So I was surprised to see this short Journal of Finance article by Richard McEnally back in 1974, A Note on the Return Behavior of High Risk Common Stocks.
HIGH RISK COMMON STOCKS, it is frequently observed, do not appear to generate returns commensurate with the level of associated risk...Soldofsky and Miller constructed indices of the returns from six investment advisory service common stock quality classes over the 1951-1966 period. They found that the geometric mean annual return from the lowest grade of stocks was less than half the equivalent return from any of the less risky classes....
He mentions another paper that found negative risk-reward data, and one where the relation was small. He then lists 4 reasons this could be happening:
1) The Asness, Frazzini and Pedersen (2011) hypothesis that people are constrained on the in equity exposure, so overload on high beta stocks, causing the Security Market Line to be almost flat (but still positive!).
2) Lower taxes on the capital gains, that are relatively more important for risky stocks.
3) Overconfidence in risky stocks
4) Positive skewness preference
Clearly this was a prescient piece! But, then he also presents his own data and takes all the stocks alive from 1945-65, and sorts them into low and high beta groups, and looks at their monthly returns, which is a massively flawed way of looking at these but strangely still pops up. Clearly this is a nonintuitive bias though I find it rather obvious.
I think this highlights you should judge academic work piecemeal, as often there's good and bad insights contained within the same work.