Specifying main effective factors on domestic steel Price: Case study
of Iran steel market
Mehdi Mirghaderi, Mohammad H. Askari Tabatabaei, J. Parvizian
sm.mirghaderi@in.iut.ac.ir, mh.askaritabatabaei@ in.iut.ac.ir, japa@ cc.iut.ac.ir
Isfahan University of Technology
Department of Industrial Engineering
Abstract.
Traditionally, production and consumption of steel are indicators of the development process in
a society. In the world village, domestic and world indexes and prices highly affect each other.
Differences between the supply and current demand are the main factors determining the steel
price. In this paper, by an Artificial Neural Network we specify indispensable variable to identify
the supply and demand for steel. Time of Contract, Expected Price, Volume of supply, Prior
contract volume, Global Steel Price, Time between contract and transfer, Previous Price,
Volume of Demand, Different Between Supply & Demand, Exchange Rate are specified as the
main variables. Then, by a System Dynamics approach we analyze intricate system functioning
against changing economic parameters in Iran.
Keywords: System Dynamics, Iran, Artificial Neural Network, Delphi method
1. Introduction
Today, forecasting of product prices
of holding companies play a crucial role in
both domestic and world markets. Steel with
respect to oil is rank two in fundamental
industries all over the world. Companies by
predicting the price easily can alter their
production plan. Steel market as an
infrastructure industry plays an indispensible
role both domestic and world market. Real
estate market and automobile industry are
two main markets which are based on steel
market. During the last two years,
economical Tsunami faces market with lots
of adversity. Meanwhile, predicting the
price and product planning show their
necessity much more than before. Iran
imports and also exports steel. Foolad
mobarakeh as the main steel producer
cannot satisfies all types of demanded steel
in domestic market. It satisfies about 45
percent of domestic demand. In addition,
Mobarakeh produces more than domestic
demand. Therefore, it exports extra
production. Iran’s special behavior as both
importer and exporter __ intensifies
indispensability of forecasting domestic
steel price. Almost all imported steel come
from China. China with the highest
economic growth during the last decade and
highest average economic growth during the
last 25 years play an important role in world
steel market. Furthermore, high share in this
industry and rise of China to become world
largest iron and steel producer and consumer
since the late 1990s [9] and lower price in
steel market highly affects other suppliers.
Total crude steel production for the
66 countries reporting to the World Steel
Association in January 2010 was estimated
to be 109.2 million tonnes, an increase of
25.9% on January 2009. In Asia, China,
India, South Korea, and Taiwan are main
producers. Iran produces about 1% of global
demand. North America, Europe, Middle
East are main markets of Iran’s production.
Selling strategy in Iran due to diversity of
productions is based on demand. In other
word, Iran’s steel industry produce
exclusively produce requested products.
Domestic steel market is based on auction.
Monthly contracts are usual in industry.
Domestic producer sells its product
in auction. Ever month all demander request
their proposed price and total amount of
their need. Then based on these proposals,
global steel price, marginal benefit, and cost
make decision about its customers. Based on
their prices domestic steel sells its product in
the same order as prices order. More info
about the structure of contacts can be
obtained in Mobarakeh Steel Co. website
[3].
In this paper, by a system Dynamics
approach, we try to specific main factors
which highly affect domestic steel market.
The rest of this paper in organized as
follows. In section 2, literature review and
some work in this area presented. In section
3, we discuss about the main factor of steel
pricing. Section 4 presents dynamics model.
The conclusion and summary are described
in section 5.
2. Literature review
There are wide varieties of different
method to forecasting price. Time series,
VARs techniques, Neural Networks, System
Dynamics, Linear models, econometrics,
segmentation, extrapolation, input-output,
simulation, judgment and so on are some
main methods of prediction [8]. H.F. Zou et
al [4] compared the predictive performance
of ARIMA, artificial neural network and the
linear combination models for forecasting
wheat price in Chinese market. Dean T.
Chen et al [5] investigated a composite
approach, using vector autoregressions to
determine the future values of exogenous
variables of the structural model in cotton
market. Multi-dimensional testing
procedures were adopted to evaluate the
accuracy of forecasts. Faroog Malik et al [7]
Found that the mean absolute forecasting
error and the mean square forecasting error
is reduced by applying cascaded neural
network relative to conventional artificial
neural networks and popular linear models.
M. Khashei et al [8], based on the basic
concepts of ANNs and fuzzy regression
models, proposed a new hybrid method that
yields more accurate results with incomplete
data sets. SHARIF et al [10] Forecasting
Multilevel Technological Substitution by
System Dynamics Modeling. Mohapatra et
al [11] Incorporating Delphi Results in
System Dynamics Models in Case of
Indian Tea Industry. In addition there some
studies like Jianguo jia [12] in price
modeling. Rahdari [6] analyzed the effect of
global steel price auction on Iran’s steel
stock market’s price.
In this study, we analyze domestic
steel market by a System Dynamics
approach. There are lots of factors which
can influence domestic steel price. In this
study by Delphi method critical factors of
pricing has been choose. Then by a neural
network method weight of each sector has
been analyzed. Finally a system Dynamics
approach shows the relation between price
and these main factors.
3. Specifying main factors in pricing
As mentioned above, neural network
is one of forecasting methods. In this paper
we used Artificial Neural Network (ANN) to
weighting domestic steel price’s main
variables. Delphi method has been used as
our variable selection method.
Select, design, and modeling a neural
network form effectiveness and efficiency of
it. In other word, choosing a suitable neural
network is the first step of using neural
network as an anticipating method. Among
neural networks ANN has been use widely
in literature review. Different types of ANN
have been used in forecasting [13].
According to literature review more than
95% of neural network’s application in trade
and market’s issues was in multilayer feed
forward (MLFF) neural networks [14]. Most
researchers found this network as the best
method for forecasting. Therefore, in this
study a multi layer perceptron (MLP)
network has been used. However, some
other types of neural networks like RBF,
PNN, and GRNN can be useful in predicting
price.
All neural networks are
combinations of nodes and layers. A MLFF
or MLP neural network has three different
layers in general; an input layer, middle
layer or hidden layer (in some cases more
than one layer), and an output. Number of
layers and number of neurons in every layer
specify a given neural network. There are a
lot of different ways to specify architecture
of an ANN in literature review [15]. But,
most of these methods are so complex to be
useful. Besides, degrees of applicability of
them in real world’s problem are
occasionally lower than the real demand.
Most of expert believes that trial and error
would be respond of this difficulty. In this
study, trial and error has been used in most
cases.
Every MLP Network has one input
and one output layer. Therefore in a MLFF
Neural Network defining number of middle
layers would be the case. One middle layer
for every neural network in order to estimate
every complex non-linear function would
satisfies our accuracy. Applying one hidden
layer for a neural network could lead to
some problems in cases that there are lots of
neurons. It seems in which using two middle
layers could have desirable results.
However, there is no exact method to denote
number of middle layers. As, number of
middle layers have correlation with number
of neurons, we define number of layers with
respect to number of neurons.
Characterizing number of input,
output, and hidden layer’s neuron is the next
step. Specifying number of input’s layer’s
neuron would be the most important
decision in MLP architectures. Usually
numbers of input variables which affect the
main variable and number of input neurons
are the same. In our case number of input
neuron has been choose as equal as number
of variables. To specify effective variables
on steel’s price there are lots of methods.
Here is the procedure of defining effective
variables. First, by Delphi method we define
primary variables which can be used as
input layer in Neural Network. Then, we
find the most suitable variables. Weight of
variables will be defined on the last step.
The procedure of Delphi method mentioned
as follows.
In Delphi method, all information
about the problem including literature
review, related articles and papers, most
critical variables in this issue, and main
financial variables which has been used
before in anticipating are sent to experts.
Then, experts define indispensible variables.
Afterward, opinion of every expert
exclusively would be evaluated. These
results would be sorted and analyzed. Then,
once more, results of next step will put
under supervision and evaluating of experts
for final regards. Finally, the result will form
our variables. The result of accepted
variables in Iran domestic steel market has
been shown in Table 1.
Row Name of Variable
1 Type of Contract
2 Time of Contract
3 Expected Price
4 Volume of supply
5 Prior contract volume
6 — Global Steel Price
7
8
9
Time between contract and transfer
Time of transfer
Previous Price
10 Volume of Demand
11 Different Between Supply & Demand
12 Exchange Rate
Table 1 - Result of Delphi Method
Specifying middle layer’s neurons is
the next step. In this ANN the following
tules has been used in order to denote No of
neurons. In every layer, first choice is n (as
the same as the number of input neurons). In
general the minimum number of neurons is
[n/2]-1 and maximum is 2n+1. As we are
going to anticipating steel’s price, the output
layer has one neuron. Back propagation has
been used as the learning algorithm [14].
This algorithm has a popularity of about
95% in most trade and financial problems.
After some trial and error a solitaire
layer neural network has been chose. 10
input neurons, 4 hidden neurons, and one
output neuron is the structure of our ANN.
The result of this Neural Network leads us to
final variables which are highly influencing
in steel’s price forecasting. The results are
presented in Table 2.
Row Name of Variable
1 Time of Contract
Expected Price
Volume of supply
Prior contract volume
Global Steel Price
Time between contract and transfer
Previous Price
Volume of Demand
Different Between Supply & Demand
10 Exchange Rate
Table 2 - Result of specified ANN
CoyuauRwn
4. Dynamics model
In this section by a System
Dynamics approach and a causal model we
demonstrate relations between main
influencing factors on steel pricing. With
respect to complexity of assessing price’s
systems, varieties and intensity of drastic
variables, and dynamic nature of affectivity
of variables modeling price with classical
approach cannot lead us to a desirable
resolution. On the other hand, including
some parameters like contract time,
demander expected price, supplier expected
price, and global steel price in classical
models faces the solution with some
difficulties. Layer-based and indirect
analysis in SD mechanism and utilizing
graphs to illustrate quality parameters made
SD approach a unique one in these types of
problems.
Difference in demand and supply are
source of forming price. In an auction in
steel stock market excess of demand in
compare with supply leads to rivalry
between demanders. This phenomenon leads
to higher proposed prices by demanders.
The highest prices in order are the winners
of auction. Supplier and demander expected
prices are forecasting of suppliers and
demanders about the future of steel’ price.
Proposed volume, proposed prices, base
proposed price, and production plan are
impressible by expected price. Time series
as one of the forecasting method has been
used commonly to anticipating expected
price. Prior price and current price also
affecting expected price.
Analysis shows that effect of
increase in price on expected prices can be
positive or negative. Figure 1 shows that
increasing demander expected price result in
increasing demander’s demand. This leads
to higher price.
Automobile industries and real estate
market are the main demanders’ market of
steel producer’s companies. According to
nature of these markets, contract time is in
contact with this market.
Global steel price (world steel price),
which dramatically influencing by world
economic growth, is one of the primary
indexes on raw material cost, and so on.
Better global economical situation leads to
more demand in raw materials’ market. This
results in more external demand and more
total demand. Not need to shed light on that
increasing in raw materials affect the other
mentioned factors in a one or two much
delay. Finally this leads to changing in
domestic price (by affect of global steel
price).
Supplier expected price and prior
contract volume which are the main source
of PP, capacity of production, and base
proposed price highly influenced by global
steel price. Difference of cost and income
shows benefit of a company. Benefit is the
most critical item in leading a company.
More benefit persuades a company to
produce more. More produce result in
decrease in price. Besides, government
strategy to restricting import, and tax in
imported steel changes price. In addition,
government’s strategy in exchange rate may
have a benefit for steel companies.
‘amount of supplier Lp world steel
import my pected price |"
price
prior price| t
+ + |demander expected
price
base proposed pric
+
word
4, [Product pa eciRbthic
Yl
prior contract
volume supply Price |
+
total demand
income cost +f Ate
+ +
exchange
Real Estate} automobile]
Market market
external
demand
rate
Figure 1
5. Conclusion
In this paper we describe a practical problem
in pricing. Domestic steel pricing plays a
crucial role in local markets. An ANN
method evaluates the proposed affective
A
lcontract timp
1
mateals
rata
variables and select best of them. Then, a
SD causal approach shows the relation and
correlation between these main variables.
References
[1] http://portal.mobarakeh
steel.ir/portal/page/portal/W ebSite/Home
[2] http://www.steelonthenet.com/
[3] http://portal.mobarakeh-
steel.ir/portal/page/portal/W ebSite/Procurement/Pu
rchase% 20T ransactions
[4] H.F. Zou, G.P. Xia, F.T. Yang, H.Y. Wang, An
investigation and comparison of artificial neural
network and time series models for Chinese food
grain price forecasting, Neurocomputing 70 (2007)
2913-2923
[5] Dean T. Chen, David A. Bessle, Forecasting
monthly cotton price: Structural and time series
approaches Intemational Journal of Forecasting,
Volume 6, Issue 1, 1990, Pages 103-113
[6] Hossein Rahdari, Hossein Rahdari, A tefeh
Ramezankhani, Pouyan Foroughi,A nalyzing the
Effect of Global Steel Price Fluctuation on Iran's,
Steel-Maker Stock Prices: A System Dynamics
Approach, System Dynamics Society. July 26 —
30, 2009. Albuquerque, New Mexico, USA
[7] Faroog Malik, Mahdi Nasereddin, Forecasting
output using oil prices: A cascaded artificial neural
network approach, Journal of Economics and
Business 58 (2006) 168-180
[8] Robert Fildes, The forecasting journals and
their contribution to forecasting research: Citation
analysis and expert opinion, International Journal
of Forecasting 22 (2006) 415-432
[9] Mingming Hu, Stefan Pauliuk, Tao Wang,
Gjalt Huppes, Ester van der V oet, Daniel B.
Miller, Iron and steel in Chinese residential
buildings: A dynamic analysis, Resources,
Conservation and Recycling, article in press
[10] M. NAWAZ SHARIF and CHOWDHURY
KABIR, System Dynamics Modeling for
Forecasting
Multilevel Technological Substitution,
TECHNOLOGICAL FORECASTING AND
SOCIAL CHANGE 9, 89-1 12 (1976)
[11] PRATAP K. J. MOHAPATRA, MADHABC.
BORA, and KAILAS C. SAHU, Incorporating
Delphi Results in System Dynamics Models: A
Case of Indian Tea Industry, TBCHNOLOGICAL
FORECASTING AND SOCIAL CHANGE 25,
159-177 (1984)
[12] Jianguo Jia, Yu Lu, Jianxun Chu, System
Dynamics Modeling for Medicine Price Policy in
China, July 26 — 30, 2009. Albuquerque, New
Mexico, USA
[13] Smith K A Jatinder N.D., neural network in
business: techniques and applications for the
operation researcher, international journal of
forecasting 14, 35-62, 1998
[14] Zhang G, Patuwo, B. E., Hu, M Y, forecasting
with artificial neural networks: the state of the art,
International Journal of Forecasting, Volume 14,
Issue 1, 1 March 1998, Pages 35-62
[15] Bo K. Wong, Y akup Selvi, Neural network
applications in finance: A review and analysis of
literature (1990-1996) Information &
Management, Volume 34, Issue 3, 1 October 1998,
Pages 129-139
(16]
