Discussion in 'Risk Management' started by Ghost of Cutten, May 16, 2011.
75% win rate, 1.5-1 payoff (i.e. you make 1500 for each 1000 you risk)?
Assuming a) the stats continue to be representative of the strategy's performance, b) the stats mean you are using a fixed target and stop, c) that your 75% winners were distributed more or less evenly throughout your sample (i.e. and so were not bunched all at the beginning), and d) that you are trying to maximise income over time rather than generate returns with a low volatility month to month ...
... then I would bet the bank on this trade (unless you have something else even better!).
Remember, 75% win rate means a resonable probability of needing to survive a run of 5 or 6 losers in a row; bear this in mind when sizing.
15% of the amount you can afford to lose per trade. After win, reinvest all winnings in next trade. And then rebase back to 15%. thats what I do.
This is assuming:
your strategy is scalable.
you are not falling into trap as hedge fund computers may detect same pattern as you at the same time or a bit after, because there is suddenly enough sample availabble to detect.
helps if there is also fundamental input....
edit: my numbers win ratio etc are slightly different. But risk adjusted as suggested.
It's a spread play between two related markets. I was thinking of risking 5% and wondering if it was too big. My drawdown tolerance is in the 20-25% range.
At 75% win rate, you have something like 2% - 9% chance of seeing a run of 5 - 6 consecutive losers in a 100-trade period... risking 5% per trade puts you 25% - 30% down after 5 - 6 consecutive losers.
On the other hand, if you are wrong about the 75% win rate and it is actually more like say 60%, then you have a reasonable chance (2% - 5%) of seeing 8 - 9 losers in a row during a 100-trade period ... that would be a 40% - 45% drawdown.
Conclusion? At 5% risk per trade, there is a reasonable chance your metal may be tested ...
With the numbers you've provided, the 95% confidence level is a drawdown of 23% per 100 trades. If you went to 3% per-trade the 95% confidence level drops to 14%. The expected maximum losing streak at 99% confidence is only 3.32 for 100 trades. At 1,000 trades it grows to about 5.
i look at it differently.
If your max drawdown in 7 net consecutive losses, say, then winning first trade will increase your skin to 8.5 losses. From random generator, for win 70% and win/loss 1.5:1 I found there is only 4 occasions out of 1000 where I would blow up starting witn 7 bullets.
One should assume that say after 30 attempts you will be up with large confidence.
Trick is: are the odds really 0.75 and winloss 1.5:1 !?!??! Or you just start trading this thing after streak of wins which is invariably followed by string of losses.
Would you mind posting the math for calculating those confidence levels please? Or if you have a link to something that gives the formula.
Could you post the math for working that out please? Cheers.
Here's the link for the streaks:
The math is simply taking the loss rate (.25 and doing power math times the number of trades to reach certainty 1.0)
(.25^3.32) * 100 = 1
The drawdowns were calculated using a Monte-Carlo program I developed. There were 100,000 runs of 100 trades in each run using the info. supplied. The assumptions are the trades are independent, random, and 0 std. dev., from the win/loss ratio described. Naturally, trades will not all be either a 1500 winner or 1000 loser so it's best to use a lower bet size to absorb the variances. However, Std. dev. of trades is not a major determinant of drawdowns.
Separate names with a comma.