())
print(f"Data for period runs {self.start_date} to {self.end_date}")
self.data = input_data
elif is_bloomberg:
print("Data likely downloaded from Yahoo Finance?")
pass
def prep_fund_data(df_path, date_col="Date"):
"""Prep fund data (csv) using char to date and setting 'Date' as the index
Args:
df_path (str): Path to dataframe
date_col (str): Date column labelled in bloomberg dataframe
Returns:
df (dataframe): Dataframe with date columns converted to np.datetime64
"""
df = pd.read_csv(df_path)
df = char_to_date(df)
assert date_col in df.columns, f"The date column: {date_col} is not specified in the function"
df.set_index('Date', inplace=True)
return df
if __name__ == "main":
RUN_DATE_STR = dt.datetime.now().strftime("%Y-%m-%d")
RUN_DATE_DT = np.datetime64(RUN_DATE_STR)
WORK_DIR = r"/Users/philip_p/python/projects"
OUTPUT_DIR = os.path.join(WORK_DIR, "output", "efficient_frontier")
TODAY_OUTPUT_DIR = os.path.join(OUTPUT_DIR, RUN_DATE_STR)
if not os.path.exists(TODAY_OUTPUT_DIR):
os.mkdir(TODAY_OUTPUT_DIR)
example_df = pd.read_csv(get_data_path("example_data.csv"),
index_col='Date')
example_df = utils_date.char_to_date(example_df)
example_df.dropna(axis=0, inplace=True)
# EXPLORATORY DATA ANALYSIS
# --------------
# plot the prices - using plot_raw_data method defined above
# plot_raw_data(df= df, file_name="price_plot", y_label="Price (p)", to_close=True)
# # Now look at 1 day returns (should follow something resembling a normal distribution)
# returns = df.pct_change()
# plot_raw_data(df=returns, file_name="returns_plot",
# y_label="Daily Return (%)", to_close=True)
# Explore the methods
daily_return = example_df.pct_change()
import utils_date
plt.style.use('seaborn')
pd.set_option('display.max_columns', 5)
wkdir = "C://Users//Philip//Documents//python//"
inputFolder = wkdir + "input/"
inputDir = wkdir + "input/"
outputFolder = wkdir + "output/"
# "example_data.csv", "example_data_na.csv" has NA rows
# df = pd.read_csv(inputDir + 'example_data.csv') #, parse_dates=True)
df = pd.read_csv(inputDir + "funds_stocks_2019.csv")
df = utils_date.char_to_date(df) #convert all dates to np datetime64
df.set_index('Date', inplace=True)
# deal with returns not time series of prices, as prices are non-stationary transform the time
# series so that it becomes stationary. If the non-stationary is a random walk with or without
# drift, it is transformed to a stationary process by differencing - it is now a stationary
# stochastic (random probability distribution) process if time series data also exhibits a
# deterministic trend, spurious results can be avoided by detrending if non-stationary time
# series are both stochastic and deterministic at the same time, differencing and detrending
# should be applied - differencing removes the trend in variance and detrending removes
# determinstic trend
def log_daily_returns(data):
"""Give log daily returns"""
log_daily_return = data.apply(lambda x: np.log(x) - np.log(x.shift(1)))[1:]