310 likes | 502 Vues
Time Series Analysis. Predicting future sales from past numbers. What is it?. This is not as difficult as it first appears so do not panic! It is used to forecast future sales from past data. If we have a sales pattern that has grown like this one . Sales. Time.
E N D
Time Series Analysis Predicting future sales from past numbers
What is it? • This is not as difficult as it first appears so do not panic! • It is used to forecast future sales from past data. If we have a sales pattern that has grown like this one ...
Sales Time
Then we can predict the future Sales N.B. May not always happen! Time
But what if it looked like this? Sales Time
This is where time series analysis comes in. • It ‘irons out’ the peaks and troughs in data to give a roughly smooth line which you can then extend into the future • N.B. The further into the future you go the less reliable the extrapolation becomes.
How does it do this? • It does this by taking an average of figures over a time period, for example an average of the 4 quarters in a year. Then the first time period is dropped off the average and the next one is added.
For example • Mon Tues Wed First number • Tues Wed Thurs Second number • Wed Thurs Fri Third number • Thurs Fri Sat Fourth number • And so on…. • It is called a moving average.
For example…. It is important that the answer is between feb and March and not with either one. This is called Centering 10+11+13+7 4 =10.25
Then… 10.25 11+13+7+12 4 =10.75
And then… 10.25 10.75 13+7+12+13 4 =11.25
Until… 10.25 10.75 11.25 12 11.75 12 12.5 12.5 12
If it is still not flat enough… • Then you can take an average of the moving average (usually only 2 each time) • This also centres the data for you automatically. (So it falls exactly on a month and not between two)
Like so… 10.25 10.75 11.25 12 11.75 12 12.5 12.5 12
How accurate? • How often did the actual sales coincide with the trend? • Why should this start to happen in the future?
Variation • Calculate the difference between the moving ave figures and the actual numbers. • Positive and negatives are important. • This is why we centred the data!
10.25 10.75 11.25 12 11.75 12 12.5 12.5 12
Therefore… • -2.5-4+0.375+1.125+4.125-6.5+0.5+2.75 8 =-0.515 Therefore, on average the actual line is 0.515 below the trend line. So we should allow for this in our extrapolation
Variation • Then find the average variation, and just add (or subtract) it to the extrapolated prediction. • This will on average cover the variations between the trend and the actual numbers.
We can do better however. Seasonal Variations
Seasonal what? • Many firms have busy and quiet times • For example….
We can take account of this • If sales in January are always low then we can make our predictions more accurate by taking this into account. • Instead of working out the variance for all time periods we could look at the variance for Januarys, then Februarys, then all Marches and so on.
How? • In the same way. Find the average variation for the time you are interested in and then add it to (or take it away from) your prediction.
Beware • Past performance does not mean it will carry on for ever • The further into the future you go the less reliable it becomes.