[TravisVOX]

I'm pretty sure I know the answer but I wanted to toss it out in case I'm missing something or others have more thoughts.

If a Kelly value when applied to my bankroll produces a wager less than the accepted minimum, does that particular play become a non-bet at that point?

For example, if the minimum bet is $2, but the Kelly value comes out to $1.80... non-bet, or round-up to a bet or...? What if the value comes out to $0.80?

[NunyaBidness]

If the bet size is that small, then the edge, and more importantly $ per hour is so miniscule that you'd be better off using your time on something else.

[yak merchant]

I assume the $2 is only an example. That is why it is prefaced with "for example". But regardless, full kelly at -110 says bet ~2.35% of your bankroll if your model has a play that wins 53.5%. So the answer probably comes down to how accurate your model is. If it is based on unlimited observations (i.e. when to split Aces against 16) then you round up, if it is based on a whole lot of clean data and the variance is low, then you may round up. But if it's based on a couple of seasons of noisy data I'd probably pass. But more likely the previous answer is correct if your minimum bet is bigger than 1/2% of your bankroll, you either need to find a place with a lower minimum bet or grow your bankroll because even at 1/4 kelly staking, if at -110 odds your model says you have a very narrow win expectation of 53.4% you will need to bet more than 1/2% of your bankroll.

[TravisVOX]

Thanks, yak.

I'm working with horse racing in this particular project utilizing projected payoffs of exotic wagers. Thus, in a race, there can be hundreds of combinations yielding small edges which, when the takeout is applied, can create low Kelly values. Because bet minimums in racing can be $1 or $2, the bankroll requirement to reach that minimum with such a small Kelly can be quite large.

[HeeeHAWWWW]

Are you using fully-Kelly? If so don't round up, because you run the risk of straying into -EG territory.

[bihon]

If a Kelly value when applied to my bankroll produces a wager less than the accepted minimum...

And that happens quite often even with the reasonable edge regardless of the starting bankroll and that is exactly why I am strongly against Kelly in betting. Too aggressive.
That's why some will advise a part of it such as 1/2, 1/4 etc, but it is the same as saying 1/2 of Martingale is more applicable.
Yes, it is about maximizing profits, but only if you don't get busted and I have seen tests where even 25% edge finished betting nanodollars.

[yak merchant]

Got it. So it’s been a long time since I dealt with parimutuels, but one of the successful greyhound players I conversed with approximately 15 years ago, was doing exactly what you are doing. He built a model to predict Tri/super payouts using exacta/quinella payouts and other entry specific stats such as speed figures, etc. He then would compare the payouts of those models with another model that had projected probabilities of actual finish to create his tickets. I used a similar method, but was using tote board scraping for just quinellas/exactas so my "payouts” varied less.

I’ll definitely preface this with I’m no expert, but the methodology seems sound to me, but I would really have to*think about the staking plan, as my guess is you would need a much higher projected edge as a buffer and would have to be careful on low probability (high payout) projections, as your ticket(s) could easily move the payout into -EV*territory. Unlike fixed odds betting, you have not only have to worry about error in your “results” model, you have to worry about error in your “payout” model, along with the fact that one guy playing numbers that day at the track could cut payouts in half. You definitely need to make sure you recalculate the projected payouts using pool size and the tickets you are going to buy. When I was doing it I would buy 1 ticket for a combination if it was an overlay of X% and 2 tickets if it was an overlay of Y%, etc. If you applied Kelly to this my guess is it could get crazy fast, especially in the tails. *

So…. long story short even if your “results” model is awesome, my guess is that in Parimutuel you have to be very careful with Kelly, and may go as far to posit that it could be a bad idea on high variance/hard to predict exotics as it will cause you to over stake certain combinations.