I have encountered a rather elegant argument about the limitations of empirically testing for market efficiency, involving the central point that we do not know whether a result is due to the "true behaviour" of the market or due to the model used to simulate that behaviour.
Unfortunately I have not been able to retrieve this argument online, nor any publications relating specifically to this, which may be due to the fact that I do not know how this argument is typically referred to in the literature.
In particular, I would like to understand how we can interpret any empirical results regarding market anomalies or market efficiency when taking into account the above important limitation. Say, for example, we use the Fama-French-Carhart model in order to examine whether a particular portfolio formation strategy leads to abnormal returns. If our $\alpha$ is positive and significant, how can we know this is due to an actual anomaly (on which we have based our portfolio formation), rather than a bias in our model, which has underestimated market returns (other than the fact that the model usually has a decent predictive power)?
I greatly appreciate any clarification or resources!
