Backing the Draw on Betfair: What Happens If You Just Keep Collecting?

The draw might be the most ordinary bet in football.

Every Match Odds market gives us the same three basic choices: Home, Draw or Away. Punters have been backing draws forever. There are countless draw systems built around goals, league tables, previous results, expected goals, team strength and every other conceivable filter.

Our latest Research Lab experiment starts somewhere much simpler.

What happens if we stop trying to identify which matches are likely to draw and just start collecting the draw itself?

No claim that we've found a draw strategy. No carefully selected league where draws have performed particularly well. No retrospective search for the perfect odds range.

Just another prospective dataset beginning its life.

And after its first 140 live bets, it has already produced some cracking examples.

From Saudi Arabia v Qatar to an 11.50 draw in El Salvador

One of the first bets in the dataset was Saudi Arabia v Qatar, with the draw matched at 3.45.

It won.

Nothing particularly remarkable about that. Odds of 3.45 imply that the market considered the draw a perfectly plausible outcome. If you're going to collect draws across football, you're going to see plenty around that sort of price.

But keep collecting and things become more interesting.

On 3 October, Tolima v Boyaca Chico produced a winning draw at 6.60. The following day, CD Castellon v AD Ceuta did exactly the same thing, again at 6.60.

Then came Lysekloster v Jerv in the Norwegian 2nd Division.

The draw was matched at 9.60.

It won.

And then, early on 5 October, the collector picked up CD Fas v CD Inca in the Salvadoran Primera Division.

Average matched price on the draw:

11.50.

Another winner.

That's one of the things I already like about running these collectors globally. Within a tiny opening sample we've moved through international football, Spain, Norway, Colombia, El Salvador, women's football, youth football and competitions most British punters probably aren't analysing over breakfast.

The collector doesn't need an opinion on any of them.

It just records the market and what happened.

140 bets, 61 competitions and a very strange little dataset

At this first snapshot, we've accumulated 140 bets across 61 different competitions. Where country codes are available, we've already captured football from 34 countries, and the actual spread is broader because a number of international competitions don't carry a conventional country code in the export.

The prices are interesting too. Across all 140 bets, the average matched odds are 4.61, with individual prices ranging from 2.84 all the way to 18.00.

We've had 40 winning draws and 100 losers, producing an early strike rate of 28.57%.

Using our standard level one-unit calculation and allowing for 2% commission on winning profit, the collector currently stands at:

+34.74 units after 140 bets.

Again, that looks lovely.

Again, it means very little at this stage.

The reason is sitting right there in the individual results. A draw at 11.50 contributes 10.29 units after 2% commission from a single one-unit bet. The 9.60 winner contributes another 8.43 units. Those two matches alone therefore account for nearly 19 units of the current result.

Remove a couple of big winners from a 140-bet sample and the picture changes dramatically.

That's not a problem with the dataset.

That's something the dataset can eventually teach us about variance.

The losing bets matter just as much

It's easy to write about the spectacular winners because they're fun. But there are already 100 losing bets sitting beside them.

The longest losing sequence in this opening sample is 19 consecutive bets.

That number interests me almost as much as the 11.50 winner.

A draw collector naturally produces a strange psychological experience. You can lose repeatedly and then hit a relatively large-priced winner that repairs several previous losses at once. That makes the order in which results arrive important to how the strategy feels, even when the underlying mathematics hasn't changed.

At 140 bets, though, we're nowhere near having enough evidence to characterise those sequences properly.

What happens over 1,000 bets?

How frequently do losing runs of 15, 20 or 30 appear? Does the average price remain somewhere around 4.6 once ordinary domestic schedules return after the international break? Do the occasional 8.0, 10.0 or 12.0 draws compensate sufficiently for the much larger number of losing bets?

Those are much better questions than asking whether +34.74 units means backing draws works.

It doesn't tell us that.

Why I'm not filtering anything yet

This is probably the point at which traditional betting-system research would start slicing.

Perhaps Colombian draws look good. Maybe international matches look poor. Perhaps prices between 3.5 and 5.0 produce a better return. Maybe women's football behaves differently. Perhaps we should exclude enormous draw prices.

With 140 bets, you could probably find something that looks impressive if you tortured the spreadsheet for long enough.

I'm deliberately not interested.

The collector has already reached 61 competitions. That's exactly what I want at this stage: breadth, not optimisation.

Let the losing bets stay.

Let the 18.00 shots lose.

Let the occasional 9.60 or 11.50 draw land.

Let international football sit beside the Norwegian 2nd Division, the English WSL, Colombian football and whatever else enters the market.

First build the dataset. Then ask it questions.

Imagine looking back at this from bet 10,000

That's ultimately what makes these experiments interesting to me.

The first 140 observations are almost disposable from a conclusions point of view. They're entertaining because we're watching the dataset being born, but their real value comes if they eventually become the first 140 rows of something much larger.

At 5,000 or 10,000 bets, we could have a substantial prospective dataset containing thousands of real football Match Odds markets across countries, competitions and price ranges.

Then the questions become much more interesting.

How often does a football match actually finish level across the markets we've captured? How closely did market prices correspond with the eventual frequency of draws? What did the distribution of losing runs look like? How much of the total return came from relatively rare high-priced winners? Did apparent patterns visible after 500 bets survive another 5,000?

Perhaps the eventual conclusion is incredibly simple: blindly backing the draw loses money.

Fine.

The purpose of collecting data isn't to force every research question to produce a profitable betting system. The value is having enough properly collected evidence to find out what actually happened.

For now we have 140 bets, 40 draws, 100 losses, 61 competitions, average odds of 4.61, a 19-bet losing run and +34.74 units after commission.

We've also got an 11.50 draw in El Salvador and a 9.60 draw in the Norwegian second division doing a fair amount of heavy lifting.

That's a great opening chapter.

It isn't the conclusion.

Research Lab note: This is a live research project, not a betting system or betting recommendation. Early profit or loss is not treated as evidence of a sustainable edge. The collector exists to accumulate live-market observations under consistent research conditions for later analysis.