The museum is busiest in quarter 3, but by how much? The seasonal component of a quarter measures how far its values usually sit above or below the trend. Add it to the trend line, extended into the future, and you have a forecast. This lesson calculates both, and says how far to trust the forecast.
about 25 min1 simulation1 practice question
Lesson 3 of 3Show all lessonsHide
Try the code and the simulations beside the lesson.
What you will be able to do
calculate seasonal variations and seasonal components from a table of values and moving averages;
find them from a time series graph and its trend line;
forecast a future value from the trend line and a seasonal component, and state the assumptions and limits.
Seasonal variation
The seasonal variation at a time is how far the real value sits above or below the trend there:
seasonal variation=actual value−centred moving average
For the museum at quarter 3 of year 1: 64−51=13. The value is 13 thousand above the trend. At quarter 4 of year 1: 42−53=−11, so 11 thousand below.
Quarter t
Actual
Centred MA
3
64
51
4
42
53
5
46
55
6
64
57
7
72
58.5
8
50
60.5
9
50
63
10
76
65
The seasonal variations are 13, −11, −9, 7, 13.5, −10.5, −13 and 11.
The seasonal component
Each season appears twice. The seasonal component of a season is the mean of its seasonal variations. It is also called the average seasonal variation.
Quarter 1: 2−9+(−13)=−11.
Quarter 2: 27+11=9.
Quarter 3: 213+13.5=13.25.
Quarter 4: 2−11+(−10.5)=−10.75.
The four components add up to 0.5, close to zero. That is a good check: over a whole year, the seasons above the trend balance those below it.
From a graph and its trend line
Sometimes you have a graph with a trend line instead of a table. Then read the trend value from the line at a time, and use it in place of the moving average.
The museum's trend line is y=45+2t. At quarter 7 it gives 45+14=59. The actual value is 72, so the seasonal variation is 72−59=13. A value read from a graph is never exact, so components found this way are a little rougher.
Forecasting
To forecast a future value:
Extend the trend line to the future time and read off the trend value.
Add the seasonal component of that season, with its sign.
Quarter 1 of year 4 is t=13. The trend value is 45+2×13=71. The quarter 1 component is −11. The forecast is 71+(−11)=60 thousand visitors.
For quarter 3 of year 4, t=15: 45+30+13.25=88.25 thousand.
The museum's trend line y = 45 + 2t drawn solid over quarters 1 to 12 and dashed over quarters 12 to 16, with the twelve data values as circles. Four star markers show the forecasts for year 4: 60 at quarter 13, 82 at quarter 14, 88.25 at quarter 15 and 66.25 at quarter 16. A short vertical arrow at each forecast runs from the trend line to the forecast.
How far can we trust a forecast?
A forecast rests on assumptions.
The trend carries on in the same straight line.
The seasonal pattern repeats in the same way.
Random variation is small. A forecast cannot predict a strike, a storm or a new exhibition.
And it has limits.
The further ahead, the less reliable. Extending the line far beyond the data is extrapolation, as with a line of best fit.
A line drawn by eye differs a little from person to person, so the forecasts differ too.
Few years of data give components that rest on only two or three values each.
Practise with new numbers
Try it in Python
The program finds the seasonal variations of the museum data, averages them into components, and forecasts all four quarters of year 4.
Your turn.
The usual mistakes
Subtracting the wrong way. Seasonal variation is actual minus trend, so a busy season gives a positive value.
Forgetting the sign of a component. A component of −11 lowers the forecast.
Using one season's variation as its component. Average all the variations of that season.
Forecasting from the last data value. Start from the trend line, not from the last point.
Remember these
Seasonal variation = actual − trend (a centred moving average, or a value read from the trend line).
Seasonal component = the mean of the seasonal variations of one season.
Forecast = trend value + seasonal component.
A forecast assumes that the trend and the pattern carry on. It is less reliable further ahead.
Check yourself
Try each one on your own first. If you are stuck, open a hint. Then check with Show the answer.
The actual value in a quarter is 82, and the centred moving average there is 70. Find the seasonal variation.
The seasonal variations of quarter 2 are 12, 11 and 12. Find the seasonal component.
A trend line gives 200 at quarter 14. That season's component is −18. Forecast the value.
Three seasonal components are 6, −4 and 9. Roughly what should the fourth be?
Give two reasons why a forecast for three years ahead is less reliable than one for next quarter.
Enroll free in Time Series and Moving Averages
Every lesson is open to read. Enrolling also lets you:
Mark lessons done and carry on where you stopped.
Take the tests of this lesson and its course.
Count the course towards the certificate for its chapter.