Reverse Strategy Ideas? | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Reverse Martingale strategy 101 replies

Entering a new position - reverse strategy 15 replies

SBOR II - Session Breakout and Reverse strategy (universal) 34 replies

EURUSD Session Breakout and Reverse Strategy 50 replies

Need help coding reverse strategy of this EA 0 replies

Attachments: Reverse Strategy Ideas?
Exit Attachments

Reverse Strategy Ideas?

  • #1
  • First Post: Jan 22, 2009 8:31am Jan 22, 2009 8:31am
Hello All, I just have some thoughts that I would like to put forth and my logic behind them and the results from testing them that I would like to share that are currently baffling me to no end and would like to hear you all's thoughts on why this is or perhaps how I can solve this problem and determine if its even worth pursuing. I'm a very experienced programmer so this isn't a question of my skill or ability just merely am not understanding why this is not working when every piece of logic says that it should. So heres the problem...

I'm sure for those of us that have coded many EAs more often than not they blow the account pretty quickly the first couple of runs and if not most of the EAs perform similarly even after months of coding. I understand this is largely a function of the back test quality in mt4, but even with 90% quality from your brokers feed should give you for the most part consistent results whether it be winning or losing. So here is what I propose

Given a Strategy that merely enters the market in a direction based on a set of entry rules. It has SL: 10pips, TP: 5pips and the strategy just lets the trade run till either of these levels are hit

The Results:
Trades: 679. Wins: 254 (37%) Losses: 425 ( 63%)

Obviously this strategy would blow an account pretty quick. So my idea is this. Why not just do the opposite of everything in the strategy, if we get a long signal, then we short. And instead of 10pip SL we now use a 5pip stop and instead of 5pip TP we use a 10pip TP so now all those previous stop outs are cut in half pip wise. And all those previous wins are larger and occur more often since we are entering in the opposite direction. So I coded this. Exact same entry rules, exact same timeframe and dates, exact same data and backtest quality even checked that it was entering at exact same prices and times etc.

The Results:
Trades: 679. Wins: 31 (4.5%) Losses : 648 (95.5%)

This does not make any sense at all to me. I know 90% quality is not 100% so its hard to prove this exactly, but those are pretty drastic differences When technically all the previous losses should now be wins and vice versa thus our new win % should have been (63%) and this should be a very profitable strategy. Any thoughts or ideas would be greatly appreciated. Thanks!
  • #2
  • Jan 22, 2009 8:41am Jan 22, 2009 8:41am
I've come across similar "problems".


The main reason is that the market is in a particular state. It just happens to work for that particular setup. Was that test EU during one of its longer trending periods?
  • #3
  • Jan 22, 2009 8:49am Jan 22, 2009 8:49am
Yep it was EU, Jan 01 2009 - Today. I've done it over several periods under different market conditions and the same results. I understand the market may be in a particular state at that point of the set up. But still logically by doing the reverse shouldn't the previous losses now be wins? If everything else is a constant (dates, market conditions in those dates, entry price etc)
I understand this may not work on a different period. But in an isolated test with everything remaining constant except for changing the direction one enters and switching the SL and TP then the results should be exact opposite. Do you know why this is not true?
  • #4
  • Jan 22, 2009 9:44am Jan 22, 2009 9:44am
I have no means to refute it if that's what you're saying. Your theory as I see it is correct.

The point where this theory falls apart is what happens going forward.
  • #5
  • Jan 22, 2009 9:54am Jan 22, 2009 9:54am
You might also have a scenario that has an element of curve-fitting to it. Its hard to say without knowing more about your system. 679 trades may (or may not) be a lot of trades (over time, I mean) depending on your strategy. It doesnt strike me as a particularly high number.
  • #6
  • Jan 22, 2009 9:57am Jan 22, 2009 9:57am
679 trades in 22 days? (Less weekends, 15).?! What then makes a good sample size? or sample density for that matter.
  • #7
  • Edited 10:06am Jan 22, 2009 10:00am | Edited 10:06am
In response to Post 4:

Right, and I understand that but am currently only concerned with why it does not work when used over the same period, not forward testing. For that it should do exactly as expected but it doesn't. Do you have any idea why this wouldn't work just using the same dates and data? Have you tried it before? Or perhaps you would know of a better way to backtest, given all the EAs you've coded? I currently can get 90-92% backtest from IBFX data center. And of course the 90% 1m data from alpari. Perhaps the theory I've proposed is correct just a matter of the modelling quality even though it is 90%+
  • #8
  • Jan 22, 2009 10:05am Jan 22, 2009 10:05am
actually your result makes perfect sense. have you account for spread in your stop loss? if not... this is easily explained.

Stop Loss

assuming spread is fixed at 2 pips throughout. before reversing. your stop loss was 10 pips which means price needs to move 8 pips against you to hit.

after you reversed, the stop loss became 5 pips, meaning price only need to move 3 pips against you for you to get stop out.

Take Profit

before reversing. TP was set at 5 pips. after accounting for spread, price needs to move 7 pips in your favor to hit TP.

after reversing, TP is now 10 pips. that means 12 pips is required to move in your favor in order to hit TP.

summary

after reversing, you've increased your chances of hitting stop with a lower stoploss. and decreased your chances of hitting TP with a higher TP target.
Get the picture?

regard,
Zen
Programming for a better future.
  • #9
  • Jan 22, 2009 10:06am Jan 22, 2009 10:06am
You can't reverse something that is not symmetric. Period.
  • #10
  • Jan 22, 2009 10:09am Jan 22, 2009 10:09am
Quoting Ronald Raygun
679 trades in 22 days? (Less weekends, 15).?! What then makes a good sample size? or sample density for that matter.
Sample sizes should be measured in time, not necessarily trades. Market conditions (and volatility) go beyond 22 days
  • #11
  • Jan 22, 2009 10:53am Jan 22, 2009 10:53am
Quoting LongToBeFree
Sample sizes should be measured in time, not necessarily trades. Market conditions (and volatility) go beyond 22 days
The sample size and time is really irrelevant here. I am merely concerned about reproducing the exact opposite of a given set of results keeping everything about a system constant except entry direction and switching SL, TP. One should be able to do this given a timeframe and set of data whether it be with a system that makes a 100 trades or 1000 no matter what the volatility is.

I am in class right now so I will get on in about an hour to respond to the other posts and post the strategy that I am using. However again, the strategy is equally irrelevant. As long as the entry rules are the same it does not matter, we are just concerned about producing the exact opposite.
  • #12
  • Jan 22, 2009 12:13pm Jan 22, 2009 12:13pm
It's possible to get up to 99% modeling quality.

You need an older build of MT4 (208 and earlier I believe), and tick data. I have both, but nowhere near enough tick data to make decent backtests. And tack on the problems I have in keeping clean data, and gaps in the data, it's very painful to work with.
  • #13
  • Jan 22, 2009 12:24pm Jan 22, 2009 12:24pm
Quote
This does not make any sense at all to me.
Makes perfect sense.

Quote
Trades: 679. Wins: 254 (37%) Losses: 425 ( 63%)
Let's assume a neutral distribution.

Now if you double the TP to match the SL you'd also cut your wins in half, ie:

679 trades. 127 wins, 552 losses.

Now if you further half the SL, that means the SL is twice is likely to get
hit, cutting your wins in half again.

679 trades. 63 wins. 616 losses.

At that point the number of wins is no longer statistically significant, and variance over time could be +/- 20% of the overall.

What I'm getting at is this: Your rules are more or less random and have no edge. All you're doing is messing with the TP and SL and getting results that more or less match a random, neutral distribution.

Any wins based on these rules are due to luck. But over enough time your account will trend to 0 if you trade it.

Quote
But still logically by doing the reverse shouldn't the previous losses now be wins?
If you had kept the TP and SL the same, it probably would've been fairly close (plus/minus sample variance if you change the dates). But you didn't. Even if you had, this method would not be profitable. Even if 254 wins to 425 losses is the mean, and you flip them to get 425 wins and 254 losses, that's still less than 2:1, so you won't break even on the .50 ROI.

This is not a spread issue, altho spread would certainly come into play when factoring in profit. It's a simple statistical problem. It's what happens when you don't have an edge.

Edit:
Quote
Sample sizes should be measured in time, not necessarily trades. Market conditions (and volatility) go beyond 22 days

Should be based on both. You need a wide assortment of trades, but you also need to experience a lot of different market conditions.
  • #14
  • Jan 22, 2009 1:48pm Jan 22, 2009 1:48pm
Quoting forexarmy
Hello All, I just have some thoughts that I would like to put forth and my logic behind them and the results from testing them that I would like to share that are currently baffling me to no end and would like to hear you all's thoughts on why this is or perhaps how I can solve this problem and determine if its even worth pursuing. I'm a very experienced programmer so this isn't a question of my skill or ability just merely am not understanding why this is not working when every piece of logic says that it should. So heres the problem...

I'm sure...

it's because the majority of losses come from the little fake outs that happen during consolidation periods or whatever. you do get some wins though because it will call the trades right sometimes. but now, since you are going the opposite way, all the little fakeouts remain, but since you are now going against your original winners, you now lose those as well. it's like losing all the fakeouts and the trades you would have one the first way. and by the way, those little fakeouts usually lose no matter which way you trade them. it's like entering the market and exiting at the same time. here's a pic to explain what im talking about. this is using a basic MA cross strategy:

http://i203.photobucket.com/albums/a...example111.jpg

there are five wins and ten losses (5/10) so basically, 33% wins, 66% losses. now let's reverse the trades:

http://i203.photobucket.com/albums/a...example222.jpg

now we have two wins and thirteen losses (2/13) or 13.33% wins and 86.66% losses. those little fakes are the things that get you. the only way to have a profitable system is to filter those out.

i hope this helps to explain it a bit


oh, and this is all based purely on the number of wins and losses, no SLs or TPs. rabid's post talks more about those.
  • #15
  • Jan 22, 2009 2:10pm Jan 22, 2009 2:10pm
Quoting forexarmy
The sample size and time is really irrelevant here. I am merely concerned about reproducing the exact opposite of a given set of results keeping everything about a system constant except entry direction and switching SL, TP. One should be able to do this given a timeframe and set of data whether it be with a system that makes a 100 trades or 1000 no matter what the volatility is.
Youre not thinking this through.

The time period does matter because it drives volatility. Here's an example:

Take a full candle (close is lower than open) with decent sized tails on both ends. You say to yourself, "market stunk today, ended down'. Does that mean that everyone lost money that day? Of course not. You know the result of the candle, but how was it formed?

  1. Did the market have a very early rally and stay at its high for 80% of the day only to nosedive at the end?
  2. Did it start down but a mid-day rally created the high only to have an epic fight by the bears bring it back down?
  3. Was it up & down all day and it just so happened to end down - given another hour it would have been up?


Now, this is just one candle, one day. So, how many days of differing volatility to you want to expose your EA to?

The answer is: A LOT OF THEM!

So, in the short time period you posted, its possible that you have inadvertantly done some curve-fitting of your own. Maybe you picked the series of candles that favored quick down movements and slow upwards movements (stopping you out more). Or maybe you didn't. Only way to find out is to keep on testing.

  • #16
  • Jan 22, 2009 4:33pm Jan 22, 2009 4:33pm
Wow lots of great discussion going on in here! Thanks for all the great replies guys! Unfortunately though I think my intentions are still being misinterpretted.

I'm not worried about how this ea will perform forward testing or under every single market condition or over 10 years worth of data or on every time frame or all the things you all have suggested. All I am trying to do is produce the exact opposite results given a fixed time interval. That time interval doesn't matter if its one day or 2 weeks or 10 years. If my EA makes 10 trades in 1 month, great. I don't care if 90% of them fail. I just want to figure out a systematic way to reverse whatever strategy was used whether it be random entry or a strategy with a 'edge' even if it happened to not be profitable. By reversing I mean exactly this..

Original Trade:
Long TP:5, SL:10
The trade hit the SL and lost.

So what does this mean? This means that throughout the course of the trade my TP was never hit but yet my stop loss was. So therefor by reversing the direction of the trade I will enter at the exact same level, price will move on exactly as it did and now it will hit my TP which was previously the SL so now that trade is a winner. Nothing complex going on here.

Many of you may say this is stupid and you need to account for all market conditions or whatever but thats not the point, its the concept at hand here. What if we have a system backtested over 5 years with thousands of trades and ran across all market conditions with what we believe to be an edge but when we test it the results are 80% losers, 20% winners.

Now we can apply the same concepts demonstrated above in that single trade and apply it to the whole system effectively so we now have 80% wins and 20% losers. over the exact same time interval and settings.

Again many of you may say this doesn't matter cause its not forward testing and somehow it didn't account for all market conditons. The point by being able to do this is perhaps to reveal some things about our systems that fail, and open us up to a new edge just by reversing the trades. Maybe the strategy was a ranging strategy playing selling tops and buying bottoms, the system was backtested over 2 years and failed. But by reversing the trades and getting profitable results you now realize that hey this EA is doing a great job of recognizing set ups, but they are meant for a breakout strategy instead etc.

Take a look at the 2 charts I've posted and you'll see I got this method of reversing to work today. My previous % loss is now the % win exactly. and you can see the trades all occur at the same level of their previous counterparts and the results have all been flipped exactly as well.

Thanks guys, keeps the ideas coming. The charts posted show that this can be done. And before any of you ridicule the R:R here with TP:4 and SL:20 I want to stress that again thats not the concern here, it may not be smart, yes I agree. But we are looking for a systematic method to apply to any system that fails to reverse it. It just so happened with this particular EA I was testing it on thats how the levels turned out to be.
Attached Image(s) (click to enlarge)
Click to Enlarge Name: regeatradesf.jpg Size: 88 KB
Click to Enlarge Name: reversetradesf.jpg Size: 98 KB
  • #17
  • Jan 23, 2009 1:07am Jan 23, 2009 1:07am
Quote
So what does this mean? This means that throughout the course of the trade my TP was never hit but yet my stop loss was. So therefor by reversing the direction of the trade I will enter at the exact same level, price will move on exactly as it did and now it will hit my TP which was previously the SL so now that trade is a winner. Nothing complex going on here.

Unless it goes 6 pips in one direction, then 11 back.

If you want to "reverse" the system rules you have to keep the MM, SL and profit targets. You can't cut them in half, if you do, you change the rules and the outcome.

Quote
Now we can apply the same concepts demonstrated above in that single trade and apply it to the whole system effectively so we now have 80% wins and 20% losers. over the exact same time interval and settings.

As was already said, you can't reverse something that's not symmetrical. If you flip the expectancy, atleast in the cases mentioned above, you'd just end up cutting your ROI proportionally.
  • #18
  • Jan 23, 2009 7:55am Jan 23, 2009 7:55am
Quoting Rabid
Unless it goes 6 pips in one direction, then 11 back.

If you want to "reverse" the system rules you have to keep the MM, SL and profit targets. You can't cut them in half, if you do, you change the rules and the outcome.
My reasons for "cutting" my SL and TP in half or switching them around was to maintain the same MM and probability that price would hit that level.

If the original trade hits my SL of 5 and never hits my TP of 10 that means when I reverse it I must make my previous SL my new TP. If I were to keep them the same then it would not truly be the same trade cause now price would have to go an extra 5 pips in that same direction where it previously hit the SL just to hit my new TP

Example
EURUSD price is 1.3000 I enter Long
SL at 1.2995
TP at 1.3010
Essentially because of 2pip spread price must go 3pips against me to hit SL and 12 pips in my favor to hit TP

Result. Loss, price hit SL at 1.2995

Reverse the trade
EURUSD price is 1.3000 I enter short
SL at 1.3014
TP at 1.2999
Essentially because of 2pip spread price must go 12pips against me to hit SL and 3pips in my favor to hit TP.

Result: Win, price hit TP at 1.2999

See how by "cutting my TP in half" (really just switching my SL and TP) I maintain the same probability that price will hit that level when reversing the trade? Otherwise if I maintained a 10pip TP still then price would have to go an extra 5pips south in my short trade to be a win. It would not be the same trade. Hopefully my example cleared that up a bit. The spread is really only the only problem here when everything else takes place in a constant environment when backtesting.

I understand the problem with symmetry as well, no way around that so this will never be perfect. Again that's largely a result of the spread when talking about this closed environment of backtest data. the Expectancy remains the same since I maintain the same probabilities for price going a certain direction.
  • #19
  • Jan 23, 2009 9:38am Jan 23, 2009 9:38am
Quote
My reasons for "cutting" my SL and TP in half or switching them around was to maintain the same MM and probability that price would hit that level.

Uhm, but you're not.

Quote
If the original trade hits my SL of 5 and never hits my TP of 10 that means when I reverse it I must make my previous SL my new TP. If I were to keep them the same then it would not truly be the same trade cause now price would have to go an extra 5 pips in that same direction where it previously hit the SL just to hit my new TP

Right, but price doesn't move straight in a line. It wiggles, which means price can move back 5 pips before moving forward 10 pips. If your SL is at 5 pips, you're getting stuck in the wiggle.

Quote
EURUSD price is 1.3000 I enter short
SL at 1.3014
TP at 1.2999

I think your numbers are wrong. If you're setting your TP 1 pip and a 14 pip SL then price won't even pay your spread. If you go short at bid 1.3000 your breakeven would be 1.3000 - spread*point, spreads are tacked on to the ask, not the bid, so you pay them when you close the trade. That's unlike buying at the ask, where your ask includes the spread.

Quote
See how by "cutting my TP in half" (really just switching my SL and TP) I maintain the same probability that price will hit that level when reversing the trade? Otherwise if I maintained a 10pip TP still then price would have to go an extra 5pips south in my short trade to be a win. It would not be the same trade. Hopefully my example cleared that up a bit. The spread is really only the only problem here when everything else takes place in a constant environment when backtesting.

No, it doesn't. If price moved in a straight line, maybe, but it moves back and forth. The closest exit will be first executed in this instance. You can't get a perfect reverse of the trade, hence the symmetry problem, the best you can do is invert the ruleset.
  • #20
  • Jan 23, 2009 9:42am Jan 23, 2009 9:42am
Quoting Rabid
Right, but price doesn't move straight in a line. It wiggles, which means price can move back 5 pips before moving forward 10 pips. If your SL is at 5 pips, you're getting stuck in the wiggle.
Thats exactly right and thats were the OP is getting lost. The price is not linear!
0 traders viewing now