The volcanic forcing data used in Otto et al 2013 was processed with a 5 year running mean filter, for no apparent reason.
[Nic Lewis , a co-author of the paper, has contacted me to assure me that it was a 3 year, not a 5 year, running mean filter that was used.]
This has the effect of spreading the effect of the volcanoes back in time before the event. This creates a fortuitous and spurious correlation with the natural variations in temperature happening around the time of the eruption.
This will thus create a spurious correlation giving the impression that the model is capturing this key feature of the temperature record in its hindcasts.