Hi Friends,
We have experimented with different systems for about 3 weeks and the experience has been thrilling. Though no real money was involved the excitement and feelings were no less than real trading. The experiment started just before the substantial crash. Here I remember, Sanjay's question when I had done back testing of four different strategies under different market conditions. Refer to:
http://www.traderji.com/63837-post308.html
He had asked me why I had taken entry near the beginning of market crash. Now we also did the same. And the result was almost all Buys were hit by 5% stoploss (The stoploss concept is now being discussed actively by Ajay, Karthik and others). However, the MABIUTS-HB hit only one stoploss, other Buys were terminated by SELL signals, although resulting in some loss, but the losses were not as high as stoploss. On going through the algorithm to determine why stoploss was hit in one case, I have found that the criteria for BUY/SELL conditions given in the program are not adequate. In this regard the live trading excercise has helped in discovering the shortcomings of the system. I am now describing the shortcomings and the remedy. I would like the members of 'System Implementation Excercise' to go through it and suggest whether I can change over to the modification during the rest of the excercise.
The present MABIUTS-HB system gives a BUY signal when the histogram rises above the Bollinger Band Bottom (BBB) to the extent of at least 1% of the BBB. Similarly it gives a SELL signal when the histogram falls below the Bollinger Band Top (BBT) by atleast 1% of BBT.This arrangement gives the signals with much less whipsaws. The drawbacks I discovered are:
1) If the histogram crosses the Band to the extent of less than 1% there is no BUY/SELL signal and next day if it goes beyond 1% we do not get any signal. This results in missing BUY/SELL opportunities which may be crucial ones. Therefore, I want to remove this restriction of 1% level of crossover. This may introduce a few more whipsaws but would be worth taking in order not to miss good opprtunities of entering/exiting the trades.
2) In an uptrend due to a small pull down, the histogram crosses below the BBT, it results in a SELL signal and we follow it and sell the stock. We then wait for a BUY signal. Now imagine, the uptrends continues and histogram does not fall below the BBB. There is no BUY signal. Instead the histogram may again go above BBT and later fall below it generating a SELL signal again, at a much higher price. But we can not take advantage of this SELL signal as we have already sold and there was no buy signal in between. To solve this we must introduce another BUY condition. This is done by generating a BUY signal when the histogram crosses above the BBT. [In AFL this is written as Cross(D,BBT)]. I would like to introduce this extra condition and follow it in the rest of the testing excercise.
3) A smilar situation arises in down trend. After histogram crosses above BBB generating BUY signal, it may again fall below it (meaning prices fall further) and later go above BBB generating another BUY signal without a SELL signal in between. This prevents us from taking advantage of lower prices. To avoid this we must have a SELL condition when the histogram crosses and goes below the BBB [Cross(BBB,D)]. I would like to make this change and use it in the rest of testing process.
Introduction of above three modification may introduce a few more whipsaws but it makes the system perform better and restrict the losses to much lower levels in case of sudden crashes. Members may please go through the reasonings above and approve the changes in the AFL to be used henceforth.
It was decided in the chat room session earlier that any member wanting changes to the system should present the case to the members with reasons and change only if the members approve. Therefore, I am presenting my case here and seeking your approval.
I am attaching the modifued AFL herewith. Change the .txt extension to .AFL before tesing it in AmiBroker.
Regards
-Anant
