So I've started looking at stats on the Football. Have a good source that can give all the normal stuff, plus each goal scored and the minute scored in.
Now the primary question is sample size and historical relevance.
Having stats back to, say, 2005 might seem like a good plan...but football has changed a fair bit in that time (and from what I can see anecdotally) tends to cycle at least every 4 years as World Cups promote new tactical approaches/strategies, and some times even every 2 years.
As an example, a high pressing, high energy game is currently in vogue, where even as recently as the last Euro Championships, sitting back and counter attacking at pace was the favoured approach...with these different styles potentially having big impacts on when and how goals are scored.
So, the question is...do you look for long term statistical re-occurences, or short term trends (say 1 or 2 seasons)?
The problem with the short term (especially if basing on an individual league and say goal timings) is a lack of data (results in the 10s) which is likely to be statistically insignificant. If looking for a longer term trend you see that anomalies occur at the season by season level, which would leave a strategy based on them looking unprofitable.
I'm' trying to build some stats based on these stats...my presumption is for a Strat/Edge to be worth it, it basically needs to have around a 10% advantage from an Odds perspective (market inefficiency) over the observed statistics to account for some profit, anomalies, and the 5% commission (for those of us new to this).
And then finally, if spotting an inefficiency based on stats with outcomes, would you look to back/lay the inefficiency and run to the actual outcome, and assume long term success, or look to try and trade in and out within the market (I.e. trying to spot when the inefficiency is negated by the market's combined wisdom) which seems much harder to do if basing a strategy or edge on observed statistical outcomes.
Interested to know thoughts and approaches that have been successful or unsuccessful (obviously no actual edges, but the methodologies applied when looking for them and experience of what has worked or not worked at a general level).
Query about use of Stats
Statistics is only half of the data. The other half is the prices available when that data was relevant. You can only backtest an idea if you know what odds you could have opened and traded out at. It could well be that you've found a trend in events, but if that's already reflected in the market prices then there's no way you can profit from it.
So for me, I need to collect live football market prices so that I can know exactly when and at what prices I could have traded in/out at. You see it could be that at some times the odds move out to that 10% edge which you're looking for (or however you're measuring your trigger), and sometimes it'll drift away. Markets prices are constantly changing, sometimes the price is favourable for your strategy, and sometimes it's not. The key is identifying when that happens, and that means sec by sec prices to backtest against.
At least this is what I do, and it works for me. Depending on what leagues/games you're looking at, it means collecting a lot of data though. Betfair are releasing a historical data service soon, however it's not going to be cheap either. So my preference will still be to collect it myself.
So for me, I need to collect live football market prices so that I can know exactly when and at what prices I could have traded in/out at. You see it could be that at some times the odds move out to that 10% edge which you're looking for (or however you're measuring your trigger), and sometimes it'll drift away. Markets prices are constantly changing, sometimes the price is favourable for your strategy, and sometimes it's not. The key is identifying when that happens, and that means sec by sec prices to backtest against.
At least this is what I do, and it works for me. Depending on what leagues/games you're looking at, it means collecting a lot of data though. Betfair are releasing a historical data service soon, however it's not going to be cheap either. So my preference will still be to collect it myself.
Thanks Xitian...my thoughts exactly...based on Stats I could setup some automations that look for a price at a point in term based on historical outcomes and then in some instances it might get favorably matched and not in others.
But without then being able to assess subsequent price changes to know when the market returns to efficiency (when to get out), the only way to go would be to let the open run to conclusion and look for a wins over the long term.
But it doesn't answer the question of what constitutes a significantly relevant sample size to identify a long term predictive trend in a football context.
But without then being able to assess subsequent price changes to know when the market returns to efficiency (when to get out), the only way to go would be to let the open run to conclusion and look for a wins over the long term.
But it doesn't answer the question of what constitutes a significantly relevant sample size to identify a long term predictive trend in a football context.
I can't answer definitively with regards to Soccer Mystic, but again, my experience would tell me to backtest it, which would require you to store what odds Soccer Mystic is predicting for you since you're effectively using it as your trigger condition. So the question you're asking is - If I had followed these trigger criteria in the past, would it have generated a profit? And in order to backtest an answer to that question, you'll need historical Soccer Mystic data. I have no idea if it's easy/possible to save/export live Soccer Mystic data.
The other question you're trying to answer - Is the profit shown from my backtest statistically significant? That one's a bit harder to answer without serious statistics knowledge (which we seem to slightly lack amongst the forum posters!). Personally I'd do two things:
1. Over the sample that you have, measure the overall profit and standard deviation. Does the standard deviation look to be stable within that sample? (That one was a tip from PWebb). How many standard deviations was the overall profit worth? You effectively want there to be very little chance that an outlier will wipe out any profits.
2. Tweak some of your entry/exit parameters and see whether you arrive at a similar profit. If your profit varies wildly, seems uncorrelated with your parameters, sometimes shows an overall loss, then this means your strategy is not robust, and it's likely that the original set of parameters were just lucky.
In the end, even if you have something which shows to be statistically significant, you still need to have the stomach to be able to handle the ups and downs. Football can be particularly up and down because you're effectively depending on goals (or not) which can net a large profit or loss each time. You need to ask yourself whether you're prepared to ride those ups and downs. For example I don't like any strategies which have a chance of being in a drawdown for more than a few days. My football is profitable on 70% of the days, which means it's very unlikely to have 4 continuous losing days. If that happens I know something is up and I can pause it. So in summary, not only do you need to decide if something is statistically significant, you need to ask - Is this type of strategy appropriate to my risk appetite?
The other question you're trying to answer - Is the profit shown from my backtest statistically significant? That one's a bit harder to answer without serious statistics knowledge (which we seem to slightly lack amongst the forum posters!). Personally I'd do two things:
1. Over the sample that you have, measure the overall profit and standard deviation. Does the standard deviation look to be stable within that sample? (That one was a tip from PWebb). How many standard deviations was the overall profit worth? You effectively want there to be very little chance that an outlier will wipe out any profits.
2. Tweak some of your entry/exit parameters and see whether you arrive at a similar profit. If your profit varies wildly, seems uncorrelated with your parameters, sometimes shows an overall loss, then this means your strategy is not robust, and it's likely that the original set of parameters were just lucky.
In the end, even if you have something which shows to be statistically significant, you still need to have the stomach to be able to handle the ups and downs. Football can be particularly up and down because you're effectively depending on goals (or not) which can net a large profit or loss each time. You need to ask yourself whether you're prepared to ride those ups and downs. For example I don't like any strategies which have a chance of being in a drawdown for more than a few days. My football is profitable on 70% of the days, which means it's very unlikely to have 4 continuous losing days. If that happens I know something is up and I can pause it. So in summary, not only do you need to decide if something is statistically significant, you need to ask - Is this type of strategy appropriate to my risk appetite?
Soccer Mystic (like Tennis Trader) is modelled from the prices of several thousand historic matches, it is designed to be a guide as to where the prices will move should either team score at a given moment - so there will be a deviation in some matches but over the long term it will remain accurate to within a few percent.
Their is several videos tutorials on how to use and get the best from it
https://www.youtube.com/playlist?list=P ... 8385C5B599
I have spent the last 5 years looking at the effect of the time of the opening goal to discover that the betting industry does not factor the time of the opening goal and the effectStubbo wrote: ↑Mon May 15, 2017 3:41 pmSo I've started looking at stats on the Football. Have a good source that can give all the normal stuff, plus each goal scored and the minute scored in.
Now the primary question is sample size and historical relevance.
Having stats back to, say, 2005 might seem like a good plan...but football has changed a fair bit in that time (and from what I can see anecdotally) tends to cycle at least every 4 years as World Cups promote new tactical approaches/strategies, and some times even every 2 years.
As an example, a high pressing, high energy game is currently in vogue, where even as recently as the last Euro Championships, sitting back and counter attacking at pace was the favoured approach...with these different styles potentially having big impacts on when and how goals are scored.
So, the question is...do you look for long term statistical re-occurences, or short term trends (say 1 or 2 seasons)?
The problem with the short term (especially if basing on an individual league and say goal timings) is a lack of data (results in the 10s) which is likely to be statistically insignificant. If looking for a longer term trend you see that anomalies occur at the season by season level, which would leave a strategy based on them looking unprofitable.
I'm' trying to build some stats based on these stats...my presumption is for a Strat/Edge to be worth it, it basically needs to have around a 10% advantage from an Odds perspective (market inefficiency) over the observed statistics to account for some profit, anomalies, and the 5% commission (for those of us new to this).
And then finally, if spotting an inefficiency based on stats with outcomes, would you look to back/lay the inefficiency and run to the actual outcome, and assume long term success, or look to try and trade in and out within the market (I.e. trying to spot when the inefficiency is negated by the market's combined wisdom) which seems much harder to do if basing a strategy or edge on observed statistical outcomes.
Interested to know thoughts and approaches that have been successful or unsuccessful (obviously no actual edges, but the methodologies applied when looking for them and experience of what has worked or not worked at a general level).
When you say the "Betting Industry" Cassini, are you talking about Bookmakers or Punters? Assumedly on Betfair Exchange as an example it is factored in via the wisdom of the crowd, with the exception maybe of under/over reaction based on surprise/expected events, or aspects of herd mentality creeping in.
With respect regular in play bookmakers, don't their odds just feed through based on the price at the Exchanges?
Would be good to understand a bit more about your methodology of discovery and (at a personal level) I'd be interested to understand your 5 years of effort which sounds like a mammoth undertaking!
Having done this, are you now finding it straightforward to capitalise on it?
With respect regular in play bookmakers, don't their odds just feed through based on the price at the Exchanges?
Would be good to understand a bit more about your methodology of discovery and (at a personal level) I'd be interested to understand your 5 years of effort which sounds like a mammoth undertaking!
Having done this, are you now finding it straightforward to capitalise on it?
Very interesting post thanks Xtian - how do you collect second-by-second data from Betfair?xitian wrote: ↑Mon May 15, 2017 4:01 pmStatistics is only half of the data. The other half is the prices available when that data was relevant. You can only backtest an idea if you know what odds you could have opened and traded out at. It could well be that you've found a trend in events, but if that's already reflected in the market prices then there's no way you can profit from it.
So for me, I need to collect live football market prices so that I can know exactly when and at what prices I could have traded in/out at. You see it could be that at some times the odds move out to that 10% edge which you're looking for (or however you're measuring your trigger), and sometimes it'll drift away. Markets prices are constantly changing, sometimes the price is favourable for your strategy, and sometimes it's not. The key is identifying when that happens, and that means sec by sec prices to backtest against.
At least this is what I do, and it works for me. Depending on what leagues/games you're looking at, it means collecting a lot of data though. Betfair are releasing a historical data service soon, however it's not going to be cheap either. So my preference will still be to collect it myself.
Stubbo wrote: ↑Tue May 16, 2017 9:12 amWhen you say the "Betting Industry" Cassini, are you talking about Bookmakers or Punters? Assumedly on Betfair Exchange as an example it is factored in via the wisdom of the crowd, with the exception maybe of under/over reaction based on surprise/expected events, or aspects of herd mentality creeping in.
With respect regular in play bookmakers, don't their odds just feed through based on the price at the Exchanges?
Would be good to understand a bit more about your methodology of discovery and (at a personal level) I'd be interested to understand your 5 years of effort which sounds like a mammoth undertaking!
Having done this, are you now finding it straightforward to capitalise on it?
I am giving FREE in play trading analysis in real time > must be going quite well as over 600 000 " hits " in 2 years and not one mention on social media
I write my own programs to collect data directly from the API, however I understand that BetAngel should be able to do something similar from spreadsheet automation. Have a search of the forum. I'm sure I've seen people talk about one for pre-race horses, but I'm not sure about football markets.