Accurate measurement of bank risk is a matter of considerable importance for bank regulation and supervision. Current practices in most countries emphasize reliance on financial statement data for assessing banks’ risk. However, the possibility of increased reliance on market-based risk indicators has been a topic for academic and regulatory debate for a long time. Market monitoring of bank risk has typically been tested by regressing market-based risk indicators on various benchmark indicators (such as accounting ratios and credit ratings) to detect whether the market tracks bank risk. This approach overlooks the methodological ‘unobservability’ problem that testing one imperfect proxy indicator against another, when the true value (in this case, a bank’s ‘true’ risk) is unknown, must yield limited conclusions as to the appropriateness of either indicator – particularly in the event of failure to establish a significant association. This paper assesses the relative information content of different risk indicators indirectly by associating the divergence between these indicators with the institutional setting. Empirical results for a large panel of banks worldwide suggest that market-based indicators are often more accurate than accounting indicators for high levels of institutional quality. In particular, spreads on subordinated debt may be more informative than either equity-based or accounting-based measures if the institutional conditions for market discipline to function are favourable. In addition, a combination measure incorporating both accounting and market data has superior accuracy regardless of the level of institutional quality, indicating that market data may contain complementary information on risk. These results cast doubt on the validity of the conclusions drawn in several previous studies that reject market discipline based on the finding that market-based risk indicators do not correspond well with various standard non-market indicators.
bank risk risk indicators subordinated debt market discipline panel data