Irish Commission on Credit Unions, March 2012
APPENDIX 4: Efficiency Study by Professor Donal McKillop and Barry Quinn
This study investigates Irish credit union performance over the period 2002 to 2010. The empirical analysis uses a two-stage approach. The first stage measures efficiency by a data envelopment analysis (DEA) estimator, which explicitly incorporates the production of undesirable outputs such as bad loans in the modelling process. In modelling the productive process of credit unions, each credit union is viewed as employing a set of inputs (capital expenditure, labour expenditure and interest paid on deposits and dividends paid on shares) to produce a set of desirable outputs such as various types of loans and investments as well as shares and deposits. Unfortunately some loans and investments turn out to be ‘bad’ loans and ‘bad’ investments which must be written off as bad debt. Producer-specific performance measures which permit credit unions to be credited for the reduction of undesirable outputs as well as for increasing desirable outputs and decreasing inputs are thus obtained 36. The performance measure (efficiency score) obtained for each credit union ranges from 0 (highly inefficient in transforming inputs into outputs) to 1 (highly efficient in transforming inputs into outputs). In the second stage of the analysis how certain factors influence the efficiency scores of credit unions is examined. Factors that prove important are capital strength, liquidity, asset size and common bond type.
Table 2 : Factors Influencing Performance
Parameter estimates and bias corrected confidence intervals
|Regressors||Estimated Coefficients||Lower 5%||Upper 5%|
|Surplus funds/asset ratio||-0.0623||-0.04||0.18|
|Asset size squared||-0.121***||-0.13||-0.1|
(i)The regressand is the bootstrap-based bias-corrected DEA estimate of the unobserved efficiency score of each credit union with a higher score value indicating greater efficiency. A positive estimated coefficient thus implies that an increase in a explanatory factor increases efficiency.
(ii) ***, **, * denote significance from zero at the 1%, 5% and 10% levels using the bootstrap-estimated confidence intervals.
(iii) The truncated regression estimation with bootstrap is based on Simar and Wilson (2007), Algorithm 1, using 1,000 bootstrap replications for the confidence intervals of the estimated truncated regression coefficients.