← All methods

TBATS · Pro

TBATS-lite

Trigonometric seasonality stand-in for multiple seasonalities (Pro).

What it assumes

Trigonometric seasonality (lite stand-in for full TBATS).

When to use

Multiple seasonalities (hourly energy).

When to avoid

Very short series.

Knobs

Fourier + trend (Pro lite).

How it works

Full TBATS handles several seasonal periods (hour-of-day and day-of-week together). The Pro lite in this engine is a trigonometric / Fourier + trend stand-in, not a complete De Livera–Hyndman–Snyder state space. It is aimed at longer series such as hourly energy. Very short n is excluded. Excel Forecast Sheet cannot tell that story at all.

Sample forecast question

Sixteen hours of load: night is quiet, evening peaks. What is the TBATS-lite forecast for the next eight hours?

Step-by-step on these numbers

Sample series (MW)
PeriodActualFitted
06:005855.20
07:005454.38
08:005150.84
09:004950.53
10:004849.49
11:005053.67
12:006261.63
13:007874.82
14:008687.69
15:008484.73
16:008079.04
17:007776.59
18:007673.40
19:007875.44
20:008281.26
21:008892.30
Forecast
PeriodForecast95% interval
22:00103.0298.70 – 107.35
23:00100.0793.95 – 106.19
00:0094.3886.89 – 101.88
01:0091.9283.27 – 100.58
02:0088.7479.06 – 98.41
03:0090.7880.18 – 101.38
04:0096.5985.14 – 108.04
05:00107.6395.39 – 119.87

Parameters the engine found

  • cp = 8
  • m = 8
  • k = 9
  • b0 = 42.105
  • b1 = 4.061
  • b2 = -2.144
  • b3 = 2.940
  • b4 = 9.909
  • b5 = -0.261
  • b6 = 2.114
  • b7 = 0.212
  • b8 = 1.070
  1. The question

    Sixteen hours of load: night is quiet, evening peaks. What is the TBATS-lite forecast for the next eight hours? Sample hours: 06:00=58, 07:00=54, 08:00=51, 09:00=49, 10:00=48, 11:00=50, 12:00=62, 13:00=78, 14:00=86, 15:00=84, 16:00=80, 17:00=77, 18:00=76, 19:00=78, 20:00=82, 21:00=88 (MW).

  2. Lite stand-in

    Full TBATS handles several seasonal periods. This engine’s Pro lite is a trigonometric / Fourier + piecewise trend stand-in (same family as the piecewise-trend page), not the complete De Livera–Hyndman–Snyder state space.

  3. This sample

    n = 16, seasonal period m = 8 (intra-day-ish wave on 16 hours). Changepoint at 8; 9 OLS terms.

  4. Forecast

    Project the Fourier + trend design forward. 22:00: 103.02; 23:00: 100.07; 00:00: 94.38; 01:00: 91.92; 02:00: 88.74; 03:00: 90.78; 04:00: 96.59; 05:00: 107.63.

  5. Interval

    Residual σ = 2.208. For h = 1 the 95% band is [98.70, 107.35] around 103.02. The engine labels this a residual-Gaussian heuristic (σ√h), not a simulation interval.

Graph of this sample

06:00 · 58 21:00 22:00
TBATS-lite on the canned sample: actuals, fitted, and a 8-step forecast. Actual Fitted Forecast
The chart plots the canned table on this page (actuals, in-sample fitted, forecast, 95% interval). Not your Excel series. Free analysis never uploads raw data.

Educational only. Not investment, weather, or operational advice. In the add-in, rank is rolling-origin MASE — “best supported among candidates on this series.”

Use it in the Excel Add-in