Every couple of years I make some attempt to condense the principles of financial modeling:
All securities bear risk.
The essential problem of financial modeling is to determine the expected return that goes with a given estimated (future) risk.
To determine the expected return of a given risky security, note that there are three ways of lowering its risk:
Dilution: combining the security with a riskless bond;
Diversification: combining the security with other uncorrelated securities;
Hedging; combining the security with another anti-correlated security to cancel one aspect of its risk.
If by these means you can lower the estimated risk of a portfolio that contains the security to zero, it is logical that you should expect to earn the return of a riskless bond, assumed known. The expected return of the given security can then be pinned down from this constraint.
All of this is perfectly reasonable a priori. Where it fails in real life is that one cannot estimate risk correctly.