Elsevier

Information Sciences

Volume 375, 1 January 2017, Pages 296-313
Information Sciences

Financial modeling and improvement planning for the life insurance industry by using a rough knowledge based hybrid MCDM model

https://doi.org/10.1016/j.ins.2016.09.055Get rights and content

Abstract

Financial modeling for the life insurance industry involves two main difficulties: (1) Selecting the minimal and critical variables for modeling while considering the impreciseness and interrelationships among the numerous attributes and (2) measuring plausible synergy effects among variables and dimensions that might cause undesirable biases for an evaluation model. To overcome these difficulties, this paper proposes a two-stage hybrid approach: Rough financial knowledge is retrieved first, and then the obtained core attributes are measured and synthesized using fuzzy-integral-based decision methods. The main innovation of this study is the use of rough knowledge retrieval procedures and fuzzy measures for exploring the synergy effects on financial performance. This approach is expected to support insurers to systematically improve their financial performance. A group of life insurance companies in Taiwan was analyzed, and the findings support the existence of interrelated synergy effects among the core criteria. In addition, five companies were examined to illustrate financial performance improvement planning with this approach. This study bridges the gap between advanced soft computing techniques and pragmatic financial modeling in a dynamic business environment.

Introduction

The financial industry is critical to the stability of a nation's economy. Therefore, since the global financial crisis of 2008, increasing interest has been shown in examining the financial performance (FP) of the financial industry, especially for the benefit of stakeholders (e.g., investors, management, and governments). Two mainstream research topics are related to the FP of financial institutions (mainly banks and insurance companies), namely bankruptcy and financial failure prediction [29], and FP evaluation for credit scoring and investment purposes [43]. The present study focused on the evaluation of the FP of life insurance companies. The methodologies adopted in previous research [12] for assessing the FP of financial institutions can be categorized as follows: (1) methodologies based on statistical models, (2) methodologies based on machine learning and soft-computing techniques, and (3) methodologies involving multiple criteria decision-making (MCDM) methods [61]. In addition, because of the dynamics and complexity of business environments, hybrid approaches (i.e., the integration of more than two methodologies) are increasing rapidly. For example, under the framework of SMAA TRI [54], Angilella and Mazzù [3] proposed the ELECTRE TRI [63] for building a judgmental rating model to support the financing decisions for small and medium-sized enterprises (SMEs). In their model, qualitative judgments and quantitative financial data were considered and integrated for rating SMEs.

Conventional social science studies have mainly been based on statistical methods for identifying the relationships between selected financial variables and the subsequent FP of financial companies. Methods such as discriminant analysis, factor analysis, principal component analysis (PCA), and logit regression have been widely adopted in previous research. Following the work of Altman [1], discriminant analysis has been widely used for analyzing the financial failure of companies [28]. Subsequently, the logit-regression-based synthesized score approach (Z-score) presented in the influential papers [1], [2] became prominent; for example, West [60] used logit regression along with factor analysis to measure the FP of banks. Other statistical methods, such as probit analysis and PCA, have been applied or combined for predicting the FP of banks [8]. Kumar and Ravi [28] presented a systematic review of this research area. However, regarding the statistical approach, the following three aspects are questionable: (1) The assumption of independence among the variables, (2) certain probabilistic assumptions, and (3) the additive-type aggregation in the synthesized score approach. In a complex financial environment, interrelationships among variables often exist; however, owing to the limitations of statistical methods, certain interrelated or nonadditive-type influences cannot be measured or modeled accurately.

Machine learning techniques (e.g., artificial neural networks (ANNs), genetic algorithms (GAs), decision trees (DTs), and support vector machines (SVMs)) are useful for determining nonlinear relationships among data sets. Most machine learning techniques do not require the probabilistic distribution of data to be assumed; therefore, they are more practical for real business applications. A recent survey of machine learning methods used for predicting financial crises [30] suggested that bankruptcy prediction and credit scoring could be regarded as classification problems in machine learning. The survey categorized the adopted techniques as single or hybrid classifiers. Of the various classifiers, the ANN-related techniques (e.g., back-propagation, self-organizing map, and competitive learning neural networks) are probably the most dominant for financial prediction problems. For example, a previous study [7] compared the financial distress prediction results of ANN techniques with statistical methods for life insurance companies. The back-propagation ANN technique was found to outperform traditional statistical methods. ANN techniques mimic the learning mechanism of the brain, and the learned results are stored in the connections between neurons; the learning process and learned results are often criticized as a “black box,” implying that understanding the outcomes is difficult [13]. Some single classifiers, such as SVM [59] and GA [37], have a similar drawback. Hybrid classifiers often involve the integration of two or more techniques (e.g., GA + ANN+ ARIMA [58] and DT + ANN + SVM [24]). One of the techniques is used for performing the initial classification, and the others are used for the tuning parameters of the hybrid models [23]. Generally, machine-learning-based studies have focused on increasing the accuracy of classification or prediction.

Soft computing techniques (mainly the fuzzy set [64] and rough set [39] theories, which are discussed here) are based on solid mathematical foundations, and are used for modeling the impreciseness or uncertainty in a system. They have been widely applied in engineering [66] and social economics [56], [66]. One of the key advantages of soft computing techniques is their logical reasoning capability, which can help obtain meaningful knowledge (i.e., logic or rules) for solving a problem. These techniques are often integrated with machine learning techniques for solving the FP evaluation problem. For example, ANN techniques and fuzzy inference were integrated to address the credit scoring problem [33] and the adaptive-network-based fuzzy inference system was integrated with the dominance-based rough set approach (DRSA) for evaluating the FP of banks [45]. As suggested by the aforementioned survey [30], soft-computing-integrated classification techniques appear to be the most promising direction for future research on FP prediction.

The third category of methodology used for assessing the FP or operational efficiency of financial institutions comprises MCDM methods. Recently, Fethi and Pasiouras [13] and Zopounidis et al. [67] presented updated reviews on this approach. The fuzzy set theory is commonly incorporated in MCDM methods for FP evaluation problems [34], [44] to mimic the imprecise judgments and reasoning of decision-makers (DMs). The MCDM approach considers multiple criteria (the terms “attributes” and “criteria” are used interchangeably in this paper) simultaneously to make ranking or selection decisions, and is based on utility theory, as developed in economics [65]. Other methods that involve the construction of decision models on the basis of pairwise comparisons between criteria are also used; of these methods, analytic hierarchy process (AHP) extended methods have attracted the most attention. The original AHP [41] is based on the assumption of the independence of criteria. The generalized analytic network process (ANP) [42] allows internetwork relationships in its model. The generalized ANP has been adopted for evaluating the performance of wealth management banks [62] and commercial banks [46]. The MCDM approach is based on the experience and knowledge of DMs and experts, and is therefore appropriate for in-depth investigations of relationships between the predefined criteria (attributes) of a complex problem, or system [65].

FP modeling for financial institutions involves difficulties related to (1) the selection of the minimal and critical variables, (2) the clarification of cause–effect relationships among variables, and (3) the measurement of plausible synergy effects among the criteria and dimensions. The complexity of imprecise and conjoint relationships among dimensions and criteria cannot be measured or modeled accurately using statistical models, a single soft computing technique, or an MCDM method. Therefore, a novel hybrid approach is proposed in this paper. In addition, although a considerable number of studies have adopted various methods and techniques to increase the classification accuracy of FP evaluation, or devise a system of logic or rules for the problem, limited research has been undertaken to diagnose the FP of financial companies for improvement planning. Only a few recent studies [46], [48], [49] have pursued this direction. Moreover, the present study attempts to obtain more constructive implications regarding improvement planning for addressing the FP modeling problem.

To overcome the aforementioned difficulties, this study proposes a two-stage approach. In the first stage, the learning capability of a soft computing technique is used to retrieve rough knowledge. In the second stage, the core attributes are synthesized using a nonadditive-type fuzzy aggregator to construct a hybrid MCDM model. The expected contributions of the proposed approach are as follows: (1) Retrieving core attributes and rough financial knowledge (decision rules) for enabling in-depth analyses, (2) refining the rough knowledge by identifying cause–effect influences among the core attributes, (3) measuring the synergy effects among dimensions and criteria for performing accurate FP evaluations, and (4) facilitating systematic improvement planning for insurance companies.

The innovations of this study can be outlined in two parts, namely modeling and business applications. Regarding modeling, a novel mechanism is devised for retrieving rough knowledge from historical data, and a nonadditive approach is applied to measure the plausible synergy effects among variables. In literature, the classical financial optimization model (i.e., mean-variance theory [35]) has strengths regarding obtaining optimal results under the expected returns and risk. Mean-variance theory has been applied in many fields (e.g., in supply chain risk analysis [10]); however, it is based on the presumed statistical distributions of data, which have the limitations of the aforementioned statistical approach (e.g., the independence of variables). This study differs from the classical financial optimization model in emphasizing exploring the imprecise patterns and knowledge provided by historical data, which requires fewer assumptions for financial modeling. In the era of “big data,” developing ways of leveraging the strengths of machine learning and soft computing techniques to realize accurate modeling is a challenging and yet valuable research topic.

For business applications, new tools for facilitating systematic FP improvement are proposed, namely internetwork relationship maps (INRMs) and directional flow graphs (DFGs). Thus, this study is expected to provide an enhanced understanding of how multidisciplinary methodologies can be integrated to obtain implicit and critical knowledge regarding FP modeling, thereby bridging the gap between academia and practice. Empirical cases of registered life insurance companies in Taiwan were analyzed to illustrate the proposed approach.

The remainder of this paper is organized as follows: Section 2 introduces the background and discusses the development of existing methodologies. Section 3 elucidates the proposed hybrid approach. In Section 4, a group of registered life insurance companies in Taiwan are analyzed as empirical cases. The results and implications of the current study are discussed in Section 5. Finally, Section 6 concludes the study and presents suggestions for future research.

Section snippets

Preliminary

This section briefly introduces the background of the present study and discusses the development of methods used for FP prediction and the evaluation problem. The purposes of adopting each method and technique in the proposed approach are also explained and discussed.

Rough-knowledge-based hybrid approach

In this section, the framework and procedure of the proposed approach are introduced. The approach is divided into two parts: (1) DRSA for investigating imprecise patterns and granules of knowledge by obtaining core attributes and decision rules, and (2) a DRSA-core-attribute-based hybrid decision model for evaluating the nonadditive performance effects among the attributes (criteria).

Empirical cases from the life insurance industry

In view of the importance of the life insurance industry to the stability of the national economy, in 2009 Taiwan made it mandatory for all registered life insurance companies to report their operational and financial performance to the public. To illustrate the application of the proposed nonadditive hybrid approach to FP diagnoses, openly accessible data from the life insurance industry in Taiwan were analyzed.

Results and discussions

To examine the proposed approach, the data sets of five life insurance companies were obtained from the testing set. The five companies were Zurich International Insurance, Taiwan Branch (A), Fubon Life Insurance (B), China Trust Life Insurance (C), Prudential Life Insurance, Taiwan Branch (D), and First Aviva Insurance (E). The entire testing set (27 alternatives) was used to transform the raw data of the five companies into performance scores ranging from 0 (worst) to 10 (best) for each

Concluding remarks

In this study, a hybrid approach for the FP diagnosis of life insurance companies was proposed. The attribute reduction and implicit knowledge retrieval capabilities of DRSA helped induct rough financial knowledge from historical data in the first stage. In the next stage, a hybrid decision model constructed using the DANP and fuzzy integral methods was used to refine the rough financial knowledge in two ways: (1) the cause–effect influence relations among the core dimensions or criteria were

Acknowledgments

We are grateful for the valuable opinions and suggestions from the Editor-in-Chief Prof. Pedrycz and Reviewers, which have helped us improve this study in many aspects; also, the funding supports from the Ministry of Science and Technology of Taiwan under the grant numbers MOST-104-2410-H-305-052-MY3 and MOST-104-2410-H-034-064 are appreciated.

Reference (67)

  • J.J. Huang et al.

    Two-stage genetic programming (2SGP) for the credit scoring model

    Appl. Math. Comput.

    (2006)
  • C. Hung et al.

    A selective ensemble based on expected probabilities for bankruptcy prediction

    Expert Syst. Appl.

    (2009)
  • K. Ishii et al.

    A model of human evaluation process using fuzzy measure

    int. J. Man-mach. Stud.

    (1985)
  • Y.C. Ko et al.

    An extended fuzzy measure on competitiveness correlation based on WCY

    Knowl.-Based Syst.

    (2013)
  • J.J.H. Liou et al.

    A dominance-based rough set approach to customer behavior in the airline market

    Inf. Sci.

    (2010)
  • J.J.H. Liou et al.

    A fuzzy integral-based model for supplier evaluation and improvement

    Inf. Sc.

    (2014)
  • R. Malhotra et al.

    Differentiating between good credits and bad credits using neuro-fuzzy systems

    Eur. J. Operat. Res.

    (2002)
  • A. Mardani et al.

    Fuzzy multiple criteria decision-making techniques and applications–Two decades review from 1994 to 2014

    Expert Syst. Appl.

    (2015)
  • T. Murofushi et al.

    An interpretation of fuzzy measures and the Choquet integral as an integral with respect to a fuzzy measure

    Fuzzy Sets Syst.

    (1989)
  • C.S. Ong et al.

    Building credit scoring models using genetic programming

    Expert Syst. Appl.

    (2005)
  • Y.P. Ou Yang et al.

    A VIKOR technique based on DEMATEL and ANP for information security risk control assessment

    Inf. Sci.

    (2013)
  • P. Sevastjanov et al.

    Stock screening with use of multiple criteria decision making and optimization

    Omega

    (2009)
  • K.Y. Shen et al.

    A new approach and insightful financial diagnoses for the IT industry based on a hybrid MADM model

    Knowl.-Based Syst.

    (2015)
  • K.Y. Shen et al.

    Combining VIKOR-DANP model for glamor stock selection and stock performance improvement

    Knowl.-Based Syst.

    (2014)
  • S. Tsumoto

    Automated extraction of medical expert system rules from clinical databases based on rough set theory

    Inf. Sci.

    (1998)
  • J.J. Wang et al.

    Stock index forecasting based on a hybrid model

    Omega

    (2012)
  • R.C. West

    A factor-analytic approach to bank condition

    J. Bank. Finance

    (1985)
  • C.R. Wu et al.

    Evaluating business performance of wealth management banks

    Eur. J. Operat. Res.

    (2010)
  • L.A. Zadeh

    Fuzzy sets

    Inf. Contr.

    (1965)
  • C. Zopounidis et al.

    Multiple criteria decision aiding for finance: an updated bibliographic survey

    Eur. J. Operat. Res.

    (2015)
  • E.I. Altman

    Financial ratios, discriminant analysis and the prediction of corporate bankruptcy

    J. Finance

    (1968)
  • E.I. Altman

    Predicting Financial Distress of Companies: Revisiting the Z-Score and ZETA Models. Stern School of Business

    (2000)
  • K. Atanassov
    (2014)
  • Cited by (63)

    View all citing articles on Scopus
    View full text