=
OPTIMAL_FORCASTING_MODELS_FOR_THE_OIL_=
GAS_INDUSTRY_DAILY
DOW JONES_INDEX_CONOCOPHILLIPS_VIRTUAL_INTERNET_OFFICE
=
=
OPTIMAL FORCASTING MODELS FOR THE OIL =
&
GAS INDUSTRY: A STUDY OF THE =
DAILY DOW JONES U.S. OIL & GAS IN=
DEX AND
CONOCOPHILLIPS WITH THE A=
ID OF
THE VIRTUAL INTERNET OFFICE
=
<BF>Avi Rushinek, Ph.D. <RO>
<BF>Sara F. Rushinek, Ph.D. <RO> * **
=
=
=
This is
a study of the challenges and opportunities for improved forecasting techni=
ques
of the Oil and Gas Industry using the Virtual Internet Office.
Government and non-profit accounting issues are at the center of these
challenges since Oil and Gas companies are regulated. Due to the recent
advances of the Internet, the Virtual Internet Office (VIO) is no longer ju=
st a
theory, it is the reality which accounting oil and gas industry experts nee=
d to
know and apply. The authors discuss the importance of proper controls=
in regards
to budgeting and forecasting, and the use of appropriate quantitative metho=
ds such
as regression, trend, and statistical analyses.
Since this industry is regulated, the more important controls include a variety of forecasting models that governments and the industry can use to negotiate rates in banking and tax-planning application= s. This study uses some examples to d= emonstrate the importance of due diligence in applying the proper forecasting models.<= span style=3D'mso-spacerun:yes'> It shows the trade-offs between ea= se of use and simplicity of some models compared to other models that are more complex but have better accuracy rates.
This is especially important, since both the consumer countries, such as the United States, countries in Europe, and China, as we= ll as the supplier countries from the Oil Cartel, Middle East, Venezuela, and = Russia all deal with forecasting the markets and the industry on a regular and continuous basis. It is impor= tant to note that statistics can mislead when using inappropriate models, which = may be particularly common in regulated offshore cartel type industries, such as Oil and Gas industry.
=
<BF>Introduction<RO>
=
The oil and gas industry faces an environment that is becoming increasingly complex with changing local, national and internation= al regulations and standards. De= mands are pushing oil and gas companies to implement processes to mitigate risk, raise quality, ensure compliance and reduce losses.
The most common internal controls include budgeting and variance analysis which are based on forecasting techniques. Such forecasting methods can be ra= ther misleading if sufficient due diligence is not applied. This study demonstrates how to app= ly due diligence and the errors that may result from failure to apply such control= s.
This is a study of trend line and cash-flow forecasts.= Our research creates a budget audit for the Dow Jones U.S. Oil & Gas In= dex (^DJUSEN) and ConocoPhillips (COP), an integrated petroleum company. = This study tests the univariate, multivariate linear, and nonlinear modeling techniques, to determine which one is the best in terms of accuracy of statistical measures. An exam= ple is the correlation between actual and predicted values.
Likewise, our study conducts hypothesis testing to see whether the null hypothesis of no correlation can be rejected in favor of an alternate hypothesis. Furthermore, traders may consider integrating s= uch analysis with accounting software such as Microsoft Office Accounting (MOA)= , as well as spreadsheet analysis, such as Microsoft Excel using test and graphi= cs methods.
This study also develops an automated forecasting trend analysis based web query. Such methods will facilitate proper cash management and integration with Excel accounting software, and professional trading software such as Wealth-Lab and Trade-Station using its proprietary= Easy Language.
=
<BF>Information Technology (IT) And
=
As the global economy becomes =
more
dependent on computerized systems, oil and gas companies are relying more a=
nd
more on information technology for business operations.[1]
Oil companies are also invest=
ing
heavily in developing information systems that can create or sustain
competitive advantages in the market. Today’s Virtual Internet Offi=
ce
systems are becoming flexible enough to help oil and gas companies anticipa=
te
business needs and adjust ahead of market curves. Our study proposes using =
the
Virtual Office to help with optimal predictive modeling and strategic decis=
ion
making.
<BF>Automation and Forecasts of the Stock Mar=
ket
Trends in the Oil and Gas Industry<RO>
The Oil and Gas Industry continually monitors and forecasts its stock market position. Our research has added to the literature by developing ways using an internet approach to predict which oil companies represent the best values today and can deliver the most solid growth in the future.[2]
We also researched the role of the forensic accountant= in evaluating the forensic audits of stock market forecasting in the internati= onal oil and gas industry. We demonstrated financial models for online broker litigation.[3]
We continued adding on to the research by looking at forecasting models in the oil and gas industry. Since all companies currently refu= se to disclose their forecasting models, we devised a reversed accounting theory = that suggests ways for outsiders to use the theory on companies to anticipate th= eir performance. Or in the case of fraud, the forecast can be a red flag that a company may be cooking their b= ooks due to the fact that their performance is “too good to be true”= .[4][5]
One service that is rare in forecasting stock market prices in the Oil & Gas industry is the virtual office service. This is primarily due to the fact = that the technology is fairly new. The virtual office approach will lead to the installation of a web server even at an offshore location. This w= ay the business can be run from anywhere in the world. All it means, for example, is that an oil firm can drop a web server in the Bahamas, offering= any of the above services from anywhere in the world to any client in the world, without stepping foot in the offshore territory. With some local support, t= he local technician will ensure that the web server in up and running (working= ).
Our current study extends the research by looking at t= rend line and cash-flow forecasts of the Oil & Gas industry as a whole and f= orecasts individual oil companies. We want to know if the industry and company behave in the same pattern and whi= ch models will do the best in forecasting stock prices in the short term. Short term may be one, two or thre= e day forecasts. Various modeling techniques will be examined to determine the best results.
<BF>Polynomial
(Poly) Model a Univariate Non-linear Forecasting Model Trend Line OHLC Char=
t<RO>
Polynomial Model a Univariate Non-linear Forecasting M= odel Trend Line describes the relationship between the Julian date and the stock market values, along with an Open-High-Low-Close (OHLC) stock chart. = Dow Jones U.S. Oil & Gas Index (^DJUSEN) is an industry index of oil and ga= s stocks in which ConocoPhillips (COP) comprises one of the companies. This st= udy investigates which forecasting method is most accurate for daily stock quote forecast, for a single industry and a single company. Our assessment reveals that consistently the Polynomial Trend Line (Poly) is the most accu= rate model based on its R-Square value.
<BF>Data
Analysis and Data Sources for an Automated System<RO>
Data has been retrieved from Yahoo daily historical qu= otes for the Dow Jones U.S. Oil & Gas Index (^DJUSEN) and ConocoPhillips (CO= P).[6] ConcoPhillips is an integrated pet= roleum company with operations in nearly 40 countries and assets of approximately = 173 billion dollars. It is the th= ird largest integrated energy company in the United States.[7]
Using Excel Web Query, the data can be updated automat= ically and recalculated on a regular basis after market hours, so it will be updat= ed, when the trader is ready to trade. Once the data has been refreshed, = the spreadsheet recalculates all the models automatically, reproducing updated analysis.
=
<BF>Log. Expo, Linear & Power Models - Univariate Forecasting Model Trend Lines (Open High Low & Close) Chart= <RO>
=
Table 1 shows the daily Dow Jones U.S. Oil & Gas I= ndex (^DJUSEN) as of Monday, November 19, 2007, from Yahoo financial, historical quotes. The table headings are the models, rank, algorithms, R-square, and = the described functions.
Table 1
|
Model |
Rank |
Algorithm |
R-Square |
Function |
|
Monday, November 19, |
|
||
|
1 |
1 |
Poly |
R²=
=3D 0.693
|
y =3D 3E-07x6 - 0.065x5 + 648=
7.x4
- 3E+08x3 + 1E+13x2 - 2E+17x + 1E+21 |
|||||
|
2<= o:p> |
2<= o:p> |
Log |
R² =3D 0.293 |
y =3D -=
2116ln(x)
+ 22459 |
|
|
|
|
|
|
3 |
3<= o:p> |
Expo |
R² =3D 0.287 |
y =3D 1=
E+17e-8E-0x |
|
|
|
|
|
|
4<= o:p> |
4<= o:p> |
Linear |
R²=
=3D 0.281
|
y =3D -=
0.534x +
21685 |
|
|
|
|
|
|
5 |
5<= o:p> |
Power |
R² =3D 0.224 |
y =3D 2=
E+12x-27.0 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
The attributes Model, Rank, Algorithm, R-Square, and Function show how model 1 ranks as number 1. Thus, the top ranked model i= s the Polynomial (Poly). The highest R² is 0.693, and this explains why we rank this function, y =3D 3E-07x6 - 0.065x5 + 6487.x4 - 3E+08x3 + 1E+13x2 - 2E+17x + 1E+21, as the best model. The next model number 2 is ranked = as second which is the Logarithmic (Log) model. The number 3 rank is Exponential (Expo). The Linear model follows with numb= er 4. The Power model ranked number 5 making it last of the rankings..
<BF> ConocoPhillips (COP) Company Ranks Polynomial (Poly) as the Top Model<RO>
Just like its own industry index, ConocoPhillips (COP) Company ranks the Polynomial as the best model. As in the industry oil index, the = Poly ranks number one as the other models follow with lower rankings. This confirms our initial notion t= hat the Poly is the best Univariate Forecasting Model. However, the Polynomial model is a= lso more complicated than the linear models.&n= bsp; Therefore, many traders are using the linear models, especially to calculate budget numbers for accounting purposes since it is much easier to understand.
Table 2 shows the same picture for the individual comp= any, as we saw for the industry index, namely that the Polynomial is a superior univariate stock market forecasting model.= Furthermore, we may be able to see the point of entry into a trade a= fter the Poly trend line reaches a minimum point on the Monday, turning upward f= or the next day.
=
= Table 2 uses COP stock market price data to produce the following models: [8]
No. Rank Model &n= bsp; R**2 Function
Table 2=
a
1 1 Pol=
y R²
=3D 0.88 y =3D 1E-07x6-0.033x5+3290.x4-2E+08x3+5E+12x2-8E+16x+5E+20
2 2 Log=
&=
nbsp; R² =3D 0.678 y =3D -7265ln(x)=
+ 76958
3 2 Lin=
ear
R² =3D 0.678 y =3D -0.184x + 7349
4 3 Exp=
o R²
=3D 0.672 y =3D 2E+39e-0.00x
5 3 Power R² =3D 0.672 y =3D 2E+30x-86.2
Typically, COP demonstrates the pattern of the entire industry, the Poly model is the best univariate forecaster.
=
|
Table 2 |
|
|
|
|
|
|
|
||
|
No=
.. |
Rank |
Model |
R**2 |
Function |
|
|
|
|
|
|
1<= o:p> |
1 |
Poly |
R² =3D 0.88 |
y =3D 1E-07x6 - 0.033x5=
+ 3290.x4 - 2E+08x3 + 5E+12x2 - 8E+16x +
5E+20 |
|||||
|
2<= o:p> |
2 |
Log |
R² =3D 0.678 |
y =3D -7265ln(x) + 76958 |
|
|
|
|
|
|
3<= o:p> |
2 |
Linear |
R² =3D 0.678 |
y =3D -0.184x + 7349 |
|
|
|
|
|
|
4<= o:p> |
3 |
Expo |
R² =3D 0.672 |
y =3D 2E+39e-0.00x |
|
|
|
|
|
|
5<= o:p> |
3 |
Power |
R² =3D 0.672 |
y =3D 2E+30x-86.2 |
|
|
|
|
|
|
Table 2b |
|
|
|
|
|
|
|
|
|
|
|
|
|
Tuesday,
November 20, 2007 |
|
|
|
|
|
|
|
|
Budget |
=3DFORECAST($A$3+1,$G$3:G67,$A$3:$A$67) |
|
|
|
|
|
|
|
|
Budget |
Sale |
$ 83.70 |
Wednesday, November 21, 2007 |
Close |
|
|
|
|
|
|
Budget |
Cost |
$ 79.29 |
Tuesday, November 20, 2007 |
Close |
|
|
|
|
|
|
Budget |
Gain |
$
4.41 |
1 Day Lag Budget Gain |
|
|
|
|
|
|
|
|
=3DIF(K12>0,"Gain","Loss") |
|
|
|
|
|
|
||
|
|
Actual |
|
Wednesday,
November 21, 2007 |
|
|
|
|
|
|
|
Actual |
Sale |
$ 77.82 |
Close Wednesday |
|
|
|
|
|
|
|
Actual |
Cost |
$ (78.67) |
Open Wednesday |
|
|
|
|
|
|
|
Actual |
Loss |
$ (0.85) |
=3DIF(K17>0,"Gain","Loss") |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
=
=
=
= Table 2b shows a forecasted gain due to a short trade, expecting the stock market to= go down by about $4.41. This is = based on a linear forecasting model as follows:&n= bsp; = ; &n= bsp; = ; &n= bsp; = ;
= &= nbsp; &nbs= p; &= nbsp; Tuesday, November 20, 2007 &n= bsp; = ; &n= bsp; = ; &n= bsp;
= &= nbsp; Budget &= nbsp; =3DFORECAST($A$3+1,$G$3:G67,$A$3:$A$67) &= nbsp; &nbs= p; &= nbsp; &nbs= p; &= nbsp;
= Budget &= nbsp; Sale $ 83.70 Wednesday, Novemb= er 21, 2007 Close &= nbsp; &nbs= p; &= nbsp; &nbs= p;
= Budget &= nbsp; Cost $ 79.29 Tuesday, November= 20, 2007 Close &nb= sp; = &nb= sp; =
= Budget &= nbsp; Gain $ 4.41 1 Day Lag B= udget Gain &nb= sp; = &nb= sp; = &nb= sp;
= &= nbsp; =3DIF(K12>0,"Gain","Loss")<= span style=3D'mso-tab-count:6'> &= nbsp; &nbs= p; &= nbsp; &nbs= p; &= nbsp;
= &= nbsp; Actual &= nbsp; Wednesday, November 21, 2007 &n= bsp;
=
= Table 2b also shows how the trader will end up losing money due to the reversal of t= he linear downtrend that can be seen in a Polynomial chart of the 6th level. Following is the actual performance, resulting in a loss of about $.85 per share, as a result of a forecasting error:
= &= nbsp; &nbs= p; &= nbsp; &nbs= p; &= nbsp;
=
Actual Sale =
;
$ 77.=
82 Close Wednesday
= Actual Cost = ; $ (78.67) Open Wednes= day &= nbsp; &nbs= p; &= nbsp; &nbs= p; &= nbsp;
= Actual Loss = ; $ (0.= 85) =3DIF(K17&g= t;0,"Gain","Loss") &= nbsp; &nbs= p; &= nbsp; &nbs= p; &= nbsp;
= &= nbsp; &nbs= p; &= nbsp; &nbs= p; &= nbsp; &nbs= p; &= nbsp; &nbs= p;
= &= nbsp; &nbs= p; &= nbsp; &nbs= p; &= nbsp; &nbs= p; &= nbsp; &nbs= p;
= This analysis demonstrates how important due diligence is not only with an individual company, COP, but also for the entire Oil and Gas Industry. It shows that the Polynomial (Poly= ) model would normally be superior to the linear model, yet most traders and researchers, including commercial trading software developers (such as TradeStation, WealthLab, and eSignal) provide only built-in linear regressi= on functions, but not Polynomial functions.&n= bsp; Due to the simplicity and the greater ease of use, traders and expert witnesses prefer the linear model. <= /span>However, the linear model violates some basic assumptions, and therefore, is less accurate than the polynomial model. Therefore, due diligence dictates t= hat we should consider more than one model, especially whenever the models conflict.
=
= As in the current case, the linear is trending down, with a negative slope of -0.= 184x as the following function shows:
=
=
Linear &n=
bsp; R&s=
up2;
=3D 0.678 y =3D
-0.184x + 7349
=
= At the same time the Polynomial function shows an uptrend about the trading day of Wednesday, based on the following function:
=
=
Poly R²
=3D 0.88 &nbs=
p; y
=3D 1E-07x6 - 0.033x5 + 3290.x4 - 2E+08x3 + 5E+12x2 - 8E+16x + 5E+20
=
= Ignoring the conflict, may lead the trader and the expert to bet on a downturn in the price and on a short trade, when in fact there was an upturn in the stock of COP and a gain in the price.
=
= Due diligence behooves the analyst to look at multiple models and review their = pros and cons objectively leading to a well balanced opinion.
=
<BF>SUMMARY, CONCLUSIONS, & IMPLICATIONS<RO>
This study tests the daily Dow Jones U.S. Oil & Gas Index (^DJUSEN) as of Monday, November 19, 2007, from Yahoo financial, historical quotes using Logarithmic, Exponential, Linear, Polynomial, and P= ower Models. It uses Univariate Forecasting Models with Trend Lines of the Open, High, Low, and Closing (OH= LC) price of the day. We show that looking at only these charts could well mislead the analysts. One needs to consider both linear = and polynomial models. ConocoPhil= lips (COP), an oil and gas industry sample company ranks the polynomial model as= the better model. This is in agre= ement with the Oil & Gas industry index that also shows that the polynomial model = is a better predictor than the linear model.&nb= sp; This further demonstrates that due diligence requires the review of polynomial models in addition to the linear models to promote reliability a= nd accuracy of the analysis.
The conclusion of this paper is that the Virtual Inter= net Office can aid in automating continuous model forecasting techniques. Through time, the models can be te= sted and enhanced. Forecasting models within various oil and gas companies can be compared with the Oil & Gas Index and back tested to improve predictability. It is importa= nt to adapt due to the swift changes in the Oil and Gas markets.
=
* The authors are professors at the University of Miami
** = The authors will provide additional appendices, tables, charts, and program code listings, upon written request, to save space some information has been omitted.
[1] McLa= ne, Blake : “Overlooked considerations when investing in an IT infrastructure”; Oil and Gas Financial Journal, Vol 4 = Issue 12, December 2007.
[2] Rush= inek, A. and Rushinek, S., “Oil and Gas Market Sector & Company Relative Strength Congruity Trading Theory (MSCRSCTT): An Internet Automation Approa= ch Applied to Chevron Corporation’s Stocks”, Oil, Gas, & Energy Quarterly, June 2000, Vol. 48, No.4, 671-682.=
[3] Rushinek, A. and Rushinek, S., “Forensic Audits of Stock Market Forec= asts for Texaco & the International Oil and Gas Industry & Financial Mod= els for Online Broker Litigation”, Oil, Gas, & Energy Quarterly, June 2003, Vol. 51, No.4, 659-678.
[4] Rushinek, A. and Rushinek, S., “Internet Web Server Comparative Forecasting Sales Models: Reversed Accounting of Texaco Corp & the Oil & Gas Industry”, Oil, Gas, & Energy Quarterly, December 2005, Vol. 54, No.2, 341-350.
[5] Rushinek, A. and Rushinek, S., “Tax Fraud Forecasting for the Petrole= um Refining Industry: Financial Ratios & Reversed Accounting Theory Applie= d to Shell Oil”, Oil, Gas, & T= ax Quarterly, March 1997, Vol. 45, No.3, 341-350.
[7] ConocoPhillips www.conocophillips.com
[8] ConocoPhillips stock price data http://finance.yahoo.com/q/hp?s=3DCOP