Wygładzanie wykładnicze i quiz Holta-Wintersa

Odpowiedzi i wyjaśnienia pojawiają się w miarę rozwiązywania quizu.

1. W prostym wygładzaniu wykładniczym, co robi większa alfa?
   • It gives the newest observation more influence on the updated level
   • It increases the number of seasonal periods
   • It sets every forecast equal to the sample maximum
   • It removes the level state
2. Which component does Holt's method add beyond simple exponential smoothing?
   • A seasonal state
   • A trend state
   • A moving-average error component
   • A multiplicative response link
3. Kiedy addytywna sezonowość może być rozsądnym wyborem?
   • There is no recurring seasonal pattern to model
   • Seasonal amplitude grows in proportion to the series level
   • Seasonal effects remain roughly fixed in absolute size
   • Seasonal variation should be omitted from the forecast
4. Kiedy można rozważyć sezonowość multiplikatywną?
   • The seasonal period is unknown by definition
   • Seasonal effects remain the same number of units
   • All observed values are negative
   • Seasonal variation scales with the series level
5. Co kontroluje parametr wygładzania beta w metodzie Holta?
   • The weight for updating the level
   • How the trend state is updated
   • The weight for updating the seasonal state
   • The damping factor for extrapolating the trend
6. Które ustawienie określa liczbę obserwacji w cyklu sezonowym?
   • The smoothing weight used for seasonal updates
   • The number of periods used in seasonal differencing
   • The number of observations in a complete seasonal cycle
   • The number of forecast steps returned
7. Why compare a Holt-Winters model with a seasonal-naive baseline?
   • To force all smoothing parameters to zero
   • To prove the seasonal period is exact
   • To avoid inspecting residual seasonality
   • To check whether added smoothing structure improves later forecasts
8. How does a damped trend affect longer-horizon extrapolation?
   • It reduces how strongly the trend continues into longer forecasts
   • It extrapolates the estimated trend unchanged at every horizon
   • It removes the seasonal state after the first forecast
   • It forces the level update to zero after one step