On the top left are the changes for the players in my sample from season 15 to season 16. Ideally for this purpose they should have received no salary-relevant training, but I include a few who apparently have received a little bit, among them our sole outlier Bertas Seporaitis at the top end. Still the pattern is clear, you can see where the graph 'turns a corner' at around $280k.

On the top right you see the compounded salary changes since the beginning of dynamic salaries in season 11(?). In the low salary regions where almost all players are, the differences are not big - decreases between 15% and 20% of original salaries.

On the bottom left is the picture that emerges if one tries to find a "salary deflation coefficient" for season 16 in the form of an exponent, as we have done in the last seasons. Some may remember that already for season 15 some irregularities emerged between very low and higher salaries under such a model, and with the further compound changes of this season it becomes clear that even below 'monster territory' there is no single coefficient that fits all players - there appears to be a nonlinear decrease of the coefficient from very low to medium and high salaries. Above $280k the old 'salary deflation coefficient' model breaks down completely, as there is now a second, nonlinear relationship in place here. [Edit: The title of that diagram is a bit misleading, it is not about the change from s15->s16, but from origin to s16]

On the bottom right you see the top left graph again, but this time with the salaries drawn on a logarithmic scale on the X-axis. In such a view the change of the decrease over the salary range appears linear again, with two different gradients for 'normal' and then for 'monster' salaries above the $280k (in season 15 terms) salary point. Our Lithuanian behemoth Bertoraitis at the top end messes things up just a little bit because he trained. This graph explains why I suggest the new model for dynamic salary change as described in the previous message - the fit with the calculated coefficients is very good, with errors of much less than 1% on untrained players.

Now the interesting part, which is a bit speculative: With such a structure for the dynamic salary formula, the BB's can in the future tune 'normal' and 'monster' salaries independently. For example monster salaries could be left constant in the future while normal salaries continue to decrease at a slow pace, swallowing up the lower parts of the monster-curve - or vice versa, monster salaries could continue to fall and normal salaries remain near constant. For monster-owners like myself this is something new to think about...

(Josef Ka)