Automated Racing Betting
-
Hey guys, I'm Caroline, one of the few female members at BTC. I've been a member for a little over a year but have not been active on the forum at all.
It's taken me a really long time to figure out what works and doesn't work me. I came to this rather accidentally through matched betting which I started in 2018. Although I made some great money with that for a couple of years, as my accounts got restricted one by one I very quickly realised that matched betting was not a long term income source - and that led me to betting and trading.
Over the last few years I have tried all sorts of things - trading and betting on football, horse racing, tennis and even other more niche sports! I started out trying to follow successful traders and tipsters and started to move away from "human" systems and move towards more stats based work.
I also realised very quickly that my utter disinterest in the sport of football in particular meant that I was never going to be a good trader. I cannot bring myself to watch a football match (I do like tennis though!) and I still don't understand the offside rule despite dating a football fanatic for 10 years lol!
Around 2020 I realised that if I was ever going to make a consistent profit from sports I needed to completely remove any human bias and emotion (especially my own) and work with strategies that are purely data driven.
I'm a software developer by trade so I have been able to use my dev skills to help me with things like data gathering and analysis for creating strategies.
There are many things that I learned about myself over the last year or two but what I have noticed recently whilst taking more of an interest in the forum, is that my style seems to be very different to most people on here, and so I thought that doing a bit of a blog may be useful to others? This is a summary of what I do now:
-
- Bet purely on strategies I have developed myself - I will never blindly follow what anyone else does.
-
- I bet not trade. Set and forget if you like. I place my bets and hope they win. No trading out, no stress!
-
- I now only bet on a strategy that I can automate. After spending a year testing several football strategies manually by paper trading them and enduring the insanity of 14 hour days every weekend only to determine that they were not profitable, rather than crying about it, I decide to focus on automation only.
-
- I like very high volume strategies - lots and lots and LOTS of bets. I will explain why in a later post.
-
- I will only put money on a strategy I am extremely confident in. I have strict mathematical formulas for determining that confidence level. Again, I'll explain this in another post.
-
- I take a long term view. I completely ignore the day to day (which can be painful at times!), and the week to week. Sometimes I even have to ignore the month to month.
-
- I manage my bank roll by grouping strategies according to their risk profile (another topic for a post!) and have a separate bank for each group. Each of those banks will have stakes set according to that risk. Higher risk = lower stakes obviously. I don't use percentage staking at all - the way I set stakes is very different to most people here.
Currently I am betting only on horse racing. I have a mixture of back and lay strategies. I have plans for automated football betting and possibly some in-play tennis strategies but they are a long way off.
Despite the years of study and testing this stuff, I am still not profitable. However, I think 2023 will be the year that changes yet. Though January has not had a good start to my strategies!
So that's enough waffle. Any anyone interested in this kind of betting? I think most people at BTC are footy traders (that's why I originally joined) but for me personally, the addition of the racing software has been a game changer and I'm seeing a bit more interest over on the racing threads since it's launch.
If you guys are interested in a blog of this kind then I'll post on a semi-regular basis. I'm only going to do P&L reports on a monthly basis as the short term is so irrelevant with this style of betting. (Variance is a killer for me).
Tbh, there's not much point posting much unless I actually start to see a solid profit, but I'm hoping to see that by the end of this month
-
-
@caroline-middlebrook said in Automated Racing Betting:
@john-folan said in Automated Racing Betting:
I’ve got a max drawdown calculator on my horse resources page(for excel). I normally use that and then times that by 3 to work out my bank.
3 times! You take a very conservative approach then.
That’s for each strategy. Some of the drawdown percentages are quite low to be fair
-
@john-folan said in Automated Racing Betting:
I’ve got a max drawdown calculator on my horse resources page(for excel). I normally use that and then times that by 3 to work out my bank.
3 times! You take a very conservative approach then.
-
I’ve got a max drawdown calculator on my horse resources page(for excel). I normally use that and then times that by 3 to work out my bank.
-
My New Approach to Staking & Bank Management
In January I had a BIG loss, but having looked more closely at the results, wasn't actually out of the bounds of expectation for the strategies involved.
However, it did highlight a problem with my bank management.
So first of all I obviously build each strategy individually using BTC's wonderful new software; analysing and testing parameters to death until I have come up with something that looks good, passes the math test (which I'll explain in another post!), and doesn't raise any red flags.
For each strategy I calculate the maximum drawdown that had been reached in the backtesting, and also cross check this against an online tool that estimates it just to make sure nothing is off. If I run that strategy in isolation, I would create a bank with the number of points equal to 1.5x the MDD.
So for example, if it had reached a max drawdown of 58 points, I'd round that up to 60, multiply by 1.5 and set that bank at 90 points. Each bet is placed at 1 point, so I have 90 bets available for that bank.
I do this for each strategy BUT... what I had been doing, was combining strategies that had a similar risk profile together to create a bigger strategy and used a single combined bank for them. This was a mistake!!!
At first glance it seems to make sense. Martin actually talks about this quite a lot in his videos, talking about having a portfolio of different strategies and how very often a loss in one of them will be counteracted by a win in another. That's the theory.
But in practice, due to variance, (and sod's law) there's nothing to stop EVERY strategy from going into a downswing at the same time. That's what happened to me in January (all except one, which broke even). Individually, the DD of each strategy was within the parameters I would expect. But combined together, the combined drawdown was a lot less than the individual drawdowns added together so I thought I could get away with leveraging my bank.
I was wrong.
Thankfully, I was on small stakes and it wasn't a disaster. It's taught me a expensive lesson though. So for February, I have once again separated each individual strategy. Each has it's own bank. I've also cut out the ones which had the highest variance.
Stake Setting
In terms of Return on Capital all strategies perform around the same - with about an 8x or 9x return over a year. As such, I want my banks for each strategy to be the same in terms of £££.
I've rounded up the banks a little so that they are all a multiple of the lowest risk one. I have the following:
1 at 12 points (A)
2 at 24 points (B,C)
1 at 30 points (D)
1 at 60 points (E)Then I adjust stakes accordingly. So... if I stake £1 per point for strategy E which needs the full 60 points, then for strategy D which only needs 30 points, I'll stake £2. Strats B & C need 24 points, so they get 2.5x the stake at £2.50. Lastly the lowest risk strategy (A) only needs a 12 point bank and I can stake £5 on that one.
Previously I was staking the same £ amount on each one and this also cost me as when the higher risk ones inevitably had substantial losses this hurt my bottom line much more than when the lower risk ones were in a drawdown.
I hope that makes sense!
-
@ian-parry yes I've been tracking them daily and all but one had a big downturn in January. It was very strange really!
-
@caroline-middlebrook jesus! what are the strategies and how long did you forward test them before going live?
-
@caroline-middlebrook This are very interesting results and to some degree mirror similar results I've seen with strategies I found using the horse software. When you use the software to check results for the strategy you are using does it show a similar downturn in January?
-
370 Points Lost!!!!
So January so far has been frankly catastrophic so far and I've simply had to re-evaluate everything before tomorrow as it's a busy Saturday; the busiest day we've seen in a while after the constant meeting cancellations in this cold weather.
So first of all, this is a chart showing the performance of all the strategies combined together. It looked very healthy and was based on 181k bets, so a very large sample size. I went live on January 1st and it just went downhill from there!
I've highlighted the drawdown that's occurred in January and that was taken a couple of days ago and yesterday and today were also heavily losing days so you can visualise a further dip there.This particular chart plots 1 point per day from the start of the data which goes back to the beginning of March 2021.
Another reason I was very confident going live was that when I broke the results down into months, not only had there never been a losing month, but the lowest month still made over 80 points. The BTC back data at the beginning of this year had 22 months of data.
Now there is one other thing to mention, and that is that all of these strategies are automated using BFBotManager and the results did not 100% match the bets made by BTC. This is something I will continue to track over time to get an idea of how my bots perform in the real world. But for the most part, the same horses were bet on, with some small discrepancies in the odds matched.
So what now?
I can't continue as is after such a huge loss. My bank can't sustain any more of it!
So, another thing to mention is that two of the strategies are extremely high volume and make up the bulk of all the bets. They are both quite low ROI but show extremely good profit potential over the long term simply due to their volume. However, another issue I ran into here was that both of these strategies start placing bets some hours before the race starts which means that the liability builds up to high levels before the first race kicks off. And due to the sheer number of bets, I had to keep the stakes low to be able to handle the liabilities. That was a good thing for this month of course, but is perhaps not the most efficient use of my bankroll.
So for now, I have decided to drop the two big strategies and also put on pause all of the back strategies. I had 4 back and 3 lays.
What I did next was to re-analyse every lay strategy with January's disastrous month included. There actually wasn't a whole lot of difference over the long term actually so it certainly looks like variance wasn't on my side. However, the fact that there was such a huge loss in just 27 days when previously, there had never been less than 80 points profit in a month means that I have to change something.
I had a few other lay strategies that were not activated. What I do is first look at each strat in isolation and then look at how they perform when combined together. Taking out the very high volume one changed the profile of the combinations quite a bit.
So, starting tomorrow I have 5 small lay strategies active and that's it. The total volume of bets analysed here is 17k. Here is the chart of these 5 lay strategies plotted over every day as before.
It looks less smooth than the previous one but being only lay bets it's pretty low risk. The highest drawdown over this period is just 26 points.
So I will set aside the initial loss of the 370 points and will start tracking separately from tomorrow as the strategies have significantly changed. I do hope it goes better than the first try!
-
I'll explain how I calculate drawdowns in another post. It's something that has evolved a lot for me over the last couple of years.
-
I came across a blog post from smart betting club which looks into probabilities of losing runs based on strike rate. It might be of interest?
https://smartbettingclub.com/blog/betting-strike-rate/ -
@caroline-middlebrook said in Automated Racing Betting:
As Keith Driscoll says. Laying is more stable than backing and is an excellent place to start. Get a solid footing down and then look for backing later for the real fun.
Yes I figured that out recently. I took my profitable but volatile back strategies and basically inverted them to create opposite lay strategies. These now form the backbone of my bets. I do still have the back bets but at much smaller stakes. Combined together, they seem to compliment each other nicely.
I’m a big fan of a variable portfolio. Something about eggs in one basket. The trick is the separate bank for each.
-
As Keith Driscoll says. Laying is more stable than backing and is an excellent place to start. Get a solid footing down and then look for backing later for the real fun.
Yes I figured that out recently. I took my profitable but volatile back strategies and basically inverted them to create opposite lay strategies. These now form the backbone of my bets. I do still have the back bets but at much smaller stakes. Combined together, they seem to compliment each other nicely.
-
@ryan said in Automated Racing Betting:
@caroline-middlebrook One of the people I point people to on the horses who isn't just making play money and is doing very well is @John-folan.
He understands the software very well and has built some brilliant things on it.
The software is excellent. You just need to take your time with it and learn how to use it properly. What affects what and so on.
As Keith Driscoll says. Laying is more stable than backing and is an excellent place to start. Get a solid footing down and then look for backing later for the real fun.
-
@caroline-middlebrook One of the people I point people to on the horses who isn't just making play money and is doing very well is @John-folan.
He understands the software very well and has built some brilliant things on it.
-
@caroline said in Automated Racing Betting:
@caroline-middlebrook
I have wondered, If you had a service that gave you a percentage of each horse/teams chance of winning then blindly backed/layed every one at say 10% above/below that chance whether you would make money.
Obviously not all bets would get taken, and you would have to calculate whether the percentages given were a fair reflection on average over the long term
Getting the percentages is not so hard but calculating whether they are accurate over the long term is the tricky bit. Anyone have any ideas?This is exactly what led me to BTC. I was trying to find something that would tell me the expected probability or the odds (one can be converted to the other) and then use that as a baseline price, and then I would add 10% to that price and if I could get that on Betfair, then I'd bet on it (or paper trade it). I tried various football data services and eventually BTC.
The trouble I kept running into was just lack of data! As explained in my first rant on variance, I just couldn't get enough data to be able to be SURE that the service had accurate data. When I joined BTC this is exactly what I did with the football. I tested it for a year but sadly I did not understand enough about football to know exactly what to track so the experiment ultimately failed.
I'll explain in another post more about variance, and sample size in particular and show the calculations behind it. But generally speaking, unless betting at extremely low odds (below evens) or having a very good ROI (more than 10%), you will need more data than you can gather in a single season. With football you then have the problem of the shifting dynamics of the teams then rendering the past season's data as irrelevant! Thus, it's VERY difficult to be able to truly assess a profitability of such a service in football.
Racing has a LOT more horses to bet on - hundreds in any given day, but then the odds are generally way higher which means that the data required is even more.
There was one service that I found for racing which had data going back several years which gave me over 42k results to analyse and that was good starting point. I analysed it to death and came up with a strategy that fitted by betting style.
Sadly, it still fell short because it was based on betting with bookies and of course I am restricted everywhere so I could only get a few quid on. I then tried getting the same prices on exchanges, but the value was only to be found early in the morning and then I basically hit the same problem - I could only get a few quid on. I managed to get decent profit from it, but the price of the service was expensive and ate into the profits which were severely capped due to these issues so I ultimately abandoned it
I'm hoping that the BTC back testing software will allow us to develop strategies that we can bet on with good liquidity so that they can scale in order to produce a solid income and not just play money.
-
@caroline-middlebrook
I have wondered, If you had a service that gave you a percentage of each horse/teams chance of winning then blindly backed/layed every one at say 10% above/below that chance whether you would make money.
Obviously not all bets would get taken, and you would have to calculate whether the percentages given were a fair reflection on average over the long term
Getting the percentages is not so hard but calculating whether they are accurate over the long term is the tricky bit. Anyone have any ideas? -
Variance, Part 1
Ahhh variance - the absolute bane of my life as a sports bettor! I had never even heard of the term until I started doing casino offers some years back. I kinda got the concept but I had no idea just how.... variable, everything could be.
One thing that tripped me up is that other than a few failed dabbles, I have always done straight betting, and not trading. I listened to Martin talk about how many trades he likes to see before assessing a strategy and he usually says something along the lines of 200-300 trades, 3-6 months etc.
But that's for trading, not betting, and there is a HUGE difference which I absolutely did not understand at the beginning. At the end of 2021 I started testing 5 football strategies - variations of goal markets. Overs / unders / first half goals / BTTS.
I diligently paper traded them, recording all my data in a spreadsheet week after week. At weekends, this meant long hours as all my strategies are high volume.
My approach is to gather as much data as possible by testing across a large data set and then to filter down afterwards. So, I'd test on a wide odds range with the aim of using that data to be able to filter it down to a narrower, more profitable range. That has always been my approach, though that has changed now with the addition of the racing software.
But anyway, I remember wanting to be "very" sure about my data before filtering so I decided that I would wait until I had a full 500 results on a strategy before analysing it further so that I could be sure that variance was not a factor. Oh how wrong I was!
December 2021 and the first 2 months of 2022 showed great promise for several of my strategies and I was sure I was set for life at that point lol! After some more months I finally had my 500 results and I analysed further, testing odds ranges and some other parameters to come up with a smaller subset with a higher ROI and and this point I started to put money on the strategies - thankfully, only £2 stakes.
As the spring rolled on and we went into summer my previously profitable strategies starting showing massive losses. In hindsight, I now suspect it was due to the end of season effect but at that time, I didn't know enough about football to really understand the seasons and I didn't track where in the season any particular league was.
Around this time the racing software was released and I decided to put the football betting on hold. I may come back to it at some point, but only after BTC releases the back-testing software for football
But one thing that alarmed me was the variance of those footy strategies, especially ones that were betting at what I would consider low odds - around evens. I see people talking on the forums about the maximum losing streak but I find that to be a rather meaningless stat, and instead I track the actual drawdown and maximum drawdown of each strategy.
I see everywhere guidelines that state something along the lines of "use 1% or 1% of your bank and you'll be fine". But again, that advice is generally for trading, not for betting.
I work in points, not percentages and I set up multiple banks. I'll explain the ins and outs of my bank management another time. But anyway, to bet 1% of the bank would mean to me that a 100-point bank size is needed, and to bet 2% of bank would mean to me that only a 50 point bank is needed. I always bet 1 point in every bet, regardless of odds. For lay bets, I lay to a liability and that liability is 1 point.
50 points seems like loads for a strategy that's betting around evens right??
Maybe! But quite possibly not! It took me lots and lots of reading to fully understand how to evaluate a strategy and they are most certainly NOT all the same.
The real question is, how do you know whether the data you have shows a real, genuine profitable edge, or is just luck? We humans tend to vastly under-estimate how much data is really needed to be able to determine a real edge, and we have an uncanny ability to spot patterns in data that is utterly random.
This is something that has tripped me up time and time again when using tipsters, and when evaluating my own strategies.
I think I've rambled on enough for now, but next time I visit this topic (and I have a lot more to say about variance), I'll discuss a particular formula that I use which allows me to rate my confidence in a strategy and answer the question of whether or not it is just luck.
January 2023 update so far...
I'm talking about variance because so far, 3 weeks into January and my strategies are performing terribly! (Currently 160 point loss!) But is it just variance?? It should be! The drawdown is well within the expected parameters of the strategies but bloody hell, it's so hard to swallow when it's actually happening with real money!
That's undoubtedly the thing I struggle the most with - staying the course when in a downswing. I tend to bet at higher odds than most and so my strategies are way more volatile than most which makes this extremely difficult at times.
Still, I'm hoping for a turnaround during the final week of January because over the last 22 months of data (I'm using the BTC data for all of these) there has not been a single month that made less than 50 points, so to end in a large loss would be a massive deviation from the norm.
-
Hi Caroline,
I was very happy to find this blog as I believe we are similar in our approach and quite different to many other betfair traders. I am also a software developer and bet/trade solely with automation. I did try manual trading for a while but it wasn't for me. I'm not the best at removing emotion from my decision-making but my bots are complete experts at it!
I've also been focusing on the horse racing software recently and have a few strategies running automatically. One question I've been asking myself more and more recently is at what point can we say a strategy has positive EV and thus increase the stakes. I've been running a few simulations to answer that question and once done I hope to start a blog here too and share the results. Keep an eye out!
Anyway great to talk to someone who approaches this in a similar way I do. Hope to stay in touch.