Research Framework and Empirical Models
3.5 Research Design
3.5.1 Regression Model
3.5.1.2 Models for Discipline Imposed by Bond Holders
104 coverage of maximum IDR 2 billion. However, the other two periods only have 2 quarters which is considered insufficient to get a reliable result. The current study, therefore, will not estimate a separate regression for these two observations.
Table 3.1 Deposit Coverage under the Limited Deposit Insurance Scheme
Time Period Coverage Number of Quarters
March 2006 – Sept 2006 Maximum IDR 5 billion 2 Quarters Sept 2006 – March 2007 Maximum IDR 1 billion 2 Quarters March 2007- Oct 2008 Maximum IDR 100 million 6 Quarters Oct 2008 – Dec 2011 Maximum IDR 2 billion 13 Quarters
Source: LPS Annual Report 2009
105 in the study by Flannery and Sorescu (1996), the present analysis employs the following specifications to estimate the determinants of bond yield spread:
YIELD_SP𝑖,𝑡 = 𝛼0+ Β1YIELD_SPI,T−1+ Β2 (BANK_FUNDAMENTAL)I,T−1+ Β3(MACRO_VARIABLE)I,T + Β4RATINGI,T+ Β5BANK_SIZEI,T+ Β6(BOND_SIZE)I,T+ Β7STA_BANKI,T+ Β8SUBDEBTI,T+ ΕIT
(4)
YIELD_SP i,t = the difference between bank bond yield and government risk free bonds
BANK_FUNDAMENTALi,t = vectors of bank idiosyncratic risks that represent the CAMEL ratios which consists of CAR, NPL, OPEX, NIM, and DER MACRO_VARIABLE,i,t = vectors of macroeconomic variables which consists of
GDP_RT, EXC_RT, and BI_RT
RATINGi,t = credit rating score issued by rating agencies BANK_SIZEi,t = bank total assets
BOND_SIZEi,t = principal amount of bond issued by a bank
STA_BANKi,t = dummy variable of government ownership (set = 1 for state bank, and 0 otherwise)
SUBDEBTi,t = dummy variable of type of bond (set = 1 for sub-debt, and 0 otherwise)
εi,t = random error terms
The dependent variable, YIELD_SPit, is the difference between bank bond yield and government risk free bonds at time t (in logarithmic form). The vector of bank fundamentals is included with a lag difference to account for the delay in the availability of financial statement information. The use of lagged variables is important as a means to capture, not only the monitoring power, but also the influence of market players on financial institutions (Bliss & Flannery, 2002). The following sub-section explains in more detail the justification for, and measurement of, these variables.
106 Bank Fundamental Variables
Bank fundamentals, as independent variables, are bank characteristics related to the bank’s financial condition, which are similar to those used in the CAMEL rating system of banks as mentioned in Section 3.4.1. In this study, the CAMEL financial ratio is utilized as an indicator of bank risk. This approach is widely used in previous studies, such as in Flannery and Sorescu (1996), Deyoung et al. (2001), and Evanoff and Wall (2001). The Indonesian banking regulator employs the CAMEL ratios to assess bank soundness, therefore these ratios convey relevant information for bond holders in relation to the financial health of banks (Bank Indonesia, 2012a). Independent variables used in the bond holder regression model are described below (the calculation methods of the variables are provided in page 98-101):
1. CAR: The capital adequacy variable is expected to have a negative effect on the yield spread (Evanoff & Wall, 2001) because this ratio measures the ability of banks to absorb a reasonable level of losses before becoming insolvent (Bank for International Settlements, 2006).
2. NPL: This study uses the lagged first difference of the non-performing loans ratio in the regression, as used in Wu and Bowe (2012). The coefficient of this variable measures the impact that changes in the quality of bank assets have on the dependent variable. The present analysis focuses on the impact of the periodic flow of non-performing loans, rather than on the accumulated stock of non-performing loans, as a determinant of the bond yield. The larger the non-performing loan ratio, the greater the likelihood of a loss, and the larger the required bond spread by investors. Hence, an increase in the NPL variable is expected to have a positive effect on the yield spread (Jagtiani et al., 2002;
Wang et al., 2010).
3. OPEX: The management efficiency variable is expected to a have a positive effect on the yield spread because banks that are less efficient are expected to face higher levels of expenditure (Martinez-Peria & Schmukler, 2001).
4. ROA: The earning quality variable is expected to a have negative correlation with the yield spread because it measures the efficiency of a bank’s use of its assets to generate net earnings (Avery et al., 1988). Therefore, an increase in bank profitability would reduce the probability of default, consequently lowering bond yield spreads (Jagtiani et al., 2002).
107 5. NIM: Similar to the ROA, the net interest margin coefficient is expected to have a
negative correlation with the yield spread because an increase in the NIM ratio would increase bank profitability and, consequently, reduce the probability of default.
6. DER: This study uses a debt to equity ratio to represent liquidity risks36 as well as a leverage indicator. This ratio represents the relative proportion of shareholder's equity and debt used to finance a company's assets. The higher ratio indicates higher risk since banks are more dependent on debt to finance its assets. Therefore, this leverage coefficient is expected to have a positive correlation on yield spread (Mendonça & Villela Loures, 2009).
7. BOND_SIZE: Other aspects of liquidity risk are related to the uncertainty with respect to the time and cost of transactions in selling bonds in a bond market at any given time.
Therefore, investors normally demand an adequate yield spread as compensation for high transaction costs, especially on less liquid bonds (Crabbe & Turner, 1995). However, this liquidity risk of bonds is quite difficult to determine since this liquidity is not directly observable at the market. Adapting the model used by Menz (2010), Sironi (2003) and Avery et al. (1988), this study uses the amount or size of bond issuance as a proxy for liquidity risk. The existing literature suggests that bonds with greater par value or principal amount are normally more liquid than smaller bonds. The expectation is, therefore, a negative relationship between bond par value and yield spreads.
8. RATING: Previous studies have empirically confirmed the influence of credit rating issued by rating agencies on bond yield spreads (Avery et al., 1988; Flannery & Sorescu, 1996; Jagtiani et al., 2002; Mendonça & Villela Loures, 2009; Morgan & Stiroh, 2001;
Sironi, 2003). This rating represents the probability of default in securities, thus the higher probability of default is expected to increase the yield of bonds required by investors.
Following Sironi (2003), Jagtiani et al. (2002), and Avery et al. (1988) the ratings are transformed into cardinal numbers, with higher ratings being represented by smaller numbers. For example, the AAA rating as the highest quality of bonds has the scale of 1, whereas the BBB- rating as the lowest investment grade has the scale of 10. An increase in the rating scale (e.g. from BBB- (scale of 10) to BBB (scale of 9)) should then lead to a lower credit spread. Thus, a negative correlation between the yield spread and the bond rating is expected.
36 The published quarterly financial statements of the Indonesian banks use LDR as a liquidity indicator.
However, for assessing the bond yield, the literature commonly employs the DER rather than the LDR ratio.
108 9. BANK_SIZE: Bank total assets, as a proxy for bank size, may be indicative either of greater diversification benefits or of greater liquidity of the bank’s bonds (Flannery &
Sorescu, 1996). Moreover, a larger institution may be perceived as TBTF, and seen as having a lower probability of default by conservative investors (Jagtiani et al., 2002;
Levy-Yeyati et al., 2004a; Wang et al., 2010). Therefore, the behaviour of bond holders toward large banks is expected to be more risk neutral since these banks are most likely to be rescued by a government in the event of defaults. Consequently, this study expects an inverse relationship between bank size and bond yield spread (Flannery & Sorescu, 1996;
Jagtiani et al., 2002; Levy-Yeyati et al., 2004a; Wang et al., 2010).
10. STA_BANK: The previous empirical evidence suggests that the government ownership of banks might affect the extent of market discipline (Levy-Yeyati et al., 2004a; Sironi, 2003). Since these government banks are most likely to be bailed out, the behaviour of bond holders toward state banks is expected to be more risk neutral. A dummy variable, STA_BANK, is included in the model to test the hypothesis of bond holder sensitivity to the risk of default of state banks.
Macroeconomic Variables
Macroeconomic factors play a significant role in determining bond risk premium, as discussed in Chapter 2. The empirical evidence maintains that a bond’s rate of return could not be completely explained by the risks of an individual institution, therefore the use of macroeconomic indicators as control variables becomes necessary (Sironi, 2003). Following Mendonça and Villela Loures (2009), for this purpose three control variables are considered:
the growth of gross domestic products (GDP_RT), exchange rate (EXC_RT), and the central bank’s interest rate (BI_RT)37. Previous studies, particularly in emerging markets, reveal mixed results about the relationship between these variables and the yield spreads. Therefore, as Indonesia faced volatile economic conditions during the study period, the present study did not predict a particular sign on the correlations between the systemic risk factors and the macroeconomic variables (Hadad et al., 2011).
37 The central bank (BI) interest rate (BI Rate) is the policy rate reflecting the monetary policy stance adopted by BI and announced to the public. The rate is issued by the Board of Governors. It is implemented in the BI monetary operations conducted by means of liquidity management on the money market to achieve the monetary policy operational target.
109 3.5.1.3 Models for Discipline Imposed by Equity Holders
The form of market discipline in capital markets is demonstrated by the ability of equity holders to evaluate the financial condition of a bank (monitoring phase) and by the responsiveness of the bank management to investors’ stock-return assessment (influencing phase) (Bliss & Flannery, 2002). Shareholder ability to assess the riskiness of publicly-traded banks is indicated by the fluctuation of the bank share prices as shareholders react to the announcements of bank financial indicators such as CAMEL ratios (Bliss & Flannery, 2002;
Caner et al., 2012). Therefore, the bank risk monitoring behaviour of equity holders exists if there is a significant relationship between equity returns and risk measures obtained from the financial statements of banks (Bliss & Flannery, 2002).
The model to measure discipline by shareholders is constructed upon the growth of share price or equity return as a dependent variable. The equity returns are calculated by taking the difference between the stock return of a bank at the closing date and the return of a short-term investment alternative (Bliss & Flannery, 2002; Caner et al., 2012). In the present study, the yield of the BI certificate (SBI)38 is used as an available investment alternative for investors in Indonesia. Bank stock prices and returns fluctuate according to the risks taken by banks as signaled by the CAMEL ratios after controlling for other bank specific variables and macroeconomic conditions, amongst other factors. With an intention to make a comparison of the empirical results for each discipline agent, the monitoring model of equity holders is designed to be similar to the models for evaluating market discipline by depositors and bond holders. The following is the model used:
EQT_RTI,T= Α0+ Β1 (BANK_FUNDAMENTAL)I,T−1+ Β2(MACRO_VARIABLE)I,T + Β3BANK_SIZEI,T+ Β4STA_BANKI,T+ ΕIT
(5)
EQT_RT i,t = the bank equity return is represented as the growth of the bank share price
BANK_FUNDAMENTALi,t = vector of bank idiosyncratic risks that represent the CAMEL ratios which consists of CAR, NPL, OPEX, NIM, and LDR
38 The BI certificate (SBI) is a Rupiah-denominated security issued by BI in recognition of short-term debt and comprises one of the instruments used in Open Market Operations. The term of SBIs is at least 1 month and no more than 12 months. SBI may be held by banks and other parties as stipulated by BI, and are negotiable. SBI may be purchased on the primary market and traded on the secondary market under repurchase agreements (repo) or in outright purchase/sale.
110 MACRO_VARIABLE,i,t = vector of macroeconomic variables which consists of
GDP_RT, EXC_RT, and BI_RT BANK_SIZEi,t = bank total assets
STA_BANKi,t = dummy variable of government ownership; 1 for state bank, and 0 otherwise
εi,t = random error terms
In line with the regression models for depositors and debt holders, the vector of bank fundamentals for this model is included with a lag difference to account for the fact that the financial statement information is available to the public only with delay. The following sub- section explains in more detail the justification for, and measurement of, these variables.
Bank Fundamental Variables
Bank fundamentals are represented by CAMEL financial ratios, similar to the approach used to investigate discipline by depositors and bond holders in Sections 3.5.1.1 and 3.5.1.2. The use of CAMEL ratios as indicators of bank risk have previously been used in, for example, Bliss and Flannery (2002) and Caner et al. (2012). Independent variables used in the shareholder regression model are described in detail as follows (the calculation methods of the variables are provided in page 98-101):
1. CAR: The capital adequacy variable represents the level of capital to absorb losses before becoming insolvent. Therefore, this variable is expected to have a positive effect on equity returns (Beighley et al., 1975; Berger, Davies, & Flannery, 2000; Caner et al., 2012;
Shome et al., 1986).
2. NPL: The non-performing loans ratio is used as an indicator for asset quality. An increase in the NPL ratio indicates a low return for the bank. Therefore, this variable is expected to have a negative effect on equity returns (Berger et al., 2000; Caner et al., 2012).
3. OPEX: The management efficiency variable is measured using the OPEX ratio. This ratio is expected to a have a negative correlation with equity returns (Caner et al., 2012) because an increase in the OPEX ratio might reduce the wealth creation ability of a bank.
111 4. ROA: The earning quality variable is closely associated with the value maximization of
shareholders and shares are priced on the basis of bank predicted future performance (Berger et al., 2000). Hence, this ratio is predicted to a have positive correlation with equity returns (Beighley et al., 1975; Berger et al., 2000).
5. NIM: The net interest margin coefficient is often employed to measure bank profitability, similar to the ROA ratio. NIM is expected to have a positive correlation with equity returns because an increase in the NIM ratio would decrease the ability of banks to generate wealth.
6. LDR: Similar to the regression model for depositors, proxy for liquidity risk is assessed by the loan to deposit ratio (LDR). Generally, banks with a large volume of liquid assets are perceived to be safer, therefore the liquidity coefficient is expected to have a positive correlation with equity returns.
7. BANK_SIZE: Bank total assets, used as a proxy for bank size, are included as a control variable to evaluate the influence of the TBTF doctrine on shareholder behaviour (Beighley et al., 1975; Berger et al., 2000).
8. STA_BANK: Previous studies of market discipline suggest that government ownership of banks might affect the extent of market discipline (Levy-Yeyati et al., 2004a; Sironi, 2003). Similar to other models, STA_BANK is included as a dummy variable in order to test the sensitivity of shareholders to the risk profile of state banks.
Macroeconomic Variables
Macroeconomics factors play significant roles in determining equity returns, as discussed in Chapter 2. The empirical evidence supports the view that equity returns could not be completely explained by the risks of an individual institution, which increases the importance of using macroeconomic indicators as control variables (Sironi, 2003). The present study incorporates three control variables: gross domestic products (GDP_RT), exchange rate (EXC_RT), and the central bank’s interest rate (BI_RT). Previous studies, particularly among emerging markets, reveal mixed results on the relationship between these variables and the yield spreads. No particular a priori direction of relationship is postulated.
112 3.5.2 Statistical Methods
Prior studies on the empirical analysis of market discipline imposed by market participants have commonly employed static panel data, with an extensive use of fixed or random effect panel models. Martinez-Peria and Schmukler (2001), Ghosh and Das (2003), Nier and Baumann (2006), for example, used static panel data to investigate discipline by depositors while Flannery and Sorescu (1996), Mendonça and Villela Loures (2009), Deyoung et al.
(2001), Menz (2010), and Hwang and Min (2013) used this procedure to measure the sensitivity of bond holders on bank fundamentals.
Recent studies, however, suggest the employment of the generalized method of moments (GMM) estimator developed for dynamic models of panel data by Arellano and Bover (1995).
This approach has been adopted by studies, such as those of Cubillas et al. (2012), Wu and Bowe (2012), Hadad et al. (2011), and Karas et al. (2010). The dynamic relationships are characterized by the presence of a lagged dependent variable among the regressors (Baltagi, 2008). The GMM has several inherent advantages over panel models because this methodology is specifically designed to address three relevant econometric issues: (1) the first difference specification of the GMM models potentially reduces any inconsistency in the estimates arising from unobservable heterogeneity (or unobservable bank-specific effects) across banking institutions; (2) the autoregressive process in the data regarding the behaviour of the dependent variables (i.e. the growth rate of deposits is likely to exhibit some degree of persistency, resulting in autocorrelation. This can be accommodated through the inclusion of the lagged dependent variable on the right-hand side of the estimated equation); and (3) the likely endogeneity of the explanatory variables. The panel estimator controls for this potential endogeneity by using instruments based on lagged values of the explanatory variables (Martinez-Peria & Schmukler, 2001). Taken together, the GMM procedures are expected to yield more consistent estimators based on the ability to control for potential endogeneity, unobserved heterogeneity, and persistence in the dependent variable. Based on the arguments, to test the hypotheses, this study employs the GMM estimator.