Zero-Inflation Revisited

I had a couple of questions about Elo based ratings, one I can no longer find but fortunately I recall the gist of it, and the second was on my clarification about how I handled the zero-inflation problem in this recent post.

To address the second one first, the comment from Alkalmazasok was this:

Hi Cassini! I'm a bit confused here... You say you track the goal expectancy against the actual number of goals scored. The zero-inflation comes in when you compute the probabilities for different number of goals (with the help of the goal exp.). So how does the comparison of goal expectancies and the actual results say anything about the zero-inflation?

Zero-inflation is a problem with the Poisson Distribution when applied to football - essentially it under estimates the likelihood of a team scoring zero goals and to a lesser extent, of it scoring one goal. The input into the formula is the number of goals you expect a team to score, a number that will not usually be an integer, but something like 1.67341. When you plug this number into Poisson, it tells you what percentage of the time the time the team will score 0 goals, one goal, two goals etc. The numbers for zero and one are too low, and need to be 'inflated' with a manual adjustment.

Cassini Top 10

My solution to this, and there are most likely better ways out there, was to experiment a little and come up with an inflation formula that appeared to make the correct, or close to correct adjustments. In simple numbers, if the goal expectancy is 1.5, Poisson tells me that 22.3% of the time, the team will not score, and 33.5% of the time they will score one goal. If over a period of time my goal expectancy of 1.5 is on average correct, but the team scores zero goals 30% of the time, and one goal 36% of the time, then I can adjust my inflation accordingly. If my average is 1.5 and over a large number of matches teams are scoring 3 goals, that's a problem with the expected number of goals.

Club Elo Top 10

The first question was how do I determine the starting ratings for each team, and again, this is a best guess, although one good resource is to use the numbers already calculated by a site such as clubelo.com. Once you start applying your own bespoke adjustments to them, they will settle down soon enough. I started my own ratings about five or six years ago now, using league points along with the EA coefficients for each league as a starting point, and the rankings are not hugely dissimilar to those at ClubElo although my scale appears to be a little more stretched out. Paris St Germain suffer because Ligue 1 is rated the lowest league, but as points also transfer between clubs as a result of Champions League and Europa League matches, a run in the Champions League will help them. Since the ratings are used for domestic league games, how Bayern Munich compare to say Manchester City is of academic interest only anyway.

Weekly Draw Down

Over the last five full rounds of matches, I’ll actually include Wednesday’s AS Roma 4:2 Parma game to make a nice round 50 matches and easier percentage calculations, just 13 (26%) ended as draws. Exclude France, and the numbers drop to 8 draws from 40 (20%), as France saw five of Saturday’s six matches end all-square for at least something of a mini-tsunami.

In the midst of something of a draw-drought, I thought it might be interesting to look at where the actual draws fall each week, if time permits.

First the season numbers across the big five, led as usual by France with 29%, then followed by Serie A at 25%, with the EPL, La Liga and the Bundesliga all at 22%.

England

In alphabetical league order then, we will start with the English Premier League where the two draws last weekend fell at West Bromwich Albion v Cardiff City (3.82) followed by Arsenal v Manchester City (3.64) and neither of these two results was too shocking. The game at WBA was a high-scoring 3:3, with a dramatic finish seeing the hosts take a late 3:2 lead only for the visitors to score an even later equaliser. A six goal game ending even is always something of a fluke, but West Bromwich Albion are the top drawers in the EPL so far, having now drawn 14 of their 31 games. The Arsenal v Manchester City game was a more traditional low-scoring 1:1 affair, with a slightly weaker side (based on current rating and not league position) hosting a stronger side – conventional wisdom being that home advantage makes up the short coming in home team strength.

France

Ligue 1 was the pièce de résistance for the weekend, or more specifically Saturday’s six fixtures were, with no less than five resulting in draws. Europe’s current King of the Draws club Montpellier were of course featured (0:0 at home to Valenciennes at 3.44), as were Ajaccio 2:2 Toulouse (3.52), Nantes 0:0 Bordeaux (3.07), Reims 1:1 Lorient (3.21) and Sochaux 1:1 Olympique Marseille (3.23). Ajaccio , Reims, Toulouse and Bordeaux have all drawn at last 10 of their 31 matches to date, so it was more of a surprise that Evian Thonon Gaillard v Monaco (3.39) didn’t finish level! (ETG won 1:0) Unfortunately with a strike rate of 29%, selecting draws in this league is rather easier than most, and the prices reflect this.

Germany

In Germany, the two draws were both something of a surprise, although anyone who follows my Bundeslayga system won’t have been totally shocked to see Bayer Leverkusen fail to beat Eintracht Braunschweig and those who correctly value the motivational component of games may well have considered crowned champions Bayern Munich to be rather short to win a now rather meaningless game at home to a mid-table Hoffenheim.

While an understanding of human nature might have suggested a sub-par performance was in prospect for a game sandwiched between the highs of winning the Bundesliga in Berlin on Tuesday and the importance of a Champions League Quarter Final first leg at Arsenal a week later, few would have would have expected a previously 100% at home Bayern Munich and having with a goals for / against margin of 39 to 7, to come up with a 3:3 draw. The Pinnacle Sports price was 8.61 for the draw (the 13th longest in this league so far this season).

The slightly more predictable draw at Bayer Leverkusen was priced at 5.42. Bottom placed opponents Eintracht Braunschweig did win the home encounter, and picked up 6 points in their last 5 games, but with Leverkusen motivated by a possible Champions League spot, a draw wasn’t too many peoples expected outcome.

Italy

Serie A’s two draws were Samdoria v Fiorentina which finished 0:0 (3.34), Samdoria’s first such result of the season, and Livorno drawing 2:2 (3.58) with Internazionale - Fiorentina’s rivals for a Champions League spot and the current top-drawing side in Italy who thus missed an opportunity to close the three point gap.

Spain

Finally La Liga, where there were yet another two draws with the games at Osasuna (1:1 v Real Sociedad 3.44) and Villareal (1:1 v Elche) finishing level at 4.48. Villareal and Elche were both promoted last season, but with Villareal failing to live up to their nickname (Yellow Submarine) and playing anything like relegation candidates they might have been expected to gain three points from Elche, although the latter have been in draw form of late with four from their previous eight matches as they attempt to claw their way to safety. Aside from doomed Real Betis, nine teams are within six points of relegation. 11 draws from 31 matches puts Elche second in La Liga behind fellow strugglers Real Valladolid. Levante are actually the team with the most 0:0 / 1:1 draws as five of Real Valladolid’s draws have been 2:2.

Top Drawers Per League

With April here, there will be the inevitable odds-on draws in Serie A (and B) and an increasing number of matches where motivation is a factor. We’ve already seen the Bayern Munich v Hoffenheim game generate a result that would have been unlikely just a couple of weeks ago, and so their remaining fixtures will be interesting to monitor. As each week passes, more and more teams will find themselves with nothing to play for, sometimes opposing teams with something to play for and sometimes teams that are similarly not too concerned about the outcome, and maybe taking the opportunity to try out new players and combinations.

I haven’t seen this to be an issue in previous seasons, and with my rating based system an objective one, I am not inclined to make subjective changes. I would rather make a note that one or two teams don’t ‘need’ the win and review the results at the end of the season. I hasten to add that if you are planning on investing your retirement savings on such games, don’t.

Here Be Dragons

In the whole scheme of possible health issues, a back sprain is pretty low on the scale, but having done myself such an injury during a lunchtime run on Thursday, I have to say it is very painful and restrictive and the drugs prescribed don't seem  to be helping too much either. (They send me to sleep, but the pain persists during my waking hours). One plus is that I now have a lot more sympathy for my 80+ year old parents and their difficulties when exiting from cars. All this is a long-winded way of explaining why I have been a little quiet on here for the past few days despite being unable to go to work - just sitting down for more than a few minutes was most uncomfortable.

Feeling a little stronger this morning, so without further ado, here are the weekend updates for the FTL, and it does make for brighter reading for most entrants.

The "In Profit" entries are back up to six, and they are:

Every one of the six made profits, the big winners being Webbo and Fedslam who both made their fortunes tipping Crystal Palace to beat Chelsea. More of the same please boys. Fedslam moves up from £75 to within range of £225 in one fell swoop. Leaders Cassini Value Selections had two out of three, with West Bromwich Albion appearing to have the hat-trick all set with a 2:0 and then a late 3:2 lead, but they couldn't hold off the mighty Reds, aka Cardiff City Dragons or whatever their identity is these days. Skeeve had a profitable weekend too, and the unglamorous Bundeslayga selections made another small profit thanks mostly to Bayer Leverkusen dropping points at home to bottom of the the table Eintracht Braunschweig.

The "In Contention" section continues to be small in number, just three, and dominated by the XX Draws:

With 7 draws from 13 selections, and two more going down to 87th minute goals, it was a long overdue good weekend for these, with the draws up 10.39 points and up four places. The Unders version made a small loss, as did Hofs Hackers, but they remain little changed in 7th and 8th respectively.

As for the remaining entries:

In terms of points, the big winner was The Football Analyst +6.18 points, (bounty liability down to £275) and Football Elite was robbed by his selection of Crystal Palace as a DNB selections which under FTL rules is recorded as a lay of Chelsea. I believe Matt's Official numbers are a lot better than those reflected here.

The only other profitable struggler was Punters' Friend Neil who made 2.57 points. The biggest loser of the weekend was Peter Nordsted's Drawmaster which dropped 2.56 points and three places, a total that would normally be quite reasonable but in a good week looks rather bad. In fact Peter had the next two worst returns as well, with his Official Premier Betting account bets dropping 1.85 points, and the modified FTL entries down by 0.82 points. Rubicon and Jamie A dropped 0.16 and 0.03 points.  After 5,000 bets for the season, the totals now look like this, but at least some 36.34 points better than the last update, and at this rate another ten rounds of matches and we'll be in profit:

Professionals sub-set lookslike this:

The X-Factor

An interesting idea from the product development team at Betfair is the new variable handicap market. One issue with opening up a market for a particular point spread or total is that should the line change, a not uncommon occurrence in a sport such as basketball, the prices on both sides are no longer close to the 50/50, evens, 2.0 mark, but can be off by quite a bit.

The solution Betfair have previously chosen is to create new markets, but again, this new line can prove to be 'off' also, and you end up with several markets. Below is a screenshot showing four totals markets on one game, spanning a difference of 20 points, and two handicap markets.

The result of several markets is that liquidity suffers. More choice is not always better, and this is one situation where this is certainly true.

So Betfair's quite ingenious answer is to introduce the 'variable' market from today. I must admit, at first this made no sense to me. Variable speed limits might work well enough, but variable handicaps?

After reading the rules on this, it appears there is either a loop-hole, or I am missing something rather obvious.

Early results show that the optimal trading strategy is to back Team at as high a price as possible, and then lay at the lowest price possible before the game starts. A second strategy would appear to do the exact opposite, i.e. lay Team at the lowest price possible and back it back at the highest.

Team always wins, the tricky part is selecting the correct one, and determining the value of x. This is where that 'A' Level in Pure Mathematics and Algebra comes in useful; solving a simple equation for the value of x.

Betfair's Rules say "Who will win this match with the stated handicap applied?" but offer little help on how to solve for x, which is good for those with expertise in solving equations.

From reading Betfair's Press Release on this new market, it appears that they are anxious to return to the days where "Sharp Minds" had the advantage, 

Le Problème de la France

The deep dive into the state of draws this season continues, and one thing stands out. There is a problem in France.

I mentioned yesterday that a winning or losing system isn't necessarily an indicator that the system is any good. The phrase I used was that a rising tide lifts all boats, which basically means that a sequence of results either above or below the long-term and expected averages can be deceiving. Regarding draws in the top five European leagues, 2013-14 is almost certain to be a season long sequence with draws under represented across the board. One of the advantages of including five leagues is that the results should be less volatile than if you select just one league to follow. This season appears to be an exception with the draw percentages down on both the 5 year and 10 year averages in every league.

Here are the current season numbers and the 10 and 5 year averages:

It looks like the worst league should be England, the best France, but as the post title hints at, this is not the case.  The next screenshot shows this season. The last five full rounds (50 matches each) in each league are shown, with the number of draws and the returns if backing every draw in those 50. 

The "All X" column shows the futility of backing every draw, at least the futility of doing so at Pinnacle Prices, and the other columns show the results of backing the sub 4.0 draws and the performance of the XX Draws.

Finally, Betslayer left a comment suggesting that it is not the total draws that are hurting, but that: 

...surely it has to do with the number of low scoring draws not just draws? (as that is your target area). I have done a lot of research in this area and surprisingly this season is a great one for under 2.5 draws (across 14 European leagues).

Fourteen leagues is a significantly larger pool than the XX Draws 'draw' from, and it appears to be the other nine leagues where his statement that backing 0:0s or 1:1s is working this season. 

In England, Germany, Italy and Spain, the Under 2.5 goal draws are all down on last season, (England 16.3% from 20%, Germany 14.8% from 18.6%, Italy 18% from 20.5% and Spain 15.7% from 16.8%) and in France they are about the same (slightly up at 22.6% over 21.3%).   

FTL Midweek Carnage Update And Tsunami Warning

The midweek results were no kinder to most FTL entries than the previous weekend. One entry made a profit, with thirteen making a loss. The eight idle entries were the beneficiaries. I have added a new column on the far right showing the change on the previous published table.

The "in-profit" entries are down to five, and they are:

The Bundeslayga were the one profitable entry with the lay of Borussia Dortmund in the Revierderby a winner finishing 0:0. 

The "in contention" group is up to three members, and they are:

No positional changes, but all three lost as shown in the new column.

As for the "losers", you can see from the final column that all who were active lost a little more this week.

Peter Nordsted was the biggest loser with none of his Premier Betting selections winning, and no Drawmaster winners either, but Jamie A, the XX Draws and Rubicon didn't fare much better. The Football Analyst reduces his bounty liability to £300 by virtue of Peter's woes rather than his own success, and the Totals section of the spreadsheet makes for depressing reading. 

For those of us looking for draws, the last few weeks have seen what might be described in business terms, as a challenging environment. Skeeve left a comment on my post regarding his selections, or more his staking and multiples strategy, which ended with a PS:

I hope this upcoming weekend is the one when you nail all the draws and get back to profit :)

The problem I have with that statement is not the sentiment, which is appreciated, but that it implies the draws are there to be nailed. However this season, the draws have been notable for their absence across all five of the leagues I follow.

It doesn’t take a genius to figure out that a system looking for a particular result will struggle when the frequency of those results drops below it’s expected and long-term average, and will look great when they exceed expectations. The key is whether the selections being made are performing better or worse than a set of selections chosen at random. Your index fund might have been up 22% in 2013, but if the index it uses as a benchmark was up by 35%, that was a poor year for the manager. Something I will perhaps incorporate into the FTL next season is likely a benchmark or two.

Looking at the EPL, the last 50 matches have resulted in just 6 draws. Overall this season, this league is on course for its lowest draw total since the outlier of 2005-06 (20.3%). With the 10 year average at 26.3%, it's clear that a draw system will struggle, but unless something fundamental has changed, there will be times when the draws exceed expectations. In other words, when judging selections, it is important to look at more than just the P&L. A rising tide lifts all boats.

Backing all draws in the EPL would have lost 50.12 points. Limiting bets to the sub 4.0 selections would have lost 19.31 points (I used 4.0 because it is rare for an XX Draw selection to be priced greater than this) so it's some comfort that the EPL XX Draws are up 4.22 points. 

La Liga is my best performing league this season, and I will look at them along with the others this weekend. One plus to the poor run is perhaps seeing that the selections have value, even if the tide has ebbed. I'm hoping the receding tide it is due to a coming tsunami. 

Zero-Inflation

As the carnage continues unabated in the FTL table, (I'll update the numbers after tonight's matches complete the midweek schedule), I thought I'd address the topic of zero-inflation, a subject I mentioned in a post I wrote for Betting Expert back in 2012, and which a few people have asked me about since that time, including one recent request for help on Twitter. In my post, I wrote:

The intricacies of the Poisson distribution need not be fully understood for you to make use of it, because Microsoft’s Excel has a built-in Poisson function. Before we look at Poisson in action, it is important to know that several studies have found that the probability of draws is underestimated by Poisson.

The reason for this is that the probability of zero, and to a lesser extent 1, is under-estimated by Poisson, which is why an adjustment needs to be made to the output. A search of the Internet for ‘zero-inflation’ – a fancy term for increasing the probability of zero – will reveal a number of studies, and some ideas on how to apply this.

The zero-inflation in football is a challenge, and my attempts to solve the problem involved a certain amount of trial and error since there is, to my knowledge at least, no set formula for football models. Basically, I made an initial 'best guess' and then tracked my expected (predicted) scores against the actual scores. It's not a quick fix, but over a period of time (you probably need at least a season), you can see if your calculated goal expectancies are reasonably close to the actual results.

If they are not, then you adjust the inflation parameter accordingly. I track my goal expectancies in 0.1 increments so over time, an expectation of say 1 goals should see an average close to 1 goals. Some bands will be off by more than others, for example my 0.7 band currently averages 0.702 which is excellent, but my 0.9 band averages 0.995 which is not so good (off by 10.5%). If your expectancies are all lower or higher, then you need to make an adjustment, but if some are higher and some lower, the difference is probably noise, and the extreme expectancies will have a relatively low sample size.

Whatever numbers you inflate the zero and one by, a deflation needs to take place on the other outcomes to compensate since the combined probabilities of all possible goal totals cannot be greater than 1.0. The two goals bears the brunt of the deflation in my model, with less on the three and less again on the four, five and so on.

Stripped Naked - How Tips Are Best Viewed

A losing weekend for many of us, especially for the 'bigger' names in the FTL table. Myself, Skeeve, Peter Nordsted, The Football Analyst and Football Elite all had fairly sizeable losses.

Here is the profitable section of the table with a few comments:

The Cassini Value Selections were a rare ray of sunshine, making another 1.01 points with four winners from seven selections, and the three losing teams interestingly all scored first. A couple of selections in Spain couldn't hold onto leads, first was Real Vallodolid on Saturday and then on Sunday the slightly more anticipated and watched El Clasico where Real Madrid led 2-1 before it all got rather Messi - and Aston Villa let their lead over Stoke City slip as well.

Also interesting was that the prices on every selection shortened, most significantly Norwich City who shortened to 2.14 from an earlier 2.2.

Skeeve had a small  loss (0.88 points) on his FTL customised entry (more on that later), while his Official selections (also slightly customised) lost 4.01 points.

The XX Bundeslayga selections made a small profit of 0.25 points with one winner from two, while Fedslam lost 1.93 points.

No changes in position for the top five, but Webbo picked up 1.55 points to move into fourth cash place ahead of the XX Draws (Under 2.5) which lost 0.75 points.

I usually have a 'contenders' screenshot at this point showing the entries down by less than 10 points, but only Hofs Hackers falls into that category this week, so here are all the losers:

On a weekend where there were just eight draws in the 49 top five leagues across Europe, Peter Nordsted's Drawmaster managed to find two of them and moves up three places. The XX Draws managed to find exactly none of them for an embarrassing wipe-out, and drops down three places. Football Elite lost 3.43 points, but managed to move up one place, while Jamie A had a poor weekend losing 7.55 points. The Football Analyst lost 6.0 points last midweek, and another 3.81 at the weekend, and the bounty is now up to £325 with 13 ahead of him and sitting on a loss of 32.21 points on the season just behind Premier Betting who had another poor weekend. Their modified FTL entry saw all 10 selections lose, but the Official entry picked up a winner on a half-time market for a loss of 8.5 points. Neil took his total selections up to 650 but lost another 6.31 points

All in all a pretty awful weekend. Even Jonny G couldn't beat the lack of draws although he did hit two from 13 selections for a loss of just 5.79 points.

I mentioned Skeeve and his modified selections earlier, and here is why. In his weekend update email to subscribers, he wrote:

We're now at -31.6 with the doubles for the 2013/14 season so far, which is half of the recommended bank for the doubles (the Asian Handicaps are now at a small +3.2 point profit for the season after making a bit less than +10 point profit since January, so we're currently a bit more than a third of the bank down with both sets of picks combined). So they're obviously rubbish all of a sudden, aren't they? I've lost my edge, the bookmakers win, it's time for retirement, hello sunshine.

But wait, how come I'm one of the most successful services for the current season participating in Cassini's friendly tipster league with a +15 point profit at a 15% ROI? Does Cassini live in some sort of a parallel universe? This season, fortunately for him - yes, he does. The thing is that Cassini is recording my bets as one-point flat-stake singles (and ignoring the Skrill South parts of the doubles because of odds-recording issues). With so many half-won-half-lost double bets this season, it's easy to understand what makes the difference.

As Skeeve says, there are a few differences between what Skeeve says to do, and what I actually do. (I can be difficult, so I am told).

Firstly, as the Football Data web site doesn't include the Conference South, I discard those selections from the FTL entry, so if the South selections have been poor this season, that's my gain and Skeeve's loss.

Secondly, if Skeeve selects a double or a treble, I simply make that tip into two or three single bets, with one point on each. Skeeve might have the same selection in multiple bets - I only count the selection one time.

Thirdly, if Skeeve selects a team to win by -0.5 or -1.5, the bet is recorded as a single on that team to win. (Football Data don't include the AH prices). And, as I do for others, any DNB, (0), or +0.5 bets are recorded as lays of the opposition.

So whereas following Skeeve's recommended bets exactly as prescribed would mean a loss to date of 28.4 points, the more conservative approach of betting (non Conference South) selections as singles is up 14.16 points.

Betting Skeeve's selections as prescribed except to a level one point stake would be in profit by 7.29 points, which suggests to me that the problem is not with the selections, it is with the stakes and the combining of bets into multiples.

This weekend is a case in point - Skeeve had four doubles and an AH bet. All four doubles lost, a suggested stake of 10 points down the pan.

The four doubles were:

Kidderminster Harriers and Luton Town 4 points
Forest Green Rovers and Macclesfield Town 2 points
Forest Green Rovers and Sutton United 2 points
Macclesfield Town and Sutton United 2 points

So it's an expensive afternoon when Sutton United can't score at bottom of the table Dorchester Town and Forest Green Rovers can't score at home to Welling United and Kidderminster Harriers can only draw at home to Southport. Luton Town and Macclesfield Town both won, but their placement meant their wins were wasted.

Backed the Cassini way, you would have had four singles bets, and four points risked.

Kidderminster Harriers Lost -1.00
Luton Town Won +0.31
Forest Green Rovers Lost -1.00
Macclesfield Town Won +0.74

Net Loss: -0.95 points

Skeeve also had a selection of Dartford to win at Hyde (4 points at 1.99) which they duly did, although at a Pinnacle price of 2.07. Here's another area where Skeeve is leaking points. Those 0.08 points add up over the course of a season, and it is to his credit that his recorded prices are always on the low side. No Bet Victor / bet365 / Paddy Power or unattainable prices for Skeeve.

Backing in multiples and with multi-point stakes can easily work the other way of course, but I don't like them. To me, there is no logic in linking, say, Forest green Rovers and Sutton United together in one bet, and unless you know your edge, betting three, four, five points whatever can look very much like chasing. The only thing that makes a bet worth four points rather than one is that you have a huge edge.

The true measure of the quality of a tipster's selections is the return on them when they are backed as singles with level stakes. Stripped naked, their beauty, or ugliness, is all too apparent.  Had all Skeeve's selections (the multliples) won, his profit would have been 13.48 points. That's a 23.48 point swing on five selections, and you can see how such a possible variation in returns makes it hard to judge the quality accurately.

To conclude, I would say that stripped naked, Skeeve's tips aren't bad at all. The FTL compares all tipsters as equitably as possible, and compared to the big hitters of Football Elite, The Football Analyst, and Peter 'Weatherman' Nordsted, Skeeve is currently well ahead, and the only one in profit.

Six Down

Six years ago today, on a Friday the 21st March, the Chilean architect and painter Guillermo Jullian de la Fuente died, but the day is now remembered and celebrated as that when a young Cassini revealed himself to the world.

Or to put it another way, the Green All Over blog was launched.

1,746 posts later, and the darn thing is still going strong with 639,000 hits per my counter, and according to Blogger, some 807,000 hits. Will we hit a million by this time next year? Will the blog still be active this time next year?

The first two comments were from Graeme Dand (before he turned into TFA) and Mark Iverson, both still in short trousers at that time, and both still very much alive and well in the betting / trading world.

I have a standing joke that I trot out every anniversary, which might not be side-splittingly hilarious, but I like my traditions, so here we go.

Not all the 1,746 posts have been hugely interesting admittedly, but the other 1,745 have been...

Certainly not many betting related blogs have been around for so long, and I probably wouldn't have expected to still be here six years on back in 2008. There are a few reasons why I'm still here. One is that I am still profitable - as any blogger will tell you, it's easy to blog when things are going well, not so easy when they are not. The golden days have gone with the Super Premium Charge, but it was always just a part-time hobby, and in many ways the money was secondary to the intellectual challenge of trying to find edges and competing with others.

It helps that I am multi-dimensional so far as my interests in sports are concerned too. If your focus is on one sport, it's not so easy to find new angles to discuss.

The blog has evolved too, as have my Elo based ratings. Looking back, some of my ideas were rather basic, (OK, embarrassing), but there is no shame in that so long as you learn from your mistakes and improve.

The FTL gives me regular content too, at least during the football season, as do the comments and mentions I find of the blog around the blogosphere. And then there are my 'campaigns' to call out nonsense, unfairness or wrong doings where I see it, and in betting, there is a lot of that.

And this year of course, there was one individual's attempts to silence me with his ludicrous claim to have reported me to the police - my 'crime'? Linking to another blog post. That was an interesting saga which generated a lot of additional hits.

Oh yes, and a thick skin helps too. I'll be out celebrating for a couple of days, but I will be back for a seventh year of stimulating content. Happy Spring.

Tingly Loins

That Steve over at Daily 25 is becoming more and more of a character each day. In his latest post, on the topic of Graeme Dand's 52 systems that make up the TFA, he came up with a couple of corkers.

I like my blog writing the same way I like my sex, in and out quickly and leaving someone very unsatisfied.

The hint that there is possibly a connection between blogging and sex is interesting. Some blog daily, some blog more than once a day, while others need longer recovery periods between posts.

Graeme does have amazing stamina when in action, but he's not in action that often. Weekly is usually good enough for Steve, but his choice of words is sometimes disturbing:

I know Cassini’s loins get all tingly when I mention him on this blog, but he has a timely blog post which helps with this point. He was saying how Peter Nordsted (A tipster) had posted an under bet for subscribers to his service and then posted the opposite side on twitter. This is a blatant example of what I was trying to get at. With so many different bets (and overlapping bets) it is easier for a tipster to point to the winners. 

As I mentioned, the main theme of Steve's post wasn't my loins, but Graeme's abundance of systems. It's something I've commented before, and it does seem unnecessary, especially when you consider the limited number of matches that make up the pool Graeme draws from each week. 46 league matches plus 12 Conference Premier games generating enough bets to warrant 40+ systems does seem a little excessive.

It's not like an investment fund choosing from thousands of companies of different sizes and across different sectors. This is one country and the top five leagues in one sport.

Of course, Graeme's selections can and do appear in one or many systems, but it does make it difficult to assess how good the ratings are. For me, the measure of a system is one point per selection - are you up or down?

When you vary stakes and combine selections in multiples or select opposing selections for different audiences, it's easy to see why accusations of smoke and mirrors have some merit. Although not perfect by any means, the FTL does at least apply one point per selection, all of which are singles, and uses the same source for all prices allowing a like for like comparison to be made. Yes, Pinnacle's prices can often be beaten, and Draw No Bets and Asian Handicap bets have to be recorded in ways that the selector didn't advise, but over time, who is finding value and who isn't will show through.

The other point I want to make here is about how it is possible to find so many value selections from such a relatively small list of selections. My pool is the top five European leagues, usually 49 matches per week, and of those there are maybe 13 or 14 are draw / under value bets on average, and the Value Selections for this season total 75 - which is 0.054% (1 in 20) of the matches. I do look for a 10% minimum edge on these bets, and maybe should settle for less, ROI for show and all that, but I'll see this season out before making any changes.

A comment on the Football Form Labs guest post, more of a sales-pitch than I was expecting, misses the point that most football bettors are not full-time syndicates with millions of other people's money to invest, but part-timers / hobbyists looking to make a few extra quid for themselves:

Great stuff Will > high turnover and low margin is the only way to win in the long run which is what every betting syndicate around the globe is doing.

Pointless having 4/5 bets a weekend . It will not work. You need a big rod in a big sea.

The statement that "high turnover and low margin is the only way to win" is clearly incorrect, and I would disagree with the 4/5 bets a weekend being pointless line too.

If the point is to make a little extra money and not spend all weekend watching sports, then I would say this is the perfect set up. What "it will not work" means I'm not sure. For many of us, it works just fine. What one person is looking for in betting, as in life, is not necessarily what others are looking for.

On the subject of the Fidens Fund, I had one email expressing concerns. It's not a new idea, and Centaur Sports famously went bust in 2012 doing the same thing.

Back when the fund launched in 2010, the directors of Galileo told CNBC that they were using high-level software that took five years to develop, and expected to a 15-25% return on investment.

But last week investors got a letter saying the fund was being liquidated and all the money was gone.

On paper, this seems like it should have worked (Mark Cuban even endorsed the idea in 2004). Plenty of professional gamblers make their living handicapping sports and finding value in point spreads. And with all sorts of cash and technology at its disposal, you'd think a fund like this could generate some decent returns.

So what went wrong?

According to the Journal, the fund told investors that they lost it all due to "sheer bad luck".

Sheer bad luck, or just an idea that is fundamentally flawed? Sports betting markets are a small fraction of the size of financial markets and strategies that work in the latter won't work in the former. In betting markets, it's an advantage to be small - your £20 can be easily bet at top prices, your £20,000  not so easily, and your £75,000 or more? Prices will start to shorten and your edge / value will decrease.

If Fidens do have a sustainable edge, why would the operators not just invest their own money and compound it?

At the average edge of 3.76% they claim (over 9,400 bets), an edge found on 3,000 bets a year, 'just' £1million would be worth about £1,757,596 after a year, if risking the stated conservative 0.05% of the total fund.

Why they would need those additional minimum £20k deposits is the big question prospective investors should be asking. The additional money won't mean better prices - there's no bulk discount in betting.

An edge of 3.76% over 9,400 bets is very impressive, as is finding 250 value bets a month, but as stakes rise and best prices are that much harder to get on at, that number can only come down.

28 investors with just under £1.5 million invested means an average investment of around £50k. The 'performance fees' applied at the end of each quarter (at a rate of 50% on the profit in that quarter) would generate, using the numbers provided, about £3,785 per account. Multiply that by 28, and goodness! That's £424k a year.

Stand by for news of the launch of my new Cassini Value Fund. My loins are tingling.

Grab Your Raincoat

I seem to have a few threads going at the moment, which is good. At my age, it's good to keep the brain exercised. 

No prices or rationale yet from Peter Nordsted and his 'opposing tipping' strategy, but Hejik had this comment:

Good post Cassini and as usual I'd agree with much of it.

I find Peter's methods of operation astonishing to be quite honest and am extremely confused as to why he is so widely respected as a 'Tipster' and 'Betting Writer'.

I must stress this is merely my opinion but I've found his level of knowledge regards all things betting somewhat wanting and most of his ideas are just back-fitted nonsense and apparent guesswork (rigidly banding teams) and the sort of thing I was trying to make work in my first 6 months of betting.

The evident lack of progress and knowledge development over such a long period of time is quite startling and his approach to self-improvement, in a betting context, would appear to be denial of the very need for it.

I stress there's no personal ill-will as Peter actually seems a real gent on the face of things, but when you put yourself up as some kind of expert yet it's apparent to experienced people that you're anything but, you must be prepared to accept an amount of criticism of your ideas.

As I've said elsewhere, recent events would suggest that people need to "wise the hell up" and start questioning things a bit more.

The rigid banding idea is, of course, nonsense as Hejik says. Whether it is Peter or anyone else, the logic is akin to that of deciding to carry an umbrella on March 25th every year, because since records began, it has rained on that day more often than not.

To my twisted way of thinking, it is rather more relevant to check the weather forecast that morning. The weather on March 25th 2013, 2012, 2011, ....1892 is data, but it's useless data if you want to know what to wear on March 25th 2014.

Last night, Pete's 'model' suggested a bet on the Colorado Avalanche @ 1.91 playing at the Winnipeg Jets. I say 'model' in quotes, because to me, a model spits out a probability, and you bet if you can find a price that beats that probability.

A model is not something that tells you that March 25th was wet 70 times in the last 100 years, and 4 times in the last decade, is 80% wet in Leap Years and in years ending in a 1, it has rained every year except when there was a World War in progress or Halley's Comet was paying Earth a visit. That is trivia, all very interesting to some perhaps, but it is not useful data to base your clothing choice on.

Did Peter's 'model' suggest Colorado should be priced at 1.81, and thus (allowing for 5% padding) 1.91 was value? No. The selection was made because it just so happens that a narrow band which includes the 1.91 opening Colorado price, has in the past been profitable.

A subsequent email from Peter said:

I have also had an email on whether Colorado is still a pick as they have drifted out to 2.04.

The answer is yes as the selections are all based on the opening prices.

Presumably the answer would have been yes if the question had been "Is Colorado still a pick as they have shortened to 1.09?" since the selection is based on the opening price?  At what price are these selections recorded in the record books? 1.09, 1.91, 2.04?

Selections have to be based on value, (using opening price is ridiculous), and if you don't have a clue what the 'true' price should be, you are just wasting your time and ultimately money, although I doubt that anyone plays a 'system' like this for real money.

Colorado lost 4:5 after leading 1:0 by the way, although irrelevant to the issue.

Like football clubs, weather patterns change from week to week, month to month and year to year. Using previous weather records for a specific date as a predictive tool is as silly as using Manchester United's record versus Liverpool since 2005 to calculate the probability of a win against them in 2014. That it rained heavily on April 1st in 2012 and 2013 doesn't mean it will rain again in 2014. Look out the window and check the forecast if you want to know whether to grab your raincoat.

Danny left this comment:

I am not a fan of using current form for a couple of reasons.

One is that it's something punters put too much emphasis on and it tends to drive the price on a team too low as a result. And it's even worse if the team has a big following, try getting value on Leeds for example if they have won their last four matches.

Another reason is if you are talking about the last six games you are looking at a very small sample and luck may well have played some part. Also home and away form are very different animals so is there much point in lumping them both together?

Market theory would suggest teams in good recent form could get over backed so I would argue price action is a better way to find value than form.

Every team has it's price!

When have Leeds ever won their last four matches? If form leads to teams being backed too short, then there is an opportunity on the lay side. The trouble with basing selections on price is that you are effectively trying to reverse engineer the selection.

If you input your (useful) data into your proven model, it will generate probabilities which can be compared with the market's assessments, and value bets jump out at you.

If you start with the prices, it's not so easy to objectively say what the true price should be and explain why the market has it wrong.

Colorado are not value at 1.91 because the market at one time thought they were a 1.91 shot. They may be value at 1.91, or they may not. Had they opened at 1.89 would they have been value? 1.9?  Such arbitrary distinctions based on lines drawn in the sand while looking back, make no sense.

2% chance of rain today according to the forecast. I'll leave the brolly at home. 

Football Form Labs Guest Post

My recent post in which I took a quick look at Football Form Labs triggered a response from Will Wilde, the Head Trader from that organisation. Since I have no experience of them, and my observations were based on a very quick look at a few web pages and screenshots, when Will asked for the opportunity to guest post here, I had no problems at all.

Here is Will's post, unedited and in full, and its presence here should not be taken as an endorsement of their products or trading services, merely as another opinion from the other side.

Thanks Will.

===============================================================
Hi Cassini

Thanks for taking the time to publish your article on Football Form Labs.

I can’t see on our logs that you have actually taken out a trial of either Form Lab MAX or Form Lab Black. Both are available on a free 14-day no-obligation trial. You can cancel the subscription at any time during the 14 days and you won’t be charge. If you haven’t you might find this reply a bit confusing and long-winded!

Form Lab Black launched in May 2011 and we have been providing Daily Report recommendations since then. You can see a list of all Daily Report recommendations that we have published on the P&L section of the Daily Reports tab in Form Lab MAX and Form Lab Black.

Form Lab MAX contains Daily Reports, Game Notes, Match Data and Pre-Match Query. In addition Form Lab Black includes In-Play Query, Players and Live Logic additionally. To avoid me droning on the two user guides are published in full here:

http://downloads.footballformlabs.com/formlabblack/welcome.pdf

http://downloads.footballformlabs.com/formlabmax/welcome.pdf

In short, Daily Reports are editorial recommendations, trends and identification of key players missing from that day’s matches. Match Data gives a snapshot including 20+ Form Tables for all of the 50+ leagues we cover, Game Notes are an automated selection of a strong trend or trends in a match (not necessarily signifying value), Pre-Match Query allows you to query our 50+ leagues over the past five seasons, In-Play Query (as the name suggests) does the same as Pre-Match Query but allows you to add the dimension of any match situation into your query (based on time elapsed, match position etc), Players lets you query the effect of an individual or a combination of players across nine leagues over the past five years (for example you could query the database on how Manchester United do when Rooney and Van Persie start since Van Persie joined Manchester United when away from home: W16-D6-L3 when both have started netting on average 2.16 times per match compared to W4-D2-L3 when one or both have been absent as they’ve scored just 1.67 goals per game. Obviously we have sample size issues but you get the idea), Live Logic is the smartest of all of the tools available in Form Lab Black. Live Logic is customisable in-play bet prompts every five minutes or whenever a goal is scored for all 50+ leagues that we cover.

With any software of this nature it is very easy to pick ‘holes’ and therefore some intelligence is required when deciding the query you should make and how much reliance you should put on the past results.

Onto your questions about the Pre-Match Query output, on the Pre-Match Query tab you will see you have a number of options to choose when querying a match. For example, if you toggle between CP (current position) and FP (finishing position) on Position or, the generally more useful, Opposition Position your query results if FP is selected will only bring up matches where you faced teams that ended the season in that position, regardless of where in the league they were when you played your match during the season. For example, your games against teams that finished in the top-10 would only show up in a query of Opposition Position FP Top-Half. This isn’t to say that Current Position is of no use, as that team was a top-half team at that point in the season. Naturally, you encounter issues at the early stages of any season and those games aren’t included in the algorithm that produces the automated Game Notes. You can also query on Form, Time of the year/season and under the current manager. The ability to see how other teams may have performed when in a similar run of form or league position as the team in question, and at a particular stage of the season, is also a very good indicator of whether the price on offer is value. As an aside, on any of the query tabs on Form Labs if you hover over the ‘i’ you will see Tool Tips on how to get the most out of the software.

Regarding your comment about the output page you will see that in any output query you have a Summary of the results below in the box above. The Odds Conv row is just the odds conversion of the percentage. Nothing special there, but to those less au fait with decimal odds than yourself it can provide helpful. The much more interesting row is the P&L row which is what would have happened had you backed that result in the matches you have chosen. These odds are based on the average odds across 25 bookmakers at kick-off. The four-game sample you mention isn’t interesting from an odds conversion or P&L basis however Pre-Match Query allows you to query not just on the teams in question but also the full league. So for instance if we select ‘All Teams Home’ in Pre-Match Query and ‘Seasons 08/09-13/14’ in ‘Period’ and Generate for the Premier League we can see the Summary of all 2191 Premier League games since 2008/09. You can narrow this query down again by selecting team / form type or, perhaps, a period in the season. In this example, the Odds Conv and the P&L are more interesting than a four-game sample of Swansea’s home games.

Football Form Labs was designed as a betting tool to help bettors. Both Form Lab MAX and Black give users access to our database in an easy-to-use interface where they can form their own strategies and quickly bring up relevant past results from a huge variety of domestic leagues. With the addition of a consistently profitable editorial section (Daily Reports) we feel that at £30 a month for MAX and £60 a month for Black (with discounts available for quarterly or annual subscriptions) that they offer great value to the keen bettor.

Fidens exists as a separate entity to Form Labs. I won’t regurgitate everything that is written in our prospectus: http://fidens.footballformlabs.com/fidensprospectus.pdf but Fidens has many great selling points:

1) Is completely transparent. Investors can see every bet that has been placed on their account through a simple online login.

2) We don’t hold any of the investor’s money. We use an independent broker to place all of our bets. It is all held in the investor’s broker account and they can withdraw their investment without penalty at any time.

3) All profits for UK investors (and investors in many other countries) are completely tax-free.

4) We are extremely low-risk placing less than 0.50% of FUM on each trade.

5) We are extremely high-turnover. We bet in 27 leagues around the world and place over 3000 bets per year.

6) Our average edge per bet over 9,400 bets is 3.76%.

Fidens is unique. Syndicates of the past have tended to be secret, unscalable and unsustainable but Fidens is completely different. We currently have 28 investors with just under £1.5m under management. I’m happy to provide referees (away from here) if you wish. Fidens, in its current format, is scalable to approximately £15m under management and we hope to be at that level within 18 months.

Thanks again,

Will Wilde

Head Trader, Football Form Labs

Eyebrow Raising

As promised, here is a more detailed update on the last few days of FTL action going back to last Tuesday (March 11th). First, the profitable entries:

The Cassini Value Selections had a poor weekend, with just one winner (Stoke City) and three losses, for an overall loss of 1.71 points, but still a healthy ROI%. Hoffenheim managed to lose a 2:0 lead as well, which hurt. Augsburg lost 1:2 (after leading) and Elche drew 0:0.

Skeeve made small profits (less than a point) on both entries while the recent run of good results for the XX Bundeslayga system added another 4.44 points with three winners out of three. Fedslam  made 6.62 points to remain in 5th place (3rd money spot) but a large lead over 6th place and the XX Draws (Under) selections which were up 2.25 points.

The zero to -10 points group is:

Once again, a reminder that my results for Football Elite treat the DNB bets as lays of the opposed team, and use Pinnacle Sports prices and so will be different from their official results. As mentioned yesterday, The Football Analyst was the big loser here, and I shall spare Graeme the agony of repeating myself. The XX Draws had a roller coaster weekend which started with four winners from five (including a fluked 3:3) and everything going swimmingly before they went the remainder of Saturday and Sunday without a single draw from 12 matches. Monday was almost two for two, but a 90th minute goal by Napoli took away one winner and they ended with an overall loss of 1.07 points from 18 matches.

 Forza Fizzer emailed me to announce that he is retiring gracefully from the FTL. He led the table in early November, up by 13.72 points at that time, but only two (both less than one point) profits since then have seen a slow slide down the table. The fizz has fizzled out.

Funnily enough, Forza Fizzer has had almost exactly the same number of selections as Premier Betting and Rubicon, and is in the same ball-park on results too, with a few more wins to his name.

And then there are the Jonny G draw selections which after 34 selections, are improving, but down by 14.48 points after the latest round.

As most readers will know, I do like my draws, and it's strange that the draw-hunter entries (XX Draws, Drawmaster and now Jonny G) are all finding them hard to come by, yet Football Analyst Graeme is tripping over them. He has 11 Draws Systems, and all 11 are in profit.

And speaking of multiple systems, Anonymous commented that:

Considering you have about 4 separate entries in your own league table it's a bit rich to criticise Pete. I'm sure many people running systems will have 'overlaps' or multiple selections just because of the way the data is interpreted. Maybe Peter's brief from Matchbook sometimes means the result will be at odds with his own view and much like you cherry picking and highlighting whichever of your 'entries' is doing best Pete will do much the same.

Not for the first time, Anonymous misses the key point, which is that this issue is not about cherry-picking, overlapping selections or data interpretation - it is about giving contradictory tips. Diametrically opposed selections. Heads and Tails. Both cannot be value. 

Not that it is relevant to the debate, but yes, I have four separate entries in the FTL table, of which two (the XX Draws and Under) are closely related (more cousins than siblings perhaps) and of the other two, one is specifically targeting Home lays in the Bundesliga, while the other is looking for Home winners. There is no conflict of interest. There could well be an overlap, for example if the Away price is so far off that both the Draw and the Home prices could be value, but you won't see a contradiction such as my tipping Heads on here and Tails to my subscribers. 

The issue is that if you have a profitable model, as Peter claims (although some eyebrows might be raised* at that claim given the losses so far this season) it makes no sense to over-ride the (profitable) model by tipping the opposing (and thus losing) selections. Why would an experienced professional allow subjectivity to question the output of a long-term winning model?   

As a reminder of the issue, the match in question was Saturday's Aston Villa v Chelsea match. Premier Betting Account subscribers were advised:

Fair enough - Pete's selections are from a model so the Unders and the Draw, overlapping selections, were both identified as value. If Under 1.5 goals was value at 3.75, the Under 2.5 would have been value at around 1.93. Allowing for an edge of 10%, this means that Pete's model would have generated a price of 1.75. Even if Pete works off a narrower 5% edge, then we have a price of 1.83 for the Under 2.5 and the Over would therefore be at about 2.20.  

Pete's Matchbook preview concludes:

So my recommended trade for this game is Over 2.5 goals at around 2.00.

How does this make any sense? The 'profitable' model says the true Overs price is 2.2 - so value (assuming a 5% edge) would be at 2.31, and that is a LONG way from the recommended 2.0.

Either Peter is recognising that his model is not, after all, profitable, and has no confidence in its selections, (bad news for Premier Betting Account subscribers) or Pete is advising bets he knows are not value on Matchbook (bad news for Matchbook's readers). True, the Matchbook reviews are just noise, with no real substance to the 'analysis', and are there just to attract clicks and bets, but to suggest what you know is poor value as a recommended bet seems very poor form.

Although confirmation of Peter's model's price on the Under would be nice to know, we don't need it. Which ever way you slice it, one of Peter's audiences is getting poor advice. The issue is not that the advice overlaps; it is that the advice contradicts, and to maintain credibility, you need to be consistent. It would be nice to get a clear explanation from Peter, but I'm not sure there is one.
-------------------------------------------------------------------------------------------------------
* I once told a friend that she drew her eyebrows on too high. She looked surprised.