Analyze On The Forecasting Demand

Jul. 2005, Volume 4, No.7 (Serial No.25)

China-USA Business Review, ISSN 1537-1514, USA

Analysis on the Forecasting Demand of Front-line Worker
Zhenzhu Zhang∗

Tianjin University

Abstract: The importance of personnel forecast is pointed out in this paper, and also the author expounds
comparative analysis through instances among several methods of personnel forecast.
Key words: forecasting methods time series analysis causal forecasting

1. Introduction of Forecasting Methods
The normal forecasting methods are divided into three types, which are qualitative forecasting, time series
forecasting and causal forecasting [1].
Qualitative forecasting is used in the environment that lacks the statistic history information or the turn in the
course of events [2]. The primary methods are manager’s opinion, jury of executive opinion, sales force composite,
consumer market survey, Delphi method, and so on.
The method of time series analysis is used in condition that has enough statistic history information. The
types are set out in Table 1.
Table 1 The Methods of Time Series Analysis
The name of
forecasting
method
Last-value

Account method
Forecasting value
= Last value

[F]
[L]

Character

Application

No relativity between
value

Unstable time series

Average

F = Average of all values

Notable relativity between
value

Quite stable time
series

Moving
average

F = Average of the latest n values

Relativity between
contiguous value, the
data’s number reflect the

degree of stabilization

Moderate stable time
series

Adjusting α to adapt the
different stabilization

Time series from
unstable to quite
stable

Change slowly or change
by chance

The equal value of
probability
distributing changes
upwards or
downwards

Exponential
smoothing

F = α*L + (1-α)*(Last Forecasting Value)
[LF] α ∈

( 0,1) α: Smoothing constant

F = α* L + (1-α)*(LF)

+ (Trend estimate) [T]
T = β (The latest trend)[TL] + (1-β) (Last
Exponential
smoothing with
trend

∗

( )

time trend estimate) β ∈ 0,1
TL = α (L - Reciprocal second time
value) + (1-α) (Last time forecasting
value - Reciprocal second time
forecasting value)
β: Trend smoothing constant

Zhenzhu Zhang (1974-), female, Ph.D. candidate of School of Management, Tianjin University, Tianjin, China, Postcode: 300072;
Main research field: Supply Chain Management; Tel: 13011332163; E-mail:

64


Analysis on the Forecasting Demand of Front-line Worker

There is another method belonging to the time series analysis, which is called ARIMA or Box-Jenkins
method. It is so complicated that is achieved by software. It is applied in the problem, which has visible time
variety.
Causal forecasting confirms the linear or nonlinear relations between dependent variable and independent
variable through XY (Scatter) chart of observation data. When there are many independent variables causing the

variety of dependent variable, we call it as multi-linear regression. On the great mass of application, the nonlinear
relation can be changed into linear relation through forecasting the relation of dependent variable and independent
variable [3].
In the course of applying the forecasting, the work named model diagnose needs to choose or improve on the
original model according to the series of target value. The target value which is used to judge the model is mainly
as mean absolute deviation (MAD), mean square (MSE), R2, Adj.R2 and so on. The iterative work to diagnose
model make the target value achieve the ideal precision. Then, the opposite model of forecasting gains
optimization. In this paper, we use MAD and MSE as the target value. Their formulas are as follows [4]:
MAD the sum of forecasting error / the time of forecasting;
MSE the square sum of forecasting error / the time of forecasting

2. Analyzing from Example
The company A is a non-shop that sells commodities for pregnant woman and baby through call center to
order and confirm the price. There, the products are mainly comprised by high quality eatable nurture, clothing,
toy, washing commodity, interrelated books, magazines and remembrance. Every year, the company posts the
catalogs of product to plenty of users or potential consumers. The users are told to purchase through the telephone
number, which is printed on the catalog and then is connected to call center. Usually, we estimate the number of
client representation in the specific period of time through the statistic of call numbers. In this paper, we use the
above forecasting model to analyze the forecasting of front-line worker from this example. Now we know the sum
of call numbers in every quarter in the last three years as follows.
Table 2 The Every Quarter’s Total Call Numbers and Sales of Company A in the Last Three Years
The first year

Name

The second year

The third year

1

2

3

4

1

2

3

4

1

2

3

4

Call
number

6809

6465

6569

8266

7257

7064

7784

8724

6992

6822

7949

9650

Sale
century

4894

4703

4748

5844

5192

5086

5511

6107

5052

4985

5576

6647

We use excel and the Software-Crystal Ball 2000 to process modular arithmetic. In Table 2, the method of
causal forecasting uses sales as independent variable and the linear regression equation as follows:
y a b x, through this model, we use the method of least squares to confirm, then we get a = -1223.86; b =
1.6324. The result is in Table 3.
Table 3 The Target Value of Every Forecasting Method
The name of forecasting method

MAD

MSE

65

Analysis on the Forecasting Demand of Front-line Worker
Last-value

295

145909

Average

400

242876

Moving average

437

238816

Exponential smoothing

324

157836

Exponential smoothing with trend

345

180796

Cell linear regression

35

1838

Note: In Exponential smoothing α = 0.5; In Exponential smoothing with trend α = 0.3, β = 0.3.

From Table 3, we know that the call numbers have obvious season fluctuation and are closely correlative
with distribution. So the target value of linear regression is far less than the others. The method of causal
forecasting is more appropriate than the other methods in this case.
References:
1. Shuangzeng Hu, Ming Zhang. Logistics System Engineering, Tsinghua University Press, 2000: 58-60
2. Chunshan Feng, Jiachun Wu, Jiang Fu. Research on the Integrate Exertion of the Qualitative and Quantitative Forecasting
Method (Natural Science Edition), Journal of Donghua University, 2004(6): 114
3. Yongnin Jia. Apply the Method Forecasting to the Decision-making, Railway Communication Signal, 2004(5): 12
4. Frederick S. Hillie, Mark S. Hiller. Introduction to Management Science, China Financial & Economic Publishing: 550-562

(Edited by Dragon, Joy and Sun)

66