Efficiency of Horse Racing Markets on Betfair

[spreadbetting]

One thing I have started doing recently is analysing the 'edge' my bets get against SP (as a function of amount staked) this is helping me pinpoint when the market at 'x' was wrong. Of course looking at individual races its almost pointless but over just 1 day you tend to get a value which holds true long term. Would love to see how this value (edge) varies from manual trader to trader and if it has any relevance to their pnl.

I do similar things to look for inefficiencies within these supposedly 'efficient' markets. The only way we profit from the markets as traders or punters is if we're getting value prices on a consistent basis.

A simple exercise for traders to find where they're obtaining most value is to download their betting history and simply sum the total of their back bets and the total of their lay bets. In an ideal world both figures should be positive otherwise, for various reasons, they're throwing away value when either opening or closing bets especially if they continually open from a fixed angle of backing or laying.

[Derek27]

One thing I have started doing recently is analysing the 'edge' my bets get against SP (as a function of amount staked) this is helping me pinpoint when the market at 'x' was wrong. Of course looking at individual races its almost pointless but over just 1 day you tend to get a value which holds true long term. Would love to see how this value (edge) varies from manual trader to trader and if it has any relevance to their pnl.

I do similar things to look for inefficiencies within these supposedly 'efficient' markets. The only way we profit from the markets as traders or punters is if we're getting value prices on a consistent basis.

A simple exercise for traders to find where they're obtaining most value is to download their betting history and simply sum the total of their back bets and the total of their lay bets. In an ideal world both figures should be positive otherwise, for various reasons, they're throwing away value when either opening or closing bets especially if they continually open from a fixed angle of backing or laying.

I can see where you're both coming from and the potential benefits of it, but I don't think it would help me much if I applied it to my trading history.

If I scalp a horse several times between 6.2 - 6.4, that horse could end up at 9 or 10. Such analysis would suggest the back bet was poor value and the lay bet was good, but it doesn't really matter where the horse ends up. Cold trading is largely about being safe and securing a profit without undue risk.

[spreadbetting]

I can see where you're both coming from and the potential benefits of it, but I don't think it would help me much if I applied it to my trading history.

If I scalp a horse several times between 6.2 - 6.4, that horse could end up at 9 or 10. Such analysis would suggest the back bet was poor value and the lay bet was good, but it doesn't really matter where the horse ends up. Cold trading is largely about being safe and securing a profit without undue risk.

Simply adding up your backs and lays has no relation with the final price of the horse if we assume your scalping could be equally split between drifters/steamers , it will just simply show if you're trading efficiently backing/laying and not throwing away value unnecessarily. If however you're continually laying drifters and closing early then it's something you should consider amending to increase your profits. I just assessed my last 3 months bets on my greyhound bot and the profit was more or less split equally over 55,852 bets so hard to see how I can tweak it. It actually surprised me just how close they were over a long period. Over the last month 19000 bets it a 75/25 split Back/Lay

[ShaunWhite]

The market is efficient or not depending on how you measure it. I make money from inefficient markets but can also prove they are completely efficient.

+1 Exactly.

[Iron]

Hi Derek

No, I didn't, but I don't think it makes a difference.

I think I can see where you might be coming from.

Let's say we had a million 5 horse races and assigned each horse odds of 5.0, from horses that should be odds on to ones that should be starting at 1000. By laying all of the horses at 5.0, then you would more or less break even (if you didn't have commission to pay). The overall picture would be one of extreme efficiency, but that would be misleading.

Similarly, if every time a market has a 15% chance horse goes off at 30% it makes an equal mistake in the opposite direction, you might expect the long-term picture to be one of extreme efficiency.

However, despite there being mistakes in individual markets, if we were to look at 10,000 horses that went off at 3.0, you would expect near as dammit a third of them to have won, which shows that, on aggregate, the market is extremely efficient (even though it might be wrong on individual races, possibly most of the time). That suggests to me that a hell of a lot of information is being accurately assimilated into the market.

Jeff

I think you may have applied the wrong test.

Did you apply random percentages adding up to 100% for each race?

[Derek27]

Hi Derek

No, I didn't, but I don't think it makes a difference.

I think I can see where you might be coming from.

Let's say we had a million 5 horse races and assigned each horse odds of 5.0, from horses that should be odds on to ones that should be starting at 1000. By laying all of the horses at 5.0, then you would more or less break even (if you didn't have commission to pay). The overall picture would be one of extreme efficiency, but that would be misleading.

It makes all the differences Jeff. If you didn't balance the overrounds you're probably comparing a 370% overround to a 100% overround - a comparison cannot be make.

As I explained in the coin-tossing example, looking at backing/laying all or a specific group, whether it's under 10s, over 10s, odd number saddle cloths, even number saddle cloths, horses beginning with the letter 'A', etc. you'll break even with a 100% overround, so the test doesn't give any indication as to market efficiency.

The true test of how efficient BSP is, is can a gambler with good knowledge of horse racing make a profit betting at BSP or prices at the off. The obvious answer is yes, and this proves that BSP is not efficient. If they were it would not be possible for a gambler to make much more than a few percent profit.

[Iron]

As I explained in the coin-tossing example, looking at backing/laying all or a specific group, whether it's under 10s, over 10s, odd number saddle cloths, even number saddle cloths, horses beginning with the letter 'A', etc. you'll break even with a 100% overround, so the test doesn't give any indication as to market efficiency.

It's an interesting point.

I take your point that, providing the odds add up to 100%, you will more or less break even long term if we discount commission. I was initially sceptical, but I've run a simulation on a spreadsheet, and it checks out, as the errors in one direction cancel out the errors in the other.

However, it's harder to explain away the fact that, for example, of the 126 horses that set off with BSPs of 3.0 in 2017, 34% won. Or the fact that, of the 241 horses that set off with BSPs of 4.0, 26% won. Does that kind of accuracy not suggest that, at worst, the BSP is there or thereabouts, over a large sample of horses?

The true test of how efficient BSP is, is can a gambler with good knowledge of horse racing make a profit betting at BSP or prices at the off. The obvious answer is yes, and this proves that BSP is not efficient.

It doesn't - it proves, at most, that the BSP can be inefficient for individual horses (which nobody disputes).

As an aside, I doubt that many people can beat the BSP long-term, but I could be wrong.

Jeff

[LinusP]

As I explained in the coin-tossing example, looking at backing/laying all or a specific group, whether it's under 10s, over 10s, odd number saddle cloths, even number saddle cloths, horses beginning with the letter 'A', etc. you'll break even with a 100% overround, so the test doesn't give any indication as to market efficiency.

It's an interesting point.

I take your point that, providing the odds add up to 100%, you will more or less break even long term if we discount commission. I was initially sceptical, but I've run a simulation on a spreadsheet, and it checks out, as the errors in one direction cancel out the errors in the other.

However, it's harder to explain away the fact that, for example, of the 126 horses that set off with BSPs of 3.0 in 2017, 34% won. Or the fact that, of the 241 horses that set off with BSPs of 4.0, 26% won. Does that kind of accuracy not suggest that, at worst, the BSP is there or thereabouts, over a large sample of horses?

The true test of how efficient BSP is, is can a gambler with good knowledge of horse racing make a profit betting at BSP or prices at the off. The obvious answer is yes, and this proves that BSP is not efficient.

It doesn't - it proves, at most, that the BSP can be inefficient for individual horses (which nobody disputes).

As an aside, I doubt that many people can beat the BSP long-term, but I could be wrong.

Jeff

You would be surprised Jeff.

[Iron]

As an experiment, I sorted all horses in the 2017 horse racing data file by ascending BSP, and then deleted duplicates by the 'race date and time' column, meaning I was left with just one horse per race - the horse with the lowest BSP in that race.

That prevents the over-round issue that Derek was referring to distorting the results.

The sample size I was left with was 13,000. Average odds 3.41.

The commission-free lay ROI over the sample was 0.22%, suggesting very high market efficiency at BSP over a large sample.

Jeff

[Iron]

Quite possibly.

The more I learn about the markets, the more I appreciate how little I know.

Jeff

You would be surprised Jeff.

[spreadbetting]

As an experiment, I sorted all horses in the 2017 horse racing data file by ascending BSP, and then deleted duplicates by the 'race date and time' column, meaning I was left with just one horse per race - the horse with the lowest BSP in that race.

That prevents the over-round issue that Derek was referring to distorting the results.

The sample size I was left with was 13,000. Average odds 3.41.

The commission-free lay ROI over the sample was 0.22%, suggesting very high market efficiency at BSP over a large sample.

Jeff

But what was the overround on those markets, Jeff, you'd really need to adjust the BSP to take into account any over/underround otherwise you may be simply getting your 0.22% ROI from the fact there are more layers than backers in the BSP markets. Just did a quick average BSP overround and it came out at 99.875% from just under 122,000 markets.

[Iron]

Hi SB

I also don't know how BSP is calculated - I'm not sure anyone outside of Betfair does - but my understanding is that it approximates to the average price of the money traded just before the off, so surely is a good guide to how efficient the market is at that point? I'm not saying you're wrong, btw - I keep an open mind.

Jeff

But what was the overround on those markets, Jeff, you'd really need to adjust the BSP to take into account any over/underround otherwise you may be simply getting your 0.22% ROI from the fact there are more layers than backers in the BSP markets.

[spreadbetting]

The BSP isn't set to return to 100% is all I was pointing out, Jeff, can be under or overround. So your ROI could just as easily be coming from the fact those markets were overbroke, unless you tweak your figures to that theoretical 100% they'd be flawed.

[ShaunWhite]

Only 536 people have actually been high enough above the earth to see it as a globe, perhaps, unless that is a conspiracy. But 7bn people live on it, I'm plumping for flat.

i dont believe for a minute your an idiot flat earther, so i can only assume this is part of your contrarian approach to trading (going against the crowd).

It's been a while since I've been at 40,000 feet, but I thought you could just about see the curvature of the earth from that height ?

i think you need a lot higher 15 to 20 miles

You can see it at sea level.

When a ship approaches you see its mast before the superstructure, or if you're approaching land you see the mountain tops first. Known since the first canoe ventured too far offshore, certainly by all the circumnavigators.

You're all only thinking of seeing the curve in one plane.

[Derek27]

Hi SB

I also don't know how BSP is calculated - I'm not sure anyone outside of Betfair does - but my understanding is that it approximates to the average price of the money traded just before the off, so surely is a good guide to how efficient the market is at that point? I'm not saying you're wrong, btw - I keep an open mind.

Jeff

But what was the overround on those markets, Jeff, you'd really need to adjust the BSP to take into account any over/underround otherwise you may be simply getting your 0.22% ROI from the fact there are more layers than backers in the BSP markets.

It is quite complicated but Betfair did explain how it works at the time.

A provisional price is calculated from SP bets. If it's below exchange prices at the off, exchange back bets get matched with SP money and that raises the SP. Conversely, if it's above current prices unmatched lay orders get matched to reduce SP.