“The long-term relationship between enterprise risk management and bank performance: the missing link in Nigeria”

This study investigates the relationship between Enterprise Risk Management adoption and implementation, and the performance of banks using a sample of four out of the seven Strategically Important Banks (SIB) listed on the Nigerian Stock Exchange covering the period from 2005 q1 to 2015 q2. In this study, the authors determined a measure for Enterprise Risk Management (ERM) adoption or implementation (ERM index) us-ing an integrated Enterprise Risk Management measurement model for the banking sector suggested by Soliman and Mukhtar (2017). A time series Johansen’s cointegration test was used to obtain evidence of the long-term association between ERM and performance, while Vector Error Correction Model (VECM) analysis was performed to gather evidence of causality relationship between ERM and performance. Finally, Generalized Impulse Response Function was used to obtain evidence of how performance responds to the introduction of a shock on Enterprise Risk Management. This study makes significant contributions to the existing body of knowledge, as it yields the first Enterprise Risk Management-performance-based empirical results that indicate a long-term relationship, causation effects, in addition to responding to performance ERM.


INTRODUCTION
In recent years, Enterprise Risk Management (ERM) as a discipline has received unprecedented interest and international attention (Arena et al., 2010).This has resulted in many leading organizations adopting ERM as their risk management framework (Beasley et al., 2006).Empirical investigations into the global level of ERM adoption have indicated that an increasing number of firms are embracing ERM.For example, a survey conducted by Deloitte in 2008 on 151 firms in various regions: North America 56, South America 24, Europe 68 and others three firms, found a growing interest in ERM among firms with a majority of respondents; 64 percent in Europe and 62 percent in North America.Such results indicate that their interest in ERM was higher compared with the previous year to the current survey.The survey also found that 56 percent of the firms had implemented ERM programs for no more than two years before the year of the survey; thus signalling the growing trend in ERM adoption (Deloitte, 2008).
An important feature of an ERM programmed embarked upon by any firm is that it is designed for a medium to long-term benefit rather than the short-term.In this regard, one would expect ERM-performance based studies to be focused on determining the medium to long-term effects of ERM adoption and implementation.This is, however, not the case, as all the ERM-performance related studies conducted before this study have used data analysis techniques that do not capture the medium to long-term effects.We note that the study undertaken by Pagach and Warr (2010) attempted to investigate the effects of adopting ERM on a firm's long-term performance, but failed to use time series data analysis techniques that could provide evidence of the durable relationship between ERM and performance.To fill this major research gap, we adopted time series data analysis techniques such as cointegration, causality and impulse response tests to establish the presence or otherwise of any long-term, causal and impulse response relationships between ERM and performance.Our main objective of performing time series data analysis is to explain some important relationships among our variables, such as: the long-term relationship, causal relationship and impulse response effects.Such analysis is vital to our study because implementation of ERM by firms is a long process.Hence, the benefit of implementing ERM is expected to manifest in the medium and long term rather than in the short term (Pagach et al., 2010;Eckles et al., 2011).We note that, despite the vital medium to long-term features of ERM benefits, there is no study on ERM-performance relationship that is based on time series analysis aimed at establishing long-term, causal and impulse response relationships.The research question that remains unanswered in the ERM literature is whether there is a long-term relationship between ERM and performance.Therefore, our time series analysis is novel in this research area.The main contribution of this study is that, for the first time, it provides financial sector regulators and operators with an empirical evidence of the medium to long-term effects of ERM.The remainder of this paper is structured as follows.First, we present the conceptual framework followed by a summary of relevant and related literature on the empirical evidence of the relationship between ERM and performance.We then present the methodology, followed by an analysis of the empirical results.Finally, we conclude by summarizing our findings and provide a conclusion.

LITERATURE REVIEW
ERM is generally considered as a paradigm shift from 'silo-based' approach of managing risk to a holistic approach to managing risk.There is a theoretical expectation that, when firms adopt and/or implement an ERM program, they derive several benefits.These include keeping all their risks within a defined risk tolerance limits, and improvement in performance and market values, among other benefits (Pagach & Warr, 2010, 2011;Arena et al., 2010).Prior to this study, other empirical studies have attempted to determine whether the theoretical anticipated benefits of ERM are achieved.Indeed, some studies (see for example, The ERM-performance related studies cited above have used data analysis techniques that are not capable of determining the medium to long-term effects of ERM on the performance of firms, however, the study by Pagach and Warr (2010) attempted to investigate the effects of adopting ERM on firm's long-term performance.Because ERM programs are designed and implemented over a long period, the results of an ERM implementation are expected to manifest over a longterm horizon fully (Pagach et al., 2010;Eckles et al., 2011).Nonetheless, the literature on ERM has not paid the required attention to the long-term relationship between ERM implementation and performance.
According to The Casualty Actuarial Society (2003), ERM enables firms to monitor risks from all sources, thereby resulting in increasing the firms' short-and long-term value to stakeholders.Additionally, Pagach et al. (2010) and Eckles et al. (2011) have argued that a firm's investment in an ERM program takes a long time for the expected benefits to be realized.This implies a theoretical expectation of long-term relationship of ERM and performance.An empirical study conducted by Obalola et al. (2014), which investigated the effects of ERM adoption by Nigerian insurance firms, using data collected over a 10-year period, found evidence of continuous improvement in the performance of insurance firms that adopted ERM.Although the focus of that study was not to determine any long-term relationship between ERM and performance, the fact that 10-year data was used indicates that the results of the study cover a long-term horizon.In another empirical study by Pagach and Warr (2010), the effects of ERM adoption on long-term performance of firms were investigated but little impact of ERM adoption on the performance of firms was revealed.
One major theoretical expectation of ERM is that its adoption causes or results in better performance of the adopting firm.This is indicated in several definitions of ERM such as: Arena et al. (2010) and Hoyt et al. (2010).This is an implicit indication that ERM adoption is expected in theory to have a causality effect on firms' performance.Although to the best of our knowledge no empirical study has conducted a direct investigation into the causality relationship between ERM and performance, we note that certain empirical studies have adopted methodologies from which we can infer the examination of causality relationship.For example, the study by Hoyt et al. (2010), which used the announcement of Chief Risk Officer (CRO) appointment or presence of similar position in a firm to signal ERM adoption, found a positive and significant relationship between announcement of CRO appointment (a proxy for ERM) and performance.Conversely, Pagach and Warr (2010) used the announcement of CRO appointment as a signal to ERM adoption, but failed to find significant evidence that announcement of CRO appointment (a proxy for ERM) caused changes in performance measures.We argue here that, the fact that the reactions of performance measures were tested against one-off event (announcement of CRO appointment) implies an expectation that the one-off event could cause an effect on the selected performance measures.In other words, if a selected performance measure of a firm, such as share price, is expected to react simply because the firm has announced the appointment of CRO, then the reaction of the share price, in this case, can be said to have been caused by the announcement of CRO appointment (proxy for ERM).
Previous research studies were mainly focused on the short-term effects of ERM with little or no evidence given on the long-term relationship between ERM and bank performance.The main contribution of this study is that, for the first time, it provides empirical evidence of the medium to long-term effects of ERM on bank performance using various statistical techniques.

CONCEPTUAL FRAMEWORK AND RESEARCH HYPOTHESES
A critical foundation upon which this study rests is the fact that, theoretically, ERM adoption or implementation by firms is expected to translate into improved performance and value of firms.This is achieved by decreasing earnings and stock-price volatility, reducing external capital and increasing capital efficiency ( The conceptual framework for the model is presented in Figure 1.
The conceptual framework presented in Figure 1 indicates that ERM has short and long-term association with, and causation effects on, firm value.Moreover, firm value responds to shock on ERM.

METHODOLOGY
To facilitate our time series analysis, we selected an important category of banks in the Nigerian banking sector, which is the group of banks designated by the CBN as SIBs.Out of the total of eight banks that make up this list of SIBs in Nigeria, we eliminated one bank that is neither listed on the Nigerian Stock Exchange nor has an international authorization.The remaining seven SIBs were grouped by total assets as of December 31, 2014, and selected the top four banks regarding total assets.We argue that using a much smaller sample for our time series analysis will facilitate more indepth analysis that can be reasonably generalized to the entire population of similar banks.
The main independent variable in this study is ERM index while two dependent variables (i.e. return on average equity -ROAE and firm value -FV) were used.For the ERM index, we used an integrated model for measuring ERM implementation for the banking sector that was first suggested by Soliman and Mukhtar (2017) as a way of addressing the serious limitations associated with ERM measurements.We obtained the relevant quantitative and qualitative information as required by the integrated ERM models from the published quarterly and annual reports of the selected banks, which we applied to the relevant model to obtain the ERM index for the relevant periods for each bank.
For return on average equity, profit after tax divided by average equity was obtained for all the sampled banks over the period under study.With regard to the firm value, we used the Tobin's Q, which is considered the most commonly used measure of firm value in empirical risk management studies (Hoyt et

Empirical results and discussion
First, we performed ADF unit root tests to determine the suitability of our variables of interest (i.e.ERMI, ROAE and FV) for co-integration analysis.
To conclude that a variable is suitable for co-integration test, such variable must be non-stationary (i.e. has a unit root) at level and stationary (i.e.no unit root) at first difference.To achieve such an

ERM index
Performance measures The next step is to carry out Johansen's co-integration test using the optimal lag value of seven based on Akaike's Information Criterion.The test was run under the null hypothesis that 'there is no co-integration' and the alternative hypothesis whereby 'there is co-integration'.Both null and alternative hypotheses were related to levels of cointegration being test.
The results presented in Note: Lag Length: 0 (Automatic -based on SIC, max lag = 9).An asterisk indicates significant at 5% significance level.fects of ERM adoption on the long-term performance of firms but found little impact of ERM adoption on performance.
We then conducted the VEC Granger causality test using FV as the dependent variable with ERMI and ROAE as the independent variables.
From Table 3 C(1) p-value of 0.21 percent, which is less than 5 percent (see Table 4).The combined effects of the negative sign of D(FV) C(1) Coefficient and the D(ERMI) C(1) p-value, which is less than 5 percent, imply that there is a long-term causality relationship from ERMI and ROAE to FV.We can, therefore, conclude that, in the long run, our independent variables (ERMI and ROAE) cause our dependent variable FV.
With regard to establishing the possibility of a short-run relationship between our variables, the Wald test result of the ERMI coefficient indicates a Chi-square statistic of 27.7860 and a corresponding p-value of 0.02 percent (see Table 4).Since the corresponding p-value is less than 5 percent, we reject the null hypothesis (there is no short-run causality from the independent to the dependent variables) and accept the alternative hypothesis.Hence, we conclude that the coefficients of the independent variable are zero, meaning that there is short-run causality from the ERMI to FV.
For the second Wald test of this model (ROAE), the results indicate a Chi-square statistic of 19.8608 and a corresponding p-value of 0.59 percent, as presented in Table 4. Again, since the corresponding p-value is less than 5 percent, we reject the null hypothesis and accept the alternative hypothesis, thus concluding that there is short-run causality from ROAE to FV.
The overall result of our VECM is that it provides strong evidence of causality relationship from ERMI and ROAE to FV in the long term, alongside short-run individual causality relationship from ERMI to FV.Based on this evidence, we can, therefore, conclude that ERM has both long-and short-term causality rela- tionships with the performance of Nigerian banks as measured by FV.
Another important test performed as part of data analysis is the impulse response analysis (IRA) in which we examined the speed with which a 1 percent standard deviation shock in one variable affects the current, as well as future values of other variables over a specified period.In running our IRA, we introduced positive shock of one standard deviation of the variables in the model to see the reactions of the variables.
Because our data is gathered on a quarterly basis, we selected 42 periods, which is equivalent to the length of our time series data.Moreover, this length of time horizon provides us with a long-term view of the response of performance measures to an impulse on ERM.
The results of impulse response analyses as presented in Figure 2 yield interesting reactions of different variables to an introduction of one standard deviation positive shock to other variables.Although we value the importance of the response of each of the variables to an introduction of a positive shock on all other variables, we consider the response of ROAE and FV to a positive shock on ERMI to be more relevant to our study.This is because the focus of our study is to determine the effects if any, of ERMI on performance (FV and ROAE).From the impulse response graph, FV and ROAE responded differently to a positive shock on ERMI both within the range of positive values without assuming negative values.The response of FV to a positive shock on ERMI was relatively neutral at the beginning but later fluctuated mildly within the first five quarters, followed by more volatile fluctuations up to the tenth quarter before returning to mild fluctuation and finally assuming equilibrium after quarter twenty.This implies that if the ERMI of a firm receives a positive shock, it takes about three quarters for it to impact positively the FV of the firm with the impact increasing along quarters and gradually returning to equilibrium.Conversely, a positive shock on ERM receives an almost immediate response from ROAE, which fluctuates sharply within the first 10 quarters, assuming mildness in the fluctuation after quarter 10 and, thereafter, settling at equilibrium after quarter 20.This implies that, if the ERM of a firm receives a positive shock, ROAE responds immediately with the magnitude of response reducing over time until equilibrium is restored after year 20.
It is noteworthy that the two variables in this study (ROAE and FV) exhibit some similarities regarding their response to an introduction of one standard deviation positive shock on the main variable (ERMI).In the first place, both ROAE and FV responded by series of fluctuations but did not assume zero values within the 42 quarter time horizon.This is consistent with our expectation that ROAE and FV are positively related to ERM.Hence, a positive effect on ERM should, all other things being equal, result in a positive response from ROAE and FV.Though some downward trends were observed, we note that such downward trends were not sufficiently significant to push ROAE and FV to assume zero values; hence, maintaining their positive relationships with ERMI.The second major similarities exhibited by the response of ROAE and FV to a positive shock on ERM is the fact that equilibrium levels were restored after quarter 20, which is five year horizon, thus reflecting the long-term association between ERMI and performance of firms.
Relating our observed trend of responds of performance measures to a shock in ERM, we argue that it corroborates the findings of the positive relationship between ERM and performance of banks (both long-and short-term) as well as the causality relationship established between ERM and performance of banks.We also note that, in responding to a positive shock to ERM, our performance measures did not assume negative values.

CONCLUSION
In this study, we found significant empirical evidence that ERM does affect the performance of Nigerian banks in both the short and long terms.Moreover, the study has provided empirical evidence that ERM has a causation effect on the performance of Nigerian banks and evidence was obtained that the performance of Nigerians banks responds positively to a positive shock on ERM.The evidence of cointegration obtained implied that ERM implementation by banks has a long-term association with their performance.In other words, our cointegration results provided significant evidence of long-run association between ERM and performance of Nigerian banks.Furthermore, the study provided evidence of the causality relationship between ERM and performance measures; thereby implying that the adoption or implementation of an ERM program leads to improvement in the performance of Nigerian banks, both in the short and long terms.Furthermore, this study provides evidence of how the performance of Nigerian banks (represented by ROAE and FV) responds to the introduction of one standard deviation positive shock to ERM.The evidence obtained from our IRA indicates that the performance of Nigerian banks responds positively to the introduction of a positive shock on ERM with series of volatile upward and downward trends in the first five quarters, assuming more stable trends thereafter and finally returning to equilibrium in quarter 20.
Based on this study and considering the strength of the evidence obtained, we affirm the hypotheses for this study and conclude that the adoption and implementation of ERM by Nigerian banks are positively related to their performance in the long run.Moreover, the performance of Nigerian banks responds  positively to a positive shock on ERM.The relationship between ERM and bank performance provides evidence to policy makers on the significant role that ERM can play in enhancing performance.
We argue that, by making such a significant contribution to the literature on ERM and bank performance, we have set the pace for future researchers to consider using more comprehensive performance indicator models in measuring ERM implementation and the impact on performance, which will go a long way to improve the quality of research in this overlooked research area.

Figure 1 .
Figure 1.The conceptual framework for the study

Figure
Figure 2. Graphical result of generalized impulse response analysis The results of ADF unit root test, in Table1, for ERMI, ROAE and FV at level show that non-stationarity cannot be rejected at five percent significance level.Moreover, the ADF at first difference of ERM, ROAE and FV indicates that the null hypothesis of non-stationarity is rejected in all three cases at 5 percent significance level.

Table 2 .
Johansen co-integration test Note: An asterisk denotes rejection of the hypothesis at the 5% significance level; Trace test indicates 3 co-integrating equation(s) at the 5% significance level; Max-eigenvalue test indicates 3 co-integrating equation(s) at the 5% significance level.

Table 3 .
Granger causality test under VEC environment Banks and Bank Systems, Volume 13, Issue 1, 2018 negative sign of D(FV) C(1) Coefficient, which gives an initial indication of the possibility of long-run causality from ERMI and ROAE to FV.This is further corroborated by the D(ERMI)

Table 4 .
VECM with Firm Value (FV) as the dependent variable 2. Graphical result of generalized impulse response analysis