Week 3: Generating Process & Smoothing Methods

Question 1

What do smoothing methods do?

Answer 1

Eliminate randomness using some type of averaging method
- Helps identify underlying component
- Used for short-medium term predictions
- Good for routine basis forecasting

Question 2

What is entire average method?

Answer 2

Appropriate predictor if time series is horizontal (has LEVEL) throughout the whole duration, with no other underlying components - bad predictions

• EXTREME: All observations have the same weight

Question 3

What is the naive method?

Answer 3

• Appropriate predictor if time series is horizontal (has LEVEL) throughout the whole duration, with no other underlying components
• Previous period becomes this periods forecast
• Places 100% weight in most recent observation

Question 4

What is Moving Average (MAq)?

Answer 4

Appropriate predictor if time series is horizontal (has LEVEL) throughout the whole duration, with no other underlying components

• Average of a certain number of the most recent observations
• All prior observations have zero weight

Question 5

Is MAq arbitrary?

Answer 5

Yes, but has factors influencing it -
1. VOLATILITY - The more volatile, the more observations needed

2. LENGTH: Short time series - short MAq
3. PREDICTIVE PERFORMANCE: Can try different variations, the one with the smallest error criterion should be selected

Question 6

How does Simple Exponential Smoothing (SES) distribute weights?

Answer 6

• Appropriate predictor if time series is horizontal (has LEVEL) throughout the whole duration, with no other underlying components
• Puts more weight in Yt-1 (most recent observation), as it it goes backwards, weighting reduces

Question 7

Is SES smoothing constant, alpha (a) arbitrary?

Answer 7

Yes, but has factors influencing it -
1. VOLATILITY - The more volatile, the lower the value of alpha needed to smooth the time series

3. PREDICTIVE PERFORMANCE: Can try different alpha variations, the one with the smallest error criterion should be selected

Question 8

What happens when alpha is lower?

Answer 8

When alpha is lower, more weight is given to the observations in the past

Question 9

What value is the smoothing constant?

Answer 9

0-1 (inclusive)

Question 10

Where should you place more weight in forecasting?

Answer 10

More recent values are most likely to be better predictors and should be given more weight

Question 11

What are the systematic components of time series are influenced by ?

Answer 11

Relevant variables & random components

Question 12

What is the random component?

Answer 12

Unobservable changes, unpredictable events

Question 13

Forecasting requires the model to accurately predict what?

Answer 13

Another sample - the future.

It is worthwhile to try to find a general model that will explain/predict both past and future samples

Question 14

What are the random component assumptions?

Answer 14

1. Zero Mean
2. Constant VARIANCE
3. Symmetric Distribution
4. Uncorrelated with residuals at other points in time (independently derived)

Question 15

How to select best model

Answer 15

This will depend on the type of patterns observed in the time series and the type of model (time series/causal)

• A good forecaster will use several error functions as indicators of forecast performance of models . The lower the better