Justifiable?

[Foxy Dread]

I have a model that is making 10% on all bets, however if I discarded bets where the model is finding extraordinarily high value ie 20%, the margin goes up to 26%.

Is it reasonable to assume that the model´s weakness at the higher margins is because of a piece of missing information and that explains the poor performance in this value range or does the decline in performance suggest something else?

[HeeeHAWWWW]

Is it reasonable to assume that the model´s weakness at the higher margins is because of a piece of missing information and that explains the poor performance in this value range or does the decline in performance suggest something else?

Does often seem to be the case, yes. It might be worth capping the edge (especially if you Kelly stake) - I've found 12% is a good tradeoff, not sacrificing too much profit but drastically cutting the variance.

By the way, you'll almost inevitably find a model producing 10% is not going to continue that over a decent sample.

[Foxy Dread]

The sample size is small, I´m hoping for 3% so any fall off whilst not welcome wouldn´t be unexpected.

[Nick@SI]

Just my opinion but I'd say probably is missing something. That said it is not easy to quantify everything doing line up, weather, ref/ump adjustments along with splits among other things just not going to be easy unless got a couple people working on it. Hopefully can continue to be solid for you capping it at 12% or so like HeeeHAWWWW suggests probably best idea.

[Topo]

Be careful with models. If you incorporate variables, make sure they are logical. Don't just adjust things here and there to produce the best statistical outcome from the past. Those adjustments you make may be like cherry picking the history to produce a false sense of security going forward.

[TravisVOX]

Test your current model against data set aside to make sure you're not over fitting past data.

[Foxy Dread]

Yes, the model has only been tested against new data.

[pringles]

sample size & timeframe ?

[HUY]

I have a model that is making 10% on all bets, however if I discarded bets where the model is finding extraordinarily high value ie 20%, the margin goes up to 26%.

Is it reasonable to assume that the model´s weakness at the higher margins is because of a piece of missing information and that explains the poor performance in this value range or does the decline in performance suggest something else?

What sport and what kind of bets (moneyline, spread etc.) is your model suggesting?