"Stupid" system for ES

I am trading live here in my Journal.

http://www.elitetrader.com/vb/showthread.php?threadid=48230

I love to trade and it is my life's work. I apologize for posting so much, I just can't keep myself away from ET.

You traders are great people. I joined the Denver Trading Group and maybe I can get more involved there and less involved here...

Michael B.




Quote from 1000MILESHIGH:

you jump on every sys bandwagon to come down the pike here

i would be interested to know if you made any money -ever- trading
More...
 
I backtested it on my 5-minutes minidow data. I am on European Central Time so that "buying the close of 1555", which in turn means buying the open of 1600, really means 0955 am and 10 am EST:

Inputs: x(6), y(4), z(12);

{*** entries ***}
If time=1555 and c < c[x] Then Begin
buy this Bar at c;
value2=c;
End;

If time=1555 and c > c[x] Then Begin
sell this Bar at c;
value2=c;
End;

{*** exits ***}
exitlong at value2+y limit;
exitlong value2-z stop;

exitshort at value2-y limit;
exitshort value2+z stop;

----

I optimised x, y, z, to find out many interesting things.

"Z" likes to be given a high value, 10-20 ticks and even more, the higher it is the more profitable the system is. Not strange because that's the stoploss - it likes to be high. That's why when I try to trade with small stoplosses I end up losing more money than with larger ones. That would encourage me to have no stoplosses at all, but then sooner or later you get that one trade that wipes you out completely. At any rate, with a small stoploss as he asked to test, the system loses money, even if it had zero commissions.

"y", take profit, likes a smaller value, like 6-10 ticks, which would be half as many on the SP I suppose.

"x" is how many periods we have to go back to a value that's indicative of a later reversal at 10 am. It seems that we should go back to 9.30 rather than to just 5 minutes earlier to find the required higher value for longs, and lower for shorts, which we need to trigger a trade. But from 9.30 to 9.50 works.

If we make all these changes, higher stoploss, further "previous close"...the system becomes profitable, but not very profitable. It makes 800 dollars with 1 minidow in 6 months on the long trades, and it makes just 100 dollars with the short trades.

The good news is that if I reverse the "> previous close" condition, it makes less money, so it means the idea of a reversal at 10 am is not wrong, but it's not enough for a system, at least as far as my limited testing capabilities tell me. I mean the system you thought of, with a much larger stoploss than you said, seems to work, but not well enough to be used, since you risk too much money to make too little.
 
A 'stupid system' I've thought of pretains to buying based on Highs/Lows of previous bar. Ex....buy on open of a 3 min bar when the previous 3 minute bar fails to make a new low. take profits at high, or whenever. Stop below low of previous bar.
 
I could set up and run the back test in about 15 minutes, but it will be pointless and waste of time(I've run many such tests). To begin with, it was not specified which direction the trade will take (with or against the previous bar). But, I am pretty sure that that is irrelevant. Previous bar trades are non predictive (random). (Even when they are slightly biased in one direction (say 1/2 hour reversals), the "edge" is so faint it does not overcome costs.) Picking a trade at random and then having a stop 1/2 the size of the cap is surely going to be a losing strategy. Costs alone will burry you. God forbid something like that backtests well - run like hell :D all it means is that you have managed to fit a prediction on which way 10:00 o'clock numbers surprises (utterly random) occur. This has been my experience. Hope it helps.
 
Quote from MAD10:

I could set up and run the back test in about 15 minutes, but it will be pointless and waste of time(I've run many such tests). To begin with, it was not specified which direction the trade will take (with or against the previous bar). But, I am pretty sure that that is irrelevant. Previous bar trades are non predictive (random). (Even when they are slightly biased in one direction (say 1/2 hour reversals), the "edge" is so faint it does not overcome costs.) Picking a trade at random and then having a stop 1/2 the size of the cap is surely going to be a losing strategy. Costs alone will burry you. God forbid something like that backtests well - run like hell :D all it means is that you have managed to fit a prediction on which way 10:00 o'clock numbers surprises (utterly random) occur. This has been my experience. Hope it helps.
More...

I have a tendency to post quickly and not explain things.

The concept I was trying to describe above is more of a scalp type system, w/ only exits/entries based on previous bar....direction would be based on the trend of the 3 minute bars, it would be a fade. I.E. 5 consecutive 3 min bars continue to *not* make higher lows...fade w/ this concept when it does.

Def. a system that for the average ET poster would have very high transaction costs and be useless. But if you are a scalper and like this sort of thing, there may or may not be an exploitable edge there.
 
Quote from MAD10:

I could set up and run the back test in about 15 minutes, but it will be pointless and waste of time(I've run many such tests). To begin with, it was not specified which direction the trade will take (with or against the previous bar). But, I am pretty sure that that is irrelevant. Previous bar trades are non predictive (random). (Even when they are slightly biased in one direction (say 1/2 hour reversals), the "edge" is so faint it does not overcome costs.) Picking a trade at random and then having a stop 1/2 the size of the cap is surely going to be a losing strategy. Costs alone will burry you. God forbid something like that backtests well - run like hell :D all it means is that you have managed to fit a prediction on which way 10:00 o'clock numbers surprises (utterly random) occur. This has been my experience. Hope it helps.
More...

Believe it or not that was the point. If we take a random entry and do the 2:1 Risk/Reward thing - what happens? In other words, it's like trying to cheat the market. The probability of up and down movements are both 50% but you want the moves to be continutoes to the degree that you can cut off loss at 1R while the wins go straight to 2R without retracintg to -1R. Err did that make any sense?:)

I have a theory.... maybe call it "edge elimination" theory. If you enter randomly 50/50 with 1:1 R/R than you will get 50 50 %s in reality. But if you try to mess with R/R thru fixed targets then the actual probabilities will adjust thermselves to even out the edge back to 0.


Example: 50%*1avgWin-50%*1avgLoss = 0 Edge
Now you try to be smart and make that avgWin 2 times avgLoss
and you hope , you salivate for 50%*2R-50%*1R=50%*R= + edge

but in reality the % proability of a win will now drop to roughly 33% ie.

33%*2R-66%*1R = 0 edge again ! :)

test it, see if im right on this hypothesis.
 
Here are the results of a quickly set-up backtest. I have only printed selected stats (their names are descriptive).
Note: The tests were done with the big contract SP and the look-in-bar resolution is only a minute bar (where the stop was taken before the cap). I believe that my cost assumptions are reasonable. (The resutls may not exactly match someone elses backtest due to some proprietary procedures for sizing and performance calculation)

The one thing that jumps out is that:
(As expected) All losses are attributable to costs. Without costs, It is a breakeven strategy.


**Capital: 2472197
**RETURNS AND RATIOS
1 year 3 years 5 years Since Incep
Annualized Return -150.96 -87.08 -85.25 -87.71
Volatility 56.04 39.66 33.55 32.74
Sharpe Ratio -2.69 -2.20 -2.54 -2.68
1 year 3 years 5 years Since Incep
Annualized Return -150.96 -87.08 -85.25 -87.71
Capped Volatility 50.33 36.28 31.02 29.93
Capped Sharpe -3.00 -2.40 -2.75 -2.93

**Yearly percentage returns
1997 -59.12
1998 -43.92
1999 -72.97
2000 -50.22
2001 -57.46
2002 -45.62
2003 -55.22
2004 -84.20
2005 -8.61
------
Edge -0.1744 wins 0 losses 9 ratio 0.0 PLUS 0 zeros

**YEARLY DOLLAR RETURNS
1997 128 (1,461,630)
1998 252 (1,085,729)
1999 252 (1,804,036)
2000 252 (1,241,623)
2001 248 (1,420,539)
2002 252 (1,127,693)
2003 252 (1,365,092)
2004 252 (2,081,665)
2005 82 (212,763)
------ ------------ ------------ ------------
(11,800,770)
Edge -0.1832 wins 0 losses 9 ratio 0.0 PLUS 0 zeros


**Monthly Stats
Edge -0.1832 wins 17 losses 77 ratio 18.1 PLUS 0 zeros


#Tick No.Trans Lots/Tran TotCost AvgCost/Trans
sp 0 0 11,006,132 0
----------------------------------------------------
All 0 0 11006132 0

sp WinTrades:721 LosingTrades:1228
AvgWin:2.3235 AvgLoss:-1.4191
AvgWtime:11 AvgLtime:7
sp stopped:1227 capped:721 noFollow:0 expired:1
 
thanks for the back test. it proved my hypothesis, or my grand theory of "edge elimination tendency". Yes? Where do we go from here though?
 
You must log in or register to reply here.
Back
Top