[Bao Jingyan]

Somewhat concerned by what seems to be some extreme variance. Here's the scenario:

I bet 1200 football matches over a 5 week period for a 113% ROI. All games were flat stakes, 1x2 markets, and odds between 2.0 and 4.5. I thought this was robust sample which seemed to indicate my handicapping was almost certainly profitable.

However, in the last 3 weeks I've bet around 600-700 games for a 91% ROI. Both these sample sizes seem significant. I can see nothing that has changed in between the two samples.

Am I overreacting to variance? I thought the liklihood of running at 113% over 1200 games betting on fairly short odds was basically neglible if 'true' ROI was negative, but then equally it seems extremely unlikely that if 'true' ROI was negative, I'd be able to run at just 91% ROI for a 600-700 match sample?

Do I have to assume that something has changed in the way I am picking matches, or is this within the 'to-be-expected' variance range?

[tukkk]

cool story

[LT Profits]

cool story

Why? Seems plausible? He is using ROI as literally meaning return, so when he says 113%, he means profit of 13% and 91%i is approximately a LOSS of 9%.

[Bao Jingyan]

Why? Seems plausible? He is using ROI as literally meaning return, so when he says 113%, he means profit of 13% and 91%i is approximately a LOSS of 9%.

Exactly. To be honest I wouldn't give a **** about variance if it was the difference between £213 or £191 back for every £100 wagered...

[tukkk]

Im sorry im not used to using ROI% over a period of time, im using it bet-based

If you say your ROI over a 5week period is (i dont understand if it is 113% or 13% over 5 weeks) then your ROI% per single bet is very small and your variance is huuuuge

[Bao Jingyan]

Im sorry im not used to using ROI% over a period of time, im using it bet-based If you say your ROI over a 5week period is (i dont understand if it is 113% or 13% over 5 weeks) then your ROI% per single bet is very small and your variance is huuuuge

ROI is an average over the time period and all bets. e.g. 1200 bets @ £10 a bet would return a total of £13560, or £1560 profit.

Basically every £10 bet is returning a £1.30 profit.

[tukkk]

in the gamblers world ROI meaning anything other than return per $ wagered is pretty pointless imo

[smoke a bowl]

Alright Tuk, we get that the terminology was a bit confusing. However now that we understand that what he meant was this(over a 1200 game sample his roi was 13% and then over the next 600-700 samples his roi was -9%), what do you think of his situation?

[tukkk]

given that you went 13% over 1,2k in just 5 weeks indicates a very small edge so you can easily go - approx 9% in the next 700 bets
im not saying its wrong but pushing over 30 games/day is a double-edged sword , you might not realize when you are -EV in some spots and you surely need a serious bankroll management to handle it all

[Bao Jingyan]

given that you went 13% over 1,2k in just 5 weeks indicates a very small edge so you can easily go - approx 9% in the next 700 bets im not saying its wrong but pushing over 30 games/day is a double-edged sword , you might not realize when you are -EV in some spots and you surely need a serious bankroll management to handle it all

Agree we are working with a small edge here whatever, but the idea was always to find lots of small edges rather than the occassional big one. Realistically finding a big edge (say 20%+) is extremely difficult in markets where bookies are already taking 8% juice. The theory is also that the magnitude of games should even out variance assuming the edge is genuine. Running at +13% for 1200 games was also enough to quadruple the BR staking just 0.8% per game - so low risk, high reward.

The thing I'm confused about here though is that the maths doesn't seem to add up. You say +13% is a small edge, but to achieve that over a large sample (1300) suggests it is close to genuine. In fact histograms suggest (unless I'm going wrong here) that if my true edge was 0%, I wouldn't achieve a +13% ROI over a 1200 game sample in 1000 rerunnings of the situation.

That doesn't mean that I wasn't running well during this period, but even conservatively (i.e., I was running very well) it still suggests a very strong chance of a genuine edge of around +7%, but then equally if this was genuine, then to run at -9% for 600 games (not as big a sample, but still signigicant) doesn't make sense either....

[Dark Horse]

Your sample size, for either period, is big enough. The problem is your test period. In a short time window it is possible to be hit by short term fluctuations. When the window is wider, at least a year, those short term fluctuations tend to even out, and you should have a better grip on your true edge.

[Bao Jingyan]

Your sample size, for either period, is big enough. The problem is your test period. In a short time window it is possible to be hit by short term fluctuations. When the window is wider, at least a year, those short term fluctuations tend to even out, and you should have a better grip on your true edge.

Good point - hadn't thought of it this way. But isn't this somewhat affected by range of bets. For example, if I was just betting English Premier League matches, even with a big sample, ROI may get distorted by a short term trend of say, high scoring matches. In reality though, I'm betting games on all continents at all levels - isn't a short term fluctuation that is consistent across the entire world as unlikely as simply running really badly for a long period of time? Unless a fluctuation is directly linked to time, surely it should even itself out over x amount of games, regardless of time.

I think a good example here is that the second 600 game sample occurred during the final few weeks of the season in most European leagues. This creates weird situations where certain teams have more motivation to win certain games, so could be an explanation for the drop in fortunes. But actually I've performed slightly better in leagues coming towards the end of the season than ones starting or halfway through. And regardless, at the same time I'm betting a load of games in leagues not near the end of the season, and the downturn is consistent in these as well.

Another problem with testing a static handicapping formula over a long time period is that you open yourself up more to the possibility of (a) long term fundemental shifts in the way football is played (b) changes in how bookies price games. Of course this is all the more reason to always be looking to improve, but doesn't it also somewhat negate the notion that time period of sample is as important as number of games?

[tukkk]

Running at +13% for 1200 games was also enough to quadruple the BR staking just 0.8% per game - so low risk, high reward.

The thing I'm confused about here though is that the maths doesn't seem to add up. You say +13% is a small edge, but to achieve that over a large sample (1300) suggests it is close to genuine. In fact histograms suggest (unless I'm going wrong here) that if my true edge was 0%, I wouldn't achieve a +13% ROI over a 1200 game sample in 1000 rerunnings of the situation.

That doesn't mean that I wasn't running well during this period, but even conservatively (i.e., I was running very well) it still suggests a very strong chance of a genuine edge of around +7%, but then equally if this was genuine, then to run at -9% for 600 games (not as big a sample, but still signigicant) doesn't make sense either....

Now you are stating that you had an ROI of 13% PER GAME ?!
Given that information I say its impossible to sustain 13% ROI when betting that volume, even if you achieve 2% ROI doing that, you´re basically god.

This thread has been really confusing

[LT Profits]

Now you are stating that you had an ROI of 13% PER GAME ?!
Given that information I say its impossible to sustain 13% ROI when betting that volume, even if you achieve 2% ROI doing that, you´re basically god.

This thread has been really confusing

Semantics again. Having a 13% ROI over 1300 games and having an average ROI of 13% per game is the exact same thing.

And I agree that 13% is an almost impossible hold unless you are in a small market, which soccer certainly is not. As Dark Horse pointed out, give the guy a year. He still has an ROI of approximately 5.3% over 2000 plays.

[LT Profits]

Semantics again. Having a 13% ROI over 1300 games and having an average ROI of 13% per game is the exact same thing.

And I agree that 13% is an almost impossible hold unless you are in a small market, which soccer certainly is not. As Dark Horse pointed out, give the guy a year. He still has an ROI of approximately 5.3% over 2000 plays.

Sorry, that was based on 13% over 1300 plays and -9% over 700 plays, Revising the former to 1200 plays as the OP stated yields about 4.9% ROI over 1900 plays.

[Bao Jingyan]

Now you are stating that you had an ROI of 13% PER GAME ?! Given that information I say its impossible to sustain 13% ROI when betting that volume, even if you achieve 2% ROI doing that, you´re basically god. This thread has been really confusing

Sorry for any confusion. Yes, for the first 1200 game sample, average ROI per game was +13%, by which I mean across those 1200 matches I was making £0.13 profit on every £1.00 wagered at flat stakes. I'm not sure how best to express this.

You are right, it would seem it is impossible to sustain that, given that for the last 600 games I've been losing £0.09 on every £1.00 staked. What is confusing me is that it also seems to be impossible to average +13% over 1200 games without a positive 'true' ROI, which makes the last 600 games somewhat concerning...

[Bao Jingyan]

Sorry, that was based on 13% over 1300 plays and -9% over 700 plays, Revising the former to 1200 plays as the OP stated yields about 4.9% ROI over 1900 plays.

Correct. I would be delighted with a 4.9% ROI over 1900 plays (especially given that these 1900 plays have come in two months) if it weren't for the fact that I've been losing money for over 600 plays now...

[tukkk]

What is confusing me is that it also seems to be impossible to average +13% over 1200 games without a positive 'true' ROI, which makes the last 600 games somewhat concerning...

I think so too

But what if something has changed ? for example maybe you have been betting too big underdogs lately and you maybe suck at betting them, but you didnt know that or whatever other factor, there is so much to analyse, but it will be rewarding

I suggest comparing your price to the closing line, if you are beating that, then you shouldnt be afraid of not profiting from this regularly, since you put in massive volume

[Bao Jingyan]

I think so too But what if something has changed ? for example maybe you have been betting too big underdogs lately and you maybe suck at betting them, but you didnt know that or whatever other factor, there is so much to analyse, but it will be rewarding I suggest comparing your price to the closing line, if you are beating that, then you shouldnt be afraid of not profiting from this regularly, since you put in massive volume

Have never bet a team longer than 4.5, so variance shouldn't be too crippling. Have been beating the closing line by average 6% consistently (during the +13% run and the -9% run), but there seems there seem to be other factors at work. Something could well have changed, but I can't for the life of me work out what it is, and has it really happened so quickly?

[tukkk]

Have been beating the closing line by average 6% consistently

with your average odds its equivalent to something like 15+ cents linebeating, right?
im sorry but something here seems to be really really off
is it even possible to get 30 shots at beating soccer for 15+ cents daily
i specialize in european soccer and these numbers seem to baffle me

oh yeah, and flat betting also

---
cant put the puzzle together

[xbalto]

i imagine the variance is a lot higher than it should be in theory because the conditions for each bet are different. different conditions for every game, maybe different betting decisions, etc. overdispersion http://en.wikipedia.org/wiki/Overdispersion

[Bao Jingyan]

with your average odds its equivalent to something like 15+ cents linebeating, right? im sorry but something here seems to be really really off is it even possible to get 30 shots at beating soccer for 15+ cents daily i specialize in european soccer and these numbers seem to baffle me oh yeah, and flat betting also --- cant put the puzzle together

Not sure what 15 cents linebeating converts to, but I am indeed beating closing prices by an average of 6%. However, this isn't neccesarily profitable on its own when you take vig into account. It is a start though. I'm not just betting European football, and in general avoid the biggest Euro leagues.

[Bao Jingyan]

i imagine the variance is a lot higher than it should be in theory because the conditions for each bet are different. different conditions for every game, maybe different betting decisions, etc. overdispersion http://en.wikipedia.org/wiki/Overdispersion

Thanks for that... does seem to make some sense in that variance definitely seems to be higher than would be expected from modelling. Still concerned about how homogenous the upswing and now the downswing seem to be though - 1200 bets of consistently steady winning, followed by 600 of consistently steady losing.

[smoke a bowl]

Not sure what 15 cents linebeating converts to, but I am indeed beating closing prices by an average of 6%. However, this isn't neccesarily profitable on its own when you take vig into account. It is a start though. I'm not just betting European football, and in general avoid the biggest Euro leagues.

FWIW 6% sounds fairly extreme. Can you give an example of what you consider beating a closer by 6%? If you are truly beating closers by an average of 6% you can't run bad enough to lose long term I don't care how anyone tries to refute that.

[Bao Jingyan]

FWIW 6% sounds fairly extreme. Can you give an example of what you consider beating a closer by 6%? If you are truly beating closers by an average of 6% you can't run bad enough to lose long term I don't care how anyone tries to refute that.

I would class backing at 3.7 when a team closes at 3.48 as beating a closing line by 6%.

[smoke a bowl]

I would class backing at 3.7 when a team closes at 3.48 as beating a closing line by 6%.

OK that makes sense but to me that isn't beating the closer by 6%. 3.7 or +270 or (27.03% to break even)) as apposed to 3.48 or +248 or (28.74% to break even) is only beating the closer by 1.71% the way I calculate it.

[tukkk]

That is exactly what i said, over 2k samplesize

-you are beating the lines at an avg of 15+ cents
-you make 200+ bets in a week
-you flat bet
-you dont seem to know much about analyzing your bets

[Bao Jingyan]

That is exactly what i said, over 2k samplesize -you are beating the lines at an avg of 15+ cents -you make 200+ bets in a week -you flat bet -you dont seem to know much about analyzing your bets

Sigh. As I say, I don't really know what beating a line by 15 cents means, as I am not familiar with US odds. I've explained what I mean when I say I'm beating closing odds by 6% (reply #25). This is what I'm achieving.

Flatbetting is the only viable option really when betting this volume. Kelly would get way to complicated, especially when the bankroll fluctuates so often and bets are placed several days in advance of KO.

But anyway, not really here to justify what I'm doing - was hoping that people might have suggesting on my original dilemma: i.e. running at +13% ROI for 1200 bets, changing nothing, and then running at -9% ROI for the next 600. If you do not believe this has actually happened, please treat it as a hypothetical...

[Dark Horse]

But anyway, not really here to justify what I'm doing - was hoping that people might have suggesting on my original dilemma: i.e. running at +13% ROI for 1200 bets, changing nothing, and then running at -9% ROI for the next 600. If you do not believe this has actually happened, please treat it as a hypothetical...

Not as unusual as you may think. Time to backtest it. Provided you have access to the lines, take it back one or more years.

[smoke a bowl]

Sigh. As I say, I don't really know what beating a line by 15 cents means, as I am not familiar with US odds. I've explained what I mean when I say I'm beating closing odds by 6% (reply #25). This is what I'm achieving.

Flatbetting is the only viable option really when betting this volume. Kelly would get way to complicated, especially when the bankroll fluctuates so often and bets are placed several days in advance of KO.

But anyway, not really here to justify what I'm doing - was hoping that people might have suggesting on my original dilemma: i.e. running at +13% ROI for 1200 bets, changing nothing, and then running at -9% ROI for the next 600. If you do not believe this has actually happened, please treat it as a hypothetical...

Like Dark Horse said, do some back testing if possible. I've learned through 10s of thousands of bets that numbers get defied every now and then. Every time you think you've seen the sickest run ever (good or bad), there is another run coming that will blow your mind. Long story short here is that you have 1800 samples with a damn good earn and that shouldn't be ignored. As long as conditions haven't changed from the 1st 1200 samples to the last 600 samples I would just chalk it up as a sick run of variance and keep plugging away.

[TomG]

Does your betting method calculate a fair line for you prior to making your wager? If so, calculate your expected ROI from your model and compare it to your realized ROI over your entire sample. Run a hypothesis test if you want to quantify it precisely.

If not, then calculate your ROI over your entire sample. Assume this is your long run ROI. Then calculate the probability of achieving +13% ROI over 1200 bets and -9% ROI over 600 wagers.

Without running any numbers, I suspect you started out running way over expectation and then ran way below expectation with the overall result being approximately in line with your model's long term expected results.

[Bao Jingyan]

Thanks for the various advice. Going to keep going with it - backtesting is a little tricky in terms of scraping all the data I'd need, but will certainly keep things going and try and analyse what I've got in some of the way suggested. Will report back in another 1000 games...

[hutennis]

Your sample size, for either period, is big enough. The problem is your test period. In a short time window it is possible to be hit by short term fluctuations. When the window is wider, at least a year, those short term fluctuations tend to even out, and you should have a better grip on your true edge.

There are a couple of interesting points here I'd like to have explained please.

1. Sample size big enough...

Since we all know what is "big enough" we should have pretty good feel of what is not "big enough", right?

I see it here all the time.
"LOL at your 120 games sample size" or "Stop crying about 40 games sample size" etc.
So the idea of what is not "big enough" is certainly there.

Well, I have a question then... Let's see.

After a 1000 coin flips the results are 545 heads, 455 tails.
Now we can safely say that coin is not fair. Heads have an edge.
Probability that they don't is 0.24% with corresponding z score of 2.82.

Now, lets reduce our sample size to just 10 coin flips.

Wow, this is not even funny, right? For sure it's well pass LOL as far as sample size. It has to be at least LMAO or maybe even ROFL.
Yeh, maybe. But the thing is that in those 10 flips there was not a single tail.
Do I need another 990 tries to conclude that coin is not fair and heads have an edge?
Or probability of 0 for 10 being 0.000977 with z score 3.09 should be enough?

And if it is enough then once again it's not the size that matters, but probability (z score) of observed results.
So instead of LOL at 40 games sample size we should ROFL at 1.93 z score.

Am I missing anything?

2. Test period. Wider window, at least a year.

What the hell are you talking about????

How would independent event (coin flip, draw of the card, result of football game, horse race or tennis match) even know about time period??? Why would it care????
Independent events are not humans. Not only they don't have memory, they have no idea what the calendar is either!!!!

Do you imply that results drawn from 10000 data points collected over a year are more valid then results of 1000000 Monte Carlo simulations done in 1.3 seconds?

[Pot luck]

2. Test period. Wider window, at least a year.

What the hell are you talking about????

How would independent event (coin flip, draw of the card, result of football game, horse race or tennis match) even know about time period??? Why would it care????
Independent events are not humans. Not only they don't have memory, they have no idea what the calendar is either!!!!

Do you imply that results drawn from 10000 data points collected over a year are more valid then results of 1000000 Monte Carlo simulations done in 1.3 seconds?

I don't believe that football games are independent events. The outcomes depend, to various degrees, on past results, performances and even future fixtures. In addition, early season games are played differently to end of season games where a team has had a full season playing with each other but will be suffering from fatigue.

[hutennis]

I don't believe that football games are independent events. The outcomes depend, to various degrees, on past results, performances and even future fixtures. In addition, early season games are played differently to end of season games where a team has had a full season playing with each other but will be suffering from fatigue.

Understood.

But everything you have mentioned and I'm sure some more is factored in a process of price discovery by market (oddsmakers + betting public) so at the end, right before kick off, the outcome of the game essentially becomes just another independent coin flip albeit with with different probabilities for heads and tails.

It could be 60/40 coin flip, but it's still independent of anything that has happened before, b/c everything that has happened before is figured in those 60/40 percentage.