Unlike with baseball or basketball, you have to bet well in advance of the actual events in football. Also, 90% of the games each week occur on one day. So, many of your bets should be treated as simultaneous wagers, since they're all occurring within one betting period.

With simultaneous events, bet sizing for each game is dependent on the number of other games that you're betting on, and the edge that you have on them. You don't just bet X% on a 55% game - you may bet much more or less on that game depending on the other games occurring at the same time.

So, if you want to bet on a Thursday game and it's Sunday, then you must either treat is as simultaneous with all of the Saturday games, or you have to wait until AFTER the Thursday game to bet on all of the Saturday games. This obviously causes problems with betting openers.

You can't bet on a single game until you know every single game that you're going to bet on that week. If you're waiting on news for a particular game, injury reports, or whatever - you have to put off your entire batch of orders until you get everything set.

This causes problems when your analysis takes several days - you can't do analysis on game A and then bet on it and move to game B, because simultaneous kelly wagering is dependent, and so you have to wait until you finish. This makes it really, really difficult to bet openers, and you're betting against much sharper lines.

Do you basically have to flat bet football? Or do you just approximate some bet sizes relative to edge and then resize each week?

Halp me.

If you definitely want to bet openers, I'd go ahead and group the Thursday games with the Sunday games. Other than that, I don't see what the problem is. Yes, you have to know every bet for the week, but if you're waiting for an injury report, so is the line.

Edit: I guess it would be a problem if your analysis takes a super long time. By the time you finish analysis for your last game, the circumstances for the first game you analyzed might have changed. In that case, I dunno.

If you have some idea of how many (and what edge) plays you're likely to have, based on backtesting, you can run excel on those past weeks, and see how much your bets generally get reduced, and pre-reduce them that much.. and add on more later, if you're coming out way low and you still like it. Alternatively, if you have a lot of faith the market will agree, you can bet and then middle some amount of it off later. If the market frequently doesn't agree, you'd be left with a lot of ugly positions if you go that route.

This is generally what I figured. I ran a bunch of tests with 10-15 53-58% games, and taking the kelly sized bet as if it was 1 game and cutting it in half (so 54% --> 1.7%, 55% --> 2.75%, 56%-->3.8%, 57%-->4.85%, 58%-->5.9%) is actually remarkably close to what the final simultaneous sized wager ends up being, and is usually just a bit short (which is preferable) of the actual number.

So if I want to bet 60% Kelly on my simultaneous wagers, betting 30% of a full kelly 1 game bet and then adjusting after that seems like it could get me pretty close.

If your analysis is sound, and you multiple outs (most importantly matchbook), you should be able to hedge your "overbets" for a gain or minimal loss. So in this case you can approximate (or overbet) and adjust them later in the week.

If this question is for Bob's clients, you're getting into things that are too advanced for most of them.

ganchrow had some good posts about this on sbrforum

how do you calc kelly for 10+ bets without your computer blowing up from too much math

10 bets is well within excel solver capabilities as long as you can call them independent.

ah thanks i'll have to install excel. i usually use the one on the sbr webpage but it crashes if you put in too many at once

Ganchrow admits that the larger multi team parlays aren't nearly as important as the 2-3 team parlays anyways.

I use the SBR one for up to 15 simultaneous events and it works fine. Each event takes about 4x longer to calculate, I can do 15 in about 10 minutes.

I'll be getting into like 20-40 per week later on in the season, not sure how I'm going to handle that yet, but the differences >15 are pretty nominal I think.

I like this. Say the line always moves to you exact expected line, and you can get out at matchbook and lose 2 cents doing so. Then you would probably bet the Thursday game assuming you had no other action. If the line doesn't move at all, then you have the problem you originally stated. I assume Bob's picks would be somewhere in the middle, where the line does move, but not all the way.

Yao,

Why do you feel qualified to calculate this as you don't have any experience with KC?

All,

AMPL GIMO.

What part of my post do you have a problem with?

Nice, so I basically got it figured out.

I bet a bunch early, hedged out later, and tried a .3 multiplier which I felt would get me pretty close to 1/2 Kelly overall. I can't calculate for the full 18 since the calculator maxes out at 15, but here are the first 15 games that I had on the sched:

0.98% 1.76% 55.45%
1.02% 1.85% 55.23%
1.71% 3.17% 53.87%
1.92% 3.63% 52.86%
0.67% 1.20% 55.83%
2.14% 4.41% 48.54%
0.90% 1.42% 63.54%
0.39% 0.69% 56.04%
1.66% 3.35% 49.49%
1.30% 2.38% 54.59%
1.09% 1.99% 55.03%
1.34% 2.46% 54.51%
0.57% 0.85% 66.52%
1.62% 3.52% 46.00%
1.11% 2.02% 54.88%

Column 1 is the total amount of my bankroll that I actually wagered
Column 2 is the output for Full Kelly as described by my estimated edge and the odds that we were being given
Column 3 is the fraction of full kelly that i ended up wagering.

So as you can see, all the numbers fall between 46% and 66% which is great - I didn't come close to overbetting a single game, but I still kept it at around 1/2 Kelly across the board. Since there's so much error in my estimations before we get to the math model in week 5, I'll keep it at around .55 Kelly. After week 5, I'll bump that up to around .6 or so.

Not sure how this will change after week 5 when I start betting 30-40 games a week.

my bankroll is always as big as my equity... NOT.

I assume that the excel solver suggestion was directed at using numerical methods to approximately maximize log W -- which is a good way to go. (Then bet .5 to .65 Kelly.)