How to adapt lowpass/moving averages periods to the market?

Mithrandir777 · Post by **Mithrandir777** » Sun Sep 21, 2014 11:11 pm

Hi, I have been programming some eas with lowpass that catch for instance trends of 2 month length or 1 month length. But I found that strategies that succeed in backtests for instance EUR/USD in 2009-2014 fail miserably in 2002-2008.
Is there a way to adapt the moving average period to the market conditions? Thanks!

Cubic · Post by **Cubic** » Wed Sep 24, 2014 6:37 am

Nice! Your first post here is a million dollar question

There have been a number of attempts to use market conditions to adjust indicator parameter values on the fly:
-The Vidya (Volatility Index Dynamic Average) indicator is an exponential moving average that uses market volatility to alter the M.A.'s length,
-The Kaufman Adaptive Moving Average uses Kaufman's Efficiency Ratio to adjust the M.A.,
-The Mesa Adaptive Moving Average and other efforts have attempted to identify and use the dominant frequencies in the price waveform to adjust the M.A. length,
-Then you have self-correcting Kalman filters: http://en.wikipedia.org/wiki/Kalman_filter ...,
the Ehlers' Fractal Adaptive Moving Average, Russian work with "digital filters" rebuilt/customized periodically for the market being traded, and IIRC, I think Mark Jurik might've been adjusting Jurik Moving Average lookback or phase parameters with his Composite Fractal Behavior indicator.

You get the idea... There have been a lot of attempts at this.

Other efforts have been made to alter system rules on the fly, like this: http://www.earnforex.com/forex-e-books/ ... rategy.pdf
...others have opted for frequent re-optimization.

To some extent the answer (assuming that there really is one

) would be affected by the type of system that you are running, the market it is running on, and especially how the average is being used.

This post is barely 'scratching the surface'. I hope some of the Old Hands on this board will chime in on this important question.

Mithrandir777 · Post by **Mithrandir777** » Sat Sep 27, 2014 3:29 am

Thanks for your input Cubic, I really appreciate it! And I know it's a difficult question, I don't intend it to have a definite answer but at least some approximation to the problem. I also would appreciate if more members contribute to the discussion

I have tried using MAMA but it produced too many false signals compared to a lowpass filter. Another approach I found now -instead of rearranging the signal according to market conditions- is to filter the signals when the market has low volatility, for example using the Alligator indicator. I'll try to make a Alligator Lowpass filters and report if it gives good results.

Mithrandir777 · Post by **Mithrandir777** » Sat Sep 27, 2014 3:41 am

Cubic » Wed Sep 24, 2014 3:37 am wrote: -Then you have self-correcting Kalman filters: http://en.wikipedia.org/wiki/Kalman_filter ...,
the Ehlers' Fractal Adaptive Moving Average, Russian work with "digital filters" rebuilt/customized periodically for the market being traded, and IIRC, I think Mark Jurik might've been adjusting Jurik Moving Average lookback or phase parameters with his Composite Fractal Behavior indicator.

What are this "Russian work" with digital filters? And what is IIRC ? I didn't find it in google.

Cubic · Post by **Cubic** » Sat Sep 27, 2014 8:30 am

What are this "Russian work" with digital filters? And what is IIRC ? I didn't find it in google.

This was something that I saw years ago. It was formerly at http://www.fin-ware.com. I was just surprised to see that it is now an inactive website.

Here's what the site looked like in 2012:
http://web.archive.org/web/201203112058 ... -ware.com/

Many of the site's pages and 2 articles are preserved, but unfortunately the sample download links no longer work.

I did a little digging and it looks like http://finware.ru/ is the new online home for the company behind this tool. I have no affiliation.

~~~~

If I Recall Correctly (abbreviated as "IIRC"

), this was the toolset that I saw being discussed on some Russian trading forums. My understanding was that the spectral analysis-derived indicators were being calibrated for a specific currency pair and then re-tuned by users from time to time, to adapt to changes in market dynamics. The overall value of this approach, with these tools, is unknown. I would assume that the value, if any, would vary depending on the trading ruleset being employed.

~~~~

I just bumped into this: http://fx.qrz.ru/ and this, http://www.mql5.com/en/articles/32 - however I've only given them a quick skim.

~~~~

You may find this interesting: http://www.forexfactory.com/showthread.php?t=494141

Mithrandir777 · Post by **Mithrandir777** » Mon Oct 06, 2014 2:29 am

I have been busy with studies and work the last weeks. I have added to my script a filter that measures volatility. I could have used ATR but for some reason I found Alligator interesting so here it is the script and the results, the language used is lite-c from the Zorro platform:

Code: Select all

function run()
{
//set(PARAMETERS);
StartDate = 2002;
EndDate = 2008;
NumWFOCycles = 10;
LookBack = 2000;
vars Price = series(price());
var period = 500;
vars Trend = series(LowPass(Price,period));

Stop = ATR(100)*4;//optimize(4, 2, 8);

vars Meanness = series(MMI(Price,200));
vars Filter = series(LowPass(Meanness,period));

Alligator(series(MedPrice()));
var diff = FisherN(series(abs(rBlue-rGreen) + abs(rGreen-rRed)),500);
vars diffs = series(diff);
var Threshold = 2.5;//0.1;//1.0;//1.5;//2.5

static bool Filter2;
if (crossOver(diffs,Threshold) or crossUnder(diffs,-Threshold))
Filter2 = true;
if (crossUnder(diffs,Threshold) or crossOver(diffs,-Threshold))
Filter2 = false;

if( valley(Trend) ){
exitShort(); // close opposite position
if(falling(Filter) and Filter2)
enterLong();
} else if( peak(Trend) ) {
exitLong();
if(falling(Filter) and Filter2)
enterShort();
}
}

The results and equity curve:

Walk-Forward Test Workshop4_2Pc EUR/USD - performance report

Simulation period 01.05.2002-31.12.2008
Test period 30.11.2004-31.12.2008
WFO test cycles 9 x 2742 bars (23 weeks)
Training cycles 10 x 15538 bars (134 weeks)
Monte Carlo cycles 200
Lookback time 2000 bars (17 weeks)
Assumed slippage 10.0 sec
Spread 2.3 pips (roll -0.10/0.04)
Contracts per lot 1000.0

Gross win/loss 117$ / -44$ (+941p)
Average profit 18$/year, 1.48$/month, 0.07$/day
Max drawdown -34$ 47% (MAE -42$ 58%)
Total down time 41% (TAE 9%)
Max down time 68 weeks from Jul 2007
Largest margin 5.00$
Trade volume 17479$ (4278$/year)
Transaction costs -3.02$ spr, 0.10$ slp, -0.88$ rol
Capital required 34$

Number of trades 17 (5/year, 1/week, 1/day)
Percent winning 41%
Max win/loss 42$ / -16$
Avg trade profit 4.28$ 55.4p (+215.8p / -56.9p)
Avg trade slippage 0.01$ 0.1p (+0.6p / -0.3p)
Avg trade bars 151 (+346 / -14)
Max trade bars 542 (4 weeks)
Time in market 10%
Max open trades 1
Max loss streak 6 (uncorrelated 6)

Annual return 52%
Profit factor 2.65 (PRR 1.25)
Sharpe ratio 0.75
Kelly criterion 1.08
R2 coefficient 0.385
Ulcer index 6.0%
Prediction error 125%

Confidence level AR DDMax Capital

10% 103% 14$ 17$
20% 93% 17$ 19$
30% 85% 18$ 21$
40% 82% 20$ 22$
50% 77% 21$ 23$
60% 73% 22$ 24$
70% 67% 25$ 27$
80% 62% 28$ 29$
90% 55% 32$ 33$
95% 50% 36$ 36$
100% 33% 58$ 54$

Portfolio analysis OptF ProF Win/Loss Wgt% Cycles

EUR/USD .166 2.65 7/10 100.0 ./XX//\\\
EUR/USD:L .169 3.14 5/6 88.4 ..\///\\\
EUR/USD:S .147 1.60 2/4 11.6 .//\...\.

http://postimg.org/image/z3wui0ifz/

As you can see, I used Alligator lines closeness as a measure of volatility, I normalize their differences and when they go out of a threshold (which is higher the less volatile the market is, this threshold can be adapted to market conditions, maybe using equity curve trading for that) I consider it a signal to buy or sell.

This has turned the loser system into a winning one with good Sharpe Ratio for a trend system and an ulcer of 6%. What do you think? Do you think It has any bias or curve fitting? I look forward to hearing your inputs. Thanks!