Exploring Data Patterns

The most sophisticated forecasting model will fail if it is applied

to unreliable data.
1. Data should be reliable and accurate (‫)موثوقة و دقيقة‬.
2. Data should be relevant (‫)لها صلة بالموضوع‬. The data should
be representative of the circumstances for which they are
being used.
3. Data should be consistent (‫)مالئمة‬. When definition
concerning data collection changes, adjustment need to be
made to retain consistency in historical patterns. For
example, when government agencies change the mix or the
"market basket" used in determining a cost-of-living index.
Years ago personal computers were not part of the mix of
products being purchased by consumers; now they are.
4. Data should be timely (‫)لها عالقة بالفترة المدروسة‬.

Generally, two types of data: cross-sectional data and a

time series data.

Cross-sectional data: are observations collected at single

point in time.
A time series data: are collected over successive
increments of time

1
';vl';df

Exploring time series Data Patterns:

There are typically four general types of patterns: horizontal,
trend, seasonal, and cyclical.

2
 When data grow or decline over several time periods, a
trend pattern exists. The following Figures show the trend
component:

3
 The trend is a long term component that represents the
growth or decline in the time series over an extended period of
time.

Time Series Plot

The cyclical component is the wavelike fluctuation around

the trend.

4
When data collected over time fluctuate around a constant level
or mean, a horizontal pattern exists. This type of series is to be
stationary in its mean. Monthly sales for a food product that do
not increase or decrease consistently over an extended period
would be considered to have a horizontal pattern.
The seasonal component is a pattern that repeats itself
year after year.

Exploring time Patterns with autocorrelation analysis:

Autocorrelation is the correlation between a variable lagged one
or more time periods and itself.
n
∑ (Y t −Ȳ )( Y t−k −Ȳ )
t=k +1
rk = n
∑ (Y t −Ȳ )2
t=1 k = 0, 1, 2......

Where
r k = the autocorrelation coefficient for a lag of k periods
Ȳ= the mean of the values of the series

5
 Y t = the observation in time period t
Y t−k = the observation k time periods earlier or at time period t - k

The following is monthly sales data: table 3-1

(Y t −Ȳ )
2
t Month Yt Y t−1 Y t−2 Y t −Ȳ Y t−1−Ȳ (Y t −Ȳ ) (Y t−1−Ȳ )
1 Jan 123 -19 361
2 Feb 130 123 -12 -19 144 228
3 Mar 125 130 123 -17 -12 289 204
4 Apr 138 125 130 -4 -17 16 68
5 May 145 138 125 3 -4 9 -12
6 Jun 142 145 138 0 3 0 0
7 Jul 141 142 145 -1 0 1 0
8 Aug 146 141 142 4 -1 16 -4
9 Sep 147 146 141 5 4 25 20
10 Oct 157 147 146 15 5 225 75
11 Nov 150 157 147 8 15 64 120
12 Dec 160 150 157 18 8 324 144
0 1474 843
Example 3.1: compute the lag 1 autocorrelation
coefficient and the lag 2 autocorrelation coefficient.
n
∑ (Y t −Ȳ )(Y t−1 −Ȳ )
t =1+1 843
r1= n
= =. 572
1474
∑ (Y t −Ȳ )2
t=1

We can say that there is a positive lag 1 autocorrelation in this

time series. It is .572. This means that the successive monthly
sales are somewhat correlated with each other.

6
 160

150

140

130

120
120 130 140 150 160

Figure 3-4 Scatter Diagram for Example 3. 1

(Y t −Ȳ )
2
t Month Yt Y t−1 Y t−2 Y t −Ȳ Y t−2−Ȳ (Y t −Ȳ ) (Y t−2−Ȳ )
1 Jan 123 -19 361
2 Feb 130 123 -12 144
3 Mar 125 130 123 -17 -19 289 323
4 Apr 138 125 130 -4 -12 16 48
5 May 145 138 125 3 -17 9 -51
6 Jun 142 145 138 0 -4 0 0
7 Jul 141 142 145 -1 3 1 -3
8 Aug 146 141 142 4 0 16 0
9 Sep 147 146 141 5 -1 25 -5
10 Oct 157 147 146 15 4 225 60
11 Nov 150 157 147 8 5 64 40
12 Dec 160 150 157 18 15 324 270
0 1474 682

n
∑ (Y t −Ȳ )(Y t −2 −Ȳ )
t=2+1 682
r2= n
= =. 463
1474
∑ (Y t −Ȳ )2
t=1

We can say that there is a positive moderate lag 2

autocorrelation in this time series. It is .463. Generally, as the
number of time lag (k) increase, the magnitude of the
autocorrelation decrease.

7
Autocorrelation coefficient for different time lags for a variable
can be used to answer the following questions about a time
series:
1. Are the data random?
2. Do the data have a trend (are they nonstationary)?
3. Are the data stationary?
4. Are the data seasonal?
Are the data random?

If a series is random, the autocorrelation between Y t and

Y t−k for any time lag k are close to zero. The successive
values of a time series are not related to each other.
Do the data have a trend (are they nonstationary)? Are the
data stationary?
If a series has a trend, successive observations are highly
correlated, and the autocorrelation coefficients typically are
significantly different from zero for the first several time lags
and then gradually drop toward zero as the number of lags
increase.

Are the data seasonal?

If a series has a seasonal pattern, a significant autocorrelation
coefficient will occur at the seasonal time lag or multiples of
the seasonal lag. The seasonal lag is 4 for quarterly data and
12 for monthly data.

How does an analyst determine whether an autocorrelation

coefficient is significantly different from zero for data of
table 3-1?
8
 Quenouille (1949) and others have demonstrated that
autocorrelation coefficients of random data have a sampling
distribution that can be approximated by a normal curve with
a mean of zero and an approximated standard deviation of
1/ √ n . Knowing this, the analyst can compare the sample
autocorrelation coefficients with this theoretical sampling
distribution and determine whether, for given time lags, they
come from a population whose mean is zero.
Actually, some software packages use a slightly different
formula, as shown in Equation 3.2, to compute the standard
deviation (or standard error) of the autocorrelation
coefficients. This formula assumes that any autocorrelation
before time lag k is different from zero and any
autocorrelation at time lags greater than or equal to k is
different from zero. For an autocorrelation at time lag 1, the
standard error 1/ √ n is used.
k −1

SE ( r k )=
√ 1+ 2 ∑ r 2i

n
i =1

(3.2)

Where
SE (r k )= the standard error (estimated standard deviation) of the
autocorrelation at time lag k
r i = the autocorrelation at time lag i
k= the time lag
n= the number of observations in the time series

This computation will be demonstrated in Example 3.2

If the series is truly random, almost all of the sample
autocorrelation coefficients should lie within a range specified
by zero, plus or minus a certain number of standard error.

9
 At a specified confidence level, a series can be considered
random if each of the calculated autocorrelation coefficients is
within the interval about 0 given by 0±SE(r k ) , where the
multiplier t is an appropriate percentage point of a t distribution.

Although testing each rk to see if it is individually

significantly different from 0 is useful, it is also good practice to
examine a set of consecutive rk 's as group. We can use a
portmanteau test to see whether the test, say, of the first 10 rk
values, is significantly different from a set in which all values
are zero.
One common portmanteau test is based on the Ljung-Box Q
statistic:
m
r 2k
Q=n (n+2) ∑
k =1 n−k (3.3)

Where
n= the number of observations in the time series
k= the time lag
m= the number of time lags to be tested

r k = the sample autocorrelation function of the residuals lagged k

time periods

Are the data random?

A simple random model, often called a white noise model,
is displayed in Equation 3.4. Observation Y t is composed of
two parts: c, the overall level, and εt , which is the random error
component. It is important to note that the εt component is
assumed to be uncorrelated from period to period.
Y t =c +ε t (3.4)

10
Are the data in table 3-1 consistent with this model? This issue
will be explored in Example 3.2 and 3.3.

Example 3.2
A hypothesis test is developed to determine whether a particular
autocorrelation coefficient is significantly different from zero
for the correlation shown in Figure 3-5. The null and alternative
hypotheses for testing the significance of the lag 1 population
autocorrelation coefficient are
H 0 : ρ1 =0
H 1: : ρ1 ≠0

If the null and hypothesis is true, the test statistic

r 1 −ρ1 r −0 r1
t= = 1 =
SE( r 1 ) SE (r 1 ) SE(r 1 )

Has t distribution with df = n-1. Here, n-1 = 12-1 = 11, so for a

5% significance level, the decision rule is as follows:

If t < 2.2 or t > 2.2, reject H 0 and conclude the lag 1

autocorrelation is significantly different from 0.

The standard error of r1 is SE (r k )=√ 1/12=√ . 083=. 283 , and the

value of the test statistic becomes
r1 . 572
t= = =1 . 98
SE(r 1 ) . 289

Since -2.2 <1.98<2.2, we cannot reject H 0 : ρ1 =0 .

11
 To test for zero autocorrelation at time lag 2, we consider
H 0 : ρ2 =0
H 1: : ρ2 ≠0

and test statistic

r 2 −ρ2 r −0 r2
t= = 2 =
SE(r 2 ) SE (r 2 ) SE( r 2 )

Has t distribution with df = n-1. Here, n-1 = 12-1 = 11, so for a

5% significance level, the decision rule is as follows:

If t < 2.2 or t > 2.2, reject H 0 and conclude the lag 1

autocorrelation is significantly different from 0.

The standard error of r1 is

k −1 2−1

SE ( r k )=
√ 1+ 2 ∑
i=1
n
r 21
=

and the value of the test statistic becomes

√ 1+2 ∑ r 2i

n
i=1
=
√ 1+2( . 572)
12
=
1 . 6544
√
12
=√ . 138=. 371
,

r2 . 463
t= = =1 .25
SE(r 2 ) . 371

Since -2.2 <1.25<2.2, we cannot reject H 0 : ρ2 =0 .

An alternative way to check for significant autocorrelation is to
construct, say, 95% confidence limits centered at 0. The limits
for time lags 1 and 2 are as follows:

Lag 1: 0±t. 025 SE (r 1 ) or 0±2. 2(.289 ) → (-.636, .636)

Lag 2: 0±t. 025 SE (r 2 ) or 0±2.2(.371 ) → (-.816, .816)

Autocorrelation significantly different from 0 is indicated

whenever a value for rk falls outside the corresponding
confidence limits.

12
Example 3. 3
Do the Data Have a Trend?
If a series has a trend, a significance relationship exists between
successive series values. The autocorrelation coefficients are
typically large for the first several time lags and then gradually
drop toward zero as the number of lags increases.
A stationary time series is one whose basic statistical
properties, such as the mean and variance remain constant over
time. Consequently a series that varies about a fixed level (no
growth or decline) over time is said to be stationary. A series
that contains trend is said to be nonstationary. The
autocorrelation coefficients for a stationary series decline to zero
fairly rapidly, generally after the second or third time lag. On
the other hand, sample autocorrelation for nonstationary series
remain fairly large for several time periods. Often, to analyze
nonstationary series, the trend is removed before additional
modeling occurs. The procedures discussed in chapter 9 use this
approach.
A method called differencing can be used to remove the
trend from nonstationary series. The data originally presented in
table 3-1 are shown again in figure 3-8, column A. the Y t
values are lagged one period, Y t−1 , are shown in column B. the
differences, Y t −Y t−1 (column A – column B), are shown in
column C. for example the first value of differences is Y 2−Y 1 =
130-123=7. Note the upward growth or trend of the VCR data
shown in Figure 3-9, Plot A. Now observe the stationary pattern
of the differenced data in figure 3-9, plot B. Differencing the
data has removed the trend.

13
 A B C D E F E
1 Yt Y t−1 Differences
2 123
3 130 123 7
4 125 130 -5
5 138 125 13
6 145 138 7
7 142 145 -3
8 141 142 -1
9 146 141 5
10 147 146 1
11 157 147 10
12 150 157 -7
13 160 150 10
Figure 3-8 Excel Results of Differencing VCR Data of Example 3.1

160
yt
150

140

130

120
1 2 3 4 5 6 7 8 9 10 11 12

Figure 3-9 Time series plots the VCR Data and Differenced VCR Data for
Example 3-1

14
 Differences
15

10

-5

-10
1 2 3 4 5 6 7 8 9 10 11

Example 3.4
An analyst for Sears is assigned the task of forecasting operating
revenue for 2005. She gathers the data for the years 1955 to
2004, shown in table 3-4. The data are plotted as time series in
figure 3-10. Notice that, although Sears operating revenues were
declining over the 2000-2004 period, the general trend over the
entire 1955-2004 time frame is up. First, the analyst computes a
95% confidence interval for the autocorrelation coefficients at
time lag 1 using 0±Z . 025 1 / √n where, for large samples, the
standard normal .025 points has replaced the corresponding t
distribution percentage point:
0±1.96 √ 1/50
0±.277

Table 3-4 Yearly operating revenue for Sears, 1955-2004, for Example 3.4
Year Yt Year Yt Year Yt Year Yt Year Yt
1955 3307 1967 7296 1979 17514 1991 57242 2003 23253
1956 3556 1968 8178 1980 25195 1992 52345 2004 19701
1957 3601 1969 8844 1981 27357 1993 50838
1958 3721 1970 9251 1982 30020 1994 54559
1959 4036 1971 10006 1983 35883 1995 34925
1960 4134 1972 10991 1984 38828 1996 38236
1961 4268 1973 12306 1985 40715 1997 41296
1962 4578 1974 13101 1986 44282 1998 41322
1963 5093 1975 13639 1987 48440 1999 41071
1964 5716 1976 14950 1988 50251 2000 40937
1965 6357 1977 17224 1989 53794 2001 36151
1966 6769 1978 17946 1990 55972 2002 30762

15
 welfjk

Operating Revenue
70000

60000

50000

40000

30000

20000

10000

Next the analyst runs the data on Minitab and produces the
autocorrelation function shown in Figure 3-11. Upon the
examination, the analyst notices that the autocorrelation for the
first four time lags are significantly different from zero (.96, .92,
.87, and .81) and that the values then gradually drop to zero. As
a final check, the analyst looks at the Q statistic for 10 times
lags. He LBQ is 300.56, which is greater than the chi-square
value 18.3 (the upper .05 point of a chi-square distribution with
10 degrees of freedom). This result indicates the autocorrelation
for the first 10 lags as a group are significantly different from
zero. The analyst decides that the data are highly autocorrelated
and exhibit trendlike behavior.
The analyst suspects that the series can be differenced to
remove the trend and to create a stationary series. He differences
the data (see Minitab applications section at end of the chapter),
and the results are shown in Figure 3-12. The differenced series
shows no evidence of a trend, and the autocorrelation function,
16
shown in Figure 3-13, appears to support this conclusion.
Examining Figure 3-13, the analyst notes that the
autocorrelation coefficient at time lag 3, .32 is significantly
different from zero (tested at the .05 significance level). The
autocorrelations at lags other than lag 3 are small, and the LBQ
statistics for 10 lags is also relatively small, so there is little
evidence to suggest the differenced data are autocorrelated. Yet
the analyst wonders whether there is some pattern in these data
can be modeled by one of the more advanced forecasting
techniques discussed in Chapter 9.
Are the Data Seasonal?
If quarterly data with a seasonal pattern are analyzed, first
quarters tend to look alike, second quarters tend to look alike,
and so forth, and a significant autocorrelation coefficient will
appear at lag 4. If monthly data are analyzed, a significant
autocorrelation coefficient will appear at lag 12. That is January
will correlate with other Januarys, February will correlate with
other Februarys, and so on. Example 3.5 discusses a series that
is seasonal.
Example 3.5
Perkin is an analyst for the Costal Marine Corporation. Perkin
gathers the data shown in table 3-5 for the quarterly sales of the
corporation from 1994 to 2006 and plots them as the time series
Table 3-5 Quarterly sales for Costal Marine Corporation,
1994-2006, for example 3.5
year 31-Dec 31-Mar 30-Jun 30-Sep
1994 147.6 251.8 273.1 249.1
1995 139.3 221.2 260.2 259.5
1996 140.5 245.5 298.8 287.0
1997 168.8 322.6 393.5 404.3
1998 259.7 401.1 464.6 479.7
1999 264.4 402.6 411.3 385.9
2000 232.7 309.2 310.7 293.0
2001 205.1 234.4 285.4 285.7

17
 2002 193.2 263.7 292.5 315.2
2003 178.3 274.5 295.4 286.4
2004 190.8 263.5 318.8 305.5
2005 242.6 318.8 329.6 338.2
2006 232.1 285.6 291.0 281.4

graph shown in figure 3-14. Next, he computes a large-sample

600

500

400

300

200

100

95% confidence interval for the autocorrelation coefficient at

time lag 1: 0±1.96 √ 1/52
0±.272

Then Perkin computes the autocorrelation coefficients shown in

figure 3-15. He notes that the autocorrelation coefficients at time
lags 1 and 4 are significantly different from zero ( r 1 =.39>. 272
and r 4=. 74>.333 ). He concludes that the corporation sales are
seasonal on a quarterly basis.

18
Choosing a Forecasting Technique

Some of the questions that must be considered before deciding

on the most appropriate forecasting technique for a particular
problem follow:

 Why is a forecast needed?

 Who will use the forecast?
 What are the characteristics of the variable data?
 What time period is to be forecasted?
 What are the minimum data requirements?
 How much accuracy is desired?
 What will the forecast cost?

To select the appropriate forecasting technique properly, the

forecaster must be able to accomplish the following:

 Define the nature of the forecasting problem.

 Explain the nature of the data under investigation.
 Describe the capability and limitation of potentially useful
forecasting technique.
 Develop some predetermined criteria on which the
selection decision can be made.
A major factor influencing the selection of a forecasting
technique is the identification and understanding of historical
patterns of the data. If trend, cyclical, or seasonal patterns can be
recognized, then techniques that are capable of effectively
extrapolating these patterns can be selected.

19
Forecasting Technique for Stationary Data
A Stationary Series is one whose mean value is not changing
over time.
It is important to recognize that stationary data do not
necessarily vary randomly about the mean level. Stationary
series can be autocorrelated.
Stationary forecasting technique are used in the following
circumstances:

 The forces generating a series have stabilized, and the

environment in which the series is relatively unchanging.
Example:
The number of breakdowns per week on an assembly line
having a uniform production rate.

 A very simple model is needed because of a lack of data or

for ease of explanation or implementation.
Example:
When a business is new and very few historical data are
available.

 Stability may be obtained by making simple corrections

for factors such as population growth or inflation.
Example:
Changing income to per capita income amounts.

 The series may be transformed into a stable one.

20
Example:
Transforming a series by taking logarithms, square roots, or
differences.

 The series is a set of forecast errors from a forecasting

technique that is considered adequate.
Example:
(See Example 3.7 on p. 85.)
Techniques that should be considered when forecasting
stationary series include naive method, simple averaging
methods, moving averages, and autoregressive moving averages
(ARMA) models (Box-Jenkins methods).

Forecasting Technique for Data with trend

A trend in time series is a persistent, long-term growth or
decline.
For a tending time series, the level of the series is not constant.
It is common for economic time series to contain a trend.
Forecasting technique for trending data are used in the following
circumstances:

 Increased productivity and new technology lead to change

in lifestile.
Example:
The demand for electronic components, which increased with
advent of the computer.
21
  Increasing population causes increases in demand for
goods and services.
Example:
Increases of the sales revenues of computer goods.

 The purchasing power of the dollar affects the economic

variables due to inflation.
Example:
Salaries, production costs, and prices.
Techniques that should be considered when forecasting tending
series include moving averages, holt's linear exponential
smoothing, simple regression, growth curves, exponential
models, and autoregressive integrated moving average
(ARIMA) models (Box-Jenkins methods).

Forecasting Technique for Seasonal Data

A time series is a series with a pattern of change that repeat
itself year after year.
Forecasting technique for seasonal data are used in the
following circumstances:

 Weather influences the variable of interest.

Example:
Summer and winter influence activities (e.g., sports such as
skiing), and clothing.

 The annual calendar influences the variable of interest.

22
Example:
Retail sales influenced by holidays.
Techniques that should be considered when forecasting seasonal
series include classical decomposition, Census X-12, Winter's
exponential smoothing, multiple regression, and autoregressive
integrated moving average (ARIMA) models (Box-Jenkins
methods).

Forecasting Technique for Cyclical Data

A cyclical effect is as wavelike fluctuation around the trend.
Decomposition methods can be extended to analyze cyclical
data.
Forecasting technique for seasonal data are used in the
following circumstances:

 The business cycle influences the variable of interest.

Example:
Variables like economic, market, and competition factors.

 Shifts in popular tastes occur.

Example:
Shifts like fashions, music, and food.

 Shifts in population occur.

Example:
Shifts like because of war, famines, epidemics and natural
disasters.

23
Techniques that should be considered when forecasting cyclical
series include classical decomposition, economic indicators,
econometric models, multiple regression, and autoregressive
integrated moving average (ARIMA) models (Box-Jenkins
methods).

24