def main():
# Prevent truncating display of DataFrame
pd.set_option('display.width', None)
goog_data = data.get_google_data('data/goog_data_large.pkl',
start_date='2001-01-01',
end_date='2018-01-01')
# Features are open - close price and high - low price
# Target is difference in daily close price
goog_data, x, y = finml.create_regression_trading_condition(goog_data)
# Split into train (80%) and test data sets
x_train, x_test, y_train, y_test = finml.create_train_split_group(x, y)
# Ridge regression model with regularisation parameter 0.1
ridge = linear_model.Ridge(alpha=10000)
ridge.fit(x_train, y_train)
print('Coefficients:\n', ridge.coef_)
# Evaluate model on training data
print(f'Mean squared error on training data: '
f'{mean_squared_error(y_train, ridge.predict(x_train))}')
# Explained variance score: 1 is perfect prediction
print(f'Variance score on training data: '
f'{r2_score(y_train, ridge.predict(x_train))}')
# Evaluate model on test data
print(f'Mean squared error on test data: '
f'{mean_squared_error(y_test, ridge.predict(x_test))}')
# Explained variance score: 1 is perfect prediction
print(f'Variance score on test data: '
f'{r2_score(y_test, ridge.predict(x_test))}')
# LASSO regression model to predict prices and calculate strategy returns
goog_data['Predicted_signal'] = ridge.predict(x)
goog_data['GOOG_Returns'] = \
np.log(goog_data['Close'] / goog_data['Close'].shift(1))
print(goog_data)
# Calculate cumulative returns for the test data (train data onwards)
cum_goog_return = \
finml.calculate_return(goog_data, split_value=len(x_train),
symbol='GOOG')
cum_strategy_return = \
finml.calculate_strategy_return(goog_data, split_value=len(x_train))
finml.plot_chart(cum_goog_return, cum_strategy_return, symbol='GOOG')
sharpe = finml.sharpe_ratio(cum_strategy_return, cum_goog_return)
print(f'Sharpe ratio: {sharpe}')
# Display plots and block
plt.show()
def main():
# Prevent truncating display of DataFrame
pd.set_option('display.width', None)
goog_data = data.get_google_data('data/goog_data_large.pkl',
start_date='2001-01-01',
end_date='2018-01-01')
# Features are open - close price and high - low price
# Target is based on difference in daily close price and is +1 if future
# close price is higher than the current close price and -1 if the future
# close price is lower than the current close price.
goog_data, x, y = finml.create_classification_trading_condition(goog_data)
# Split into train (80%) and test data sets
x_train, x_test, y_train, y_test = finml.create_train_split_group(x, y)
# KNN classification model using K=15
knn = KNeighborsClassifier(n_neighbors=15)
y_train_array = y_train.to_numpy().ravel()
knn.fit(x_train, y_train_array)
accuracy_train = accuracy_score(y_train, knn.predict(x_train))
print('Training accuracy:', accuracy_train)
accuracy_test = accuracy_score(y_test, knn.predict(x_test))
print('Test accuracy:', accuracy_test)
# Predict whether the price goes up or down
goog_data['Predicted_signal'] = knn.predict(x)
goog_data['GOOG_Returns'] = \
np.log(goog_data['Close'] / goog_data['Close'].shift(1))
print(goog_data)
# Calculate cumulative returns for the test data (train data onwards)
cum_goog_return = \
finml.calculate_return(goog_data, split_value=len(x_train),
symbol='GOOG')
cum_strategy_return = \
finml.calculate_strategy_return(goog_data, split_value=len(x_train))
finml.plot_chart(cum_goog_return, cum_strategy_return, symbol='GOOG')
sharpe = finml.sharpe_ratio(cum_strategy_return, cum_goog_return)
print(f'Sharpe ratio: {sharpe}')
# Display plots and block
plt.show()