def get_returns_series(ticker_files, current_date, n_days, rolling_lookback,
datetime_format):
"""
Calculates and returns the VAE embeddings at the given date.
"""
x_all = {}
x_mean_all = {}
x_std_all = {}
filenames_all = {}
# tickers_all = []
lookback_start_dates = {}
for file in ticker_files:
df = read_csv(file, datetime_format)
# Get the lookback period ending at the current date.
df = df.loc[:current_date]
df = df.iloc[-(rolling_lookback + n_days):]
# Calculate normalised returns.
df['returns'] = df['close'].pct_change()
df['mean'] = df['returns'].rolling(rolling_lookback).mean()
df['std'] = df['returns'].rolling(rolling_lookback).std()
df['returns'] = (df['returns'] - df['mean']) / df['std']
df = df.dropna()
# We should be left with the number of lookback days.
if df.shape[0] != n_days:
continue
x = np.array(df['returns'])
x_mean = np.array(df['mean'])
x_std = np.array(df['std'])
ticker = get_ticker_from_filename(file)
x_all[ticker] = x
x_mean_all[ticker] = x_mean
x_std_all[ticker] = x_std
filenames_all[ticker] = file
lookback_start_dates[ticker] = df.index[0]
# x_all.append(x)
# x_mean_all.append(x_mean)
# x_std_all.append(x_std)
# filenames_all.append(file)
# # tickers_all.append(get_ticker_from_filename(file))
# lookback_start_dates.append(df.index[0])
return x_all, x_mean_all, x_std_all, filenames_all, lookback_start_dates
def data():
return read_csv("datasets/csv/isbsg10.arff.csv")
def data():
return read_csv("datasets/csv/albrecht.arff.csv")
def data():
return read_csv("datasets/csv/maxwell.arff.csv")
def output_spreads(pairs_vae, filenames_all, current_date, n_backtest_days,
returns_lookback, backtest_results, trade_results,
backtest_results_file, out_dir, datetime_format):
# Plot the rolling mean of the spread, for each pair that we backtested.
# This is for generating features and targets for a supervised "profitability" predictor.
# This requires the traded pairs to be the same as the backtested pairs (in config file).
for pair_ind, pair in enumerate(pairs_vae):
ticker_1, ticker_2 = pair[0], pair[1]
filename_1 = filenames_all[ticker_1]
filename_2 = filenames_all[ticker_2]
df_1 = read_csv(filename_1, datetime_format)
df_1 = df_1.loc[:current_date]
df_1 = df_1.iloc[-(returns_lookback + n_backtest_days):]
df_2 = read_csv(filename_2, datetime_format)
df_2 = df_2.loc[:current_date]
df_2 = df_2.iloc[-(returns_lookback + n_backtest_days):]
# The spread between the two tickers.
df_1['spread'] = df_1['close'] / df_2['close']
# The rolling mean and std dev. of the spread.
df_1['spread_mean'] = df_1['spread'].rolling(20).mean()
df_1['spread_std'] = df_1['spread'].rolling(20).std()
# Normalise the mean and std. dev.
df_1['spread_norm'] = (df_1['spread'] -
df_1['spread'].mean()) / df_1['spread'].std()
df_1['spread_norm_mean'] = df_1['spread_norm'].rolling(20).mean()
df_1['spread_norm_std'] = df_1['spread_norm'].rolling(20).std()
# df_1['spread_mean_norm'] = (df_1['spread_mean'] - df_1['spread_mean'].mean()) / df_1['spread_mean'].std()
# df_1['spread_std_norm'] = (df_1['spread_std'] - df_1['spread_std'].mean()) / df_1['spread_std'].std()
# backtest_sharpe = backtest_sharpes[pair_ind]
backtest_sharpe = backtest_results[(pair[0], pair[1])][2]
backtest_profitable = 1 if backtest_sharpe > 0 else 0
print(pair)
trade_sharpe = trade_results[(pair[0], pair[1])][2]
trade_profitable = 1 if trade_sharpe > 0 else 0
fig, axes = plt.subplots(2, 1, squeeze=False)
ax = axes[0, 0]
ax.plot(df_1['spread'], c='blue')
ax.plot(df_1['spread_mean'], c='orange')
ax.plot(df_1['spread_mean'] + 2 * df_1['spread_std'], c='green')
ax.plot(df_1['spread_mean'] - 2 * df_1['spread_std'], c='green')
ax.grid()
ax.set_title(f'Sharpe {backtest_sharpe}')
ax = axes[1, 0]
ax.plot(df_1['spread_norm'], c='blue')
ax.plot(df_1['spread_norm_mean'], c='orange')
ax.plot(df_1['spread_norm_mean'] + 2 * df_1['spread_norm_std'],
c='green')
ax.plot(df_1['spread_norm_mean'] - 2 * df_1['spread_norm_std'],
c='green')
ax.grid()
plt.savefig(out_dir / f'backtest_spread_{ticker_1}-{ticker_2}.png')
plt.close(fig)
df_1 = df_1.dropna()
results_str = f'{current_date}, {ticker_1}, {ticker_2}, '
spread_norm_means = list(df_1['spread_norm_mean'])
spread_norm_stds = list(df_1['spread_norm_std'])
for spread_mean in spread_norm_means:
results_str += f'{spread_mean}, '
for spread_std in spread_norm_stds:
results_str += f'{spread_std}, '
results_str += f'{backtest_sharpe}, {backtest_profitable}, {trade_sharpe}, {trade_profitable}'
backtest_results_file.write(f'{results_str}\n')
backtest_results_file.flush()
def data():
return read_csv("datasets/csv/finnish.arff.csv")
def data():
return read_csv("datasets/csv/china.arff.csv")
def data():
return read_csv("datasets/csv/desharnais.arff.csv")
def data():
return read_csv("datasets/csv/miyazaki.arff.csv")
def data():
return read_csv("datasets/csv/kemerer.arff.csv")
def data():
return read_csv("datasets/csv/kitchenhamm.arff.csv")