Expected Value vs Expected Growth (Kelly criterion Part I)

[barryt]

Re: what is your Bankroll?

Nice discussion on bankroll size, but shouldn't we constrain ourselves to what's in our ADW account(s) for next bet purposes.

a second question is when to switch from flat betting to Kelly criteria.i.e.when is there enough data to actually to start believing we are winning say :70% of our bets at odds of 1.91?. I have 2 seasons of data, 30 games. Should I start Kelly, 1/2 Kelly, 1/4 kelly

im sorta new to using the Kelly criteria and have just used your Kelly calculator and read your article(little over my head), but I seem to get the concept. I skiimmed that chapter in Dick Mitchell's "common sense betting"
trust the questions are not too simple nor mundane!

[barryt]

Using the Kelly Calculator

Using the Kelly calculator for 3 simultaneous independent event with Kelly@1, the output say bet $61 per game. With Kelly set at 1/2 ,output says bet $83. Seems counterintuitive to me
inputs are 3 games : series1: odds -110: win% 79.4:Kelly =1, and .5

I'm presuming I' M doing something wrong..not the calculator.

[HeeeHAWWWW]

Try putting in some realistic numbers :-) Say 54% win rate.

[agendaman]

the kelly criterion doesnt acct. for extended losing streaks-i am a mathematician and this stuff has all been tried before in long run you will go flat broke only successful one i know of is by edward thorpe-card counting-since gone to rest with shoes and 8 decks

[HeeeHAWWWW]

Warren Buffet hasn't done too badly using it.

[u21c3f6]

the kelly criterion doesnt acct. for extended losing streaks-i am a mathematician and this stuff has all been tried before in long run you will go flat broke only successful one i know of is by edward thorpe-card counting-since gone to rest with shoes and 8 decks

I am sorry, but Kelly most certainly does account for extended losing streaks. What you must be referring to is bad inputs for one’s edge. If you are truly a mathematician, then surely you have to agree that Kelly is optimal given correct inputs for one’s edge.
Joe.

[barryt]

Try putting in some realistic numbers :-) Say 54% win rate.

Are you saying that the calculator only works for a certain range of win rates.? The 75% win rate has been achieved over 3 year span. Is actually 31-10 at present and 8-4 last 12.
So I wanted to find out the appropriate bet size for such a win% with games going on simultaneously and the Kelly calculator didn't seem to work properly.
it was not a facetious enquiry.

[LT Profits]

Are you saying that the calculator only works for a certain range of win rates.? The 75% win rate has been achieved over 3 year span. Is actually 31-10 at present and 8-4 last 12.
So I wanted to find out the appropriate bet size for such a win% with games going on simultaneously and the Kelly calculator didn't seem to work properly.
it was not a facetious enquiry.

His point was that 75% at -110 is not sustainable long term, so pointless to expect it to be.

[thom321]

For Kelly to "work", it does require you to not use an implied edge that is higher than what you could achieve over the long term i.e. over an infinite number of bets. The fact that your bet history is 3 years is actually irrelevant. What is relevant is the total number of bets, which in your case is 41. So your sample is very small and your realized win rate over such as small sample is not a reliable forecast for you long term expected win %.

To assume that you can achieve a 75% win rate at an average odds of -110 in the long run is not realistic. Even if you were to realize that kind of long term win rate and edge and you followed Kelly, with a 75% win rate and odds of -110 (52.4% implied win probability) and hence risking 43% of your bankroll per bet, over a large number of bets you should expect a drawdown of 99%. If you were to continue to bet at your current pace (41 bets over 36 months) you should achieve an annualized return of about 175% (average/median scenario). While that might be optimal in the sense that following Kelly would give the highest annualized return given a known edge, it is not practical given the very large drawdown.

Using Kelly when counting cards in Blackjack does make more sense since your edge is known.

Edit: LT Profit posted while I was writing so that is why part of my post is simply rehashing what he said.

Are you saying that the calculator only works for a certain range of win rates.? The 75% win rate has been achieved over 3 year span. Is actually 31-10 at present and 8-4 last 12.
So I wanted to find out the appropriate bet size for such a win% with games going on simultaneously and the Kelly calculator didn't seem to work properly.
it was not a facetious enquiry.

[HeeeHAWWWW]

Are you saying that the calculator only works for a certain range of win rates.? The 75% win rate has been achieved over 3 year span. Is actually 31-10 at present and 8-4 last 12.
So I wanted to find out the appropriate bet size for such a win% with games going on simultaneously and the Kelly calculator didn't seem to work properly.
it was not a facetious enquiry.

Sorry, wasn't meant that way - stuff in text often comes across wrong. It does appear the calculator errors out at the extremes, above about 40% edge in your scenario.

Unfortunately, there's no way those sort of edges are plausible :-( At a guess, I'd suggest you probably have a good edge there, but on top of that have also been very lucky. For example, a binomial on a 57% win/rate would still put your results in the top 1% of the distribution (or top 0.1% on 54%).

[smoke a bowl]

the kelly criterion doesnt acct. for extended losing streaks-i am a mathematician and this stuff has all been tried before in long run you will go flat broke only successful one i know of is by edward thorpe-card counting-since gone to rest with shoes and 8 decks

Of course it accounts for long losing steaks. The only thing it doesn't account for is the player being wrong about what they perceive the edge to be on a given play.

[barryt]

Sorry, wasn't meant that way - stuff in text often comes across wrong. It does appear the calculator errors out at the extremes, above about 40% edge in your scenario.

Unfortunately, there's no way those sort of edges are plausible :-( At a guess, I'd suggest you probably have a good edge there, but on top of that have also been very lucky. For example, a binomial on a 57% win/rate would still put your results in the top 1% of the distribution (or top 0.1% on 54%).

In the interest of clarity, here's .the "system". I didn't invent it :just followed it with success.

in the NBA playoffs, 2nd game of the series ONLY: BET ON ,ATS, the team that lost the first game SU. It's that simple.j
its now 31-11 =75% edge = 42.5% L3 years including 8-5 this year with one round to go; the finals.(i feelits better than 75% in the first round)
IT has some logic to it and the results obviously have exceeded expectations, and 3 years is a short run, but we are all only here for a short run....LOL..It doesn't work for hockey nor baseball as there really is no ATS in those games.
tx for the reply, I don't need to use Kelly any more as I've reached my MAX BET I.e."choke" level
barryt
PS thanx to THOM and LT for their replies also. I agree it's not a predictable result going forward, but what really is? The win probability is not known in advance, only the ODDS. it's not Blackjack nor coin tossing.

[bihon]

It does appear the calculator errors out at the extremes...

That is the right answer.
The edge being unrealistic is irrelevant.

[tsty]

sorry if this is a bump but great post bro

was looking for stuff like this

[Miz]

You are doing the right thing, looking at this older stuff. This guy is legendary.

[ChuckyTheGoat]

Props to Ganch. It was always a head-scratcher:

* If u have a series of +EV bets, how could betting be a bad thing?

Makes sense now. If u overbet your edge, you can expect your B/R to plummet.