I've been in the market for a number of years by now and have done reasonably well without ever having used the tech analysis approach. That's because - (a) I don't know how to work TA and - (b) I have always considered TA to be an agreed approach whereby investors acted in unison by following the prescribed interpretations of the charts. A bit like traffic lights. However, I have no doubts that the method does work because there are many TA practitioners who do get good results. Sometime back a thought crossed my mind that if TA can work for individual stocks - why can't a program using TA be written to automatically select stocks which will appreciate in value? For all I know there may be a number of such programs already in use but, I thought, it would be one hell of a challenge to write another one. But I'd have to do it my way.
The reason why I had to do it my way is because I have no idea how commercial stock market programs work. I soon found that writing this program wasn't exactly a walk in the park but, after a lot of time and lots of effort, I am now seeing a light at the end of the tunnel. About a month back I started running tests to establish if and how my approach performs and am happy to say that at this stage the results are encouraging being better than bank interest by a factor or two.
To test my theories I run two systems, one processing the latest data whilst the other one is running data which is several months old. This second system has selection parameters built in enabling it to identify those stocks which are likely to go up. At the end of each market day's run I "buy" exactly $1,000 worth of every share that was selected this giving me the numbers of every share "bought". I then display all these stocks using charts that are current, and use the current prices to get the final values.
I started running tests using data as from 07/08/06 and processed every market day, the last one being 24/11/06 - a total of 80 trading days. The runs were terminated progressively starting from 06/02/07 and finishing on 16/02/07. The results are as follows -
Number of shares selected 275
Amount "invested" $275,000
Amount "returned" $371,812
Amount "gained" $96,812
% "Gained" 35.2%
Six shares more than doubled in value and four shares lost more than 40%. Overall 64 shares returned losses whilst 211 shares returned gains. On an average there were 3.4 selections per day. More stocks per day were selected in the August-September period than in the October-November period. Also, gains were lower in the latter period, possibly due to the reduction in run time.
Of course I am ignoring brokerages and CGT.
There were a couple of unexpected results -
1. I expected to see best gains from stocks that had dropped down to the bottom third of their annual price range. That wasn't the case. Best gains came from stocks that were in the top third of their annual price range. Very few came from the bottom range or even the middle range. This could be due to the parameters used in the selection filter.
2. If I was monitoring these stocks from the day they were "bought", the final result wouldn't have been as good. This is because I'd have sold a lot of them on the first significant dip. As it was they dipped and corrected and came up with reasonable final results.
At this stage I believe that better results are possible by doing some vetting of the selected stocks before "buying" them. Some improvement can be gained by modifying the selection filter, though this will require some deep thinking. Also, this program was written using only the daily closing prices without giving any consideration to volumes, except for rejecting all stocks whose annual daily average of shares traded was below 100k. I had looked at volumes but so far haven't been able to recognise any consistent patterns which I could use in this program. Certainly a possibility in the future.
I am looking forward to receiving some comments.
anon
The reason why I had to do it my way is because I have no idea how commercial stock market programs work. I soon found that writing this program wasn't exactly a walk in the park but, after a lot of time and lots of effort, I am now seeing a light at the end of the tunnel. About a month back I started running tests to establish if and how my approach performs and am happy to say that at this stage the results are encouraging being better than bank interest by a factor or two.
To test my theories I run two systems, one processing the latest data whilst the other one is running data which is several months old. This second system has selection parameters built in enabling it to identify those stocks which are likely to go up. At the end of each market day's run I "buy" exactly $1,000 worth of every share that was selected this giving me the numbers of every share "bought". I then display all these stocks using charts that are current, and use the current prices to get the final values.
I started running tests using data as from 07/08/06 and processed every market day, the last one being 24/11/06 - a total of 80 trading days. The runs were terminated progressively starting from 06/02/07 and finishing on 16/02/07. The results are as follows -
Number of shares selected 275
Amount "invested" $275,000
Amount "returned" $371,812
Amount "gained" $96,812
% "Gained" 35.2%
Six shares more than doubled in value and four shares lost more than 40%. Overall 64 shares returned losses whilst 211 shares returned gains. On an average there were 3.4 selections per day. More stocks per day were selected in the August-September period than in the October-November period. Also, gains were lower in the latter period, possibly due to the reduction in run time.
Of course I am ignoring brokerages and CGT.
There were a couple of unexpected results -
1. I expected to see best gains from stocks that had dropped down to the bottom third of their annual price range. That wasn't the case. Best gains came from stocks that were in the top third of their annual price range. Very few came from the bottom range or even the middle range. This could be due to the parameters used in the selection filter.
2. If I was monitoring these stocks from the day they were "bought", the final result wouldn't have been as good. This is because I'd have sold a lot of them on the first significant dip. As it was they dipped and corrected and came up with reasonable final results.
At this stage I believe that better results are possible by doing some vetting of the selected stocks before "buying" them. Some improvement can be gained by modifying the selection filter, though this will require some deep thinking. Also, this program was written using only the daily closing prices without giving any consideration to volumes, except for rejecting all stocks whose annual daily average of shares traded was below 100k. I had looked at volumes but so far haven't been able to recognise any consistent patterns which I could use in this program. Certainly a possibility in the future.
I am looking forward to receiving some comments.
anon