On the remote chance you care about this… I found a couple more instances of “February 29", some “Murder of ___” pages, “Talk: 2014”, and assorted stuff like that. As well as a couple dozen duplicate entries. Probably nothing that would alter the results you reported. There are ~ 100 entries with 1 edit where I think the 1 is an error. Anyway.
I took the cleaned-up list and applied your celebrity formula. I then counted deaths by year for the entire list, and for the top 2000 entries according to the formula. In the case of the top 2k there was a weak linear trend (r2 = 0.5). The ratio of predicted to actual deaths had its top two values in 2016 and 1994 and bottom two values in 2000, 2001, and 2004.
For the entire list there was no linear trend (r2 = 0.1). Instead I took the ratio to the average. Top two values were in 2016 and 1993; bottom two values were in 2001, ‘02, ‘03, ‘04 and ‘05.
Doesn’t look terribly related to the business cycle. The overall pattern with the two series is similar, with peaks in 93–94 and 2016 and a trough in the early 2000s.
Spreadsheet at http://lumma.org/temp/DeadCelebs.xlsx