so here we are with a system that downloads eod data each evening, uploads to a database and then calculates some indicators
psars are calculated on close, open, high, low and vol (although psar on vol is a little strange to say the least because the numbers involved are so large), as well as psars of open up and close down and high up and low down (psars trip at the top and bottom of trends, so these switch the up and down basis)
to make sure we have a trend we need the other indicators:
adx, vi and st are calculated as per wikipedia.
psars on vi and macd are also calculated just for fun and have proven very useful, as is the psar on the close psar
I enabled the embedded web server in the database and output reports to a web page, my ip is static and the database allows the creation of http code - I have programmed an http5 canvas reporting system. OEL has apache 2.2, may use this if I split the server into web and dbase in the future, but simple is the name at the moment
the interesting analysis happens when I link the indicators... as these are the start of the stock picking process I have called these alpha's.
alpha's consist of statements such as
show me all companies which have an average volume greater than 10 x 10,000 / close price (indicating that buying and selling of the shares will not be a problem), that over the last 10 days have had a minimum st of 0, +adx is greater than -adx, psar on vi is positive but was negative over the last 3 days, psar on macd is positive but was negative over the 3 previous days and the psar on close is positive with the previous 3 days being negative, and give me all companies that follow this rule from 1st Jan 2010 to the present day
this is approximately alpha 12, an alpha that works very well
the coding of alpha's interface is also via a web page, so the development cycle is I pick which alpha I want on a web page, edit, save, run, check resulting company charts, all via the web.
for live alpha, I merely run and due diligence the companies coming out with the date of today.
I run this statement from a specific date and can gauge effectiveness of alpha's, at the same time have programmed a routine that automatically goes through the alpha's and indicates which are good and which are not, this takes a lot of processing. current analysis is choosing companies at random and checking out movement, if price goes up after indicated date, then choose another company until I find one that does not follow the up rule and start the process of finding out what indicator runs against the other companies. I do this iteratively, each time changing the code and then running and checking again. each time the number of returned companies that follow the rule reduces - the art is to not restrict the companies to an extent that no companies are returned, but return all companies that increase in value over the proceeding days.
the production of a fully tested alpha takes weeks
the result is that I have built up and am building a set of alpha's that under specific circumstances indicate a good probability that the company share price will go up...