What does PF=1.075 tell us?
- Thread starter trd
- Start date
You're confused, because in this discussion of Profit Factor, noone bothered to state the formula. % Wins is a variable of this simple equation. BTW, even if you have a system or method with a PF > 3, you obviously have to consider frequency of trades as well. For instance, you won't be making a living with a system that only daytrades ONCE on a 5M chart every two months.
Profit Factor = (PW * AW) / (PL * AL)
â¢PW = Probability of a trade being a winning trade
â¢AW = Average win size
â¢PL = Probability of a trade being a losing trade
â¢AL = Average loss size
edit: Think of % Wins as sort of a smoothing variable. You can have tremendous results with only 35%, but as it goes up you get a smoother equity curve.
Thanks for the explanation. When I do the math on a system I am currently tracking I get very similar results for your formula vs gross profit/gross loss. I understand what your saying about % wins. My thinking was just flat out wrong as far as % wins in relation to the PF.
I just assumed the higher the % profit the better the system in relation to the PF. This is just not the case. I appreciate you taking the time to comment. Thx! Jim PS maybe you could take a look at this thread I started and comment on a few questions in the first post or two. Thx again for your time! http://elitetrader.com/vb/showthread.php?s=&threadid=168540
WP is just a number. It is mostly based on a trader's money management, rather than its profitability. If you take small losses you generally have low WP. If you take large losses, you generaly have a higher WP. But this has nothing to do with a system's value.
Profit Factor tells you how much you win compared to hos much you lose. THAT has value, although you really need to consider things like max DD, total# of trades, length that method is in service (3 years means a lot more than a couple months), sharpe/sortino, and a few other stats. They tell you how successful the system is, rather than relying on your intuition or opinion.
One thing I would like to point out is that as soon as someone says that this or that is unimportant or can't be used, I can almost guarantee you that someone is in fact using it successfully. There are too many blanket statements IMO. There is more than one way to do just about anything.
Win % is a very important consideration for me and is not meaningless if you know how to apply it.
For example, if I develop two systems, one has a PF of 1.2 and the other has a PF of 1.1, which one should I use? If we don't consider Win %, it might appear that the PF 1.2 would be the choice. Now, if I added the following info: For system PF 1.2 you risk one to make five and for system PF 1.1 you risk one to make one, now which system is your choice?
For me, I will choose system PF 1.1. "Kelly" gives me that answer. For PF 1.2 a "safe" amount to invest (wager) would be 20%/5 = 4%. For PF 1.1 a "safe" amount is 10%/1 = 10%. Therefore, I would only make 4%*.2 or .8% on PF 1.2 but 10%*.1 or 1% with PF 1.1. Only if I wanted to increase my level of risk on PF 1.2 would I be able to match the higher profitabilty of PF 1.1. In addition, you will be able to multiply your bankroll faster and more consistently with PF 1.1.
Joe.
Win % is a very important consideration for me and is not meaningless if you know how to apply it.
For example, if I develop two systems, one has a PF of 1.2 and the other has a PF of 1.1, which one should I use? If we don't consider Win %, it might appear that the PF 1.2 would be the choice. Now, if I added the following info: For system PF 1.2 you risk one to make five and for system PF 1.1 you risk one to make one, now which system is your choice?
For me, I will choose system PF 1.1. "Kelly" gives me that answer. For PF 1.2 a "safe" amount to invest (wager) would be 20%/5 = 4%. For PF 1.1 a "safe" amount is 10%/1 = 10%. Therefore, I would only make 4%*.2 or .8% on PF 1.2 but 10%*.1 or 1% with PF 1.1. Only if I wanted to increase my level of risk on PF 1.2 would I be able to match the higher profitabilty of PF 1.1. In addition, you will be able to multiply your bankroll faster and more consistently with PF 1.1.
Joe.
Off topic...
But a PF with 1.00001 and 10.0+ doesn't help if the market changes and the system stops working... All the talk about backtesting measurement (It's important in its own way) is only 1/2 the story. All the Tradestation, WealthLab, or other kinds of system performance reports have 0 (zero) value when it comes to predicting("probabilistically") whether it will make money.
The question you should be asking is, "My model generated a PF of 1.075 up until now. What is the probability of the PF performing in the norm for the next year(month, week, day)"... MC VaR is one way to deal with it but it gets really tricky with using the same profile distribution, skewing the results against real market changes. (Of course, I use it... but because it centers itself to equity curve analysis, I use MC VaR for generating Risk Management schemes.) Throwing random data generated by Brownian doesn't help much because there is no market specific tendency unless you tweak it, but that's much too technical for a retail trader...
The bigger problem is... there is no single formula, it all depends on the model you're testing. (eg. Short term scalping = High Sharpe... Trend-following = High RRR...) Model development doesn't stop at signals and position sizing. It should really go into testing for risk management, technological implementation, etc. etc.
I've been caught up with it for a long time but the systematic traders need to start focusing from a different standpoint. More so, I blame the software vendors for being way too incompetent with how backtesting is done, and unfortunately they are the ones who set the standards for model assessment. Seriously... all these software vendors (I can imagine Tickzoom, TradesStudio people and others joining in saying, "You can do what you mention with our software" but they have no clue) and your retail brokers are more to blame for not providing regular / retail traders for the appropriate access to "equal" trading.
(For a truly valuable software package to exist, the software developers need to start testing whether their tests are actually significant enough to help the user make money. But they're too stupid to understand so we end up writing our own stuff...
)
Good luck trading all.
But a PF with 1.00001 and 10.0+ doesn't help if the market changes and the system stops working... All the talk about backtesting measurement (It's important in its own way) is only 1/2 the story. All the Tradestation, WealthLab, or other kinds of system performance reports have 0 (zero) value when it comes to predicting("probabilistically") whether it will make money.
The question you should be asking is, "My model generated a PF of 1.075 up until now. What is the probability of the PF performing in the norm for the next year(month, week, day)"... MC VaR is one way to deal with it but it gets really tricky with using the same profile distribution, skewing the results against real market changes. (Of course, I use it... but because it centers itself to equity curve analysis, I use MC VaR for generating Risk Management schemes.) Throwing random data generated by Brownian doesn't help much because there is no market specific tendency unless you tweak it, but that's much too technical for a retail trader...
The bigger problem is... there is no single formula, it all depends on the model you're testing. (eg. Short term scalping = High Sharpe... Trend-following = High RRR...) Model development doesn't stop at signals and position sizing. It should really go into testing for risk management, technological implementation, etc. etc.
I've been caught up with it for a long time but the systematic traders need to start focusing from a different standpoint. More so, I blame the software vendors for being way too incompetent with how backtesting is done, and unfortunately they are the ones who set the standards for model assessment. Seriously... all these software vendors (I can imagine Tickzoom, TradesStudio people and others joining in saying, "You can do what you mention with our software" but they have no clue) and your retail brokers are more to blame for not providing regular / retail traders for the appropriate access to "equal" trading.
(For a truly valuable software package to exist, the software developers need to start testing whether their tests are actually significant enough to help the user make money. But they're too stupid to understand so we end up writing our own stuff...
)Good luck trading all.
+1. Backtesting is mostly good for predicting the past. I have seen so many people (ET is just one place) where newbies run around with excitement, because they finally have a successful backtest. Then they walk forward a week in the market, and think their "ship has come in." Unfortunately, it sank at the dock.
no, you're the one who's confused...
profit factor = profit/loss
what you posted is expectancy.
Mr. Webster, you don't make money through definitions. You make money through an understanding of the numbers. I've seen the above called at least 3 different names and I've seen at least 3 equations for expectancy. They all basically come down to the same thing.........should you bother to trade the system/method or not.
Through simple plug & chug he can see W% is not as important as maybe he thought it was. And, again frequency of trades is important.