Time Series and Forecasting - Study Notes
Chapter summary
Statistical data collected, observed or recorded at successive intervals of time form a time series. Its analysis compares past and present data, measures chronological variation and gives a basis for business forecasts. Box and Jenkins, whose names are attached to auto regressive moving average models, open the chapter.
Definitions
- Morris Hamburg: a time series is a set of statistical observations arranged in chronological order.
- Ya-Lun-Chou: a collection of readings belonging to different time periods of some economic variable or composite of variables.
- W.Z. Hirsch: the main objective of analysing a time series is to understand, interpret and evaluate change in economic phenomena.
Uses of time series
- Predicts future values from previously observed values.
- Identifies fluctuations in economics and business.
- Helps evaluate current achievements.
- Used in pattern recognition, signal processing, weather forecasting and earthquake prediction.
Components and models
| Component | Meaning | Example |
|---|---|---|
| Secular trend (T) | Long term upward or downward tendency | Growth of population; deaths falling with better medical care |
| Seasonal variation (S) | Fluctuations within a year over seasons | Lower ice cream sales from November to March |
| Cyclical variation (C) | Up and down swings about the trend over more than a year | Trade cycle: prosperity, decline, depression, recovery |
| Irregular variation (R) | Sporadic changes not due to the other three | Fire loss in a factory |
- Additive model \( Y = T + C + S + R \): components independent; all in the same unit.
- Multiplicative model \( Y = T \times C \times S \times R \): components interdependent; their logarithms are additive.
Methods of measuring trend
1. Graphical (free hand) method
Plot time on the X-axis and values on the Y-axis and draw a smooth curve with about as many points above as below it. Merits: simple, no calculation, works for non-linear trend. Demerits: subjective; different statisticians get different values, so it is not reliable.
2. Semi-averages method
- Divide the series into two equal halves; with an odd number of years omit the middle year.
- Place the average of each half at the middle of its period.
- Yearly change = difference of semi-averages / gap between central years; add or subtract it for other years.
- A negative difference means a decreasing trend.
- Merits: simple, few calculations. Demerits: only for linear trend; affected by extreme values.
3. Moving averages method
- 3-year: add three values, place the moving total against the median year, divide by 3.
- 4-year: place 4-year totals between the 2nd and 3rd years, add consecutive pairs to centre them, and divide the centred total by 8.
- A longer period gives a smoother curve.
- Merits: easy to apply; useful for periodic fluctuations; same result for different persons.
- Demerits: less effective for non-periodic data; choosing the period is difficult; no trend values for the first and last few years.
4. Method of least squares
The sum of squares of deviations of actual and computed values is least, so the line \( y = a + bx \) is the line of best fit. Merits: removes personal bias; gives trend values for all periods; allows forecasts. Demerits: harder calculations; new data need recalculation; ignores cyclical, seasonal and irregular movements; valid only for the immediate future.
| Number of years | Coded variable |
|---|---|
| Odd | \( X = x - \text{first year} \), \( u = X - A \) (A the middle value) |
| Even | \( u = 2X - (n - 1) \) |
Seasonal variation
Needs quarterly, monthly, weekly or daily data. Causes: seasons, weather conditions and social customs. Four methods: simple averages, ratio to trend, percentage moving average and link relatives; the book uses simple averages: arrange the data season-wise, find seasonal averages, find the grand average, and express each seasonal average as a percentage of it.
Cyclical and irregular variation
Cyclic movements are mainly due to the trade cycle. Irregular variation, also called erratic fluctuation, is the residual after trend, seasonal and cyclic variation; Patterson calls it non-recurring and sporadic, and no statistical technique measures or isolates it.
Forecasting
Forecasting analyses past and present conditions to arrive at rough estimates of future conditions. T.S. Lewis and R.A. Fox: using the knowledge we have at one time to estimate what will happen at some future moment.
- Short-term: days, weeks or months; found using seasonal variation.
- Medium-term: one to two years.
- Long-term: many years; found using trend; used for strategic planning.