Encuestas realizadas a los Padres de familia

To test the capital management hypothesis and the hypotheses stated in section 6.4, this study uses a Tobit fixed-effects panel regression. The Tobit panel regression model is used because the dependent variable ‘regulatory provisions’ is strictly non-negative. The Tobit model overcomes the problem with OLS being inconsistent as a result of provisions being bounded at zero. Using a latent variable censored at zero it can be shown that the Tobit estimator retains consistency. Results show no difference in the sign and significance of variable coefficients under a fixed-effect panel specification. The general model specification is set up as follows:

y_𝐢∗_,𝐭 =𝛂_𝐢+𝛃_𝟏CRISK_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛆_𝐢,𝐭 (6.7)

𝐏𝐑𝐎𝐕𝒊,𝒕 = �

y_𝐢∗_,𝐭 𝐢𝐟 y_𝐢∗_,𝐭> 𝟎

𝟎 𝐢𝐟 y_𝐢∗_,𝐭≤ 𝟎 (6.8) y_i∗_,t = Latent variable taking the value of PROV_𝑖,𝑡, and zero if negative.

αi = Panel fixed effects intercept term.

PROV𝑖,𝑡 = ratio of provisions to outstanding loans. Either, Specific provisions (SPROVi,t) the

General Reserve for Credit Losses (GRCLi,t) or Total Provisions (TOTPROVi,t), depending

on the model. All regulatory provisions variables are included as ratios to Total Loans. CRISK_i,t = Credit risk proxy as specified in section 6.3 (IMP_i,t, IMPPD_i,t)

CRBP_i,t = The regulatory capital ratio in excess of the minimum regulatory requirement and before provisions and taxes, as either Tier 1 capital ratio before provisions and taxes (T1BPi,t)

6.6.1. Full Sample Model

The first set of models examines the whole sample of Australian banks, both standardised and IRB banks operating across the entire sample from March 2004 to December 2011 during Basel I and Basel II periods. The model tests the capital management hypothesis across the sample. The model also tests for the presence of earnings management, in line with hypothesis 9. The models are Tobit regressions, specified as follows;

SPROV_𝐢,𝐭 =𝛂_𝐢+𝛃_𝟏CRISK_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛃_𝟑EBTP_𝐢,𝐭+𝛆_𝐢,𝐭 (6.9)

GRCL_𝐢,𝐭 =𝛂_𝐢+𝛃_𝟏CRISK_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛃_𝟑EBTP_𝐢,𝐭+𝛆_𝐢,𝐭 (6.10)

TOTPROV_𝐢,𝐭= 𝛂_𝐢+𝛃_𝟏CRISK_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛃_𝟑EBTP_𝐢,𝐭+𝛆_𝐢,𝐭 (6.11)

6.6.2. Standardised Bank Model

The second set of models examines the sample of Australian banks reporting under the standardised approach to Basel II throughout the full sample period March 2004 to December 2011. The models test the capital management hypothesis in line with hypotheses 1-3. The models are Tobit regressions specified as follows;

SPROV_𝐢,𝐭 =𝛂_𝐢+𝛃_𝟏CRISK_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛆_𝐢,𝐭 (6.12)

GRCL_𝐢,𝐭 =𝛂_𝐢+𝛃_𝟏CRISK_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛆_𝐢,𝐭 (6.13)

6.6.3. IRB Bank Models

The third set of models examines the sample of Australian banks reporting under the IRB approach to Basel II throughout the full sample period March 2004 to December 2011. The models test the capital management hypothesis in line with hypotheses 4-6. The models are Tobit regressions specified as follows;

SPROV_𝐢,𝐭= 𝛂_𝐢+𝛃_𝟏CRISK_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛃_𝟑BASELII×CRBP_𝐢,𝐭+𝛃_𝟒BASELII+𝛆_𝐢,𝐭 (6.15)

GRCL_𝐢,𝐭= 𝛂_𝐢+𝛃_𝟏CRISK_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛃_𝟑BASELII×CRBP_𝐢,𝐭+𝛃_𝟒BASELII+𝛆_𝐢,𝐭 (6.16)

TOTPROV_𝐢,𝐭= 𝛂_𝐢+𝛃_𝟏CRISK_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛃_𝟑BASELII×CRBP_𝐢,𝐭+𝛃_𝟒BASELII+

𝛆𝐢,𝐭 (6.17)

Where:

BASELII = dummy variable taking the value of 1 where the bank is reporting under Basel II, and zero otherwise.

BASELII×CRBP_i,t = interaction effect between Basel II and the Regulatory Capital Ratio (before provisions and taxes), interpreted as the additional effect on the relationship between CRBPi,t and PROVi,t that occurs as a result of the change to Basel II IRB reporting.

6.6.4. Eligible Provisions Model

The fourth set of models tests hypotheses 3 and 4. The sample is for all IRB banks reporting eligible provisions (EL) during the Basel II period, from March 2008 to December 2011. The three models separately examine eligible provisions on defaulted exposures, eligible

provisions on non-defaulted exposures and total eligible provisions. The models are Tobit regressions, specified as follows;

EP-Defaulted_𝐢,𝐭 =𝛂_𝐢+𝛃_𝟏CRISK_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛆_𝐢,𝐭 (6.18)

EP-Non Defaulted_𝐢,𝐭 =𝛂_𝐢+𝛃_𝟏CRISK_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛆_𝐢,𝐭 (6.19)

EP_𝐢,𝐭 =𝛂_𝐢+𝛃_𝟏CRISK_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛆_𝐢,𝐭 (6.20)

6.6.5. Eligible Provisions Model after Expected Loss

The final set of models examines the relationship between eligible provisions and regulatory capital after controlling for expected loss. As mentioned, banks are required to set eligible provisions in line with expected loss outcomes or be capital charged for the shortfall between EP and EL (50% to Tier 1 and 50% to Tier 2). This final model specification examines hypothesis 7 and 8 to determine if banks use eligible provisions to capital manage after controlling for the incentive to provision in line with expected loss estimates from internal capital models.

EP-Defaulted_𝐢,𝐭 =𝛂_𝐢+𝛃_𝟏EL-Defaulted_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛆_𝐢,𝐭 (6.21)

EP-Non Defaulted_𝐢,𝐭 =𝛂_𝐢+𝛃_𝟏EL-Non Defaulted_𝐢,𝐭+𝛃_𝟐CRBP_𝐢,𝐭+𝛆_𝐢,𝐭 (6.22)

6.7. Results

Table 6.4 shows results for all banks over the sample period. Results for 12 models are shown, split into three panels A, B and C, based on the dependent variable used in the model; specific provisions, GRCL and total provisions, respectively. Within each panel there are two separate groups of models based on the regulatory capital ratio used: Tier 1 capital ratio and Total capital ratio. Within each group there are two models, one for each banks-specific credit risk variable used: (1) impairments ratio to total loans and (2) impaired and past due ratio to total loans. The model fit is sufficiently high across the 12 model specifications, with pseudo-R2 of close to 70%. Bank-specific credit risk variables and regulatory capital ratios show high levels of significance across the 12 models in Table 6.4.

6.7.1. Full Sample Results for Earnings Management

The coefficient for EBTP is negative and insignificant for all dependent variables, SPROV, GRCL and TOTPROV for the results presented in Table 6.4 over the whole sample of banks. The null hypothesis of no earnings management (Hypothesis 9) cannot be rejected, indicating Australian banks have not used provisions as a tool to manage earnings over the sample period. Similar results were found for EBTP when tested separately for standardised and IRB banks. These results are not reported but are available on request.

6.7.2. Full Sample Results for Capital Management

Results for the regulatory capital variables (T1BP and TCBP) are positive and significant as shown in Table 6.4 for specific provisions, GRCL and total provisions. This result is in line with banks using provisions as a tool to influence their capital position, limiting provisions when capital ratios are relatively lower and provisioning more when regulatory capital is

The difference in provisions for a bank with a regulatory capital ratio in the lower quartile and that of a bank with a higher regulatory capital ratio in the upper quartile is shown in Table 6.4 for each model specification. Results vary between specific provisions, GRCL and total provisions, averaging differences of 0.105%, 0.06% and 0.145% respectively. These differences are considerable when compared with the average provisioning levels across the sample of 0.26%, 0.66% and 0.91% of total loans respectively.

Table 6.4

The effect of bank capital ratios on bank provisioning practices, Tobit regression analysis

This table examines the relation between bank capital adequacy ratios and bank loan-loss provisions. The sample period is March 2004 to December 2011. Numbers in parentheses are t-statistics. ** indicates significance at the 5% level.

Panel A: Specific provisions