St. Petersburg University
Graduate School of Management
[Master in Corporate Finance]
IMPLEMENTATION OF SIMULTANEOUS
EQUATIONS MODEL TO FORECASTING RESIDUAL
INCOME OF RUSSIAN OIL COMPANIES
Master’s Thesis by the 2nd year student
Sennikov Igor
Research advisor:
Associate Professor, Garanina T. A.
St. Petersburg
[2016]
Statement about the independent character of the master thesis
I, Sennikov Igor, (second) year master student, Master in Corporate Finance, program
«Management», state that my master thesis on the topic «Implementation of simultaneous
equations model to forecasting residual income of Russian oil companies», which is presented to
the Master Office to be submitted to the Official Defense Committee for the public defense, does
not contain any elements of plagiarism.
All direct borrowings from printed and electronic sources, as well as from master theses,
PhD and doctorate theses which were defended earlier, have appropriate references.
I am aware that according to paragraph 9.7.1. of Guidelines for instruction in major
curriculum programs of higher and secondary professional education at St.Petersburg University
«A master thesis must be completed by each of the degree candidates individually under the
supervision of his or her advisor», and according to paragraph 51 of Charter of the Federal State
Institution of Higher Education Saint-Petersburg State University «a student can be expelled
from St.Petersburg University for submitting of the course or graduation qualification work
developed by other person (persons)».
(Student’s signature)
26/05/2016 (Date)
Заявление о самостоятельном характере выполнения выпускной
квалификационной работы
Я, Сенников Игорь, студент второго курса магистратуры направления
«Менеджмент», заявляю, что в моей магистерской диссертации на тему «Применение
системы одновременных уравнений для прогнозирования остаточного дохода российский
нефтяных компаний», представленной в службу обеспечения программ магистратуры для
последующей передачи в государственную аттестационную комиссию для публичной
защиты, не содержится элементов плагиата.
Все прямые заимствования из печатных и электронных источников, а также из
защищенных ранее выпускных квалификационных работ, кандидатских и докторских
диссертаций имеют соответствующие ссылки.
Мне известно содержание п. 9.7.1 Правил обучения по основным образовательным
программам высшего и среднего профессионального образования в СПбГУ о том, что
«ВКР выполняется индивидуально каждым студентом под руководством назначенного ему
научного руководителя», и п. 51 Устава федерального государственного бюджетного
образовательного учреждения высшего образования «Санкт-Петербургский
государственный университет»
о том, что «студент подлежит отчислению из Санкт2
Петербургского университета за представление курсовой или выпускной
квалификационной работы, выполненной другим лицом (лицами)».
(Подпись студента)
26/05/2016 (Дата)
3
Abstract
Master Student's Name
Master Thesis Title
Igor Sennikov
Implementation of simultaneous equations
model to forecasting residual income of
Faculty
Main field of study
Year
Academic Advisor's Name
Description of the goal, tasks and main
Russian oil companies
Graduate School of Management
Corporate Finance
2016
Garanina T. A.
Main goal is development of simultaneous
results
equations model to forecasting residual
operating income and verification on the
sample of Russian oil companies. Main tasks conduct literature review on accounting based
valuation and choose the applicable method of
valuation; extend the chosen valuation model,
taking into account the features of oil industry
in Russia; collect the data for model
estimation; applicate the Simultaneous
Equations Model to the extended valuation
model; develop Visual Basic Application for
forecasting purposes of residual operating
income of Russian oil companies; make
managerial implications of the obtained
results. As a result of current research, there
was elaboration of forecasting model and
Keywords
Virtual Basic Application development.
Value Based Management, Valuation, Oil,
Residual Income, Simultaneous Equations.
Аннотация
4
Автор
Название магистерской диссертации
Игорь Сенников
Применение системы одновременных
уравнений для прогнозирования
остаточного дохода российский нефтяных
Факультет
Направление подготовки
Год
Научный руководитель
Описание цели, задач и основных
компаний
Высшая Школа Менеджмента
Корпоративные Финансы
2016
Гаранина Т. А.
Основная цель данной работы заключается
результатов
в разработке модели одновременных
уравнений и проверка модели на примере
российских нефтяных компаний. Задачи –
изучение литературы и выбор подходящей
модели оценки; расширение существующей
модели оценки, учитывая специфику
нефтяной отрасли; сбор данных для
выполнения эконометрических расчетов;
применение системы одновременных
уравнений для расширенной модели;
разработка программы в среде VBA; вывод
управленческих решений, основанных на
применении разработанной модели. В
результате проведенного исследования,
была разработана улучшенная модель для
оценки российских нефтяных компаний, а
также разработано приложение в среде
Ключевые слова
VBA для оценки.
Ценностно-ориентированный менеджмент,
Оценка, Нефть, Остаточный доход,
Одновременные Уравнения.
5
Table of Contents
Statement about the independent character of the master thesis..................................................... 2
Abstract............................................................................................................................................4
Introduction..................................................................................................................................... 7
Chapter 1. Accounting approaches to company’s valuation............................................................9
1.1 Value-Based Management......................................................................................................9
1.2 Residual Income Models......................................................................................................13
1.3 Simultaneous Equations Model........................................................................................... 18
Chapter 2. Methodology................................................................................................................ 25
2.1 Estimation Model Development and Data Description....................................................... 25
2.1.1 Application of Simultaneously Equations Model..........................................................25
2.1.2 Data and Descriptive Statistics......................................................................................29
2.1.3 Application Three-Stage Least Squares Initial Estimation............................................36
2.1.4 Extended model.............................................................................................................37
2.1.5 Generalized Two-Stage Least Squares Estimation........................................................43
2.2 Practical implications...........................................................................................................46
2.2.1 Application of Panel Vector Autoregression................................................................. 46
2.2.2 Virtual Basic for Application Program Description...................................................... 49
Conclusion..................................................................................................................................... 52
References..................................................................................................................................... 53
6
Introduction
Companies’ valuation is an essential part of investing activity. There is special discipline,
which main purposes are to create value and to measure it precisely – value based management.
Current masterwork is devoted to work with second part – valuation process.
There are different valuation model with different techniques and assumptions. Also,
from industry to industry, there is a necessity to apply different valuation techniques and use
industry-specific assumptions. In this masterwork, main valuation model are considered and
highlighted their main features. In present time, almost all valuation model are quite universal,
which could be good for general valuation purposes. On the other hand, there is the one main
disadvantage - they are not always effective in implementation for different industries. Main
accent is done on specific, narrow-specialized valuation model for industry valuation purposes.
Next important question, discussed in this masterwork - choosing the suitable and
interesting industry, which is open for improvements in valuation sphere due to its specific
operations. Russian oil industry was chosen for that role. In current masterwork eight biggest oil
companies are considered, still not the whole industry, so it will be more accurate to mention
companies’ valuation.
There was a separation an oil companies from gas extracting, because of existence of
some significant differences in processes and performance indicators. Actually, one important
mention about not working with whole industry is that this masterwork is focused only on
upstream operation as on the most important part for oil companies. Only upstream oil
companies have specific performance indicators like oil production or available oil reserves.
Due to working with production and exploration oil processes, main valuation model, on
which whole research is focused on – residual operating income model. This choice was done
after deep investigation of existing valuation models.
Next important research findings in the current masterwork is deep investigation and
study of simultaneous equation model, as a possible key agent for improving of model accuracy.
Based on previous researches done with study of simultaneous equations model, there found a
fact that this model could be implemented and used for improving of residual operating income
model – specially one unknown variable which need to be forecasted – residual income.
Residual operating income model was broadly studied and widely used for different
research purposes. Moreover, simultaneous equations model is quite known tool, especially for
economic purposes. But combination of these two model for improving of accuracy purposes
was never studied before. Also, if we are talking about implementation and development of
different valuation models, they are done predominantly for set of companies from different
countries or especially for USA companies. There is the main relevance of current masterwork.
7
As further improvement, researches done is this work can be extended and use not only for
Russian oil companies, but for oil companies based in different countries. Of course, for these
purposes, special assumption are needed to be done.
To summarize purposes of current work, this master work is dedicated to study of
Russian oil industry. The main purpose is to create a valuation model specifically for this
industry. This choice was done not accidently: oil industry has a significant share in formation of
Russian budget.
While performing research, very important is to precisely set a research goal, which will
be the main result of the work to achieve. In our case it is:
Development of simultaneous equations model to forecasting residual operating
income and verification on the sample of Russian oil companies.
It is very hard to achieve the research goal without setting a plan of steps with clear
research objectives. They are the next:
Conduct literature review on accounting based valuation and choose the
applicable method of valuation.
Extend the chosen valuation model, taking into account the features of oil industry
in Russia.
Collect the data for model estimation.
Applicate the Simultaneous Equations Model to the extended valuation model
Develop Visual Basic Application for forecasting purposes of residual operating
income of Russian oil companies.
Make managerial implications of the obtained results.
Results of current research done has a very strong managerial implication for a case of
working with real oil companies.
8
Chapter 1. Accounting approaches to company’s valuation
1.1 Value-Based Management
Before investing money into some company and even think about it, we need to
understand the main purpose of each company. However, it is not so easy to formulate this aim
precisely. Nowadays, there are two main theories.
Shareholders theory. According to Jensen (Jensen, 2001), main purpose of each
company is to increase its value for shareholders. In this theory, author means
also creditors, and by increasing its welfare, society receives positive impact.
Stakeholders theory. Second important theory is presented by Wallace (Wallace,
2003) and implicates not only shareholders and creditors, but all people somehow
involved in company’s operations and functioning. This theory present function
of the company as multi-factor.
Determining and finding this firm’s value is the main purpose of Value-Based
Management (VBM). For successful achievement of this aim, we need to clearly understand,
what is included is VBM and what steps we need to undertake for goal reaching. Understanding
of this mechanism help us in in determining crucial valuation points. Pursuant to Copeland
(Copeland, Koller, Murrin, 1995), structure of VBM can be drawn as next:
Figure 1.1 Structure of VBM
9
As we can see from the picture, effective increasing and value determining is not possible
without clear and effective management. More clear definition of difference between
“governance” and “management” is given by one of professors from Graduate School of
Management (Bukhvalov, 2005). We will not stop on governance and management and will go
more into valuation part.
In this masterwork, VBM was described as a necessary introduction for valuationstrategy for company. According to Volkov (2006), Value-Based Management consists of four
main modules: valuation, strategy, finance and corporate governance.
Valuation module. This module presumes choosing of appropriate valuation model
and all necessary assumptions and procedures for determining company’s value.
In addition, it implements monitoring system and crucial drivers for new value
creation.
Strategy module. Determines precise link between company’s value and its
business-strategies.
Finance module. Describes company’s financial policies, which directed to value
creating.
Corporate Governance module. Aimed on negotiations of shareholders’ and
managers’ interests.
This masterwork is aimed for valuation purposes, so we will concentrate only on one
module. We will examine valuation module in details, because valuation module is a core
architecture for VBM. Implementing this module means to consider and make necessary
decisions in its three parts (Volkov, 2006):
Which model to choose?
Which core indicators to choose?
How to implement it through value-drivers creation?
This masterwork will be built according to these steps. But before making any decision
about model, we need first to determine main purposes of all valuation models. It is quite
obvious that we need to determine company’s value. First, we need to answer the question: what
is the company’s value and on which type we need to be more focused? We have three main
indicators:
Book Value. This indicator based on difference between total assets and total
liabilities – very common measure could be seen in every Balance Sheet. Book
Value measure historical value of assets, not trading one. In other words, book
value of company is that its shareholder should receive in case of liquidation.
However, as assets could be trading at different value (actually, market value), so
10
this case is unlikely. Nevertheless, we could use book value as useful benchmark
for determining its value, even having recent study with loud name “Size Matters,
Book Value Does Not!” (Lambert, Hubner, 2014), which reconsider whole usage
of Fama-French CAPM model (Fama, French, 2004).
Market Value. This is indicator of what investors are willing to pay for company’s
assets. This indicator could differ significantly for book value and depends more
on investor’s expectations. Defining the market value of private companies is
quite difficult task, because they are not publically traded. Many adjustments need
to be done or many peers need to be taken to define this market value precisely.
Recent study (Destri, Picone, Mina, 2012) offers implementation of new
technologies for determining market value. We will not stop on them, because the
only thing we are interested in this masterwork about market value – determining,
is our target firm overvalued or undervalued.
Fundamental Value. Also, can be called as “intrinsic value” (Penman, 2010) or
“warranted value” (McTaggart, Kontes, Mankins, 1994). It is true actual value of
a company, the most interesting kind of value for us. The purpose of many
valuation tool is to precisely derive this company’s value, which can be quite
complicated not only for private companies. Fundamental value, opposite to
market, is driven by internal expectations, not public. Deep analysis of main
differences is presented in Brainard’s fundamental work (Brainard, Shapiro,
Shoven, 1990). We will not stop on describing of its features, just highlight that
this firm’s value is the most interesting for us.
When we are speaking about company’s value, we are most interested in fundamental
value. Fundamental value is only “true” indicator of company’s performance. According to
Damodaran (2003), company can have fundamental value if only creates flows to shareholders
or stakeholders. Nevertheless, choosing market, book or fundamental value will not cancel the
main balance equation. We will write it for fundamental value, as the most interesting indicator
for us:
V A (V F )=V E +V D ,
Where:
V A - fundamental value of assets (or V F - fundamental value of firm),
V E - fundamental value of equity,
V D - fundamental value of debt.
11
(1.1)
According to the formula (1), we can derive two main approaches to the fundamental
valuation of equity.
Operational Approach. Following this method, we determine fundamental value
of the firm as sum of discounted flows, associated with firm’s assets. Then it is
necessary to subtract discounted outflows, associated with debt. By doing this, we
receive fundamental value of company.
Capital Approach. Fundamental value of firm determined as discounted future
flows to shareholders.
Main difference in these approaches is in underlined assumptions and the process of
calculation of book value of debt. Here, in line with operational approach for calculating the
value of debt we need to discount all possible outflows, associated with debt. However, if we
need to calculate amount of debt using capital approach, it will be sufficient just to take book
value of debt.
Next interesting thing is that almost all valuation models can be viewed as equivalent
under certain assumptions. Volkov and Rudenko (2009) highlighted core assumptions for an
achieving of equivalence between certain models:
Discounted free cash flow model (DFCFM) and residual operating income model
(ReOIM) equivalence. Can be achieved by using the same discount rate and
general assumptions about calculation the terminal value (for example, by
applying constant earnings before interest or taking constant return on net assets).
Discounted free cash flow model (DFCFM) and residual earnings model (REM)
equivalence. Can be achieved by implementing two conditions: using general
assumption about calculation of terminal value and by calculating of WACC
based on fundamental value of financing sources for DFCFM.
Residual operating income model (ReOIM) and residual earnings model (REM)
equivalence. Can be achieved by calculating of weighted average cost of capital
based on fundamental value of financing sources for ReOIM.
Having two main approaches underlined, now we need to define what kind of flows we
may use. Here we have:
Cash flows
Residual Income flows
For better understanding and highlighting the core valuation models, table below is
displaying the most common approaches:
Table 1.1 Classification of valuation models (Volkov, 2006)
12
Cash Flows
V FCF
E =
(DFCFM)
∞
FCF j
∑ (1+k
j=1
j
W)
V DDM
=
E
−D BV
- free cash flow, k W -
FCF
d
∞
∑ (1+kj
j=1
E)
j
d - dividends, k E - cost of
equity
weighted average cost of capital,
D BV - book value of debt
IncomeResidual
Types of flows
Approaches to equity valuation
Operational
Capital
Discounted Cash Flows Models (DCFM)
Discounted Free Cash Flows Model
Dividend Discount Model (DDM)
Residual Income Model (RIM)
Residual Operating Income Model
Residual Earnings Model (REM)
OIM
V Re
=
E
[
(ReOIM)
∞
NA 0 + ∑
j=1
ReOI j
(1+k W )
NA - net assets, ReOI
j
]
∞
V REM
= E0+ ∑
E
−D 0
- residual
RE j
(1+k E ) j
- residual earnings, E
j=1
RE
operating income
-
equity
After consideration the most common valuation model, it is time to move further for
identification of main differences and features. Next part is dedicated to a closer look at residual
income models.
1.2 Residual Income Models
Company valuation is a crucial field for modern corporate finance. The aim of each
company is to increase its market value; therefore, there is a need of precise valuation models.
Who is a target for “holding” company’s value – stakeholders or shareholders – another field for
discussions (Jensen, 2001). This master work is dedicated to valuation models. Plenty of them
were invented in recent years. The most accurate of them were described with implementation
for Russian market by Rutgayzer (2007).
One important group, widely used by many financiers is valuation methods based only on
operating profitability and growth prospective. This paper focuses on one method from this
group - Residual Income model (RIM) (or Residual Earnings model), developed by Ohlson
(1995). He mentioned and proved relationship between accounting data such earnings, book
value, dividends and market value of a company.
Residual Income model was popularized by Penman (2007) in his textbooks dedicated to
financial analysis. In Russia, this method was studied by Volkov (2006).
13
From the very beginning, this valuation method found many researchers, who claimed
this method as superior by its accuracy. First, RIM was tested using large simple of firms during
15-years period prior to study (Bernard, 1995). Author concluded this model as effective and
more preferable to use for short time period predictions. Many later articles were dedicated to
analyzing of this model, implementing it for different companies and different industries.
Penman and Sougiannis (1998) support the superiority of accounting-based methods over
methods, which use cash flows. Then, Francis at al. (2000) stated better accuracy of RIM,
comparing to dividend discount model (DDM) and discounted cash flow model (DCF). Later,
Lundholm and O’Keefe (2001) stated that there are should be no difference between DCF and
RIM, because they are using the same underlying assumptions.
Nevertheless, RIM is universal method and could not provide same accurate results for
different industries. Perek (2012) in his study did not find significant difference between DCF
and RIM for Turkish textile industry.
Next step in improving accounting based models was done by Bergmann (2013). Author
developed her own model for improving income forecast as an input into valuation models. The
main purpose was to use publicly available information, instead of using just information,
provided by financial analysts. Barth (2005) used this information as a component to his equity
valuation model, specifying it as “other information”. He developed three linear information
valuation models (LIM) and tested them. His model were based on earnings too, as in
Bergmann’s (2013) work. The main difference was that Barth (2005) tested influence of
disaggregated earnings on equity forecast prediction. The accuracy of each model were tested by
comparing errors, without using analysts’ forecasts as benchmarks.
However, earlier, Ohlson (2001) in his later work was offering not to avoid analysts’
forecasts, but to use them as a proxy for new-derived models. Cheng (2005) supported this point
of view, mentioning that some useful information like industry characteristics can be found only
in analysts’ forecasts. Not all researchers support using analysts’ forecasts as benchmarks, or
somehow trying to implement them into valuation model. Richardson and Tinaikar (2004) earlier
wrote that it “is a tautology that one can outperform a parsimonious relation when analyst
forecasted abnormal earnings are added to the model containing only accounting realizations” (p.
228).
To find a compromise, Bergmann (2013), in addition to her simultaneous equations
model (SEM) used not only method of comparing errors, but benchmarks as autoregressive
model and random walk model. As a result, model of income forecasting, more stable and
accurate for crisis years, were developed. In this paper, inspired by work of authors from
Graduate School of Management (Bukhvalov et al. 2012), new model derives from universal one
and applies for reality of Russian oil industry.
14
This masterwork is dedicated to one question – how to improve Residual Income
Valuation Model (RIM or RIV) and make more precise for current industry. First, we need to
determine this model and decide which weak parts could be improved. Significant contribution
to creation of this model was done by Ohlson (1995). Nevertheless, first attempts to formulize
and derive this method were done much earlier – by Edwards & Bell (1961) and then, later by
Peasnell (1982). In the simplest form, we could write this formula as:
RI t
∞
V 0 =BV 0 + ∑
t=1
( 1+r )r ,
(1.2)
where:
V 0 - value of a company at time zero,
BV 0 - book value of company at time zero,
RI t - residual income at time t ,
- interest rate.
One important note – the formula (1.2) could be applied for mature company only, with a
r
constant growth rate g . Next step is an adjusting the formula for company, having a fact this
company is still growing. Here we need to introduce “terminal value” – a company’s present
value of all cash flows from some point of time in future (time m ), when we, as analysts,
expect company will have constant growth rate. Different approaches could be done to this
calculation, including most popular one – Gordon Growth Model (Gordon and Shapiro, 1956).
More complicated and formal approach could be done according to Levin and Olsson (2000). In
our masterwork we will not stop on deep analysis of these methods, just rewrite RIV formula in
the next way for still growing companies:
m−1
V 0 =BV 0 + ∑
t=1
RI t
Tm
+
(1+r )r ( 1+r )m−1 ,
(1.3)
where:
V 0 - value of a company at time zero,
BV 0 - book value of company at time zero,
RI t - residual income at time t ,
r
- interest rate,
T m - terminal value at time
m .
Therefore, the main purpose of this model is to determine the “true” value of a company
for present moment. Looking at this formula and having assumption like constant interest rate
r
, one important question arises – how correctly forecast residual income? To answer this
15
question, we need to dig into structure of residual income and understand the algorithm for its
deriving.
Here we need to make some distinguishing between residual income and residual
earnings. For that purpose, we need to determine components of earnings. According to Penman
(2007), three crucial components are in the scheme:
Figure 1.2 Components of Earnings and Book Value
First, let us rewrite residual earning model for equity valuation:
V 0E=CSE 0 +
RE 1
1
( 1+r E )
+
RE 2
2
(1+r E )
+
RE 3
(1+r E )3
+...
(1.4)
where:
V E - value of equity,
CSE - common shareholders’ equity,
RE
- residual earnings,
r E - cost of equity.
Second, we need to write equation, how we receive residual earnings:
RE t =Earnings t −r E⋅CSE t−1
(1.5)
Where:
CSE - common shareholders’ equity,
RE
- residual earnings,
r E - cost of equity.
REM anchors valuation of equity on the book value of equity. The result is forecasted
earnings minus required return on book value. Now, if we take residual earnings from net
operating assets, we receive residual operating earnings (ReOI). Here we consider the most
important valuation model for current masterwork. The formula is given below:
16
[
∞
ReOI j
j=1
(1+k W ) j
OIM
V Re
= NA 0 + ∑
E
]
−D 0
(1.6)
Where:
NA - net assets,
ReOI - residual operating income,
D - debt,
k W - weighted average cost of capital.
As for improving this model for oil companies’ valuation, we are most interested in
residual operating income, which calculated as:
ReOI t =OI t⋅(1−tax )−WACC⋅NOA t−1 ,
(1.7)
For further work, we need to determine operating income and derive it with a usage of
sales and operational expenditures:
ReOI t =(SALES t −OPEX t )⋅(1−tax )−WACC⋅NOA t−1 ,
(1.8)
Where:
SALES - total revenues,
OPEX - operational expenditures,
WACC - weighted average cost of capital,
NOA - net operating assets.
The formula (1.8) is the most important for us and improvement of its variables is main
purpose of this masterwork.
This part was dedicated to model choosing and explaining the main features of the
decision. Next part will consider the simultaneous equations model and uncover, why this model
is so important for developing of residual operating income.
1.3 Simultaneous Equations Model
What is simultaneous equations model (SEM) and how it is different from structural
equations model? SEM is one form of structural equations, also SEM have a “structural” part in
their building form. A lot of attention were paid to study of SEM by Cowles Commission 1, the
foundation at Yale University, which purpose is conducting various of economic researches. This
1http://cowles.yale.edu/ Official website of Cowles Commission
17
foundation is very important for econometric development and its researches are highly cited in
many scientific and econometric articles.
The difference between SEM and the whole group of structural equations were
underlined by Reiss and Wolak (2007). Simultaneous equation model is the model, where several
equations with mutually or jointly dependent endogenous variables. For better understanding of
that kind of structural equations, where is the best way to present the most popular economic
functions: supply and demand. McFadden (1999) firstly proposed this simple explanation.
Demand function:
Q dt =α 0 +α 1 Pt +α 2 Y t +u1t ; α 1 α 2 <0
(1.9)
Supply function:
Q st =β 0 + β 1 P t +β 2 Z t +u2t ; β1 β 2 >0
(1.10)
Equilibrium condition:
Q dt =Q st
(1.11)
Where:
Qdt - quantity demand,
Q st - quantity supply,
P
Y
Z
- price,
- disposal income,
- raw materials costs,
t - time.
By combining demand function (1.9), supply function (1.10) and the main equilibrium
(1.11), we receive a system of simultaneous equations. In received system both supply and
demand functions needed to be found simultaneously.
In this masterwork two types of variables will be mentioned several times: endogenous
and exogenous. For further work, there is a need of precise definition of them.
Endogenous variables
If we are talking about endogenous variables in a system of simultaneous
equations model, those variables are jointly depended. Their values are
determined within the model. In other words, we need to find exactly their value.
Exogenous variables
Those variables in a context of the system of simultaneous equations have
specified values from the very beginning. They are predetermined outside the
system, in other words, they are given. They often known as non-stochastic.
18
The figure below is digging into further structure of endogenous and exogenous
variables:
Figure 1.3 Types of Variables in SEM
After defining endogenous and exogenous variables, there is one important addition of
SEM is complete or incomplete condition. We need to define endogenous variables and for that
purposes we have an amount of equations. Let us consider the next income/consumption system
of simultaneous equations:
C t =β 0 +β 1 y t +ut
Y t =C t +I t
(1.12)
Where:
C - consumption,
Y - income,
I - investment,
t - time,
u - error term.
Further, there is a necessity of precise bullet pointing the difference of SEM not from
structural equation, but from single equations model. Features of structural equation modelling
(as group, includes simultaneous equations) first was highlighted by Karimimalayer and Anuar
(2012). Earlier, a guide for structural equations modelling was compiled by Schumacker and
Lomax (1996). The table below shows the main differences:
Table 1.2 Main difference of SEM from single equation model.
Single Equation Model
Simultaneous Equations Model
In this type of model, single relationship Relationship inside this type of model
19
between variables is presented.
includes various interdependencies between
One equation.
variables.
Two and more equation without any
limitations in number.
Ordinary least squares (OLS) estimation may No possibility of using OLS, because of
be used.
interdependencies presence.
Parameters for ingle equation can be Using of just one equation in a system and
estimated just by a one variable
usage of just one variable will not help us in
overall system estimation.
Equations in SEM can be presented in different forms:
Structural form.
This form show the complete system of equations. Here, the structure of all
relationships between variables is fully presented. Endogenous variables in this
case are presented as functions of other endogenous variables, exogenous
variables and disturbance terms.
Reduced form.
The other type of expressing SEM, a model, where endogenous variables are
expressed as explicit function. This function is taken of predetermined variables.
In other words, here endogenous variables not expressed through other
endogenous variables, but only with predetermined exogenous variables and
stochastic disturbance.
When we are talking about simultaneous equations and forms of their expression, there is
a need of implementing and explanation of recursive model.
Recursive model.
A way of expression the equations with a special order of equations. Here, the
first equation includes only those variables, which are predetermined, in a right
side. Next, second equation includes not only its predetermined variables in the
right side, but also endogenous variables from first equation. All further equations
expression follows the same logic and depends on amount of equations in the
system.
As formula, recursive form of SEM could be presented as next:
Y 1=f ( x 1 , x2 ,.. ., x k u1 )
Y 2=f ( x 1 , x2 ,.. ., x k Y 1 ,u2 )
Y 3=f ( x1 , x2 ,. .. , x k Y 1 , Y 2 ,u3 )
Y 4 =. ... ...
20
(1.13)
Next, after consider all possible types of SEM and different ways of expression, it is time
to move to visual and understandable way of showing the model – graphical. But first, we need
to define the generally accepted notation of SEM (Bollen, 1989):
Table 1.3. Notation for Simultaneous Equation Models (Bollen, 1989)
Vector/Matrix
Variables
Definition
x
y
ζ
Dimensions
Endogenous variables
Exogenous variables
Disturbance terms/errors
p×1
q×1
p×1
Coefficient matrix for the exogenous variables;
p×q
Coefficients
Γ
effect of exogenous variables on endogenous
variables; direct effects of x on y
Coefficient matrix for the endogenous
Β
p× p
variables; effect of endogenous variables on
endogenous variables; direct effects of
y
on
y
Covariance Matrices
Φ
Covariance matrix of exogenous variables, x
Covariance matrix of exogenous variables,
Ψ
q×q
p× p
ζ
Next important thing about SEM is the pictorial way of representation. It called path
diagrams. Having theoretical model, for better understanding it is important to see, how variables
are related to each other. Usage of path diagrams follows some basic rules:
Observed variables are shown in rectangles
Direction of influence are denoted by usage of single-headed arrows
Double-headed arrows depict covariance, which is not uncovered in the model
Unobserved and latent variables usually depicted unenclosed, but also could be
shown in ovals.
Let us provide an example of SEM path diagram, which formula is:
y1 =γ 11 x 1 +γ 12 x 2 +ζ 1
y2 =β 21 y 1 +γ 22 x 2 +ζ 2
Where:
y - endogenous variables,
x - exogenous variables,
21
(1.14)
γ
- path coefficients for exogenous variables,
β - path coefficient for endogenous variables.
Here is quite simple example of SEM (1.14) with only two equations; also, this model is
recursive, following the logic of equations order. Figure below (1.4) depicts the given SEM
(1.14) in a graphical way:
Figure 1.4 Path Diagram
Now, it is time to link all theoretical background about SEM to an implementation for
residual operating income and discuss previous research work in this fieald. Having
interdependencies in operating income’s drivers, there is a need of using SEM instead of just
single equations. Relations between different accounting numbers are described by Tsal et al.
(2008). Simultaneous equations were widely used before for valuation purposes. Nature of SEM
is described by Wooldridge (2006), given with the meaning of simultaneity: it is “when one or
more of the explanatory variables is jointly determined with the dependent variable” (p. 546).
Recent fundamental research in company valuation, which could be counted as a framework,
was done for a case of emerging markets (Ntim et al. 2015). This paper describes correlation
between board size and firm’s valuation, showing positive results. Earlier, sales forecasting with
implementation of inventory and gross margin data was done for retail industry using SEM
(Kesavan et al. 2010). Bergmann (2013) did her study in the similar way, but focusing on income
forecasting instead of sales, and without constructing model for particular industry.
The framework, used by Bergmann (2013) and this masterwork, are suggested by
textbooks about financial analysis (Penman, 2007; Lundholm and Sloan, 2007) – having
interdependencies, forecasting process is not linear.
Bergmann (2013) in her model identified central value drivers: sales, operating
expenditures and net operating assets. This value driver – endogenous variables – were modeled
with a set of simultaneous equations. Aim of this masterwork is to complete author’s model with
industry specific exogenous variables.
Having a set of observations and SEM, there is not possible to use method of ordinary
least squares (OLS). Having interdependencies, created regression will be biased and
inconsistent. For this case, two-stage least squares method (2SLS) and three-stage least squares
22
method (3SLS) are more suitable for obtaining values of parameters. Application of these
methods into SEM are searched in details by McFadden (1999).
23
Chapter 2. Methodology
2.1 Estimation Model Development and Data Description
2.1.1 Application of Simultaneously Equations Model
The model in its first form will be align with model, proposed by Bergmann (2013).
Author did a fundamental work, building models using simultaneous equations. She proved in
this work existence of interdependencies between endogenous variables (SALES, OPEX and
NOA) and its dependency on a set of exogenous variables. The amount of observations was very
big – more than 19 thousands. The only one problem that this model covers very different
industries. Having this too universal model, it is very doubtful to receive precise results for every
industry. First step was precise determination the core of the model, proposed by Bergmann
(2013). Author proposed four models and compare it using mean errors (median average
percentage error– MdAPE; mean absolute error – MAE; root mean square error – RMSE). The
most suitable model with less errors is the next:
SALESit =α 10 +α 11 NOA it +α 12 OPEX it +α 13 SALES it−1 +α 14 OB it−1 +
+α 15 FedRate it−1 ++α 16 ΔGDP t +α 17 HI t−1 +α 18 CapIntI t −1 +α 19 MS it−1 +
+α 110 CapIntF it−1 +ε it
OPEX it =α 20 +α 21 SALESit +α 22 OPEX it−1 +α 23 MS it−1 +α 24 Δ PPI t +ηit
NOA it =α 30 +α 31 SALESit +α 32 NOA it−1 +α 33 FedRate t−1 +υit
(2.1)
Where:
SALES = net sales,
OPEX = operating expenditures,
NOA = net operating assets,
OB = order backlog,
FedRate = federal funds rate,
ΔGDP = change in gross domestic product,
HI = Herfindahl index,
CapIntI = capital intensity of industry,
CapIntF = capital intensity of firm,
MS = market share,
Δ PPI = change in product price index
Bergmann called this model (2.1) as a perfect foresight model. The estimation was done
using three-stage least squares model (3SLS). Why exactly to use this model, not ordinary least
squares (OLS)? Actually, in this case we have system of structural equations (in our case –
24
simultaneous), where we have endogenous variables (SALES, OPEX, NOA in our case) among
explanatory ones. And, by having disturbance correlated with endogenous variables, we violate
the main assumption of OLS model. Other word, estimating equations by OLS, we will force to
do it separately, without taking into accounting interdependencies between them. Further, as we
have explanatory variables among dependent in other equations, the error terms between
equations will be correlated. Finally, results of our estimation will be biased. According to
Davidson and MacKinnon (1993), 3SLS model uses an instrumental variable approach for
estimates and generalized least squares (GLS) for taking into account correlation structure in the
disturbances. Greene (2012) underlined three steps in 3SLS model:
First step
During this step, there is creation of instrumented values for all endogenous variables.
The main purpose of these instrumented values is that they can be considered as
predicted. This results from creating a regression of each endogenous variables on all
exogenous. This step is crucial one for performing whole estimation and identical to
one, which is using is two-least squares estimation (2SLS).
Second step
This step implicates an obtaining of consistent estimate for covariance matrix of
equation disturbances. As a base for this step, model is using residuals from 2SLS
estimation of each equation.
Third step
On this step, 3SLS model using estimates covariance matrix again, but now
performing GLS-type estimation. Also, all endogenous variable on the right side of
equations (which were determining) are replaced with instrumental variables.
Now, we can see a logic why this method is called “three-stage least squares”. Logic of
choosing this method is quite understandable – we simply cannot perform good estimation model
using OLS in our case. Actually, there are several methods for performing good unbiased
estimations for structural equations. One of them is two-stages least squares (2SLS). This
method is quite similar to 3SLS but have simpler estimation process: consists of two steps
instead of three. 2SLS is the extension of OLS method. According to Angrist and Imbens (1995),
we can derive principles of both steps of the model:
First step
25
There is a creating of new variables in this step. This process is performing with a use
of instrumental variables.
Second step
In this step, there is an estimation of variables, created in previous step. Then, actual
values are being replaced by new variable and model starts performing simple OLS
estimation.
Usually, 3SLS model is more preferable for simultaneous equation than 2SLS. Generally,
reasons for choosing model estimation is up to researcher, but sometimes, there are some
external limiters for models using. For example, in this masterwork statistical analysis was done
in program environment Stata 12, where usage of 3SLS estimation is limited. Nevertheless, first
step of performing technical analysis was to derive core model from model, proposed by
Bergmann. This step is necessary, because proposed foresight model contain many not-industry
specific variables, which will affect precise estimation for our chosen industry.
The core model, taken as a base for further analysis has the next form:
SALESit =α 10 +α 11 NOA it +α 12 OPEX it +α 13 SALESit −1 +ε it
OPEX it =α 20 +α 21 SALESit +α 22 OPEX it −1 +ηit
NOA it =α 30 +α 31 SALESit +α 32 NOA it−1 +υit
,
(2.2)
Where:
SALES = net sales,
OPEX = operating expenditures,
NOA = net operating assets.
The one independent variable, which could be used for further analysis is order backlog
(OB). This is firm-specific indicator, which, probably, has a huge influence on a sales forecasting
(Elliot and Uphoff, 1972). Order backlog is the sales order waiting to be filled, actually, a
negative component, whom all companies want to avoid. Unfortunately, as we a focusing current
study on Russian industry, there is no consistent information about order backlog. That probably
could be improved in a future with developing of data disclosure politics. Nevertheless, for now
it is just a possible way for future improvement.
For achievement of set goals, we need to do precise estimation and to find a between all
set variables (SALES, OPEX and NOA) and their interdependency. Actually, later we will be
adding another independent industry-specific and firm-specific variables for more precise
estimation. To find these interdependencies and influences, we will perform different regressions
and make regression analysis. Therefore, first, we need to define statement “regression analysis”
and understand the process behind it.
26
Regression analysis is a statistical process for estimation relationships between variables.
It used for statistical models – sets of taken assumptions and received data. So, in regression
analysis, we built and estimate regressions – approaches for relationship modelling. All
regression can be depicted in the next form:
Y =f ( X , β )
(2.3)
Where:
Y - dependent variable,
X - independent variable,
β - unknown parameter.
In more simple form, regression is a function of Y on
X , and the main purpose is to
estimate unknown variable β . What kind of function and how to estimate these unknown
variables are the most important research questions. There are many different types of
regressions for different purposes and we will not stop an each of them. Main types of regression
and logic of estimation will be explained further, during the current regression analysis.
Now we have an understanding, with which kind of data we will be operating and in next
part descriptive statistics will be described. But, first, we need to pay attention on statistical
software for performing regression analysis.
Usually, the choice of statistical program is up to researcher. In this masterwork, main
criteria were: friendly interface and availability for usage. Under these criteria, Stata 2 was chosen
as a main tool for regression analysis. This package was created in 1985 by StataCorp and
remain one of the most popular tool of general-purpose statistical analysis. Table below
illustrates versions of data and capabilities of this software package:
Figure 2.1 Stata Package version and capabilities
2http://www.stata.com/ Stata Package Official website
27
2.1.2 Data and Descriptive Statistics
Oil and gas exporting were driving Russian economy for many years. Exactly these
sources of energy helped Russia to recover from depression in 1999.
Three main issues arise from an oil policy inside the country:
1) Natural characteristics of oil and its reserves
2) Political governance
3) Technical and economic sides of oil exploration
This purpose of this master work is to dig into the third issue.
Even our national currency depends on world prices of crude oil. Influence of oil
exporting on monetary policy is overviewed by Russian ex-minister of finance (Kudrin, 2014).
Nevertheless, this work is not going to overestimate the value of this industry to Russia.
In many news sources, it is written that contribution of oil industry into federal budget for year
2105 was more than 43%3. We need to distinguish federal and consolidated (including regions)
budgets. Oil industry contribution into consolidated budget is less than 20% 4.
Second reason for creating valuation model for oil sector in Russia is a fact that our
country has a seventh place in amount of proven oil reserves. 5 Only six countries: Venezuela,
Saudi Arabia, Canada, Iran, Iraq and Kuwait have bigger oil reserves. However, according to the
fact that oil is a strategic raw, information about reserves could be distorted by some countries.
In previous year, there was extracted 532.9 million tons of oil (including gas condensate)
in Russia.6 In comparison with year 2014, oil extraction had increased by 1.3%. Amount of
exported oil is 244.5 million tons, which is 9.4% more than in year 2014. Share of imported oil is
45.9%.7 In 2015, weight of exported oil of all exported goods was 26.1%, in 2014 – 30.9%.
Almost 300 companies in Russia have government permission for oil extracting, having more
than 30% as vertical integrated companies.
Having such sharp decline in oil prices, there is a hard situation for all Russian energy
market. Cash inflow from oil exporting was 87 billion dollars, much less than in 2014, it was
almost 150 billion dollars.
The perspectives of Russian oil industry depends on world prices on crude oil. Moreover,
Russia needs to be aware of movements of main competitors – Saudi Arabia and USA. But oil
prices cannot be the only crucial factor in oil extracting policy. Different regions have different
costs of oil extracting – from 4 dollars in Saudi Arabia to 32 dollars for American slate oil. For
recently opened oilfield in Russia, costs are 16 dollars per barrel, for old oilfields – just 6.
3http://ria.ru/infografika/20130912/958932396.html
4http://info.minfin.ru/kons_doh_isp.php
5https://www.cia.gov/library/publications/the-world-factbook/rankorder/2244rank.html
6http://www.vedomosti.ru/business/articles/2016/02/03/626546-dobicha-nefti-rossii-rastet-blagodaryabashnefti-gazprom-nefti-tatnefti
7http://www.gks.ru/bgd/free/b04_03/IssWWW.exe/Stg/d06/38.htm
28
Many researches were dedicated specifically to forecasting of Russian oil industry.
Sample comprises top Russian oil companies. Search was done using sources, available
in library of Graduate School of Management. First step was downloading all necessary
information using Thomson Reuters Datastream. Then, due to not large sample size, to avoid any
elimination of omitted variables, all variable very checked manually with Thomson Reuters
Eikon. Because of working with top Russian oil companies, more detailed information about
some missing or questionable variables were checked in SCREEN and SPARK. All variables
measured in dollars – there is not possible precisely measure firm’s performance in rubles due to
high correlation between crude oil price and ruble’s appreciation/depreciation. All data is taken
for period last 15 years. From year 2000 until year 2014, which information was available for
most companies for the moment of writing this masterwork. This period includes increasing
price of oil from year 2000, period of world economic crisis 2007-2009 and crisis in Russia in
2014. The best conditions for testing the model.
Here is short review of eight Russian oil companies, which variables were taken and used
in current research:
Bashneft. One of the oldest Russian oil companies, established in 1932. Has 170
oil plants in three oil extraction regions in Russia – Volga-Urals province, TimanPechora and Western Siberia. Oil production is about 17 million tons per year. For
year 2014 has 3,644 million barrels of 3P (proved + probable + possible) reserves.
Classification of oil reserves type will be given later. 3.4% share in oil production
in Russia. Total revenues for December, 31, 2014 were 637,231 million of
Russian rubles with operating profit of 76,997 million of rubles. 8
RN Holding. Formerly TNK BP, company was acquired by Rosneft in 2013.
Previously one of the leading Russian oil companies, third place in amount of oil
extraction with a share of 15% in Russia. Had been operating in Western and
Eastern Siberia, Far East and Volga-Ural region. Turnover for last functioning
year was 688 billion of Russian rubles with operating profit of 202.5 billion
rubles.9
SurgutNeftegaz. Third biggest company in amount of oil production – 61 million
tons and 12% share in Russia. Company operates 22,380 wells and has a class C1
77 million tonnes of recoverable oil reserves. Company’s operations
predominantly done in Western Siberia, having 62 out of 67 oil field there. For
year 2014 company’s total sales were 863 billion of Russian rubles with pretax
profit of 1,068 million rubles.10
8http://www.bashneft.com/ Official website
9http://www.finam.ru/ Article 2014 about RN Holding’s performance
10http://www.surgutneftegas.ru/ Official website
29
Slavneft. Russian oil company. Has a eighth place in oil extraction amount with a
share of 2.9%. Operates on 31 oil fields in Khanty-Mansi Autonomous Area and
Krasnoyarsk region. In year 2014 had 7,831 million of 3P oil reserves. Oil
production is 16.2 million tons of crude oil. Total turnover was 10,334 million of
Russian rubles with total loss on sales of 807 million rubles.
Rosneft. The largest world company in oil production and amount of oil reserves.
In year 2014 oil extraction was 204.9 million of tons. Share in Russia is 37%.
Number of C1 + C2 hydrocarbon reserves is 129 million of barrels. Rosneft
operates in Western Siberia, Eastern Siberia, Timan-Pechora, Central Russia,
southern part of European Russia and the Russian Far East. Sales were 5,503
billion of rubles wuth EBITDA of 1,057 billion of rubles in 2014.
Lukoil. One of the largest Russian company with revenues exceeds 10.000 billion
rubles and net income of 350 billion rubles. Company has a 16.4% share in
Russian oil extraction amount and 2% share of world oil extraction. Lukoil
operates in 559 oil plants and has 27.8 billion barrels of 3P oil reserves. Main
company’s operations are in Western Siberia, Ural and Timan-Pechora regions.
Tarneft. Russian oil company, which primarily operates in European part of
Russia. Company has 851.5 million barrels of proved oil reserves and produce 26
million tons of oil annually. Main crude oil amount comes from six oil fields.
Turnover in 2014 was 392 billion rubles with net income of 82 billion rubles.
The list includes the eight biggest Russian oil companies. The main financial variables
were taken from Thomson Reuters Eikon, in same accounting standard – IFRS.
It was mentioned above that we operates with data in statistical models, so we
need to describe different types of data and to use it for regression analysis. We will not dig deep
into specifications of quantitative or qualitative data and highlight only three main types of data
we will be interested in current research work.
Cross-sectional data
This type of data means cross section of a study population. It can be collected by
observing many different subjects – in our case, for example, firms. The main
principal is to take different subjects and to compare their differences at one period of
time, of without taking into account any differences in time periods.
Time-series data
Different type of data, in which accent is taken for measure time points. Sample
usually is just one subject, small-scale or aggregate data taken in different periods of
30
time. Purpose of this type of data is to find differences not between subjects, but
between their measures in different points of time.
Panel data
Also called time-series cross-sectional data. This type of data is a combination of
previously described methods. It is a multidimensional data involving measures over
time. Here we obtain data about different subjects, their measurements over time and
look at both parameters simultaneously.
As we described in introductory part, main formula of residual operating income we will
be working with is following:
ReOI t =(SALES t −OPEX t )⋅(1−tax )−WACC⋅NOA t−1 ,
(2.4)
Where:
SALES - total revenues,
OPEX - operational expenditures,
WACC - weighted average cost of capital,
NOA - net operating assets.
Endogenous variables (EV) are determined inside the model system. They will be chosen
as a main value drivers. We discussed those three main value drivers previously: SALES, OPEX
and NOA. As we can see, all those variables appear in the formula (2.4). One more variable is
WACC, which will be taken separately and will not be included in simultaneous equations
model. Let us give precise determination of an each variable.
Sales
Actually, sales forecasting is the basic step of almost all forecasting models. This
variable often determines, how company will be growing and developing in the
future. The main weak point of forecasting sales is that often this step is only one
done for forecasting process (Koller, 2010). Main value drivers exactly for sales are
competition level, product substitutes, possible patents and any positive expectations,
associated with brand (Penman, 2010). Actually, amount of factors, which influence
sales is much wider, so, often, they are just not included into forecasting assumptions
(Richardson, 2010). One possible factor influence sales is interdependency with
operational expenditures and net operating assets. The purpose of this masterwork,
once again, is to create disaggregated residual operating income forecast, taking
into accounting possible interdependencies between independent variables. Sales
31
amount are taken directly from databases like Thomson Reuters Database without any
additional calculations.
Operating expenditures
Knowing sales given level, operating expenditures determines the level of profit
margin. Usually, profit margin is driven by economy of scale, consummation of
technology process, competiveness in supplier, labor sources, and so on (Penman,
2010). Koller (2010) recommends forecasting of OPEX in line according to SALES
level. So, more SALES, more OPEX and otherwise. Usually, this model should work,
logically. However, this is truth not for all cases, OPEX level can easily have other
pace, than SALES, depending on market conditions. OPEX cannot be taken directly
and needs to be calculated. The formula for deriving this variable is next:
OPEX=COGS+ SG&Α+ D&A ,
(2.5)
Where:
OPEX
COGS
SG&A
D&A
- operational expenditures,
- cost of goods sold,
- selling, general and administrative expenses,
- depreciation and amortization.
In this masterwork, we will be looking at OPEX of biggest Russian oil companies.
According to Cheng (2005), big market share leads to better manifestation of
economy of scale and more efficient production.
Net operating assets
With determining level of OPEX and knowing level of SALES, we estimate level of
profit margin. Profit margin is a measure of profitability. In contrast to profit margin,
asset turnover measures a production efficiency. It measures potential of assets to
boost and generate sales. Prediction of sales amount will help us in determining of
future net operating assets amount only indirectly, we need to take into account future
changing of working capital and long-term assets. There is a huge chain of dependent
value drivers. For example, long-term assets are determined by property, plant and
equipment ( PP&E ), which are driven, in its turn, by implementation of new
technology and new products creation. The same logic with determining drivers of
working capital by splitting it into accounts payable, inventories and accounts
receivable. At the same time, net, working capital and property, plant and equipment
can be forecast depending on sales amount, and what is why we again can implement
32
rude assumption of taking net operating assets as percentage of sales. And, again, we
found aa evidence of interdependency and that is why we will support an
implementation of simultaneous equations model. But first, we need to take a data
about net operating assets in different companies. Again, as in a case with operating
expenditures, this parameter could not be taken directly. This variable will be
calculated in accordance with a set of studies (Nissim and Penman, 2001; Callen and
Segal, 2005; Soliman, 2008). Net operating assets are defined as operating assets (OA)
minus operating liabilities (OL). So, we have a following formula:
NOA=OA−OL ,
(2.6)
Where:
NOA - net operating assets,
OA - operating assets,
OL - operating liabilities.
At the same time, we need to define operating assets and operating liabilities:
OA=TA−FA ,
(2.7)
FA=CSTI +OIA ,
(2.8)
Where: OA - operating assets,
TA - total assets,
FA - financial assets,
CSTI - cash & short-term investments,
OIA - other investments and advances.
OL=TL−FL,
FL=LTD+ DCL+ MI ,
(2.9)
(2.10)
Where: OL - operating liabilities,
TL - total liabilities,
FL - financial liabilities,
LTD - long-term debt,
DCL - debt in current liabilities,
MI - minority interest.
By combining all above, we can define net operating assets as following:
NOA=TA−CSTI −OIA−TL+LTD+ DCL+ MI .
(2.11)
We defined all three main endogenous variables. Next part will be devoted to different
ways of result estimation.
33
2.1.3 Application Three-Stage Least Squares Initial Estimation.
As for now, the main priority is to check the usage of this core model for Russian oil
industry. Without this step, we cannot process with further analysis. For these purposes, this
model were estimated as in underlying article (Bergmann, 2013) with a use of a 3SLS
estimation. Our sample consists of an eight companies for 15 years. Having some missed
variable, in total, we operated with 106 observation. Having multidimensional data, our data has
a panel, or longitudinal, type. However, for estimating system of equations (2.2), we have a right
to look only at a company-year observation and use 3SLS model, without paying a lot attention
now for determining each observation’s company or year.
Received coefficients are presented in the table (2.1) below:
Table 2.1 Three-stage least squares estimation
Equation
(1)
SALESit
Equation (2) OPEX it
Coef. [Std. Errors]
Equation (3) NOA it
Coef. [Std. Errors]
0.3066*** [0.1431]
-0.1895 [0.4664]
Coef. [Std. Errors]
SALESit
OPEX it
NOA it
SALES it−1
OPEX it−1
NOA it−1
-0.0782 [0.2727]
0.0029 [0.0065]
1.2419*** [0.2342]
0.8424
R2
*p<0.1, **p<0.5, ***p<0.01
0.7327*** [0.2200]
0.8738*** [0.0554]
0.9337
0.7108
While estimating system of equations (2. 2), were received a P-value < 0.0000 for all
three equations, which means, we have significant equation and can use them for further
analysis. From the equation (1), we can see that only SALES amount in previous year can have
some influence on SALES in present period. Having positive coefficient and its number of
1.2419 with probability of mistake is less than 5 %, this result is quite expectable. Coefficient for
OPEX in present time is negative, which can be explained by economy of scale, but it is not
significant having that big standard error and big probability of mistake. The same situation with
net operating assets – seemingly no influence of net operating assets on sales amount in present
year. R-square for equation (1) is quite big (0.8424), so regression approximates real data point
quite well.
34
Result for operation expenditures were more expected. As in Bergmann’s (2013) model,
we receive a high level of significance for sales coefficient (0.3066 with a p < 0.05). The positive
sign could be explained as if we have more sales, we need (or, in some cases, may) spent more
money for operating expenditures. The link between operating expenditures in present and past
year confirmed. Coefficient for OPEX in previous year is 0.7327 with high level of significance
(p < 0.05). Having coefficient of determination equals 0.9337, we can say, that, generally,
regression fits the given data very good.
Last equation (3) is dedicated to finding possible dependency of net operating assets in
present year on sales level in present year and net operating assets in year t−1 . Actually, by
performing estimation, only NOA it−1
coefficient was found as significant (0.8738, p < 0.05).
The coefficient number close to one tells us about close net operating assets year-to-year amount.
Determination coefficient is 0.7108, so goodness of fit of current regression is normal one.
The reason, why some coefficients became insignificant is that we have incomplete
model with few independent variable. Having SALES, OPEX and NOA as dependent and
determining variables simultaneously will not permit us to receive precise results. Nevertheless,
by performing this step, we have found that simultaneous equations model is applicable for
Russian oil industry.
Next step will be dedicated to finding specific exogenous variables.
2.1.4 Extended model
For this moment, we are on way of deriving o new model from core one. This model will
not be completely new – it just contains some improvements for residual operating income
model, taking into account interdependency between endogenous variables.
Now, we need to decide, which exogenous variables we need to include into model. Here
is list of variables with detailed explanations, why do we need to include them into model and
influence on which parameter we expect to receive.
Oil price
By oil price, we mean price of Brent Crude, which is a benchmark for oil pricing
worldwide. This parameter should obviously influence amount of sales for oil
companies. It is quite obviously, if we will look on one of fundamental economic
equations:
R=P⋅Q ,
Where:
R - revenues or sales,
35
(2.12)
P
- price,
Q - quantity.
Fundamental implementation of oil price into econometric model was done by
Osmundsen et al. (2006). Although, the main purpose of research was to define most
appropriate measure for oil companies performance: return on average capital
employed (RoACE) or enterprise value to debt-adjusted cash flow (EV/DACF), the
significant influence of oil prices were strengthen. The model, proposed by author,
has the next form:
n
EV / DACF= A i +αP t + ∑ β i KPI it +γR it +ut ,
i=1
(2.13)
Where:
EV - enterprise value,
DACF - debt-adjusted cash flow,
P - crude oil price,
KPI - vector of key performance indicators,
R - return on average capital employed.
This model is given not for interpretation purposes, but as an example of fundamental
work findings about its significant influence. Oil prices were taken from Yahoo
Finance11 on a daily basis for more precise results calculation. Then, annual average
price were calculated. As we are working with annul data for all other variables, by
take average annual oil price, we ignore sharp changes in oil prices movements.
Nevertheless, the whole trend of oil pricing is taken into account.
Oil price volatility
That means oil price volatility? Volatility is measure of statistical dispersion on some
given object (usually, market security). In our case, it is a low responsiveness or
"inelasticity" of both demanders and suppliers of quick price adjusting in a shortterm. The most recent work done in this field was dedicated to study oil price
volatility as a main factor for oil speculation (Beidas-Strom and Pescatori, 2014). By
doing a VAR (vector autoregression) estimation, author found a significant influence
of oil price volatility on oil supply/demand equilibrium changes. Damodaran (2005)
in his book “Investment Valuation” made a fundamental analysis with a set of
commodities companies with a next implementation “value of a natural resource
company depends not just on the price of the natural resource but also on the expected
volatility in that price”. That is why we decided to include this parameter into our
11http://finance.yahoo.com/ Official website
36
model. Actual historical oil price volatility was calculated directly from given oil
prices, without searching it on external sources. By knowing daily prices and its daily
logarithmic return, we may calculate annual volatility by formula:
σ=
σ SD
√P
,
(2.14)
Where:
σ - volatility,
σ SD - standard deviation,
P - time period (day).
For calculating annual volatility, we need to add into formula 252 trading days in one
year:
σ ANNUAL=σ SD⋅√252 ,
(2.15)
Having calculated annual volatility for 15 periods, we will input it into our model.
Oil Production
In recent study “Oil and gas production handbook” (Devold, 2013), focus was done
on upstream part of oil production and on one the most important drivers – amount of
oil produced per year. As we are working with Russian oil industry, we cannot ignore
this factor as Russia is one of the world’s leaders in amount of oil produced. For the
end of 2015 situation was next:
Country (MMbbl/day)
14.4%
%
34.8%
%
%
4.8
2.8
3.5 %
%
3.6%
11.5%
%
4.6%
4.2%
%
2.8
12.5%
%
U. S. (13.7)
Saudi Arabia (11.9)
Russia (11)
China (4.6)
Canada (4.4)
Iraq (4.0)
UAE (3.5)
Iran (3.4)
Mexico (2.7)
Kuwait (2.7)
Other World (33.1)
Fig. 2.2 Daily crude oil production12
12Source: US Energy Information Administration
37
So, oil production is very important indicator of company’s performance, because it
has a high influence at least on sales amount. This influence was underlined by
Gurvich (2004) in his macroeconomical valuation of Russian oil sector. Number of oil
production amount mostly available in companies’ annual reports. Sometimes it is
given in cubic metres. Conversion factor taken the same as set by New York
Mercantile Exchange (NYMEX)13 – 6.29287 barrels per cubic metre.
Oil Reserves
Oil reserves is the current amount of technically available oil. Next very important
parameter for precise companies’ performance calculation. Then we are talking about
oil reserves, we need to determine their types:
1) Proved reserves. Classification as P1 . Those reserves has a level of confidence of
90% that this reserve oil can be extracted and used (Wright and Gallun, 2008).
Additional division if for proved developed reserves (PD) and proved undeveloped
reserves (PUD). First type reserves can be extracted with existing technical
equipment and wells or with minimal investments. Second type require additional
capital raising.
2) Probable reserves. In the oil industry proved + probable reserves referred as 2P.
Those reserves claimed as 50% confidence as being extracted.
3) Possible reserves. Proved + probable + possible - 3P classification. Has only a 10%
confidence level due to geological interpretation or current absent of technology.
Since year 2010, Security Exchange Commission (SEC) 14 allowed companies to
publish information about all types of reserves in their reports. Previously, only
information about proven reserves was allowed for publishing. In Russia, categories
A, B and C1 refer to proved reserves. Category C2 includes both probable and
possible reserves15. Russia has a huge oil reserves, but not the leading position in
world:
13http://www.cmegroup.com/company/nymex.html NYMEX official website
14http://www.sec.gov/ Security Exchange Commission, official website
15Source: Society of Petroleum Engineers
38
Venezuela
Saudi Arabia
Canada
Iran
Iraq
Russia
Kuwait
UAE
U. S.
Libya
0
50
100
150
200
250
Fig. 2.3 World proven oil reserves by country, BBL
300
16
Oil reserves is the most difficult to find data, because Russian companies are not
obliged to publish this information in their reports. Nevertheless, information was
found and presented for further research.
EV/EBITDA and EBITDA/barrel ratios
As for the last independent variables, decision was to take some industry ratios. First,
it is very popular one for different industries: enterprise value to earnings before
interest, taxes, depreciation and amortization. Second, it is more industry-specific
ration - earnings before interest, taxes, depreciation and amortization to annual oil
production in barrels. For more precise understanding why we need to use these
ratios, we derive formulas for each part of rations. Enterprise value is an economic
measure of a company in according to its market value. This metric do not take into
account capital structure of a company. That is why, if we are comparing company
with a very different level of debt, this ratio can be more precise than, for example,
Price/Earnings ratio. The general formula of EV is:
EV =CSE+ D+ MI + PE−C&CE ,
Where:
EV - enterprise value,
CSE - market value common shareholders’ equity,
D
- debt at market value,
16Billions of barrels, source: US Energy Information Administration
39
(2.16)
MI
PE
- minority interest,
- market value of preferred equity,
C&CE - cash and cash equivalents.
In denominator of this ratio there is an EBITDA. Formula can be written as next, in
according with our main variables:
EBITDA=SALES−OPEX+ D&A ,
Where:
EBITDA
(2.17)
- earnings before interest, taxes, depreciation and amortization,
SALES - total sales or revenues,
OPEX - operational expenditures,
D&A - depreciation and amortization.
Ratio EV/EBITDA is used as the most important for valuing capital-intensive
companies with a high level of depreciation and amortization. Second ratio is more
specific for oil industry and uses in denominator amount of crude oil in barrels
produced annually. This metric is widely used by oil companies, for example, by
Rosneft.
We discussed six main components, which will be included into the model as independent
variables. The table below presents the short summary about all variables and their probable
influence on SALES, OPEX or NOA:
Table 2.2 Exogenous variables.
Variable
Oil Price
Description
Expected sign
Macroeconomic factors
Oil price for a year at time t . Taken as annual + (SALES)
- (OPEX)
average value. Measured in dollars.
Oil price volatility for current year. Calculated on - (SALES)
Volatility
historical basis. Taken as average annual volatility at
Oil Price
+ (OPEX)
time t .
Oil Production
Company-specific factors
Annual oil production in barrels for year at time t .
Oil Reserves
Proved oil reserves in barrels for year at time t .
EV/EBITDA
Ratio, enterprise value of company to earnings before
+ (SALES)
- (OPEX)
+ (SALES)
+ (NOA)
- (SALES)
interest, taxes, depreciation and amortization for year + (OPEX)
EBITDA/Barrel
at time t .
Ratio, earnings before interest, taxes, depreciation and + (SALES)
amortization to annual oil production in barrels for
year at time t .
40
On this stage, we are interested in interpretation of interdependency of endogenous
variables and influence of new included variables, not in forecasting process. So, all variables
will be taken in present time t .
2.1.5 Generalized Two-Stage Least Squares Estimation
Now, there is a question of which estimation model to choose. The first estimation was
executed with a help of 3SLS model with a rough estimation of just taking company-years
observation, without paying attention on panel data. Now, we cannot perform estimation this
way. The main problem is that we have now not only endogenous variables, which are unique for
each company in each year, but some exogenous variables, which are the same for all companies
in one year. 3SLS model is not working with panel data, at least in Stata environment. The
choice was done on G2SLS model (Balestra and Varadharajan-Krishnakumar, 1987) – this a
model uses instrumental variables and 2SLS estimation for panel data. This models offer five
different estimators, which are called models with different effects. When we are speaking about
panel data estimation, we are usually mean two types of models:
Fixed Effects Model
It is a statistical model, where observed quantities are not random. The fixed effects
assumption is that individual effects are correlated with the independent variables. If
random effect assumption holds, the fixed effects model are less efficient, but when it
does not holds, we may use only fixed effects model.
Random Effects Model
Also known a variance components model. This is a type of hierarchical linear model,
whose difference of population relates to its hierarchy. In our case, random effects
model is used for panel data with assumption of no fixed effects.
Actually. There are several more models like Between-Effects model (BE) or FirstDifference model (FD), but they are not useful in this masterwork. Here, or choice will be done
between Fixed Effects model and Random Effects model. For determination, which model to
choose, usually Durbin–Wu–Hausman (also known as Hausman specification test) is used
(Hausman, 1978). This tests evaluates consistency of estimators, under null hypothesis is
Random Effects, under the alternative – Fixed Effects model.
After running the regressions using Fixed Effects model and Random Effects model
(there is an algorithm of performing this test), Hausman showed us that the more consistent
41
variant will be to use Random Effects model. Actually, there is one more implementation of
2SLS with Random Effects, except G2SLS – EC2SLS (Baltagi, 2013). But second model is not
so popular because of its specification.
After describing and including exogenous variables, our model will have next form:
SALESit =α 11 OPEX it +α 12 NOA it +α 13 SALES it−1 +α 14 Oilprice t +
+α 15 Oilvolatility t +α 16 Oilreserves t−1 +α 17 Oilproductiont−1 +
+α 18 EV / EBITDA t−1 +α 20 EBITDA /Barrel+ε it
OPEX it =α 21 SALES it +α 22 OPEX it−1 +α 23 Oilprice t +α 23 Oilvolatility t +
+α 18 EV / EBITDA t−1 +ηit
NOA it =α 31 SALESit +α 32 NOA it −1 +α 33 Oilreserves t−1 +υit
(2.18)
Where:
SALES - total sales or revenues,
OPEX - operational expenditures,
NOA - net operating assets,
Oilprice - average annual price for barrel of crude oil,
Oilvolatility - average annual volatility for price of barrel of crude oil,
Oireserves - company’s oil reserves,
Oilproduction - company’s annual oil production,
EV / EBITDA
- enterprise value to earnings before interest, taxes, depreciation and
amortization.
EBITDA /Barrel
- earnings before interest, taxes, depreciation and amortization to oil
annual production in barrels.
By performing G2SLS estimation, received coefficients are next:
Table 2.3 Generalized two-stage least squares estimation results
Equation
(1) Equation
SALES it
Coef. [Std. Errors]
SALESit
OPEX it
NOA it
SALES it−1
OPEX it−1
NOA it−1
Oilprice
Oilvolatility
Oilreserves
Oilproduction
(2)
OPEX it
Coef. [Std. Errors]
0.8629*** [0.0329]
Equation (3) NOA it
Coef. [Std. Errors]
-8.8891* [6.4647]
1.1340* [0.0363]
0.0045*** [0.0032]
0.0204 [0.0355]
0.0448 [0.0474]
0.7298*** [ 0.0985]
4.2298*** [17.0101]
-392.5103 [3380.9874]
-0.3466*** [0.2281]
14.5890*** [3.8557]
-5.7582*** [15.9715]
-1164.238 [3938.259]
5.0383 [4.8324]
42
EV / EDBITDA
EBITDA /Barrel
-199.8089 [166.8622]
21.7232 [45.7983]
2
0.9823
R
*p<0.1, **p<0.5, ***p<0.01
263.0131 [180.4787]
0.9850
0.7039
After estimating system of equations (2.16), were received a P-value < 0.0000 for all
three equations, which means, we have significant equation and can use them for further
analysis. From the equation (1), we can see that only part of coefficients become significant for
SALES influence. Coefficient of OPEX at time t is significant, but only with confidence level of
10%, sign is positive. Coefficient for NOA at time t is unexpectedly small – 0.0045 with a high
level of significance (p<0.01). In contrast with initial model (2.2), by implementing extra
exogenous variables, SALES influence at time t-1 become insignificant. This fact was quite
surprising. As it was expected, oil price has a high significance level of influence on sales. As we
can see, from the result of regression analysis, coefficient of oil volatility influence on SALES is
not significant. Both oil production and oil reserves showed a significant influence on SALES
with coefficients 14.5890 and 0.3466 respectively. Both coefficients for EV/EBITDA and
EBITDA/barrel seems to be insignificant. R-square for equation (1) is quite big (0.9823), so
regression approximates real data point quite well.
In equation (2) coefficient for SALES is significant (p<0.01) with value of 0.8629. Again,
in contrast with system of equations (2.2), endogenous variable at time t-1 is not significant (in
our case - OPEX it−1 ). Oil price coefficient is significant with value of -5.7582. Negative sign
follows the logic that with higher price OPEX will be less. In equation (2) as in equation (1),
coefficients for oil price volatility and ration EV/EBITDA become insignificant. Having
coefficient of determination equals 0.9337, we can say, that, generally, regression fits the given
data very good.
Last equation (3) is dedicated to finding possible dependency of net operating assets.
Actually, by performing estimation, two out of three coefficients was found as significant.
Coefficient of SALES at time t is found as significant, but at confidence level of only 10%. High
level of confidence showed a dependency on NOA at time t-1 (0.7298, p<0.01). Determination
coefficient is 0.7108, so goodness of fit of current regression is normal one.
By performing G2SLS regression, now we can see a set of exogenous variables investor
should pay attention while performing valuation in Residual Income Operating Model. As it was
found, investor should carefully estimate oil price, oil production and reserves, not only predict
SALES amount. But for now, it very complicated to take into account all these variables and
their interdependencies, so next part will be dedicating to building forecasting model based on
this SEM model, studied in chapter 2.
43
2.2 Practical implications
2.2.1 Application of Panel Vector Autoregression
For further performing of forecast analysis, we need first to define correct method.
Simultaneously taking interdependency of endogenous variables and lagged exogenous variables
will be technically difficult. Under these conditions, taking G2SLS estimation is not beneficial.
In this masterwork, choice was done on panel vector autoregression models (PVAR). First timeseries vector autoregression model was introduced in literature as useful alternative of
multivariate simultaneously equations models (Sims, 1980), which we used in second chapter.
The main feature of this model is that all variables is treated as endogenous, although some
restrictions exists (Abrigo and Love, 2015). From the very beginning of program development,
PVAR models did not work with panel data. It found huge implementation in different fields and
researches after adjusting settings for working with panel data (Holtz-Eakin, Newey and Rosen,
1988).
First usage of PVAR in Stata was described in details by Love and Zicchino (2006).
Actually, this package is not in-built in Stata, so there is a necessity of downloading of this extrapack. For performing estimation by PVAR and receive forecasting model, we need to take lagged
variables from all variables (endogenous and exogenous). Which variables to include was
decided in previous chapter – only those, which have a significant influence on SALES, OPEX
and NOA. Received model should have the next form:
SALES it =α 10 +α 11 SALES it−1 +α 12 OPEX it−1 +α 13 NOA it−1 +α 14 Oilprice it +
+α 15 Oilproductionit−1 +α 16 Oilreserves it −1 +ε it
OPEX it =α 20 +α 21 SALES it−1 +α 22 OPEX it−1 +α 23 NOA it−1 +α 24 Oilprice it +
+α 25 Oilproductionit−1 +α 26 Oilreserves it−1 +ηit
NOA it =α 30 +α 31 SALES it−1 +α 32 OPEX it−1 +α 33 NOA it−1 +α 34 Oilprice it +
+α 35 Oilproductionit−1 +α 36 Oilreserves it −1 +υit
(2.19)
Where:
SALES - total sales or revenues,
OPEX - operational expenditures,
NOA - net operating assets,
Oilprice - average annual price for barrel of crude oil,
Oireserves - company’s oil reserves,
Oilproduction - company’s annual oil production.
As you can see from formula (2.19), oil price is taken at time t. The reason is that oil
price is too volatile and its implementation just ruins the models. Experimentally were proved
that after inputting of lagged oil price, all coefficients become insignificant. From this
44
assumption, we have one very important notification about received model – it should go align
with oil price forecasting.
All coefficients by PVAR estimation are presented in the table (2.4) below:
Table 2.4 Panel vector autoregression estimation results
Equation
SALES it
(1) Equation
(2)
OPEX it
Equation (3) NOA it
Coef. [Std. Errors]
Coef. [Std. Errors]
-0.0954 [.5207]
Coef. [Std. Errors]
-0.9792*** [ 0.6028]
0.4094 [0.3299]
0.2467 [0.3673]
1.0526*** [0 .4272]
-0.5500*** [0.2166]
1.0540*** [0.5679]
1.2747*** [0.6473]
0.6377** [0.3642]
0.0889 [0.4439]
1.7775*** [0.7022]
Oilreserves
0.8613** [0.5788]
Oilproduction
*p<0.1, **p<0.05, ***p<0.01
-0.0917 [0.5229]
2.0023*** [0.7971]
0.8497* [0.6105]
0.6452*** [0.64523]
-0.1898 [0.3806]
-0.1627 [0.4788]
SALES it−1
OPEX it−1
NOA it−1
Oilprice
As we can see, repercussion of interdependency is highly noticeable in forecasted model,
especially for OPEX: its future value is highly dependent on today’s SALES, OPEX and NOA
performance. Dependency on future prices is necessary only for more precise calculation of
NOA. It is even better for analysts, because future oil prices are unknown for now and highly
volatile, thereby complicating it’s forecasting.
After estimating system of equations (2.19), were received a P-value < 0.0000 for all
three equations, which means, we have significant equation and can use them for further
analysis. From the equation (1), we can see that only part of coefficients become significant for
SALES influence. Both coefficients for SALES and OPEX at time t-1 is insignificant, therefore,
they will not be used for forecasting purposes. Coefficient for NOA at time t-1 is close to one –
1.054026 with a high level of significance (p<0.01). In contrast with SEM estimation (2.18), oil
price is not significant for influence on SALES. This fact was quite surprising. Both oil
production and oil reserves showed a significant influence on SALES with coefficients 0.8613
and 1.7775 respectively.
In equation (2) coefficient for SALES is significant (p<0.01) with value of -0.9794. The
same situation with significant coefficients for lagged OPEX and lagged NOA. Again, in contrast
with SEM estimation (2.2), oil price is not significant for influence on endogenous variable (in
our case, OPEX). As in equation (2), both oil production and oil reserves showed a significant
influence on SALES with coefficients 0.8497 and 2.0023 respectively.
Last equation (3) is dedicated to finding possible dependency of net operating assets.
Actually, by performing estimation, only half of coefficients was found as significant.
Coefficient of SALES at time t is found as insignificant, the same unexpected result for oil
45
reserves and oil production. Both coefficients for OPEX and NOA at time t-1 is significant, but
with different levels of confidence, p<0.01 and p<0.05, respectively. As for oil price, finally, at
least, for NOA, it become significant with a high level of confidence. We estimated coefficient
and decided, which of them we need to consider and forecast for to receive results that are more
precise. Nevertheless, for now it is quite uncomfortable to use just number, thereby there was
taken a decision to “pack” all receive data into user-friendly interface and to create a computerforecasting program.
2.2.2 Virtual Basic for Application Program Description
Received results from G2SLS and PVAR estimation are quite difficult for interpretation.
Yes, by performing estimation, we know on which variables investor should be focused mostly.
Also, we found which variables are not significant, moreover, we built forecasting model with
precise structure and dependence on lagged variables. Very important finding is building
valuation forecasting align with forecasting of oil prices, because oil price is a very volatile and
cannot be plugged into the model.
In a purpose to have stronger practical implementation and for real practical usage of
current work, was created a software in a Virtual Basic for Application (VBA) environment.
General purpose in in automatization of forecasting process. All we need is to input some
company’s variables and run the forecasting process.
The main feature of the current program is possibility to work in two different modes:
forecasting and historical checking.
Forecasting mode.
This mode is created for forecasting purposes of any oil companies. Code is based
on PVAR estimation. All received coefficients are plugged in the code. This mode
is the best suitable for Russian oil companies, but can work with oil industry in
different countries.
Historical checking mode.
This mode is created for model accuracy testing. Here, forecasting process is
using too, but with different data. All we need to do is to choose the one of eight
Russian oil companies and year of checking. All historical data automatically will
be plugged in the part of detailed companies’ data. Then, forecasting process will
be performed with historical data. Finally, there is possibility to check historical
residual operating income and forecasted residual operating income based on
model, created in current masterwork.
46
Program interface is presented below:
Picture 2.1 Oil Companies Valuator
1. Here is the part refers to the general data, which should be inputted every time for
different valuations. First field devoted to the current year of forecasting as a
starting point. It is not crucial part, but for better understanding, current year
should be inputted, and forecasting years will be taken as +1, +2, +3 respectively.
2. This box is for weighted average cost of capital. This parameter is different for
different companies and even for one company in different periods of time, so it
should be inputted manually every time.
3. Last box in general data is for tax rate. Usually it is 20% for Russian companies,
but decision of manual input was taken for a case of changing in legislation or
working with different countries.
4. This part is devoted for more detailed data about company’s performance. This
part is filling only for forecasting purposed, not for historical checking of model
accuracy. First box is for SALES amount in millions of dollars.
5. Second box in this part is for operational expenses in millions of dollars.
6. Next box represents net operating assets amount expressed in millions of dollars
too.
7. Further, here is a place for annual oil production amount in barrels.
47
8. This box is created for oil reserves amount in barrels. Having many companies
with resumption rate of more than 100%, this number is slightly increasing from
year to year, but not significantly.
9. Last free fields in this part are devoted to crude oil prices for next three years. As
it was mentioned above, this parameter is quite volatile and hard to predict, so oil
prices could not be implemented into the model.
10. Here is the short instruction, which explains how to use two different modes of
the program.
11. This part is useful for checking mode – it automatically loads and fills empty
boxes with companies’ parameters. Here is the choice of eight Russian oil
companies, whose data was gathered for current research.
12. This button launch forecasting mode.
13. In these three boxes next three forecasting years are shown.
14. Here forecasted residual operating income is shown after forecasting process.
15. Button activates historical checking mode with all historical data loaded for
companies’ performance and, at the same time forecasting process based on
historical data.
16. Here posted historical results and years. To receive any data here, last possible
year with available information – 2014.
17. In these three boxes historical residual operating income is shown
Current VBA program was created for more comfortable valuation process. Code in this
program is based on PVAR estimation coefficients received during research process. This
program is open for further adjustments and code improvement.
48
Conclusion
Studying of Russian oil industry showed the importance of developing and improvement
of existing valuation models. Having almost all existing valuation models universal, that fact
makes them inefficient for working with features of different industries. In the first chapter, all
main principles of value-based management were studied, as a basement for further research.
While performing current masterwork, main valuation models were studied, divided into
groups and their main features are discussed. Then, residual operating income was highlighted
from other valuation models and taken for further improvement. Within the model, focus was
done on only one variable, which needed to be forecasted – operating income.
For further improvement and possible adjustments, significant part of literature review
was devoted to simultaneous equations model. Studying of articles written in this field showed a
possibility of implementation of simultaneous equation model into residual operating income
model for improvement purposes.
In a developed model three main exogenous variables were taken – sales, operating
expenditures and net operating assets. Model was checked on the set of eight biggest Russian oil
companies by 3SLS estimation. Estimation results showed an possibility of using developed
model. Then, created model was extended by adding exogenous variables – crude oil price, its
volatility, oil production, oil reserves and two ratios – EV/EBITDA and EBITDA/Barrel. G2SLS
model estimation indicated significance of only oil price, production and reserves coefficients.
Finally, forecasting model was created, based on previously received estimation results.
This last step was performed by PVAR estimation tool.
Received forecasting model has almost all necessary information to work with valuation
of Russian oil companies. One important note – current mode needed to be used with align of
crude oil price forecasting – as it is a very volatile variable, which could not be directly
implemented into developed model.
For more strong managerial implication and for more comfortable way of using
developed model, there was a decision of creating a VBA program. This tool could be used for
valuation purposes, with a set of minimum data which needed to be inputted.
There is a possible ways of improvement of created model and application. However,
current masterwork created a basement of using simultaneous equation model within residual
operating income model to forecasting the value of Russian oil companies.
49
References
1. Abrigo, M., Love, I. 2015. “Estimation of Panel Vector Autoregression in Stata: a
Package of Programs”. Discussion Paper. University of Hawai`i at Manoa, 45 p.
2. Angrist, J. D., Imbens, J. W. 1995. “Two-Stage Least Squares Estimation of Average
Causal Effects in Models With Variable Treatment Intensit”. Journal of the American
Statistical Association 90: pp. 434-446.
3. Balestra, P., Varadharajan-Krishnakumar, J. 1987. “Full information estimations of a
system of simultaneous equations with error component structure”. Econometric Theory
3: pp. 223–246.
4. Baltagi, B. H. 2013. Econometric Analysis of Panel Data. Chichester, UK: Wiley, 467 p.
5. Beidas-Strom, S., Pescatori, A. 2014. "Oil Price Volatility And The Role Of Speculation".
International Monetary Fund 14: pp. 1-28.
6. Bergmann, I. 2014. “Accounting based valuation and ‘other information’ – A
simultaneous equations approach”. Discussion Paper, 2014 International Accounting
Section Mid-Year Meeting and Doctoral/New Faculty Consortium, 56 p.
7. Bernard, V. I. 1995. “The Feltham-Ohlson Framework: Implications for Empiricists”.
Contemporary Accounting Research 11(2): pp. 733–747.
8. Bollen, K. A. 1989. Structural Equations with Latent Variables. New York: Wiley, 528 p.
9. Bukhvalov, A. V. 2005. “Corporate governance as an object for scientific researches”
[“Korporativnoe upravlenie kak ob"ekt nauchnyh issledovanij”]. Russian Management
Journal 3(3), pp 81-96.
10. Callen, J. L., Segal, D. 2005. “Empirical Tests of the Feltham-Ohlson (1995) Model”.
Review of Accounting Studies 10(4): pp. 409–429.
11. Copeland, T. E., Koller, T., Murrin, J. 1995. Valuation: measuring and managing the
value of companies. New York: Wiley, 558 p.
12. Damodaran, A. 2002. Investment Valuation. New York: Wiley, 921 p.
13. Davidson, R. Mackinnon, J. D. 1993. “Estimation and inference in econometrics”. New
York: Oxford University. 871 p.
14. Destri, M. L., Mina, A., Picone, P. M. 2012 Corporate Social Performance: Paradoxes,
Pitfalls and Pathways To The Better World, Charlotte, NC, USA: Information Age
Publishing, 657 p.
15. Devold, H. 2013. Oil and gas production handbook: an introduction to oil and gas
production, transport, refining and petrochemical industry. Oslo: ABB, 921 p.
16. Elliot, J. W., Uphoff, H. L. 1972. “Predicting the near term profit and loss statement with
an econometric model: a feasibility study”. Journal of Accounting Research 13: pp 14991526.
17. Fama, E. F., French. K. R. 2004. “The Capital Asset Pricing Model: Theory and
Evidence”. Journal of Economic Perspectives 18(3): pp 25–46.
18. Greene, W. H. 2012. Econometric Analysis. New York: Pearson, 1241 p.
50
19. Gurvich, E. T. 2004. “Macroeconomic valuation of the role of Russian oil industry”
[“Makroehkonomicheskaya ocenka roli rossijskogo neftegazovogo sektora”]. Questions
of economics 10: pp. 1-36.
20. Holtz-Eakin, D., Newey W., Rosen H. S 1988. “Estimating vector autoregressions with
panel data”. Econometrica, 56(6): pp. 1371-1395.
21. Hausman, J. A. (November 1978). "Specification Tests in Econometrics". Econometrica
46(6): pp. 1251–1271.
22. Jensen M. 2001. “Value Maximisation, Stakeholder Theory, and the Corporate Objective
Function”. European Financial Management 7(3): pp. 297-317.
23. Koller, T., Goedhart, M., Wessels, D. 2010. Valuation: Measuring and Managing the
Value of Companies. New Jersey: John Wiley & Sons, 840 p.
24. Kudrin, A. 2014. "The Influence of Oil and Gas Revenues on Russia’s Monetary and
Credit Policy". Working Paper, Gaidar Institute for Economic Policy, 27 p.
25. Lambert M., Hübner G. 2014 “Size matters, book-‐to-market does not! The F&F
empirical CAPM revisited”. 12th Paris Finance Meeting, 23 p.
26. Love, I., Zicchino, L. 2006. “Financial development and dynamic investment behavior:
Evidence from panel VAR”. The Quarterly Review of Economics and Finance, 46(2): pp.
190-210.
27. McFadden, D. 1999. Chapter 6. Simultaneous equations. Resource document. Lecture
notes for Economics, Berkeley, 39 p.
28. Nissim, D., Penman, S. 2001. “Ratio Analysis and Equity Valuation: From Research to
Practice”. Review of Accounting Studies 6 (1): pp. 109-154.
29. Ohlson, J.A. 1995. “Earnings, Book Values, and Dividends in Equity Valuation”.
Contemporary Accounting Research 11(2): pp. 661-687.
30. Ohlson, J.A. 2001. “Earnings, Book Values, and Dividends in Equity Valuation: An
Empirical Perspective”. Contemporary Accounting Research 18(1): pp. 107-120.
31. Osmundsen, P., Asche, P., Misund, B., Mohn, K. 2006. “Valuation of International Oil
Companies”. The Energy Journal 27(3): pp. 49-64.
32. Penman, S. H. 2010. Financial Statement Analysis and Security Valuation. New York:
McGraw-Hill/Irwin, 768 p.
33. Penman, S.H., Sougiannis, T. 1998. “A Comparison of Dividend, Cash Flow, and
Earnings Approaches to Equity Valuation”. Contemporary Accounting Research 15(3):
pp. 343- 383.
34. Reiss, P. C., Wolak, F. A. 2007. “Structural econometric modeling: rationales and
examples from industrial organization”. Handbook of Econometrics 6: pp. 4280-4317
35. Richardson, S., Tuna, I., Wysocki, P. 2010. “Accounting Anomalies and Fundamental
Analysis: A Review of Recent Research Advances”. Working Paper. New York
Conference, 37 p.
36. Rutgauzer, V. M. 2007. Business Valuation [Ocenka stoimosti biznesa]. Moscow:
Maroseyka, 448 p.
37. Sims, C.A. 1980. “Macroeconomics and reality”. Econometrica, 48(1), pp. 1-48.
51
38. Soliman, M. T. 2008. “The Use of DuPont Analysis by Market Participants”. The
Accounting Review 83 (3): pp. 823-853.
39. Volkov, D. L. 2008. “Theory of value-based management: financial and accounting
aspects” [“Teoriya cennostno-orientirovannogo menedzhmenta: finansovyj i
buhgalterskij aspekty”]. Saint Petersburg: GSOM, 320 p.
40. Volkov, D. L., Rudenko, A. 2009. “Value-based approach in managing of international
high-tech companies” Corporate Finance 2(10): pp. 5-27.
41. Wallace, J. S. 2003. “Value maximization and stakeholder theory: compatible or not?”
Journal of Applied Corporate Finance 15(3): pp. 120–127.
42. Wright, C. J., Gallun, R. A. 2008. Fundamentals of Oil & Gas Accounting. New York:
PenWell Books, 750 p.
52
Отзывы:
Авторизуйтесь, чтобы оставить отзыв