2 - Simple Exponential Smoothing Flashcards
What does a forecast algorithm do?
Predicts y(t+1) given y (1:t)
How does the simple exponential smoothing works to provide a forecast of y(t+1)?
geometrical decreasing weights
Why can we say that if t is large, the weights’ sum in the average is approximately one?
If j increases, what happens to c?
it becomes smaller
Classic form of the simple exponential smoothing algorithm
Recursive form of the simple exponential smoothing algorithm
What does the alpha parameter tell us in the recursive form?
Forecast error correction form of the simple exponential smoothing algorithm
- It is another way to write it.
- Ho moltiplicato (1-α) con y(t|t−1) e poi ho messo in evidenza α
How is the forecast function in the simple exponential smoothing?
In simple exponential smoothing, the forecast function at time t is assumed to be a constant function of h.
- Praticamente il forecast in t per t+1, t+2 e così via è sempre lo stesso, cioè è flat.
How do we tune (choose) the smoothing parameters α?
- consider a (fine) grid of values of α in (0, 1)
- for each α, use the algorithm to recursively compute the one-step-ahead forecasts, for t = 1, … , T
- “Score” the forecasts and choose the value α that gives the best predictive performance.
What are the popular measure of predictive performance? That are also the most known scores for the forecast error
Scale dependent:
- Mean Absolute Value (MAE)
- Mean Square Error (MSE)
Non scale dependent:
- Mean Absolute Percentage Value (MAPE)
Are trend and seasonal behavior envised in simple exponential smoothing?
Simple exponential smoothing does not envisage a trend nor a seasonal behavior.