Kelly criterion calculated wrong.

[yakir4123]

I wanted to implement kelly criterion in my strategy after seeing this video:
https://www.youtube.com/watch?v=_FuuYSM7yOo&ab_channel=MarcinAnforowicz

Happy to know that jesse support that and also has an example in the docs.
https://docs.jesse.trade/docs/strategies/example-strategies.html#getting-the-size-right

Unfortunate after several experiments and investigatation I found that something is off. The problem was in the method of utils.kelly_criterion

The results are completly different between these formulas to utils.kelly_criterion formula.

[saleh-mir]

I cannot make the unit tests for this to pass. I tried all sorts of help from Chad GPT and similar tools two. I don't really know much about Kelly criterion. Can you help with this please?

def kelly_criterion(win_rate: float, ratio_avg_win: float, ratio_avg_loss: float) -> float:
    return win_rate - ((1 - win_rate) * ratio_avg_loss / ratio_avg_win)

def test_kelly_criterion():
    # Scenario 1: win_rate is 60%, average win is twice the bet amount, and average loss is the bet amount
    assert abs(kelly_criterion(0.6, 2, 1) - (0.6 - 0.4/2)) < 1e-9

    # Scenario 2: Even win rate, average win and average loss are the same
    assert abs(kelly_criterion(0.5, 1, 1)) < 1e-9  # Expected result is 0

    # Scenario 3: win_rate is 70%, average win is three times the bet amount, average loss is twice the bet amount
    assert abs(kelly_criterion(0.7, 3, 2) - (0.7 - 0.6/3)) < 1e-9

    # Scenario 4: win_rate is 40%, average win is half the bet amount, average loss is four times the bet amount
    assert abs(kelly_criterion(0.4, 0.5, 4) - (0.4 - 2.4)) < 1e-9

    # Scenario 5: win_rate is 80%, average win is 1.5 times the bet amount, average loss is 0.5 times the bet amount
    assert abs(kelly_criterion(0.8, 1.5, 0.5) - (0.8 - 0.1/1.5)) < 1e-9