Experiments in Technical Analysis

karthikmarar · #**731** 5th February 2007, 11:16 PM

Quote:

Originally Posted by asnavale

Hi Karthik,

You are right. The Histogram method has inherent problem of whipsaws. I have already used the Hi-Lo threshold and MA threshold to reduce them. I have described the two options in an earlier post in this thread. You might have gone through that. I was expecting some comments from you on this. Although there was reduction in whipsaws the two alternatives had some other problems in specific cases as I have pointed out in that post. The MA threshold still gives a few more whipsaws compared to Hi-Lo threshold.

I am working on another alternative. It appears to give better performance than all the earlier variants. I will post it in a day or two. Your comments and guidance will be very useful for refining the system.

Regards

-Anant

Dear Ananth

Sorry, I did not give any comments as I had not really done a thorough check on your new modifications. Whatever I did showed clearly that these were still prone to too many whipsaws.

I was also experimenting with a dead band and all buy and sell signals will be active only beyond these bands. But these bands have to variable and one needs to adjust them based on the historical value of that particular stock. My work is incomplete as I am finding very little time nowadays. And that is one of the reasons for the MACD posts getting delayed. Do post your new idea and will surely check out and comment.

Warm regards

Karthik

uma_k · #**732** 6th February 2007, 09:57 AM

Hi Anant,

I have one request. In your modification to MABIUTS-HB system using Hi-Lo and MA thresholds you have given AFLs for each of them. Can you combine both and make a single AFL with provision for selecting either MA or HI-LO threshold? This will help in running a single AFL instead of two and visualising the differences better.

Regards

*** Uma ***

asnavale · #**733** 6th February 2007, 08:37 PM

Quote:

Originally Posted by uma_k

Hi Anant,

No new post from you for long time. Are you working on any new twist to Experiments in Technical Analysis.

Regards

*** Uma ***

Hello Uma,

I am Working ... Working ... Working ... Working ... Working ...

When I get something worth posting I will surely do. Thanks for your interest.

Regards

-Anant

kkseal · #**734** 6th February 2007, 10:02 PM

Hi Anant,

Just adding to my last post.

I use a shorter-term MA crossover filter in my own humble system. Works better than without one but i can tell you it's not perfect.

Regards,
Kalyan.

asnavale · #**735** 6th February 2007, 10:11 PM

Hi Kalyan,

I have already used this filter in the threshold based strategy. I have introduced 9-day MA for BUY and 3-day MA for SELL. Works well but not completely satisfactory. Similarly, in another option I have used last 9-day's highest Close as threshold for BUY and last 3-day's Lowest Close for SELL. This almost is same as the MA-based system. But has its own drawbacks. So, trying to find out some other alternative to filter whipsaws as much as possible.

Thanks for your interest and suggestions. Keep it up so that you may give me some idea to try out.

Regards

-Anant

asnavale · #**736** 6th February 2007, 10:47 PM

Quote:

Originally Posted by uma_k

Hi Anant,

I have one request. In your modification to MABIUTS-HB system using Hi-Lo and MA thresholds you have given AFLs for each of them. Can you combine both and make a single AFL with provision for selecting either MA or HI-LO threshold? This will help in running a single AFL instead of two and visualising the differences better.

Regards

*** Uma ***

Hello Uma,

Here is the AmiBroker AFL code you wanted. You can select the MA-threshold or Hi-Lo threshold in Parameters Window whien you right-Click in the chart.

Code:

("MABIUTS-HB : Threshold Based");
 p1  = Param("EMA Period", 15, 1, 200, 1);
 p2  = Param("Smoothening Period", 15, 1, 200, 1);
 p3  = Param("Bollinger Period", 5, 3, 25, 1);
 p4  = Param("Bollinger Band Width", 1, 0.5, 3, 0.1);
 p5  = Param("Period for Buy", 9, 3, 50, 1);
 p6  = Param("Period for Sell", 3, 3, 50, 1);
 p7  = Paramtoggle("Threshold Based On", "MA | Hi-Lo", 1);

  X  = EMA(C, p1);
  Y  = EMA(X, p2);
  D  = X - Y;
BBB  = BBandBot(D, p3, p4);
BBT  = BBandTop(D, p3, p4);

 Buy  = Cross(D, BBT) AND (C >= MA(C, p5));
 Sell = Cross(BBB, D) AND (C <= MA(C, p6));

If (p7)
          {  Buy  = Cross(D, BBT) AND (C >= HHV(C, p5));
             Sell = Cross(BBB, D) AND (C <= LLV(C, p6));
          }

Buy  = ExRem(Buy, Sell);
Sell = ExRem(Sell, Buy);

shape = Buy * shapeUpArrow + Sell * shapeDownArrow;
PlotShapes(shape, IIf(Buy, colorBrightGreen, colorRed), 0, IIf(Buy, L, H));

The same code is attached herewith as a text file. You can download the file, change the extension to .AFL and directly use in AmiBroker charts.

Regards

-Anant

uma_k · #**737** 8th February 2007, 09:55 AM

Hi Anant,

Thank you for uploading the AFL.

CONGRATULATIONS
FOR ACHIEVING 200+ POSTS

Thanks

*** Uma ***

asnavale · #**738** 8th February 2007, 07:48 PM

Thank you Uma

-Anant

asnavale · #**739** 8th February 2007, 08:04 PM

THE MABIUTS-HB WITH RSI

The Story so far:

The MABIUTS-HB system is based on difference between two Moving Averages. Any Moving Average based system is prone to whipsaws and MABIUTS-HB is no exception. I have tried to reduce the whipsaws in following ways:

By introducing Bollinger Bands of the Histogram and BUY and SELL are based on the cross-over of the Histogram and the Bollinger Bands. This has resulted in some reduction in whipsaws.
I further refined the system by introducing a filter. There are two alternatives for the Filter: one based on short term MA and another based on previous High or Low over a short period. Both of these are almost identical except in few specific cases. Both of these did reduce the whipsaws to a considerable level. But the occurrence of whipsaws is still a problem to be solved.

What is New now:

Karthik has recently suggested some kind of indicator based Filter though he did not specify which. I have tried a few different indicators and found that RSI can be used as a Filter to reduce whipsaws to a much more acceptable level, though not completely eliminating them.

I am describing now the RSI based Filter applied to MABIUTS-HB system.

The basic signal generation is the same as earlier which is based on Histogram crossing over the Bollinger Bands. But the signal is valid only when the RSI is also favourable. With few trials with different values of RSI, I found that BUY condition when RSI is above 50 and SELL condition when RSI goes below 50 are satsfactory.

The commonly used default period of 14 for RSI was found to be inadequate for this MABIUTS-HB system. After testing with different values I have chosen the RSI period of 21 which appears well suited to filter the whipsaws. Therefore, I have taken RSI(21) as the Filter.

The BUY and SELL conditions are therefore generated as follows:

BUY: Histogram crosses and goes above upper Bollinger Band AND RSI(21) is 50 or more.

SELL: Histogram crosses and goes below lower Bollinger Band AND RSI(21) is below 50.

Here the Bollinger Bands are those of the Histogram itself, with a period of 5 and width 1.

The AmiBroker AFL using this method to plot the BUY and SELL arrows on the price chart is given below. I have also attached the AFL file with this post to so that you need not type it. Just download it and directly use by changing the extension from 'txt' to 'AFL'.

Code:

("MABIUTS Histogram With RSI(21) Filter");
p1 = Param("EMA Period", 15, 1, 200, 1, 1);
p2 = Param("Smoothening Period", 15, 1, 200, 1, 1);
p3 = Param("Bollinger Band Period", 5, 3, 50, 1);
p4 = Param("Bollinger Band Width", 1, 1, 3, 1, 1);

 X = EMA(C, p1);
 Y = EMA(X, p2);
 D = X-Y;
 
BBB = BBandBot(D, p3, p4);
BBT = BBandTop(D, p3, p4);

Buy  = Cross(D, BBT) AND (RSI(21) > 49.99);
Sell = Cross(BBB, D) AND (RSI(21) < 50.00);

Buy  = ExRem(Buy, Sell);
Sell = ExRem(Sell, Buy);

Marker = Buy * shapeUpArrow + Sell * shapeDownArrow;
PlotShapes(Marker, IIf(Buy, colorGreen, colorRed), 0, IIf(Buy, L, H));

I have enclosed three figures to show the effect of the Filters. each figure has three charts of the same stock. The bottom chart is without any Filter. The middle chart is with Hi-Lo Filter. The top chart is with RSI filter. In each figure you can see that the lower chart has too many signals. the middle chart has much reduced number of signals. The top chart has very few signals. These charts clearly show the pronounced effect of RSI based filter.

However, be aware that this is not the best solution. In several cases you can still see some whipswas, but less than the other options. On a casual observation I have seen in most cases when RSI oscillates between 45 and 55 the whipsaws are more. But if we use 45 and 55 as filter values the system performs poorly. Therfore, I have retained the RSI(21) Filter value of 50.

There is a need to address this problem. Your critical evaluation, comments and suggestions are needed. Please feel free to post your opinion.

Thanks

-Anant

kkseal · #**740** 8th February 2007, 09:05 PM

Hi Anant,

Excellent charts - brought out the differences clearly. Only in the case of Wipro there was a false sell signal during a consolidation phase.

Simce your system uses a 15d EMA and a smoothened version of it as the trigger line, i am not surprised that the conventional 14d RSI didn't work for you.

This seems better than all the filters that have been tried so far.

Regards,
Kalyan.

Similar Threads for: Experiments in Technical Analysis
Thread	Thread Starter	Forum	Replies	Last Post
Technical Analysis Course	snowy	Technical Analysis	28	7th November 2008 07:37 PM
Technical analysis course	sh50	Seminars & Tutors	18	10th June 2007 03:46 PM
Hello everybody.Technical Analysis	vijay chauhan	Introductions	2	30th May 2007 03:09 PM
Technical Analysis	crosscurrency	Forex	0	30th October 2005 11:00 PM