Vlad

23-01-2006, 11:15 AM

I was thinking another day how performance ratings are calculated and find the process less than satisfactory. The first well known issue is that average rating is used instead of doing it game by game. This one is easy to fix and this is what I believe Bill is doing in Australia.

The second problem is that actual ratings are used in calculations. This might be not so good idea because the ratings are quite often lagging. Take example of Raymond in the last championship. His actual rating was 2050, while he performed about 2300. When the performance ratings of his opponents are calculated the above fact is not taken into account. This problem is easy to fix if approach that scientists use is applied. (Scientists have a very similar problem. They rank scientific journals. There are cross references from one journal to another. There is original ranking, which is used to calculate new ranking based on references a particular journal receives from other journals. The process is repeated until the ranking converges.)

In performance rating terms the process can be amended in the following way. After old performance ratings are calculated, they are used as initial ratings in the process and using them new performance ratings are calculated. The process is repeated until performance ratings converge. The process will give true performance rating which does not depend on your initial rating.

The second problem is that actual ratings are used in calculations. This might be not so good idea because the ratings are quite often lagging. Take example of Raymond in the last championship. His actual rating was 2050, while he performed about 2300. When the performance ratings of his opponents are calculated the above fact is not taken into account. This problem is easy to fix if approach that scientists use is applied. (Scientists have a very similar problem. They rank scientific journals. There are cross references from one journal to another. There is original ranking, which is used to calculate new ranking based on references a particular journal receives from other journals. The process is repeated until the ranking converges.)

In performance rating terms the process can be amended in the following way. After old performance ratings are calculated, they are used as initial ratings in the process and using them new performance ratings are calculated. The process is repeated until performance ratings converge. The process will give true performance rating which does not depend on your initial rating.