International Journal of Financial
Management (IJFM)
ISSN(P): 2319-491X; ISSN(E): 2319-4928
Vol. 5, Issue 4, Jun – Jul 2016; 23-36
© IASET
PREDICT CRUDE OIL PRICES IN THE INTERNATIONAL MARKET
AN ALTERNATIVE FORECASTING TECHNIQUE
SIAGH AHMED RAMZI1 & HACHEMANE MOULOUD2
1
Assistant Professor at Faculty of Economics, Commerce and Management Sciences
Kasdi Merbah University Ouargla, Algeria
2
Professor and Director of Research, Lecturer at Algiers University 3,
Faculty of Economics and Management Sciences, Expert in Econometrics and Fisheries Economics
ABSTRACT
The severe clearly unrest in the oil market reflected in the volatility of oil prices, allowed to consider the
efficiency of the various forecasting techniques applied. In order to stand on the ability of some models to provide
forecasts for decision-makers and to investigate the presence of any sort of seasonality and identify the type of the trend
component crude oil prices we used weekly, monthly and annually data collected from the U S Energy Information
Administration for the period 1987 to 2015. Appropriate forecasting technique mainly based on a two forms of the "double
exponential smoothing" model related to Brown's Linear (i.e., double) Exponential Smoothing and the Holt-Winters
Double Exponential Smoothing referred to a more sophisticated version of the first ones is model. After construct
short-term forecasts for weekly, monthly and annual data. The “Evaluation statistics” section shows the RMSE and RSS
statistics for each of the forecasts. Results have shown the importance of the non-parametric models in providing
reasonable forecasts supported by appropriate statistical tests and indicated absence of seasonal effects and also economic
cycles of them in terms of their compounds. The slump in the prices will deepen unless the OPEC and non OPEC
producers sit together to avoid any further prices collapse.
KEYWORDS: Crude Oil Prices, Forecasting, Double Exponential Smoothing, Brown's And Holt’s Models
INTRODUCTION
Oil is an important source of revenues for oil-exporting countries. Considering that the price of oil is one of the
most important economic indicators in the world, its changes is watching of both producers and consumers, in addition to
the decision-makers and actors in the financial markets and international. The price volatility disturbed a lot of
underdeveloped economies and those mainly dependent on oil export. In this context, and given the importance of this
price, seen by many as a measure of the performance of the global economy, the first economists particular importance to
the process of predicting future trends across harness a lot of effort to provide appropriate and effective and applicable in
different circumstances tools. Because changes in the price of oil can have significant effects on macroeconomic dynamics,
it is not surprising that many researchers have applied various econometric techniques that help to model and to forecast
changes in the price of oil (Alquist et al., 2013).The great turmoil in oil markets across the world renewal of the
importance of these techniques in light of the sensitive changes that have defined time series on the phenomenon studied.
Therefore, making reliable forecasts of the price of crude oil become interest for a wide range of applications. For example,
www.iaset.us
editor@iaset.us
24
Siagh Ahmed Ramzi & Hachemane Mouloud
policymakers, government and central banks forecasters view the price of oil as one of the key variables in generating
economic projections and in assessing macroeconomic risks and policies and to explore alternative hypothetical forecast
scenarios.The complexity in prices volatility resorting to the non-free models which requiring stationary or inertness
component availability on time series and absence of anomalous values process.Given process complexity of the use of
macro regression models caused by the multiplicity of actors variables in the oil market from the economic factors,
political and geostrategic. Huntington (1994)indicate the failed of structural models in the prediction of oil prices in the
nineties of the last century. In the same way, Artus et al. (2010) argue that the experience has shown how much is a
complex process to predict the evolution of oil prices. This work focuses on the study of the importance of simple models
in providing acceptable predictions can rely on their results in the short and medium term(Artus, Chevalier, Chalmin and
D’Autume, 2010).
And that any attempt to model the phenomenon to predict without regard to random blowing out of hand and
jumps in the price are not well-defined (mis-specified) productive conciliation weak (poor fitted data) data. So will this
work in very sensitive conditions characterized by a period of post-global financial crisis and turmoil in the political world
with the spread of proxy wars and conflicts in the Arab and Islamic worlds, and it increased, and spread the risk. At the
same time it is assumed that the crude oil price properties in the future will be the same as in the past and the present.
A more complex extrapolation scheme that involves a study of the dynamics of changes in characteristics of the crude oil
price with due regard for such dynamics within the forecasting interval is less frequently used in forecasting.
Economists and econometricians are pressed to make short- and medium-term predictions on oil prices since they
are volatile and depend on various situations. They use a range of forecasting tools and highly complicated mathematical
models, which either focus on financials (using spot and future prices), or supply and demand considerations (quantifying
variables and testing their explanatory power). The models used most often are regression-based structural models,
time-series analysis, Bayesian autoregressive models, and dynamic stochastic general equilibrium graphs. Because
economists are still undecided as to which method is most reliable, they use a combination of them all to get the most
accurate oil prices answer forecasting. The complexity of the model does not mean superiority over simple wording. Many
econometricians have gone to prefer simple to complex models for a number of considerations. Therefore, this article will
only deal with exponential smoothing models and will briefly cover more simple models proposed by Holt and Brown.
Thus, the article will discourse topics related to the short-term forecasting of crude oil price sequences with slowly
changing characteristics on the basis of exponential smoothing models. The "short-term forecasting" should in this case
mean forecasting a small number of time intervals ahead instead of forecasting for a period of less than a year as it is
usually assumed in economics.
Another important goal is to use both methods for forecasting oil prices, as well as comparing the quality of the
forecasts and recommending the method best suited to the actual data. The comparative performance of these models is
then considered1to show the practical difference, on their forecasting ability, using comparison of the root mean square
errors, is conducted on the models’ forecasts for a sample excluded from the modeling process. We examined the
difference between the known data and the predicted data using the residual sum of squares (RSS) and the root mean
squared error RMSE. This accurate prediction of oil prices is difficult but very important for governments, companies, and
investors. The more accurate the forecasts of oil prices, the greater the impact they have on improving the accuracy of
1
Alternative of Chow tests approach of the error differences between actual price and the price.
Impact Factor (JCC): 2.4528
NAAS Rating: 2.25
Predict Crude Oil Prices in the International Market an Alternative Forecasting Technique
25
forecasts of a wide range of macroeconomic results.
THE OIL PRICE SPECIFICATIONS AND APPROACH OF ABILITY TO FORECAST
Among oil price forecasting literature, there are two main groups of forecasting methods generally used by
experts. First ones based on quantitative approach represent numerical and quantitative variables that impact on oil prices
and include the econometrics methods and non-standard methods. Among them, econometrics models incorporate three
classes of models: time series models, financial models, and structural models. On another side, the main non-standard
methods which are recently applied are artificial neural networks and support vector machines. The second group,
qualitative methods estimate impact of infrequent events such as wars and natural events on oil prices; these approaches
recently gained more reputation among oil price research. However, between various types of qualitative forecasting
methods there are small number of them that are applied to forecast oil prices, such as Delphi method, belief networks,
fuzzy logic and expert systems, and web text mining method (Bashiri Behmiri and Pires Manso, 2013).
At this point, if a variety of opinion was observed among econometrics researchers on appropriate forecasting
models (Figure1), especially on finance and oil prices thus to the links of these topics to risk, some ones believe about that
functional form play an important role in selecting process of the appropriate technique. For this, the researcher have
tendency to prefer using non-linear relationships in those areas, despite the multiplicity of forms in addition to the lack of
efficient testing tools orientate preference between certain non-linear forms.Huntington (1994) pointed to the structural
models based on supply and demand failed to predict oil prices in the nineties of the last century. He attributed to several
reasons including the difficult accuracy to predict the variable (GDP) in particular for developing countries, and also, the
weakness to predict the oil offer outside OPEC Organization. In these circumstances, econometricians believed that the use
of simple models may have a better performance than other models available in the short term while things may get
complicated in the case of most forecasts and the long-term average due to changes that occur at different levels of
economic and environmental ones. Therefore, economists and econometricians are pressed to make short- and
medium-term predictions on oil prices since they are volatile and depend on various situations. They use a range of models
but are still undecided as to which method is most reliable; they use a combination of them all to get the most accurate oil
prices answer forecasting. In this context we can find the usage of the time series models which generally include three
main categories: naïve models, exponential smoothing models and auto-regressive models such as ARIMA2 and
ARCH3/GARCH4 family models and can be applies to forecasting crude oil prices. We denote that Pindyck (1999)and
Sadorsky (2006)have compares different types of forecasting models, including the random walk, historical mean, moving
average, exponential smoothing, linear regression model and autoregressive models and argued their adaptation to oil
prices modeling. But, the last two decades have notably seen significant developments in exponential smoothing (ES),
which has become one of the most noteworthy forecasting strategies. Exponential smoothing was established as a classical
method of analysis for forecasting different econometric and financial real-time data prospects (Tavakkoli et al., 2015).
The results obtained from data smoothing are operative, simple, accurate and easy to communicate and understand (Broze
and Mélard, 1990).
2
Autoregressive integrated moving average.
Autoregressive conditional heteroskedastisity.
4
Generalized autoregressive conditional heteroskedastisity.
3
www.iaset.us
editor@iaset.us
26
Siagh Ahmed Ramzi & Hachemane Mouloud
Figure 1: Crude Oil Price Forecasting Classification Techniques
The origin of exponential smoothing forecasting technique was initiated from the operations research activities of
the US Navy during the Second World War. Under theses researches in 1944, Robert G. Brown develops a model
essentially an exponentially weighted moving average applied to continuous data and in the early 1950s extended the
approach to discrete data, developing models that could deal with trends and seasonal patterns below a particular
application in forecasting the demand for spare parts in Navy inventory systems, which was so successful in terms of
forecast accuracy and data storage savings (Gardner Jr., 1985). The methodology was formalized, generalized and
extended in Brown (1963), Brown and Meyer (1961) and D’Esopo (1961). On similar way Charles C. Holt (Holt, 1957)
developed a similar method for exponential smoothing of trending time series. His works gained acceptance with Winters
(1960), which tested the methods on sales data with such success that they became known as the Holt–Winters forecasting
system. Further developments were soon made by fascinated researchers, in view of it being a formulation of ‘adaptive
forecasting’ for the technique (Mills, 2011). The whole area of exponential smoothing developed rapidly during the 1960s,
with more consideration to parameter selection, choice of starting values for the recurrence relationships used to compute
forecasts, the monitoring of such forecasts, and the adaptive control of the smoothing parameters (Gardner Jr., 1985).
Harrison (1967) provided the first synthesis of the exponential smoothing methodology, showing that the various types of
exponential smoothing were all particular forms of the Box and Jenkins (1962) polynomial predictor. Recently Holt (2004)
give reflection on the genesis of the method. By technical econometrics view, exponential smoothing has proven through
the years to be a useful technique in many forecasting situations. It was first suggested by Holt (1957) and generalized
Winters (1960). Although ad hoc exponential smoothing (ES) methods which defines an extended class of ES methods and
offers a theoretical foundation for analysis have been employed with recent methodological developments and have
embedded a collection of ad hoc approaches of these models in a modern dynamic nonlinear model framework
(Alquist, Kilian and Vigfusson, 2013).
Finally as Alquist, Kilian and Vigfusson (2013) conclude, although there are a number of works dealing with the
problem of predicting the price of oil, it is difficult to reconcile the seemingly conflicting results. Some issues can
enumerated, on definition of the oil price variable and its expressed terms (nominal or real), on chosen period for
estimation and evaluation, on how the forecast accuracy is evaluated, on whether tests of statistical significance are
provided, on conducting analysis in-sample or out-of-sample and whether the methods are parametric or non-parametric.
Impact Factor (JCC): 2.4528
NAAS Rating: 2.25
Predict Crude Oil Prices in the International Market an Alternative Forecasting Technique
27
The limit issue of this work is to try to predict oil prices in the international market relying only on moving averages
weighted exponential model.
BUILDING FORECAST METHODOLOGY
In this study, data for weekly, monthly and annually crude oil prices (in US dollar) for the period 1987
May 2015 November, giving about 343 observations is used. The data was gotten from the U S Energy Information
Administration (EIA) data survey. In the present work, we focused on an in -sample period, which means that the time
series between 1987 M05 and 2015 M11 served to generate the forecasts of models, whereas the time series between 2015
M12 and 2016 M06 served as out -of - sample data against which the accuracies of the forecasts were measured. We will
evaluate models’ forecast accuracy using the evaluation sample forecast series. Each of the evaluation sample forecast
series contains actual data until November 2015, and then forecast data from December 2015 until June 2016. Based on
visual assessment of the time series of the spot price of crude oil is plotted (Figure 2), we can conclude that there is a
systematic component in the form of a linear increasing trend and small random fluctuations.
Figure 2: Brent Crude Oil Monthly Price (1987-2015)
Figure 3: Brent Crude Oil Monthly Price (2008-2015)
www.iaset.us
editor@iaset.us
28
Siagh Ahmed Ramzi & Hachemane Mouloud
Thinking about this case by application of Box-Jenkins method checking be sticky (Hachemane, 2010). The
strong volatility chronicled series especiallybetween2008 and 2015(Figure 3) have cemented the idea of searching for
suitable alternatives methods. Also, this method poses several questions about how the application of severe structural
changes in the series characterized by abnormal and may consider it values its problematic when estimating model
parameters and let hold a snooze test time series in these circumstances. To apply this technique remains conditional on
adding to the above extent linear trend component that did not materialize in time series of variable oil prices. Its appears
signs of non-linearity in this component and statistically denied its linear shape by the strong significance of BDS5 test
(Brooks C., 2014)which came for different dimensions and complies with the general understanding in the literature of
random time series analysis and not its potential interpretation of non-linear vehicle for the general direction of a series of
oil prices weakening results appreciation under this approach.
Table 1: BDS Linearity Test of Oil Prices Series
BDS Test for PB
Date: 06/10/16 Time: 19:23
Sample: 1987M05 2016M12
Included Observations: 356
Dimension BDS Statistic
2
0.187355
3
0.317485
4
0.404690
5
0.461507
6
0.497278
Std. Error
0.004091
0.006491
0.007716
0.008028
0.007727
Raw epsilon
Pairs within epsilon
Triples within epsilon
52.57993
82729.00 V-Statistic
21480079 V-Statistic
Dimension
2
3
4
5
6
c(m,n)
0.679357
0.661221
0.644196
0.627935
0.612277
C(m,n)
39614.00
38331.00
37125.00
35975.00
34871.00
z-Statistic
45.79759
48.91172
52.44811
57.48921
64.35261
Prob.
0.0000
0.0000
0.0000
0.0000
0.0000
0.703185
0.532296
C(1,n-(m-1)) c(1,n-(m-1)) c(1,n-(m-1))^k
40901.00
0.701429
0.492002
40608.00
0.700500
0.343736
40316.00
0.699566
0.239505
40025.00
0.698626
0.166427
39716.00
0.697347
0.114999
Abuse of the forms, the time series can be considered non stationary (as reflected in correlation function) and
attempt to build a model based on the first differences for the Brent prices variable and then predict short-period assuming
six months and then compare it with the moving averages simple model mentioned above. As expected, this attempt was a
disappointing since the model fails tracking the inflexion points and so unable keep up with the decline in prices. In this
context, lot of studies show and concur on the weakness of these models in prediction (Bashiri Behmiri and Pires Manso,
2013) due to their linear trait and hence was unable to track non-linear series of Brent prices data, add to the structural
change and extreme volatility.
5
The BDS test (W. A. Brock, W. Dechert and J. Scheinkman) detects nonlinear serial dependence in time series.
Impact Factor (JCC): 2.4528
NAAS Rating: 2.25
29
Predict Crude Oil Prices in the International Market an Alternative Forecasting Technique
Figure 4: Brent Crude Oil Weekly Price
The Figure 4 reflect the strength of the volatility of the time series, making them non stationary series and this has
consequences on applying random time series analysis technique. It may also springs to mind adopting simple techniques
to correct volatility likely moving average for example, or considering volatility as part of outliers and handled in a manner
to facilitate the forecasting process and provides acceptable predicted values. In addition, crude oil prices have challenged
the forecasting abilities of such models. When, Hendry (2006) proposed the double-differenced forecasting model as
alternative, he argue that, when time series are subject to infrequent trend changes, the no-change forecast may be
improved upon by extrapolating today’s oil price at the most recent growth rate. Going so far, another strategy is to
extrapolate from recent trends. Given that oil prices have been persistently trending at times, it is natural to consider a
random walk model with local drift designed to capture “short-term fore cast ability”.
To avoid all the above we wanted to measure appropriate exponentially weighted moving average models
(EWMA) for such forecasting process and investigate the feasibility of the results in the short term (ie, within six months
when using monthly data may reach two years in the annual dataset). And, follow events more closely we focused
intentionally our prediction using weekly data for easy interactivity with the results and ability to the instantly comparison
because of the availability of information. Evoke that these models are characterized amendment feature meaning updating
forecasts when any new information from the oil markets was available. The study of this type of time series in terms of
their compounds showed the absence of seasonal effects and also economic cycles of them. This result is easy to test the
appropriate forecasting technique mainly of Brown's Linear (i.e., double) Exponential and Holt’s methods with two
parameters. Since both methods are used in the same circumstances. The spotlight of the simplest time-varying trend
model which is Brown's linear (i.e., double) exponential smoothing (LES) model, reveal two different smoothed series that
are centered at different points in time witch forecasting is based on an extrapolation of a line through the two centers.
SIMPLEST CRUDE OIL PRICE FORECASTING TECHNIQUES
Forecasting methods used in this study were selected based on a detailed graphical analysis of the variable and
due to the nature of the identified components, by simplest method allied to the exponential smoothing models. Usually
acknowledged, the name 'exponential smoothing' is attributed here to the use of the exponential window function during
convolution. It is no longer attributed to Holt, Winters & Brown. The defining equation for simple exponential smoothing
is expressed as follows:
www.iaset.us
editor@iaset.us
30
Siagh Ahmed Ramzi & Hachemane Mouloud
=
+ (1 − )
Where Lt denotes the singly-smoothed series, Yt, the past observations and α is the data smoothing factor. Note
that the weights assigned to previous observations are in general proportional to the terms of the geometric progression
{1,(1−α),(1−α)2,(1−α)3,...} which is a discrete kind of an exponential function, so this is where the name for this smoothing
method originated.
But when there is a trend in the data simple exponential smoothing does not do well6 which is inconvenient. In
such situations, numerous methods were conceived under the "double exponential smoothing" or "second-order
exponential smoothing", which is the recursive application of an exponential filter twice witch introduce a term to take into
account of a series exhibiting a trend. This slope component is itself updated via exponential smoothing7. Therefore, due to
the nature of the data for forecasting the price of oil expressed above, the double exponential smoothing model was used,
which generally described by two forms of the smoothing model related to Brown's8 Linear (i.e., double) Exponential
Smoothing and the second, referred to a more sophisticated version of this model, the Holt-Winters9Double Exponential
Smoothing.
The algebraic form of Brown’s Linear (double) Exponential Smoothing model can be expressed in the smoothing
equations as the following formula:
=
= (
+ (1 − )
−
) + (1 − )
The interpretation of the trend-smoothing constant γ is analogous to that of the level-smoothing constant α.
Referring to as "Holt-Winters double exponential smoothing", this method divides the time series data into two
components: the level, Lt, and the trend, Tt. These two components can be described by the following formula:
For the Level:
=
+ (1 − )(
−
)⋯0 <
<1
Where, the trend coefficient is also updated by a similar exponential smoothing. To distinguish the trend updating
smoothing weight from that for the intercept, γ is used instead.
For the Growth:
= (
−
) + (1 − )
⋯0 <
<1
Where Lt Represent the smoothed value for time t, and Tt is the best estimate of the trend at time t. For parameters
α is the data smoothing weight factor, 0 < α < 1, and γ is the trend smoothing factor, 0 <γ< 1.Hence for a given set
parameters, these two components are calculated and Lt is used to obtain the double exponential smoothing of the data at
time t. Furthermore, the sum of the level and trend components at time t can be used as the one-step-ahead (t + 1) forecast.
6
NIST/SEMATECH ‘e-Handbook of Statistical Methods’, http://www.itl.nist.gov/div898/handbook/, date: April 26, 2016.
"Model: Second-Order Exponential Smoothing". SAP AG. Retrieved April 23, 2016
8
‘Averaging and Exponential Smoothing Models’.duke.edu. Retrieved April 25, 2016.
9
KALEKAR, P. S. (2006) Time series Forecasting using Holt-Winters Exponential Smoothing. In Book Time series Forecasting using
Holt-Winters Exponential Smoothing (ed^eds Editor). City: Kanwal Rekhi School of Information Technology, p.4.
7
Impact Factor (JCC): 2.4528
NAAS Rating: 2.25
31
Predict Crude Oil Prices in the International Market an Alternative Forecasting Technique
There are actually various well-known forecasting methods that are based only on the analysis of past values of a
time laps. The main instrument of these methods is the extrapolation scheme where the sequence properties identified at a
certain time lag go beyond its limits. It’s evident in this time series, whether monthly, quarterly or yearly they take a
similar behavior and do not differ only in terms of volatility intensity. The series also maintains all characteristics and their
compounds which allow adopting the appropriate cycle and according to the horizon of forecasting. Despite that Gardner
and McKenzie (1985) suggested that the trends should be “damped” to be more conservative for longer forecast horizons10.
Our work is limited on different forms, including annual to predict the average term by investigate weekly, monthly and
annually prices to the short-term forecasting.
Finally, after construct short-term forecasts for several future steps, the “Evaluation statistics” section shows the
RMSE and RSS statistics for each of the forecasts, along with the methods. Using residual sum of squares (RSS) to
measures the total deviation of the response values from the fit to the response values which is Sum of squared differences
between the observed and estimated response11. Minimizes values prove that the model has a smaller random error
component. Also, Root Mean Squared Error (RMSE) known as the fit standard error or the standard error of the regression.
It is an estimate of the standard deviation of the random component in the data. The use of RMSE is very common and it
makes an excellent general purpose error metric for numerical predictions. A RSS & RMSE values closer to 0 indicates a
fit that is more useful for prediction.
=∑(
−
=
∑(
)
−
)
RESULTS FORECASTING
To stand on the importance of Holt’s smooth exponential forecasting technique, the weekly prices predictions
have been set from the first week of the year 2016 up to the first week of February. This period was selected based on the
availability of data to assess the any historical forecasts within the studied sample. Results were as follows:
Table 2: Brent Price and Five (05) Weeks Horizon Forecasting
Weeks
31 Dec 2015
07 Jan 2016
14 Jan 2016
21 Jan 2016
28 Jan 2016
04 Feb 2016
Price
36.56
34.19
29.10
27.76
31.75
32.18
Forecasting from 07 Jan 2016
35.20
35.74
33.23
32.27
31.31
30.35
Forecast Error
1.36
1.554.134.510.44
1.83
Where the RMSE, root mean square error reached 2.54. This test is important while comparing between models,
but in this case it is advisable strengthening other tests.
10
11
Two smoothing parameters α and γ (with values between 0 and 1), and one damping parameter 0 < φ < 1
quantity minimizes by the least squares estimator.
www.iaset.us
editor@iaset.us
32
Siagh Ahmed Ramzi & Hachemane Mouloud
Figure 5: Test the Regressive Prediction Significance
In addition to the previous RMSE test we want to use again the regression technique to determine the importance
of prediction. Before this we used the figure above which shows the viewing values and forecasts compatibility curve in
terms of the general trend. We support these tests by regression equation where values are represented by dependent
variable and predictions as independent variables.
Table 3: Assessing Regressive Forecasting
Dependent Variable: YT
Method: Least Squares
Date: 04/06/16 Time: 10:30
Sample: 1 7
Included observations: 7
Variable Coefficient Std. Error
C
0.293086
17.26175
FT
0.964348
0.515463
t-Statistic
0.016979
1.870839
Prob.
0.9871
0.1203
This estimate underlines the importance of these forecasts from statistical point of view which is compatible with
requirements of the test, where we accept the null hypothesis for intercept which is equal to zero, and the slope is equal to
one.
We adopt now monthly data to forecast Brent price in international markets. These data showed that the general
trend for the price of Brent (according to Holt two smoothing parameters) will continue to decay to levels below thirtieth
(30) $. Note if current conditions remained predominant and non-OPEC exporting countries did not contribute to reduce
prices. We note also, that Algeria's oil light deserts type and content of sulfur therefore no require additional operations
process to preserve environment and rich by gas oil, preferably used in diesel extraction. Also it outclasses by reach four
(4$) dollars on price terms the Brent price. The general behavior of this series is no different than the weekly data curve
only in terms of volatility; therefore we expect the same behavior of future forecasts. We set the evaluation sample to
“2015M12 2016M6”, giving us twelve months of forecasts to evaluate. We choose to evaluate each of the available
smoothing methods, and set the training sample for the Least-squares, Mean square error and MSE ranks methods to be
“2015M12 2016M6”.
Impact Factor (JCC): 2.4528
NAAS Rating: 2.25
33
Predict Crude Oil Prices in the International Market an Alternative Forecasting Technique
Table 4: Forecasting up to June 2016
Periods
Observed
Forecast
2015.10
48.43
43.24
2015.11
44.27
47.60
2015.12
38.01
42.70
2016.01
30.70
40.64
2016.02
32.18
38.58
2016.03
na
36.52
2016.04
na
34.45
2016.05
na
32.39
2016.06
na
30.33
The Holt monthly smoothing results represented in particular to estimate the monthly price at $ 42.70 for
December 2015 while $ 44.27 is observed (Actual data) for the same month to reach the value of forecast 30.33 $ in June
2016. These forecasts reflect an expected trend towards continues deterioration of prices for exporters, a conditional
scenario of economic survival and political conditions intact and without any agreement between the members of the
organization and non-oil producing countries.
Figure 6: Tracking Performance
As observed on Figure 5 and Figure 6, in magnitude terms the different forecasts adding excess of estimate
(overestimation) when the peaks and underestimation (underestimation) on lows which observed in the smooth case, the
sensitivity of the forecasting process and how difficult, both in term and the short or medium term. And also, prices decay
may be largely due to the improved ability of the United States in unconventional oil production, which is developed by
technology that allow an important diminish of costs extraction (Mohaddes and Raissi, 2015)12. As well as the ability of
US producers in adjusting and control production levels in interaction with price fluctuation, which means its ability to
balance between overall demand and aggregate supply and that it might explain continued decline in prices.
Desire to tradeoff between the two approaches, Brown’s double exponential smoothing and Holt’s smoothing
parameters, the forecasting process was conducted within the sample (Ex-ante forecast) and accuracy of model fitting was
evaluated by means of two measures: the residual sum of squares (RSS) and the root-mean-square error (RMSE). The
fitting errors obtained for both compared models are presented in Table 5.
Table 5: Assess the Forecast and Differentiation between the Two Way:
Holt with Two Parameters (HW) and Double Exponential Smoothing (DES)
Method
DES
HW
12
Period
11-2015
11-2015
Alpha
0.77
1
Beta
0.37
RSS
7377
7130
RMSE
4.63
4.56
In “The U.S. Oil Supply Revolution and the Global Economy”, p13.
www.iaset.us
editor@iaset.us
34
Siagh Ahmed Ramzi & Hachemane Mouloud
According to these indicators, forecasts are evaluated by standard statistics of forecast accuracy. The common
criterion of forecast performance RSS&RMSE showing that Holt’s model provides much more accurate forecasts than
double exponential smoothing , which should come as no surprise as α=1 is greater than 0.77, the value below which
(Harrison, 1967) found little difference between the RMSEs of the two approaches.
By against, the prediction failure of the double exponential smoothing may be due to the linear nature of the
model, which does not correspond to the behavior of oil prices series and thus adequate with most research and studies
above(Mills, 2011).
By reverting to the annual data, the results on applying the two previous approaches have been differentiated and
the Holt’s forecasts were more optimistic. The superiority of the Holt’s method is clearly observed (see Table 6) in terms
of total of residual sum of squares (RSS) and root-mean-square error (RMSE).
Table 6: Annual Prices Forecasts for Brent Crude 2014-2017
Base Forecast 2013
DES
2014
117.75
2015
103.80
2016
49.57
2017
35.64
RSS
7177
RMSE
15.73
HW
108.98
99.39
52.74
53.16
6242
14.67
Brent crude oil prices are forecast to average $43/b in 2016 and $52/b in 2017, respectively. However, the current
values of futures and options contracts suggest high uncertainty in the price outlook13. Growing global oil supply
disruptions, rising oil demand, and falling U.S. crude oil production contributed to the price increase.
CONCLUSIONS
Although registered predict variation, the general trend in prices may continue to decline or be unstable at current
low levels production. With circumstances of no arrangement to standing productions level at least close to January 2016
level by the producing countries. The fear now is a continuation of this scenario for the history will repeat itself again and
same circumstances beyond the eighties and its accompanying no economic or social policies.
In forecasting the price of crude oil many of the pioneers of econometrics and modeling has gone to prefer simple
to complex models for a number of considerations, the most important and cost predictability. We have seen different
combination of Exponential Smoothing Model in term of time laps (weekly, monthly and annually) and forecasting
horizon. This related to the short-term forecasting of crude oil price sequences with slowly changing characteristics on the
basis of exponential smoothing models. The forecasting method which minimize mean squared error have select Holt’s
method almost optimized better in-sample. The comparative performance of these models is considered to show the
practical difference, on their forecasting ability. Where, accurate prediction of oil prices is difficult but very important for
governments, companies and investors. The more accurate the forecasts of oil prices, the greater the impact they have on
improving the accuracy of forecasts of a wide range of macroeconomic results. The complexity of the model does not
mean superiority over simple wording. This suggested some of key questions that arise in forecasting the price of crude oil.
What do applied forecasters need to know about the choice of model specifications? How useful are oil futures prices in
forecasting the spot price of oil?
13
In U.S. Energy Information Administration | Short-Term Energy Outlook (Analysis & Projections), June 2016,
http://www.eia.gov/forecasts/steo/.
Impact Factor (JCC): 2.4528
NAAS Rating: 2.25
35
Predict Crude Oil Prices in the International Market an Alternative Forecasting Technique
REFERENCES
1.
ALQUIST, R., KILIAN, L. and VIGFUSSON, R. J. (2013) Forecasting the price of oil. . In Handbook of Economic
Forecasting: Elsevier B.V., 427-507.
2.
ARTUS, P., CHEVALIER, J.-M., CHALMIN, P. and D’AUTUME, A. (2010) Les effets d’un prix du pétrole élevé et
volatil. In Book Les effets d’un prix du pétrole élevé et volatil (ed^eds Editor). City: Paris Dauphine University.
3.
BASHIRI BEHMIRI, N. and PIRES MANSO, J. R. (2013) Crude Oil Price Forecasting Techniques: a Comprehensive
Review of Literature. Alternative Investment Analyst Review, p32.
4.
BOX, G. E. P. and JENKINS, G. M. (1962) Some statistical aspects of adaptive optimization and control. Journal of
the Royal Statistical Society Series B 24, 297-343.
5.
BROOKS C. (2014) Introductory Econometrics for Finance.
6.
BROWN, R. G. (1963) Smoothing, Forecasting and Prediction of Discrete Time Series: NJ: Prentice-Hall.
7.
BROWN, R. G. and MEYER, R. F. (1961) The fundamental theorem of exponential smooth-ing. Operations
Research 9, 673-87.
8.
BROZE, L. and MÉLARD, G. (1990) Exponential Smoothing: Estimation by Maximum Likelihood. Journal of
Forecasting 9, 445-455.
9.
D’ESOPO, D. A. (1961) A note on forecasting by the exponential smoothing operator. Operations Research 9,
686-7.
10. GARDNER, J., E. S. and MCKENZIE, E. (1985) Forecasting trends in time series. Management Science 31 12371246.
11. GARDNER JR., E. S. (1985) Exponential smoothing: the state of the art. Journal of Forecasting 4, 1-28.
12. HACHEMANE, M. (2010) Time series and forecasting techniques. In Book Time series and forecasting techniques
(ed^eds Editor). City.
13. HARRISON, P. J. (1967) Exponential smoothing and short-term sales forecasting. Management Science 13, p821.
14. HENDRY, D. (2006) Robustifying Forecasts from Equilibrium-Correction Systems. Journal of Econometrics 135,
399-426.
15. HOLT, C. C. (1957) Forecasting Seasonals and Trends by Exponentially Weighted Moving Averages. In Book
Forecasting Seasonals and Trends by Exponentially Weighted Moving Averages (ed^eds Editor). City: Carnegie
Institute of Technology.
16. HOLT, C. C. (2004) Forecasting seasonals and trends by exponentially weighted moving averages. International
Journal of Forecasting 20, 5-10.
17. HUNTINGTON, H. G. (1994) Oil price forecasting in the 1980s: what went wrong? The Energy Journal, 1-22.
18. KALEKAR, P. S. (2006) Time series Forecasting using Holt-Winters Exponential Smoothing. In Book Time series
Forecasting using Holt-Winters Exponential Smoothing (ed^eds Editor). City: Kanwal Rekhi School of
www.iaset.us
editor@iaset.us
36
Siagh Ahmed Ramzi & Hachemane Mouloud
Information Technology, p.4.
19. MOHADDES, K. and RAISSI, M. (2015) The U.S. Oil Supply Revolution and the Global Economy. WP No. 15/259,
IMF Working Papers International Monetary Fund, p13.
20. MILLS, T. C. (2011) The Foundations of Modern Time Series Analysis: Palgrave Macmillan UK.
21. PINDYCK, R. S. (1999) The Long-run Evolution of Energy Prices. . The Energy Journal 20, pp 1-27.
22. SADORSKY, P. (2006) Modeling and forecasting petroleum futures volatility. Energy Economics 28, pp 467-488.
23. TAVAKKOLI, A., HEMMASI, A. H., TALAEIPOUR, M., BAZYAR, B. and TAJDINI, A. (2015) Forecasting of
Particleboard Consumption in Iran Using Univariate Time Series Models. BioResources 10, 2032-2043.
24. WINTERS, P. R. (1960) Forecasting sales by exponentially weighted moving averages. Management Science 6,
324-42.
Impact Factor (JCC): 2.4528
NAAS Rating: 2.25