Anybody has some idea/data?

How much does your winrate decrease when you rank up?
In my example of ranking up an account that I opened after a break:

Rank/games/winrate[%]
1k/13/77
1d/23/70
2d/20/70
3d/20/70
4d - I should play games instead of thinking about ELO

1: You getting a 70% score against Tygem 1d-3d suggests your Elo rating is roughly 150 Elo higher than the average of your opponents up to now.
2: The EGF does not use Elo ratings. The EGF rating system has rank widths of 100 GoR by definition, but 100 GoR does not equal 100 Elo. So EGF ranks typically are not 100 Elo wide and an answer to your question may not be really meaningful in terms of EGF ranks or any other go rating system.
3: Tygem does not use Elo ratings either.
4: It seems that Tygem ranks are pretty much all over the place, because there are many sandbaggers and cheaters, most games are blitz and newcomers can set their initial rank to what they want (up to 3d).

That being said, according to this OGS 2018 survey, Tygem 1d-8d roughly corresponds to the range of EGF 4k-5d and Tygem 4d roughly corresponds to EGF 1d.

Thank you gennan for useful information.

I know Tygem does not use Elo, but effectively one could try to use Elo to describe the system.

For instance, compute the average winrate in even games between random players from ranks G and G+1, which then translate to "Elo width" of rank G.
This probably depends on G.

Then one can repeat the same for players from G-1 and G+1, and see if Elo model applies (which would suggest the Elo width of 2 ranks is twice the width of one rank).


But my data is obviously insufficient to describe it.

One could guess from the Tygem's promotion system, that the first winrate should be around 65%, since one needs 70% winrate for promotion in some streak. This would correspond to Elo width of ~ 100. But I have no idea if this is indeed the case.

Note that the threshold is different according to the level: to promote from 5d to 6d you need to win 15 of your last 20 games (75%).

Earlier this year I analysed the 1 million tournament game results in the EGD (spanning the time period 1996-2019) and used logistic regression to calculate the Elo widths of EGF ranks. It does vary over time, but overall it roughly looks like this:
Someone else found similar results: https://wismuth.com/elo/calculator.html (see Elo per stone in go on that page).