Kelly with correlated outcomes

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • Tomato
    SBR MVP
    • 01-29-09
    • 1251

    #1
    Kelly with correlated outcomes
    Denmark-Netherlands

    Pinnacle probabilities:
    Netherlands -179 = 62.958%
    Draw +309 = 23.992%
    Denmark +652 = 13.049%

    Netherlands -1 +102 = 49.6975%
    Denmark +1 -109 = 51.3025%

    BetOnline odds:
    Draw +350 = 22.2222% = 7.967% EDGE
    Denmark +1 +110 = 47.619% = 7.735% EDGE

    Should you take into consideration the correlation between the draw and +1 (If draw wins, +1 +110 wins) and adjust the structure of your bets (and how would you do this), or treat them as separate markets and bet both full Kelly?
  • Optional
    Administrator
    • 06-10-10
    • 61395

    #2
    I think you should take the correlation into account if trying to follow Kelly, as it assumes independent bets. (but don't know how)
    .
    Comment
    • Pancho sanza
      SBR Sharp
      • 10-18-07
      • 386

      #3
      Originally posted by Tomato
      Denmark-Netherlands

      Pinnacle probabilities:
      Netherlands -179 = 62.958%
      Draw +309 = 23.992%
      Denmark +652 = 13.049%

      Netherlands -1 +102 = 49.6975%
      Denmark +1 -109 = 51.3025%

      BetOnline odds:
      Draw +350 = 22.2222% = 7.967% EDGE
      Denmark +1 +110 = 47.619% = 7.735% EDGE

      Should you take into consideration the correlation between the draw and +1 (If draw wins, +1 +110 wins) and adjust the structure of your bets (and how would you do this), or treat them as separate markets and bet both full Kelly?
      Yes you should consider the correlation.

      There are 4 unique outcomes, Denmark Win, Draw, Netherlands by exactly 1 , Netherlands By > 1, you need the probabilities (P) for each. You have all but Netherlands by exactly 1 which you can figure out from the other probs.

      Next, calculate the log of your ending bankroll under each unique outcome, the formula is P*LOG(END BANKROLL), end bankroll is just Sum of opening bankroll + $$ W/L under each scenario.

      Sum these amounts and maximize them using excel solver.
      Comment
      • Ganchrow
        SBR Hall of Famer
        • 08-28-05
        • 5011

        #4
        Originally posted by Tomato
        Denmark-Netherlands

        Pinnacle probabilities:
        Netherlands -179 = 62.958%
        Draw +309 = 23.992%
        Denmark +652 = 13.049%

        Netherlands -1 +102 = 49.6975%
        Denmark +1 -109 = 51.3025%

        BetOnline odds:
        Draw +350 = 22.2222% = 7.967% EDGE
        Denmark +1 +110 = 47.619% = 7.735% EDGE

        Should you take into consideration the correlation between the draw and +1 (If draw wins, +1 +110 wins) and adjust the structure of your bets (and how would you do this), or treat them as separate markets and bet both full Kelly?
        Of course you need to take correlation into account.

        Based on Pinnacle these are the underlying probs for the spread bet:


        And the conditional probabilities (i.e., conditioned on the Netherlands NOT winning by exactly 1) for the ML Bet:

        (Note the slight error correction shown in red.)

        So the unconditional probabilities and bet results for each of the four possible relevant outcomes are

        So if we label the bet on the draw as XDr and the bet on Denmark as XD+1 then out full Kelly expected utility function looks like this:
        E[U(XDr, XD+1)] =
        13.0492% * log(1 - XDr + 1.1*XD+1) +
        23.9927% * log(1 + 3.5*XDr + 1.1*XD+1) +
        27.7970% * log(1 - XDr) +
        35.1611% + log(1 - XDr - XD+1)

        Which when maximized wrt XDr and XD+1, yields the following full-Kelly wagers:

        Some years back, I posted the attached spreadsheet to optimize just these sorts of straightforward ML/Spread combinations. It's simplistic and reflects the work of a younger, gentler, sloppier man (), but it appears I wasn't totally inept and it does do the trick. NO warranties, either expressed or implied.

        Note that on the spreadsheet the ML refers to the Draw bet which, perhaps somewhat confusingly is marked as a "Win" if the teams Draw and a loss otherwise.
        Attached Files
        Comment
        • Tomato
          SBR MVP
          • 01-29-09
          • 1251

          #5
          Thank you Pancho villa and Ganchrow.

          Tomato is just starting to bet soccer and implement Kelly with his edge and this is very helpful.
          Comment
          • ojs69
            SBR High Roller
            • 01-31-10
            • 197

            #6
            Good stuff you guys
            Comment
            • roon
              SBR Hustler
              • 05-13-10
              • 93

              #7
              I am new to the mathematical approach to sports wagering. Am I missing something in these calculations? I get

              Netherlands -179 = 64.2%
              Draw +309 = 24.4%
              Denmark +652 = 13.3%

              when I add these percentages, it comes to 101.9, so I would assume this is where I am off and I need to perform a second calculation. Do I take the 64.2 and divide by the 101.9 .642/1.019? And the again with the draw and Denmark? When I do this I get the correct percentages from the first post. Is this to compensate for the vig?

              I would really appreciate any help. I am pretty strong with math and am very intrigued with these concepts. I'm just beginning to put this info into practice. Gonna take awhile though...
              Comment
              • Poogs
                SBR High Roller
                • 04-05-10
                • 116

                #8
                The extra 1.9 is the theoretical hold of the book.
                Comment
                SBR Contests
                Collapse
                Top-Rated US Sportsbooks
                Collapse
                Working...