Flat Races - Defining features

+1 vote
Anyone have any tips for modelling the flat races?  I did great on the jump races with a strike rate of over 40% at prices of 3/1 and above but that model doesn't work very well on the flat races.  I get a place finish is over 80% of my races so I feel like there must be some feature that I am missing that is critical to flat races.
asked May 8, 2018 by DataScientist Handicapper (530 points)

2 Answers

0 votes
Can you say what features you have already included - then an omission might be more obvious?

Other than that, it may be worth restricting any model to 2 year old races only, 3 year old races only, or all age handicaps - since the rate of improvement in flat horses, being the equine equivalent of young adolescents, bears little resemblence to jumps horses, bar the older age groups.
answered May 8, 2018 by colin Guru (12,370 points)
All the features I use are derived from the smartform database, about 70 in total.

I’ll give the age restrictions a try, I’ve tried to generalise it so it’s not restricted to any type of race but it’s just not quite hitting the mark.

Some features I don’t currently include are finishing times and distance beaten.
I was going to attempt to build a rating system based on the above but I feel like it’s going to be a lot of work and I don’t know how much value it will add to my model.
Am assuming you have trainer and jockey strike combined, trainer strike at course, distance etc?
The other thing that seems to be important on the Flat - perhaps more so than jumps? - is Class.  Unfortunately this is not available on most Irish races (unless group races) but you can manufacture a proxy for that based on race type and prize money.
Last but not least is the sire strike rate - perhaps not so important over jumps but hugely influential on the flat in terms of ground preferences, precocity (ie. early maturing / late maturing offspring) and distance preferences.
I have horse/jockey and jockey/trainer and trainer/course but I don't have trainer/distance...I can add it and see how it goes.
For class I have things like moved up/down class but I don't actually have a class variable since it's the same for all horses in the race.
I'm finding the Irish races very hit and miss, I am still +'ve after 68 bets but I feel like there is a lot more work to do there.
There is so much info in the database that I haven't even considered yet too!!
On class, it's more a case of "class strike rate".  Eg. if 2 trainers may have the same overall strike rate but totally different strike rates in group 1 races.
Ah ok I get you, well I can add them too and see what the impacts would have been on todays races.
There are so many combinations of horse/jockey/trainer/course/distance etc. and I only use a handful of them. The algo I us will just reduce the coefficients to zero if they have no impact so it won't hurt to try them
0 votes
I have only just read your request for assistance with regard to your disparity between flat selections and national hunt selections.

I have been writing and developing systems for nearly thirty years and have developed a system that used between ninety and a hundred factors factors and its success was in selecting high priced winners if I chose the second rated horse and ignored the top rated.  One by-product of this system was the emerging fact that horses that dropped in class were not certain winners.  In fact what emerged from the data was the fact that if a horse was upgraded by a move of two classes upwards it was more likely to win than in any other change of class.

Many other facts emerged from using this system which dispelled many apocryphal sayings used in horse racing..

The basis of the data used was from the site of Adrian Massey who produced data of exceptional value.  He discontinued producing his statistics and that meant I had to find data from elsewhere.

There are several sites where data can be obtained and I have tried them all.  The biggest bugbear with this approach is that the site will change the layout of the data at any time and translation of the data then becomes a problem.  As I am more interested in systems and programming than betting I find it an interesting exercise.

As mentioned earlier, the use of factors to determine the best selections in a race is of importance.  How many?

Currently I use five.  Form, Jockey,Trainer, Days Since Last Run and Weight.  This system has been in use for several years and has a strike rate of 30% to 70% for all types of racing.

Today the strike rate was 52%  for all flat meetings.  

It isn't necessary to split the atom to achieve success.  It's what you do with what you have that is the path to success.
answered May 19, 2018 by Awbower Novice (200 points)
I use a slightly simpler version of a model proposed by Noah Silverman which uses conditional logistic regression, the model basically reduces the coefficients to zero if they are not useful.

I actually found limiting myself to 2/3 year old horses had a negative impact on its predictive accuracy.

I’ve also tried limiting races such that all runners have had at least 3 races so my weighted averages have a value also made no noticeable impact on the model.

My strike rate on sellers, claimers, novice, maidens etc. are all similar.

The difficulty is reducing the 30 or so predictions a day down to a manageable number. I feel like it’s missing a key ingredient because the wins/losses are faulty evenly distributed across all race/track types.
I repeat what I commented in the last paragraph of my reply to you in a different way.  "There is more than one way of skinning a cat!"

I don't say this to be offensive but to point out to you that your use of a convoluted process is not as successful
as my simple method (compared to yours).

The method I use is to download the race cards I wish to process from the internet.  This data is run through a program that extracts the DSLR from the data.

A second program then calculates the ratings for each horse from the five factors I mentioned in my reply and produces a listing from which I extract the top three rated horses.

Yesterday my strike rate was 57%.

Below are my selections for Redcar today.

 Redcar 21 May 2018
 14:05 Watch Racing UK Anywhere Novice Auction Stakes (Class 5) (2YO only)
 5  Hurricane Five       D O'Meara            D Tudhope             101
 7  Zebzardee            R A Fahey            P Hanagan             091
 10 Singe Du Nord        N Tinkler            Lewis Edmunds         090
 14:35 Racing UK HD On Sky 426 Handicap (Div 1) (Class 6) (3YO only)
 2  Laith Alareen        Ivan Furtado         Jane Elliott          139
 4  Furni Factors        Ronald Thompson      Lewis Edmunds         216
 5  Terri Rules          Miss J Feilden       J Quinn               213
 15:05 Racing UK HD On Sky 426 Handicap (Div 2) (Class 6) (3YO only)
 2  Liamba               D O'Meara            D Tudhope             215
 4  Shay C               D Carroll            Ger O'Neill           231
 5  Eeh Bah Gum          T D Easterby         D Allan               148
 15:35 Pat And Les Holmes Celebration Fillies' Novice Stakes (Class 5) (3YO plus)
 2  Hermosita            R Varian             David Egan            129
 3  Amandine             D M Simcock          J P Spencer           120
 7  Pride's Gold         S Crisford           S De Sousa            127
 16:05 Watch Racing UK In Stunning HD Handicap (Class 3) (3YO plus)
 5  Confessional         T D Easterby         Cam Hardie            196
 7  Monks Stand          J Mackie             Barry McHugh          129
 8  Merry Banter         P T Midgley          P Mulrennan           126
 16:35 Racing UK Straight Mile Fillies' Handicap (Class 5) (4YO plus)
 1  Ce La Vie            K Dalgleish          Connor Beasley        154
 2  Kilbaha Lady         N Tinkler            Lewis Edmunds         213
 4  Make On Madam        J L Eyre             D Allan               126
 17:05 Racing UK Profits Returned To Racing Median Auction Maiden Stakes (Class 6) (3YO to 5YO)
 6  Towton               J D Bethell          G Lee                 089
 8  Bollin Joan          T D Easterby         D Allan               094
 10 Savaanah             R Charlton           S De Sousa            131
 17:35 Watch Race Replays At racinguk.com Handicap (Class 6) (3YO only)
 3  Boko Fittleworth     Ben Pauling          P Hanagan             159
 5  If We Can Can        M Johnston           S De Sousa            245
 8  Rock On Bertie       N Tinkler            Lewis Edmunds         214

As I view the above I think you should be able to follow my selections.  The final figure at the end of each horse is my calculated rating.  You will note that the highest rated is not always the winner.  In fact Laith Alareen wasn't in the same ballpark as the top two selections but won at 14/1.

The method I use is to DUTCH all three.  I never bet on odds-on races and if the potential winnings are less than
66% of my stake I don't back.

On the question of odds-on horse I had a friend who only backed them if they were my top selection.  He bet thousands (out of my league) but smiled all the way to the bank.

To save space I haven't listed my Carlisle selections but have had winners in Cosmic Law, Seen The Lyte and Aclimatise so far.
Depends on your measure of success, I am currently building a system that is fully automated, it uses the smartform data as well as some other data that I scrape and builds a model, make the predictions and then places the bets on the exchange without any work from myself.

I've always fancied a dutching model though, what is your ROI?
I have reviewed the type of system you describe with the addition of a loss recovery feature and it was very profitable.  The target for the Dutching was to achieve a profit of £6 for every race on the day. However, in order to achieve a decent profit it was necessary to stake in excess of £1000 on occasions.  I reckon a bank of at least  £2 to £3000 would be required.

As you note because of the low prices of the selections the returns from Dutching will be low.

Initially I started by Dutching only on the prices below 11 in the betting.  Very rarely did I achieve a return in excess of the total stakes.  It was profitable but not at a rate which I deemed viable.

I then reverted to the method I have described in my first post.  I used my own selections.

I Dutched two horses in the 17.45 at Carlisle today as one on my selections was a non-runner.  One was priced at 4.6 (the winner) the other at 14.00.  The profit was 2.3 times the amount staked.  Unachievable when Dutching low prices.

I chose badly today which races to bet on as the RoI was only 23%.  Yesterday it was 78% and the day before 81%.

Use the following as a motivation for your efforts.  If the profit you make in a day is only a £1 and you make it every day, you will have made over £360 in a year.

I admire your efforts and hopes and wish you success.  

You will inevitably have losses and it is the method of avoiding those losses that you need to be aware of.  My RoI today being an example.
I had to answer you last night in the midst of preparing for my holiday.  I am now officially on holiday but have brought the results of the review I did of the automated Dutching and Loss Recovery system.

I mentioned the high staking.  Over the period in which I monitored the system the Overall RoI was 10.69 and the profit generated was £869.79 (Staked- £8132.07).
 
This confirms your belief that Dutching in the way you perceive it will be as above.

I achieve a better RoI because I know that favourites don't win every race (even odds-on ones).  I know that short-priced horses don't bring decent profits when there are more than one in a Dutch.

Using my selections was an obvious move as I knew I would achieve at least 30% success if I backed them all.  No profit though!

Unlike TSB software developers I fully test my programs before committing to their use.  Once productive I monitor the results every day for two reasons:

1 - to ensure the program is working as I designed it
2 - to note any results that have a distinct pattern. (re my earlier comment on change of Class)

I have been developing systems and writing programs with regard to Horse Racing for over twenty-five years.  I initially specialised in favourites performance only.  Down to which price (SP) the favourite would win at and would lose at.

I have a file containing data on favourites only in every race over the last 21 years.  From this data I can generate over a hundred statistics relating to every favourites performance at every meeting in the UK.  In spite of this I couldn't tell you how the favourite would perform in any race with certainty.

I have developed over a hundred systems and written in excess of 500 programs in the search for the Holy Grail of selections.
I view my current method as the nearest I will ever achieve.

For the past twenty odd years I have been in full pursuit of my target, working day and night to achieve my aim.

Now I spend half-an-hour to an hour a day to create my selections and list them.

I will have to look at the automated system mentioned in the blogs to see if I can utilise this and free my self from being glued to the computer when racing is on.

I append my selections for today.

 Nottingham 22 May 2018
 13:50 188bet Mobile Bet10 Get20 Ebf Novice Stakes (Class 5) (2YO only)
 1  Barend Boy           Hugo Palmer          James Doyle           094
 4  Dixieland            R Hannon             Tom Marquand          156
 6  My Dear Friend       R M Beckett          Oisin Murphy          096
 14:20 Champions League Final Betting At 188bet Novice Median Auction Stakes (Class 5) (3YO only)
 5  Ibraz                R Varian             Jim Crowley           122
 6  Infrastructure       M Meade              James Doyle           117
 11 Papelaine            Eve Johnson Houghton Tom Marquand          161
 14:55 Daily Racing Specials At 188bet Classified Stakes (Class 5) (3YO only)
 1  Austin Powers        M Johnston           P McDonald            215
 7  Mashaheer            W J Haggas           Jim Crowley           213
 9  Tebay                M Dods               Callum Rodriguez      218
 15:25 Read Silvestre De Sousa At 188bet Fillies' Handicap (Class 3) (3YO only)
 1  Verandah             J H M Gosden         R Havlin              228
 3  Caiya                Eve Johnson Houghton S De Sousa            113
 6  Rasima               R Varian             David Egan            116
 16:00 188bet Casino British Stallion Studs Ebf Fillies' Handicap (Class 4) (3YO only)
 1  Affina               S Crisford           James Doyle           130
 2  Blanchefleur         R Hannon             Tom Marquand          199
 3  Sea Youmzain         M Johnston           P McDonald            131
 16:35 Download The App At 188bet Handicap (Class 5) (4YO plus)
 1  Airton               J D Bethell          Jim Crowley           115
 5  Northwest Frontier   R A Fahey            T Hamilton            232
 7  Brigadoon            M Appleby            Alistair Rawlinson    132
 17:05 Follow 188bet On Twitter Handicap (Class 5) (3YO only)
 2  Bustam               J J Quinn            S De Sousa            233
 6  Gabrial The Devil    R A Fahey            T Hamilton            225
 11 Belated Breath       H Morrison           Oisin Murphy          235
 Chepstow 22 May 2018
 14:10 188Bet Mobile Bet10 Get20 Fillies' Novice Stakes (Plus 10) (Class 5) (2YO only)
 4  Kg Passion           P D Evans            Fran Berry            201
 5  Lady Wolf            B R Millman          William Carson        199
 7  Popping Corks        R Hannon             P J Dobbs             211
 14:45 Download The App At 188bet Handicap (Class 5) (3YO plus)
 9  Cooperess            D Burchell           Hollie Doyle          189
 10 Peak Storm           J G M O'Shea         Kate Leahy            183
 11 Living Leader        Grace Harris         L Morris              197
 15:15 Bet & Watch At 188bet.co.uk Handicap (Class 6) (3YO only)
 3  Superego             C G Cox              A Kirby               212
 4  Thunder North        D R Lanigan          P Cosgrave            190
 7  Kachumba             Rae Guest            Dane O'Neill          243
 15:45 Pickwick On Course Bookmakers Handicap (Class 4) (3YO plus)
 5  Pastfact             M S Saunders         L P Keniry            154
 6  Bungee Jump          Grace Harris         Cameron Noble         143
 7  Global Tango         C Hills              G Mosse               167
 16:20 Thanks Keith Ottesen Its Been Amazing Fillies' Handicap (Class 5) (4YO plus)
 1  Sunshineandbubbles   J Candlish           Rossa Ryan            206
 2  Iconic Belle         M R Channon          Nicola Currie         120
 3  Dora's Field         W S Kittow           Fran Berry            150
 16:55 ReeceR Group Ltd Novice Stakes (Class 5) (3YO plus)
 5  Samson's Reach       R J Price            Joshua Bryan          099
 8  Almoghared           J H M Gosden         Dane O'Neill          128
 12 Podemos              R M Beckett          R Kingscote           104
 17:25 Coopers Marquees Handicap (Class 5) (4YO plus)
 2  Sonnetist            P D Evans            Fran Berry            213
 5  The Last Debutante   M Johnston           F Norton              146
 7  What A Scorcher      Mrs N S Evans        L Morris              176

In the 14.20 at Nottingham the RoI was 28% (SP's staked: 6/4, 7/4 and 66/1).  In the 17.05 at Nottingham the RoI was 133% (SP's staked 9/4, 10/1 and 4/1)
Wow it sounds like you've been in this game a long time! I started a few years ago after watching a video on YouTube from Noah Silverman who talks of building on a model Bill Benter created for the HK races.

My current system is just a win only method since I don't know how to place ew bets on the exchange but from January 1 2018 it has made 849 points profit using £1 stakes including yesterday (which was a small loss of 2.4 points).  I'm sure this could be improved on if I placed a place bet on some of the longer odds runners.  
I also take SP since I'm not doing this manually and I don't know what to do if my bet isn't matched.

I am going to back test all my daily selections using your method of picking 3 horses to dutch in each race and see what it looks like.

Previously I had been trying to ducth a percentage and I found that it wasn't profitable enough and required a very high strike rate - and like you say big stakes otherwise you're only making buttons.
Your rate of return of 849 points is near enough a profit of 5.9 points a day. A return not to be snuffed at.  The variables in this return are how many bets are made to achieve this figure and the RoI.  Dependent on these figures is how I would use this return figure to improve the financial return.

Your suggestion of using the place market to me is a no-no.  I went through a period when I backed all my selections that were 10.0 and above on Betfair.  The profit from a winner at a price of over 200.0 was soon evaporated by the number of selections unplaced.  To me the exercise was interesting but unrewarding and soon abandoned.  Place backing only on Betfair is the method to use, not each-way betting as it reduces the staking value.

If I had Dutched that winner just imagine the RoI I would have achieved.

The exercise of gambling is to achieve the best financial return one can.  Win only does this.

Many experts advocate multiple systems should be used to achieve better returns.  I don't subscribe to this view as it is spreading your bets on systems that have varying strike rates.  It rings of a defeatist attitude.

The thinking I apply when designing systems is that horses are like individuals and cannot read.  They are unaware that they have a good or bad strike rate, that they are favourite or an outsider. Trainer stats are a guide only, they don't take into account how many times the trainer has raced an unfit horse in order to improve its handicap when it enters it into a higher class race.  (Re my earlier point on horses improving when the Class of race is raised)

All my decisions are based on past achievements from which I calculate a balance of probabilities.  My system dictates and I never interfere with the selections.  Personal prejudices are forbidden.

You are on the right path in thinking about improving your current method.  You have a strong base on which to achieve improvement.  Stick to that only!

Don't bet if the return is too low.  If the return is greater than the stakes you can soon recover any losses without resorting to the drastic method used by the system I reviewed.  Don't bet on every race only the ones that bring the return described above.
I've been looking at the prices from 11am which is when I collect them from the market to the SP on Betfair (suggested by Colin above), it looks like if I traded out on my position I could reduce my loss substantially.  Of course there are races where I still win despite the price drifting out so much but they don't compensate for the amount of losses.

I'm going to test it for a few weeks and see how it works out in real time but based on a cut off of a 20% drift I have lost 300 points profit since the start of the year!
...