Scribd
Upload
4 views3 pages

SARIMA Model Selection Checklist

This document outlines a checklist for selecting the best SARIMA model in R, focusing on achieving accuracy and parsimony. Key steps include ensuring stationarity, model identification, estimation, selection criteria, residual diagnostics, and evaluating forecast accuracy. The document emphasizes the importance of using diagnostic tests and model comparison metrics like AIC, BIC, and forecast error measures.

Uploaded by

Zaw Khin
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
4 views3 pages

SARIMA Model Selection Checklist

This document outlines a checklist for selecting the best SARIMA model in R, focusing on achieving accuracy and parsimony. Key steps include ensuring stationarity, model identification, estimation, selection criteria, residual diagnostics, and evaluating forecast accuracy. The document emphasizes the importance of using diagnostic tests and model comparison metrics like AIC, BIC, and forecast error measures.

Uploaded by

Zaw Khin
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 3

Checklist for Selecting the Best SARIMA

Model in R
This document provides a structured checklist for selecting the best SARIMA model. The
goal is to achieve the most accurate and parsimonious model while satisfying all diagnostic
checks.

1. Stationarity

- Ensure the time series is stationary in both trend and seasonality.


- Use:
- First differencing (d)
- Seasonal differencing (D)
- Tests:
- Augmented Dickey-Fuller (ADF): tseries::adf.test(Y)
- KPSS Test: tseries::kpss.test(Y)

2. Model Identification

- Determine non-seasonal and seasonal orders:


- Non-seasonal: (p, d, q)
- Seasonal: (P, D, Q)[s]
- Use ACF and PACF plots:
- acf(Y) → MA and SMA components (q, Q)
- pacf(Y) → AR and SAR components (p, P)
- Look for significant lags at seasonal intervals (e.g., 12, 24).

3. Model Estimation

- Fit SARIMA models manually using:


arima(Y, order = c(p,d,q), seasonal = list(order = c(P,D,Q), period = s))
- Use auto.arima from forecast package to automatically select model:
forecast::auto.arima(Y, seasonal = TRUE)
4. Model Selection Criteria

- Evaluate and compare models using:


- AIC (Akaike Information Criterion)
- BIC (Bayesian Information Criterion)
- AICc (corrected AIC for small samples)
- Prefer models with lower values.

5. Residual Diagnostics

- Residuals should be white noise (uncorrelated and normally distributed).


- Check for autocorrelation:
- ACF plot of residuals
- Ljung-Box Test: Box.test(residuals, lag = 20, type = "Ljung-Box")
- Normality tests:
- Histogram and Q-Q plot
- Shapiro-Wilk Test: shapiro.test(residuals)

6. Forecast Accuracy

- Evaluate predictions on holdout/test data using:


- MAE (Mean Absolute Error)
- RMSE (Root Mean Squared Error)
- MAPE (Mean Absolute Percentage Error)
- Use accuracy(forecast, actual_values) to compute metrics.

7. Summary: Criteria to Consider


Factor Goal

Stationarity Use differencing to achieve

AIC/BIC/AICc As low as possible

Residual Autocorrelation None (Ljung-Box, ACF)

Residual Normality Approximately normal (Shapiro, QQ plot)

Forecast Accuracy Low MAE, RMSE, MAPE


Model Parsimony Simpler model if accuracy is similar

You might also like