A dropping odds strategy of "bet on every outcome whose price fell" does not work on our history: on the match result such bets landed less often than even the reduced price implied. But a falling price is a useful filter, and some slices are worth testing over a long run.
Articles about dropping odds usually stay general: "a drop is a signal, but there are no guarantees". We decided to count. The xG.Football tracker follows bookmaker prices on the result and the goal total of matches and records every noticeable drop. After the match we see whether the outcome landed. Below is what the data shows, how to read it and why "dropping odds don't work" would also be too simple a conclusion. In short, a blind strategy on falling odds loses to the margin, and betting on dropping odds only makes sense as a tested hypothesis.
What we counted and how
The data is every settled signal of the tracker from 1 September to 30 October 2026, 1493 in total. We update this piece as data accumulates; the first version of 24 October rested on 906 signals.
- Signal — the median price of an outcome fell 10% or more from the opening, with the price below the opening at most bookmakers (60% or more). One-off drops at a single bookmaker do not count.
- Size of the drop — how far the price fell by the close. Some signals bounce back before kickoff: the odds fell 10%, then returned. Those are in the "under 10%" row.
- Landed — the share of outcomes that won.
- Implied — how many wins the price "promised" at the signal (1 / odds). It includes the bookmaker margin, so the fair probability is slightly lower.
- ROI at closing — the result of a flat one-unit stake on every signal at the last price before the match.
No selection in hindsight: every signal is in the tables, good or bad.
Results by market
| Market | Signals | Landed | Implied by price | ROI at closing |
|---|---|---|---|---|
| Match result (1X2) | 1113 | 32.1% | 36.8% | −15.7% |
| Goal total | 380 | 54.2% | 54.8% | −1.9% |
Match result. This is the main and surprising result. Outcomes whose odds dropped landed less often than even the reduced price implied: 32.1% against 36.8%. A one-unit stake on each would have returned about −16%. In other words, the market on average "overloaded" these outcomes: more money came in than their real chances justified.
Goal total. The picture is flatter here: 54.2% landed against 54.8% implied and −1.9% at closing. The first version of this piece showed +3.1% on 224 bets: more data removed that plus. There is no visible edge on totals.
Size of the drop: bigger is not better
| Drop by the close (1X2) | Signals | Landed | ROI at closing |
|---|---|---|---|
| Under 10% (price bounced back) | 183 | 35.0% | −2.1% |
| 10–15% | 518 | 33.2% | −20.2% |
| 15–25% | 310 | 31.6% | −13.4% |
| 25% and more | 102 | 22.5% | −23.5% |
Intuition says the stronger the drop, the more confident the market. On the match result it went the other way. The sharpest drops, from 25%, landed least often: 22.5% against 26.3% implied by the price. Of those 102 signals, 82 came from leagues outside the site's list, with −4.8% at closing there. The site leagues had 20 such signals and none landed, although the price implied 21.2%. Twenty matches prove nothing, but the group is worth watching.
Keep in mind that the drop by the close is only known just before the match: at the moment of the signal you do not yet know which row of the table it will fall into.
On totals the picture is uneven: with a 10–15% drop (237 signals) 55.3% landed and the result was −1.2%, with 15–25% (70 signals) 47.1% and −19.3%, and with prices that bounced back (72 signals) 58.3% and +14.3%. The groups are small, and this spread says nothing yet.
Site leagues and the rest
We looked separately at matches of the leagues that have pages and predictions on the site, and at all the others. On the match result the site leagues did worse: 303 signals, 26.4% landed and −29.8% at closing. Other leagues: 810 signals, 34.2% and −10.4%. This gap held after the data update. A possible explanation is that in well-known leagues the line is accurate from the opening, and a strong drop there is more often public money on a popular club than news the market had not priced in yet. On a sample this size it is a hypothesis, not a conclusion.
So dropping odds are useless? No
"Don't bet on every drop" is not the same as "a falling price means nothing". A drop shows where money and information went. That is valuable for analysis:
- As a filter. A match with a strong line move deserves a closer look: there is almost always team news or motivation behind it.
- As a check of your own estimate. If the market moves against your prediction, find the reason before the bet, not after.
- As a source of hypotheses. The average over all signals is negative, but some slices behave differently. But such slices need rechecking. The first version of this piece gave an example: the match result outside the site leagues, a drop of 20% to 25% by the close, 45 signals, 37.8% landed against 35.6% implied and a result around zero. On 75 signals it now shows 30.7% against 33.6% and −19.5%. The hypothesis did not hold, which is the usual fate of a pretty slice on a small sample.
A lasting edge, if there is one, only shows over a long run — hundreds of outcomes and several seasons. Why a small sample misleads and how to tell luck from a pattern is covered in how to measure prediction accuracy.
How to test your own dropping odds theories
The tables above are averages over all signals. Your idea may be narrower: "underdog priced from 3.00, drop from 15%, totals only in top leagues". For such checks there is the Theories tab on the dropping odds page. Its results are recounted every day, and the history grows with them.
With Premium you can:
- Set the hypothesis with filters — market, outcome, side, opening odds, size of the drop, leagues, signals only.
- See the result on history — number of outcomes, how many landed, what the price implied and ROI at closing, split by size of the drop and by odds.
- Open the matches themselves — clicking any row opens the matches behind the number on the right, with a chart of the line movement.
- Save the theory to "My theories" and follow how many new outcomes were added since your last look and how they did.
- Turn on Telegram alerts — the bot sends a match as soon as it gets a new signal that fits the theory.
That way an idea is tested on future matches instead of being fitted to past ones. If a theory holds for several months and hundreds of outcomes, that is grounds for a decision. If not, you lost time, not your bankroll. How much to stake once an idea is found is a separate question, covered in bankroll management.
What dropping odds are and why odds fall at all is explained in detail in what dropping odds are.
Limits of this test
- A short period. Two months of data. The numbers will update, and some conclusions may shift, as the result on totals already has.
- Uneven coverage. Signals grew noticeably in the last week of the period. The signal rule did not change, but on 25 October we changed how the tracker's passes are shared between days, so the latest matches may be covered more densely.
- Small groups. Some slices have 10–40 signals. We only quote large groups and do not use small ones as arguments.
- The closing price is not your price. A real bet is placed earlier and at a specific bookmaker. Betting at the moment of the signal on the match result would have given a similar result, around −15%.
- This is not betting advice. We show how the market behaved, not what to do with money.
The short version
| Question | Answer |
|---|---|
| Does a dropping odds strategy work | Betting on every match result with a falling price returned about −16% at closing on 1113 signals |
| What about totals | 54.2% landed against 54.8% implied and about −2%: no visible edge |
| Is a bigger drop better? | No: on the match result drops of 25% and more landed least often |
| Why follow dropping odds then | As a filter and a source of hypotheses: a drop shows where money and information went |
| How to test your dropping odds strategy | Set the filters in the Theories tab, save the theory and track new outcomes and alerts |