Exploring Data Patterns and

Choosing a Forecasting Technique

Reading: Hanke and Wichern, chap
2/3
• Generally two types of data are of interest to a forecaster:
1. Cross –Sectional Data:
• The objective is to examine this data, then extend the
revealed relationship to the larger population.
• E.g., if we collect data on the age and current maintenance
cost of nine buses run by DTC, a scatter diagram can help us
visualize the relationship between age and maintenance
cost.
• The diagram suggests that if we want to forecast annual
maintenance cost, age can be a useful variable.
2. Time Series:
• Four general types of patterns—horizontal, trend, seasonal
and cyclical.
• When data observations fluctuate around a constant level or
mean, a horizontal pattern exists. This type of series is called
stationary in its mean. E.g., monthly sales for a product that
do not increase or decrease consistently over time would
display a horizontal pattern.
• When data observations grow or decline over an extended
period of time, a trend pattern exists.
• The figure shows a long-term trend (growth) of a time series
variable—housing costs.
• The movement of the variable has generally been upwards
over the 20 year time span shown. Some of the forces that
affect and help explain the trend of a series are population,
price inflation, technological change, consumer preferences
etc.
• Definition: The trend is the long-term component that
represents the growth or decline in the time series over an
extended period of time.
• A cyclical pattern exists, when observations exhibit rises and
falls that are not of a fixed period.
• The cyclical component is the wavelike fluctuation around the
trend that is affected by general economic conditions—
business cycles.
• Definition: The cyclical component is the wave-like
fluctuation around the trend.
• The figure above shows a time series with a cyclical
component. The peak corresponds to an expansion in the
economy (boom) and the valley shows a downturn.
• When fluctuations are influenced by seasonal factors, a
seasonal pattern exists. This pattern tends to repeat itself
year after year.
• For a monthly series, the seasonal component measures the
variability of the series each month.
• For a quarterly series, there are 4 seasonal elements
corresponding to each quarter.
• The figure shows that electrical usage for Washington
residential customers is highest in the first quarter (winter) of
each year.
• Definition: The seasonal component is a pattern of change
that repeats itself year after year.
 Exploring Data Patterns Using Autocorrelation Analysis

• Autocorrelation is the correlation between a variable lagged

one or more periods and itself.
• When a variable is measured over time, observations in
different time periods are often correlated. This correlation is
measured using the autocorrelation coefficient.
• We can also study data patterns including components like
trend and seasonality using autocorrelation.
• The next slide shows a store’s sales data for a year.
Lag k autocorrelation coefficient between observation Yt and
Yt-k which are k periods apart can be calculated by the above
formula.

Y¯ = mean of the series

Yt = observation in time period t
Yt-k = observation k time periods earlier or at time period t-k
• The scatter diagram of the data shows that lag 1
autocorrelation should be positive.
• The autocorrelation coefficient , ϒ1 , for lag 1 is 0.572.
• This implies that successive monthly sales are somewhat
correlated.
• This information gives the store owner some insights into his
time series data on sales.
• The lag 2 autocorrelation coefficient ϒ2 = 0.463 < ϒ1
• Generally as k (lags) increases, the autocorrelation
coefficients become smaller.
• If we plot the different ϒ values for different time lags against
time/lags we get the auto-correlation function , ACF.
• Also called a correlogram—graph of the autocorrelations for
various lags of a time series.
• The horizontal axis at the bottom of the graph shows each
time lag 1, 2, 3 and so on.
• The vertical axis shows the possible range of an
autocorrelation coefficient, -1 to +1.
• The horizontal line in the middle of the graph represents
autocorrelations of zero.
• The three vertical lines in the graph at the three time lags
show the respective autocorrelation coefficients.
• The statistical software usually shows you additional
information to check if these coefficient values are
significantly different from zero.
• The ACF coefficients can also answer many other questions:
1. Are the data random?
If a series is random, autocorrelations between Yt and Yt-k for any
lag k are close to zero. This implies that successive values for a
time series are not related to each other.
2. Do the data have a trend?
If a series has a trend, successive observations are highly
correlated and the autocorrelation coefficients are significantly
different from zero for the first several time lags and then
gradually drop to zero as number of lags increases. The
autocorrelation coefficient for lag 1 is often very large and
declines for successive lags.
3. Do the data have a seasonal pattern?
If series has a seasonal pattern, a significant autocorrelation
coefficient will occur at the seasonal time lag or multiples of the
seasonal lag. E.g., 4 for quarterly data and 12 for monthly data.
 Choosing a Forecasting Technique
• For Stationary Data
A stationary series is one whose mean (and other
characteristics) don’t change over time. This happens when
demand patterns influencing the series are stable.
Forecasting a stationary series involves using
the available history of the series to estimate its mean—which
then becomes the forecast for future periods.
E.g., naïve methods, simple averaging methods,
moving averages, ARMA models
• For Data with a Trend
A time series is said to have a trend if its average value
changes over time.
Forecasting techniques for trending data are used when
a. Increased productivity and new technology lead to changes
in lifestyle. E.g., demand for electronic components increased
with advent of computers
b. Increasing population causes increase in demand for G&S.
E.g., sales revenues of consumer goods, demand for energy
c. Purchasing power of the rupee affects economic variables
due to inflation. E.g., salaries, production costs and prices
d. Market acceptance increases. E.g., growth period in the
lifecycle of a new product.
Techniques: moving averages, Holt’s linear exponential
smoothing, simple regression, growth curves, exponential
models, ARIMA
• For Data with Seasonality
Time series with a pattern of change that repeats itself
year after year.
To develop a seasonal forecast we select either a
multiplicative or additive decomposition method and then
estimate seasonal indices from the history of the series.
These indices are used to include or remove seasonal
effects from forecasts. Process—seasonally adjusting data.
Forecasting techniques are used for seasonal data
when weather influences the variable of interest or the
annual calendar influences variable of interest.????

Techniques: classical decomposition, Winter’s exponential

smoothing, multiple regression, ARIMA models
• For Cyclical Data
Cyclical patterns are difficult to model because they
are not stable.
The wave-like pattern around a trend rarely repeats
itself after fixed intervals and its magnitude also varies.
Forecasting techniques are used for cyclical data
when:
a. The business cycle influences the variable of interest.
b. Shifts in popular tastes occur.
c. Shifts in population occur.
d. Shifts in product life cycle occur.

Techniques: classical decomposition, economic indicators,

econometric models, multiple regression and ARIMA models.