Sales forecasting and time-series
| English | Chinese | Pinyin |
|---|---|---|
| sales forecasting | 销售预测 | xiāo shòu yù cè |
| time series | 时间序列 | shí jiān xù liè |
| trend | 趋势 | qū shì |
| seasonal | 季节性 | jì jié xìng |
| moving average | 移动平均 | yí dòng píng jūn |
| extrapolation | 外推 | wài tuī |
Predicting next quarter's sales
- How many units should the factory make next quarter? Order too much and stock piles up; too little and you lose sales.
- Sales forecasting 销售预测 uses past data to predict the future.
Sales forecast lab
sales = trend + seasonal effect
Move along a sales trend and see how forecast and actual can separate.
A time series is data that is:
A time series records a variable over regular periods.
The long-run underlying direction of a time series is its ______.
The trend is what moving averages reveal.
Time-series 时间序列 analysis
- A time series is data recorded over regular intervals (e.g. quarterly sales).
- It splits into a long-run trend 趋势, seasonal 季节性 variation, and random noise.
Sales wobble seasonally around an underlying upward trend.
Sales over four quarters are 30, 34, 28, 32. What is the 4-quarter moving average?
(30 + 34 + 28 + 32) ÷ 4 = 124 ÷ 4 = 31.
The main purpose of a moving average is to:
It smooths the data so the trend stands out.
Moving averages 移动平均
- A moving average smooths out short-term ups and downs to reveal the trend.
- A 4-quarter moving average averages each set of four consecutive quarters.
Worked example. Sales of 20, 24, 18, 26 over four quarters average to (20+24+18+26) ÷ 4 = 22 — the smoothed trend value, with the seasonal swings averaged out.
Sales wobble seasonally around a smooth underlying trend
Extrapolation assumes the future will follow the past trend.
That assumption fails if conditions change.
Extrapolation 外推 and its limits
- Extrapolation extends the past trend into the future to forecast.
- It assumes the future behaves like the past — risky if conditions change (a new rival, a recession).
Forecasts are not facts. Extrapolation only works while the trend holds. A sudden shock — new competitor, technology, downturn — can make a confident forecast badly wrong.
You've got it
- a time series splits into trend, seasonal variation and noise
- a moving average smooths data to reveal the underlying trend
- extrapolation projects the trend forward — risky if conditions change