Wednesday 6 February 2013

recap III

Quite a few things have changed since the last update and so lets do another one.

This is all about getting a system that I have built called 'oasis' to deliver triggers as to when to buy shares, currently geared to the London Stock Exchange as I am based in the UK, but with the ability to easily integrate other exchanges - I have run tests on NYSE, NASDAQ and FX and these worked well, so am just waiting for someone to say they would like to see these and I will reinstate. 

oasis brings down via ftp, the end of day data each evening, cleans and uploads to a database. And then the calculations start.

We have 3 trigger processes, Candlesticks, Trace and Betas.

Candlesticks

This is the usual Candlestick algorithms with added bespoke relationship algorithms. We have 8 indicators related to the relationship between periods (day).
We calculate for each company the Markov Switching Multifractal (MSM) for each individual company for these 8 Candlestick indicators. 
Each evening we compare the daily calculated indicators against the MSMs and Tweet any that have a greater than 11.5% rise indicated greater than 95% of the time.
The Tweet contains 2 specific indicators, for want of better names I have called Tweedle Dee and Tweedle Dum which are average percentage price increase and percentage times that this has risen respectively. The Tweet also contains the number of times this has triggered and the day's volume as per the average 3 monthly volume, and are labelled LSE_cs.


Trace

After End Of Day data upload all indicators are calculated (PSAR, MACD, ADX, VI, RSI, Stochastics, Bollingers and quite a few mixed combinations of these such as the PSAR on the MACD)
Each one of these indicators and combinations has been assigned an alphabetic letter for the trigger, and for every combination of 6 of these indicators we have calculated the MSM prior to the End Of Day data upload. I noticed that more than 6 indicators resulted in very limited results. Examples of this are given in this blog here.
We then compare the daily calculated indicators with the stored MSM.
As an indicator the calculation of the MSM takes about 1.5 hours per company, and takes about 4 weeks of processing to complete the whole LSE exchange companies. This process is scheduled to run constantly.
All the trigger indicators are condensed into 2 indicators again called Tweedle Dee and Tweedle Dum, which are essentially the number of indicators triggered and the number of times this has been triggered.
After daily comparisons have been made, any companies that meet the specific criteria are tweeted under LSE_trace. The Tweet also contains the number of times this has triggered and the day's volume as per the average 3 monthly volume.


Beta

This is a fully bespoke and manual set of instructions the user can enter and test. This incorporates an HTML sql interface to the database along with full reporting capabilities.


There are two servers, one production that tweets and one development that tests new algorithms.
The production server is a 16 core Xeon processor, multiple solid state disks and 32 gigs of RAM.
The development server is a 12 core Xeon with ssds and 24 gigs of RAM.
Both servers run Oracle Enterprise Linux and Oracle 11g Database.

Email or comment if you would like more information

--- 
when does the Narwhal bacon ?

No comments:

Post a Comment