Lesson 7 of 7 · 20 min

Forecasting and solving time series problems

Forecast future values by extending the trend line and re-seasonalising with the seasonal index, judge how reliable a forecast is, and carry out a full time series analysis.

What you will be able to do

  • Forecast a future value: find its tt, substitute into the trend line, then multiply by the seasonal index.
  • Judge the reliability of a forecast: extrapolation, strength of the trend, and whether the pattern will continue.
  • Carry out a full time series analysis to answer a practical question.

The holiday-park manager now has everything she needs: seasonal indices for each quarter and a trend line for the deseasonalised guest nights. To forecast next summer she runs the analysis backwards. The trend line says what an average quarter will look like then; the seasonal index turns that into the quarter she actually cares about.

Forecasting from a trend line

  1. Find t for the period you want

    If t=1t = 1 is Q1 2022, then Q1 2025 is t=13t = 13.

  2. Substitute into the trend line

    Deseasonalised guest nights =0.556×13+28.885=36.113= 0.556 \times 13 + 28.885 = 36.113 thousand. This is the trend value: a deseasonalised forecast.

  3. Re-seasonalise

    Forecast == trend value ×\times seasonal index =36.113×1.262=45.57= 36.113 \times 1.262 = 45.57 thousand.

  4. Answer in context

    About 45 600 guest nights are forecast for Q1 2025.

Interactive

Forecasting guest nights

The dashed line extends the trend; the hollow points are the re-seasonalised forecasts. Change the horizon to forecast further ahead — and ask yourself how much you would trust the last one.

Worked example

Forecast a year of guest nights

  • Complex familiar

For the holiday park, deseasonalised guest nights=0.556t+28.885\text{deseasonalised guest nights} = 0.556t + 28.885 (thousands, t=1t = 1 is Q1 2022). The seasonal indices are Q1 1.262, Q2 0.698, Q3 0.964 and Q4 1.076.

Forecast the guest nights for each quarter of 2025, in thousands, correct to one decimal place.

4 steps

Try it yourself first, then check the working one step at a time.

Method

  1. Find tt for each period to be forecast.

  2. Substitute into the trend line to get the trend (deseasonalised) value.

  3. Multiply by the season's index.

Watch out for

  • Using t=1t = 1 to 44 for 2025, or t=2025t = 2025.

  • Stopping at the trend value (36.1 for Q1) without multiplying by the index.

Using deseasonalised guest nights=0.556t+28.885\text{deseasonalised guest nights} = 0.556t + 28.885 (thousands, t=1t = 1 is Q1 2022) and the Q2 seasonal index 0.698, forecast the number of guest nights in Q2 2025, in thousands, correct to one decimal place.

in thousand

Round to 1 decimal place.

Key idea

How reliable is a forecast?

A forecast is always an extrapolation: it assumes the trend and the seasonal pattern carry on beyond the data. Judge it on three things:

  1. How far ahead? One or two periods past the data is far more reliable than five years.
  2. How strong is the trend? A trend line with rr close to ±1\pm 1 (fitted to the deseasonalised values) and small irregular fluctuations gives more reliable forecasts.
  3. Will the pattern continue? New competition, a pandemic, a price change or a cyclone can break both the trend and the seasonal pattern.

The holiday-park data runs from Q1 2022 to Q4 2024. Which forecast is likely to be the least reliable?

Choose one answer

A full time series analysis

  1. Plot and describe

    Construct the time series plot. Describe the trend, seasonality and irregular fluctuations in context.

  2. Smooth if it helps

    A moving mean or median (odd window) shows the trend through the noise. Use the median if there are outliers.

  3. Seasonal indices

    If there is seasonality, calculate the indices by the average percentage method (complete years only) and check their sum.

  4. Deseasonalise

    Divide each value by its seasonal index.

  5. Fit the trend line

    Least-squares line of deseasonalised value against tt. Interpret the slope.

  6. Forecast and evaluate

    Trend value × seasonal index. Answer the question in context, and comment on reliability.

Worked example

Should the store hire for Christmas? A full time series analysis

  • Complex unfamiliar
  • Technology active

A Townsville outdoor and camping store recorded its quarterly sales ($'000). Data are illustrative.

YearQ1Q2Q3Q4
2021113105123164
2022124113129182
2023130125141190
2024142130151206

The owner hires an extra full-time staff member for any quarter with forecast sales above $220 000. Should she hire one for Q4 2025? Carry out a time series analysis to decide.

Quarterly sales ($'000), t=1t = 1 is Q1 2021.

6 steps

Try it yourself first, then check the working one step at a time.

Method

  1. Plot and describe the series.

  2. Find seasonal indices by the average percentage method.

  3. Deseasonalise and fit a trend line against tt.

  4. Forecast (trend value × index), then answer the question with a reliability comment.

Watch out for

  • Comparing the trend value (169.41) with $220 000 — it must be re-seasonalised first.

  • Using t=17t = 17 (Q1 2025) instead of t=20t = 20 for Q4 2025.

  • Giving a yes/no answer with no comment on how close the forecast is to the threshold.

Exam tip

Answer the question that was asked

Problem-solving questions end with a decision: should the store hire extra staff, will the dam reach capacity, is the tourism campaign working? Finish with a sentence that answers it directly, using your forecast, and add a reliability comment ("this is an extrapolation four quarters beyond the data, but the trend is strong (r=0.96r = 0.96), so the forecast is reasonably reliable if conditions don't change").

Common mistake

Forgetting to re-seasonalise

The trend line gives deseasonalised values. A forecast of "36.1 thousand guest nights in Q1 2025" is the trend value, and it is too low for a peak quarter by about 9 thousand. Always multiply by the season's index unless the question asks for the trend (deseasonalised) value.

Summary

  • Forecast = trend value × seasonal index, where the trend value comes from substituting tt into the trend line.

  • Get tt right: count on from t=1t = 1, four per year for quarters, twelve per year for months.

  • Every forecast is an extrapolation. It is more reliable close to the data, with a strong trend and a stable seasonal pattern.

  • In a practical problem: plot, describe, find indices, deseasonalise, fit the trend, forecast, then answer the question with a reliability comment.

Finished the lesson?

Try the 2 quick checks above and the lesson completes itself, or mark it complete now.

Lesson 7 of 70% read

Save your progress with a free account