Interestingly, this is the original 1963 paper from RCA labs, that posited the LED cooling effect.
Evidence of Refrigerating Action by Means of Photon Emission in Semiconductor Diodes
More than 90% of the photons emitted from forward-biased GaAs diodes have energies $h\ensuremath{\nu}$ higher than the applied voltage $V$. Thus, a portion of…
journals.aps.org
It was their way of trying to explain an experimental anomaly. But although their hypothesis wasn't actually tested (no temperature drop could be measured), the electroluminescent cooling effect seems to have become part of established LED theory.
Unfortunately, the full paper (available via the usual work-arounds, for those without institutional access) is rather opaque.