Croatian Operational Research Review 409

CRORR 8(2017), 409–427

The art of company financial modelling

Zoran Lukić1,†
1
Investment Banking, Zagrebačka banka, Savska 60, 10000 Zagreb, Croatia
E-mail: 〈[email protected] 〉

Abstract. In corporate finance, the term financial modelling denotes a widely used techni-
que of comprehensive customised quantification of a company’s entire operations. Even
though not mathematically strict, such models exhibit descriptive, explanatory and
predictive qualities. The paper elaborates on the main steps and principles for building
financial models of companies. It also identifies required assumptions and certain statistical
properties of well-constructed models. Furthermore, it describes the use of such models for
decision support purposes, supplemented by an illustrative example. Finally, it discusses
general characteristics and concerns associated with appropriate model construction and
use.

Keywords: financial modelling, corporate finance, decision support, business plan,

fundamental analysis

Received: September 30, 2016; accepted: July 18, 2017; available online: November 30,
2017

DOI: 10.17535/crorr.2017.0026

1. Introduction

The title resembles a stereotyped expression describing financial modelling of a

company as “more of an art than a science”. This perceived duality is further sup-
ported by a review of the literature. On the one hand, a large number of practical
tutorials are available, but limited to the constructional aspect of such modelling.
On the other hand, extensive academic research is restricted to financial forecasts
resulting from such models, and provided by publicly available equity analyst
reports. An integrated approach, connecting both the modelling process and out-
put analysis is the motivation behind this paper.
This paper presents a broad corporate use of company financial models, as evident
from previous research, supported by a survey conducted among several large
Croatian companies. It continues with a description of the constructional concept
of a model, including model assumptions and inherent statistical properties of
standard model structures. Furthermore, the purpose of establishing models and

†
Corresponding author

http://www.hdoi.hr/crorr-journal ©2017 Croatian Operational Research Society

410 Zoran Lukić

its applications are also presented using the example of a company modelled for
valuation purposes. Finally, a discussion on characteristics supports the adequacy
of such models for the purposes presented in this paper, subject to the reasonable
and responsible approach of both authors and users of the models.

2. Financial modelling in literature and practice

In general, financial modelling involves replicating the characteristics, behaviour

and financial implications of a single financial asset or portfolio of assets, ranging
from loan and lease agreements to various securities and derivatives. The classical
models, such as Markowitz portfolio selection and Black-Scholes option pricing,
are part of standard academic syllabi (e.g. [2]). Given that “time and uncertainty
are the central elements that influence financial behaviour” [32], a considerable
amount of literature is devoted to the modelling of financial assets as time series,
using stochastic mathematical apparatuses (e.g. [34], [52]), as well as neural
network architectures (e.g. [54]).
As opposed to various other asset classes, equity securities by definition exclude
pre-defined cash flows to holders, suitable for modelling. Instead, the aim of equity
modelling is assessment of security value, using the two different main approaches:
technical analysis, for listed securities, and fundamental analysis, applicable also
to non-listed ownership stakes. Technical theory assumes that “market value is
determined solely by the interaction of supply and demand” [28] and focuses,
therefore, exclusively on market data time series (prices and volumes of traded
securities). In contrast, fundamental theory “believes that each security has an
intrinsic value which depends upon its earning potential” [28], and is primarily
interested in the underlying company and economic factors that influence its
financial results.
Conceptualised in the 1930s [49], the fundamental analysis was initially restricted
to a company’s past and present [23], but has subsequently been extended to
projections of future business performance [51]. The principal tool and prerequisite
for this type of analysis is decomposition of a business into factors, replication of
its past and forecast of the future in the form of a quantified model. In addition
to investment purposes, the same modelling principles are applied for internal
corporate business planning, with particularly rapid development alongside
computer technology in the 1970s and 1980s [46], [35]. Simultaneously, it attracted
significant attention from operational researchers (e.g. [37], [46], [27], [38], [36]
and [20]). Consequently, a practical technique emerged, specialised for designing
and using company financial models, interchangeably also referred to as corporate
planning models (business plans). To distinguish it from other types of financial
models, the term Company Financial Model (CFM) is used further in the text.
 The art of company financial modelling 411

Literature on equity investment analyses differentiates between sell-side and buy-

side research (e.g. [15], [25] and [24]). The former is performed by investment
banks and is generally available to the public, whereas the latter is private and
confidential, produced by investors themselves, or their advisors. Numerous
research papers on sell-side analysis have investigated its accuracy (e.g. [7], [22]
and [5]) and impact on financial markets (e.g. [9], [29] and [30]). While explicitly
or implicitly recognised that CFM’s are the underlying tools for both kind of
investment analyses, recent papers have primarily addressed the outputs of such
models, i.e. published financial projections of companies. Except for a few cases
(e.g. [53]), the interest of contemporary research in inherent design, properties
and characteristics of CFM’s has decreased.
Disproportionally to recent academic research, there is an abundance of practical
tutorials and training material (e.g. [6], [45], [50], [4], [14] and [3]), providing
detailed guides and descriptions of steps for constructing and using CFMs.
Moreover, the international financial modelling competition ModelOff was initi-
ated in 2012, with several thousand participants getting involved annually.
To support the clear importance and penetration of CFM in corporations, a survey
was conducted in September 2016 among 5 large Croatian companies: Adris
Grupa, Atlantic Grupa, Hrvatski Telekom, Podravka and Zagrebačka banka. The
participants were asked (i) whether they use financial modelling (for analysis,
planning, valuation, financial management or other decision support purposes),
and (ii) if yes, where did the modellers learn the required skills. For the survey,
financial modelling was described as “the quantification of business operations
(using spreadsheet programme) and aimed to replicate historical results and
project the future business of the observed company”, which is the proposed
working definition of CFM. As expected, all participating companies confirmed
the use of CFM, as described above. The frequency of the indicated sources of
skills is summarised in the following table:

Source of skills #
At the current or previous employer (colleagues or internal training) 5
External training, course or professional specialisation (e.g. CFA) 3
Education at domestic universities 2
Education at foreign universities 1
Table 1: Sources of skills required for financial modelling (multiple answers possible); a
survey among 5 large Croatian companies, September 2016

The survey results are consistent with the observed and previous findings [37]
regarding the general acceptance of CFMs among corporations. The sources are
consistent with the noticed practical, rather academic inclination.
412 Zoran Lukić

As a final remark, the author of this paper has had the privilege during his advi-
sory career to build a number of CFMs for international institutional clients, and
use them for various analytical and decision supporting purposes.

3. Construction of a model

A CFM is comprehensive quantification (numerical representation) of a certain

business activity. It is designed to specifically represent a specific business entity,
not a generalisation or abstraction applicable for various entities. Its form is a
table (matrix) with columns representing time periods (usually fiscal years), and
the rows containing different line items. The historical part shows results of past
business events (actuals), while the future part displays a possible numerical
realisation of future business events (forecasts, projections).
According to [14], “financial modelling is the construction and use of a spreadsheet
depiction of a company’s or an individual’s past, present, or future business
operations”. A spreadsheet programme is the primary tools for constructing a
CFM. The appearance and development of spreadsheets has catalysed modern
financial modelling [4], enabling sequences of computational links between cells,
as well as transparently displaying numerical inputs and outputs in real time.
All business activities are represented by financial statements (balance sheet,
income statement and cash flow statement). Financial statements are, therefore,
the first element of a CFM. Furthermore, financial statements are monetary
records of underlying business operations, quantified by various operational vari-
ables and indicators. These operational items are the second element of a CFM.
Each financial line item has its corresponding operational factors, just as each
event in an entity’s real business life has financial consequences. The natural goal
is, therefore, to establish functional links between operational and financial
elements, which is the key constructive superstructure of CFM that transcends
from a mere list of data into an organised logical system.
The first type of functional link is at a purely financial level and follows accounting
rules. The second type is factorisation of financial items into quantities and prices,
and subsequently, establishment of relations using macroeconomic or market
indicators. Factorisation is crucial for all significant financial items, while less
important items can be approximated or aggregated. The amount and depth of
knowledge about a company directly impacts a modeller’s ability to identify the
importance relevant items and construct appropriate relations. An absence of
these links significantly deteriorates the quality of the model.
The third type of link aims to establish functional relations between various ope-
rational variables. These relations are complex, controversial and generally
unknown. For example, it is unclear a priori as to how exactly an incremental
unit of advertising space influences aggregate demand, market share, product
pricing and, ultimately, sales revenues. Furthermore, these relations largely
 The art of company financial modelling 413

depend on the specific sector, market and individual company, making them
difficult to determine or possibly generalise. To establish reasonable functional
relations, modellers use market studies, statistical analyses, approximations and
heuristics, or rely on sector expertise and knowledge of inherent business logic.
Construction of a CFM begins with input of historical financials, and ends with
projections of future financial statements, as stated in [3]: “a fully integrated
model derives and projects the three main financial statements of a business”. In
the historical part, financial inputs are combined with various observable operati-
onal indicators to extract and define a limited set of main model drivers. An ideal
CFM has only two kinds of drivers: (i) macroeconomic and market indicators
(exogenous); and (ii) operational variables subject to a sole strategic (business,
managerial) choice of the company (endogenous) [53]. The number of main drivers
is significantly reduced by establishing the above-mentioned third type of links.
In the future part of a CFM, construction continues in the opposite direction:
only inputted forecasts are the main drivers. Projections of all other operational
variables, quantities and prices depend on the driver forecasts, while the projected
financial statements are the consequence of the former. A typical CFM
construction plan is summarised in the following table:

Step Contents Information source

1 Model dimensioning
2 Input of historical financials and type 1 links Financial audit
3 Identification of operational factors and type 2 Due diligence
links
4 Identification of drivers and type 3 links Sector and market
expertise
5 Construction of deterministic financial routines Due diligence
6 Forecast of drivers Sector and market
expertise
7 Derivation of operational projections from Due diligence
drivers
8 Derivation of financial projections from Financial audit
operational
9 Model audit
Table 2: Plan for constructing a CFM

The plan starts with preparatory step 1, and ends with controlling step 9. As
described earlier, steps related to the past (2, 3, 4) extract information for the
historical part, while steps related to the future (6, 7, 8) construct projections for
the future part of a CPM. In between these steps, the aim of step 5 is to build
various computational financial routines (e.g. interest income and expense, debt
414 Zoran Lukić

repayment, dividend policy, depreciation and amortisation, tax levy and loss carry
forward, potential revolving debt drawdown, etc.) which are pre-defined contrac-
tually, by legal and regulatory framework, or accounting rules.
To accomplish each step, the table indicates the most common information source.
Regarding the financial level (steps 2, 8), in principle it is sufficient to analyse the
standard financial audit report with its accompanying notes. Assessment of the
operational level (steps 3, 7) requires having access to confidential business
information. This information is collected and verified through commercial,
technical and legal due diligence of the company (some portion of operational
information may be available in periodic management reports). Additionally, the
information for step 5 is obtained through financial and tax due diligence.
Finally, the assumptions level (steps 4, 6) requires sector and market expertise.
This means a deep knowledge and understanding of technological processes and
the business model used by the company for transforming available resources and
market circumstances into cash flows. As mentioned earlier, the tasks imple-
mented at this level (identification of drivers, forecasts of drivers and establish-
ment of type 3 links) are controversial and subject to a critique due to the lack
of sufficient scientific grounds. For this reason, a competent sector and market
expert is crucial for the implementation, not for acting as an undisputable
authority, but to make reasonable assumptions where scientific evidence is not
available.

3.1. Model assumptions

CFM assumptions consist of two subgroups: forecasts of the driving variables; and
the relationships (“type 3 links”) among them. Both are non-exact, hence
controversial elements. The output of sell-side equity analysts’ models is being
continuously researched as to the accuracy of earnings forecasts, including tests
whether the forecasts support rational expectations, adaptive expectations or
implicit expectations hypothesis (e.g. [40], [1]). Many of them have detected
systematic forecasting errors, thus rejecting the rationality hypothesis (e.g. [41],
[8]). Several different ways of forecasting are listed in [16]: informal, expert
judgement, extrapolation, leading indicators, surveys, time series and econometric
systems of equations. In practice, the prevailing forecasting method is best
described as an extended expert judgment; an opportune use of all, depending on
the circumstances. For example, if an expert judges that a strong historical trend
of a certain driving variable has a prevailing impact on the future, then the ex
post tests would most likely indicate supporting evidence for the adaptive expecta-
tions hypothesis. It is not that the modeller has ex ante selected extrapolation as
the forecasting method. Restricting the universe of available options prior to
examination of each concrete modelling case is disadvantageous. Similar to
 The art of company financial modelling 415

forecasts, the judgment approach is used in constructing links between drivers:

the assumed relationships are generally heuristic, approximate and robust.
Despite these severe methodological inconsistencies, sell-side analysts have been
repeatedly reported as being superior in forecasting, compared to other pre-
defined, theoretically well-founded approaches, such as univariate time series
models ([39], [11], [10] and [12]). The practice indicates that automated or
algorithmic forecasting methods, in absence of conclusive expert judgment, often
result in the predictive deficiency of a CFM. Similarly, the theoretically unknown
relation that has been despite constructed heuristically and controlled by the
expert, can improve the predictive properties of a CFM.

3.2. Statistical properties of the model

In addition to judgmental forecasts, the constructional mechanics of a CFM has

a significant impact on the statistical properties of projections, and is an area
requiring further research.
Modellers intuitively recognise the benefits of disaggregated financial forecasts:
the possible overstatement of one component is offset by the possible understate-
ment of the other; thus a forecast error is reduced. For example, rather than
forecasting total revenues, the CFM standard is to decompose and perform
modelling of different revenue streams separately. In case separate forecasts (or
forecast errors) form a sequence of independent stochastic variables (generally not
identically distributed), which are uniformly bounded, with the sum of variances
approaching infinity, then the Lindeberg condition is automatically satisfied and
the Lindeberg-Feller Central Limit Theorem ensures that the standardised sum
of such variables converges in distribution to a standard normal stochastic
variable [44].
However, the variables in CFM are typically not independent, but interconnected
crosswise. Nevertheless, the error variance is still reduced by disaggregation, if the
variables are bounded and if the corresponding standard deviations have certain
properties as the uniform distribution. Formally:

Proposition: Let the following bounded stochastic variables (generally not

independent and not identically distributed) have the following support intervals
for respective probability density functions:

X ... a , b  ; X i ... a i , bi  , i  1,..., n ;  

n n
a i 1
ai , b  b .
i 1 i

Let the corresponding standard deviations depend only on the length of the
support intervals, in the following form:

 X  f ( c ) , c  b  a ;  X  f (ci ) , ci  bi  ai , i  1,..., n .
i
416 Zoran Lukić

Assume that function f is superadditive where required. Then,

Var i 1 X i  Var X .
n

Proof:
Var i 1 X i  i 1 f ( ci )  2i  j Cov X i , X j
n 2 n
 
 i 1 f (ci ) 2  2i j f ( ci )  f (c j )
n

  n
i 1
f ( ci )    f   c 
2 n
i 1 i
2
 Var X . Q.E.D.

Uniformly distributed, as do many other commonly used variables, comply with

the above requirements, which are not overly restrictive. For example, all convex
functions, non-positive at zero, are superadditive on the non-negative domain.
Another variance reduction constructional property of CFMs is due to relati-
onships between the driving variables (“type 3 links”). The CFM standard is to
link different revenue streams with corresponding costs, to move in the same
direction, thus avoiding “the free lunch” misconception. On a profit level, revenues
and costs are subtracted. If revenues and costs are stochastic and positively
correlated, the profit variance is decreased due to an altered covariance sign. This
property is similar to antithetic variates method for variance reduction, as
described in literature on Monte Carlo simulations (e.g. [43]).

4. Uses of the model

The historical part of a CFM is used for analysing past business performance, by
controlling departments in advanced corporate organisations as well as external
analysts. The CFM’s quantifying completeness and conjunction of financial and
operational factors provide a powerful managerial analytical base. It helps to
identify and explain various hidden phenomena that influenced past results,
ranging from trivial (e.g. accounting inconsistencies) to strategically important
(e.g. changes in underlying key business drivers). Such analysis is essential in
identifying the starting point and critical areas when designing a company’s
strategy.
The prerequisite for use of the future part is to input initial forecasted values of
drivers, consisting of an estimated future macro/market scenario and a defined
company strategy in the context of macro/market assumptions. The model then
outputs future operational and financial implications of the chosen strategy. In
subsequent iterations, the macro/market scenario can be altered and the strategy
adjusted accordingly to optimise business performance projections. This optimi-
 The art of company financial modelling 417

zation procedure provides valuable support in the decision-making process for

defining mid- to long-term strategic goals, as well as feasible trajectories of all
components in a consistent manner.
An important purpose of a CFM is, therefore, selection, optimisation and
evaluation of company strategy. This kind of quantified strategy is, on the other
hand, an input for the model used for various other decision-supporting exercises.
For example, CFM output can be further used for the following purposes: budge-
ting, business planning, investment decisions, return analysis, valuation, credit
analysis, identification of synergies, integration and re-organisation analyses,
profitability management, capital structure adjustments, identification of fina-
ncing requirements, risk management, stress testing, financial management inclu-
ding solvency and liquidity, improvement of business processes, etc.
The use of the model starts with a fixed set of assumptions representing the most
probable future, the base case. To assess the uncertainty of projections, the
modelling tutorials (e.g. [4], [14], [45] and [50]) suggest a variation of the assumed
inputs and observation of changes in model outputs. These experiments are refer-
red to as sensitivity analysis, scenario analysis and simulation modelling [17].
Sensitivity analysis monitors changes in a model output resulting from the
variation of a single input variable (i.e. its time series), while retaining others as
unchanged. Scenario analysis applies several pre-defined input scenarios, with
different input values for all variables. In addition to the base (the most probable)
case, scenario analysis is regularly used to determine two other possible, yet
extreme, working cases – the worst (pessimistic, downside) and the best (opti-
mistic, upside).
Thus far, the three examined cases (worst, base, best) provide a slightly clearer
view of future uncertainty, by indicating the range of possible outcomes and the
mode. However, other statistical properties of future projections remain
unexplained. Simulation modelling is, therefore, a natural extension of the ana-
lysis, whereby each input is modelled as a stochastic variable, by using a
spreadsheet pseudorandom number generator. This approach requires additional
assumptions regarding probability distributions of inputs, and appropriate
transformation of generated pseudorandom numbers. In practice, however, the
simulation can be simplified by assuming uniform distributions for all inputs.
(This practice has been discussed in the previous section.) For this purpose, each
input requires only the estimation of a realistic and possible range of outcomes
(as in scenario analysis), provided by experienced company management or an
educated sector/market expert. After running the simulation, generated outputs
are organised into empirical distributions, suitable for applying standard
statistical tests of various hypotheses and drawing probabilistic conclusions about
business projections.
Any conclusions arising from the aforementioned experiments refer to the
constructed model, not a real-world company. On the other hand, experimenting
418 Zoran Lukić

with an actual business is impossible. Therefore, if the model realistically

replicates a company, the described techniques provide consistent analytical tools,
to be used with an awareness of all the embedded advantages and limitations.

4.1. An illustrative example

This example demonstrates the benefits of modelling for company valuation, in

the specific case of Slovak Telekom, the national integrated telecommunications
operator in Slovakia. In May 2015, Deutsche Telekom announced the acquisition
of the remaining 49% stake in Slovak Telekom from the Government of Slovakia
for EUR 900 million, thus becoming the sole shareholder. The target company
has not been listed, implying the non-existence of a reference market price, as well
as lack of research incorporating financial projections and valuation. While both
transaction parties had access to inside information necessary for modelling and
valuation, this example relies only on publicly available information (e.g. the
company’s annual reports).
The standard method for valuing a minority stake is relative valuation using
market multiples of comparable companies. A natural peer group here consists of
listed incumbent operators in new EU members in neighbouring CEE. For valu-
ation purposes, the financial indicators of Slovak Telekom are multiplied by the
appropriate median market multiples from the following table:

As of 31.03.2015 2012A 2013A 2014A 2015F 2016F 2017F

EV/Sales 1.3x 1.4x 1.5x 1.5x 1.5x 1.5x
EV/EBITDA 3.5x 4.1x 4.3x 4.5x 4.6x 4.6x
P/E 9.6x 18.3x 15.0x 17.0x 14.2x 13.1x
Table 3: Median market multiples. The peer group consists of Magyar Telekom, O2 CZ
(former Český Telecom) and Orange Polska (former Telekomunikacja Polska). For each
peer, the multiples are computed as of 31 March 2015 (historical multiples are based on
actual financials, while the forward-looking multiples are based on Bloomberg consensus
estimates). The table contains the median values of computed peer multiples.

Although close, the median valuation range based solely on historical financials
does not encompass the transaction value of EUR 900 million:

EUR mn 2012A 2013A 2014A

Sales 827 809 768
EBITDA 347 306 264
Net income 63 49 44
Net cash 676
Table 4: Slovak Telekom historical financial performance.
 The art of company financial modelling 419

EUR mn Min. Med. Max.

EV/Sales 862 879 887
EV/EBITDA 888 919 947
P/E 296 321 442
Median 862 879 887
Table 5: Slovak Telekom 49% equity value, based on market multiples applied on
historical financials.

For the application of forward-looking multiples, the key financials are projected
using the constructed company model. Without insight into operational details,
the model is relatively simplistic. Revenue items are factorised into subscribers
(quantities) and average revenue per access (prices), with subscribers being furt-
her linked to sector and market indicators (market share, service penetration,
population). Expenditure items are modelled by margins only, except for
regulatory costs. The exogenous drivers are benchmarked versus the peer markets,
while the endogenous are forecasted consistently with sector standards and the
company’s strategy, broadly indicated by the management in the annual reports.
As the “forward-looking multiples are more accurate predictors of value than
historical multiples” [31], the median valuation range based on financial projecti-
ons (including the projected net cash at 1Q 2015) expectedly contains the actual
transaction value:

EUR mn Min. Med. Max.

EV/Sales 895 905 906
EV/EBITDA 963 977 978
P/E 523 558 670
Median 895 905 906
Table 6: Slovak Telekom’s 49% equity value, based on forward market multiples applied
on projected 2015 – 2017 financials and Net debt (cash) as of 1Q 2015.

Both median valuation ranges (using actuals and projections) are relatively nar-
row. They do not reflect the uncertainty of assumptions, but only the different
growth/profitability characteristics of target, compared to peers. Had the growth
(profitability) been equal, the valuations would have been constant across the
years (multiples).
The previous forward-looking valuation is based on fixed assumed future values
of model variables (hereafter the “base” case). To assess future uncertainty, the
scenario analysis observes two additional scenarios, the “low” and the “high” case.
For that purpose, a total of 12 model variables are used as input drivers, and
assigned additionally assumed low and high future values respectively.
420 Zoran Lukić

For the simulation analysis, the above mentioned 12 input drivers are modelled
as random variables, each uniformly distributed on the previously defined low-
high respective intervals. The simulation is implemented in Microsoft Excel, using
the built-in pseudorandom number generator, and generated 10,000 times.
In addition to median, the scenario and simulation analysis is performed also for
the weighted average, both taken from 3x3 forward multiples valuation results.
The marginal weights for future years (2015 – 2017) as well as used multiples
(sales, EBITDA and earnings) are selected according to assumed relative
importance: (1/2, 1/3, 1/6) for years, and (1/3, 1/2, 1/6) for multiples,
respectively.
As expected, the simulated median valuation is less dispersed than the weighted
average. Despite inputs being uniformly distributed over identical ranges as for
scenarios, the simulated Bayesian central 95% credible intervals (as defined in [26]
and [21]) are only ±4.0% and ±7.9% wide. For comparison, the ranges between
the low and high case scenarios (equal to distribution support) are respectively
3.2 and 2.1 times wider.
The output results of the scenario and simulation analysis are summarised below:

Scenario analysis: Median Wgt. Avg.

Low case 800 752
Base case 905 889
High case 1,028 1,050

Simulation model: Median Wgt. Avg.

Mean 903 889
Coeff. of variation 2.1% 4.3%
95% credible int.: ±4.0% ±7.9%
Minimum 867 818
Maximum 939 959
Table 7: Scenario analysis and simulation modelling output, for median and weighted
average valuation based on forward market multiples.
 The art of company financial modelling 421

Figure 1: Median valuation: simulated output distribution histogram and central 95%
credible interval.

Figure 2: Weighted average valuation: simulated output distribution histogram and

central 95% credible interval.

The conclusion of the illustrative example is that, given the model and input
distributions, the Slovak Telekom 49% equity value on 31 March 2015 lies within
EUR 867 – 939 million (median) or EUR 818 – 959 million (weighted average)
with 95% probability.

5. Discussion

Are CFMs correctly called models? According to [19], models are miniature repre-
sentations, imitations, descriptions that assist in visualisation. CFMs certainly are
simplified replications of real businesses. Furthermore, the purpose of conceptual
models is to describe, explain and predict reality [19]. Comprehensive quantifica-
tion of a business obviously is a descriptive rationalisation that decreases the risk
of inconsistency and incompleteness. Explanatory characteristics are achieved by
decomposition of monetary aggregates first into quantities and prices, then into
other underlying variables, with quantitative relationships among them. As
emphasised in [27], “the equations comprising such models form a kind of
knowledge base which can be used to generate explanations”. Finally, CFMs (or,
equivalently, corporate planning models) are by definition predictive models, with
422 Zoran Lukić

the objective to “provide (...) a forecast of what is likely to happen given the
occurrence of certain events over which the managers have little or no control”
[20]. Thus, the term model is justified.
Are CFMs mathematical models? Some authors are explicit that “a financial
(planning) model is a mathematical model describing the interrelationships among
financial variables of the firm” [48]. Moreover, “financial model is a system of
mathematical equations, logic and data that describes the relationships among
financial and operating variables”, that falls “into two types: simulation, better
known as ‘what if’ models, and optimization models” [47]. But, aren’t mathema-
tical models substantially more general and abstract constructs than CFMs? For
example, the mathematical model for coin toss is a binary uniform random
variable, while the mathematical model for vibration of an elastic rod is a partial
differential equation with boundary-initial conditions. Both are appli-cable to any
particular object, anytime. In contrast, each CFM is constructed for a specific
company, for a particular point in time. It does not exist separately from the
concrete object of application, and is not applicable to other objects. Naturally, a
CFM itself is a system of equations, and it uses mathematical models as sub-tools,
such as regression or time series analysis. Yet the primary purpose of CFMs is
numerical representation of a single object, not a functional representation of the
whole class. The reasons above indicate that CFMs do not qualify as mathematical
models.
Attempts to increase generality or applicability of CFM typically result in a
decrease of its explanatory and predictive qualities. It is the prevailing practice,
therefore, to construct customised CFMs that are specific to one company only.

5.1. Responsibility of modellers

As CFMs are not well defined, but company-specific and subject to judgmental
assumptions, the model structure and ultimately the output is at the discretion
of modellers. But, isn’t that unacceptably subjective and susceptible to manipu-
lation? Research has detected examples of sell-side analyst bias, both behavioural
and attributable to the conflicted interest (e.g. [42], [55] and [33]). However, in
the European Union and other advanced jurisdictions, the release of sell-side
equity research is an intensely regulated investment activity, subject to regulation
by financial services authorities. Moreover, each research report is undersigned by
its author, providing the name and contacts for possible inquiries. In addition to
legal requirements, the industry has developed important self-regulatory mecha-
nisms. One example is the continuous ranking of analysts according to forecasting
accuracy, but more importantly, the professional associations, such as the CFA
Institute, require members to adhere to ethical codes and standards of professional
conduct [13]. The impression of absolute freedom in modelling is illusory, in
 The art of company financial modelling 423

contrast with the actual continuous surveillance and strict professional

responsibility of the modellers.
The responsibility for the modelling outputs and the reasonable use of results is
on the modeller, not the model. The proper use of models is to support decisions,
not to substitute reasoning or to dictate conclusions. Motivated by the recent
financial crisis, the Financial Modellers’ Manifesto was published in 2009 by
prominent financial mathematicians [18]. Even though it is devoted to
algorithmic, well-defined models of financial securities, it contains the principles
that are also applicable to the construction and use of CFMs, summarised in The
Modellers’ Hippocratic Oath. It postulates the distinction between the financial
reality and its modelled representation; prohibits overconfidence in models and
promotion of such beliefs; forbids unreasonable model oversimplifications; and
calls for conscious conduct. In accordance with these principles, and given the
extensive use of CFMs as supporting tools for influential decisions, it is imperative
for CFMs to remain within a comprehensive structure, built for specific objects,
and based on educated judgement.

6. Conclusion

This paper presents the observed extensive use of CFMs in corporate finance,
equally for corporate business planning and fundamental investment analysis.
Such models are regarded by business professionals as important decision suppor-
ting tools due to exhibited descriptive, explanatory and predictive qualities.
The CFM output projections are to be considered as a possible hypothetical
future, consistent with other model assumptions, and suitable for scenario and
simulation analysis. The assumptions are predominantly based on expert judg-
ment, while the models are customised for a specific company. The observed
practice indicates that automatisation, algorithmisation or generalisation can be
significantly detrimental to CFM’s explanatory and predictive strengths.
Colloquially expressed, a good CFM is tailor-made, hand-made unicum; an
expensive craftwork. Thus, the emphasis is on the responsibility of modeller as
well as the awareness of the recipients, particularly under overconfident market
circumstances.
Recent academic research is mostly restricted to publicly available model outputs
by equity analysts. Even though expectedly weaker than private, due to the lack
of confidential inside information, published equity analysts’ forecasts are
repeatedly confirmed superior relative to predictions by other methods, such as
univariate time series models. This paper argues that the use of CFMs, as
described herein, contributes significantly to this advantage. It has been shown
that, by construction, CFMs possess important forecast error variance reduction
properties.
424 Zoran Lukić

The suggestion therefore for future research into equity analysts’ forecasts is to
incorporate analyses of underlying financial models. Increasing the understanding
of relations between the model characteristics and output properties importantly
requires analysing model forecasts in the context of expectations theory, as well
as studying and testing regular model designs for various statistical implications.
A limitation of this paper is identification and discussion of only two significant
CFM features: judgmental assumptions and variance reduction mechanics. The
recommendation is for further research to determine other properties, both
beneficial and non-beneficial for the quality of CFMs. Such research may
eventually lead to codification of modelling elements and procedures, in order to
improve scientific foundations of CFMs.

References

[1] Ackert, L. F. and Hunter, W. C. (1995). Rational expectations and security

analysts’ earnings forecasts. Financial Review, 30(3), 427–443.
[2] Aljinović, Z., Marasović, B. and Šego, B. (2011). Financijsko modeliranje (2nd
ed.). Split: Ekonomski fakultet Sveučilišta u Splitu.
[3] Ascoli, E. (2013). Modeling (4th ed.). London: Adkins, Matchett & Toy.
[4] Avon, J. (2013). The Handbook of Financial Modeling: A Practical Approach
to Creating and Implementing Valuation Projection Models. New York:
Apress.
[5] Barber, B., Lehavy, R., McNichols, M. and Trueman, B. (2001). Can investors
profit from the prophets? Security analyst recommendations and stock
returns. The Journal of Finance, 56(2), 531–563.
[6] Benninga, S. (2014). Financial Modeling (4th ed.). Cambridge, MA: MIT
Press.
[7] Bonini, S., Zanetti, L., Bianchini, R. and Salvi, A. (2010). Target price
accuracy in equity research. Journal of Business Finance & Accounting, 37(9-
10), 1177–1217.
[8] Bradshaw, M. T. (2011). Analysts’ forecasts: what do we know after decades
of work? Available at SSRN: http://ssrn.com/abstract=1880339 [Accessed
18/09/16].
[9] Brav, A. and Lehavy, R. (2003). An empirical analysis of analysts’ target
prices: Short-term informativeness and long-term dynamics. The Journal of
Finance, 58(5), 1933–1968.
[10] Brown, L. D., Hagerman, R. L., Griffin, P. A. and Zmijewski, M. E. (1987).
Security analyst superiority relative to univariate time-series models in
forecasting quarterly earnings. Journal of Accounting and Economics, 9(1),
61–87.
 The art of company financial modelling 425

[11] Brown, L. D. and Rozeff, M. S. (1978). The superiority of analyst forecasts as

measures of expectations: evidence from earnings. The Journal of Finance,
33(1), 1–16.
[12] Call, A. C., Chen, S. and Tong, Y. H. (2013). Are analysts’ cash flow forecasts
naïve extensions of their own earnings forecasts? Contemporary Accounting
Research, 30(2), 438–465.
[13] CFA Institute (2014). Standards of Practice Handbook (11th ed.). Charlot-
tesville, VA: CFA Institute.
[14] Charnes, J. M. (2012). Financial modeling with Crystal Ball and Excel (2nd
ed.). Hoboken, N.J.: John Wiley & Sons.
[15] Cheng, Y., Liu, M. H. and Qian, J. (2006). Buy-Side Analysts, Sell-Side
Analysts, and Investment Decisions of Money Managers. Journal of Financial
and Quantitative Analysis, Vol. 41 No. 1, 51–83.
[16] Clements, M. P. and Hendry, D. F. (2002). An Overview of Economic Fore-
casting. In Clements, M. P. and Hendry, D. F. (Eds.). A companion to
economic forecasting (pp. 1–18). Malden, MA: Blackwell Publishing.
[17] Damodaran, A. Probabilistic Approaches to Risk: Scenario Analysis, Decision
Trees and Simulations. Available at: http://people.stern.nyu.edu/adamodar
/pdfiles/papers/probabilistic.pdf [Accessed 18/09/16].
[18] Derman, E. and Wilmott, P. (2009). The financial modelers’ manifesto.
Available at SSRN: http://ssrn.com/abstract=1324878 [Accessed 24/09/16].
[19] Dym, C. L. (2004). Principles of Mathematical Modeling (2nd ed.). Amster-
dam; Boston: Elsevier Academic Press.
[20] Frazier, G. D. (1969). A Corporate Long Range Planning Model. Business
Economics Vol. 4, No. 4, 62–65.
[21] Garthwaite, P. H., Jolliffe, I. T. and Jones, B. (2002). Statistical Inference
(2nd ed.). New York: Oxford University Press.
[22] Gleason, C. A., Bruce Johnson, W. and Li, H. (2013). Valuation model use
and the price target performance of sell-side equity analysts. Contemporary
Accounting Research, 30(1), 80–115.
[23] Graham, B. and Dodd, D. L. (2008). Security Analysis: Principles and
Technique (6th ed.). New York: McGraw–Hill.
[24] Groysberg, B., Healy, P. and Chapman, C. (2008). Buy-side vs. sell-side
analysts’ earnings forecasts. Financial Analysts Journal, Vol. 64.4, 25–39.
[25] Hobbs, J. and Singh, V. (2015). A comparison of buy-side and sell-side ana-
lysts. Review of Financial Economics, Vol. 24, 42–51.
[26] Koop, G., Poirier, D. J. and Tobias, J. L. (2007). Bayesian Econometric
Methods (Econometric Exercises vol. 7). New York: Cambridge University
Press.
[27] Kosy, D. W. and Wise, B. P. (1984). Self-Explanatory Financial Planning
Models. In Proceedings of the 4th National Conference on Artificial Intelli-
426 Zoran Lukić

gence, AAAI–84 (pp. 176–181). Austin, TX: Association for the Advancement
of Artificial Intelligence.
[28] Levy, R. A. (1966). Conceptual Foundations of Technical Analysis. Financial
Analysts Journal, Vol. 22.4, 83–89.
[29] Li, W. and McDowell, E. A. (2011). Investor affect, investor status and the
influence of analyst reports. Journal of Finance and Accountancy, 8, 1.
[30] Loh, R. K. and Stulz, R. M. (2011). When are analyst recommendation
changes influential? Review of Financial Studies, 24(2), 593–627.
[31] McKinsey & Company, Koller, T., Goedhart, M. and Wessels, D. (2010).
Valuation: Measuring and Managing the Value of Companies (5th ed.).
Hoboken, N.J.: John Wiley & Sons.
[32] Merton, R. C. (1995). Influence of mathematical models in finance on practice:
past, present and future. In Howison, S. D., Kelly, F. P. and Wilmot, P.
(Eds.). Mathematical Models in Finance (pp. 1–14). London: Chapman & Hall
for The Royal Society.
[33] Michaely, R. and Womack, K. L. (1999). Conflict of interest and the
credibility of underwriter analyst recommendations. Review of financial studi-
es, 12(4), 653–686.
[34] Mills, T. C. and Markellos R. N. (2008). The Econometric Modelling of
Financial Time Series (3rd ed.). Cambridge: Cambridge University Press.
[35] Minch, R. P. (1989). Logic Programming as a Paradigm for Financial Mode-
ling. MIS Quarterly, Vol. 13 No. 1, 65–84.
[36] Naylor, T. H. and Gattis, D. R. (1976). Corporate planning models. California
Management Review, 18(4), 69–78.
[37] Naylor, T. H. and Schauland, H. (1976). A survey of users of corporate
planning models. Management Science, 22(9), 927–937.
[38] Newton, J. K. (1985). The Decision Support Role of Operational Research in
Corporate Planning. In Rand, G. K. and Eglese, R. W. (Eds.). Further Deve-
lopments in Operational Research. Oxford: Operational Research Society,
Pergamon Press.
[39] O’Brien, P. C. (1988). Analysts’ forecasts as earnings expectations. Journal
of accounting and Economics, 10(1), 53–83.
[40] Ramnath, S., Rock, S. and Shane, P. B. (2008). Financial Analysts’ Forecasts
and Stock Recommendations: A Review of the Research. Hanover, MA: Now
Publishers.
[41] Ramnath, S., Rock, S. and Shane, P. (2008a). The financial analyst
forecasting literature: A taxonomy with suggestions for further research.
International Journal of Forecasting, 24(1), 34–75.
[42] Richardson, S., Teoh, S. H. and Wysocki, P. D. (2004). The walk-down to
beatable analyst forecasts: The role of equity issuance and insider trading
incentives. Contemporary accounting research, 21(4), 885–924.
 The art of company financial modelling 427

[43] Robert, C. and Casella, G. (1999). Monte Carlo Statistical Methods. New
York: Springer Science & Business Media.
[44] Sarapa, N. (1992). Teorija vjerojatnosti (2nd ed.). Zagreb: Školska knjiga.
[45] Sengupta, C. (2004). Financial Modeling using Excel and VBA. Hoboken,
N.J.: John Wiley & Sons.
[46] Shim, J. K. and McGlade, R. (1984). The use of corporate planning models:
past, present and future. Journal of the Operational Research Society, 35(10),
885–893.
[47] Shim, J. K. and Siegel, J. G. (2007). Handbook of Financial Analysis,
Forecasting, and Modeling (3rd ed.). Chicago: CCH.
[48] Siegel, J. G. and Shim, J. K. (2000). Accounting Handbook (3rd ed.). New
York: Barron’s Educational Series.
[49] Stabile, D. R. (2005). Forerunners of Modern Financial Economics: A Random
Walk in the History of Economic Thought. Cheltenham: Edward Elgar
Publishing.
[50] Stein Fairhurst, D. (2012). Using Excel for Business Analysis: a Guide to
Financial Modelling Fundamentals. Singapore: John Wiley & Sons.
[51] Thomsett, M. C. (1998). Mastering Fundamental Analysis. Chicago:
Dearborn Financial Publishing.
[52] Vondraček, Z. (2008). Financijsko modeliranje – predavanja. Available at:
https://web.math.pmf.unizg.hr/~vondra/2013-14.html [Accessed 06/09/16].
[53] Wild, J. J. (2011). Measurement of a financial model of the firm: A field study.
Journal of Applied Business Research (JABR), 9(4), 83–92.
[54] Zekić, M. (1996). Neural networks for time-series predictions in finance and
investing. In Hunjak, T., Martić, Lj. and Neralić, L. (Eds.). Proceedings: 6th
International Conference on Operational Research. Zagreb: Croatian Operati-
onal Research Society.
[55] Zitzewitz, E. (2001). Measuring herding and exaggeration by equity analysts
and other opinion sellers. Available at SSRN: http://ssrn.com/abstract
=405441 [Accessed 24/09/16].