This thesis proposes a theory of inefficient markets that uses limited rational choice as a central trait and I call it the theory of Homo comperiens. The theory limits the alternatives and states that the subjects are aware of and only allow them to have rational preference relations on the limited action set and state set, i.e. limited rationality is introduced. With limited rational choice, I drive a wedge between the market price and the intrinsic value and thus create an arbitrage market. In the theory, the subjects are allowed to gain knowledge about something that they previously were unaware of. As the discovery proceeds, the arbitrage opportunities disappear, and the market prices regress towards the intrinsic values. The theory is applied to firms and market-pricing models for a Homo comperiens environment is a result. The application of the theory to firms also leads to testable propositions that I test on a uniquely comprehensive Swedish accounting database that cover the years 1978—1994. Hypotheses are tested which argues that risk-adjusted residual rates-of-returns exist. The null hypotheses argue that risk-adjusted residual rates-of-returns do not exist (since they assume a no-arbitrage market). The null hypotheses are rejected in favor of their alternatives at a 0.0 percent significance level. The tests use approximately 22,200 observations. I also test hypotheses which argue that risk-adjusted residual rates-of-returns regress to zero with time. The null hypotheses are randomly walking risk-adjusted residual rates-of-returns, which are rejected in favor of the alternative hypotheses. The hypotheses are tested using panel regression models and goodness-of-fit tests. I reject the null hypotheses of random walk at a 0.0 percent significance level. Finally, the results are validated using out-of-sample predictions where my models compete with random-walk predictions. It finds that the absolute prediction errors from my models are between 12 to 24 percent less than the errors from the random walk model. These results are significant at a 0.0 percent significance level.