This is a machine learning extension of the Elliott Wave principle.

- Blogs
- Elliottware: machine learning extension of Elliott Wave principle
- Elliottware with Neuroshell

- Public
- 2 likes
jaguar1637 liked this 3545 days ago

JohnLast liked this 3543 days ago

By JohnLast 3545 days ago Comments (5)

Here I tested the Elliotware approach of the Elliott wave principle with Neuroshell, I tried 3 basic strategies that are popular systems:

I used the training period on 1 h time frame, from the 01.09 to 26.09. And the green bars show the out of sample trading. Maybe it was mistake that I did not used the train and validate possibility. I used the full optimization option.

1. The Regression slope system. This system is based on linear regression slope with different bar lenght.

The system was very profitable.

PREDICTION

euro60m.cht

EURUSD (EURUSD)

Trading

Percent Change in Open 10 60 min bars into the future from the next open

2011.10.04 9:11:41 PM

GENERAL ANALYSIS

Statistic

Input Start Date 2011.09.23 7:00:00 PM

Input End Date 2011.10.05

Output Start Date 2011.08.31 5:00:00 PM

Output End Date 2011.10.05

Number of Bars 170

Average Error 0.472

Correlation (r) -0.038

R-squared -0.0334

Mean Squared Error 0.3045

% Correct Sign 52.94

Number of Trades 11

Return on Trades 2.7%

Annualized Return 87.7%

Long Entry Threshold -0.02

Long Exit Threshold -0.2

Short Entry Threshold -0.2

Short Exit Threshold -0.02

PREDICTION

euro60m.cht

EURUSD (EURUSD)

Trading

Percent Change in Open 10 60 min bars into the future from the next open

2011.10.04 9:12:28 PM

TRADING STATISTICS

Performance Statistic All Trades Long Only Short Only

Start Date 2011.09.23 7:00:00 PM

End Date 2011.10.05

Beginning Price 1.34759

Ending Price 1.32404

Change in Price -0.02355

Percent Change in Price -1.7%

Annual Percent Change in Price -56.9%

Return on Trades 2.7% 0.5% 2.2%

Annual Return on Trades 87.7% 16.9% 70.8%

Return on Account 2.7% 0.5% 2.2%

Annual Return on Account 87.3% 16.5% 70.8%

Net Profit $ 0.03664 $ 0.00693 $ 0.02971

Gross Profit $ 0.05791 $ 0.01342 $ 0.04449

Gross Loss $ 0.02127 $ 0.00649 $ 0.01478

Ratio Gross Profit/Loss 2.72 2.07 3.01

Percent Profitable Trades 54.5% 40.0% 66.7%

Number Trades 11 5 6

Number Winning Trades 6 2 4

Number Losing Trades 5 3 2

Largest Winning Trade Profit $ 0.03969 $ 0.00919 $ 0.03969

Largest Losing Trade Loss $ 0.00805 $ 0.00475 $ 0.00805

Average Trade Profit $ 0.00 $ 0.00 $ 0.00

Average Winning Trade Profit $ 0.01 $ 0.01 $ 0.01

Average Losing Trade Loss $ 0.00 $ 0.00 $ 0.01

Ratio Avg Win/Avg Loss 2.27 3.10 1.51

Maximum Consecutive Winners 3 1 2

Maximum Consecutive Losers 3 2 1

Average Trade Span 16 bars 10 bars 21 bars

Average Winning Trade Span 21 bars 12 bars 26 bars

Average Losing Trade Span 9 bars 9 bars 10 bars

Longest Trade Span 72 bars 16 bars 72 bars

Longest Winning Trade Span 72 bars 16 bars 72 bars

Longest Losing Trade Span 15 bars 15 bars 14 bars

Largest Units Traded 1 1 1

Largest Winning Units Traded 1 1 1

Largest Losing Units Traded 1 1 1

Average Units Traded 1 1 1

Average Winning Units Traded 1 1 1

Average Losing Units Traded 1 1 1

Commissions Paid $ 0.00 $ 0.00 $ 0.00

Maximum Drawdown $ 0.01808 $ 0.00962 $ 0.01546

Maximum Open Trade Drawdown $ 0.01276 $ 0.00962 $ 0.01276

Required Account Size $ 1.36678 $ 1.36678 $ 1.36678

2. MACD inputs. With this system MACD is used as an input.

PREDICTION

euro60m.cht

EURUSD (EURUSD)

Trading

Percent Change in Open 10 60 min bars into the future from the next open

2011.10.04 9:14:25 PM

GENERAL ANALYSIS

Statistic

Input Start Date 2011.09.23 7:00:00 PM

Input End Date 2011.10.05

Output Start Date 2011.08.31 5:00:00 PM

Output End Date 2011.10.05

Number of Bars 170

Average Error 0.455

Correlation (r) 0.187

R-squared 0.0242

Mean Squared Error 0.2876

% Correct Sign 55.29

Number of Trades 1

Return on Trades 1.7%

Annualized Return 56.9%

Long Entry Threshold -0.11

Long Exit Threshold -0.25

Short Entry Threshold -0.11

Short Exit Threshold 0.05

PREDICTION

euro60m.cht

EURUSD (EURUSD)

Trading

Percent Change in Open 10 60 min bars into the future from the next open

2011.10.04 9:15:09 PM

TRADING STATISTICS

Performance Statistic All Trades Long Only Short Only

Start Date 2011.09.23 7:00:00 PM

End Date 2011.10.05

Beginning Price 1.34759

Ending Price 1.32404

Change in Price -0.02355

Percent Change in Price -1.7%

Annual Percent Change in Price -56.9%

Return on Trades 1.7% 0.0% 1.7%

Annual Return on Trades 56.9% 0.0% 56.9%

Return on Account 1.7% 0.0% 1.7%

Annual Return on Account 56.9% 0.0% 56.9%

Net Profit $ 0.02355 $ 0.00 $ 0.02355

Gross Profit $ 0.02355 $ 0.00 $ 0.02355

Gross Loss $ 0.00 $ 0.00 $ 0.00

Ratio Gross Profit/Loss 0.00 0.00 0.00

Percent Profitable Trades 100.0% 0.0% 100.0%

Number Trades 1 0 1

Number Winning Trades 1 0 1

Number Losing Trades 0 0 0

Largest Winning Trade Profit $ 0.02355 $ 0.00 $ 0.02355

Largest Losing Trade Loss $ 0.00 $ 0.00 $ 0.00

Average Trade Profit $ 0.02 $ 0.00 $ 0.02

Average Winning Trade Profit $ 0.02 $ 0.00 $ 0.02

Average Losing Trade Loss $ 0.00 $ 0.00 $ 0.00

Ratio Avg Win/Avg Loss 0.00 0.00 0.00

Maximum Consecutive Winners 1 0 1

Maximum Consecutive Losers 0 0 0

Average Trade Span 169 bars 0 bars 169 bars

Average Winning Trade Span 169 bars 0 bars 169 bars

Average Losing Trade Span 0 bars 0 bars 0 bars

Longest Trade Span 169 bars 0 bars 169 bars

Longest Winning Trade Span 169 bars 0 bars 169 bars

Longest Losing Trade Span 0 bars 0 bars 0 bars

Largest Units Traded 1 0 1

Largest Winning Units Traded 1 0 1

Largest Losing Units Traded 0 0 0

Average Units Traded 1 0 1

Average Winning Units Traded 1 0 1

Average Losing Units Traded 0 0 0

Commissions Paid $ 0.00 $ 0.00 $ 0.00

Maximum Drawdown $ 0.02132 $ 0.00 $ 0.02132

Maximum Open Trade Drawdown $ 0.02132 $ 0.00 $ 0.02132

Required Account Size $ 1.34759 $ 0.00 $ 1.34759

3. Stochastic inputs. In this system it is used slow and fast stochastic inputs.

Here it was BAAAD

PREDICTION

euro60m.cht

EURUSD (EURUSD)

Trading

Percent Change in Open 10 60 min bars into the future from the next open

2011.10.04 9:15:43 PM

GENERAL ANALYSIS

Statistic

Input Start Date 2011.09.23 7:00:00 PM

Input End Date 2011.10.05

Output Start Date 2011.08.31 5:00:00 PM

Output End Date 2011.10.05

Number of Bars 170

Average Error 0.499

Correlation (r) 0.138

R-squared -0.136

Mean Squared Error 0.3349

% Correct Sign 53.53

Number of Trades 6

Return on Trades -1.8%

Annualized Return -59.6%

Long Entry Threshold -0.17

Long Exit Threshold -0.17

Short Entry Threshold -0.17

Short Exit Threshold 0.08

PREDICTION

euro60m.cht

EURUSD (EURUSD)

Trading

Percent Change in Open 10 60 min bars into the future from the next open

2011.10.04 9:16:29 PM

TRADING STATISTICS

Performance Statistic All Trades Long Only Short Only

Start Date 2011.09.23 7:00:00 PM

End Date 2011.10.05

Beginning Price 1.34759

Ending Price 1.32404

Change in Price -0.02355

Percent Change in Price -1.7%

Annual Percent Change in Price -56.9%

Return on Trades -1.8% -1.8% 0.0%

Annual Return on Trades -59.6% -58.1% -1.5%

Return on Account -1.8% -1.7% 0.0%

Annual Return on Account -58.3% -57.0% -1.4%

Net Profit $ -0.02473 $ -0.02414 $ -0.00059

Gross Profit $ 0.02014 $ 0.00442 $ 0.01572

Gross Loss $ 0.04487 $ 0.02856 $ 0.01631

Ratio Gross Profit/Loss 0.45 0.15 0.96

Percent Profitable Trades 50.0% 33.3% 66.7%

Number Trades 6 3 3

Number Winning Trades 3 1 2

Number Losing Trades 3 2 1

Largest Winning Trade Profit $ 0.00923 $ 0.00442 $ 0.00923

Largest Losing Trade Loss $ 0.02283 $ 0.02283 $ 0.01631

Average Trade Profit $ 0.00 $ -0.01 $ 0.00

Average Winning Trade Profit $ 0.01 $ 0.00 $ 0.01

Average Losing Trade Loss $ 0.01 $ 0.01 $ 0.02

Ratio Avg Win/Avg Loss 0.45 0.31 0.48

Maximum Consecutive Winners 2 1 2

Maximum Consecutive Losers 2 1 1

Average Trade Span 29 bars 27 bars 31 bars

Average Winning Trade Span 28 bars 31 bars 27 bars

Average Losing Trade Span 30 bars 24 bars 40 bars

Longest Trade Span 40 bars 31 bars 40 bars

Longest Winning Trade Span 39 bars 31 bars 39 bars

Longest Losing Trade Span 40 bars 31 bars 40 bars

Largest Units Traded 1 1 1

Largest Winning Units Traded 1 1 1

Largest Losing Units Traded 1 1 1

Average Units Traded 1 1 1

Average Winning Units Traded 1 1 1

Average Losing Units Traded 1 1 1

Commissions Paid $ 0.00 $ 0.00 $ 0.00

Maximum Drawdown $ 0.03873 $ 0.02478 $ 0.02489

Maximum Open Trade Drawdown $ 0.02489 $ 0.02347 $ 0.02489

Required Account Size $ 1.38021 $ 1.38021 $ 1.38021

От 23 септември 2011 |

## Comments

Very interesting

I am surprised about the bad result w/ the Stochastic inputs (may be use the calculation from Napoli. I will put this indicator for download)

You should try instead this input : Laguerre_Filter

Well this net was going bad since the first 24 h. The next test would be to use 4 stochastic nets they all will produce different results. That is why on my opinion it is not wise to base trading decisions on the result on only one net. We should use panels of them or combine them with other filter.

The positive thing is that the fitst net learned the time series well and gave good entries in the range and when the trend impulse came it did not made a counter trend prediction as most of the nets do most of the time lol.

Well this net was going bad since the first 24 h. The next test would be to use 4 stochastic nets they all will produce different results. That is why on my opinion it is not wise to base trading decisions on the result on only one net. We should use panels of them or combine them with other filter.

The positive thing is that the fitst net learned the time series well and gave good entries in the range and when the trend impulse came it did not made a counter trend prediction as most of the nets do most of the time lol.

I want to hsow some recent test of Elliotware approach with the Neuroshel Day Trader. The Neural net was a probabilistic network. I find this addon for the software one of the best. Adaptive net (Classify 2 with 2 inputs) with two inputs (2 wavelets) The idea was to predict the Wavelt filter Daubechie4 with 2 inputs:

Wavelet filter Daubechie4 (Calculates direct Daubechies-4 wavelet transform of X over the last n points. Applies threshold filter to the result (sets to zero all wavelet coefficients less than T% of the largest one except the Smoothing coefficient; the value of Smoothing (bias) coefficient is always set to zero). Calculates inverse Daubechies-4 wavelet transform of the result of filtering. Returns the result of the inverse transform corresponding to the current time point. )

WaveletvalDaub4 (Calculates discrete Daubechies-4 wavelet transform of X over the last n points.))

What matters here is the choice of the traininf and testing period according to the Elliotware approach.

I wish we had such wavelet indicators for Metatrader we would make a good use of them. However I think there are some implementations of probabilistic neural nets for Metatrader out there.

Here I want to show another shot with the continuation of the system, and you will see why trading with neural nets is not so easy. Here we have a fractal break - out and the system starts to counter the trend.

Why doing so? I think that it is doing so by its design, as it was optimized very well on the previous market conditions. Whenever the market switches its regime those conditions change and the system stops performing.

Even if I was using adaptive nets I was not using the genetic optimizer but the common proprietary neural net of Neuroshell, the turbo prop.

Here the switch of market conditions was perfectly clear, the break - out changed the local characteristics fundamentally. So all this discussion is about regognizing changes of market states. The problem is not in the neural net here, it is just the market that changes its local charecteristics.