TESLTMS (Triple Exponential Smoothing with Linear Trend and Multiplicative Seasonality)
TESLTMS (Triple Exponential Smoothing with Linear Trend and Multiplicative Seasonality) function, we use the hw() function with seasonal = "multiplicative".
This is the preferred model for data where seasonal spikes grow in proportion to the trend.
Example: - A 20% increase every December becomes a larger absolute value each year as the trend increases.
Zero Constraint (Important)
Multiplicative models perform division by seasonal factors.
Rule:
- If your data contains zeros or negative numbers, this function will crash
- In such cases, use:
- TESLTAS (Additive Seasonality Model)
Seasonal Multiplier Concept
-
Additive Model:
→ “Add 50 to the forecast” -
Multiplicative Model:
→ “Multiply the forecast by 1.15”
Key Idea:
- Seasonal effects grow as the trend grows
- Spikes become larger over time
Parameters
- Alpha (α): Level (base smoothing)
- Beta (β): Trend (slope smoothing)
- Gamma (γ): Seasonality (seasonal smoothing)
Key Points
- Uses Holt-Winters exponential smoothing
- Best for growing seasonal patterns
- Seasonal effects are percentage-based
- Requires strictly positive data
- Implemented using
hw(seasonal = "multiplicative")
When to Use
Use TESLTMS when: - Seasonality increases with trend - Data shows proportional seasonal growth - Example: sales, demand, traffic with growth cycles
When NOT to Use
- Data contains 0 values
- Data contains negative values
- Seasonal pattern is constant (use additive instead)