Diagnostics for asset pricing models

Abstract

The validity of asset pricing models implies white-noise pricing errors (PEs). However, we find that the PEs of six well-known factor models all exhibit a significant reversal pattern and are predictable by their lagged values up to 12 months. Moreover, the predictability of the PEs can produce substantial economic profits. Similar conclusions hold for recently developed machine learning models too. Additional analysis reveals that the significant PE profits cannot be explained by common behavioral biases. Our results imply that much remains to be done and there is a great need to develop new asset pricing models.