Measuring and Assessing Risk Culture

Abstract

Weak Risk Culture was a driver of banks’ failures after the financial crisis of 2008 (PCBS, 2013). Regulators (FSB, 2014) and practitioners (Deloitte Australia, 2012) developed frameworks to analyse and enforce RC in financial institutions; new pillar III discipline is pushing banks to develop a strong RC (BCBS, 2015). No empirical evidence exists in literature that link RC and banks’ stability. The following chapter tries to fill this gap, it uses Financial Stability Board’s framework (Financial Stability Board, 2014) of Sound Risk Culture (SRC) and extracts the vocabulary associated with a strong RC. Then, it applies quantitative text analysis (QTA) on banks’ disclosure to build a SRC Index. Finally, it analyses the correlation between the index and the Z-score of the banks.

1 Introduction

A weak risk culture was one of the drivers of banking crisis of 2008 (Parliamentary Commission on Banking Standards 2013). The interest of risk culture is now common to practitioners, regulators and academics. For instance, HSBC states that “Establishing and maintaining a strong link culture is of fundamental importance in ensuring the sustainable success of an organization and to the reestablishment of trust of financial institutions and the banking sector’’. It is therefore not surprising that banking regulators (Financial Stability Board 2014) and practitioners (Deloitte Australia 2012; Institute of Risk Management 2012) developed frameworks to analyze and enhance risk culture in financial institutions: e.g. a new pillar III discipline is pushing banks to develop a strong risk culture (BCBS 2015). The New York Fed Governor, Dudley (2015) notes “In the last year, we have seen emerging approaches to supervision that aim to address culture, conduct and governance. These methods are being developed in a number of jurisdictions”.

Although there is a great interest and discussion on how to improve bank risk culture, there is surprisingly no empirical evidence about the bank risk culture and how banks’ risk culture is related to stability. This chapter aims to fill this gap: specifically, we provide evidence that among the four components for a sound risk culture, identified by the FSB, Tone-from-the-top is the most significant for banks’ stability. Institutions with a higher level of intervention of the Board and Senior Management show, on average, a higher solidity between 2004 and 2014, after controlling for dimension, impaired loans, annual average sector’s performance and average bank’s performance throw considered period.

In this chapter, we aim to provide empirical evidence about the risk culture of European banks. For this purpose, we measure banks’ risk culture by using a text analysis technique build on the Financial Stability Board’s framework (Financial Stability Board 2014) of sound risk culture. Specifically, we identify a set of words (labelled “bag of words”) capturing the bank’s risk culture orientation based on the Financial Stability Board’s 2014 framework and we apply a quantitative text analysis (QTA) on banks’ public disclosure to build an index, labelled as “Sound Risk Culture Indicator” (SCRI). In the final step, we measure the correlation between our SCRI index and the Z-score of the banks. Our sample includes 147 of the major European banks between 2004 and 2014.

This chapter is organized as follows. Section 7.2 presents a brief discussion on methods used in risk culture measurement and the advantages to use QTA. Then, Sect. 7.3 illustrates SRCI’s computation, Sect. 7.4 describes the sample and Sect. 7.5 reports the analysis and the results. Section 7.6 concludes the chapter.

2 How to Measure Banks’ Risk Culture

The first necessary step to measure bank’s risk culture is to define the concept of culture. In our framework, we focus on the organizational culture definition, i.e. “culture is people’s beliefs, values and attributes” (Schein 2010). “Beliefs and values” are the core (deepest) levels of culture: they can be observed only from inside companies by using interviews, ethnographic studies or questionnaire. Conversely, “Attributes” is the most visible level of culture and it refers to any behaviour that is observable from outside the organization (e.g. disclosure, buildings, history, performance, risk-taking, etc.): companies’ attribute can be analyzed by indirect methodologies (as e.g. a text analysis).

There are various approaches to measure firm’s culture that would enable a researcher to measure companies’ beliefs, values and attributes and each measurement approach has its pros and cons (e.g. Schneider 2000). As such, the choice of the technique to measure corporate culture depends on the aims of the analysis. Qualitative methodologies (as interviews) can provide researchers with a deeper understanding of firms’ culture, but are very time-consuming. First, we need to recognize that values disclosed by firms are not useful to evaluate culture (Guiso et al. 2015). Generally, practitioners and consultants use such methodologies to interview board members, CEOs and top management with the aims of evaluating the main challenges in risk taking, and then they evaluate existing culture in the company with questionnaires to employees (Deloitte Australia 2012). This process applies to (one or few) companies to measure culture beliefs and attributes, it requires time to be set up and results cannot be safely compared across competitors. In order to measure corporate culture for a large number of companies, the most effective approach is to focus on companies’ attributes: although these do not provide a direct measure of company’s culture, “attributes” provide analysts with some visible effects of companies’ risk culture. Attributes are easier to be identified and evaluated in an objective manner using quantitative methods.

In this chapter, we focus on largest European banks and so we rely on text analysis to measure their risk culture. Specifically, we use a QTA, i.e. a method widely used in past papers dealing with corporate culture (Fiordelisi and Ricci 2014; Carretta et al. 2010, 2015a, b). The QTA has various advantages: it is objective, replicable and it is able to measure soft characteristics such as attention to client, risk sentiment or Tone-from-the-top . For our purposes, Richter (2014) is an interesting example of the QTA usage since it builds a quantitative measure of risk culture in financial institutions . Specifically, Richter (2014) develops a risk culture Intensity Index using QTA on the annual reports of the 30 biggest banks in Germany, between 2008 and 2011 and he provided evidence that risk culture changes over time.

3 Estimation Procedure

The underline assumption behind the QTA use to estimate the bank risk culture is that the bank’s culture is mirrored in the words (vocabulary) used by this bank. As such, our estimation procedure is based on the following four steps: first, we need to identify some “items” of a sound risk culture. Second, we need to define a “bag of words” capturing each of the items defined in the previous step. Third, we need to identify appropriate company’s documents where we believe the risk culture is mirrored; and, in final, we can run a QTA and estimate a score capturing the risk culture.

Regarding our first step, we identify the attributes of a risk culture based on the framework presented by Financial Stability Board (2014) “Guidance on supervisory interaction with financial institutions on risk culture”, which includes contributions given by Senior Supervisors Group (2009), Group Trinity, KPMG, McKinsey and IIF. In this framework, FSB identifies four minimum characteristics of a good risk culture: Tone-from-the-top (TFT), Accountability (ACC), Effective and Challenging Communication (COM) and Incentives (INC).

The “Tone-from-the-top (TFT)” attribute suggests that Board and Senior Management have the main responsibility to develop the RC within the bank, e.g. by supporting the implementation of Risk Appetite at all organizational levels (FSB 2014). The second attribute, labelled as “Accountability (ACC)” suggests that a company is able to take into account its risk-taking only if it has the know-how to recognize risks and it has escalation processes (as whisper blowing) to report treats. The third risk culture attribute is labelled as “Effective Communication and Challenge”: specifically, a careful risk management requires not only an intense data flows but also an open communication culture inside the organization: e.g. the top management must encourage alternative views and pay attention to the risk management’s advisory. In final, a company needs a system of “Incentives” (INC) in the form of rewards and penalties related to its risk indicators at all levels (from the CEO to loan officers), and these incentives should be not only monetary but also related to training, job rotation and successions planning to develop risk culture (Bianchi and Carretta 2016).

In the second step, we identify the bag of words. This step is based on Financial Stability Board’s framework of sound risk culture (Financial Stability Board 2014). In Table 7.1, we report a summary of the indicators that the FSB identifies for each of the four attribute (above discussed) and we collected the words from the FSB’s description. Since most of these words would be too generic taken stand-alone, we grouped two and three words where these together gives a well-defined and unique meaning to a sentence. For example, the word “board” and “lead” stand-alone would be not sufficient to identify the leadership attitude of the board (ID 4.1.a in Table 7.1), but these words taken together would show a guidance role of the board. The combination of the words selected (labelled as “Extracted Sentences”) are reported in Table 7.1, where each coma-separated term in the first group is searched with each term of the second group. These combinations of words are thereafter labelled as “Groups-Meaning-Units” (GMUs).

Table 7.1 Examples of terms extracted by FSB’s framework

Since some GMUs reported in Table 7.1 are associated to more than one indicator, we deleted each GMU that refers to more than one FSB’s items. The subclasses (SCs) reported in Table 7.1 are obtained aggregating GMUs with the same first group’s terms or MUs with just one word in each C. In Table 7.2, we show the outcome of this process.

Table 7.2 Grouped original MUs by characteristics and sub-characteristics

Each extracted MU expresses a certain meaning and it is used to look for this meaning in the text. One limitation is that the same concept may be stated in different ways, not just with the words included in the MU. To face this limit, the method includes new MUs, these are composed by synonymous of each term taken from the Thesaurus Collins Dictionary. This results in 9361 MUs considered.

We count how many time MUs appear in the banks’ disclosure. We divide the text into sentences (consecutive words contained between a blank line and a dot or between two dots) and add one to a MU-specific count for each sentence containing the MU.

In the third step, we run the QTA by using the NLTK Python package. Specifically, we count a term irrespectively whether it appears as adverb, adjectives, verb or noun (considering not the term itself but its root without the suffix). Table 7.3 shows the final list of MUs considered. SC from 1 to 5 contains MUs composed by words extracted from TFT’s indicators and their synonymous, SC from 6 to 8 contains the ones from COM’s indicators, SC from 9 to 14 the ones from COM’s indicators and the rest refers to INC’s indicators. Each Group 1’s term is aggregated with each term of Group 2 and 3 to create the MUs.

Table 7.3 Final MUs’ list

Once defined MUs, we defined the Sound Risk Culture Indicator (SRCI) as:

$${\text{SRCI}}_{it} = {\text{TFT}}_{it} + {\text{ACC}}_{it} + {\text{COM}}_{it} + {\text{INC}}_{it}$$

(7.1)

$${\text{TFT}}_{it} = \frac{{N_{it}^{\text{TFT}} }}{{{\text{tw}}_{it} }}$$

(7.2)

$${\text{ACC}}_{it} = \frac{{N_{it}^{\text{ACC}} }}{{{\text{tw}}_{it} }}$$

(7.3)

$${\text{COM}}_{it} = \frac{{N_{it}^{\text{COM}} }}{{{\text{tw}}_{it} }}$$

(7.4)

$${\text{INC}}_{it} = \frac{{N_{it}^{\text{INC}} }}{{{\text{tw}}_{it} }}$$

(7.5)

where ${\text{TFT}}_{it} ,{\text{ACC}}_{it} ,{\text{COM}}_{it} \,{\text{and INC}}_{it}$ are, respectively, the indicators for Tone-from-the-top , Accountability , Communication and Incentives for the bank i at time t. $N_{it}^{\text{TFT}}$, $N_{it}^{\text{ACC}}$, $N_{it}^{\text{COM}}$ and $N_{it}^{\text{INC}}$ are the total number of sentences containing a MUs associated with the correspondent C. ${\text{tw}}_{it}$ is the total number of words in the bank- i’s disclosure at time t. Dividing by total number of words corrects indicator for the distortions due to a longer or shorter bank’s disclosure.

4 Data

Our QTA focuses on annual and Pillar 3 reports of the largest banks (both listed and non-listed) in the eurozone and the UK between 2004 and 2004. Listed banks have been identified by using Datastream, while non-listed banks were selected looking at the ECB’s register of Significant Supervised Entities (SSE). As such, we identified 247 banks, but we have to limit the analysis to 147 banks that have (at least) one report in the English language.

In Tables 7.4, 7.5 and 7.6, we report various descriptive statistics of our sample. Looking at the number of banks, most are from Italy, Germany, France and the UK (67% of the Eurozone Bank’s total asset ).

Table 7.4 Sample composition: number of banks by country

Table 7.5 Sample composition: number of banks by year

Table 7.6 Sample composition: number of listed and non-listed banks

Figure 7.1 shows the mean SRCI and mean tw’s growth rates per year, mean tw increases in the whole period suggesting an increasing attention for disclosure’s transparency (perhaps, due to a stricter regulation over time). As shown in Table 7.7, SRCI and the four components have long tails and, consequently, we winsorize these variables at the 10% level.

Fig. 7.1

Source Authors’ Elaboration

Mean tw and SRCI’ growth rate per year.

Table 7.7 Descriptive statistics

Table 7.8 shows the correlation between the four components of SRCI and other variables, such as total assets and impaired loans on gross loans. Interestingly, we do not find a very high correlation among the four components: overall, this suggests that the four components really measure different aspects of bank risk culture. A higher TFT is positively associated with bank’s dimension and negatively associated with impaired loans.

Table 7.8 Correlation table

5 The Relationship Between Risk Culture and Stability

To analyze the relationship between the SRCI four components and bank stability, we run the following OLS model:

$$\begin{aligned} Z_{it} & = \beta_{1} {\text{TFT}}_{{i\left( {t - 1} \right)}} + \beta_{2} {\text{ACC}}_{{i\left( {t - 1} \right)}} + \beta_{3} {\text{COM}}_{{i\left( {t - 1} \right)}} + + \beta_{4} {\text{INC}}_{i(t - 1) } \\ & \quad + \omega_{1} \ln \left( {\frac{{{\text{IMP}}_{it} }}{{{\text{TL}}_{i(t - 1)} }}} \right) + \omega_{2} \ln \left( {{\text{TA}}_{i(t - 1)} } \right)\alpha_{t} + \varphi_{j} + \epsilon_{it} \\ \end{aligned}$$

(7.6)

where $Z_{it}$ is measure the bank i’s stability (i.e. the ratio between the sum of bank ROA and equity ratio and the ROA standard deviation, i.e. $Z_{it} = \frac{{{\text{ROA}}_{it} + \frac{{{\text{E}}_{it} }}{{{\text{TA}}_{it} }}}}{{{\text{sd}}({\text{ROA}}_{it} )}}$); $\ln \left( {{\text{TA}}_{it} } \right)$ is the one year lag of natural logarithm of the total asset ; $\ln \left( {{\text{IMP}}_{it} /{\text{TL}}_{it} } \right)$ is the one year lag of natural logarithm of impaired loans on total loans. $\varphi_{ j}$ and $\alpha_{t}$ the country- and time-fixed effects. We included the lag values of four components because we hypostasize that risk culture–stability relationship is not simultaneous.

As shown in Table 7.9, the estimated coefficients for TFT and ACC are statistically significant, respectively, at 5 and 10% confidence level: TFT displays a positive link with bank’s stability, while ACC shows a negative link. Our results are consistent with past studies that suggested the primary role of Tone-from-the-top in determining institution’s risk culture. Specifically, the FSB (2014) stress that Board and Senior Management have the main responsibility in setting risk culture and Power et al. (2013), after interviews with different financial institutions , recognize that the development of a good TFT is the common element of all risk culture changing programs set after the crisis.

Table 7.9 The relationship between risk culture and banks’ stability

6 Conclusion

A weak risk culture was one of the drivers of banking crisis of 2008 (Parliamentary Commission on Banking Standards 2013) and there is great attention toward risk culture among to practitioners, regulators and academics. Surprisingly, there is no empirical evidence about the relationship between bank risk culture and stability. Our chapter aims fill this gap: specifically, we focus on the FSB framework and we provide evidence that the Tone-from-the-top feature is the most significant component of the risk culture and this is associated to a greater banks’ stability. Furthermore, banks with a higher level of intervention of the Board and Senior Management show, on average, a higher solidity between 2004 and 2014, after controlling for dimension, impaired loans, annual average sector’s performance and average bank’s performance throw considered period.

Notes

• 1. For example in 4.1.a is searched “board” AND “lead” OR “board” AND “example” OR “management” AND “lead” OR “management” AND “example” and so on.

• 2. Subclass has no particular meaning, they are used just to simplify text analysis and are obtained aggregating MUs within the same category that have same first or second word inside.

• 3. http://www.collinsdictionary.com/.

• 4. Source: aggregated balance sheet of euro area monetary financial institutions, excluding the Eurosystem August 2016.