Credit Spreads with the Kelly Criterion

[Alexpung]

You can do some programming and Monte Carlo it. Don't even need the equation.

 

[MKTrader]

From his post:
95% winners (so 19 out of 20 win)
Average winner = $20
Average loser = $5,000

Horrible expectancy. Unless he made a typo, I think it can be dismissed right there, regardless if he uses Kelly or any other money management system? Martingale might be fun with those $5,000 losses, though...

I'm not sure if anyone has really let this sink in...I hope at least the TS has. He's trying to make $20/trade (which could easily be less with unfavorable bid/ask in options) and will lose $4,650 on every 20 trades on average. Money management (Kelly or otherwise) is a moot point.

 

[dcwriter2]

Ok, maybe I am plugging the data in wrong, but with the call half of an iron condor way out of the money, say, with a delta on the short of about -97 (and the long wing a little further out), and a 97% win rate, risking say 5,000K to make $20, how would I do the Kelly? I can't believe there is no edge with a delta like that.

 

[taowave]

How wide is your spread,i.e percent of spot??

Your protective wing is "a little further out",but you still have a max loss of 5,000 while taking in 20 bucks??

makes no sense

 

[gaussian]

Ok, maybe I am plugging the data in wrong, but with the call half of an iron condor way out of the money, say, with a delta on the short of about -97 (and the long wing a little further out), and a 97% win rate, risking say 5,000K to make $20, how would I do the Kelly? I can't believe there is no edge with a delta like that.

Kelly doesnt tell you the edge it tells you the bankroll percentage you should bet for mathematically optimizing your expected return.

You have no edge in this trade.

import random
from statistics import mean

random.seed()


def trade(p):
    """
    Flips a biased coin with probability of success p. With p in (0, 1]
    """
    if random.random() <= p:
        return True

    return False

def simulate():
    """
    Simulates 100 trades of the worst system ever.
    """
    bankroll = 10000

    for x in range(0, 100):
        if trade(0.95):
            bankroll += 20
        else:
            bankroll -= 5000

    return bankroll
       

def main():
    # 10000 Monte Carlo simulations of the worst system ever
    # excluding any actual costs.

    bankrolls = []
    for x in range(0, 10000):
      bankrolls.append(simulate())

    print(f"Average Bankroll: {mean(bankrolls)}")

if "__name__" == "__main__":
    main()

Here is a dumb python script that I took 5 seconds to write to monte carlo simulate your strategy using 10000 repeated runs of 100 trades. It never turns a profit because your expected loss greatly exceeds your possible gains.


This makes sense mathematically too:

E[Trade] = P(gain) * gain_dollars + P(loss) * loss_dollars
E[Trade] = 0.95 * 20 + 0.05 * (-5000)
E[Trade] = -231

So you are expected to lose 231 dollars per trade assuming you had a large enough bank roll to run these trades as expected.

 

[dcwriter2]

Tks. Makes sense.