Inferring the score of a tennis match from in-play betting exchange markets

Over the past few years, betting exchanges have attracted a large variety of customers ranging from casual human participants to sophisticated automated trading agents. Professional tennis matches give rise to a number of betting exchange markets, which vary in liquidity. A problem faced by participants in tennis-related betting exchange markets is the lack of a reliable, low-latency and low-cost point-to-point score feed. Using existing quantitative tennis models, this paper demonstrates that it is possible to infer the score of a tennis match solely from live Match Odds betting exchange market data, assuming it has sufficient liquidity. By comparing the implied odds generated by our quantitative model during play with market data retrieved from the betting exchange, we devise an algorithm that detects when a point is scored by either player. This algorithm is further refined by identifying scenarios where false positives or misses occur and heuristically correcting for them. Testing this algorithm using live matches, we demonstrate that this idea is not only feasible but in fact is also capable of deducing the score of entire sets with few errors. While errors are still common and can lead to a derailment of the detection algorithm, with more work as well as improved data collection, the system has the potential of becoming a precise tool for score inference.