12. Predictions with iNZight

Predicting with iNZight

Predictions uses the Holt-Winters method, which puts greater emphasis on recent data

Graph of predictions with 95% upper and lower limits for predictions (95% confidence interval for prediction interval)

And predictions in table form

By default will give two years of predictions

Holt-Winters method

uses exponential smoothing where recent data has greater weight, and old data has low weight.
- The weighting changes by a constant ratio (exponential decay).
Smooth's the trend, calculates the seasonal component and then puts these together and produces a prediction.
It is an Additive Model:
- Model = trend + seasonal effect.

Predictions need to be rounded appropriately and with units (remember units and scale eg 1000 000km²)

Discuss the predictions in context

Confidence Intervals: We can be 95% sure that the predictions for (time period) will be between ___ and ___

Exemplar

Prediction Video (you tube video by Priscilla Allan)

Discuss IN CONTEXT and with appropriate rounding & units

eg the total number of visitors to NZ in Jan 2013 is predicted to be 198 300

We can be 95% confident the number of visitors to NZ for Jan 2013 will be between 187000 and 209500 people.

The is a reasonably large variation in the 95% prediction limits (A variation of 27600 people between the limits in June 2013)

Be careful if the end of the trend line has been influenced by the position in the seasonal cycle of the end point.

Here the raw data finishes with more recent data below the trend line, so the trend line drops at the end.

As a result the predictions are low.

Here the last 3 data values have been removed to level the end of the trend line.

Notice the predictions are more appropriate.

Here 3 more data values have been removed making the trend line rise.

Notice the predictions are higher.

Here are the three situations together:

This discussion is worth noting and exploring for EXCELLENCE.

Google Sites

Report abuse