This paper a develops novel statistical test of whether individual factor risk premia

are identified from return data in multi-factor models. We give a necessary and

sufficient condition for population identification of individual risk premia, which we

call the kernel-orthogonality condition. This condition is weaker than the standard

rank condition commonly assumed for linear factor models. Under misspecification,

our condition ensures point identification of the risk premium with minimal pricing

error. We show how to test this restriction directly in reduced-rank models. Finally,

we apply our test methodology to assess identification of risk premia associated with

consumption growth and intermediary leverage.

We document that the carry of crypto futures, i.e. the difference between futures and spot prices, can become very large (up to 60% p.a.) and varies strongly over time. This behavior is most consistent with the existence of a highly volatile crypto convenience yield that stems from two main forces: (i) trend-chasing and attention by smaller investors seeking leveraged upside exposure to crypto assets in boom periods, and (ii) the relative scarcity of "arbitrage" capital taking the other side through a cash and carry position. Engaging in the latter is risky due to spikes in margins and liquidations amid drawdowns. The interplay between these two forces, and the involved high leverage, may help explain why severe market crashes are a frequent feature of crypto markets.