MTS Vs Random Entry - A Paradox | Forex Factory

Options

Permalinks
Print Thread

Similar Threads

Scripts for Limit entry, Stop entry, Market entry 47 replies

Random entry proves strength of money management. 18 replies

Paradox System: post your entry here 12 replies

Is random really random? 231 replies

Why technical indicators and MTS fail... 14 replies

MTS Vs Random Entry - A Paradox

  • #1
  • First Post: Dec 10, 2005 10:45am Dec 10, 2005 10:45am
Hi all...

I found this paradox... I have been doing HUNDREDS of automated backtests....
I've found these scary things:

a)more indicators does not nessesary means better results (I did this before I calculated indicator correlations). I tried a to program Elders trend following system and from what I remember, I stoped the idea...

b) Stop loss values and trailing stops changed has had drastic effects on my results....

c)Recently for my stops/exits (a crucial part) I began to use "adaptive" values rather than preset numbers... Didn't help much except I could test the strategy on numerous markets without having to guess point B.

I also manually went through a certain cool fibonacci strategy that I've thought off... Again it had variable exits (price projections).... It worked well in 2005 on many currencies, even EUR/PLN (or something like that)... I was excited, "I am going to shovel money with my excavator!!!" yet I started checking it from 2002 and it bombed....


Many automated trading systems have shown GREAT results for some period, and totally flushed the deposit on the next.... Hmmm... So I might have had 90-99% winning percentage for some time, but eventually a series of lossess have put the system into a minus....

Then I ran a totally random system which was constantly in the market, reversing at trailing stops and stoplossess (which were super tight).... Guess what, after a year it came out in minor pluses....


HMMMMM................... A paradox....
Anyone can shed some light?
  • #2
  • Dec 10, 2005 12:17pm Dec 10, 2005 12:17pm
Fanat,

I dont know if this would work, but what if you moved the stop loss points, on this new system you are working on, to 2 standard deviations from the median price. ( 2 is a suggestion, maybe more or less or some other statistical tool)

Maybe I am asking the wrong question, but what I am asking is, if there is a statistical value (for instance standard deviations) in which your new system will be still in the money most of the time if you widened your stops.

By widening your stops you would take advantage of the regression to the median phenomena and not be stopped out prematurely before price returned to the median. Assuming that returning to the median would be in your favor.

Yes there may be a big negative pip number you might be carrying but assuming you were trading in the direction of the major trend, price according to regression, would come back to your median.

An example. I use (standard deviation) in the example as a possible statistical device.
You determine that the fundamental move on your pair would be long, and you go long. Then you determined what a relatively safe (standard deviation) from your entry would be, based on price history for a month or more. Then you set a catastrophic stop at that point below your entry price. You also set a limit stop at the upper (standard deviation). Then you let your entry price run, letting it range into positive and negative numbers, even big negative numbers because your stop is far enough away and you have faith in the fundamental trend, your entry point, and price probably not hitting the (standard deviation) due to regression to the median.

If your entry goes into positive numbers you can look at the limit stop (standard deviation) as a high point where price probably will not hit before it regresses back to the median, and you would want to take profit as price approached that point.

Key understandings are that negative pips are not losses unless you close the trade. The size of your account may not let you trade this way. You can add or take off positions as price moves in or out of your favor. You are paying a lot of attention to the fundamental direction of the pair, and trade in that direction. Moves of 100 pips or so will not excite or anguish you if the move falls into the range you have set for regression to the median.

I would like to know what you think, and what statistical devices you might use to get an edge.

Lou
  • #3
  • Dec 24, 2005 7:02pm Dec 24, 2005 7:02pm
Hi,
Quoting Fanat
b) Stop loss values and trailing stops changed has had drastic effects on my results....

c)Recently for my stops/exits (a crucial part) I began to use "adaptive" values rather than preset numbers... Didn't help much except I could test the strategy on numerous markets without having to guess point B.

I also manually went through a certain cool fibonacci strategy that I've thought off... Again it had variable exits (price projections).... It worked well in 2005 on many currencies, even EUR/PLN (or something like that)... I was excited, "I am going to shovel money with my excavator!!!" yet I started checking it from 2002 and it bombed....

Many automated trading systems have shown GREAT results for some period, and totally flushed the deposit on the next.... Hmmm... So I might have had 90-99% winning percentage for some time, but eventually a series of lossess have put the system into a minus....
Ditto

Merry Christmas
Alan
  • #4
  • Apr 12, 2008 8:33am Apr 12, 2008 8:33am
Quoting Fanat
Hi all...

I found this paradox... I have been doing HUNDREDS of automated backtests....
I've found these scary things:

a)more indicators does not nessesary means better results (I did this before I calculated indicator correlations). I tried a to program Elders trend following system and from what I remember, I stoped the idea...

b) Stop loss values and trailing stops changed has had drastic effects on my results....

c)Recently for my stops/exits (a crucial part) I began to use "adaptive" values rather than preset numbers... Didn't help much except I could test the strategy on numerous markets without having to guess point B.

I also manually went through a certain cool fibonacci strategy that I've thought off... Again it had variable exits (price projections).... It worked well in 2005 on many currencies, even EUR/PLN (or something like that)... I was excited, "I am going to shovel money with my excavator!!!" yet I started checking it from 2002 and it bombed....


Many automated trading systems have shown GREAT results for some period, and totally flushed the deposit on the next.... Hmmm... So I might have had 90-99% winning percentage for some time, but eventually a series of lossess have put the system into a minus....

Then I ran a totally random system which was constantly in the market, reversing at trailing stops and stoplossess (which were super tight).... Guess what, after a year it came out in minor pluses....


HMMMMM................... A paradox....
Anyone can shed some light?
Backtests are not reliable.
1. The farther in time you go back the higher the time frame the tester uses and the results become more obscured. If you backtest 2 weeks the tester uses a 1M time frame. If you go back 1 year or more it may use the bars from a 4H time frame. By the time you go to 2002 its probably testing off of weekly bars. So the farther back you go the more obscure and unreliable your results are.
2. Currencies often gap over the weekends there by screwing up your TP, SL and TS. Making your tests unreliable.
3. Backtests do not put in the adjustments many companies make at news times for currency spreads.
So if you want reliable accuracy of backtests they are best for a couple of weeks and then there are too many other factors to make them reliable.

Forward testing is the only reliable way to test an EA and that can take months of patient testing which most people do not do. I recommend testing the EA on 18 different currencies at one time on a live demo and during a month it will come across about every scenario it will probably face. Then open a micro account and use it trading pennies, when it passes those tests then you might have the confidence to try real testing. Good luck.
  • #5
  • Jan 1, 2011 2:32am Jan 1, 2011 2:32am
I'm new to posting on FF and I trade a very small live account so take what I say with a grain of salt:

Price moves based off of two sided volume: people dumping money into a currency or pulling from it.

When people decide to take action on either side of this determines when the price moves up or down.

Figure out when people get confidence in a pair and buy in. Figure out when people get scared and dump it. Trading funky systems seem to be like chasing ghosts.

The reason most systems do not work is because they don't deal with S&R, mostly just moving averages crossing or something of the like.

Just my $0.02

Have a happy new year! 2011!!!!
  • #6
  • Jan 1, 2011 5:47am Jan 1, 2011 5:47am
Quoting nanningbob
1. The farther in time you go back the higher the time frame the tester uses and the results become more obscured. If you backtest 2 weeks the tester uses a 1M time frame. If you go back 1 year or more it may use the bars from a 4H time frame. By the time you go to 2002 its probably testing off of weekly bars. So the farther back you go the more obscure and unreliable your results are.
What kind of crap testing software do you use? Aren't you the guy that puts out EAs? I don't run backtests but with the testing software I have I can see any timeframe as far back as I have data for.

To OP, where's the paradox? You got positive results with random entries or exits, why not? Ever heard of the story where the darts beat the experts' stock recommendations? Consider this: the "system" you are comparing your random method to was an inherently losing system, therefore it should not be surprising that a random method may be more or less 50/50, in your case more than less, and beat your losing system. Of course if you haven't included a spread, your slight positive may actually be a slight negative.
  • #7
  • Last Post: Jan 23, 2019 4:53am Jan 23, 2019 4:53am
Quoting Fanat
Hi all... I found this paradox... I have been doing HUNDREDS of automated backtests.... I've found these scary things: a)more indicators does not nessesary means better results (I did this before I calculated indicator correlations). I tried a to program Elders trend following system and from what I remember, I stoped the idea... b) Stop loss values and trailing stops changed has had drastic effects on my results.... c)Recently for my stops/exits (a crucial part) I began to use "adaptive" values rather than preset numbers... Didn't help much except...
Automated backtesting?
0 traders viewing now