Official Ratings/Rankings of BTTV, ELTTL, KTTV, NÖTTV, OÖTTV, ÖTTV, StTTV, STTV, TTA, TTACT, TTNSW, TTNT, TTQ, TTSA, TTT, TTTV, TTV, TTWA, VTTV, and WTTV.

Long-Term Stable, Quick-Reacting Table Tennis Ratings via Poisson Jumps

Abstract

It is not easy for a rating system to be stable over decades and also react quickly to rapidly-improving players.

Ratings Central has been providing table tennis ratings for clubs, organizations, and countries around the world since 2004. As of July 2025, the database contained over 120,000 players and 5.6 million matches. The Bayesian rating algorithm (in the same family as the Elo, Glicko, and Glicko-2 algorithms) used a normal mean-zero random walk to model how a player’s playing strength changed in time.

In 2018, the Austrian Table Tennis Association (our largest customer) reported that the ratings for players who had been playing for over a decade were deflating 30–40 points per year. Also, rapidly-improving players (typically juniors) were sometimes underrated. Fixing the priors that the Austrian Table Tennis Association had used reduced the deflation to 10 points per year. Nevertheless, the problems were real.

To fix the problems, we modified the temporal-update model by adding a Poisson jump process.

To determine the size and probability of the jump, most helpful were

We set the jump to 200 rating points. We set the probability of a jump so that the mean of the temporal update was +7 rating points a year. Ratings Central already used discrete probability distributions to implement the algorithm, so it was not too difficult to incorporate the Poisson jumps into the temporal-update model.

This fixed both problems.

Ratings Central started using the improved temporal update model in 2019. All historical data was reprocessed using the new model.

For future work, we propose an algorithm for rating individual doubles players.

Presentation