In reply to https://towardsdatascience.com/deep-learning-in-finance-9e088cb17c03

AR_1_on_gold.py

%matplotlib notebook

import pandas_datareader as pdr
from sklearn.linear_model import LinearRegression 
from sklearn.metrics import mean_squared_error


# Download gold data (ETF)
gld = pdr.get_data_yahoo('GLD',start='1984-01-01')

# Use AR(1), i.e. previous price predicts next
gld['Prev'] = gld['Adj Close'].shift(1)
X = gld['Prev']
y = gld['Adj Close']

# 80% train, 20% test
t = int(0.8*len(gld)) 
X_train = X[1:t]
y_train = y[1:t]
X_test = X[t:] 
y_test = y[t:]

# Fit and predict
linear = LinearRegression().fit(X_train.values[:,None],y_train.values[:,None]) 
predicted_price = linear.predict(X_test.values[:,None])

print(mean_squared_error(y_test.values[:,None], predicted_price))

y.plot()
plt.plot(X_test.index, predicted_price)