I am not a Kelly man, but I can tell you how I back test Kelly in this case. I use a program to go back to each previous day and run the model for that day. I then use the odds for that day to compute the model's expected return for a variety of wagers. For those with positive expected returns I generate a pseudo wager using Kelly (or whatever bet sizing method I am testing). The program then grades the wager based on the actual results. I then analyze the returns of these pseudo wagers. It's quite simple to back test.
-
Comment -
what do you use as your bet at odds?Comment -
Mainly the SBR archived odds that I downloaded. I don't know what these odds are (book and time of odds). I also have saved the odds pages and can select a particular book's last listed odds.Comment -
that is the big problem, for backtesting we dont really know what line we would have gotten and that can have a huge impact on results.
if we use -110 and it really was -117 at the time we would have bet it we would really have a different picture at the endComment -
last listed odds are not very close to actual bet at lines, this is why I say backtesting accurately staking methods is difficult unless you have all the necessary componentsComment -
Comment -
And like clockwork all arguments return to “nobody on the planet can beat the closing line”.Comment -
I run my model on 1000s of games come up with my line and then compare that to the closing line and get an average "off" number, then any game that is off by more than that number i bet.Comment -
oh no, I know my edge is the difference of what I bet at and what it closes at. I only use this as my method to try and find value that MIGHT exist and for sure not how much. like today I have the Knicks +8 (model) and bet at +12 and I do not expect it to come to +8 but I HOPE it comes off 12 enough to cover the margin. I think it will because over 1000s of bets I am avg of 1.22 of the closing line with my model so anything more than that is likely to come closer to my number BUT in no way is my model more accurate than the closing number. I hope that clarifies, I write bad so it might have just made it more muddy LOLComment -
No, I get it.
But I still think it illustrates that you do try to estimate an edge.
I think just about everyone does.
From the gut feel bettor who thinks something is "value", to the semi serious bettor who really can spot anomalies from experience, right up to serious modelers.
I can see how your way works well enough for you to not want to over complicate it or add risk you are unsure about to use Kelly.
I like Alfa's way of simplifying it too.
But if someone, pretty much anyone really, wants to concentrate on building a model with enough accuracy to benefit from Kelly, I think it can be done.
I think it's simply the hard work that stops the 99% of people choosing to do that. Not skill or brains or lack of perfection in edge calculation.
Anyway, that's my 2 cents for what they are worth.
Carry on
.Comment -
u r right optional but the fact is the estimated edge just adds another dimension of risk to the betting risk. its already risky enough no need to add more what ifs, we have enough of that without kelly. Lose because you lose not because you bet the wrong amount!
also I do like alfas idea of adjusting your stakes monthly or every 100 wagers or something to minimize draw down and regulate ROILast edited by danshan11; 01-28-19, 04:55 PM.Comment -
Actually I don’t use Kelly as I understand the maf and I work 80-100 hours a week so sports betting is nothing I have time for any longer, but I get that the mafs is hard. I don’t really think I’m mocking anybody but am fascinated in the irony of your projection. Please see post #102. It explains why if you are on the correct side of things Kelly makes sense. And I agree with you that Kelly isn’t for 99.9% of people, but your I can’t do it so nobody can attitude is what drives me crazy. There are 350 million Americans, 10 % of which bet on sports. Your insistence that nobody can do it is maddening. When your own math says 35000 people should be able to, so why couldn’t there be 10 on this forum?Comment -
post 102 is the same thing people keep saying over and over estimate your edge and back test. The problem is that we cant do that because we dont have the line we would have bet at, we can guess we can use a certain point but for me I never bet the same time I bet when the line is where I need it to be that can be when limits rise or the next morning. I think the 1 out of 100,000,000,000 can so that means you can is what drives me crazy, yes there is one guy out there out of millions that actually can just pick winners, with no rhyme or reason and crush the line so that means we can all do it. I dont think its possible in the betting world someone can actually know their edge because of injuries and last minute scratches on individual games and if you dont KNOW (not estimate) your edge kelly is just another what if and adds a variable that most people dont need. and when you say could be 10 on this forum oh yes for sure and I am sure that those 10 dont give a hoot or holler about what the heck I am saying about anything and I dont think my limited narrow and non educated opinion means anything to them. I am more for entertainment for those 10 BUT the rest should not even be thinking about kelly cause honestly besides those 10 guys, the rest dont even have an edge period!Comment -
but I would not advise that method because the actual testing is flawed because it is based on estimated edges and it also assumes you always have an edge of 1 to 6% which is nearly impossible and of course not accurate.
the issue with Kelly is this simply put
you dont know your edge on individual games and this would cause your bet on that ind game to possibly be the wrong amount, now this is very similar to the result from fixed profit betting. if I bet a fixed profit bet on a -200 favorite I bet 2 units to win 1 and if it closes at -150 I over bet that game, this could possibly have a negative compounding effect.
if you use an avg edge you are really running a risk of variance causing serious draw down with kelly
and I know alfa tried to solve the adjusted bankroll dilemma but it still is an issue and an amplified issue with kelly.
if you pull a DRH and use a model with success for 1000s of bets and now that model is garbage or variance gets you, kelly will draw you down hard.
I think this screams the safest and best method is
flat betting of betting 1% of bankroll and adjust that 1% amount on the current bankroll every 100 bets or so, might be a way to tighten that up.
I analyze bettor data and I can see clearly the couple fatal flaws they make
1 they bet a flat amount until bankroll is gone example they have 10k they bet $100 a game and do that past the adjust needed point and now they are risking way more or less than 1% on each bet, a $100 bet on a 15k bankroll is too low and will lower roi if you actually have an edge and $100 bets on a 4k bankroll is even more detrimental as it is 2.5%ish of total bankroll and that will sink you
2 the typical bettor is at a norm spot of 50-50 say but in reality they are at 48-52
so they bet 10 games and deviations tell us
they should win 5 games
and in a few rolls they will win 5 games
but they are more likely to win 1 than they are to win 9 and eventually this deviation gets them because of the less than 50 true win probability.
something like this
10 games
chance to win 5 48%
chance to win only 1 is 4%
chance to win 9 is 2%
so they are twice as likely to go bankrupt as they are to double up
these are not exact numbers but the point is valid
maybe heehaw can run the deviation ratios accurately but the point is valid.Comment -
Yeah. To clarify, since the original wasn't entirely clear, "5% errors" means a random edge estimate error between -5% and +5%.
Of course, the mean average error of zero is a rather large caveat. My guess is many +EV players have a very skewed distribution for edge estimate errors, eg -8% to +1%. Thus >> over-betting, the Kelly danger.Comment -
right or wrong? you take a large group of say totals and you say ok
the total is 205 on every game
some games are 188 and some are 235 but they average out to 205
so using 205 on every game is a good idea since it will average out to 205 anywayComment -
sure but are they close enough to base your bet size on?Comment
