analisi_econometrica_statistica_aziona_4.py

from statsmodels.tsa.statespace.sarimax import SARIMAX
from sklearn.metrics import mean_squared_error

# Drop NaNs from differencing (if any)
daily_data = df['Value'].resample('D').mean()
daily_data = daily_data.asfreq('D') # daily average
daily_data = daily_data.fillna(method='ffill')  # safety fill if any missing

subset = daily_data['2024-01-01':'2024-03-31']

# Fit a SARIMA model
model = SARIMAX(
    subset,
    order=(1, 1, 1),             # p, d, q
    seasonal_order=(1, 1, 1, 7), # P, D, Q, S (weekly seasonality)
    enforce_stationarity=False,
    enforce_invertibility=False
)

results = model.fit(disp=False)
print(results.summary())

results.plot_diagnostics(figsize=(15, 8))
plt.tight_layout()
plt.show()

forecast = results.get_forecast(steps=168)
forecast_df = forecast.summary_frame()

# Plot
plt.figure(figsize=(14, 5))
plt.plot(daily_data[-168:], label='Observed (last 7 days)')
plt.plot(forecast_df['mean'], label='Forecast (next 7 days)', color='orange')
plt.fill_between(forecast_df.index,
                 forecast_df['mean_ci_lower'],
                 forecast_df['mean_ci_upper'],
                 color='orange', alpha=0.3)
plt.title('SARIMA Forecast')
plt.xlabel('Date')
plt.ylabel('Value')
plt.legend()
plt.grid(True)
plt.tight_layout()
plt.show()