Forecasting the global price of iron ore

using neural networks

Pouria Farajian1, Nima Farajian2, Hossein Eghbali3
1
Faculty of Civil and Architecture, University of Eyvanekey, Semnan, IRAN
2
Faculty of Computer Engineering, University of Eyvanekey, Semnan, IRAN
3
Faculty of Industrial Engineering, University of Eyvanekey, Semnan, IRAN

Abstract
Today's dependence on technology increases human need for iron ore products, and
forecasts suggest that by the year 2035 Demand for steel 60 Percent increase (
Mohammadi , Soltani Mohammadi , and the donor of security 2035) For this reason,
predicting the price of metals such as iron ore using quantitative and qualitative methods
such as economic and technical study The market didn't really match the reality One of the
most common methods of pricing is the time series method. In this research, by modeling
and using series analysis At the time, with the help of dynamic neural network, the price of
iron ore has been predicted. Then, using the neural network method Synthetic and given
the monthly price of iron ore and the factors affecting its volatility, the estimated iron ore
price and then the results the obtained results were evaluated in terms of predictability.
Optimal neural network model with 3 Layer and 10 Near estimated the price of iron ore
very accurately. In this model the amount of training error is approx. %1.7 And for equal
validation %2.3 And test error %1.5. Also, regression and Data correlation at 32%
confidence level and a high correlation value With R2= 0.98 Indicates a good and accurate
model

Key words: Iron ore prices, Forecasting, Neural network

1. Introduction
There are generally two types of analysis in market and economic analysis:
A. Fundamental Analysis
B. Technical analysis

Analytics The fundamentals act qualitatively and also to some extent It's been a taste and it
depends on the type of expert. That is, on qualitative factors Affecting the target product
market) Iron ore here (Study is don. Technical analysis on digital data and time charts it works.


corresponding author, University Address: Daneshgah Street, Ayatollah Taleghani Blvd, EyvaneKey
E-mail : Nimafarajian@eyc.ac.ir

Electronic copy available at: https://ssrn.com/abstract=3969975

https://ssrn.com/abstract=3792463
 In this study, due to quantitative data, technical analysis was used. However, although technical
analysis Well done, without careful analysis, it is impossible to study and study the economy.
Of course, the prediction with Using a neural network can reduce our dependence on
fundamental analysis. In this research technical factors all have Small amounts are examined.
First, the type of data and their nature are described. be. Then by analyzing the effect data They
are described, followed by the introduction of the desired neural network and technical
analysis, Summary of topics is expressed. Today's world depends on technology, Increases
human need for iron ore products. Forecasts indicate that the world population per year2030
with the growth rate 2% to 8.3 There will be a billion people. Thus, consume iron ore per
year2030 It will require about 20 billion tons of iron ore.

Table 1. World Bank iron ore price

forecast and comparison with real prices

Year Expected price Real price

2013 127.6 94.3

2014 91.4 64.7
2015 52.8 36.8
2016 46.5 23.7
2017 41.1 -
2018 42.4 -

2. Necessity of research
In the last two decades the price of metals (Especially iron ore) It has set special records in world
markets the figure 1 It shows the price changes of iron ore over the last 30 years the iron and steel
market is one of the turbulent markets that Predicting the right future can have a positive impact
on decisions. Predicting iron ore prices and forecasts Correct, it can facilitate the decision-making
process of buying and selling it in global markets and the best time to run deals and Determined
investments. Therefore, accurate forecasting of iron ore prices is very important.

Figure1: Price changes of iron ore grains 62 percent over the past 30 years

Electronic copy available at: https://ssrn.com/abstract=3969975

https://ssrn.com/abstract=3792463
 3. research goal

The purpose of this study is to introduce a method to predict the price of iron ore Which can
influence the important factors affecting the price Consider the iron ore well and make a more
realistic forecast of the future of iron ore prices in the world. In this research, we try to use a
method that can explain the complexities mentioned above on iron ore price changes Consider it
well and provide a better forecast. According to the said use of neural networks for performance
Their specificity and teachability are probably the most appropriate response to the stated target
response -That is, a better forecast for iron ore prices-Will be achieved. For this reason, the
following objectives are answered as far as possible:

A. Review of previous research on iron ore price forecasts

B. Providing a neural network model and comparing its results with the World Bank forecasts

4. Research assumptions

The assumptions of this research also include:

• Estimate price for cutie 61-62 Magnetite iron ore is performed.

• FOB prices are a consideration.

5. Research Background:
Table 2: Parameters to Investigate Factors Affecting Iron Ore Price

Moghadam Daneshgar Hafez

Rafiee˛Tayyeb Nia
˛Monjazy˛Mehrdanesh
2009
2015 2013 2015

Shipping fare gold price production High affinity for the iron ore market with steel
rate Steel
interior

Steel prices Steel prices Significant growth in warehouse inventory and

surplus supply
Steel
prices

Oil prices Oil prices Request Significant weakness of the Chinese housing
Steel market
interior

Electronic copy available at: https://ssrn.com/abstract=3969975

https://ssrn.com/abstract=3792463
Iron ore demand US Inflation Rate exchange
rate
Weakening China's economic growth rate

Dollar price US interest rate Scrap Increase the amount of investment to extract Iron
prices ore from mines in countries such as Brazil and
Australia

Euro's Price Price Iron Apply the decision of many major manufacturers
ore Iron ore in the world to increase production Itself
Dow Jones Stocks to eliminate small producers

Supply rate Iron US GDP The growing growth of investments in

ore transportation Rail and marine interior

Scrap prices Significant decrease in the state rights of small

mines Australia
Price of aluminum

Euro to Euro ratio World Steel

Dollar Production
Slowing China's GDP growth

Worldwide production
of iron ore

China's GDP

Iron ore prices

Continued Table 2: Parameters to Investigate Factors Affecting Iron Ore Price

Rafiee˛Tayyeb Nia Moghadam˛monjazy˛merdanesh Daneshgar Hafez 2015

2009 2015 2013

Crude steel Crude oil prices Price Finished Worldwide production of iron ore
production China Iron ore

Electronic copy available at: https://ssrn.com/abstract=3969975

https://ssrn.com/abstract=3792463
 Industry Focus Index BDI Interest rate World Steel Production
China Steel

BDI 1indicator Iron ore prices Inflation Worldwide volume of iron ore
exchange

Import of iron ore Consumption China's population growth

China Iron ore china

Exchange volume Domestic World population growth

World Iron Ore production
China

Domestic Price China's crude steel production

production Iron international
ore china Iron ore

Index USDX2 The global demand for iron ore

6. Introducing the data

In general, many parameters can be considered to study the effect of iron ore prices. In the
selection of parameters, the following were considered in this study:

• Collecting Parameters: By studying articles, research and scientific papers related to the
subject of research, the parameters that influence it Various researchers have studied them,
summarized in Table 2.

• Select the appropriate timeframe: According to (A) Drastic changes in the price of iron ore
in recent years, (b) modification Iron ore pricing from year to month (F) The more
important role decisions of the world's three major iron ore producers (BIG-3) In
determining the price of this mineral, it was decided that from 2013 to 2016 data that Is
provided monthly, use in model building to maximize the effect of the foregoing in the
model Implemented.

1
Baltic Dry Index
2
US Doller Index

Electronic copy available at: https://ssrn.com/abstract=3969975

https://ssrn.com/abstract=3792463
 • Selection of appropriate parameters: According to the parameters listed in Table 3 selected
parameters First of all by Researchers have used it and, secondly, it is possible to collect it
on a monthly basis.

Based on the above, the following parameters were selected to build and test the model:

1. Worldwide production rate of iron ore

2. China iron ore production rate
3. Iron ore imports to China
4. gold price
5. Dollar Index
6. Crude oil prices
7. Freight Cost Index
8. Price of aluminum
9. Dow Jones stock index
10. US interest rate
11. Steel prices
12. Euro's Price

Table3. appropriate parameters

Electronic copy available at: https://ssrn.com/abstract=3969975

https://ssrn.com/abstract=3792463
It is necessary to perform the analysis of the price of iron ore and to predict its changes in order to
perform the analysis That data is first analyzed in a general way to be divided.
In economic analysis, data are generally divided into two categories: exogenous and endogenous
in this study There is also a classification according to which the data are:

• Endogenous factors: Global iron ore production, China's iron ore production and China's
largest iron ore imports Iron ore.

• Exogenous factors: Gold prices, crude oil prices, aluminum prices, steel prices, US interest rates,
Dow Jones stock index, Dollars, Euros and Transportation Costs (Based on the Baltic Index) Which
are factors outside the scope of iron ore

7. Analyze the data relationship or price of iron ore

The results of the parameters mentioned in Table 1 are shown separately That study one-
dimensional parameters Effective in analyzing the iron ore market does not help much .Even the
main economic parameters such as supply and demand, as well as displacement Iron ore also failed
Much of the overall, seasonal or sudden fluctuation of iron ore prices justifies So should this
Examine the parameter together. But putting together all these parameters and studying them in a
multidimensional way in traditional ways at the same time it is very difficult, because reading tools
on paper more than three dimensions are intangible and do not produce good results to the for this
reason, the nature-based innovative approach was used. One of the most effective methods is
neural network .However, the output of the neural network method effectively depends on the
number and composition of the input parameters, and sometimes over or under. Becoming a
parameter can have a huge impact on the accuracy of the neural network model in order to
construct an efficient neural network model, a good combination of the parameters introduced in
Table 1 should be introduced as inputs. Network used the criterion of solidarity alone cannot
Indicates the effectiveness of a network input variable and may even be one the high correlation
variable will have a negative impact on the network output This is due to the shared dependency
of this variable and the target variable on one the third variable is. One of the best ways to select a
feature in machine learning is the CFS3 method This method has usability issues Regression and
neural networks with numerical and continuous objective function have excellent results compared
to other classical methods. The choice is variable. (HALL 1999) In this research, this method is
also used to select the optimal combination of input variables Network used .CFS method is a
based method It is a correlation and its main idea is that variables are input to the network They
are suitable to have high correlation with the target variable and to be independent of each other
In the method of a function A heuristic is used to investigate the properties and the optimal
combination of properties is chosen accordingly. By applying the algorithm CSF on the data collected
in the previous section, the following 7 features as optimal and optimal combination The neural network
input variables were specified as follows:

1. Worldwide iron ore production

3
Correlation based Feature Selection

Electronic copy available at: https://ssrn.com/abstract=3969975

https://ssrn.com/abstract=3792463
 2. China iron ore production rate
3. Iron ore imports to China
4. gold price
5. Dollar Index
6. Crude oil prices
7. Freight Cost Index

8. Price analysis using neural network

➢ Explanation of the neural network used

The neural network used in this research is the network MLP Is leading This network has three
general sections of the input layer, The middle layer (the hidden layer) and the output layer Global
iron ore production data, iron ore production in China, iron ore imports China, Gold, Crude Oil,
Dollar and Transportation Costs Used as Neural Network Inputs And the price of magnetite rock
also Specified as output and target function for prediction In the middle layer section, a middle
layer with 10 neurons was used ) Doubt ( Input data with ratio %70,%15 AND %15 They were
divided for training, testing and accreditation sections, respectively. The excitation function used
in the network is sigmoidal function and the weights between neurons are also optimized using the
Levon-Berg algorithm of Macquarie Were.

Figure 2: Your view of the neural network used

The performance of this network is estimated using the data of the last three months to predict the
fourth month. Then with Using the generated data and the previous two months of the fifth month
and so on, the whole time period is predicted. To do the job Initially, the initial model and its order
are identified and its parameters are calculated. Next, check the fit of the network and the number
of neurons the middle layer runs at 10 neurons for the optimal state. The criterion of error is the
calculation of the mean square root of the error that represents the accuracy Network results.

➢ Neural Network Results

As can be seen in Table (2), the amount of training error is approximate %1.7 And for equal
validation% 2.3And test error% 1.5 That is, the error of training and testing is less than validation,
indicating that the phenomenon of over-fitting has not occurred.

Electronic copy available at: https://ssrn.com/abstract=3969975

https://ssrn.com/abstract=3792463
By comparing the results with reality (Target values) The amount of correlation between network
results and reality is acceptable. This theme is on Figure 3 can be seen.

Figure 3: Predicted values of iron ore prices using neural network.

Table 4: Regression information at 95% confidence level

Figure 3: Correlation and Linear Regression of Predicted Model and Real Data

9. Discuss the results

In this section, the results obtained after validation in the upper and lower part of the regression
as well as the absence of over-fitting error the model is examined and analyzed .The obtained
model shows high correlation with iron ore price data the results of the network are valuable
Because there is currently a model that can bring all endogenous and exogenous factors together.

Electronic copy available at: https://ssrn.com/abstract=3969975

https://ssrn.com/abstract=3792463
The iron ore market can predict the price of this important mineral. That is, consider all the factors
with all the complexities mentioned and Provide an analysis of the price of iron ore with a accuracy
of about 2%. The regression model presented in Error! Reference source not found It was a
statistically stable and valid model Because they have responded positively to the t-student and p
-value tests. This indicates the high power of the proposed model as well as the expressive a new
concept in economic analysis and financial forecasting for important financial and economic
decisions.

➢ Comparison of Results with Forecasts of Other Sources

Designed to demonstrate the high value of work and the accuracy of the neural network, the results
are consistent with World Bank forecasts, Forecast the Neoma InstituteLuff forecast, forecast 4
presented in Moghadam et al., As well as prices available in the iron ore market. Is compared. This
comparison is shown in Table 5.

Table 5: Comparison of results with forecasts of other sources and actual market prices

Iron ore prices

The results Mean prediction

year World bank Neoma Medium forecast
Real data of this of Moghadam et
forecast prediction luff
research al

2013 94.3 92.2 127˛6 135˛4 127 129.2

2014 69.7 65.1 91.6 96.9 122 92.5

2015 36.8 37.1 52.8 55˛8 120 54.6

2016 23.7 24.7 46.5 50 119 49.1

2017 - 22.1 41.1 54 116 43.5

2018 - 21 42.4 57 115 45.2

As it can be seen from the results of Table (4), the accuracy of the studied method has shown good
accuracy compared to the actual market data, and Compared to World Bank Forecasts, the Neoma
Institute, the Lough Forecast Medium and the Medium Forecast has done. However, neural

4
www.kneoma.com

Electronic copy available at: https://ssrn.com/abstract=3969975

https://ssrn.com/abstract=3792463
network results and forecasts confirm the downward trend in iron ore prices by 2017. With
Regardless of the results of this study, the World Bank forecasts the closest prediction to real
values of iron ore prices, so Figure 4 compares actual prices, neural network forecasts and World
Bank forecasts.

10.Conclusion
Two types of endogenous and exogenous agents were identified by studies on the iron ore market. One-
dimensional study (study of individual effective parameters) It is not possible to analyze the iron ore market
and all effective parameters must be Studying together, the neural network was chosen as one of the most
effective methods. The MLP neural network used in this network research the leading one had three general
sections of the input layer, the middle layer )The hidden layer with 10 neurons and the outer layer( Input
data with ratio 70%,15% AND 15% For training, testing and accreditation sections respectively Allocated.
The stimulation function used in the network is sigmoidal function and the weights between the neurons
using the algorithm Leven-Beurge Maquart Optimized. The amount of training error is approx. 1.7% And
for equal validation 2.3% And test error 1.5% And is a more fitting phenomenon in the model Nothing
happened. The correlation value (at 95% confidence level) between the predicted results and the actual
values is about 98% which represents a very high value. The model shows. The obtained model with high
correlation with iron ore price data indicates that the results are valuable Network. This model can combine
all endogenous and exogenous factors in the iron ore market to raise the price of this important mineral
with an error in About 2%. Finally, the model is statistically stable and valid) P Value, T STUDENT
positive tests( have given This demonstrates the robustness of the proposed model and also introduces a
new concept in economic analysis and financial forecasting for decisions It is important financially and
economically.

Reference
Hafez Target Investment Consulting Company. (2015)"Factors Influencing Iron Ore Price Reduction".
scholar, Mohammad Reza. "Factors Influencing Electricity on Steel in Today's Iran Market." (2013)
Economics Journal - No. 3014
Rafiee, Majid, and Mohammedi Asir Tayeb Nia. “Investigating the Influential Parameters on Iron Ore Price
Using Neural Network Modeling.” Symposium Steel 88. Yazd: Iranian Iron and Steel Society, 882-
875.1388
Mohammedi, AS, Soltani Mohammadi, and H. “Prediction of iron ore prices using time series models.”
Journal of Mining Development, : No. 44, February and March: 92-88: 2013
Moghaddam, Mohammad Reza, Masoud Manji, and Amir Hossein Mehrdanesh. "Evaluation of factors
affecting iron ore prices using artificial neural networks."The 3rd Iranian Open Mines Conference. 1394-
Hall, Mark A. "Correlation-based feature selection for machine learning." PhD diss., The University of
Waikato, 1999.
Li, H, B Wang, E Ren, and C Wu. "Empirical analysis of the influencing factors on iron ore prices."
Artificial Intelligence, Management Science and Electronic Commerce (AIMSEC), 2011 2nd International
Conference on, 2011: 3004-3008.
Loof, A, A Malanichev, and I Khobotilov. "Long-term iron ore price modeling: Marginal costs vs. incentive
price." Resources Policy, 38, 2013: 558-567.
Wenhui, C, L Yalin, and J Yong. "Influencing factors analysis of China’s iron import price: Based on
quantile regression model." Resources Policy 48, 2016: 68-76.

Electronic copy available at: https://ssrn.com/abstract=3969975

https://ssrn.com/abstract=3792463
Ye, H, and J Li. "Analysis of the factors affecting the price of imported iron ore in China." J. Econ. Issues
10, 2012: 119-122.

Electronic copy available at: https://ssrn.com/abstract=3969975

https://ssrn.com/abstract=3792463