Beating Betfair Book Proposal

Introduction

The horse betting landscape has changed dramatically over the last 20 years. The internet, as in many other areas of life, has redesigned the way we bet. Back in 1997 I was teaching an introductory class in Internet based programming for the web at De Montfort University. To kick off the first session I wanted to take an overview of the way the internet was heading and what opportunities it offered. I needed an example of how it could radically change our way of doing business and my interest in betting gave me a perfect example. I told the students that selling jeans or t shirts was all well and good and certainly had a place on the net but the real winners would be those people who invented something that people wanted but before the internet, could not be fulfilled. As way of an example I showed them Interbet.com. You probably have never heard of it. It was a US based site and to my knowledge was in fact the first internet person to person betting site, well before Betfair emerged. I have to admit it was a bit of a ghost town in terms of liquidity but it allowed me to show the students a prime example of something totally new that could turn an industry on its head. Of course there are better examples of totally new products like Facebook but betting even to non betting students was something they could understand. I am not sure if I ruined the lives of any of those students but there have been no court cases so far.

Betfair arrived around the turn of the millennia and its arrival revolutionized the way we bet. If you cannot agree with that then you have sadly been left behind or perhaps I should say its has changed your betting but you are blissfully unaware.

Exchange betting is not the only tech’ led initiative that has changed how we bet. Back in 1986 I was making regular visits to London’s Colingdale newspaper library. On arrival I would fill in a form requesting a period of Sporting Life back issues, say January 1975 to July 1975. After waiting around half an hour, if you were lucky, an old guy would appear wheeling a trolley. The trolley had two purposes, to stop the old guy from falling over and to deliver newspapers to the waiting customers. When you finally got your hands on 6 months of back issues you set to work like a Dickens character on the night before Christmas, analyzing the current system you hoped would take you off this small island and onto a privately owned one. If you worked quickly without breaks you might get a second six month batch in but usually it was back to the train station.

Today we can buy or scrape data from the internet, write or buy some software to analyse it and get the answer that took me months to find out in a matter of seconds. In this data driven age betting has not escaped the big data revolution and I would suggest that the most prized skill any would be successful punter can possess today is Computer Programming skill. It gives you access to the three main skills I aim to introduce in this book.

The first is data gathering via web scraping. You can buy commercially, out of the box data for horse racing but web scraping allows you to gather data that perhaps is not currently available in such a format. We will use Python programming language to gather data.

The second skill is data modelling. Gathering data is all well and good but what are you going to do with it. How are you going to utilize it so that on a daily basis it will tell you what horse to bet and perhaps what horse to lay. To answer this question we will look at Machine Learning using Python once again and the Python Scikit-learn library.

Finally if you intend to spend time on more meaningful activities like installing solar panels on that island of yours you will need the computer to automatically carry out your daily modelling and auto execute your bets and lays. Once again Python will be the vehicle of choice along with the Betfair APING.

If you have done some rudimentary programming before then what follows should be within your compass. If you have not done any programming then I will suggest a decent Youtube introduction which will give you a foothold.

If you think this book may be for you then leave a comment in the comments section or you find me on Twitter at @smartersig

Brier Skill Score and Horseracing

Tags

Machine Learning and horse racing

When developing machine learning models for horse racing we quite rightly need some way to evaluate how successful they are. Horse race betting is a bit different to predicting face recognition or breast cancer diagnosis because these applications are all about accuracy of predictions. With betting, accuracy clearly has a part to play, but its not the complete picture. A less accurate model (in terms of predicting winners) could be more profitable and for that reason we tend to focus on profit. The two most common profit measurements are flat stake profit and variable stake profit. The former simply means putting £1 on every selection, for example the top rated in our ratings. Variable staking means we place a stake set to win £1 relative to the odds. So for example a 2/1 shot would have a bet of 50p placed on it. Of course in both examples the stake can be whatever you want it to be.

The advantage of variable stake monitoring is that it is not prone to inflation from one or two big priced winners which may give you a never to be repeated profit that sends you skipping off to remortgage your house. The variable stake monitoring does not suffer from this and gives a more realistic impression of possible future performance.

So what about the more traditional Machine Learning performance metrics, should we bin them when developing ML models and simply focus on profit/loss ?. Probably not, a mixture of metrics can help give us more confidence if all of them are showing improved signs over a rival model.

Horse Racing models often have a degree of inbalanced data. That is is to say that the thing we are trying to predict (win or lose) usually contains far more zeros than one’s, after all our lines of horse data will clearly contain more losers than winners unless we have engineered the data in some way.

One metric that is useful for inbalanced data sets is the Brier Score and what I am about to describe is its close cousin the Brier Skill Score

First of all what is a Brier Score. Imagine we have a three horse race with the following

horse, model probability, W/L (1 means won 0 means lost)

Al Boum Photo, 0.5, 1
Lost In Translation, 0.3, 0
Native River, 0.2, 0

So our model gave Al Boum Photo a 0.5 chance and he won the race.

The Brier score for these 3 lines of data would be

((0.5 – 1)^2 + (0.3 – 0)^2 + (0.2 – 0)^2) / 3 = 0.1266

Where ^2 simply means ‘squared’

Looking at the above you can hopefully see that if the lower rated horses tend to lose and higher rated horses tend to win we will get a lower Brier score than if races were predicted the other way round. This is why a lower Brier Score means a ‘better’ score.

Next up is the Brier Skill Score (BSS). This measures the Brier Score against some other measure, after all stating that the score above is 0.1266 does not give you an instinctive feeling of how good or bad it is. We just know its better than 0.2 for example.

The BSS is calculated by first working out some sort of measure we can compare to. In this case we will opt for a baseline measure of simply predicting all horses with a value of 0.33. Why 0.33, well because that is the percentage of 1’s in the sample set. Obviously across many races this will come out at more like 0.1 or thereabouts. With the 0.33 for every horse we can now calculate a Brier Score based on probabilities of 0.33 for every horse. What we are doing is using an average likelyhood for the prediction probability of each horse. Substituting this in we get

((0.33 – 1)^2 + (0.33 – 0)^2 + (0.33 – 0)^2) / 3 = 0.2222

Now to calculate the BSS we divide the models Brier score by the Naive predictions Brier score and then subtract this from 1

1 – (0.1266 / 0.2222) = 0.4302

Negative values mean the model has less predictive value than naive baseline probabilities. Positive values (max = 1) mean the model is beating naive baseline predictions. Our one sample 3 horse race is clearly kicking butt but over many races that score would certainly come down but if your model is any good, hopefully stay above zero. More importantly if you modify a model and your BSS score go’s up then you can be hopeful that the changes are worth sticking with.

King George 2019

This years KG seems to pivot around whether Cyrname will be as effective at 24f as he is at 20f and of course by effective we mean more than just will he be as good performance wise because as punters we are not in the game of being right about how good a horse is we are in the game of being right about how wrong other punters are. Cryname could be equally effective at 24f as 20f but still be a good lay or a good bet at the price. This is the conundrum that most punters starting out fail to appreciate and usually never aquire an understanding of.

In analysing Cryname as a focus horse for this race we could examine how horses with his type of essential credentials have faired betting wise in the past. AE values provide a vehicle for doing this, they give a picture of not simply how many horse have won when moving from 20f to 24f after winning at 20f but a picture of how much you would have won or lost if you had backed them.

Looking first at horse’s moving up from 20f to a 24f Chase race of any type we have an AE value of 1.01. this means to bookmaker SP prices after deducting over-round you would have won 1p in every pound bet. Obviously you cannot bet without over-round so that actual profit would disappear but we are using these AEs as a measure of under or over bet by the public.

Now if we look at horse’s that have simply won last time running over 24f regardless of the distance of their last race we get an AE of 0.98. Together these two AE values do not look like a case for opposing Cryname. However when we venture into checking those horse that ran over 24f after previously winning at 20f we get a damaging AE of 0.86. This is a pretty b=negative AE value and suggests that the combination of trying a new distance from their last race plus being bet by the public for winning last time out sends these horses off at too short a price.

Finally I checked whether non handicaps were any better. Perhaps handicap wins force a trainer to try a new distance in response to an anchoring weight increase. The AE for non handicaps came in at 0.88.

This prompts me to lay Cryname particularly at the 2.6 currently available.

Merry Christmas to all readers of my blog, responses welcome in the comments section

Using KNN to Evaluate Final Sectionals

There has been some talk recently of the eye catching debut of Waldkonig at Wolverhampton from the Gosden stable. The trouble with eye catching winners is that out eyes have a habit of deceiving us and our pockets pick up the tab. This can particularly be the case when a horse wins off an uneven gallop. A horse quickening off a slowish pace will always look more impressive than a horse grinding out a distance between first and second off a fast run race.

This is where sectional times can be of assistance at least in giving an objective method of analysing performances where varying pace scenarios have unfolded. To do this I decided to try the humble K Nearest Neighbor Machine Learning algorithm. It is an algorithm thats pretty easy to get your head around so lets take a look.

With regard to sectional times I divided the 8f Wolves races into three sections. First call, second call and final call. These are essentially how fast in seconds was the first third of the race how fast was the middle third and how fast was the final third. Now what I want to do is compare Waldkonig’s first and second call with the corresponding calls for all 2018 and 2019 handicap races where a horse was running in a class 5 handicap and finished in the first three. The question however is what do we mean by compare. This is where the kth nearest neighbor algorithm kicks in. I used the Euclidean distance between the two sets of first and second calls. OK at this stage I am probably losing you so lets make a concrete example. Imagine we are comparing Waldkonigs calls of 35.11 and 40.95 with the calls from another horse in another race, let us call him supercaller and his call times are 36.0 and 38.0. To compare the two we apply the following formula

(35.11 – 36.0)^2 + (40.95 – 38.0)^2

^2 simply means raise to power two. We then find the square root of this and we have a measure of the similarity or closeness of one vector to the other. If we find the smallest five of these calculations we have effectively found the five results from past results that ran closest to Waldkonigs first and second calls. Now if we take the average of those five horses’s final calls we can use this as a prediction of what Walkonig should have run for his final call.

It turns out that Waldkonig should have run (if he is a typical class five placed handicapper) 36.42 whereas in fact he ran 34.65. He beat the KNN average final call by 1.77 secs.

OK so does that make him a Derby winner, well I will cover that in a following blog if there is interest in this piece but in the meantime let me give you one horse that received far less attention and ironically from the same stable. Millicent Fawcett ran far faster earlier calls than Waldkonig and her predicted final call comes out at 37.778 and yet she ran 35.79 beating her predicted final call by 1.988. Interestingly as well is that she is by the same sire as Waldkonig

A Brave New Betting World

Tags

Machine Learning Betting

So you are off to the races on Saturday. It’s 2025 and your smart friend has invited you. You are travelling in his fully auto piloted hydrogen car, the two of you having a beer in the back seat while the car finds its way to Ascot. Your friend is excited because he has the 2/1 fav ante post at 5/1 to win 5,000 WeChatDig the current goto digital currency backed up by Blockchain technology. He is pissed off though because as a pro gambler Blockchain’s logging of every transaction in a Government accessible log means the threat of taxing the winnings of serious gamblers is now almost ready to go live. This is infuriating as he has not yet fully paid off the final payment on his chipped brain implant that enables him personal neural access via 5G to any information and data that desktop Google used to provide, an invaluable facility for plugging in late race information, via his own brain, into his own personal deep learning Neural Network which runs on his Quantum Computer phone. But you are not worried about all this, you are just pissed off that you are half way to Ascot and you forgot to pick up a copy of the Racing Post from the newsagent.

Which of these two do you feel most aligned with, because if it’s the Racing Post reader then you are going to pretty soon feel like you wandered into one of Chevley Park’s stud farm boxes and got mistaken for a mare. The betting landscape is changing although you would not think so listening to the popular media still driven by a fool in the ring and droves of writers/tipsters ‘burning the midnight oil’. I am struggling to think of one media contributor has input which reflects the current direction of successful pro betting.

Is it all gloom?, if you are not part of the Artificial Intelligence approach to sports betting are you just one of the monkeys trying to type a Shakespeare play ?. Probably, if we are talking meaningful amounts of money here, after all we have to remember that the small bet turnover, large bet size avenue has been closed down by bookmakers.

So how best to re skill and create the best possible chance of punting success (please don’t suggest getting your Racing Post delivered). The first thing I would do is learn to code. For some people this is unattainable but for others a few months on line could get you up and running especially if the course was geared towards the application to sports betting. The second approach is to gain access to AI through a more intuitive, easy to use interface. On both of these points I have been considering offering some input. The former would be in the shape of a one or two day course in Python programming covering topics and skills that I use on a daily basis. On the latter I have made some progress on a prototype. You can access a short video demo’ at https://www.youtube.com/watch?v=EEqxwJqjx7g

You may think what I am describing is alarmist, a fantasy even if you cherry pick the most likely points of the message. Well do not expect to read something similar in the popular media. Their job is to protect their sponsors by keeping you ill informed and locked into a way of betting that can only lead to one result. You can choose to know a lot about Racing or a lot about data, I know which one I would choose.

Using GroupKFold to access races in Python

Tags

GroupKFold Python Horse racing

Apologies to those of you not interested in Machine Learning and Python but I wanted to get this out there so that if my code is incorrect (it appears to work) or perhaps there is a more efficient approach, then some one will kindly put me right.

When you build a ML model you can use a technique called K fold cross validation. It simply means splitting you data into, as an example, 5 partitions, and then training your model on partition 1 to 4 and testing it on partition 5. This happens five times with each times having a different test partition eg- train on partition 1,2,3 and 5 but test on 4.

Now the problem is that with race data you really want to split on whole race boundaries. You do not want, for example, half the field from a given race ending up in the train data and half in the test data.

To handle this we can use GroupKFold

The idea behind GroupKFold is that the data is first grouped on an identifier, in our case the RaceId and then the folds or partitions as I called them are created on the groups. Below is the small sample data, followed by the code.

RaceId,Track,Horse,Dlto,Penulto,Age,Rating,Bfsp,FinPos
1,Nemarket,Mill Reef,13,56,4,85.5,3.5,1
1,Newmarket,Kingston,34,23,4,76.2,7.5,0
1,Newmarket,Nijinsky,27,,4,95,10.2,0
2,Sandown,Red Rum,98,23,5,90,5.4,0
2,Sandown,Henbit,101,54,4,85,20.4,1
2,Sandown,Troy,22,32,4,98,1.9,0
2,Sandown,Wollow,36,23,4,87,2.2,0
2,Sandown,The Minstrel,44,67,4,88,5.8,0
2,Sandown,Try My Best,34,53,4,82,3.2,0
2,Sandown,Tromos,62,73,4,65,6.2,0
3,Bath,Sea Pgeon,47,35,4,81,20.0,1
3,Bath,Monksfield,59,5,4,78,11.4,0
3,Bath,Night Nurse,12,15,6,62,4.2,0
3,Bath,Birds Nest,14,78,5,53,3.2,0
4,York,Frankel,25,17,4,100,1.9,1
4,York,Brigadier Gerard,23,67,3,90,2.9,0
4,York,Dubai Millenium,89,23,4,85,4.8,0
5,York,Posse,23,56,4,82,5.6,0
5,York,El Gran Senor,32,21,4,100,2.3,1
5,York,Radetsky,67,21,7,70,12.4,0

The code

import numpy as np
import pandas as pd

from sklearn.ensemble import RandomForestClassifier

m = RandomForestClassifier()

from sklearn.model_selection import GroupKFold

## read in the data ##
df = pd.read_csv(‘testcsv.csv’)

## take a look at the data ##
print (“The input data”)
print (df)

## keep it simple remove rows with missing data ##
df = df.dropna(axis=0)

print (“df”)
print (df)

## create group indices ##
groups = df[‘RaceId’]

## take a look at what this produces
print (“The groups indices”)
print (groups)

## pull out the fields required for the model ##
X = df[[‘Dlto’, ‘Penulto’]]
y = df[[‘FinPos’]]

print (“y”)
print (y)

## create instance of groupKFold set n_splits as required ##
gkf = GroupKFold(n_splits=2)

## loop through n_splits times createing index’s into the arrays for the allocated rows ##
for train_index, test_index in gkf.split(X, y, groups=groups):

print(“TRAIN:”, train_index, “TEST:”, test_index)

X_train, X_test = X.iloc[train_index], X.iloc[test_index]
y_train, y_test = y.iloc[train_index], y.iloc[test_index]
print (“— X train test —-“)
print(X_train, X_test)
print (“— Y train test—“)
print (y_train, y_test)

## train a model on X_train using y_train as target features ##
y_train = np.ravel(y_train)
model = m.fit(X_train,y_train)

## test model on X_test ##
preds = model.predict_proba(X_test)
print (‘Predictions’)
print (preds)

print (“————–Split done———–“)

NOTE – When viewing the output you are probably expecting the dropna command to physically remove row number 2 as it contains a NaN, it certainly does not appear when the df is printed out. But if you examine the indices and the group links into the df you will see that it still links in to row 3 for Red Rum. In other words it does not shuffle down the df entries when executing a dropna. I was fooled by this for a while. Thanks to @amuellerml for nudging me in this direction

A Brief Look Back

My first exposure to horse betting could have been a disaster or a blessing depend on where you end up 40 years down the line. I met a friend at College back in 1975 and as an ex pro footballer he had already caught the bug via long afternoons after training with nothing to do. My first bet was 3 x dollar doubles and a dollar treble. I think it was shillings actually but we called then dollar doubles for some unknown reason., Yes you guessed it, they all won. I went into that 1976/77 winter hoping The Minstrel would win everything and his stable companion Cloonlara would do likewise in the filly classics. Cloonlara never trained on but the Minstrel won the Derby.

I remember being fascinated by the wonderful names of those 2yos, Tachypous still sticks in my mind and also the wonderful secret code of race reading appealed to my psyche. What did nearest finish mean or hld up.

I am a great believer in the idea that your the people you hang out with will greatly influence your betting and although me and my mate had some great success along with failures I had begun my betting career with a strong bias against handicaps which I now realise was a mistake. My belief that there was no point arguing with the handicapper who’s full time job was to make all the horse’s theoretically dead heat was incorrect but it took me a few years to realise this.

Years passed with mainly losing years although my pre disposition to be non addictive and the thought I put into bets even from an unsound perspective generally meant I lost small. The media had already polluted my thinking, straight jacketing my outlook into finding winners instead of finding value. I remember a line from Van De Wheil who some of you may remember writing in the Raceform Handicap book readers letters page. “There is a winner in every race but is there a bet” or words to that effect. I think VDW should have spelt that out a little more clearly instead of expecting everyone to understand what he meant.

I have always had an interest in crowd betting or the wisdom of crowds perhaps due to my first really great betting year. My buddy from college and I teamed up in 1984 and for a flat season bet together with a no bet unless we agree rule. It probably helped that he was in Brecon Wales and a new wet behind the ears bookmaker had opened who thought it was a safe idea to bet in the morning to Sporting Life forecast SP’s…..DOH!. By the end of the season he offered to pay us by giving us his car which we refused in favour of installments. Late in the season we met up in Rotherham where by then I was lecturing at a college in Computer Science and split up the money over the table in an Italian restaurant. I still think this approach has potential for a small group of punters perhaps with bet suggestions put forward and then annonymous voting on bet or no bet.

By the late 80’s I was lecturing in Computer Science at De Montfort University, in fact before that in 1987 I had spent many months travelling to London to spend hours in the Colingdale newspaper library researching past copies of the Sporting Life for a wonderfully naive system I had put together. I recall at the time thinking that if this data could only be in machine readable form my computing skills would allow me to unpick it. I did not realise that a couple of guys would implement this in the mid 90’s and revolutionize the way we look at betting. The system had a poor year when I took it into the field and if it was not for Mystiko winning the Guineas that year (a non system bet) I would have been out of the housing market for a good while.

I was modest on the betting front during the 90’s as I was still lecturing at the University and I had opened a restaurant with my partner but in the late 90’s I discovered the first of probably two life changing decisions. The first was RSB a computer package that allowed one to interrogate dozens of horse racing variables and build systems based around them. For me and others betting had been akin to landing in a foreign town and then trying to find the railway station whilst having your nose constantly 2 inches from the ground. RSB was like being handed Google Maps. For sure there was a lot of backfitting suicide conducted by many of its users. Never have so many jockeys on certain course’s on soft ground on a Tuesday afternoon only in races with less than 10 runners been backed by hopeful naive punters but when taken with education from other sources RSB allowed a birds eye view of what really had value in Race betting. That education I mentioned came from my second smart move.

Reading a Russell Clarke article one day I discovered a place for serious minded bettors to exchange ideas and get help. It was a monthly magazine called SmartSig and an accompanying email forum for discussion. There were plenty of smart punters on there and it was through SmartSig that I began to learn rudimentary language and rules that pro bettors work to and not the misplaced find a winner mentality of the mass media. I contributed a few articles to SmartSig and attended some of the get together’s such as the lecture we all attended on Bayes theorem given by Professor Speigelhalter in Oxford.

During the noughties I gradually moved away from Lecturing and took up full time betting. This had its own complications to overcome. I also had taken over running SmartSig after the retirement of the owner Stef and an aborted relaunch by a purchaser. Renamed as SmarterSig it is still running in a diluted form to this day. During this period I also organised and ran two sports betting conferences which as far as I know were the first of its kind in the UK, certainly outside academic circles.

It is not clear when you start betting with bookmakers that if you win they will refuse your bets but I quickly discovered this to be the case. To navigate around the problem I moved from online to betting shops in the High street, taking up camp in a Costa coffe shop from 9am to 12 noon, popping out occasionaly to take a price on a horse. I was having around 3,500 bets per year and averaging around 14% on turnover profit. I soon discovered however that shops ban you quicker than online and I was soon struggling to get bets on. At this point I hired a team od students from the University, paying them £5 per hour to do the running. I calculated that I was still well ahead despite the employee overhead. Two problems arose with this tactic. The students got banned and Costa became suspicious that guys rapidly leaving and returning with wads of cash while I sat orchestrating over cups of coffee looked suspiciously like a drug operation and Costa Coffee came in line with Ladbrokes et al and banned me. To this day if you go in a UK Costa Coffee shop and try to access a betting site through there wifi you will be declined. Like to think I have contributed something to the comfort of Costa Coffee drinkers

Things had to change and I decided after a chat with a guy I knew from SmartSig who ran a tipping service that the best course would be to run a service myself and let the client manage the headache of getting on. I proofed all bets to a reputable proofing service and for 3 out of 4 years actually won their accolade of top overall tipster which included all sports not just horse racing. Part of what helped me win those awards was my approach was somewhat different to other tipsters who had no doubt been brought up on the false notion that selectivity, a few bet per month was the sign of a sound tipster. I was a volume bettor by their standards and the evaluation formula used on the proofing service favoured volume. This idea of volume was to be pivotal in my next step.

After a health scare in 2013 I decided to shelve the tipping line. Maybe I had underestimated how stressful having clients can be although at the time it did not seem to bother me. My replacement however was already in the pipeline, Starbucks, no only joking. I decided that Betfair betting was the only way to go but liquidity in the morning where I had been earning my 14 or so percent was thin so I needed to replace 3,500 bets and 14% with many many more bets and perhaps 2% ROI. This would mean bet selection and placement automation. You cannot physically place the number of bets/lays I was anticipating with a manual approach or at least not without a serious Vitamin D deficiency. I did my own coding in a mixture of Perl and Python and for the last 5 years or so have been betting solely on Betfair to my own horse ratings.

If you have read any of my other blogs entries you will know that even losing punters should be migrating to Betfair where they will lose less. What does the future hold/, well I have a keen interest in Machine Learning and its application to betting but I guess ultimately everything depends on where this sport is heading. At the moment people are not being replaced in the area of horse betting and we are losing market share. I think many of the current promotion horses being ridden by people are well intentioned and certainly welcomed by me eg sectional times. The real issue however is overall profile of horse betting. Every element of Racing is covered with grace and respect but you will have noticed that when it comes to betting a fool is rolled out to talk about it. Usually the fool in the ring is the biggest culprit, the fools on bookmaker adverts are to be expected. We need betting to be viewed in a more favourable light, one which mirrors stock marketing investing. This first of all requires allowing people to bet and certainly not protecting and covering up those that don’t allow punters to pace bets within reasonable risk managed quantities. Ultimately though we need to push the exchanges as the only place in town for all punters who have ambitions of winning or at the very least losing more slowly.

Machine Learning What Should We Predict

Tags

Machine Learning and horse racing, Machine Learning Horse racing

I have been playing around with an ML model today and the purpose of this post is to hopefully promote some discussion about potential target fields.

When you feed data to an ML algorithm you need to define input features eg is the horse a course winner along with a feature that the inputs have to predict. It is with the latter that I was running a simple experiment. I ran a model on four different target features to get a feel if one stood out from the others. The four varieties were as follows.

1. Good of old fashioned 1 if the horse won and 0 if the horse lost
2. 1 if the horse won or came second 0 otherwise
3. 1 if the horse won or came second or finished 3rd in a race with more than 8 runners
4. 1 if the horse finished in the first 4 and out ran its odds 0 otherwise

In the last case outran its odds simply meant that the horse was positionally longer in the odds than its finishing position. For example a horse finishing 2nd but went off fav would be a 0 whereas a horse finishing 4th and being 5th in the betting gets a 1

I tested for both how the top rated performed and how simply backing horses above a threshold performed. This is a quick and dirty measure but the objective is to foster some discussion hopefully on other measures for target variables.

Option 1 produced

Toprated 7998 bets 1323 wins PL after comm’ +514pts ROI +6.42% Varpl +40.6

Option 2 produced

8012 bets 1328 wins PL +265.3 ROI 3.3% Varpl +43.9

Option 3 produced

8028 Bets 1365 wins PL +413 ROI +5.15% Varpl +91.5

Option 4 produced

8056 bets 1201 wins PL +235.9 ROI +2.92% Varpl + 67.79

When it came to simply backing any horse above a certain threshold on the ratings option 3 performed best followed by option 2 and then option 1 and finally option 4

The reason for trying the various options is that unbalanced data can effect the performance of ML algorithms although the Gradient Boosting Tree based algorithm I am using suffers least. An unbalanced data set simply means fewer 1’s than 0’s. The closer you get to 50-50 on the target 1’s and 0’s the more balanced the data is. Clearly adding placed runs increases the balance.

The question however is are there other options worth throwing at the algorithm. I would be happy to receive any suggestions on other possible target fields in the comments section.

Pinatubo Gift Horse or Bet Trap

You could not escape Pinatubo’s stunning win in Ireland recently when he demolished the opposition. His leg speed made Too Dran Hot look like a candidate for next years Cheltenham Gold Cup. Nor could you perhaps ignore the avalanche of bookmaker offers for next year. I think he is 5/4 for the Guineas as I write. A recent article was posted on Twitter citing the dangers and gloom around these 2yo hotpots. For sure the Tromos, Arazi and Celtic Swings of the past are fresh still in our minds but is this because we harbour gloom more than we do success. When was the last time anyone cited the great 2yo’s that went on to succeed as 3yos. Also what do we mean by succeed, well as a punter we want to know if they are a trap or a decent betting proposition.

I had a quick look back at 2yos that had logged an official rating greater than 115. Seemed like a starting point given that Too Darn Hot logged 116. Although my data go’s back to 1999 I suspect the ratings may have changed in scale along the way as I have no qualifiers prior to 2008. The total number of bets on these horses as 3yos amounted to 50 with PL of +29.6 to Bookmaker SP. Pretty impressive especially as the Betfair SP would be a good deal better. The AE value by the way was 1.46

Here are this years bets (English races only)

11/05/2019 Anthony Van Dyck (IRE) 1 2
16/05/2019 Too Darn Hot 2 1
01/06/2019 Anthony Van Dyck (IRE) 1 6.5
18/06/2019 Too Darn Hot 3 2
21/06/2019 Pretty Pollyanna 7 10
13/07/2019 Pretty Pollyanna 4 8
27/07/2019 Anthony Van Dyck (IRE) 10 7
31/07/2019 Too Darn Hot 1 1
24/08/2019 Pretty Pollyanna 2 0.9

What if we extend the range a little to say greater than 110.

314 bets PL +0.41 AE = 1.34

Again not too bad and clearly shorter priced horses did well.

Finally 2yo’s that were rated above 100 produced a further reduction in overall performance.

2364 bets PL -691.27 AE = 1.04

It would appear from the above that great 2yo records are not something to be feared and somehow we have to get over Tromos even if you did back him for the Guineas

Betfair It’s a No Brainer

Some time ago I wrote an article which described why all punters would be better off betting on Betfair than wasting their time betting with bookmakers. You can re read the article here https://markatsmartersig.wordpress.com/2018/10/05/educating-trevor-2/

What surprised me about this article is that no one challenged me with the argument that my data was not based on intelligent selections but merely looking at all horses. To rectify this and hopefully convince a few more people that Betfair is the only game in town I decided to take a look at how our newspaper tipsters had faired with their naps during 2019 based on best early morning price at 10am versus Betfair SP. The bookmakers used for checking best 10am price were as follows

Bet365,Sky Bet,Ladbrokes,William Hill,Marathon Bet,Betfair Sportsbook,Paddy Power,Unibet,Coral,Betfred,Boylesports,Totesport,Black Type,Betstars,Betway,BetBright,10Bet,Sportingbet,188Bet,888sport,Bet Victor,Sportpesa,Spreadex

As you can see I have been very generous in the sense that no punter is going to have accounts open whilst taking top prices amongst all those bookmakers, nevertheless lets see how our paper pundits faired against BFSP. As usual for simplicity I have considered only races where there were no NR’s between 10am and the off.

Backing all selections to best AM price produced the following results across all tipster naps.

5444 Bets PL -299.67 pts ROI -5.5%

Now backing the same selections to BFSP we have

5444 bets PL -277.35 pts ROI -5.09$ after commission deducted.

Even with the world of bookmakers at their disposal our newspaper tipsters could not beat BFSP and in the long run neither will you because even if you are sharper than our tipsters your accounts will be closed down quicker than a 5f sprint on fast ground at Epsom.

If there is interest I will later drill down into individual tipsters to see if there are any patterns. One area of interest may be the profile of the tipster, perhaps lower profile tipsters working for the Carlisle Evening Gazzette do better than Racing Post ratings. For example Paul Kealy has posted a profit 36.45 to BFSP after commission