7th International Workshop on Quantitative Approaches to Software Quality (QuASoQ 2019)



       Prioritization of EA Debts Facilitating Portfolio
                            Theory
         Yoon Chow Yeong                                  Simon Hacks                                       Horst Lichter
    Universiti Teknologi Petronas Division of Network and Systems Engineering                 Research Group Software Construction
   Perak Darul Ridzuan, Malaysia        KTH Royal Institute of Technology                            RWTH Aachen University
    yeongyoonchow@gmail.com                    Stockholm, Sweden                                        Aachen, Germany
                                                  shacks@kth.se                                    lichter@swc.rwth-aachen.de



   Abstract—Implementing an enterprise architecture (EA)                 which describes the immature database design decisions [8],
 project might not always be a success due to uncertainty                the context of technical debt is still limited to the technological
 and unavailability of resources. Hitherto, we have proposed a           aspects.
 new metaphor –Enterprise Architecture Debt (EAD)–, which
 makes bad habits within EAs explicit. We anticipate that the               Over the years, technical debt becomes increasingly im-
 accumulation of EAD will negatively influence EA quality, also          portant when organizations invest huge amounts of money
 expose the business into risk.                                          in IT to stay competitive, effective, and efficient. However,
 Recognizing the importance of business-IT alignment in enter-           it is vital to align IT and business in order to realize the
 prise architecture context, this paper proposes an application of       full benefits and potentials of those IT investments [9]. From
 portfolio-based thinking and utility theory for EAD prioritization.
 For proof-of-concept purpose, we develop synthetic data using           there, the concept of Enterprise Architecture (EA) has evolved
 coarse-grained estimates to demonstrate the application of the          as a method to facilitate the alignment of IT systems and
 proposed portfolio-based approach which helps to determine the          business strategies within dynamic and complex organizations
 optimum selection of EAD to be resolved. The results show that          [10]. Consequently, the huge interest in EA resulted in vast
 our approach can help EA practitioners and management to                scientific contributions that address a broad thematic spectrum
 reason their EA investment decisions based on the EAD concept,
 with adjustable enterprises risk tolerance level.                       [11], including EA frameworks, EA management, and EA
   Index Terms—Enterprise Architecture Management, Enter-                tools. However, there is a lack of insight into the application of
 prise Architecture Debt (EAD), Portfolio Theory, EA Portfolio           the debt concept to include not only the technological aspects
 Optimization, Utility Theory                                            addressed by technical debt, but also the business aspects.
                                                                         Adapting the concept of technical debt in the EA domain,
                         I. I NTRODUCTION                                hitherto we have proposed a new metaphor “Enterprise Archi-
    Technical debt is a metaphor that had been introduced                tecture debt (EAD)” to provide a holistic view [12].
 by Cunningham [1]. In the software development industry,                   In the real world, debt is not necessarily a negative thing
 technical debt is regarded as a critical issue in terms of              to incur, same goes to EA debt to be held in an enterprise.
 the negative consequences such as increased software devel-             The danger of debt comes into place when there is no proper
 opment cost, low product quality, decreased maintainability,            debt management approach to prioritize, which debt should be
 and slowed progress to the long-term success of developing              repaid as soon as possible. We predict that managing EA debt
 software [2]. Technical debt describes the delayed technical            will be one of the critical success factors of EA implementa-
 development activities for getting short-term payoffs such as           tion and, thus, there is tremendous need to allocate resource
 a timely release of a specific software [3]. Seaman et al.              effectively to maintain the current level of profitability by
 [4] described technical debt as a situation in which software           properly managing EA debts that exist in an enterprise.
 developers accept compromises in one dimension to meet an                  Numerous studies have been dealing with the approaches to
 urgent demand in another dimension and eventually resulted              prioritize technical debt in the domain of software engineering
 in higher costs to restore the health of the system in future.          [3]–[5], [8], [13]–[16], and yet these studies do not address
    Furthermore, technical debt is explained as the effect of            the business aspects as a whole EA. To fill the research
 immature software artifacts, which requires extra effort on             gap, this study aims to extend the application of portfolio
 software maintenance in the future [5]. The concept of tech-            theory into the concept of EA debt. This will be achieved by
 nical debt reflects technical compromises that provide short-           focusing on the following research questions:
 term benefit by sacrificing the long-term health of a software
 system [6]. In view of the original idea of technical debt that         (RQ1) How can a given set of EA debt items be prioritized
 focused on the code level in software implementation, the               based on a portfolio approach?
 concept had been extended to software architecture, documen-
 tation, requirements, and testing [7]. While the technical debt            The following list of research sub-questions are emerged
 metaphor has further extended to include database design debt,          from the main research question which is mentioned as above:



Copyright © 2019 for this paper by its authors.                                                                                    45
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
                           7th International Workshop on Quantitative Approaches to Software Quality (QuASoQ 2019)


                                                                         advantage of diversification can be achieved through the
 (RQ1.1) What attributes of EA debt should be contained in a             portfolio return maximization for a given level of portfolio
 portfolio-based prioritization model?                                   risk, or the portfolio risk minimization for a given level of
                                                                         portfolio return.
 (RQ1.2) What are the process steps required to prioritize EA               The expected return of a portfolio is expressed by the
 debt items based on a portfolio thinking?                               following equation [17]:
                                                                                                             N
    This study proposes a portfolio-based approach to prioritize
                                                                                                             X
                                                                                                     E=            wi µi                       (1)
 EA debt that exists in EA implementation by incorporating the                                               i=1
 portfolio thinking and utility theory into EA. This proposed
                                                                         where E is the portfolio’s return, wi is the weight of asset i in
 approach contributes to the theoretical body of knowledge by
                                                                         the portfolio, the sum of all weights w has to be 1, and µi is
 providing a fundamental understanding on how EA debt items
                                                                         the expected return of asset i. On the other hand, the portfolio
 can be conceptualized and measured for decision-making. It
                                                                         variance of return is calculated as follows [17]:
 is strongly believed that this approach can measure, manage,
                                                                                                    N                   N
 and prioritize debt on an enterprise-wide level, which can be                                      X                   X
 valuable to EA stakeholders by avoiding massive interests on                             V (E) =         wi2 δi2 + 2         wi wj ρij        (2)
                                                                                                    i=1                 i<j
 EA debt. In light of the novel introduction of the EA debt
 metaphor, it is foreseen that EA debt decision-making would             where V (E) is the portfolio’s return variance, δi is the return
 be a worthwhile subject for future research in the EA field.            variance of individual asset i, and ρij is the covariance
    The rest of this paper is structured as follows: First, we           between the assets i and j. The portfolio’s standard deviation,
 introduce the facilitated key concepts of modern portfolio              δP can then be computed as follows:
 theory and utility theory. Second, we present in Section III                                          p
 how we apply the concepts of portfolio theory and utility                                       δP = V (E)                           (3)
 theory (Section III-A), and depict a process, which guides                 Often, MPT relies on historical variance of financial assets’
 the prioritization (Section III-B). Next, we demonstrate our            returns to measure the risk. Unfortunately, projects that involve
 approach on a fictitious case study in Section IV and present           non-financial assets commonly do not have well-defined his-
 related work (Section V). Last, we conclude our work in                 torical variance for absolute objective measurement [19]. How-
 Section VI.                                                             ever, Omisore et al. [19] asserted that this does not eliminate
                        II. K EY C ONCEPTS                               the possibility of applying MPT to non-financial assets because
                                                                         the concept is transferable to a wide range of investments as
 A. Modern Portfolio Theory
                                                                         long as the “risk” is expressed in terms of uncertainty about
    In the finance domain, Modern Portfolio Theory (MPT)                 expectations and possible losses on forecasts. Therefore, in EA
 was originally developed by Markowitz [17]. The goal of this            debt context, we express risk in term of “chance of interest
 theory is to develop an approach to determine an efficient              growth”, which brings the risk of an increased amount of
 portfolio with the maximum return at a given level of risk              required effort to resolve an EA debt item in future phase
 or the minimum risk at a given level of return. Based on                as well as the negative impacts on the EA value.
 this, decisions can be made of which types and amounts of
 financial assets in a portfolio should be invested or divested          B. Utility Function and Risk Aversion
 [17], [18]. The investments can be stocks, bonds, or other                 In general, most investors require a greater return as com-
 financial products that are characterized by a return at a certain      pensation for taking a greater risk [19]. Nevertheless, investors
 level of risk. The development of MPT was based on the rule             differ in their level of risk tolerance, which means that they
 that investors should consider expected or anticipated return as        are risk averse to varying degrees and eventually leads to
 a desirable thing, whereas variance of return as an undesirable         different utility functions [20]. The concept of utility function
 thing. In MPT, a portfolio is a weighted combination of assets          provides a way to select the optimal portfolio that yields
 in which each asset’s return and variance of the return are used        the best trade-offs of return and risk, and gives the most
 to measure the portfolio performance [17].                              satisfaction (utility) to the investors, taking their risk tolerance
    The fundamental concept behind MPT emphasizes the im-                level into consideration [21]. This can be applied in the
 portance of evaluating the relationship between price changes           enterprise context where each profit-seeking enterprise differs
 in each asset and price changes in every other asset in                 in the amount of risk it is willing to accept at a given level of
 the portfolio [19]. Each individual financial asset generates           return.
 different level of return and risk and, thus, the introduction of          To address the differences in enterprises’ risk tolerance
 MPT seeks to minimize the total variance of the investment              level, our approach proposes to prioritize EA debt portfolio
 portfolio’s return through the concept of diversification. The          for repayment based on the portfolio theory along with the
 diversification allows investors to combine different assets            principle of expected utility maximization. The utility maxi-
 whose returns are not perfectly positively correlated. By wisely        mization principle states that a rational investor acts to choose
 deciding on the proportions of various financial assets, the            an investment that maximizes the expected utility of wealth



Copyright © 2019 for this paper by its authors.                                                                                           46
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
                           7th International Workshop on Quantitative Approaches to Software Quality (QuASoQ 2019)


 among a set of feasible investment alternatives [20]. The                  With inadequate resources and other forms of constraints,
 similar concept can be applied in the context of EA debt in             there are situations where we need to compromise certain
 which an enterprise should act to invest in paying off the EA           principles, goals etc. in the first EA life cycle phase. Any
 debt portfolio that maximizes the expected utility of resources         omission of business and IT aspects inevitably leads to an
 among a set of existing EA debt portfolios.                             incomplete view of the EA concerned and may result in an
   Since risk aversion is not an objectively measurably quantity         EA debt. As described by TOGAF [24], the Architecture
 that aims for absolute measure, there is no unique utility              Development Cycle (ADM) is a method to develop an EA in
 function, which comes into place. Carlsson et al. [21] reported         a continuous and iterative manner. From there, we anticipate
 that one of the commonly employed utility function is:                  that the existing EA debt somehow needs to be repaid in
                                                                         the future EA life cycle phase and, thus, EA practitioners,
                  U (P ) = RP − 0.005 ∗ A ∗ δP2 ,                  (4)
                                                                         IT representatives, and management are accountable to make
 where RP is the expected return of a portfolio, A is an index of        decisions on which EA debt item needs to be repaid first in
 the investor’s risk aversion coefficient (a higher index indicates      order to avoid higher future cost.
 a higher level of risk averseness), and δP2 is the variance of             It is expected that the accumulation of EA debt in an EA
 the portfolio’s expected rate of return, which is the square            project will significantly affect the quality of an EA such as
 of standard deviation, δP , a measure of portfolio risk. The            its maintainability and agility in responding to the rapidly
 risk aversion coefficient is meant to be positive for all risk-         changing business environments. Oppositely, if the EA debt is
 averse investors whereas a negative index indicates a risk-             managed effectively, it is expected to increase the EA value.
 loving investor [18].                                                   In other words, EA debt is analogous to a financial asset that
    The factor of 0.005 in Equation (4) is a scaling convention          generates return at a certain amount of risk. We expect that
 and normalizing factor that allows us to express the RP and             EA debt prioritization helps to effectively pay off the EA debt
 δP2 as a percentage value instead of decimals. Adhering the             and, in turn, the EA can be adapted towards the new business
 positive affine transformation property of a utility function,          requirements.
 we are allowed to scale a utility function by translating it               In view of the similarity between financial investment and
 with the addition and/or subtraction of any constant [20]. To           incurring EA debt, we realize a potential of mapping the
 further scale down the size of variance for easy interpretation         concept of financial portfolio management to the EA debt
 in our study, we reduce the factor of 0.005 in Equation (4) to          context.
 0.001.
                                                                         A. How to Measure EA debts
    In this work, we apply the altered Equation (4) to plot risk-
 indifference curves (also known as utility curves) which allows            An organization, which intends to implement EA, consists
 us to select the attainable and optimum debt portfolio by com-          of numerous EA projects for enterprise transformation, e.g.
 bining the curves with the risk-return trade-off plots. Having          adding new business processes or retiring applications. This
 to say that the single point where one of the curves intersects         study reasons that EA debt items inevitably exist in each
 the efficient frontier is the debt portfolio that provides the best     project. EA debt items could be the failure of removing
 combination of risk-return for the risk level that is acceptable        outdated elements in diagrams, a missing implementation stan-
 for the organization.                                                   dard, undefined business role definitions, outdated technolog-
                                                                         ical structure, etc. EA debt prioritization is a decision-making
       III. A PPLYING M ODERN P ORTFOLIO T HEORY TO                      process that involves determining, which EA debt portfolio is
               E NTERPRISE A RCHITECTURE DEBT                            optimal to pay off in the current phase. By examining the EA
    In the field of Information Systems (IS), it is common               debt items across business and IT layers, an enterprise can gain
 to apply theories that originates from a diverse set of disci-          a better understanding of EA debt embedded along the process
 plines such as psychology, sociology, economics, finance, and           of EA implementation. Having EA debt properly assessed and
 computer science for problem-solving at the intersection of             being paid off in line with the long-term business mission and
 people, information technology, and organizations [22]. Based           goals, we can ensure that the resources are utilized efficiently.
 on the definition of EA debt presented by Hacks et al. [12],            In short, EA debt measurement urges a new way of thinking
 this work suggests the application of portfolio thinking into           of technical debt as an integrated part of the organization’s
 EA debt context with the aim of expanding the visibility and            business aspects.
 understanding of the newly introduced metaphor.                            In order to apply financial portfolio theory to EA debt,
    Technically, EA is responsible for translating the organiza-         we need to quantify EA debt for measurement. Therefore,
 tions strategy into projects that result in the achievement of a        we derive a set of operational definitions in line with the
 target state of the enterprise [10]. The gap between the target         financial definitions in portfolio theory that conform to the
 (to-be) architecture and the current (as-is) architecture is to         common assumptions of existing technical debt studies [5],
 be filled by identifying and implementing projects, programs,           [8]. Based on existing technical debt literature [5], [8], [14],
 or initiatives. However, the required resources to achieve the          [15], [25], each EA debt item has its associated principal
 goal of a project, program, or initiative are limited in terms of       estimate, interest estimate, and expected return. Despite the
 budget, time and performance specifications [23].                       unit used for technical debt measurement in dollars, hours,



Copyright © 2019 for this paper by its authors.                                                                                  47
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
                             7th International Workshop on Quantitative Approaches to Software Quality (QuASoQ 2019)


                                  TABLE I                                     is calculated using the following equation:
                     L IST OF MEASUREMENT ATTRIBUTES
                                                                                                                IA t
   Measurement        Operational definition
                                                                                                Rdt = (X(1 +       ) ) − X,                (5)
                                                                                                                X
   attribute
   Principal (X)      The number of hours required to resolve the EA debt     where Rdt is the individual EA debt item’s expected return at
                      in the current EA phase.                                time t, X is the principal, and IA is the interest amount of
   Interest amount    The extra hours that will be required in the future
   (IA)               phase if the EA debt is not resolved in the current
                                                                              the EA debt item.
                      phase.                                                     To fit in the MPT model, we need to determine the “weight”
   Interest           The likelihood that the interest amount will increase   of each EA debt item (wd ) and “correlations with other EA
   variance (δ)       at the point of repayment in future phase.              debt items” for each of the identified EA debt items. We
   Expected           Net benefit of resolving the EA debt in the current
   return (R)         phase and hold the debt to the future phase.            assume that an EA debt portfolioP  contains all EA debt items in
   Risk aversion      The degree of enterprise’s risk averseness.             equal proportions, in a way that, w∈W wdi = 1. On the other
   coefficient (A)                                                            hand, we adapt the idea of Guo and Seaman [5] to use cor-
                                                                              relation coefficients to represent the correlation between two
                                                                              debt items, where CORij expresses the correlation between
 people, or work units, these figures are easy to interpret and               di and dj . Since an EA debt portfolio is made up with multiple
 to handle, because they serve as a common language, allowing                 EA debt items across multiple architectural layers, determining
 trend monitoring as well as historical data comparison [16].                 the correlations between EA debts requires analysis of all
 This study opts to use working hours as measurement unit.                    EA entities embedded in EAM activities as well as interoper-
 Each measurement attribute is summarized in Table I.                         ability between architecture entities and architecture domains.
    Referring to Table I, we give an instance to provide an                   A reliable estimation of correlations could be done through
 insight into how each measurement unit can be represented.                   dependency analysis [5]. For simplicity, we consider that the
 Once an EA debt item is incurred in an EA project, a certain                 correlation coefficient would be either 1 (two debt items are
 amount of working time is required to resolve the EA debt,                   related to each other) or 0 (two debt items are unrelated to
 which is denoted by principal (X). For instance, due to an                   each other). With the value of correlation coefficients, a co-
 enterprise architect’s careless examination, one of the outdated             variance matrix can be created by computing:
 elements in a use case diagram of System A was not removed.
                                                                                                    ρij = δdi δdj CORij ,                          (6)
 This EA debt item incurs an debt which requires a principal
 of 20 min to update the use case diagram immediately, at                     where ρij is the co-variance between debt items i and j.
 time 0 (t0 ). However, if this particular debt item is held to a                With all the measurement attributes required in the MPT
 future phase (tn ), this will cause faultiness in clarifying system          model, the expected return, variance, and standard deviation of
 requirements being developed and eventually affects progress                 an EA debt portfolio can be computed using the equations (7),
 of system development.                                                       (9), and (10), respectively. The expected return of an EA debt
    As such, this EA debt item carries an interest amount                     portfolio at time t is the weighted sum of its EA debt items’
 (IA) of 16 min, which is the extra hours required in the                     expected returns:
 future to identify and correct the faultiness, which already                                                    n
                                                                                                                 X
                                                                                                      t
 brings negative impacts. Unfortunately, this interest amount                                        RP =              wdi Rdt i ,                 (7)
 is uncertain in such a way that it is assumed to fluctuate over                                                 i=1
 time depending on the scope, complexity, and impact of the                   with one constraint as presented in the following equation:
 components that the EA debt item is associated with at the time                                          n
                                                                                                          X
 of repayment. In this case, if the EA debt item is held until                                                   wdi = 1.                          (8)
 time 3, t3 , the interest amount, is very likely to increase from                                         i=1
 IAt1 = 16 to IAt4 = 96 min, because the EA debt item is                        On the other hand, the variance of the portfolio’s return,
 now not only bringing negative impacts to the planning phase,                which indicates the probabilities that the set of EA debt items
 but also the development phase. This uncertainty of interest                 will return different levels of benefits, is expressed as:
 growth rate in the future represents the risk level of an EA
                                                                                                   n                     n
 debt item, because the EA debt item with high interest growth                                     X                     X
                                                                                           δP2 =         wd2i δd2i + 2          wdi wdj ρij        (9)
 rate accumulate interest faster, which brings a higher future
                                                                                                   i=1                    i<j
 cost, which can be expressed as interest standard deviation
 (δd ). If the interest growth rate of a particular EA debt item                The EA debt portfolio’s standard deviation, δP can then be
 is not likely to grow or its growing rate is much lower than                 computed as follows:
 other EA debt items, this indicates that the debt payment can
                                                                                                            q
                                                                                                      δP = δP2                        (10)
 be deferred in a way that it carries lower risk.
    On the other hand, the expected return of an EA debt item                   As in practice, it is difficult to estimate and accurately model
 at time t can be understood as the number of working hours                   the exact amount of consequences of an EA debt item. Initially,
 that can be saved by paying the debt at t0 . The expected return             when an EA debt item associated with each layer is identified,



Copyright © 2019 for this paper by its authors.                                                                                               48
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
                           7th International Workshop on Quantitative Approaches to Software Quality (QuASoQ 2019)


 the principal, interest amount, and interest growth rate is
 estimated subjectively according to the enterprise architects
 experience. These rough estimations can then be adjusted
 using historical data that was collected throughout the EA life
 cycle. The more accurate and detailed the data is, the more
 reliable the estimation. The following section demonstrates
 how the proposed approach can be applied to reason about
 prioritization decisions.
 B. Application Process
    To apply portfolio theory and utility function to prioritize
 EA debt items for debt repayment, we need to ensure that
 the considerations mentioned in Section III-A are taken into
 account in order to map the portfolio model to EA debt
 measurement. Following, we propose a series of steps to iden-
 tify the optimal EA debt portfolio based on portfolio model
 (cf. Figure 1). On top of that, with the application of utility
 function, enterprise architects can reason and justify about
 EA debt repayment decisions based on the enterprises risk
 tolerance level. The basic steps of our proposed prioritization
 approach are stated as follows:
    1) Identify a project involved in EA implementation to
        achieve the target (to-be) architecture and let P be its
        EA debt portfolio.
    2) Identify the associated EA debt items, di | ∀i ∈
        {1, 2, ..., n}. This step is important, as not only debt
        items in the IT domain, but also in the business domain                            Fig. 1. Proposed Process Steps
        are identified.
    3) For each EA debt item, di , estimate the principal (Xdi ),
                                                                         the proposed steps described in Section III-B. Coarse-grained
        interest amount (IAdi ), interest growth rate/interest stan-
                                                                         estimates of EA debt items’ properties have been made and
        dard deviation (δdi ), weights (wdi ) and the correlations
                                                                         we acknowledge that it is sufficient for measuring the EA debt
        with other debt items (CORdi,j ).
                                                                         items for preliminary prioritization decision-making. Estimates
    4) For each EA debt item, di , determine the values of
                                                                         that are more detailed can be made when more real-world
        portfolio model, which are the expected return (Rdi )
                                                                         information is available upon which to base the estimates.
        and the covariance matrix (ρij ) using Equation (5) and
        Equation (6), respectively.                                         We considered that a Company ABC can choose from five
    5) Run the portfolio model on the available data to de-              projects to support the enterprise transition from a current EA
        termine the expected return (RP ), variance (δP2 ) and           to a target EA in order to improve business-IT alignment.
        standard deviation (δP ) of the EA debt portfolio.               Along the EA implementation life cycle, various types of EA
    6) Repeat steps 1-5 for all EA projects.                             debt items incurred in each EA project.
    7) Identify the efficient EA debt portfolios. The efficient             To provide a better understanding of our proposed approach,
        debt portfolios are the ones that lie on the efficient           the following data demonstrates the application of portfolio
        frontier and give the best return-risk trade-off if the debt     theory and utility function in the context of EA debt prior-
        is repaid at the current EA phase.                               itization to show how an optimal EA debt portfolio can be
    8) Determine the enterprise’s risk aversion coefficient. For         identified and prioritized for decision-making.
        simplicity, this study ranges risk aversion coefficients            Step 1: Identify an EA project: Project A (also known as
        from 1.0 to 5.0, with the lower number representing              EA debt portfolio A).
        higher tolerance to risk.                                           Step 2: Identify the associated EA debt items across the
    9) Apply the utility function (Equation (4)) to calculate the        four architectural layers. See Table II.
        risk-indifference curves.                                           Step 3: Estimate the principal, interest amount and interest
   10) Identify and prioritize the optimum portfolio where the           growth rate of each EA debt item. See Table III.
        utility curve intersects at the efficient frontier.                 Step 4: Compute the expected return and covariance matrix
                                                                         for each EA debt item. See Table IV.
                         IV. C ASE S TUDY                                   Step 5: Run the portfolio model to compute the expected
    For proof-of-concept purpose, we applied a synthetic case            return, variance, and standard deviation of the EA debt port-
 study and artificial data was generated accordingly based on            folio. See Table V first row.



Copyright © 2019 for this paper by its authors.                                                                                49
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
                              7th International Workshop on Quantitative Approaches to Software Quality (QuASoQ 2019)


                               TABLE II                                                               TABLE IV
                  L IST OF EAD S IN DEBT P ORTFOLIO A                                P ORTFOLIO MODEL VALUES OF DEBT P ORTFOLIO A

                               PROJECT A                                                                 PROJECT A
    ID    EAD domain                EAD description                                           Expected    Standard       Covariance matrix
                                                                             ID    Weight
    A1    Business Architecture     Fail to remove the outdated ele-                          return      deviation A1       A2     A3      A4
                                    ments in a use case diagram              A1     0.25      16 min      0.8       0.64    0.56   0.48    0.4
    A2    Application Architecture  Fail to document interface descrip-      A2     0.25      11 min      0.7       0.56    0.49   0.42    0.35
                                    tions                                    A3     0.25      14 min      0.6       0.48    0.42   0.36    0.3
    A3    Data Architecture         Lack of data model for an applica-       A4     0.25      4 min       0.5       0.4     0.35    0.3    0.25
                                    tion
    A4    Technology Architecture   Missing implementation standard
                                                                                                       TABLE V
                                                                                    R ISK AND EXPECTED RETURN OF 5 DEBT P ORTFOLIOS
                              TABLE III
 P RINCIPAL , INTEREST AMOUNT AND INTEREST GROWTH RATE ESTIMATES                    Project   Portfolio risk    Portfolio expected return
                         OF DEBT P ORTFOLIO A                                         A            0.65                     9.5
                                                                                      B           0.467                    9.333
                                  PROJECT A                                           C            0.4                     12.75
            ID    Principal         Interest      Interest                            D           0.567                    15.67
                                    amount        growth rate                         E            0.2                     8.333
            A1    20 min            16 min        80 %
            A2    16 min            11 min        70 %
            A3    23 min            14 min        60 %                                           V. R ELATED W ORK
            A4    8 min             4 min         50 %
                                                                              Despite the vast attention have been paid to technical debt,
                                                                           to our best knowledge, there is no existing approach to
    Step 6: Identify other EA projects and repeat steps 2-5.               prioritize EA debt items as this metaphor is recently proposed
 Table V displays the computed portfolio’s expected return and             by us [12]. Therefore, existing prioritization approaches have
 risk of five projects.                                                    been studied in the context of technical debt.
                                                                              Technical debt management (TDM) is composed of a se-
    Step 7: Identify the efficient EA debt portfolios. As shown
                                                                           quence set of activities to prevent technical debt from being
 in Figure 2, debt portfolio A and B are inefficient portfolio
                                                                           incurred or manage existing technical debt to maintain it under
 because other portfolios can offer higher return at the similar
                                                                           a desirable level [6]. TDM activities include TD identifi-
 level of risk or a lower risk at the similar level of return.
                                                                           cation, TD measurement, TD prioritization, TD prevention,
    Step 8: Define the enterprise’s risk aversion coefficient.             TD monitoring, TD repayment, TD documentation, and TD
 Company ABC has a risk aversion coefficient value of 2.                   communication [6]. Technical debt prioritization is considered
    Step 9: For visualization, calculate and plot the risk-                as one of the TDM activities in which the identified technical
 indifference curves. See Figure 3 for exemplary curves for                debt items are ranked based on predefined rules to decide
 the utility values of 4,6,8, and 10.                                      either immediate repayment or deferred repayment on the debt
    Step 10: Identify the optimum portfolio for prioritization             items [6]. Existing studies have discussed four decision ap-
 by solving Eq. 4 for every portfolio. The risk-return scatter             proaches to deal with technical debt prioritization for complex
 plot in Figure 4 indicates that EA debt portfolio C is the                decision-making: Cost-benefit analysis [3], Remediation cost
 optimum portfolio that provides best risk-return trade-offs and           analysis [16], Real Options [13] and Portfolio theory [4], [5].
 maximum satisfaction, as it (almost) based on the utility curve           Meanwhile, the systematic literature review on the financial
 of 10.                                                                    aspect of managing technical debt conducted by Ampatzoglou
    Our synthetic case study shows that our approach is applica-           et al. [25] concluded that the three most popular financial
 ble in general. However, further research is necessary to enable          approaches are cost/benefit analysis, real-options analysis, and
 enterprises to apply our approach in practice. Especially, steps          portfolio management.
 3 and 4 might be extremely challenging, due to missing                       Simple cost-benefit analysis: A simple cost-benefit ap-
 experience in the field. To tackle step 3, we suggest collecting          proach was proposed to prioritize technical debt in terms of
 and documenting possible EA debt items and provide them as                which classes should be re-factored first [3]. Each technical
 catalogs to the community. These catalogs can help to identify            debt item consists of the estimations of three metrics: prin-
 possible debt items and serve as a discussion basis to get a              cipal, interest probability, and interest amount. This approach
 deeper understanding of the domain.                                       prioritize code level debts based on the impact of the God
    Step 4 requires the determination of the measurement at-               classes on the software maintainability and correctness.
 tributes, which are needed to calculate the optimal portfolio.               Remediation cost analysis: Moreover, the SQALE (Soft-
 However, those attributes usually will be not obvious as for              ware Quality Assessment Based on Life-cycle Expectations)
 financial assets. Therefore, future research should elaborate on          method with Sonar tool was proposed to analyze technical
 methods that enable practitioners to assess this attributes in an         debt that associated with an application source code [16]. The
 easy manner.                                                              authors proposed the synthesis of SQALE Quality and SQALE



Copyright © 2019 for this paper by its authors.                                                                                             50
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
                                             7th International Workshop on Quantitative Approaches to Software Quality (QuASoQ 2019)


                             18                                                                                     18
                             16                                                                                     16
         Portfolio Return
                             14                             D                                                       14                          D




                                                                                                 Portfolio Return
                             12                    C                                                                12                 C
                             10                                                                                     10
                              8                        B          A                                                  8                     B        A
                                        E                                                                                   E
                              6                                                                                      6                               Utility = 4
                                                                                                                                                     Utility = 6
                              4                                                                                      4
                                                                                                                                                     Utility = 8
                              2                                                                                      2                              Utility = 10
                              0                                                                                      0
                                            0.2      0.4       0.6           0.8                                                0.2      0.4       0.6        0.8
                                                  Portfolio Risk                                                                      Portfolio Risk

                            Fig. 2. Risk-return Trade-off of Five Debt Portfolios                                        Fig. 4. The Optimal Debt Portfolio


                             18                                                           proposed to assist in decision-making in which technical debt
                             16                                                           items should be repaid and which one should be held for
                             14                                                           technical debt management [4], [5]. The measurement units
                                                                                          embedded in portfolio model, such as expected return, return
         Portfolio Return




                             12
                                                                                          variance, and return standard deviation are mapped to the
                             10                                                           context of technical debt management. Each technical debt
                              8                                                           item was viewed as an asset and the application of portfolio
                                                                  Utility = 4             mathematical formulation into a portfolio of debt items helps
                              6
                                                                  Utility = 6             software developers to decide which technical debt items
                              4                                                           should be repaid first in order to minimize the future main-
                                                                  Utility = 8
                              2                                   Utility = 10            tenance cost. Also, the portfolio theory was integrated into
                              0                                                           goal-obstacle method to specifically deal with requirements
                                            0.2      0.4       0.6           0.8          compliance debt [15].
                                                  Portfolio Risk                             Our approach differs from those works in two aspects.
                                                                                          First, we broaden the scope of technical debt to the entire
      Fig. 3. Risk-indifference curves when risk aversion coefficient=2                   organization and propose a mapping of EA debt properties to
                                                                                          portfolio theory properties. Second, the existing literature on
                                                                                          applying portfolio theory to technical debt lacks an explicit
 Analysis model to measure technical debt in terms of the                                 description of its application, while we do.
 distance between the codes current quality state and its target                             Benchmarking and portfolio matrix: Instead of prioritizing
 state to indicate the quality of an application. On top of that,                         technical debt in general, Plösch et al. [14] focus particularly
 remediation index was used to represent the remediation cost                             on design debt, which is incurred due to the violations and
 of corrective actions required to resolve the non-compliance                             non-conformance of design principles on source code level.
 associated with each component of the applications software                              The authors developed a tool called MUSE in which a
 code.                                                                                    portfolio matrix is used to prioritize the identified violations
    Real-options approach: Another existing approach is incor-                            and to communicate design debt. In the proposed approach,
 porating real options thinking into the valuation of technical                           a benchmarking-oriented measurement is applied to derive a
 debt. Technically, the concept of a real option is about a right                         quality index and categorize each design best practice into
 to make a future decision without any obligation depending                               Q0- to Q5-area based on the number of identified design best
 on the way uncertainty is resolved. In other words, purchasing                           practice violations.
 the real option is analogous to investing in paying off technical                           Based on the analysis of existing literature, it is found that
 debt that facilitates future software changes. The real options                          the limitation of the aforementioned approaches is that the
 theory was applied to effectively deal with unpredictable                                relative importance of business impacts or operations are not
 changes in system requirement engineering, time period, and                              taken into account. Therefore, the concept of linking EA debt
 development cost [13]. The proposed approach considers the                               to enterprise architecture and applying portfolio thinking is
 risk associated with technical debt decisions to manage the                              novel. While Stochel et al. [26] suggested to regard each
 value of an organization’s strategic flexibility.                                        distinct type of technical debt such as process debt as a
    Portfolio theory: Furthermore, a portfolio approach was                               debt portfolio, we suggest to map each EA project as a debt



Copyright © 2019 for this paper by its authors.                                                                                                                     51
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
                            7th International Workshop on Quantitative Approaches to Software Quality (QuASoQ 2019)


 portfolio.                                                                    [5] Y. Guo and C. Seaman, “A portfolio approach to technical debt
                                                                                   management,” in Proceeding of the 2nd working on Managing
                                                                                   technical debt - MTD ’11, 2011, p. 31. [Online]. Available:
                          VI. C ONCLUSION                                          https://resources.sei.cmu.edu/asset files/Presentation/2011 017 001
                                                                                   516999.pdfhttp://portal.acm.org/citation.cfm?doid=1985362.1985370
     Implementing EA is essential to enhance the business-IT                   [6] Z. Li, P. Avgeriou, and P. Liang, “A systematic mapping study on
 alignment in a holistic manner. However, academia, software                       technical debt and its management,” Journal of Systems and Software,
                                                                                   vol. 101, pp. 193–220, 2015.
 developers, and organizations have been focusing on technical                 [7] N. Brown, I. Ozkaya, R. Sangwan, C. Seaman, K. Sullivan, N. Za-
 debt, which deals with the quality issues on code, application,                   zworka, Y. Cai, Y. Guo, R. Kazman, M. Kim, P. Kruchten, E. Lim,
 and system level. Considering the importance of EA in creat-                      A. MacCormack, and R. Nord, “Managing technical debt in software-
                                                                                   reliant systems,” Proceedings of the FSE/SDP workshop on Future of
 ing value to organizations, this work has explored a method                       software engineering research - FoSER ’10, no. January, p. 47, 2010.
 to identify the optimal set of EA debt items, which should be                 [8] M. Albarak and R. Bahsoon, “Prioritizing technical debt in database
 repaid next. Therefore, we have elaborated on the necessary                       normalization using portfolio theory and data quality metrics,” in Pro-
                                                                                   ceedings of the 2018 International Conference on Technical Debt -
 attributes of EA debts (RQ1.1) and on the necessary process                       TechDebt ’18, 2018, pp. 31–40.
 steps (RQ1.2). To tackle (RQ1.1), we have defined a mapping                   [9] D. H. Olsen, “Enterprise Architecture management challenges in the
 from the EA debt domain to the used terminology in portfolio                      Norwegian health sector,” Procedia Computer Science, vol. 121, pp.
                                                                                   637–645, 2017.
 optimization (see table I). This mapping is used as input for the            [10] J. A. Zachman, “A Framework for Information Systems Architecture,”
 process (see figure 1) to prioritize the EA debts that answers                    IBM Systems Journal, vol. 26, pp. 276–292, 1987.
 (RQ1.2).                                                                     [11] F. Gampfer, A. Jürgens, M. Müller, and R. Buchkremer, “Past, current
                                                                                   and future trends in enterprise architectureA view beyond the horizon,”
     One of the limitations is that the portfolio-based EA debt                    Computers in Industry, vol. 100, pp. 70–84, 9 2018.
 model is developed based upon a high-level approach. This                    [12] S. Hacks, H. Höfert, J. Salentin, Y.-C. Yeong, and H. Lichter, “Towards
 means any details, such as EA debt estimation tools and                           the Definition of Enterprise Architecture Debts,” arXiv e-prints, 2019.
                                                                                   [Online]. Available: https://arxiv.org/pdf/1907.00677.pdf
 methods, are outside of the research scope. Therefore, estima-               [13] Z. S. H. Abad and G. Ruhe, “Using real options to manage Technical
 tion guidelines should be developed based on the professional                     Debt in Requirements Engineering,” 2015 IEEE 23rd International
 experienced EA practitioners to provide the reference for es-                     Requirements Engineering Conference (RE), no. January, pp. 230–235,
                                                                                   2015.
 timating the debt principal and interest value of each assessed              [14] R. Plösch, J. Bräuer, M. Saft, and C. Körner, “Design debt prioritization,”
 EA debt item.                                                                     in Proceedings of the 2018 International Conference on Technical Debt
     In the meantime, coarse-grained estimations of EA debt                        - TechDebt ’18, vol. 10, no. 18. ACM, 2018, pp. 95–104.
                                                                              [15] B. Ojameruaye and R. Bahsoon, “Systematic Elaboration of Compliance
 measurement units have been made and we acknowledge                               Requirements Using Compliance Debt and Portfolio Theory,” in 20th
 that it is sufficient for prioritizing the EA debt items for                      International Working Conference, REFSQ, Essen, Germany, 2014, pp.
 preliminary decision-making. In real-world practice, EA prac-                     152–167.
                                                                              [16] J.-L. Letouzey, “The SQALE Method for Managing Technical Debt,” in
 titioners are encouraged to substitute estimations based on                       Proceedings of the 3rd International Workshop on Managing Technical
 historical measurements of extra costs required in EA debt                        Debt (MTD12), IEEE, Zurich, Switzerland, 2012, 2012, pp. 31–36.
 repayment. More detailed planning can be made when more                      [17] H. Markowitz, “Portfolio Selection,” The Journal of Finance, vol. 7,
                                                                                   no. 1, pp. 77–91, 1952.
 historical information is available upon which to facilitate the             [18] M. Engels, “Portfolio Optimization: Beyond Markowitz,” Ph.D. disser-
 estimations.                                                                      tation, Universiteit Leiden, 2004.
     Future research within this domain is two-fold. First, it is             [19] I. Omisore, M. Yusuf, and N. I. Christopher, “The modern portfolio
                                                                                   theory as an investment decision tool,” Journal of Accounting and
 necessary to provide catalogs of EA debt items to enable                          Taxation, vol. 4, no. 2, pp. 19–28, 2012.
 practitioners to identify them in their EA. Those catalogs need              [20] J. Norstad, “An Introduction to Utility Theory,” 2011. [Online]. Avail-
 to be validated and expanded. Further, effort should be invested                  able: http://srihariseshadri.com/docs/how-much-would-you-bet/norstad
                                                                                   utility.pdf
 to develop methods, which enable the practitioners to assess                 [21] C. Carlsson, R. Fullér, and P. Majlender, “A possibilistic approach
 the measurement attributes that are needed to compute the                         to selecting portfolios with highest utility score,” Turku Centre for
 optimal portfolio. Second, further means to prioritize technical                  Computer Science, Tech. Rep., 2002.
                                                                              [22] S. Lim, T. J. V. Saldanha, S. Malladi, and N. P. Melville, “Theories
 debts need to be transferred to the domain of EA debts. Then,                     Used in Information System Research : Insights from Complex Network
 the different means need to be compared concerning their                          Analysis,” Journal of Information Technology Theory and Application,
 efficiency to determine the most efficient one.                                   vol. 14, no. 2, pp. 5–46, 2013.
                                                                              [23] I. Attarzadeh and S. H. Ow, “Project Management Practices: The Criteria
                                                                                   for Success or Failure,” Communications of the IBIMA, vol. 1, no. 28,
                             R EFERENCES                                           pp. 234–241, 2008.
                                                                              [24] The Open Group, “TOGAF.” [Online]. Available: http://www.opengroup.
  [1] W. Cunningham, “The WyCash portfolio management system,” ACM                 org/togaf
      SIGPLAN OOPS Messenger, vol. 4, no. 2, pp. 29–30, 1993.                 [25] A. Ampatzoglou, A. Ampatzoglou, A. Chatzigeorgiou, and P. Avgeriou,
  [2] E. Tom, A. Aurum, and R. Vidgen, “An exploration of technical debt,”         “The financial aspect of managing technical debt: A systematic literature
      Journal of Systems and Software, vol. 86, no. 6, pp. 1498–1516, 2013.        review,” Information and Software Technology, vol. 64, pp. 52–73, 2015.
  [3] N. Zazworka, C. Seaman, and F. Shull, “Prioritizing Design Debt         [26] M. G. Stochel, M. R. Wawrowski, and M. Rabiej, “Value-Based
      Investment Opportunities,” Proceeding of the 2nd working on Managing         Technical Debt Model and Its Application,” ICSEA 2012, The Seventh
      technical debt - MTD ’11, pp. 39–42, 2011.                                   International Conference on Software Engineering Advances, no. c, pp.
  [4] C. Seaman, Y. Guo, N. Zazworka, F. Shull, C. Izurieta, Y. Cai, and           205–212, 2012.
      A. Vetrò, “Using technical debt data in decision making: Potential
      decision approaches,” in 3rd International Workshop on Managing
      Technical Debt, MTD 2012 - Proceedings, 2012, pp. 45–48.




Copyright © 2019 for this paper by its authors.                                                                                                       52
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).