Closing line value vs the discarded side

+2.92 pt

95% CI [+1.69, +4.16] · 1,211 matches

Tennis: we beat the side we discarded by 2.92 points

Measured 21 September 2026. On the tennis matches where our model took a side, that side beat the side it discarded by 2.92 percentage points of closing line value.

This is the only demonstrated performance figure we publish. The other five sports on this site do not have one, and we do not pretend otherwise.

Demonstrated ATP & WTA 1,211 matches

Why we compare the two sides of one match

A raw closing line value journal is misleading, and ours was. It showed positive values across all six sports while recording both sides of every match. That is arithmetically impossible as an edge: if backing one side beats the close, backing the other must lose to it.

What such a journal actually measures is a baseline gap, the price taken is the best of a panel of bookmakers, the closing price is a single reference. Comparing the side the model selects against the side it discards, within the same match, cancels that gap. Both carry the same baseline, so it subtracts out. What remains is the model's choice.

What the number is not

At 30 minutes before start+4.24 pt
At 24 hours before start+1.69 pt
Its 95% interval at 24 hours[−0.43, +3.81]
Matches on the winning side52.8 %

It is not a return. Closing line value predicts profitability; it is not profitability. Nothing here says you would have made 2.92% on anything.

It is not demonstrated 24 hours out. That interval contains zero. A reader betting a day ahead is not covered by the headline, so we print the bound rather than the headline alone.

It is not a per-match promise. Barely more than half the matches landed on the winning side. The average lives in the tail, not in the typical match.

Method

The unit of observation is the match, not the price snapshot. The source holds 3,151 decision rows, because one match is sampled several times as its price moves. Counting those as independent observations would shrink the confidence interval artificially, and the interval is the part that matters most. Four snapshots of one match count once.

The measurement is rerunnable from our own export of settled decisions, dated 21 September 2026, by a script kept in the repository. Before that date the figure existed only in a working note. It has been recomputed from source and reproduced identically.

The archive behind it

Tennis odds readings kept1.8 M+
Of which closing prices430,177
Oldest reading2020-01-05

The closings are the part that matters. An archive of pre-match prices can feed a model. An archive that also holds the close can judge one, and judging is what nobody can do retroactively if the prices were not stored at the time.

Citing this page

MySyndicate tennis closing line value, 21 September 2026: +2.92 pt against the discarded side, 95% CI [+1.69, +4.16], on 1,211 matches; +1.69 pt at 24 hours, CI [-0.43, +3.81]. Quote the interval with the number. https://www.mysyndicate.io/tennis-closing-line-value