Kelly-Criterion weenies...can you chek this out?

[MonkeyF0cker]

Is a 0.5 multiplier on a 500K BR any different than a 0.25 multiplier on a $1million BR?

What Joe said. And...

You were talking about time. It's a lot easier to adjust the multiplier for different sports than to keep track of different bankrolls for every sport.

[wrongturn]

Agree in theory, 0.5 multiplier on 500K will surpass 0.25 multiplier on 1 million. But in practice, the bet amount will be constrained by book limit once your roll goes to that range, so is there any huge difference? I guess that is one of mathy's points.

[MonkeyF0cker]

Agree in theory, 0.5 multiplier on 500K will surpass 0.25 multiplier on 1 million. But in practice, the bet amount will be constrained by book limit once your roll goes to that range, so is there any huge difference? I guess that is one of mathy's points.

There is no law that you cannot bet at multiple books (assuming that you're not strictly betting off-market numbers).

[scottybweyy]

I've never liked Kelly criterion in sports because it's a bunch of exact numbers assigned according to very inexact info.

Kelly maximizes growth. Fractional Kelly reduces volatility. Very few understand how to calculate what their "bankroll" actually is, how other expenses such as moving money should impact their decisions, how the factor the juice, and how their EV confidence should be reflected in these decisions.

The estimated EV is the most important part and why IMO Kelly is not great. Your EV can change so much due to injuries, weather, roster changes, etc. Even Blackjack APs who miscount the true count due to estimating remaining deck size errors or especially a mistake in the true count can be hurt by using Kelly.

I contend that using Kelly is like picking 1-5 stars and lead pipe locks. Flat betting is the best approach for most gamblers.

[fitguy67]

i agree with many that Kelly is very often mis-applied...i feel that it is mis-applied in the same sense that you might mis-apply a perfectly-good surgical scalpel by using it to do your gardening...

here's my appoach to using (not abusing) the Kelly principle

i take my "perceived edge" to be the long-term yield from whatever intelligently-defined source of picks i am using (be it a capper's output or a model)
i do NOT even try to identify the particular edge of the next play it spits out...i just assume all plays from the same source to be of equal value long term (and the historical yield to date of the entire set of such plays is the single best estimator of it)...

thus, the Kelly concept, applied with the dilution factor i choose tells me what "to win %" I should flatly apply to all plays from this source...so I don't use Kelly as a substitue for flat betting...i use it to determine what flat% i should choose per given set of plays...<yes, the="" yield="" is="" constantly="" toggling...and="" it's="" silly="" to="" be="" micro-recalculating="" it...stick="" with="" same="" %="" till="" either="" a="" new="" equity="" peak="" hit="" (what="" i="" do)...or="" at="" specific="" #="" or="" plays="" days="" have="" passed...just="" consistent...and="" periodically="" re-calculate="" source's="" (hence="" appropriate="" each="" play="" should="" aiming="" at)...that's="" it...i="" use="" kelly="" intelligently="" size="" my="" flat="" "to-win="" %"="" bets="" (all="" of="" them="" next="" re-calc="" due)...not="" individual="" bets


example...following source A (which has an 8% yield over over a large reliably-documented set of plays) at a flat 2% is the 1/4 Kelly prescription...but for another source, B, with a 4% yield long-term yield you would use the same 1/4 Kelly to choose a 1% level...to give you a similar sort of reward/risk "ride"

using long-term historical yield% (ie. $netted/$risked) as the estimator for the edge on the next (and every subsequent) play (until the next objectively-determined recalc point)...and NOT trying to estimate/fabricate one for each new play) is the key to my approach

Kelly is great...
but trying to micro-apply it to specific plays may turn it from a good ally to an enemy</yes,>...an occasionally dangerous one (because of the unnecessary variance-magnification caused by trying to estimate INDIVIDUAL plays)

[MonkeyF0cker]

Kelly maximizes growth. Fractional Kelly reduces volatility. Very few understand how to calculate what their "bankroll" actually is, how other expenses such as moving money should impact their decisions, how the factor the juice, and how their EV confidence should be reflected in these decisions.

The estimated EV is the most important part and why IMO Kelly is not great. Your EV can change so much due to injuries, weather, roster changes, etc. Even Blackjack APs who miscount the true count due to estimating remaining deck size errors or especially a mistake in the true count can be hurt by using Kelly.

I contend that using Kelly is like picking 1-5 stars and lead pipe locks. Flat betting is the best approach for most gamblers.

http://forum.sbrforum.com/handicapper-think-tank/1571755-estimated-edge-error-simulation-kelly-conundrum.html

[TommieGunshot]

I contend that using Kelly is like picking 1-5 stars and lead pipe locks. Flat betting is the best approach for most gamblers.

Most betters don't have an edge. Kelly says for us we should bet zero. That would have a much higher return than flat bets

[chunk]

i agree with many that Kelly is very often mis-applied...i feel that it is mis-applied in the same sense that you might mis-apply a perfectly-good surgical scalpel by using it to do your gardening...

here's my appoach to using (not abusing) the Kelly principle

i take my "perceived edge" to be the long-term yield from whatever intelligently-defined source of picks i am using (be it a capper's output or a model)
i do NOT even try to identify the particular edge of the next play it spits out...i just assume all plays from the same source to be of equal value long term (and the historical yield to date of the entire set of such plays is the single best estimator of it)...

thus, the Kelly concept, applied with the dilution factor i choose tells me what "to win %" I should flatly apply to all plays from this source...so I don't use Kelly as a substitue for flat betting...i use it to determine what flat% i should choose per given set of plays...<yes, the="" yield="" is="" constantly="" toggling...and="" it's="" silly="" to="" be="" micro-recalculating="" it...stick="" with="" same="" %="" till="" either="" a="" new="" equity="" peak="" hit="" (what="" i="" do)...or="" at="" specific="" #="" or="" plays="" days="" have="" passed...just="" consistent...and="" periodically="" re-calculate="" source's="" (hence="" appropriate="" each="" play="" should="" aiming="" at)...that's="" it...i="" use="" kelly="" intelligently="" size="" my="" flat="" "to-win="" %"="" bets="" (all="" of="" them="" next="" re-calc="" due)...not="" individual="" bets


example...following source A (which has an 8% yield over over a large reliably-documented set of plays) at a flat 2% is the 1/4 Kelly prescription...but for another source, B, with a 4% yield long-term yield you would use the same 1/4 Kelly to choose a 1% level...to give you a similar sort of reward/risk "ride"

using long-term historical yield% (ie. $netted/$risked) as the estimator for the edge on the next (and every subsequent) play (until the next objectively-determined recalc point)...and NOT trying to estimate/fabricate one for each new play) is the key to my approach

Kelly is great...
but trying to micro-apply it to specific plays may turn it from a good ally to an enemy</yes,>...an occasionally dangerous one (because of the unnecessary variance-magnification caused by trying to estimate INDIVIDUAL plays)

I agree with this approach without much variation in size. Are most really that confident in perceived edge, sample size, etc? Or maybe it's about comfort level.

[fitguy67]

I agree with this approach without much variation in size. Are most really that confident in perceived edge, sample size, etc? Or maybe it's about comfort level.

Long-term yield I can calculate (not estimate...calculate)...and after LOTS of plays from a particular capper or model...the variance of it drops greatly. Combine THIS quantifiable and rather stable estimator of edge with a "dilution" factor you are comfortable with and you've got a stable %roll basic Kelly betsize. I see no need to magnify this basic betsize (except to update it--again objectively...when the underlying yield of the bet-stream drifts appreciably).

At the reverse end of the comfort scale...ever notice how after a particularly bad episode of negative variance (when the capper/model-builder/tailgater just NEEDS to get revenge)...indexes of "confidence" conveniently escalate to magnify the base bet-sizes to whatever is NEEDED.

[chunk]

[QUOTE=fitguy67;15965182]Long-term yield I can calculate (not estimate...calculate)...and after LOTS of plays from a particular capper or model...the variance of it drops greatly. Combine THIS quantifiable and rather stable estimator of edge with a "dilution" factor you are comfortable with and you've got a stable %roll basic Kelly betsize. I see no need to magnify this basic betsize (except to update it--again objectively...when the underlying yield of the bet-stream drifts appreciably).

At the reverse end of the comfort scale...ever notice how after a particularly bad episode of negative variance (when the capper/model-builder/tailgater just NEEDS to get revenge)...indexes of "confidence" conveniently escalate to magnify the base bet-sizes to whatever is NEEDED.[/QUOTE

Agree with quantifying and objectively applying strategy. I believe that reverse end scale that you refer to is called chasing.....a recipe for disaster. I suspect that most of us have been there at the front end of the learning curve.