Risk:Reward vs. 2% | Forex Factory

Options

Permalinks
Print Thread

Similar Threads

What's your TP? Fixed reward vs dynamic reward 5 replies

Low Risk, High Reward Trading Experiment 3,318 replies

Rationale behind "Risk Per Trade" and "Risk/Reward Ratio"? 8 replies

Risk Reward Ratio and Account Risk 30 replies

Results for Low Risk,High Reward 11 replies

Risk:Reward vs. 2%

  • #1
  • First Post: Oct 28, 2007 6:12pm Oct 28, 2007 6:12pm
I have a system I have created that seems to be profitable, I understand and am confident in. It backtests VERY well, but, it forward tests even better. The only issue I have left is in Risk. The system uses moving averages, some of which go over 50 4h bars. Because of this the pip value of my stop needs to be quite large, and, given my 2% rule algorithm that sizes my position based on Risk using stoploss. The system has in the past month returned hundreds of pips, but it's slow, with each trade taking a week to weeks.

The problem is that I have expectancy, and high return on each trade and the system doesn't open trades often... so using such a tiny position size returns tiny profits. I'm torn over just tossing the 2% rule but am hesitant to allow a 20% drawdown if all the pairs flip at the same time on me and I get stopped out of every single position concurrently.

If I tighten the stop the back test show dramatically lower profits as spikes seem to trip it up often. Right now the best stop I can use is a trailing stop based on a EMA 50, which is often about 80 pips (depending on the pair.).

What should I be considering next?
  • #2
  • Oct 28, 2007 6:31pm Oct 28, 2007 6:31pm
Hello,

Few questions come to mind:

  1. What is the maximal number of consecutive losses you have experienced when backtesting?
  2. What is the total pip loss in the maximal consecutive losses?
  3. What is the average winning per trade in your system?
  4. Have you taken into account rollover?

If you put an 80 pip stop loss how likely is it to experience 3 or 4 of those in a row?

For risk management I always remind myself one simpe rule - if you lose 50% you now need to gain 100% to recover the loss.

So... may be with your system you can afford 4% or 5% if you find that dropping more than 20% maximal drawdown is very unlikely.

  • #3
  • Edited Oct 30, 2007 12:05am Oct 29, 2007 12:51am | Edited Oct 30, 2007 12:05am
If you know your system's probability of winning p and its win/loss ratio r, then Kelly's Formula f = [p*(r+1)-1]/r tells you what fraction f of your account you should risk on any given trade in order to increase your bankroll at its maximum rate. Of course, this position sizing algorithm only works for strategies that have positive expectancy E = p*w+q*l, where w is the strategy's average win, q is the probability of losing (= 1-p), and l the average loss.

Kelly's criterion is the most aggressive anti-martingale money management scheme. For example, if p=.6 and r=1.5, then Kelly's Formula says to risk f = .33 = 33% of your account on every trade. Obviously, most traders don't have the balls to trade that much money on a single trade. It will inevitably lead to huge drawdowns. Also, the Kelly criterion requires that both the lot size and the account's currency be infinitely divisible, which clearly isn't the case in the real world. The Kelly criterion should therefore be seen as giving the absolute maximum amount of risk per trade.
Domo Arigato, Mr. Roboto!
  • #4
  • Oct 29, 2007 1:13am Oct 29, 2007 1:13am
You should always set your stops far enough away from the entry price so that they are not triggered by the typical intrabar volatility of whatever timeframe you choose to trade. One popular way to measure this volatility is to use the ATR. For example, if you open a long position you might consider placing the stop at entry_price-1.5*ATR. The multiple in front of the ATR value should never be less than 1.
Domo Arigato, Mr. Roboto!
  • #5
  • Oct 29, 2007 3:31am Oct 29, 2007 3:31am
Quoting fourth
I have a system I have created that seems to be profitable, I understand and am confident in. It backtests VERY well, but, it forward tests even better. The only issue I have left is in Risk. The system uses moving averages, some of which go over 50 4h bars. Because of this the pip value of my stop needs to be quite large, and, given my 2% rule algorithm that sizes my position based on Risk using stoploss. The system has in the past month returned hundreds of pips, but it's slow, with each trade taking a week to weeks.

The problem is that I have expectancy, and high return on each trade and the system doesn't open trades often... so using such a tiny position size returns tiny profits. I'm torn over just tossing the 2% rule but am hesitant to allow a 20% drawdown if all the pairs flip at the same time on me and I get stopped out of every single position concurrently.

If I tighten the stop the back test show dramatically lower profits as spikes seem to trip it up often. Right now the best stop I can use is a trailing stop based on a EMA 50, which is often about 80 pips (depending on the pair.).

What should I be considering next?
you arent doing anything wrong fourth - just keep doing what you are doing.
https://linktr.ee/Soultrader
  • #6
  • Oct 29, 2007 3:51am Oct 29, 2007 3:51am
How many trades do you have in backtest? If less than 1000, don't bother - test it on other pairs. Only then check for drowdawn, you will see that risking 2% on each trade is probably too much.
Good luck,
L
  • #7
  • Oct 29, 2007 5:13am Oct 29, 2007 5:13am
Quoting fourth
What should I be considering next?
Hi Fourth

When you talk about a profitable "system", do you mean entry, or both entry and exit? I ask this question because it sounds like, despite testing, you have unresolved issues regarding exit (i.e. stoploss point).

IMHO system performance is not just about pips gained. Expectancy = dollars gained / dollars lost (i.e. risked). Hence it needs to consider both entry and exit.

As a simple example, a 300 pip gain for an 80 pip risk is the same RR as 75 pips gain for 20 pips risk. By scaling down the pos size by a factor of 4, you can trade the larger 300 pip move with exactly the same (absolute) risk to your account, as a 75 pip move. The 300 pip move will take longer to occur, so your rate of return (i.e. ROI relative to time) will be reduced, but both risk $$ (or % of acc bal) and RR remain unchanged.

Also, keep in mind that RR changes dynamically over the course of a trade. For example, if/once the stoploss is advanced to/beyond breakeven point, risk is reduced to zero. Hence true measure of expectancy is not the initial RR but the eventual outcome in terms of dollars gained and dollars lost, totaled across all trades in the test sample.

When you talk about "all the pairs flip at the same time", I assume you're talking about simultaneous trades in different pairs? If so, you need to consider the correlation between the pairs. The closer to zero (as opposed to either highly positive, or highly negative) correlation, the closer the total exposure is to 2%, regardless of the number of trades (pairs) that are open. The closer to +100% correlation, the closer the total exposure is to N x 2%, where N is the number of pairs involved (e.g. 10% for 5 pairs). The closer to –100% correlation ("perfect" hedging), the closer the total exposure is to zero.

Whether you should exceed 2% position size depends on how positive your expectancy is, and its statistical validity. Higher expectancy can potentially be used to justify higher bet size. A Monte Carlo analysis can be very helpful. The Kelly formula gives optimum bet size, but does not take wipeout into consideration.

Re statistical validity, check out the table (headed "Sample Size", "Margin of Error") here:
http://www.tradejuice.com/system-tra...ing-system.htm

For example, let's say your testing shows net wins of $44,000 and net losses of $33,000. Hence profit factor or expectancy is 44 / 33 = 1.33
If, however, your test sample involves only 50 trades, then the table tells us that the statistical error is 14%, i.e. we can say with 95% confidence that the true expectancy lies between 1.33 x (1 – 0.14) and 1.33 x (1 + 0.14), i.e. 1.14 and 1.51
If your average trade time is days or even weeks, then it will take some time to establish a statistically valid test sample size.

If you've established a system with significantly positive expectancy, then you're way ahead of me. Congrats, and good luck.

David
  • #8
  • Oct 29, 2007 7:36am Oct 29, 2007 7:36am
Fourth, if you have such a system and the backtest turns out to be reliable you are already doing very, very well. Consider that if you can gain an average of 2% of your capital every month your are almost making 100% in one year.
Anyway one suggestion could be to evaluate the system with different risk % for every trade. In fact you say that you use a trailing stop based on a EMA. This means that when you start a position your initial stop loss is not always placed at the same pip distance from your entry. In this case I would do something like this:
1) I fix my maximum risk (like 2%) once and for all
2) when I enter the market I evaluate the initial distance of the stop loss in pips. If it is statistically very far I would risk the whole 2%, if it is say 50% closer in comparison with the statistical maximum I would risk only 1%!

Example:
A) Initial stop loss 100 pips away (and supposing that statistically this could be considered very far away) -> I risk 2%
b) Initial stop loss 50 pips away (and supposing that statistically 100 pips is considered very far away) -> I risk only 1%

You can also evaluate non linear relationship between stop loss and risk.
Example: when SL is 100 pips risk is 2% when SL is 50 pips risk is 0.8%.

Summarizing: the closer your stop loss the lower your risk.
Just try it and backtest.
.Ubeee.
  • #9
  • Oct 29, 2007 1:20pm Oct 29, 2007 1:20pm
Quoting ubeee
Summarizing: the closer your stop loss the lower your risk.
Hi Ubeee

Interesting ideas. I would like to understand more about their underlying logic.

IMHO the risk % (position size) should reflect the expectancy on a given trade, relative to other trades. For example, if one is going to double the position size on a given trade (e.g. from 1% to 2%), then one needs to have double the positive expectancy to justify the additional risk. If expectancy per trade can't be calculated accurately (most likely the case), then I believe that all setups should be considered to have equal potential, and hence position size should be consistent. Otherwise one faces the situation where, for example, one losing trade (e.g. where 2% of the account was risked) requires two winning trades (at 1%) to compensate.

Also I don't understand why one would want to place the stoploss closer, and then only for the purpose of lowering the position size. Presumably the underlying rationale is that if the stoploss is too tight, there is greater probability of being stopped out, and therefore poorer expectancy justifies less risk. But why not simply improve the odds by setting a wider stoploss in the first place?

IMHO the stoploss (in fact, all aspects of exit), are better geared around technical considerations (e.g. S/R, Fibos, trendlines, pivots) in order to enhance their bearing on profitability. Simply picking a round number like 50 or 100 pips is less likely to synchronize with price behavior. At the very least, the stoploss should be pre-set to reflect the point at which the trader believes the trade will have "failed", according to his plan. Once this point has been established, then position size can be calculated, which (for my reasons above) I believe should be consistent with every trade.

In the model that I follow, position size operates completely independently of expectancy. The latter is derived from entry and exit (see my earlier post).

Anyway, that's my approach, for whatever it's worth.

David
  • #10
  • Oct 29, 2007 4:13pm Oct 29, 2007 4:13pm
Quoting hanover
Interesting ideas. I would like to understand more about their underlying logic.
Well, reading the rest of your post you already understood. In fact you wrote:

Quoting hanover
Presumably the underlying rationale is that if the stoploss is too tight, there is greater probability of being stopped out, and therefore poorer expectancy justifies less risk.
And that's it. The problem is noise, whipsaw. The closer the stop loss the bigger the chance your stop loss will be hit.

Quoting hanover
But why not simply improve the odds by setting a wider stoploss in the first place?
Well, you can do that, if you prefer. But, warning, by doing this you are completely changing the trading system. You are writing another one, something completely different. Since Fourth wrote that he has already a trading system that suits him, and he just wants to try to improve it, he can do what I suggested without significantly altering the system. This means that the backtesting he already made remains almost all still valid, he will just note small drawdowns and a little bit more gains. If, instead, you completely change the way the system computes trail stops, everything dramatically changes. All the premises of the system, the reasonings, the goals could be compromised. You could not use anymore the EMAs, for example, for trailing stops. Yes, everything can be done, and researching alternatives is a must in every trading system development, but I preferred to give him a suggestion that already fits his system, that doesn't require radical changes or the return to the drawing board.
Of course by doing this he still uses the EMA for stop loss and this means he hasn't any degree of freedom on choosing the initial stop loss distance, hence he can only tweak the risk.

Also, with this approach there is another advantage. Since by definition he will risk his maximum (2%) only when his stop loss could be statistically considered very far away (meaning: a rare event), he will find that more often than not he will risk way less than 2% per trade. This gives him the chance to have more trades running on different pairs at the same time without fearing the worst case scenario of a 20% loss of capital if 10 of his concurrent trades went bad, that was his other concern.

Quoting hanover
Also I don't understand why one would want to place the stoploss closer, and then only for the purpose of lowering the position size.
No, no, no, absolutely no. I would never, never, never suggest this. It's just that he already decided to use EMAs for trailing stop losses. But by doing this he hasn't any degree of freedom on choosing how far away his initial stop loss will be, because it will depend on where the EMA will be at the moment he starts the trades. He only has two options:
a) decide to ignore those signals that come when the EMA is too close to the price
b) lower his risk as the closer the EMA is at the time of the signal.

I find it's more distorting of the system to ignore its signals than just tweak the risk.

I know all this seems very counter-intuitive. Well, I find a lot of things in trading are extremely counter-intuitive.

Quoting hanover
In the model that I follow, position size operates completely independently of expectancy. The latter is derived from entry and exit (see my earlier post).
But why? If you know that the expectancy is different, then why you risk the same? I prefer to risk less when the expectancy is lower. Seems natural to me.
.Ubeee.
  • #11
  • Oct 29, 2007 5:27pm Oct 29, 2007 5:27pm
That's what I luuuuuuve about forums.

One guy says 2% risk is too much, do 1000 trades backtest or forget it.

The other says that if you've made 10 trades and 6 were profitable, then you can risk 33% of your account.
  • #12
  • Oct 29, 2007 6:41pm Oct 29, 2007 6:41pm
Quoting blueruby
That's what I luuuuuuve about forums.

One guy says 2% risk is too much, do 1000 trades backtest or forget it.

The other says that if you've made 10 trades and 6 were profitable, then you can risk 33% of your account.
I think you're misunderstanding the posts.

One guys was saying to risk a constant 2% per trade. He also commented on a good way to validate a strategy.

The other guy was applying Kelly's formula as an upper limit to what should be risked. Another post mentioned using confidence intervals for establishing how "confident" one could be in a given result. This is the same premise behind why a survey of 1500 people at random can effectively reflect the data for the population of the US.

These are very useful pieces of information for a trader, as they are based upon mathematical theorems as opposed to a trader just looking at some results and going "WOW! I ARE GUNNA B DA RICH00RZ!!!!11!!!".

But to address the confidence intervals, not only do you need to have a certain sample size, but it also must be done over a sufficient time period to capture the market. There is a big difference between 50 trades in a month and 50 trades spread out over 5 years.

To efficiently back test a system you should go back as far in history as you can, and either let the system run over that time period or randomly select times for each year to get an effective market model. Depending on how confident you want to be about your results (I'd probably choose no less than 98%), you can figure out how many trades you need.

The market is only Gaussian over the long haul. Short term distributions can deviate (quite significantly) from the Gaussian. This should also be keep in mind.

~X~
  • #13
  • Oct 29, 2007 6:41pm Oct 29, 2007 6:41pm
Quoting ubeee
......But, warning, by doing this you are completely changing the trading system. You are writing another one, something completely different. Since Fourth wrote that he has already a trading system that suits him, and he just wants to try to improve it, he can do what I suggested without significantly altering the system......
OK, thanks, I think I see why you are suggesting this now, i.e. to maintain his existing entry and exit rules (since he is satisfied with his system).

Quoting ubeee
But why? If you know that the expectancy is different, then why you risk the same? I prefer to risk less when the expectancy is lower. Seems natural to me.
I agree totally, as I said at the start of the second paragraph of my previous post. The key is being able to discern which setups offer higher probability of profit than others. If this can not be determined, then I don't see any reason to deviate from risking the same % per trade.

David
  • #14
  • Oct 30, 2007 12:19am Oct 30, 2007 12:19am
In post #7 hanover mentions how one can increase upside potential while maintaining total risk unchanged by simultaneously trading uncorrelated currency pairs. Here's a thread (started by hanover himself) that delves into this crucial topic in more detail. It should be required reading for all.
Domo Arigato, Mr. Roboto!
  • #15
  • Oct 30, 2007 3:34am Oct 30, 2007 3:34am
Quoting Xyrus
I think you're misunderstanding the posts.
I think you missed my point.

The guy is apparently a new trader. He's got 11 posts. He doesn't know how big to trade starting off. He's looking for some basic guidelines. Posters come in and give him a lot of good data, but come on, folks!

fourth: if you haven't yet traded with real money, I'd suggest trading a penny a pip until you have 5 or 10 trades done. Period.

If those result in overall profit, then try a dime a pip for a week or a month or 6 months, and work your way up from there.

I don't care if your account is $500 or $500,000, there's no hurry here. The market will still be there when you're really ready. No point in giving away money even if you've got it to burn.

You've backtested, you've forward tested, you've done your homework. Start trading.

Once you know your system works in real time, with real money, and that you can stick to it, then there are plenty of threads on money management, or position sizing, on FF.

Don't use Kellys formula. Unless you are a very rare exception amongst the best of the best traders, you will blow your account in a very small number of trades. Remember it takes 100% gain to get back a 50% loss. The psychological effects of a 30 or 50% loss can't be counted.
  • #16
  • Last Post: Oct 31, 2007 3:39am Oct 31, 2007 3:39am
Thanks so much guys! Thats a lot to digest
0 traders viewing now