Dropping Odds: The Science Behind 3,509 Signals and +480 Units

What dropping odds actually are

Decimal odds are prices. A price of 2.50 on a home win says a bookmaker will pay 1.50 units of profit for every unit staked, which, before the bookmaker's margin, corresponds to an implied probability of 1 / 2.50 = 40%. When that price falls to 2.20, the implied probability rises to about 45.5%. The market has changed its mind.

At OddsDrop we measure a drop on the net payout (odds minus one), because that is the part of the price a bettor actually earns:

Most price changes are noise: small corrections, margin rebalancing, liquidity-driven adjustments. A dropping odds signal is different. It is a large, fast repricing that is unlikely to be random. Our VIP football moneyline alerts start at roughly a 20% drop in the net payout, and they are filtered further so that one outcome cannot trigger a stream of duplicate alerts.

Why Pinnacle is the reference market

Bookmakers follow two broad business models. Recreational ("soft") bookmakers run wider margins, restrict customers who win consistently and often derive their prices from other sources. Market-making ("sharp") bookmakers, of which Pinnacle is the best known, run thin margins (often around 2–3% on major football markets), accept large stakes and do not ban winning customers. Their business is volume, not the selection of losing customers.

That second model has an important side effect: professional bettors are allowed to express their opinions at size, so the price is continuously corrected by the best-informed money in the market. In the language of market microstructure, Pinnacle performs price discovery. Economists have long noted that bookmakers are not passive clearing houses. Levitt (2004) showed that bookmakers can set prices that exploit bettors' biases rather than simply balance their books, which is exactly why the type of bookmaker matters when you decide whose price to trust.

A sharp price is therefore the best available real-time estimate of the true probability. When it moves hard, the most likely explanation is that the best-informed participants have learned something.

A price move is information, and information spreads slowly

Financial economists describe markets with the efficient market hypothesis: in its semi-strong form, prices reflect all publicly available information. Betting markets come close to that ideal at kick-off (the closing price at a sharp bookmaker is famously hard to beat) but are much less efficient on the way there. Information arrives unevenly: a confirmed line-up, an injury in the warm-up, a goalkeeper illness, a change in the weather, a syndicate's model update.

The first market to absorb that information is the one that accepts informed stakes. Other bookmakers follow with a delay, because they rely on feeds, manual traders or automated copying with thresholds. The result is an information asymmetry with a timer:

1. Pinnacle reprices, say from 2.50 to 2.20.
2. For a short window (sometimes minutes, sometimes seconds) some bookmakers still offer 2.40 or 2.45.
3. That stale price now carries positive expected value, because it is offered against a probability the sharp market already rates higher.

This is often called latency arbitrage, or simply "following steam". Kaunitz, Zhong and Kreiner (2017) tested the same underlying idea academically: using the consensus of bookmaker odds as an estimate of the true probability, they identified mispriced odds across many bookmakers and made a profit betting real money, until the bookmakers restricted their accounts. The mechanism works; the practical barriers are what make it difficult.

The OddsDrop track record in numbers

Claims about betting strategies are cheap. Data is not. Every VIP signal we have ever sent is logged, settled and published on our public statistics page, including every loss. As of 24 September 2026 the football moneyline record looks like this:

Two properties of these numbers matter more than the headline.

First, the result is statistically robust. With average winning odds of about 2.05 and a 55.6% hit rate, the standard deviation of a single one-unit bet is roughly 1.02 units. Across 3,509 bets the standard error of the ROI is therefore about 1.02 / √3,509 ≈ 1.7 percentage points. A 13.7% ROI sits about eight standard errors above zero, far beyond anything that plausibly comes from luck, provided the suggested price was obtained.

Second, the breakdowns behave the way theory predicts.

Bigger drops are not better: the largest moves return the least. That looks counter-intuitive until you think about latency. The most dramatic repricings usually follow information that becomes public quickly (a confirmed line-up, a big-name absence), so slower bookmakers catch up quickly too and less stale value survives. Moderate, early moves driven by informed money leave the longest window.

Price level tells a similar story. Signals where Pinnacle's post-drop price was between 1.55 and 1.70 won 66.1% of the time across 625 signals, for a +18.7% ROI, while the 2.11–2.30 band returned +9.0%. Away-side signals (+17.1% ROI over 1,470 signals) have outperformed home-side signals (+11.7% over 2,021). Draw signals lost, but there have been only 18 of them, far too few to conclude anything, and a good reminder that small samples lie.

How to measure an edge properly

Most bettors judge themselves by profit. Professionals judge themselves by process metrics that converge much faster than profit does.

Implied probability and the overround

A bookmaker's prices for all outcomes of an event imply probabilities that add up to more than 100%. The excess, called the overround or margin, is the bookmaker's built-in edge. Removing it gives the market's fair probability, the baseline any bet has to beat:

Expected value

For a single bet at decimal odds d with true win probability p, the expected value per unit staked is:

A bet at 2.40 on an outcome the sharp market now prices at 45% has an EV of 0.45 × 2.40 − 1 = +8%. That is the mathematical content of a dropping odds signal.

Closing line value (CLV)

Because the closing price at a sharp bookmaker aggregates the most information, the difference between the price you took and that closing price, known as closing line value, is widely regarded as the best short-sample predictor of long-run profitability (Buchdahl, 2016). A bettor who consistently beats the close is very likely to be profitable over time, even through a losing month; a bettor who is winning while consistently taking worse prices than the close is very likely just lucky. We capture Pinnacle's closing price for our signals and track CLV internally for exactly this reason.

Sample size

Betting returns are noisy. At typical odds around 2.0, the standard error of ROI after 100 bets is roughly ten percentage points, enough to make a genuinely +5% strategy look like −15% or +25%. Serious conclusions need hundreds, preferably thousands, of results. That is why we publish everything and refuse to draw conclusions from any segment with only a few dozen signals.

The obstacles nobody advertises

A strategy with a strong theoretical basis and a strong record is still not easy money. These are the problems we see most often, and how serious practitioners handle them.

1. The latency window closes

Stale prices are a perishable resource. The window can be a few minutes in lower-league football and seconds in major markets. Speed of alert delivery, speed of execution and preparation (funded accounts, knowing where to look) matter as much as the signal itself.

2. Account restrictions

Recreational bookmakers profile their customers. Accounts that repeatedly take prices before they move, which is exactly what dropping-odds betting does, tend to be limited in stake size. It is the most common reason profitable approaches stop working for individuals, and Kaunitz et al. (2017) documented it in their own experiment. Spreading activity, using exchanges and brokers, and avoiding purely "sharp-looking" betting patterns are the usual mitigations.

3. False moves

Not every drop is informed. In thin markets, relatively small money can move a price, and occasionally a move is deliberately misleading. Thresholds, deduplication (we alert at most twice on the same outcome, and the second time only after a further meaningful move) and per-segment results are the defence.

4. Variance and drawdowns

A 55.6% hit rate still means 44.4% of bets lose. At that rate a run of eight consecutive losses starts roughly once every 1,200 bets: rare for any single bet, practically certain over a long career. Drawdowns lasting weeks are statistically normal even for a clearly positive record. Most strategies are not abandoned because they stopped working; they are abandoned during an ordinary drawdown.

5. Overfitting

With enough filters (league, kick-off time, odds band, drop size, side) you can make any history look spectacular. This is the multiple-comparisons problem: test fifty segments and a few will look outstanding by chance alone. Rules must be simple, economically motivated and confirmed on data they were not designed on (out-of-sample validation).

6. Edges decay

Markets learn. As more bettors and bookmakers react to the same information faster, windows shorten. An edge that is not actively maintained is an edge that is slowly disappearing.

How we keep trying to beat the odds

We treat OddsDrop as an ongoing research programme rather than a finished product. In practice that means:

• Every signal is recorded and settled, with no deletions. The published record is the same data we use internally.
• Segment analysis with minimum samples. We break results down by odds band, drop size, side, country and league, and only act on segments with enough data to mean something.
• Closing prices and CLV. We store Pinnacle's closing price for tracked signals, so we can separate genuine edge from variance long before profit alone would tell us.
• New models run silently first. New ideas are tested in the background, their results logged but not published, until they prove themselves out of sample. Our current in-play research, for example, models goals as a Poisson process (the classic approach to football scores, see Dixon & Coles, 1997) whose scoring rate is updated in a Bayesian way as a match unfolds, with exact settlement of Asian totals lines, including pushes and quarter lines.
• Separate statistics for every model version. When the logic of a product changes, its results start a new ledger. Mixing the output of an old model with a new one is one of the fastest ways to fool yourself.
• Killing ideas that don't survive. Most experiments fail. That is the point of running them.

Beating a betting market is not a one-time discovery. It is a continuous process of finding where the market is still slow, and measuring honestly whether you are right.

Using dropping-odds signals responsibly

OddsDrop is a market monitoring tool, not a tipping or advisory service, and nothing here is personal financial advice. That said, the general principles used by disciplined bettors are well established:

• Respect the minimum price. If the suggested price is gone, the bet is gone. Chasing a worse price turns a positive-EV bet into a neutral or negative one.
• Stake in units, not emotions. Flat stakes of a small, fixed fraction of a dedicated bankroll are the simplest robust approach. The Kelly criterion (Kelly, 1956) gives the growth-optimal stake for a known edge, but because real edges are estimated with error, practitioners typically bet a fraction of Kelly or use flat stakes.
• Keep records. Log the price you took and the closing price. Your own CLV will tell you whether your execution is capturing the edge.
• Judge by hundreds of bets, not by a week.

Conclusion

Dropping odds work for a reason that is economic rather than mystical: the sharpest market in the world absorbs information first, and the rest of the market takes time to catch up. That gap is measurable, and across 3,509 published football signals it has produced +480.6 units at a 13.7% return on stakes. The same gap is also fragile. It demands speed, price discipline, patience through variance, and a willingness to keep testing and discarding ideas as markets evolve. Few approaches available to bettors combine a clear theoretical basis with a public, verifiable record. That combination is what OddsDrop is built on, and it is what we keep working to improve.

References

1. Buchdahl, J. (2016). Squares & Sharps, Suckers & Sharks: The Science, Psychology & Philosophy of Gambling. High Stakes Publishing.
2. Dixon, M. J., & Coles, S. G. (1997). Modelling association football scores and inefficiencies in the football betting market. Journal of the Royal Statistical Society: Series C (Applied Statistics), 46(2), 265–280.
3. Kaunitz, L., Zhong, S., & Kreiner, J. (2017). Beating the bookies with their own numbers, and how the online sports betting market is rigged. arXiv:1710.02824.
4. Kelly, J. L. (1956). A new interpretation of information rate. Bell System Technical Journal, 35(4), 917–926.
5. Levitt, S. D. (2004). Why are gambling markets organised so differently from financial markets? The Economic Journal, 114(495), 223–246.