Historic Rating Correction Required

A very important question for a rating system like the Football Club Elo Ratings is how to choose the initial values. This is a smaller problem when continuous data for more than 50 years is available, but it becomes a big issue if your ratings start in the recent past. Many ratings you can find on the web suffer from that problem.

Sometimes however the data is not continuous, most importantly the dissolution of the Soviet League in 1991 which split in 10 different European leagues and a few more Asian leagues.

The Soviet Top League in its last edition involved clubs from Russia (6), Ukraine (6), Armenia (1) and Belarus (1). Just from looking at this league, there is not a lot we can say about the strength of Baltic, Azerbaijani or Georgian clubs, other than defining an upper bound, which is what I have done so far and this has almost certainly overestimated the Elo values of clubs from the smaller Ex-Soviet Countries to a great extent.

This contributed to an artificial overall rating inflation as these clubs lost Elo points they never should have had to other clubs in the last 20 years.

To reduce this effect, I have included 22 seasons of the Soviet second level. In its final edition in 1991, 11 Russian clubs participated, 2 from both Ukraine and Moldova, and one from each Kazakhstan, Azerbaijan, Armenia and Latvia. Georgian and Lithuanian clubs had already left the Soviet league system in 1990.

We can now state that every club from an Ex-Soviet country that is newly introduced into the system is at best as good as the worst clubs in Soviet second level, a much tighter upper bound than before.

The global effect that this artificial inflation had before is approximately 20 Elo points for every team. That means that if we do not want to change today's Elo values as we are used to them now, we can instead increase all pre-1991 Elo values by 20 points which comes into effect today.

The consequence is that today's teams will have a harder time to try to eclipse historic records, but I am convinced that this is an improved representation of reality.