Now I'm not a serious sports bettor by any stretch, aside from the occasional $5-$20 or a beer between friends.

In general I've always believed systems in sports betting to be a fad or luck. Maybe in a week or two you'll get positive results, but over time it will even out.

But what if you have a system and over a period of time you are constantly getting positive results? How long would that period of time be until you considered it works?

I'm currently testing a theory of mine in the NFL. I've yet to have a losing week and as of Monday night I'm 14-2. I'm aware it's only week 6 of the season, and it could just be a hot start. But when would you consider hey, maybe I got something here?

Let me just make it easy for you...you dont have an edge. The guys shaping the lines are betting tens of thousands of dollars at a clip. Good for you for running good for a while and all the power to you. But you will lose in the long run, im sorry to say.

This doesn't answer the question...

There are two ways really. One is to back test against historical data and the other is to just wait and see. I would caution against getting overly excited because you could easily be experincing the positive side of variance. For example, in a coin flip where we flip a fair coin 100x times we'll see results ranging on average from 40x heads/tails to 60x heads/tails 19 out of 20 times we run that test but the true probability is 50/50. So a sports bettor that is getting true 50/50 on his picks can see variance go as high as 60% and as low as 40% over 100x samples on a very regular basis and can see even larger swings but more rarely.

Depending on the system you may be able to use something like the sports database to quickly see historical results http://sportsdatabase.com/nfl/query

If your system gets results 1-2 standard deviations above the mean(say 50/50 and a std of 5 so 55-60%) over a good sample then it is almost certainly profitable long term. Of course that depends on the odds being given too. If you have a system that is 60% accurate at picking if the favorite will win that will still lose money long term on most moneylines.

It actually does. In the first couple sentences actually.

Unless, of course, you actually think a guy betting 5 dollars a pop has an edge in NFL markets. In which case Id like to know what other things you believe.

The question he is effectively asking is how many games until statistical significance to discern whether his strategy is likely profitable or not.

If you want to assume he's a fish based on the way he's talking about betting, then cool, go ahead and be a douchebag about it. It still doesn't answer the question he's asking.

OP, the answer is somewhere in the thousands, maybe ten thousands of games.

Has there even been 10k NFL games played?

There are two errors in this one sentence. Firstly you probably meant to say that if you have a good win percentage over a large sample it is probably profitable long term. But that's because if the win percentage stays high as the sample gets larger, the number of standard deviations above the mean also gets larger. But being only one standard deviation above the mean doesn't increase the strength of your evidence as the sample gets larger. It could mean that you were 55 out of 100. But if your sample was 400 it would mean that you were only 52.5%%. In other words if you measure evidence by the number of standard deviations only, large sample sizes wont strengthen that evidence much.

The second error is worse. Because, everyone including cancer researchers seem to make it. The chances that an experiment will get a result two standard deviations above the mean is about one in 40 if only random factors are at play. But that doesn't mean that if you get such a result there is a 39/40 chance that it was due to something other than random factors. In fact nothing of the sort is true. Especially when the experiment purports to show that a technique will work even though the number of techniques in the past that have actually worked have been few or none. Since way fewer than one out of 40 proposed techniques (or experimental drugs) actually work, a two standard deviation trial result is not only not enough to proclaim almost certainty that a method works you can't even claim that it probably works.

It depends how his model works. If it is consistently predicting several leading indicators that are conclusive to a final game result then he could be on to something. But if he is simply picking numbers out of a hat and going to the window then he has nothing.

lol "tens of thousands of games'

its always so obvious the people who dont really bet

Okay, then what is the answer?

In which market have you placed over 10k bets?

Were you guessing until those 10k bets came through?

That isn't relevant. It doesn't matter if I've placed 10 bets in my life or 100,000. The question OP is asking is a statistics question. And not a complicated one at that. You can google this info or even get it from past threads on 2+2...

How isn't it relevant?

How many people do you know that have cleared 10k bets from a single model?

Why would anyone bet if it took 10k bets to know if their system works or not? Are they just gambling until 10k bets clear?

Where could you even find a market where you can use a model for 10k bets without tweaking it?

You keep trying to turn it into something it's not. This is about sample size. You're asking me questions to which the answers are obvious. It isn't that complex. It also isn't that concrete either. You guys play poker too, right? Would you ask me how tf someone one tables several 100k hands within a reasonable timeframe to just figure out if they're profitable or not? No. It doesn't matter how feasible the act of trying to empirically do it is. The math stays the same.

Do you understand how stupid you sound?

Why would you even place a bet if you don't know if you are beating the market?

also lol@comparing poker where you are vsing other people to making a model where you predict an outcome of an event

Let's go ahead and say that I am indeed stupid. Cool. What is the answer to OP's question?

The sample size depends on how strong the evidence is and how plausible the theory being tested is. Also the sample can come from past results rather than future results IF the theory wasn't derived by looking at past results.

It's very easy

When I finish my **** and see that it's beating pinnacle lines

Then I place my bet because I know I have an edge

Why else would you ****ing place a bet?

back testing is for suckers

Again...how could you ever make a model (or "theory" as you interestingly call it) without using past results?

Do you think that coming in here every couple years and pointing out ridiculous, theoretical examples that have no bearing in the real world makes you look smart? And while I have your presumed attention...sitting down to play poker, playing one hand, then walking around the room for 15 minutes, over and over and over, every time you play, is rude.

If the past results you used to devise a possible winning algorithm came from a different set of data than the set of data that you were using to test it, then it is OK. If it came from the data itself you would need it to be overwhelming before you had a right to believe you were on to something.

And of course you could devise a plausible theory without ever looking at past results. Especially if it is likely that no one else had thought of it despite its plausibility. Something like theorizing that quarterbacks who injure their non throwing hand are going to more likely run to one side of the field and betting against them when that side contains the opposition's star linebacker. (I have no idea if that is actually plausible. It was just an example of what I was saying.)

you argue like a man who has never made a bet based on his own knowledge. Your examples are ridiculous. If you dont make/use a model, youll never know if what youre theorizing about has already been incorporated into line. If youre doing anything resembling handicapping, you need your own numbers. Only way to get your own numbers is use to past results. (Or buy someone elses numbers, who have used past results).

Also, in your linebacker example...dont you think that if YOU know that the QB is more likely to run a certain direction, the other team does too? And will position their star linebacker on that side? And maybe everyone already knows all this and its already been factored into the line? And youd be making sucker bet after sucker bet?

I'm also curious about different perspectives on this.

I was screwing around last year and put together a college basketball total system. Within certain parameters(ie big enough difference between my total and vegas), it went 102-64-2 (61.44%). Too small of a sample for sure. How much does it change that the lines moved 99-58-11 in my direction?? That seems to be an indication that something went right, but it could (and most likely is) simply positive variance. It would take 7 years at this rate just to get a 1K sample. Huge waste of time to keep putting in the effort to track it or is there at least a possibility of being on to something?