I've been backtesting intraday strategies, running through various entries with random exits. I've read that 55% is a good number to shoot for for winning percentage on entry, unfortunately I can only get to 50% so far. I can get a higher percentage if I make the entry criteria quite strict, but this greatly reduces the number of trades taken and I'm less inclined to trust numbers stemming from 300 or so trades as compared to several thousand. For those of you who've been through this process, what number of trades do you like to see and with what winning percentage?
winning percentage for entries
- Thread starter Kris
- Start date
The OP is referring to a testing idea on page 172 of LeBeau and Lucas's (book) on futures trading.
L&L wanted to measure the relative "goodness" of entry techniques independent of the exit used. (They also wanted to measure the relative "goodness" of exit techniques, independent of the entry used. The Big Idea was, of course: find the best entry, find the best exit, put em together, Bam. A continuous firehose blast of profits into your account. Bada bing bada boom.)
L&L settled upon a simple test for entries: after entering, hold the trade for a fixed number of bars, N, (you get to choose N), and then exit on the close. Count up the percentage of these "trades" that were profitable, neglecting commission and slippage. The entry positioned you in the right direction to profit from the N-day move. Higher winning percentage is better.
Their idea is, if you were an idiot coin-flipper, randomly entering at a random time in a random direction (long or short), and then holding for N bars, you ought to have a winning percentage of about 50%. Since you want your entry method to be "better than random", you want to see a winning percentage ON THIS TEST higher than 50%. L&L picked an arbitrary measuring stick, 55%, which certainly is higher than 50%. But so is 54% and so is 56% and so are many other arbitrary numbers. In real trading, some trades would be stopped out waaay before the Nth bar and so in real life, the winning percentage will be different than what this test indicates.
I think a fair criticism of this procedure is: it's old. It solves a problem that just doesn't exist any more. LeBeau and Lucas's book is almost 20 years old; since it was published, system testing software has gotten tremendously better. Now, unlike then, you can test zillions of entry methods WITH zillions of exit methods. You don't have to "divide and conquer" the problem by choosing an entry independently of choosing an exit. Computer hardware and software has advanced quite a bit since then.
L&L wanted to measure the relative "goodness" of entry techniques independent of the exit used. (They also wanted to measure the relative "goodness" of exit techniques, independent of the entry used. The Big Idea was, of course: find the best entry, find the best exit, put em together, Bam. A continuous firehose blast of profits into your account. Bada bing bada boom.)
L&L settled upon a simple test for entries: after entering, hold the trade for a fixed number of bars, N, (you get to choose N), and then exit on the close. Count up the percentage of these "trades" that were profitable, neglecting commission and slippage. The entry positioned you in the right direction to profit from the N-day move. Higher winning percentage is better.
Their idea is, if you were an idiot coin-flipper, randomly entering at a random time in a random direction (long or short), and then holding for N bars, you ought to have a winning percentage of about 50%. Since you want your entry method to be "better than random", you want to see a winning percentage ON THIS TEST higher than 50%. L&L picked an arbitrary measuring stick, 55%, which certainly is higher than 50%. But so is 54% and so is 56% and so are many other arbitrary numbers. In real trading, some trades would be stopped out waaay before the Nth bar and so in real life, the winning percentage will be different than what this test indicates.
I think a fair criticism of this procedure is: it's old. It solves a problem that just doesn't exist any more. LeBeau and Lucas's book is almost 20 years old; since it was published, system testing software has gotten tremendously better. Now, unlike then, you can test zillions of entry methods WITH zillions of exit methods. You don't have to "divide and conquer" the problem by choosing an entry independently of choosing an exit. Computer hardware and software has advanced quite a bit since then.
Winning percentage is a measure that beginning traders cling to. It becomes less and less relevant as you develop automated trading. Winning percentage puts an initial gleam in traderâs eyes because of the constant sales hype traders are subjected to. Those who tout âprofitableâ systems throw 60 percent or 70 percent as âgoalâ a âreal traderâ must adhere to. What baloney.
Consistent profits are what an automated trading is all about not winning percentages. An automated strategy I recently built puts out position trades in the daily time interval. It has traded at 49.35% since August 07 with minimal drawdowns.
Getting no big surprises in a smooth ascending equity curve is the objective not percentage profitable. This means if you did a linear regression analysis of your ascending profit line that the deviation from that upward progression is reasonable (that is small drawdowns). If you can keep this up in live trading you will never even look at the winning percentage.
If this is the firs step that you're taking to develop your strategy, then I'm afraid you are doing it backwards.
The first step is to identify a probable tradable system and then evaluate its robustness.
But to answer your question, you want to improve the entry and exit efficiencies in a manner that robustness is not compromised and that its effects are statistically significant as far as improving the overall performance of the system. For instance, going from 50 to 55% entry efficiency improves the overall system results by how much on a risk/adjusted basis?
Read Van Tharp. His analysis of system design based on expectancy is required reading in my opinion.
The percentage of winners is only one variable in the bigger equation of profitability. As an earlier poster pointed out, you can have a 50% win rate and still be profitable. It's more important to have bigger winners than losers.
Heck, famous trend following systems like simple moving average crossovers have a win rate of less than 50% on swing trades. They make money, but they do it by catching and riding bigger trends and taking (many) smaller losses during whipsaw periods.
The percentage of winners is only one variable in the bigger equation of profitability. As an earlier poster pointed out, you can have a 50% win rate and still be profitable. It's more important to have bigger winners than losers.
Heck, famous trend following systems like simple moving average crossovers have a win rate of less than 50% on swing trades. They make money, but they do it by catching and riding bigger trends and taking (many) smaller losses during whipsaw periods.
and by the way...winning percentage is not the most appealing metric to maximize in a mechanical trading system.
Maximization of winning percentage in a short term trading system can come at the expense of lower average profit per trade. Lower average per trade in a short term trading system (particularly intraday) is detrimental as the system is exposed to commission and slippage costs.
Maximization of winning percentage in a short term trading system can come at the expense of lower average profit per trade. Lower average per trade in a short term trading system (particularly intraday) is detrimental as the system is exposed to commission and slippage costs.