What Closing Line Value Measures

The obvious way to evaluate a model is to see whether it was right. In these markets that approach is unusually weak.

Outcomes are binary and rare. A single event produces one observation, the variance is enormous relative to any plausible edge, and accumulating enough results to distinguish a genuinely good model from a lucky one takes far longer than anyone wants to wait. A model can be sound and look poor for a long stretch, or be unsound and look excellent, and nothing in the outcome record separates those quickly.

So the field reached for a denser signal.

Prices move continuously as information arrives: team news, weather, money from informed participants. The price at close reflects everything the market absorbed before the event started, which makes it the most informed estimate available at any point in the process.

If a model consistently identifies prices that later move in its direction, that is evidence it saw something before the market finished absorbing it. And crucially, every market produces this observation, whether or not the outcome happens to go a particular way.

That density is the entire argument. It is not that price movement matters more than outcomes. It is that you get thousands of observations instead of dozens.

It suggests timing. Consistently recording prices that subsequently move toward your estimate indicates the estimate contained information the market had not yet priced.

It is denser than outcomes. Every market closes, so every observation counts, and a signal emerges in weeks rather than seasons.

It is a proxy, and proxies drift from what they proxy. Closing prices are the best available estimate and they are not truth. Markets can be systematically wrong in ways that persist through close, particularly in thin markets, unusual sports and exotic outcome types where less informed money participates.

It can be beaten without being right. A method that reliably identifies prices likely to move, for reasons unrelated to the underlying likelihood, will show strong movement statistics and no predictive validity. Movement following stale prices in slow markets is the common example.

It says nothing about magnitude. A price that moved slightly and a price that moved a lot are both movements. Weighting matters and unweighted counting overstates consistency.

So the honest framing is that it is a leading indicator with known failure modes, useful for iterating quickly and insufficient alone for concluding anything. Outcomes still have to be checked, over longer horizons, as confirmation rather than as the primary signal.

Measuring it requires timestamp discipline

Define close precisely and consistently. The last price before the event begins, from a named source, with a rule for suspended or removed markets. Any inconsistency here shows up as apparent signal.

Compare like with like. The same market, the same outcome, the same source or a defensibly normalized cross-source price. Comparing your recorded price at one book against the close at another measures the difference between books as much as anything else.

De-margin both sides before comparing. Raw prices carry a margin that can differ between your observation time and close, so a change in margin can look like a change in estimate.

Weight by magnitude and record the distribution. Average movement plus its spread is far more informative than the share of observations that moved favourably.

Segment by market type. Movement behaviour differs sharply between major markets and thin ones, and an aggregate figure hides the segments where the proxy is weakest.

Keep outcomes as the slower check. Track them, expect them to be uninformative in the short run, and treat a persistent divergence between movement and outcomes as a signal that the proxy is being gamed rather than as noise.

This page describes an evaluation method and is not betting advice.