Skip to content

Double & Triple Exponential Smoothing

1. Double Exponential Smoothing (Holt's Method)

Double Exponential Smoothing, also known as Second Order (SO) Exponential Smoothing or Holt's trend-corrected smoothing, is applied when a time series exhibits a linear trend but has no seasonal pattern. It introduces a second smoothing constant, \(\beta\), to account for the trend component (\(b_{t}\)).

Mathematical Formulation:
* Initialization:
$\(s_{0} = y_{0}\)$
* Level Equation (\(s_{t}\)):
$\(s_{t} = \alpha y_{t} + (1-\alpha)(s_{t-1} + b_{t-1})\)$
* Trend Equation (\(b_{t}\)):
$\(b_{t} = \beta(s_{t}-s_{t-1}) + (1-\beta)b_{t-1}\)$

Where:
* \(b_{t}\) is the best estimate of the trend at time \(t\).
* \(0 < \beta < 1\) is the trend smoothing factor.

2. Triple Exponential Smoothing (Holt-Winters Method)

This method is utilized for forecasting time series data that contains both a linear trend and a seasonal pattern.

Key Notations:

  • \(s_{t}\): Smoothed statistic (level).
  • \(b_{t}\): Best estimate of trend.
  • \(c_{t}\): Sequence of seasonal correction factors.
  • \(\alpha, \beta, \gamma\): Smoothing factors for level, trend, and seasonality respectively (all between \(0\) and \(1\)).
  • \(L\): Length of the seasonal cycle (e.g., \(L=12\) for monthly data).

Seasonality Types:

  • Additive Seasonality: The seasonal effect remains roughly constant as the trend changes (\(Y_{t} = T_{t} + S_{t} + e_{t}\)).
  • Multiplicative Seasonality: Seasonal fluctuations increase or decrease in proportion to the level of the time series (\(Y_{t} = T_{t} \times S_{t} \times e_{t}\)).

Recursive Equations:

Component Multiplicative Seasonality Additive Seasonality
Level (\(s_{t}\)) \(s_{t} = \alpha \frac{y_{t}}{c_{t-L}} + (1-\alpha)(s_{t-1}+b_{t-1})\) \(s_{t} = \alpha(y_{t} - c_{t-L}) + (1-\alpha)(s_{t-1}+b_{t-1})\)
Trend (\(b_{t}\)) \(b_{t} = \beta(s_{t}-s_{t-1}) + (1-\beta)b_{t-1}\) \(b_{t} = \beta(s_{t} - s_{t-1}) + (1-\beta)b_{t-1}\)
Season (\(c_{t}\)) \(c_{t} = \gamma \frac{y_{t}}{s_{t}} + (1-\gamma) c_{t-L}\) \(c_{t} = \gamma(y_{t} - s_{t-1} - b_{t-1}) + (1-\gamma) c_{t-L}\)

3. Implementation and Visualization

In practice, Holt-Winters filtering (another term for smoothing) can be implemented using statistical software like R.

```r

Example: Applying Holt-Winters without seasonal component (gamma = F)

HoltWinters(x = AirPassengers, gamma = F)