How do I interpret AIC and BIC for SARIMA model selection?

Lower AIC and BIC indicate better balance between fit and parsimony. AIC favors prediction; BIC penalizes complexity more heavily. When two models differ by less than 2 AIC units, prefer the simpler model.

What is the Ljung-Box test in SARIMA diagnostics?

The Ljung-Box test checks whether residuals are white noise (no remaining autocorrelation). A non-significant p-value (p > 0.05) means the SARIMA model has captured the series structure adequately.

How does SARIMA differ from ARIMA and Holt-Winters?

ARIMA models trend only. SARIMA adds explicit seasonal AR/MA terms and seasonal differencing, making it ideal for periodic data. Holt-Winters uses exponential smoothing for level, trend, and seasonality but offers fewer diagnostic tools than SARIMA.

What is the minimum sample size to fit a SARIMA model?

As a rule of thumb, you need at least 2 to 3 full seasonal cycles, and ideally 4 or more. For monthly data (m=12), aim for 50+ observations; for quarterly (m=4), aim for 24+.

How do I know if my SARIMA forecast is accurate?

Check residual diagnostics (Ljung-Box p > 0.05, near-zero ACF, normal Q-Q plot) and out-of-sample accuracy metrics (RMSE, MAE, MAPE). Plot fitted vs actual values; the residuals should look like white noise.

SARIMA Model Calculator – Free Online Seasonal ARIMA Time Series Forecasting Tool

SARIMA Model Calculator

A free, online SARIMA (Seasonal ARIMA) model calculator for time-series forecasting. Estimate (p,d,q)(P,D,Q)_m parameters, run residual diagnostics, view AIC/BIC, generate forecasts with prediction intervals, and export APA-ready results.

Time Series Seasonal ARIMA Forecasting Free Online APA Ready

Seasonal Pattern

Forecast + CI

ACF Lags

Residuals

📥 1. Enter Your Time Series Data

Each cluster represents one time series you want to model. The default mode accepts comma-separated values (e.g., 52, 48, 55, 61, 47, …). Cluster names are editable.

📊 Sample Dataset

🔁 Seasonal Period (m)

SARIMA Order (p, d, q) (P, D, Q) — auto if blank

Order: p=AR, d=diff, q=MA (non-seasonal); P=seasonal AR, D=seasonal diff, Q=seasonal MA.

⏩ Forecast Horizon (h)

α (significance)

Upload CSV or Excel file (.csv, .txt, .xlsx, .xls)

Click columns that should each become a cluster — selected columns will be loaded as separate clusters.

Enter values one per row. Each column = one cluster (time series).

📊 2. SARIMA Results

📈 3. Four Colorful Visualization Plots

📉 Plot 1 — Original Series + Fitted + Forecast

🔁 Plot 2 — Residuals over Time

📊 Plot 3 — ACF of Residuals

🌐 Plot 4 — Seasonal Decomposition (Trend / Seasonal)

📋 Copy summary to clipboard

🧠 4. Detailed Interpretation of Results

✍️ 5. How to Write Your Results in Research

Five ready-to-use reporting templates auto-filled from your SARIMA run. Click 📋 Copy on any card.

📐 6. Technical Formulas & Theory

SARIMA(p,d,q)(P,D,Q)_m Model

The general SARIMA model is written using backshift operator B as:

Φ_P(B^m) · φ_p(B) · (1−B)^d · (1−B^m)^D · y_t = Θ_Q(B^m) · θ_q(B) · ε_t

Where:

y_t = observed value at time t
B = backshift operator: B y_t = y_t−1
φ_p(B) = 1 − φ₁B − … − φ_pB^p (non-seasonal AR)
θ_q(B) = 1 + θ₁B + … + θ_qB^q (non-seasonal MA)
Φ_P(B^m), Θ_Q(B^m) = seasonal AR and MA polynomials
d, D = non-seasonal and seasonal differencing orders
m = seasonal period (12 for monthly, 4 for quarterly)
ε_t ~ white noise N(0, σ²)

Information Criteria

AIC = 2k − 2 ln(L̂) | BIC = k ln(n) − 2 ln(L̂)

Where: k = number of parameters, L̂ = maximised likelihood, n = sample size. Lower is better.

Ljung–Box Q Statistic (residual whiteness)

Q = n(n+2) · Σ_k=1^h [ ρ̂_k² / (n−k) ] ~ χ²(h − k_fitted)

Where: ρ̂_k = sample autocorrelation at lag k, h = number of lags tested.

Forecast Prediction Interval

ŷ_n+h ± z_α/2 · σ̂ · √(1 + ψ₁² + ψ₂² + … + ψ_h−1²)

Where: ψ_j are MA(∞) coefficients of the fitted model, σ̂ is the residual SD, z_α/2 = 1.96 for 95% CI.

🎯 7. When to Use SARIMA

This free SARIMA model calculator is designed for forecasting time series with both trend and a clear repeating seasonal pattern. Use SARIMA when:

✅ Your series has at least 2–3 full seasonal cycles of data.
✅ Visual inspection shows a clear periodic pattern (monthly, quarterly, daily-of-week).
✅ The series is non-stationary in mean and/or variance (trend, drift).
✅ You need prediction intervals, not just point forecasts.
✅ You need a parsimonious model with interpretable (p,d,q)(P,D,Q) parameters.

Real-world examples

Retail / E-commerce — monthly sales with end-of-year peak.
Energy — daily electricity demand with weekly + yearly seasonality.
Tourism — quarterly hotel occupancy and airline passengers.
Public health — weekly influenza-like-illness counts.
Wildlife / Ecology — monthly species detections and camera-trap RAI trends.

Decision tree

Plot the series. Is there a repeating cycle? → If no, use ARIMA. If yes, continue.
Run ADF on raw series. If non-stationary in mean → set d ≥ 1.
Inspect seasonal differences. If still seasonal → set D = 1.
Read ACF/PACF on differenced series → choose p, q, P, Q.
Fit, compare AIC/BIC across candidates, run Ljung–Box on residuals.

✔️ 8. Assumption Checks

📘 9. How to Use This SARIMA Model Calculator

Enter Your Data — Paste comma-separated values (e.g. 52, 48, 55, 61, 47, …), upload a CSV/Excel file (one cluster per column), or use the manual entry grid. Each cluster = one time series.
Choose a Sample Dataset — Five built-in datasets are available: Monthly Airline Passengers (m=12), Quarterly Retail Sales (m=4), Daily Web Traffic (m=7), Monthly Electricity Demand (m=12), Quarterly Tourism Arrivals (m=4).
Configure Settings — Set the seasonal period m, the SARIMA order (p,d,q)(P,D,Q), the forecast horizon h, and the significance level α.
Run the Analysis — Click 🚀 Run SARIMA Analysis. The tool fits a SARIMA model, computes residuals, and produces forecasts.
Read the Summary Cards — Quick view of AIC, BIC, log-likelihood, residual SD, and Ljung–Box p-value.
Read the Full Results Table — All parameter estimates with standard errors, z-statistics, and 95% CIs.
Examine the Four Visualizations — Original + fitted + forecast, residuals, ACF of residuals, and seasonal decomposition.
Check Assumptions — Stationarity (ADF), residual whiteness (Ljung–Box), normality of residuals.
Read the Detailed Interpretation — 5+ paragraphs explaining your model, its fit, forecast quality, and limitations.
Export Your Results — Use 📋 Download Doc for a plain-text report or 🖨️ Download PDF for a print-ready report.

❓ 10. Frequently Asked Questions

Q1. What is a SARIMA model and when should I use it?

SARIMA (Seasonal AutoRegressive Integrated Moving Average) extends ARIMA by adding seasonal AR, differencing, and MA terms. Use it whenever your time series shows both a long-run trend and a repeating seasonal pattern — for example, monthly retail sales with a December peak, quarterly tourism arrivals, or daily energy demand with a weekly cycle.

Q2. What do the parameters (p, d, q, P, D, Q, m) mean?

Lowercase p, d, q are the non-seasonal AutoRegressive order, differencing degree, and Moving-Average order. Uppercase P, D, Q are the seasonal AR, seasonal differencing, and seasonal MA orders. m is the seasonal period — 12 for monthly data with a yearly cycle, 4 for quarterly, 7 for daily-with-weekly seasonality.

Q3. How do I choose p, d, q, P, D, Q for SARIMA?

Run an Augmented Dickey-Fuller (ADF) test on the series to set d. Inspect a seasonally-differenced series to set D. Then read the ACF and PACF: a PACF cutoff at lag p suggests AR(p); an ACF cutoff at lag q suggests MA(q); spikes at lags m, 2m, 3m suggest seasonal P or Q. Compare two or three candidate models by AIC and BIC.

Q4. How do I interpret AIC and BIC in SARIMA?

Both AIC and BIC penalise model complexity. Lower values are better. AIC favours predictive accuracy; BIC penalises extra parameters more heavily and is preferred for parsimony. If two models differ by less than 2 AIC units, choose the simpler one.

Q5. What is the Ljung–Box test, and what should the p-value look like?

The Ljung–Box test checks whether residuals contain any remaining autocorrelation. The null hypothesis is that the residuals are white noise. You want a non-significant p-value (p > 0.05) — this means the SARIMA model has captured the temporal structure adequately and the residuals look random.

Q6. How large a sample do I need to fit a SARIMA model?

The minimum is two complete seasonal cycles, but three to five cycles is much better. For monthly data (m=12), aim for at least 50–60 observations. For quarterly data (m=4), aim for at least 24. With fewer than two seasonal cycles, the seasonal terms cannot be reliably estimated.

Q7. What is the difference between SARIMA, ARIMA, and Holt–Winters?

ARIMA models trend and short-run dynamics only. SARIMA adds explicit seasonal AR/MA terms. Holt–Winters uses exponential smoothing on level, trend, and seasonality; it is simpler and faster but offers fewer formal diagnostics. SARIMA is preferred when you need confidence intervals, hypothesis tests on parameters, and tight residual diagnostics.

Q8. Can I use SARIMA for non-seasonal data?

Yes. Set P = D = Q = 0 and SARIMA collapses to a standard ARIMA(p, d, q) model. But if your data has any visible seasonality, fitting SARIMA with P, D, or Q > 0 will almost always produce better forecasts than plain ARIMA.

Q9. How do I report SARIMA results in APA 7th edition format?

Use the format: A SARIMA(p,d,q)(P,D,Q)_m model was fitted to the series. The model achieved AIC = ___, BIC = ___, and log-likelihood = ___. Residual diagnostics indicated white-noise residuals (Ljung–Box Q = ___, p = ___). Forecasts for the next h periods are reported with 95% prediction intervals. The Section 5 of this tool generates five reporting templates auto-filled with your statistics.

Q10. What if my SARIMA residuals are not white noise?

If Ljung–Box rejects (p < 0.05), the model is misspecified. Try increasing p or q by 1, adding a seasonal term, or applying a Box–Cox / log transformation to stabilise variance. Re-fit and re-test until residuals look like white noise. Persistent autocorrelation may indicate a structural break, level shift, or omitted external regressor.

🎯 11. Detailed Conclusion

📚 12. References

The following peer-reviewed sources support the methodology behind this SARIMA model calculator, seasonal ARIMA forecasting tool, and time series analysis calculator. All references are cited in APA 7th edition format with DOIs / URLs where available.

Box, G. E. P., Jenkins, G. M., Reinsel, G. C., & Ljung, G. M. (2015). Time series analysis: Forecasting and control (5th ed.). John Wiley & Sons. Wiley
Hyndman, R. J., & Athanasopoulos, G. (2021). Forecasting: Principles and practice (3rd ed.). OTexts. https://otexts.com/fpp3/
Hyndman, R. J., & Khandakar, Y. (2008). Automatic time series forecasting: The forecast package for R. Journal of Statistical Software, 27(3), 1–22. https://doi.org/10.18637/jss.v027.i03
Ljung, G. M., & Box, G. E. P. (1978). On a measure of lack of fit in time series models. Biometrika, 65(2), 297–303. https://doi.org/10.1093/biomet/65.2.297
Dickey, D. A., & Fuller, W. A. (1979). Distribution of the estimators for autoregressive time series with a unit root. Journal of the American Statistical Association, 74(366a), 427–431. https://doi.org/10.1080/01621459.1979.10482531
Akaike, H. (1974). A new look at the statistical model identification. IEEE Transactions on Automatic Control, 19(6), 716–723. https://doi.org/10.1109/TAC.1974.1100705
Schwarz, G. (1978). Estimating the dimension of a model. Annals of Statistics, 6(2), 461–464. https://doi.org/10.1214/aos/1176344136
Shumway, R. H., & Stoffer, D. S. (2017). Time series analysis and its applications: With R examples (4th ed.). Springer. https://doi.org/10.1007/978-3-319-52452-8
Brockwell, P. J., & Davis, R. A. (2016). Introduction to time series and forecasting (3rd ed.). Springer. https://doi.org/10.1007/978-3-319-29854-2
Cryer, J. D., & Chan, K.-S. (2008). Time series analysis: With applications in R (2nd ed.). Springer. https://doi.org/10.1007/978-0-387-75959-3
Cleveland, R. B., Cleveland, W. S., McRae, J. E., & Terpenning, I. (1990). STL: A seasonal-trend decomposition procedure based on Loess. Journal of Official Statistics, 6(1), 3–73. JOS PDF
Makridakis, S., Spiliotis, E., & Assimakopoulos, V. (2018). The M4 competition: Results, findings, conclusion and way forward. International Journal of Forecasting, 34(4), 802–808. https://doi.org/10.1016/j.ijforecast.2018.06.001
R Core Team. (2023). R: A language and environment for statistical computing. R Foundation for Statistical Computing. https://www.R-project.org/
Seabold, S., & Perktold, J. (2010). Statsmodels: Econometric and statistical modeling with Python. Proceedings of the 9th Python in Science Conference, 92–96. https://doi.org/10.25080/Majora-92bf1922-011
STATS UNLOCK. (2025). SARIMA model calculator. https://statsunlock.com