This dissertation presents empirical evidence intended to help answer two research questions. The first question asks whether executive compensation systems appear to exploit the bias in accounting-based performance measures in order to reduce the volatility in executive compensation and to allocate incentives more effectively across the range of activities performed by the executive. The second question asks whether compensation systems systematically differ between firms that use alternative accounting methods and whether any such systematic difference helps explain accounting choice. Parameters estimated in fixed-effects endogenous switching regression models were used to test the risk-shielding and incentive-allocation hypotheses. The models were estimated across a dataset consisting of 1151 executive-year observations of annual compensation paid to 222 top-level executives in 40 oil and gas firms. The dataset was partitioned by accounting method and separate models estimated for the full cost and successful efforts partitions. The tests provided modest support for the risk-shielding and incentive-allocation hypotheses, revealing that accounting measurement bias is used to focus incentives for effort in the exploration activity and to reduce executives' exposure to production risk. The design also allowed an estimate of the proportional change in compensation that was realized from the accounting choice actually made.

This study uses concurrent verbal protocol analysis to examine the decision processes of lenders as they evaluate the financial information of a loan applicant. Of specific interest is the lenders' use of Statement of Financial Accounting Standards Board No. 95 (FAS 95), Statement of Cash Flows, in that decision process.

This study uses both basic and self-selection regression models to test three hypotheses about the effect of SFAS 87 disclosures on information asymmetry during 1985- 1987. Both types of models test the hypotheses after controlling for changes in the inventory holding and order processing costs of the spread, while the self-selection models also control for potential self-selection bias.