Paste your time series, tune options, then submit. Results appear above this form.
Use this sample to test outputs before pasting real data.
| Date | Value | Note |
|---|---|---|
| 2025-01-01 | 120 | Lower start-of-year baseline |
| 2025-03-01 | 142 | Early seasonal rise |
| 2025-06-01 | 138 | Mid-year softening |
| 2025-10-01 | 160 | Peak cycle window |
| 2025-12-01 | 170 | Year-end uplift |
| 2026-02-01 | 140 | Cycle restarts with moderate growth |
The tool estimates seasonality by checking autocorrelation at candidate lags and then building multiplicative seasonal indices.
Optional steps (smoothing, detrending, normalization) can improve signal-to-noise before measuring ACF.
- Paste your series as date,value rows or values only.
- Select data frequency to load sensible lag suggestions.
- Choose Auto lags, or enter manual lags you want tested.
- Set smoothing, missing handling, and detrending if needed.
- Press Submit, then review strength, lag, and indices above.
- Export CSV or PDF to share the detected cycle evidence.
What seasonality detection measures
Seasonality is a repeating pattern that returns on a fixed cycle, such as weekly demand, monthly revenue swings, or annual energy loads. This calculator screens a single numeric series and reports whether a dominant cycle exists, plus where the peaks and troughs sit inside that cycle. It is designed for quick model readiness checks before feature engineering and forecasting.
Autocorrelation score and decision threshold
The core signal is autocorrelation at lag k, comparing values today with values k steps earlier. For common frequencies, the suggested lags include 7 and 365 for daily data, 52 for weekly data, and 12 for monthly data. A practical starting threshold is 0.30 for noisy business series; 0.45 is stricter when false positives are costly.
Seasonal indices for interpretable cycles
After selecting the best lag, the tool builds multiplicative seasonal indices where 1.00 represents the overall mean. An index of 1.12 implies roughly 12% above-average activity at that position, while 0.90 implies about 10% below-average. These indices translate directly into planning guidance, promotion calendars, and season-aware baselines.
Preparation controls that change the outcome
Smoothing uses a moving average window to reduce short spikes that can drown a cycle. Detrending removes linear growth so the repeating pattern is measured on a stable level. Missing handling can skip gaps or interpolate them; interpolation is often helpful when sensor outages or reporting delays create holes that break autocorrelation. Normalization supports fair comparisons when magnitude changes over time.
Putting results into forecasting workflows
When seasonality is detected, you can apply seasonal differencing at the detected lag, or add seasonal features such as sine/cosine encodings, holiday flags, and cycle-position indicators. When the signal is weak, expand the lag set, review frequency consistency, and validate across multiple holdout windows. The summary panel also reports mean, standard deviation, and coefficient of variation. CV near 0.10 suggests stable levels, while values above 0.50 indicate heavy variability. Use those statistics to decide whether stronger smoothing or segment-level modeling is necessary. In general, aim for at least two full cycles so the strongest lag is not a coincidence.
What data format does the tool accept?
Paste one point per line as date,value or as a single numeric value. Dates are optional and only used for display. Keep the order consistent and avoid mixing frequencies in the same series.
How do I choose the correct frequency setting?
Pick the interval that matches your sampling: daily, weekly, monthly, or quarterly. The frequency only influences the suggested lag candidates, so select the option that reflects how many steps make a typical cycle in your data.
What does “best seasonal lag” represent?
It is the lag with the highest absolute autocorrelation among the tested lags. If monthly data returns best lag 12, values tend to repeat every 12 points, indicating an annual pattern in month-to-month measurements.
When should I enable linear detrending?
Enable it when the series grows or declines steadily and that trend overwhelms repeating cycles. Detrending stabilizes the level so autocorrelation reflects periodic behavior rather than long-run drift.
What strength threshold is reasonable?
Start around 0.30 for noisy operational series and 0.40–0.50 for cleaner metrics. Raise the threshold if you want fewer false alarms, and lower it when you prefer to catch weak seasonality for further review.
Are the seasonal indices additive or multiplicative?
They are multiplicative indices relative to the overall mean. If your use case is additive, convert by subtracting 1.0 and multiplying by the mean, or model seasonality on a transformed scale that fits your domain.