BUSINESS MODELS IN BANKING: A CLUSTER ANALYSIS USING ARCHIVAL DATA.

• Author: Lueg, Rainer
• Position: Report

1. Introduction

   Defining and measuring 'business models' has become an emerging theme in contemporary accounting research (Huelsbeck, Merchant, and Sandino 2011, Ittner, Larcker, and Randall 2003, Nielsen and Roslender 2015, Vera-Munoz, Shackell, and Buehner 2007). Specifically, banking regulators have started to rethink the current 'one size fits all' regulation model and now explore the feasibility of business model-specific regulation. In fact, every recent publication on the potential impact of regulatory ratios contains at least one section where the impact is differentiated across business models (EBA 2014:45ff, 2015, 2016:78ff). The reports reveal that two of the six Basel III ratios, namely Leverage Ratio and Net Stable Funding Ratio, show very different results depending on the type of bank. Hence, it would not be sensible to require from all banks to comply with one common threshold. In this vein, banking regulators have realized that the literature to identify business models in general--and banking business models in particular--is still in its infancy. Knowledge on this topic is not sufficiently consolidated to be ready to be applied across thousands of banks, of which some are systemically important. Although convinced that a business model-specific regulation would be appropriate, its introduction would currently face the following challenges: first, the term 'business model' is not uniquely defined. Second, manual classifications of annual report information are too time-consuming and subjective. Third, annual report tend to be biased in the sense that they report which business model the bank would like to have rather than the business model that is actually in place. This paper addresses these concerns by (i) defining business models and (ii) proposing a statistical and automated approach to identify them based on audited information.

   Looking at the literature, it is surprising that scholars and practitioners struggle with the starting point of any discussion on business models: what is actually a business model (cf. Teece 2010, Zott, Amit, and Massa 2011). As a consequence, a valid and reliable measurement of business models is practically non-existent: the literature remains fragmented with incommensurable tales of allegedly successful or failed business models, which are mostly descriptive and lack theoretical foundation and predictive ability (DaSilva and Trkman 2014, Kulins, Leonardy, and Weber 2015). Without a clear measurement of business models, their success cannot be predicted, and their relative performance or the opportunity cost of choosing an alternative business model cannot be assessed. Researchers may face these challenges of measuring business models because they have largely ignored the possibility that business models may only be determined and measured given a specific industry and context (Kulins et al. 2015). Making an analogy to the literature on strategy, Porter's (1980) work acknowledges the specificity of strategy to an industry, an advancement still missing in the business model literature (exceptions: Teece 2010, Zott and Amit 2007). So far, very few studies have broken new ground in defining and measuring business models with constructs. Wirtz et al. (2010) conduct a seminal study among 22 Web 2.0 companies. The authors categorize four non-exclusive types of business models (content, commerce, context, and connection) and show the most/least favorable links to factors that shape the market for Web 2.0 services (social networking; interaction orientation; personalization/ customization and user-added value). Kulins et al (2015) analyze 41 entrepreneurial firms, and find that three unknown specific business model configurations foster financial performance. DaSilva and Trkman (2014:382) suggest a more solid foundation in the resource based view (RBV) and transaction cost economics (TCE) and elicit that business models "represent a specific combination of resources which through transactions generate value for both customers and the organization." Sanchez and Ricart (2010) conduct comparative case studies that account for the contextual factors in low-income markets. They dismiss the idea of a general business model and derive an equifinal continuum of business models that are either isolated (resource-based, aimed at value-for-money for the customer) or interactive (complementor-based, aimed at increasing customer's willingness to pay). Huelsbeck et al. (2011) are the only researchers that statistically back-test a realized business model with proprietary data. They demonstrate that what the managers deemed to be the business model was only a poor predictor of the high realized performance.

   To further our understanding on business models beyond storytelling and descriptive checklists, we propose a measurement of business models and their changes over time using publicly available data. We deliberately choose the Western banking industry (EU and U.S.) to be sector-specific and account for context: First, the crisis of 2008 has induced substantial changes to banks' business models. Second, regulators start to explicitly require that a bank must explain the sustainability of its business models in practical terms (e.g., Deutsche Bundesbank 2007). Yet, regulators have not made clear specifications what they are looking for and lack a measurement to assess the realized business models in banks. Confirming that a quantitative approach like clustering can be used to identify business models would be good news for regulators as they could use this technique to form peers and define benchmark business models instead of screening numerous annual reports.

   Pursuing this objective, we proceed as follows: (1) Departing from the general business model literature, we offer an industry-specific definition for banks and identify six key variables as proxies. In step (2), we use cluster analysis to classify the business models of selected banks. Similarly to Ayadi et al. (2011), we find the three statistical business models 'Retail bank', 'Universal bank', and 'Investment bank'. In Europe, the universal business model is the most common one, whereas in the U.S. it is the least common one. In step (3), we back-test whether the self-reported business model of each bank is matches our classification. Our back-testing reveals that clustering with our key variables results in a 100% match for investment banks, a convincing 89.7% match for universal banks, and a low 44% match for retail banks. We conclude with good and bad news for regulators: it is good news that clustering can be used to identify business models. It is bad news that the cluster variables that separate universal and retail banks need to be refined because their low match result implies that discriminatory power is not very high. In step (4), we explore the path dependency of business model change (DaSilva and Trkman 2013) during the financial crisis. We find that banks were able to transition between a universal and a retail banking business model but that path dependency limits the flexibility of changing from or toward an investment-banking model.

   Our research makes three new contributions to the extant literature: first, we analytically define 'bank business model' and add a theoretical basis compared to previous studies. Second, we use EU and US banks allowing us to study whether some business models are more frequent in one or another jurisdiction. Third, we are the first ones to back-test whether a statistically derived business model classification matches realized business models.

2. Theoretical background

   2.1. Business models in general

   Research on business models has gained momentum during the past years (Alboge et al. 2015, Dalby et al. 2014, Friis et al. 2015, Haubro et al. 2015, Larsen et al. 2014, Lueg et al. 2015, Lueg et al. 2014, Malmmose et al. 2014). Zott and Amit (2011) survey the literature and conclude that the term 'business model' is not commonly defined. 37% of the surveyed articles study business models without defining it, such as the entire literature on banking business models. 44% use their own definitions, and 19% re-use the definitions of previous papers. The poor definition is sharply contrasted by the extensive use of the term business model: since the mid-1990s, the term has been frequently used from the dot-com bubble to the financial crisis in 2008. Whenever an industry faces a profound structural change, the discussion and research around 'the business model' gains new momentum (Zott and Amit 2011). Examples of definitions include Magretta (2002), who defines business models as "... stories that explain how enterprises work". They state who the customer is, what each customer values, and how the business makes money. Similalry, Teece (2010:172) proposes that a business model is the "manner by which the enterprise delivers value to customers, entices customers to pay for value, and converts those payments to profit." With a slightly stronger focus on operations, Wirtz et al. (2010:274) state that a business model "reflects the operational and output system of a company, and as such captures the way the firm functions and creates value." Linder and Cantrell (2000) define business model as the "organisation's core logic for creating value". These definitions are, however, not related to organizational aspects of economic theory and hence lack predictive ability. Hence, we follow the RBV- and TCE-based definition of DaSilva and Trkman (2014:382), who propose that business models "represent a specific combination of resources which through transactions generate value for both customers and the organization."

   2.2. Business models in banking

   The general definition of a business model needs to be narrowed down to the context within an industry. It is only recently that 'bank business models' have seen a revival: before the financial crisis, banks are said to follow a new...