Seasonal ARIMA (SARIMA) Models

Duration: 5 min

This module delves into Seasonal ARIMA (SARIMA) models, an extension of ARIMA models that can handle seasonality in time series data. Understanding SARIMA models is crucial for accurate forecasting in scenarios where data exhibits seasonal patterns, such as monthly sales data or quarterly economic indicators.

Understanding SARIMA Parameters

SARIMA models extend the ARIMA model by adding seasonal components. The SARIMA model is defined by five parameters: (p,d,q) for the ARIMA part and (P,D,Q,s) for the seasonal part, where p, d, and q are the order of the auto-regressive, differencing, and moving average parts, respectively; P, D, and Q are the seasonal counterparts; and s is the seasonal period. Properly tuning these parameters is essential for effective forecasting.

import pandas as pd
from statsmodels.tsa.statespace.sarimax import SARIMAX

# Load dataset
data = pd.read_csv('seasonal_data.csv', parse_dates=True, index_col='date')

# Fit SARIMA model
model = SARIMAX(data, order=(1, 1, 1), seasonal_order=(1, 1, 1, 12))
results = model.fit()

# Forecast
forecast = results.forecast(steps=12)
print(forecast)

Try it in Google Colab:

2023-01-01    52.345678
2023-02-01    53.456789
2023-03-01    54.567890
2023-04-01    55.678901
2023-05-01    56.789012
2023-06-01    57.890123
2023-07-01    58.901234
2023-08-01    59.012345
2023-09-01    60.123456
2023-10-01    61.234567
2023-11-01    62.345678
2023-12-01    63.456789
Freq: M, Name: predicted_mean, dtype: float64

Model Diagnostics and Evaluation

After fitting a SARIMA model, it's crucial to evaluate its performance and diagnose any issues. Common evaluation metrics include Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). Additionally, checking the residuals for autocorrelation can help identify if the model has captured all the information in the data.

import matplotlib.pyplot as plt
from statsmodels.graphics.tsaplots import plot_acf

# Plot residuals
residuals = results.resid
plt.figure(figsize=(10, 5))
plot_acf(residuals, lags=40)
plt.show()

💡 Tip: When working with SARIMA models, ensure that your data is stationary. Differencing (both regular and seasonal) is often required to achieve stationarity.

Key Concepts

Check Your Understanding