Huge edge in cents but tiny units?

[sycoogtit]

Okay I'm new to the model world, so don't slam me for asking a stupid question. I watched Justin's video about... I forget. But at the end he said if you can make a model that gives you an average 2 cent edge over at least 100 hundred games, then you might be on to something.

I've created two models. 1 is getting about a 2.5 cent edge per game, and the other is getting a whopping 7.5 cent edge per game! Holy crap! What I think is weird is the former has fewer games but is somehow doing better unit-wise. That seems counter-intuitive to me. I thought the bigger the edge, the more plentiful the units.

Can someone explain this to me? I'm a lowly data-miner with limited knowledge of this whole "model" thing.

[sycoogtit]

Nevermind, the latter model was flawed.

[sycoogtit]

No it wasn't flawed. I'm still confused.

[Justin7]

How large are your two samples?

[sycoogtit]

They're both 90 games.

EDIT:
Okay not exactly 90. The first has 84 and the second has 91.

[TomG]

I would have to be very confident in my model to bet an average edge of about 2 cents. Actually, I take that back. I would not bet something that thin. Different min-edges for different folks, though.

[Justin7]

2 cents on a 90-game sample is a little small. How big are the swings? Are some +50, others -40? Or is it consistently beating the no-vig number by 2.5 and 7.5 cents?

[sycoogtit]

Yeah they're pretty big. They both have swings of up to 30 each way.

[sycoogtit]

2 cents on a 90-game sample is a little small. How big are the swings? Are some +50, others -40? Or is it consistently beating the no-vig number by 2.5 and 7.5 cents?

I'm confused about this. In your video I thought you said that if the line moves 5 cents in your direction then you would have a 5 cent edge and vice-versa. So if the line moves by 20 points, don't you have to reflect that as a minus 20 or plus 20 cent edge?

[Justin7]

Yes, but you need to figure out how much "noise" there is.

Assume every line move is a random number, equally likely to be positive as negative. Your mean is 2.5. What is your standard deviation? What is your z-score for the sample?

Video was an introduction -- just a starting point.

[sycoogtit]

Oh I get it. Well crap. Maybe I should stick to data mining.

If the model produces positive units over say 200 games and it keeps a +2 cent edge, would it be considered random still?