Bet size during testing phase of new system?

[Ganchrow]

Not realizing it would have such impact, I wrote down the first vague 'definition' that came to mind. I would have to think more about a mathematically correct definition (if I should require one). Personal changes in money management and investing make this a little unclear at the moment.

That's completely fair. Nevertheless, you were initially asking how to approach bet sizing of a new system from a "mathematical point of view".

If you're truly looking to come up with an answer to this then you will first need to explicitly define your terms. This isn't my requirement, this is a requirement that follows from your own structuring of the problem.

In sports gambling my objective is not to maximize profit. This is true.

The same is true for nearly all bettors, including Kelly bettors.

One reason is that profit cannot be maximized in this field because there simply isn't enough money in the market.

I'm not following you here. There's no reason why one can't attempt to maximize profit subject to some set of capacity constraints.

To me ideal money management would find just the right balance between being profitable and low in stress. Kelly is much too stressful to my liking. Of course, this is a personal preference, and not close to a definition of any use in a mathematical sense.

Actually this represents a good start. You're essentially saying that there exists a trade-off between risk and return. This is standard in portfolio optimization theory, where a trade-off between expected value and variance is made explicit (saying, for example, that an investor might be indifferent between adding 1 unit of expected return versus subtracting 3 units of variance).

In fact, it's fairly easy to show mathematically, that this type of optimization is an approximation of Kelly where all trades (bets) are a very small percentage of the total bankroll.

[Dark Horse]

I'm not following you here. There's no reason why one can't attempt to maximize profit subject to some set of capacity constraints.

The amount of money in circulation in sports betting is nowhere near the amount of money in the stock market. At a certain point it becomes impossible to maximize profit because nobody will take that kind of action. What then? Also, on a sidenote, one could easily make the case that a high profile in sports betting, such as derived from maximizing profit, is counterproductive to maximizing profit.

In fact, it's fairly easy to show mathematically, that this type of optimization is an approximation of Kelly where all trades (bets) are a very small percentage of the total bankroll.

But aren't we back to sample sizes when we have to determine our winning expectation and bet accordingly? We seem to have a different interpretation of (small) sample sizes. You stated that my earlier mentioned initial record, while possible, was quite likely the result of error (which wasn't the case). We both expected it to self-correct.


As expected, the second week of this experiment is showing some clear self-correcting. The overall winning percentage is down to 63%. Yet greater clarity is also beginning to emerge. There appears to be a clear dividing line between overlays. Overlays smaller than 1.5 runs show a winning rate of 77%. Overlays on the other side of that line show only a 51% winning rate. So the experiment continues. What else is an experimental phase but a sharpening of the knife?

[tacomax]

The amount of money in circulation in sports betting is nowhere near the amount of money in the stock market. At a certain point it becomes impossible to maximize profit because nobody will take that kind of action. What then?

While the first sentence is true, it's simply not relevant to your argument. The fact is that the amount that you (or me or anyone else here or anywhere else for that matter) have available to wager isn't going to break the bank at the books.

[Dark Horse]

That may be true in a socialist country, but America has plenty of millionaires. And, if I remember correctly, Ganch has previously stated that, from a Kelly perspective, one's bankroll is all one's money, including all assets. Surely, when we talk about maximizing profit with the help of the Kelly system, we have to define a ceiling after which that is no longer possible. I don't want to get sidetracked into that issue in this thread, though.

Please accept my compliments for knowing what everyone here is wagering. With that kind of clairvoyance, attaining riches through sports betting should be a piece of cake. ( I would estimate that a number of people here could put down money that bookmakers would refuse to take, and so they don't).

[Dark Horse]

One more question about sample size. At what point, from a mathematical perspective, is a system no longer experimental?

I've used the Z-factor, which determines if and to what extent a winning percentage is the result of luck, but the fact that it isn't luck doesn't mean that winning percentage will hold.

One problem seems to be that, if we truly apply strict sample size qualifications, we may not have a whole lot of systems left for betting.

It might help to include most recent results (L10, L20) with any system's total numbers, to get a feel for if it's hot or cold.

[Dark Horse]

A little update (since I said I would).

Overall winning percentage of experimental system has come down to 60% (52-35). This was expected. The interesting thing is that, within the grid I laid out, a number of much higher percentage angles are surfacing. I've seen that hot-cold spot phenomenon before, but what I hadn't realized before is that those angles too, when they hit at too high a percentage (such as 70%), are bound to come down to 'reality'. For instance, overlays within the 1.5 run margin are producing at a 32-13 rate (71%), but while this angle may indeed be profitable, it would at this stage probably be unwise to increase bet size. Better to wait until the percentage settles around 58-60%.

This is just an assumption, but from what some of you have said - that a system basically cannot produce at 70%, but will have to come down -, this seems to make sense as well. So initial conclusion is that while it pays to identify angles with a high winning percentage, bet size should not be increased for those angles until they have experienced a healthy dose of gravity. You could say that an angle hasn't stabilized until then. From this perspective, to increase bet size for a system because it's been hitting at 70% would be exactly the wrong thing to do.

[Dark Horse]

Brief update on experimental system/bet sizing. Getting close to point of playing full bet size.

System record at 65-45 now, or 59%. This is right in the zone Ganch mentioned, so in that sense a sign that system is reaching a healthy degree of stabilization. The good thing is that the same system for overlays of 1.5 runs and lower is still hitting at 69% (41-18). Could 'Ganch's qualification' and a higher winning percentage live side by side?

One more week of experimentation and playing at reduced bet size.

[imgv94]

When someone has a profitable system, dont they usually keep it to themselves?

[Dark Horse]

Yes. This is not about the system itself. It is about bet sizes during the experimental phase(s) of any system. It is interesting to see that the earlier expectations of others have come true, while, at the same time, higher winning percentages within a grid can still be identified.

When judging by sample size, most proven systems are in fact experimental. So it is (somewhat) important to have the initial growth curve of a system before it levels off. Usually this information is lost as the system becomes 'official'.

Still no clarity as to what bet size is preferred, but by comparing different experimental systems in their early stages some sort of common ground may be identifiable.

[Ganchrow]

Looking over your results, DH, it's a little difficult to make a point that your optimal bet size on this strategy is anything other than zero. Let's look at what we have:

Your strategy performed 32-13 in-sample. Out-of-sample, meaning once you decided on strategy specifics, it went 33-32 or about 50.77%. That's not significantluy different from a coinflip.

The fact that your strategy did extremely well in-sample (almost by definition every strategy that reaches the testing phase will do well in-sample) is immaterial. Out-of-sample, it performed no better than a coin flip.

This goes back to what I had been saying earlier about proper data sampling procedure. Selection bias crept in to your sampling methodology and you identified a strategy, which while it may have successfully described your small data sample, it appears of no value in predicting the future.

So what's the next step? IMHO, just start over and try again.

[wrongturn]

Don't know why the winning percentage is important for judging the baseball system, unless it is strictly on the over/under bets. I would think the important indicator is the units won/lost.

[Ganchrow]

Don't know why the winning percentage is important for judging the baseball system, unless it is strictly on the over/under bets. I would think the important indicator is the units won/lost.

DH stated in this post that this strategy is based on 50/50 events.

[Red_Sux]

That may be true in a socialist country, but America has plenty of millionaires. And, if I remember correctly, Ganch has previously stated that, from a Kelly perspective, one's bankroll is all one's money, including all assets. Surely, when we talk about maximizing profit with the help of the Kelly system, we have to define a ceiling after which that is no longer possible. I don't want to get sidetracked into that issue in this thread, though.

Please accept my compliments for knowing what everyone here is wagering. With that kind of clairvoyance, attaining riches through sports betting should be a piece of cake. ( I would estimate that a number of people here could put down money that bookmakers would refuse to take, and so they don't).

if you're a whale gambler trying to get a big bet in, try smurfing. worked for money launders. i beleive no matter how much you try to bet, there is always a way to bet it.

[Dark Horse]

Your strategy performed 32-13 in-sample. Out-of-sample, meaning once you decided on strategy specifics, it went 33-32 or about 50.77%. That's not significantluy different from a coinflip.

The fact that your strategy did extremely well in-sample (almost by definition every strategy that reaches the testing phase will do well in-sample) is immaterial. Out-of-sample, it performed no better than a coin flip.

Incorrect, my friend. I now have clearly identified parameters that, for instance, give me 57-34 and 40-14 angles. I never stated I was going to play all the bets, as you suggested I should earlier. I agree that the initial sample has to be positive for a system to even be identified, but that was merely the initial step.

This goes back to what I had been saying earlier about proper data sampling procedure. Selection bias crept in to your sampling methodology and you identified a strategy, which while it may have successfully described your small data sample, it appears of no value in predicting the future.

So what's the next step? IMHO, just start over and try again.

Sure. You were right all along.

[Ganchrow]

Incorrect, my friend. I now have clearly identified parameters that, for instance, give me 57-34 and 40-14 angles. I never stated I was going to play all the bets, as you suggested I should earlier. I agree that the initial sample has to be positive for a system to even be identified, but that was merely the initial step.

I don't really know what else to say here. You're really leaving me at a loss for words.

You started off in the first post of this thread looking to approach certain bet sizing "from a mathematical point of view", which is what I've tried to do with all my responses in this thread. If you don't personally like the conclusions I've drawn by looking at this issue mathematically, that's completely reasonable and you should certainly proceed in whatever manner makes you most comfortable.

If, however, you truly are looking to approach this issue mathematically, then I'd suggest you strongly consider what I've written. If you're really looking for the "mathematical point of view" ... well to the best of my understanding this is it, like it or not. Might I have erred in my analysis? Sure. It certainly wouldn't be the first time and certainly won't be the last. So if you have any mathematical objections please feel free to share them and we can take it from there. Otherwise, I think we're probably just at one of our impasses again.

[Arilou]

Red, while you can in fact bet an essentially unlimited amount of money on any major event, and in fact you can do it at Pinnacle or Matchbook quite easily, you will pay more in order to get down, decreasing your edge. Thus, at some point your edge will reach zero.

[Dark Horse]

Ganch, count me in (just as dumbfound). I did mention that the overall percentage did indeed come down to the area that you had roughly outlined. However, while the sample size increased, there appeared clear hot and cold spots within the grid I set up. I'm sure you are familiar with grids based on overlays and different spreads, so won't go into that. I was just a little amazed at how easily you returned to the assumption that this must be based on mistakes. Nothing of this requires advanced math, so the potential for mistakes is rather limited.

In any case, the way I see it the stabilization process is close to complete. Won't be updating, since it seems to create more confusion than clarity.

[Ganchrow]

Ganch, count me in (just as dumbfound). I did mention that the overall percentage did indeed come down to the area that you had roughly outlined. However, while the sample size increased, there appeared clear hot and cold spots within the grid I set up. I'm sure you are familiar with grids based on overlays and different spreads, so won't go into that. I was just a little amazed at how easily you returned to the assumption that this must be based on mistakes. Nothing of this requires advanced math, so the potential for mistakes is rather limited.

In any case, the way I see it the stabilization process is close to complete. Won't be updating, since it seems to create more confusion than clarity.

Cool.

... and best of luck to you as always, my Friend.