Wednesday, 21 July 2010

Fractional Kelly

Joep said...

I can't stop wondering about the practical implementations of using Kelly while trading and I was wondering if you could give your insight as to why you think it's useful.

While I can see the value for using a system where you place larger bets on picks that have more value and are more likely to come in (as opposed to high odds picks), it seems to me that having to correctly price the true odds in play, calculate the Kelly while keeping in mind the available liquidity (that fluctuates a lot in play, specially on lower liquidity sports) are too much practical problems to really use Kelly as opposed to using it as a guideline. Then there is issue that you're not punting and letting all your punts run, but rather you trade out of your positions often, thereby reducing the risk and making Kelly an estimate that is much too low.
The book “Sports Investing: Profiting from Point Spreads“ wasn’t written for traders, but for punters, and although all my bigger bets are made trading in-play, I do play the handicaps and totals for smaller amounts as well. I enjoy reading books on betting and trading because there's often something of interest that you come across.

I totally agree with Joe that it is impractical to use Kelly for trading, at least not for trading in-play sports markets. It could be used when trading stocks, options or futures prices, since they move slower and there is plenty of liquidity in those markets.

As Joep says above, in sports markets, the amounts available fluctuate, often hugely, and the speed at which some markets move also rules out Kelly as a viable option. Kelly works best when you have a very good approximation for the size of your edge, and you intend to let the bet, or rather ‘investment’, run. In a fast and furious NBA or NFL game, there is often not enough time to accurately calculate the edge.

In practical terms, fractional Kelly is probably a better way to go for most of us. The trick of course, is calculating your edge in the first place.


Escapee said...

Kelly Staking near useless in the real world.....
One of its major flaws ( and it has a few ) is its accuracy is dependant on the accuracy of the 'edge' calculation which is just a guessitimate most of the time.

The object of a staking plan is to manage the risk of ruin of a betting bank against seeking maximum returns for that bank.

In the real world, Most of the People seeking out staking plans are trying out a 'system' that they don't even know if it has an edge let alone what figure it might be.

Kelly is over complicated tosh, if your brainy enough to realise the importance of a staking plan to try out a system then all you need to do is set aside a bank for it and work out how many consecutive losses you want to allow the system before you admit its failure. Divide the bank by this figure and you have your bet size.
Re-assess and adjust if the bank drops to 50% or doubles, repeat until bust or rich.

One of the so called advantages of kelly is that it adjusts the bet size according to the near fictional edge calculation.
But in reality this is poppycock! If you're running a system with odds or an 'edge' that requires a stake that varies enough to make any difference then you are actually running 2 ( or more ) different systems under the same bank and using kelly will cloud the view of how each system is perfoming.

There a three keys to winning at gambling:
1) Knowing the good side of a proposition.
2) Knowing how to manage a bankroll
and 3) knowing yourself.
( Puggy Pearson. )

Kelly is mumbo jumbo in the scale of it all.

Best of luck

Anonymous said...

For me, the main problem with Kelly is "bankroll"

What is your "bankroll"?

People will instantly answer "it's the amount of money you have set aside for betting"

Yeah right. These people would give up if they lost what they had set aside? They wouldn't dip into savings?

They'd never bet again.

That's why Kelly isn't practical. Mathematically it is sound although Cassini is right in saying fractional Kelly is normally best.

Oh, not calculating your perceived edge in running shouldn't be too difficult. You shouldn't be having a bet without an edge and to know you have an edge you need to go at least some of the way towards calculating it.