Comments for Adaptive Trading Systems Blog

Comment on Obtaining the Full Signal History when Calculating Bot Performance by jamess

jamess — Fri, 03 Dec 2010 00:10:06 +0000

Thanks Ed. You shouldn’t use a global scope array to store a signal history within the CPerformance class of an equity engine. One instance of each equity engine is created by Dakota during start up. Function CalculateAdaptationPerformance() is called by the swarm server when it requires the performance for each bot. The Signal array for the given bot is passed into the function. If you had a 2 dimensional array so that you could store the bot id and corresponding signal history then you this would work. Even if you could get the bot id (I don’t think you can), the solution isn’t as neat.

James

Comment on New Trading Systems Reporting by jamess

jamess — Thu, 02 Dec 2010 23:59:14 +0000

The system used in the examples is based on a production system that predicts the WaveOsc2 indicator using the CPredict2 with the equity engine performance lookback period decreased from 1,000 trading days to 500 trading days and the CPredictor2 lookback period also decreased from 1,000 trading days to 500. I decreased the lookback periods to speed up the run.

The original PPIP (proportion of perfect while in position) equity engine will be included in the new set of equity engines as well as some new equity engines based on different performance adaptation metrics. When I have completed the development of these equity engines I will run some monte carlo simulations for a simple system using each of the equity engines.

James

Comment on New Futures Trading Systems Reporting by jamess

jamess — Thu, 02 Dec 2010 23:49:11 +0000

There is a distinction between running a futures trading versus stock trading simulation. I haven’t posted an example of the stock trading report yet. When running the futures trading simulation the net profit for a trade is calculated by multiplying the value per point by the number of points made or lost and subtracting the trading costs. The returns are calculated against the trading capital that remains at the value set.

When running the stock trading simulation the value per point is always $1. The assumption is that trading capital is 100% invested at the traded price when opening a position and, therefore, the number of shares traded varies. As with the futures trading simulation, the trading capital remains fixed.

These two simulation types are basic and additional methods for futures and stock trading, as well as no-load mutual fund trading, will be incorporated down the track.

James

Comment on Obtaining the Full Signal History when Calculating Bot Performance by ed

ed — Thu, 02 Dec 2010 02:23:55 +0000

great insight, apparently this beats my use of global scope variable to keep a running history of the signal for use inside the bot, to look back into history, so great insight.!

Comment on New Trading Systems Reporting by ed

ed — Thu, 02 Dec 2010 02:22:30 +0000

the equity curve is really impressive, almost 8% of perfect and 53% direction correct 60% winning trades and a .59 sharpe, my eyes are not so perfect and i cant zoom in any further.
what would be the results from the ordinary non log normal ppip older engine? perhaps this
would be even more interesting.
my guess is it underperforms, but only visible to a difference plot?

thanks

Comment on New Futures Trading Systems Reporting by ed

ed — Thu, 02 Dec 2010 02:16:47 +0000

is there a distinction between the futures equity engine and one used for cash stocks?
i only ask because i dont see the difference out side of parameter setting, but i could
be forgetting something.

thanks

Comment on Predicting SP Futures Returns with the K-NN Algorithm Part II by jamess

jamess — Tue, 14 Sep 2010 02:01:45 +0000

Hi Jeff,

Thank you for the feedback. The latest round of work on the CPredictor and other software is looking very promising.

Best Regards,

James

Comment on Predicting SP Futures Returns with the K-NN Algorithm Part II by Jeffrey Mishlove

Jeffrey Mishlove — Mon, 13 Sep 2010 01:27:20 +0000

James,

This looks very interesting and exciting. I look forward to learning the results of the additional tests that you plan.

Jeff

Comment on Adaptive ARMD(2) SP500 Trading System by jamess

jamess — Sun, 05 Sep 2010 01:47:37 +0000

Hi Ed,

Thanks for making me aware of the additional trading vehicles. Eventually I will compare the trading costs associated with each of them. I know Michael Stokes is keen on the 2X leveraged no load mutual funds and I’m thinking he has thoroughly checked out the slippage etc.

Combining the signals is going to be an interesting exercise.

Best Regards,

James

Comment on Adaptive ARMD(2) SP500 Trading System by ed

ed — Sat, 04 Sep 2010 19:05:50 +0000

Hi James
not only leverage mutual funds, but try double and triple leverage.
like sds, spy, sh, sso
Direxion Large Cap Bear 3X Shares(BGZ) (correlation/basis risk here)
Direxion Small Cap Bear 3X Shares(TZA)(correlation/basis risk here)
and lets not forget vxx + xxv
there are so many leveraged ways to make or loose money, just need
to be sure its worth the bet and control the risk

and need i mention that china affect australia, and now leads us markets
by the nose, and there are triple long and triple short china etfs also
but these suckers are illiquid and move.

“this is me wondering out loud”
but i have been playing with the idea that one signal might not be correct,
instead all the dv models combine to 1 vote, all the ar models combine to
one vote, all the xyz models are also sorted and combined into the “speciality”
vote, then a final choice could be weighted sigal average, a swarm vote,
something else to control for signal quality?

end of loud thinking