20dmf

[Pascal]

Todays' bounce is lying dead in the water. this is not good, knowing that we have some POMO money washing things down (I mean "up").


Pascal

8462

[happy]

Would It be possible to get an excel/csv posting of the historical 20dmf signals. For all users here ... finding a market directional model to help stay on the right side of the trade would obviously be incredibly beneficial, and I think the 20dmf would be a great tool for that, BUT ... I want to backtest some strategies to prove that.

The goal is that when 'my' system says to go long, I would only enter the trade if the 20dmf is bullish as well. many of us in this group currently do this, but having the past signals to test trading ideas and strategies would be incredibly beneficial for all involved. Since testing using historical data obviously is the only way to know if the strategy has the potential to be profitable and what the r/r ratios would be.

Bottom line ... it would be extremely useful!!

I think just the date and the signal type (buy, neutral, sell) is all that would be needed.

Thank you in advance!

[asomani]

Thank you for the update, Pascal.

You said in this thread (http://www.effectivevolume.com/showthread.php?3419-20dmf) that you would do 20DMF 'forum updates' in one thread.

Please do that, so that if we are not in the forum, we can get still alerted to any comments you may have by subscribing to the 20DMF thread.

[Pascal]

Would It be possible to get an excel/csv posting of the historical 20dmf signals. For all users here ... finding a market directional model to help stay on the right side of the trade would obviously be incredibly beneficial, and I think the 20dmf would be a great tool for that, BUT ... I want to backtest some strategies to prove that.

The goal is that when 'my' system says to go long, I would only enter the trade if the 20dmf is bullish as well. many of us in this group currently do this, but having the past signals to test trading ideas and strategies would be incredibly beneficial for all involved. Since testing using historical data obviously is the only way to know if the strategy has the potential to be profitable and what the r/r ratios would be.

Bottom line ... it would be extremely useful!!

I think just the date and the signal type (buy, neutral, sell) is all that would be needed.

Thank you in advance!

Sure. no problem. here it is.



Pascal

8467

[happy]

Sure. no problem. here it is. - Pascal

Thank you Pascal. I'm sending you an email related to this at gmail

[dsmcdougal]

Sure. no problem. here it is.



Pascal

8467

Pascal

Can you explain why there is a difference between your recent post of 20DMF Signals vs. what was once available (at least I cannot find it) on the forum website (http://www.effectivevolume.com/_ss.php?m=ssurl&p=Reports/20DMF_Pivots_Method.pdf). i have attached a Mac Excel .xls spreadsheet of the original signals.

Thanks

Dave

[dsmcdougal]

Pascal

... vs. what was once available (at least I cannot find it)

Pascal

I found your signals table from "The Effective Volume Investment Method". It appears to be different than your recent posting of May 24, 2011. I haven't tracked down if you have already explained this elsewhere. I am still looking.

Dave

[Pascal]

Pascal

I found your signals table from "The Effective Volume Investment Method". It appears to be different than your recent posting of May 24, 2011. I haven't tracked down if you have already explained this elsewhere. I am still looking.

Dave

Good find Dave! and Good question!

This is a pretty good question indeed!

There was an evolution between December and now: the IWM robot has been developed.
How did this affect the 20DMF method? In two ways:

1. When I developed the IWM robot and looked at the statistics, I noted that when the OB/OS indicator was extremely negative (below -90), we always experienced a reversal in the next days. This basically meant that pressing on a short in such conditions was not a good idea. I thus inserted that specific condition on the 20DMF: cover shorts when OB/OS reaches -90.
2. I then had a long thought about the "porosity" parameter that was set in the 20DMF. This parameter's intention is to avoid whipsaws when the indicator barely crosses a certain level and then reverses back. The porosity parameter says that the level must be crossed by a certain arbitrary level. The porosity parameter has a noble element though: it avoids the whipsaws effect in ill-timed trading decisions. This is positive when the 20DMF is used as a stand alone indicator. However, when it is used as an element of another indicator (such as the Robot), then the porosity parameter has a negative effect: it masks or delays to communicate the information that we have crosses some level. Since the 20DMF is handing the trade decision to the robot, I decided to limit the porosity parameter as much as possible and let the robot use the 20DMF together with his other indicators - and the associated statistic tables and the floor levels - to take trading decisions that were closer to the market.

The consequence is that indeed, the 20DMF as part of the robot produces more trades than the 20DMF when it was stand alone, and also produces lower returns (As a reference, the stand alone 20DMF generated 30 trades until Dec 15, 2001, for a total SPY return of 316%, while the actual 20DMF generated 36 trades during the same period, for a return of 291% as of January 28, 2011.

However, you can see below (first figure) the IWM returns using the multi time frame pivots and the stand alone 20DMF, compared to the Robot IWM return (second figure). The difference is very obvious.



Pascal

8536
8535

[dsmcdougal]

Pascal

Thanks for your in-depth explanation. So you have two systems: the original 20DMA and the new IWM Robot.

I will forward you tomorrow some Excel calculations I did for the two system signals on various ETFs.

Dave

[dsmcdougal]

Pascal

Here is a Mac Word 2004 document that describes my calculations in detail.

Executive summary: The 20DMF shows about a 10% higher CAGR than the IWM Robot over the 7/24/2007 to 12/2/2010 time period.

I have also attached two spreadsheets that show the actual calculations and charts of the two systems. I hope you can open and read them.

Executive summary: The two charts are worth 2000 words.

Dave

[Pascal]

Pascal

Here is a Mac Word 2004 document that describes my calculations in detail.

Executive summary: The 20DMF shows about a 10% higher CAGR than the IWM Robot over the 7/24/2007 to 12/2/2010 time period.

I have also attached two spreadsheets that show the actual calculations and charts of the two systems. I hope you can open and read them.

Executive summary: The two charts are worth 2000 words.

Dave

Thanks for this work Dave.
One major difference between the 20DMF and the Robots is that the 20DMF is mainly a trend following indicator, while the Robot will switch from trend following to mean-reversal depending on the combination of its LT/ST signals. Hence, in a strong trending market, the 20DMF should perform better as it wants to stay in the trade, while the IWM robot might leave the trade at some point and then be forced back in. However, in a choppy market with many "cover your shorts" signals, the 20DMF would stay in cash, while the Robot will try to get the best out of its short term signals


Pascal

[dsmcdougal]

Thanks for your explanation. Makes sense (and profit).

Dave

[asomani]

Thanks for this work Dave.
One major difference between the 20DMF and the Robots is that the 20DMF is mainly a trend following indicator, while the Robot will switch from trend following to mean-reversal depending on the combination of its LT/ST signals. Hence, in a strong trending market, the 20DMF should perform better as it wants to stay in the trade, while the IWM robot might leave the trade at some point and then be forced back in. However, in a choppy market with many "cover your shorts" signals, the 20DMF would stay in cash, while the Robot will try to get the best out of its short term signals


Pascal

Thanks Pascal.

It is in trend-following mode about 60% of the time and mean reversion mode about 40% of the time, based on the history available, is that correct?

[senco]

I would like to use the 20DMF signal for partially hedging (with IWM or TF futures) a portfolio of long traded stocks. My underlying assumption is that my long trades produce excess return, thus a trend following index hedging will improve risk adjusted returns; and that the 20DMF is better in identifying trend than 'simple' price moving-average or breadth based trend indicators. It seemed more appropriate than the robot signal as I want trend identification and not trade decisions; also it has less frequent signal.

Pascal - any feedback on this application? Do I miss something important?


Also:

Since the 20DMF is handing the trade decision to the robot, I decided to limit the porosity parameter as much as possible

If I do use the 20DMF signal for hedging it is standalone. Would it be advisable, and practical for me as a website user, to add the hysteresis (porosity) on my own?

A related comment: Would it be possible to add to the 'zoomed' 20DMF chart (and inverse ETFs graphs) the actual indicator reading, and the moving average value? When the signal is close to zero or MA level in the chart published before close it might be difficult to eyeball whether we have a transition or not.

[Pascal]

I would like to use the 20DMF signal for partially hedging (with IWM or TF futures) a portfolio of long traded stocks. My underlying assumption is that my long trades produce excess return, thus a trend following index hedging will improve risk adjusted returns; and that the 20DMF is better in identifying trend than 'simple' price moving-average or breadth based trend indicators. It seemed more appropriate than the robot signal as I want trend identification and not trade decisions; also it has less frequent signal.

Pascal - any feedback on this application? Do I miss something important?


Also:


If I do use the 20DMF signal for hedging it is standalone. Would it be advisable, and practical for me as a website user, to add the hysteresis (porosity) on my own?

A related comment: Would it be possible to add to the 'zoomed' 20DMF chart (and inverse ETFs graphs) the actual indicator reading, and the moving average value? When the signal is close to zero or MA level in the chart published before close it might be difficult to eyeball whether we have a transition or not.

Sure. I'll see if I can add these figures.
As a trend following tool, the 20DMF is better than the robot. However, it obviously does not work so well when there is no trend.


Pascal

[asomani]

Thanks Pascal.

It is in trend-following mode about 60% of the time and mean reversion mode about 40% of the time, based on the history available, is that correct?

Reminder to Pascal. Thanks.

[Pascal]

I do not know and did not check that.
I trust that your evaluation is correct.


Pascal

[Billy]

I do not know and did not check that.
I trust that your evaluation is correct.


Pascal

I think there is some confusion here with the IWM robot performance.
60% of profits (not time!) came from LT "trend-following" trades and 40% from ST "mean-reversion" trades. But ST trades are usually active when the 20 DMF is neutral, so there is no clear correlation.
Billy

[asomani]

I think there is some confusion here with the IWM robot performance.
60% of profits (not time!) came from LT "trend-following" trades and 40% from ST "mean-reversion" trades. But ST trades are usually active when the 20 DMF is neutral, so there is no clear correlation.
Billy

Thanks for clearing that up, Billy.

So, regarding time, I guess it's a question of how often markets are trending vs. range-bound. Based on your experience and knowledge, what would you say this is? I've heard some say the market is trending 1/3 of the time and range-bound 2/3 of the time. Is this true?

Perhaps we have been spoiled in recent years (2008, 2009, 2010) with some very nice market trends.

[Billy]

Thanks for clearing that up, Billy.

So, regarding time, I guess it's a question of how often markets are trending vs. range-bound. Based on your experience and knowledge, what would you say this is? I've heard some say the market is trending 1/3 of the time and range-bound 2/3 of the time. Is this true?

Perhaps we have been spoiled in recent years (2008, 2009, 2010) with some very nice market trends.

Aly,
I've heard and read for years the same 2/3 and 1/3 proportion.
But it all really depends on the timeframe you're looking at. I know traders in this forum who deem a 1-hour trend an eternity and others who are only looking at monthly trends!
For the robot, ST is a 3-day horizon and LT is a 10-day horizon. It seems to be the best timeframe compromises for optimizing what we are trying to achieve: consistent compounded (very) long term risk-adjusted returns.
I suggest that you explore the Market Sci blog archives for their posts about research on mean-reversion and trend-following markets. They were big promoters of mean-reversion strategies until they discovered that for the last few years a shift clearly happened in favor of trend-following strategies. They explain why they think this phenomenon can only accelerate in the years ahead.
I think also that modern algo and HFT trading can only precipitate more trend-following.
Billy

[John.Jacobs.EV]

So, regarding time, I guess it's a question of how often markets are trending vs. range-bound. Based on your experience and knowledge, what would you say this is? I've heard some say the market is trending 1/3 of the time and range-bound 2/3 of the time. Is this true?

The market is not typically range bound, or flat, this much of the time. A way I've found to test this is to determine peak and trough dates combined with periods of time when the movement up and down is within a range of three percent either way. In other words, when the movement is very minor this reflects a range-bound, or flat, market that is pretty much going nowhere.

The actual percentage of time in a flat market all depends on what you require the bull and bear trending market moves to be. The following table shows a move anywhere from 7% to 20%. The date range for this on the S&P 500 is 1970 to present.

http://i1083.photobucket.com/albums/j399/john_jacobs1/Bull-Bear-Flat-1970-2011.png

So for 20% trend-moves in the S&P 500 the amount of time in a range-bound market is 24.91%. When you drop this to 7% the time in a flat market moves down to 18.97%. If you average all four you end up with 22.40%.

This is the same thing but for the periods from 2000 to present:

http://i1083.photobucket.com/albums/j399/john_jacobs1/Bull-Bear-Flat-2000-2011.png

The average of all four is 22.15%. So it is pretty clear that the last eleven years have not been all that much different than the longer period of 1970-2011.

[asomani]

The market is not typically range bound, or flat, this much of the time. A way I've found to test this is to determine peak and trough dates combined with periods of time when the movement up and down is within a range of three percent either way. In other words, when the movement is very minor this reflects a range-bound, or flat, market that is pretty much going nowhere.

The actual percentage of time in a flat market all depends on what you require the bull and bear trending market moves to be. The following table shows a move anywhere from 7% to 20%. The date range for this on the S&P 500 is 1970 to present.

http://i1083.photobucket.com/albums/j399/john_jacobs1/Bull-Bear-Flat-1970-2011.png

So for 20% trend-moves in the S&P 500 the amount of time in a range-bound market is 24.91%. When you drop this to 7% the time in a flat market moves down to 18.97%. If you average all four you end up with 22.40%.

This is the same thing but for the periods from 2000 to present:

http://i1083.photobucket.com/albums/j399/john_jacobs1/Bull-Bear-Flat-2000-2011.png

The average of all four is 22.15%. So it is pretty clear that the last eleven years have not been all that much different than the longer period of 1970-2011.

Thanks for sharing that. Excellent work.

FWIW, I did a similar study today using different criteria (a number of methods to measure how often the market is in a clear bull trend or clear bear trend). I found that for 2000-2010, the SPX was in a clear bull or clear bear trend (i.e. not a flat market) 74% of the time - meaning it was range-bound 26% of the time.

Pretty close to your numbers!

[asomani]

Aly,
I've heard and read for years the same 2/3 and 1/3 proportion.
But it all really depends on the timeframe you're looking at. I know traders in this forum who deem a 1-hour trend an eternity and others who are only looking at monthly trends!
For the robot, ST is a 3-day horizon and LT is a 10-day horizon. It seems to be the best timeframe compromises for optimizing what we are trying to achieve: consistent compounded (very) long term risk-adjusted returns.
I suggest that you explore the Market Sci blog archives for their posts about research on mean-reversion and trend-following markets. They were big promoters of mean-reversion strategies until they discovered that for the last few years a shift clearly happened in favor of trend-following strategies. They explain why they think this phenomenon can only accelerate in the years ahead.
I think also that modern algo and HFT trading can only precipitate more trend-following.
Billy

Thanks for the tips, Billy.

Will take a look at those archives and ponder on your thoughts as well.

[adam ali]

Pascal,

Could you tell me the gain/loss of the 20DMF short trade initiated on May 5th (the last Excel file I have shows it as open). Also, would it be possible to put an updated 20DMF Excel trade file on the site so it's easily accessible?

Thanks again,

Adam

[Pascal]

Pascal,

Could you tell me the gain/loss of the 20DMF short trade initiated on May 5th (the last Excel file I have shows it as open). Also, would it be possible to put an updated 20DMF Excel trade file on the site so it's easily accessible?

Thanks again,

Adam

Attached.

Pascal

9159