Of all the different markets available for each race to bet on, few match this kind of betting for simplicity.
Simplicity in not only understanding the bet itself, but also in its handicapping.
In fact, if there was any one single market that was compulsory for anyone wanting to bet on Formula One to master before moving onto other markets it would be betting on whether there will be a safety car appearance at each gp.
But what makes these markets a great place to learn how to handicap markets?
Put simply it’s the fact there there are only two betting options available to bet, and the relative ease in which these options can be assessed. No only that, you’re assessment can go as deep as you’re willing to delve. Your model can be very simple and put together in a matter of minutes, or quite deep and complex requiring hours of research and number crunching.
But regardless of what model you come up with you will gain (perhaps for the first time in your punting career), an understanding of the very fundamentals of handicapping, probability and most importantly expected value. These, along with bankroll management are the cornerstones of any successful punter.
So, where do we begin?
First of all we need to identify what we are hoping our model will develop. Of course this is very easy, we want to know the probability of a SC being deployed in a given race. To help us we will use the example of the 2015 Australian GP being held in Melbourne (note the specific circuit, not just gps as some countries have multiple circuits), and try to develop a model which we think will accurately predict whether there will be pace cars during the Formula One race.
The first question you have probably asked is “was there a SC in last year’s race?”
It so happens there was! But of course this isn’t quite enough information to give us the probability of there being one in 2015, it obviously to anyone who has watched any motorsport that the probability of pace cars isn’t 100%.
We obviously need more information. This is where you will need to do some of your own work as unfortunately there aren’t many readily available SC analyses (at least that I have found). Again you will need to identify exactly what information you are looking for before beginning your search. Are you looking for the number of races with safety cars? The number of pace car appearances? The reason for each appearance? Each research question will have its own unique answers and degree of difficulty to find.
For simple “was there a SC in a grand prix” style statistics the best resource I have found is Fistats. If you want more in-depth knowledge of why there were pace cars, or how many appearances it made during a race you may have to collate these yourself from race reports. But for now I am happy to know in which races the SC appeared at least once.
So back to our example, I now know that not only did it make an appearance at the Australian GP in 2014, but also another four times between 2008 and 2013. This gives a total of 5 appearances from 7 races, or 71% of races.
Problem solved! The probability of a SC in the 2015 Australian GP is 71%.
Or is it?
Seven races is not a very large sample size, what if in the five races with a safety car it was raining? Or maybe there were simply cars stopped in unfortunate positions purely by bad luck? Or what if there were kangaroos on the track in some years which have since been relocated?
The fact is from raw numbers we simply don’t know the answers to those questions. The 71% may be an accurate probability, but we can potentially make it better.
Lets look at another model with a much larger sample size, how often the pace cars are deployed across all circuits. Using the data from Fistats we can analyse all races from 2008 to 2014, 132 races in total. This is of course a much larger sample than simply the races held at Albert Park.
Across all 132 races pace cars appeared in 58 (44% of races). Comparing this to our earlier number we see that on average pace cars are 33% more likely to appear in Melbourne than the average race. This large difference however shows us that simply the average probability of a SC appearance in a race isn’t enough to give us an accurate prediction.
But with just this little amount of information we can actually develop a reasonable (yet simple and possibly not profitable), model. By combining the probability of a safety car in Melbourne, with the probability of pace cars appearing at any gp we can reduce the chance of random events skewing our model, yet still account for the higher historical occurrences in Melbourne. By combining the probabilities we are left with a 58% chance of pace cars. This model may actually pass on its own (but would need to be tested, see below), but again I think we can do better.
Separating Different Circuits
To make our model even more accurate I am interested in why safety cars are 33% more likely in Melbourne than the average race. By knowing this, I may be able to more accurately predict not only our example race, but races at all other circuits.
So what do I know about Melbourne as a circuit that would allow me to categorize it with other circuits as well as differentiate it from others? It is a street circuit.
Street circuits are notorious for their high number of pace car appearances. Their close walls and surfaces without built up rubber are difficult for drivers and leave very little margin for error. In theory, more accidents means more safety cars, but is this the case in reality?
To test this we compare the appearances of safety cars at street circuits in our sample (Melbourne, Monaco, Montreal, Valencia and Marina Bay), to all other circuits. While street circuits made up only 32 of our 132 race sample, they accounted for 23 of the 58 pace cars. We would expect pace cars to appear in 72% of races held at street circuits compared to only 35% at other tracks, a massive 37% difference!
It seems likely that my hypothesis about increased difficulty for the driver may have something to it. With this in mind there is another variable we can analyse which can play havoc with even the best of drivers. Rain.
Again we run an analysis of our data to see in which races it rained so we can compare the probability for wet and dry tracks. Of our 132 race sample it rained in 21 races (although our statistics do not give us the severity of the rain), in which 62% of races saw a SC. For the remaining 114 dry races pace cars were only required on 45 occasions (39%). So again, it seems as though the tougher a race is for drivers, the more likely a safety car is to appear.
Breaking down our analysis by weather and circuit type again shows that wet races have a higher percentage of safety cars (75% for street circuits, 59% for permanent circuits), compared to dry races (65% for street circuits, 30% for permanent circuits).
Digging this deep and purely relying on percentages though begins to get dangerous. While it may seem like a good idea to model a wet street race at 75% probability of safety cars, this is only drawn from a sample of 4 races. To ensure that our comparisons are statistically significant it is possible to run a Chi-square test to see if our results are significantly above the normal expected frequencies (i.e our number of pace cars would occur less than 5% of the time purely by chance).
Unsurprisingly the weather comparisons on street circuits was not significant with such a small sample size. However wet tracks verses dry tracks was significant for permanent circuits, and only sightly outside significance (p=0.056 for the stats heads), across all circuits. The difference in safety car appearances on street circuits compared to permanent circuits was also significant.
We are now armed with some pretty important information to help us develop our model. Street circuits are 37% more likely to have pace cars than a permanent circuit, and a wet race on a permanent circuit is 29% more likely to have pace cars than a dry race. We also know that weather makes no difference in whether or not there is a safety car on street circuits.
Putting Together Our Model
Previously on our very simplistic model of combining the average pace car appearance in Melbourne with the overall average probability of a SC gave us a probability of 58%.
Now knowing that street circuits have a much higher probability (72%) of a safety car occurrence we can combine that with the average Melbourne pace car appearance (71%), to give us a new probability of 71.5%. To convert this into a betting odd we simply divide 100 by our probability to get $1.40. So when looking to place a bet on there being a SC in Melbourne we would want to do so on odds greater than $1.40 to ensure we are placing positive expected value bets and will win over the long term (assuming our model is more accurate than the bookmakers).
Conversely if we wanted to bet on there being no pace cars in the race the probability would be 28.5%, or $3.51.
Lets also take a look a a few other hypothetical races to see how the model can be applied.
The Malaysian GP is on a permanent circuit, however is expecting rain for race day. Pace cars have appeared in only one race at Sepang since 2008 (ironically it was wet), giving a probability of 14%. This is combined with the probability for a wet permanent circuit (59%), to give a probability of 36.5% or $2.74.
The next race in China is again on a permanent circuit, but no rain is forecast. With 29% of races seeing safety cars since 2008 and 30% of dry permanent circuit races having pace cars the probability is 29.5% or $3.39.
For those wondering, the circuits with the highest and lowest average pace car appearances in our data set are Maria Bay with 100% of races having seen a SC, and Sepang and Sakhair each with only 14%.
Testing Your Model
The above model certainly may not be a path to riches, or it may be! As I stated earlier in the article, a much more simplistic model may actually be enough to beat the bookmakers.
To know this it is important to test many model you have and you can do this a number of different ways. One way is seeing if you were able to correctly predict whether there was a safety car in a race and whether these predictions were better than chance.
Another is to compare your modeled prices to bookmaker prices and see if you were able to profit over a period of time (remembering only to bet when there is value). This requires access to previous betting market data which could be very difficult to source for a market such as this, however you can also “paper bet” you model moving forward to see how it tracks.
In back testing our model above on the 2014 Formula 1 season the outcome with the highest probability was correct in 15 of the 19 races. This is above the expected strike rate of 50%, however does not guarantee that the model would have made a profit.
It’s also important to remember that when doing any sort of back testing do not include data from the events you are testing as this will corrupt your results and could be a very costly betting mistake!
Other Possible Factors You Could Include
By no means is the model described above perfect, it is merely something to get you thinking about what you might be able to develop yourself. With more time and the willingness to dive into statistics it would be possible to investigate any number of factors which could also be incorporated into your model.
Do rookie drivers crash more early in the season? Do cars have more failures earlier in the season or later in the season as components age? Are there more crashes when certain drivers are buried back in the pack after qualifying?
Any number of possible factors could be considered and incorporated into your model. And the great thing is once you learn to predict whether there will be a pace car in a Formula 1 race, it becomes much easier to predict many other betting markets.
Those wanting to try their luck with a punt on the SC appearing at races throughout the year can find markets available at Bet Easy, Ladbrokes, Centrebet and William Hill.