History and Development
Our first forays into betting were cautious. We didn’t immediately bet the farm on the output of the Algorithm the very first weekend. And a good thing too. It would have been horrendous. Certainly marriage ending for myself.
We’ve developed and tried many different strategies over the years, each one an improvement on the last until we meet the present day iteration. Here we’ll talk you through the development of our strategies and how each was discarded and improved in turn.
The below strategies were tried and tested over the course of just over a year (September 2019 to November 2020) over 814 matches and 17 different leagues or events, both international and domestic.
As a summary, the iterative strategies tried and discussed below were:
- Just betting on the favourite;
- Finding value bets;
- Using the Kelly Criterion to determine pot size;
- Fine tuning Kelly;
- Further tuning of Kelly.
Hopefully this serves as a guide for everyone as to how this works, no doubt for experienced bettors it may seem simplistic, but the idea was to explain in simple terms how we arrived at our present day strategy in such a way that everyone can understand and trust.
So, to begin ...
Strategy 1: Just Betting On The Favourite
Often people ask if we just bet on the favourite in each match, as indicated by the Algorithm. This ends up down, and you can see how this approach would have fared over the last year below.
The graph below shows how your total pot size would have varied with different strategies of betting on the favourite. The orange shows betting 5% of your pot each time on the favourite, the blue 20% and the grey 50%. The smaller pot size performs best, but ultimately is failing and sits at £20 at the end of the experiment.
Essentially this approach fails for various reasons, but the easiest description would be the familiar expression ‘the house always wins’. Basically the bookies have the odds stacked in their favour. If you convert the bookies odds into percentages, you can see that they all add up to over 100%, on average around 107%. This 7% can roughly be thought of as the bookies profit. It basically means that they are inflating each team’s chances of winning, and therefore will pay less than they ought to if they win.
This strategy was abandoned fairly quickly as a result …
5% Pot Finish £20 -80%
20% Pot Finish £0 -100%
50% Pot Finish £0 -100%
Strategy 2: Value Betting
Therefore, you need to identify the situation when the bookies have mispriced matches, and bet on those matches. What this means in layman's terms is that the bookies are paying out too much on one team, relative to our predicted likelihood of it coming off.
For example: the bookies have a match priced at 6/1, which gives an implied chance of 14.3%. However, the Algorithm gives the side a 20% chance of victory. This means, in the Algorithm’s eyes, that side will win 1 in 5 times, but that the bookies are paying out at the price of 1 in 7 times and therefore, over the course of enough matches, you should end up up.
The mathematical formula is:
Expected Value = (Expected Return * Prob. Win) - (Loss * Prob. Losing)
Here’s how this strategy would play out over the same course of matches.
The larger bet sizes are volatile and tend to 0 fairly quickly, which is unsurprising. Strategies that bet too large a percentage of pot size per bet will generally quickly lose money. However, betting 5% of your pot each bet seems to yield (just) a positive return over the time period (3% seems to be the optimum for this strategy), so there is clearly something in value betting and it forms the underlying principle of our strategies below.
5% Pot Finish £103 +3%
20% Pot Finish £15 -85%
50% Pot Finish £0 -100%
Strategy 3: Kelly - Optimising For Value
So we need to explore this value strategy a bit more in order to maximise the returns from it. The Kelly Criterion is just such a thing - a scientific method for maximising pot returns over time.
This is explained more fully in the link above and elsewhere on the internet but, in essence, it looks at the amount of mispricing in a match AND the likelihood of the outcome, to determine the size bet that you should make in order to maximise your returns.
The mathematical formula is:
% Pot = (Expected Return * Prob. Win) - Prob. Losing
This is the percentage that you see around the website, and in our emails and graphics - the amount of your total 'pot' that you should bet in each match.
Let’s see how this performs over time.
Not too bad. However, you can see that the ‘Full Kelly’ strategy also tends to 0 over time. This is because, despite Kelly’s claims to maximising returns over time, it is also maximally aggressive, i.e it aims to maximise returns at the fastest possible rate and in doing so, runs the risk of also losing quickly, especially if played over an infinite number of matches.
Therefore, most people usually opt for a fractional Kelly (betting a fixed fraction of the amount recommended by Kelly) for a variety of practical reasons in order to reduce volatility. You can see that a 20% fractional bet in orange generates a good, relatively consistent, return over time. We will use fractional bets going forwards in our example as the full, maximally aggressive, strategies always tended to 0 over time.
Fully Kelly Finish £0 -100%
20% Kelly Finish £211 +111%
Strategy 4: Swelly - Fine Tuning Kelly
In investigating the detail behind Kelly’s performance, we discovered a few things that perhaps could enhance the returns.
First up, we had a look at the extremely ‘risky’ matches. These were matches that one side was very heavily favoured over the other. A lot of these matches were identified as ‘value’ by Kelly, but largely just because when the numbers became very small, the relative difference between our predictions and the bookies became much larger.
Practically speaking, this meant that Kelly was identifying lots of really, really long shots as value bets, but the chance of them coming off was miniscule - think Namibia beating New Zealand in Rugby World Cup. We might give them a 1% chance of doing so, and the bookies may be paying out at 500 to 1, which is correctly identified as value by the Kelly Criterion. However, it is very unlikely to happen.
Of course, it is possible that our sample size isn’t large enough to take advantage of these matches (just one coming off would generate enormous returns), which is why we colour code these matches as yellow in our recommendations. Technically they are value bets, but we wouldn’t recommend betting on them. (More on our colour coding can be found here.)
We fine tuned these thresholds (i.e the minimum chance we required a side to win in order to bet) by league and event in order to maximise our return. This strategy was nicknamed the ‘Swelly’, a nod to Kelly and the team member who developed it.
10% Swelly Finish £188 +88%
20% Swelly Finish £277 +177%
Strategy 5: Grelly - Fine Tuning Swelly
Next we ‘experimented’ with the probability of winning that we fed into Swelly. Usually this would be the chance a side has of winning that particular match, but this was substituted with the chance of the prediction being correct (through error originally ... but hey, that’s how cornflakes and penicillin were invented!).
This yielded our best results yet, and was affectionately named the ‘Grelly’ after Kelly and the idiot who first coded the mistake.
This strategy works as it is looking more holistically at the chance the prediction will be correct, rather than the specifics of the match outcome. It also rather handily removes some of the annoying features that Kelly often turned out. For example, outrageously heavy bets were often recommended for relatively heavy favourites in matches, as it was often assumed they generally wouldn’t lose. However, speaking from experience, this wasn’t always the case … happily however, Grelly removes these large bets and instead recommends more appropriately apportioned bets across the board. A handy mistake to make.
5% Grelly Finish £297 +197%
10% Grelly Finish £695 +595%
So that’s it! I hope it makes sense and you roughly understand what goes into the betting recommendations that we make. Let us know if you have any questions.
We’re always developing here at 4Cast HQ, and next on the list is investigating tailoring and optimising some of the Grelly parameters by event, rather than overall, hoping to maximise the return from better performing events.
We are also looking at tipping handicaps as well, and should have that up and running shortly on the website.
The current performance can be seen here, along with a breakdown of the Algorithm’s return by event and over time. Fingers crossed we’re still up ...