Hi Richard,
I'm not sure how much help I can be but I'll say what I can.
- what sort of improvements you see over traditional AFL strategy if you use the same or similar feature set?
I'm not sure what counts as "traditional AFL strategy". I've never been a trader trying to optimize a single security trading system. I've always been more interested in portfolio construction and allocation as the way to manage risk and gain returns. I can say feeding traditional ROC() measurements rather than memory measurements into my ML strategies produces less than 1/4 the results. I don't know what feeding memory measurements into "traditional AFL" strategies would produce. As a lark I've looked at some computationally cheap crossover strategies using memory measurements and they essentially produce nothing.
- does ML approach improve on traditional pattern, indicator or oscillator detection?
Hmmm. Again I'm not sure what results others get doing these approaches. My approach hangs on answering a few fundamental questions about securities and portfolios of securities,
1) What is the value of cash? I answer this using a technique that produces a daily price stream for the value of a dollar (up or down), so it can be evaluated on equal footing with securities.
2) What is memory for a market or a security? For this I have a technique which quantifies the degree to which a market or security is showing memory, the magnitude, and the direction/shape of the memory.
3) Have the correlations between securities changed in a meaningful way requiring action? For this, again, I have a method for measuring this.
The metrics I mention in (1) thru (3) are proprietary. I know they give me satisfactory results but I can't compare them with results anyone is getting using "traditional" methods because I have no way of benchmarking. At one point I did make a post here offering to aggregate such a benchmarking database if others were willing to share results but there was no interest/response.
- what classifiers/models work best for financial data series?
Again, I can't give you a general answer but I have had success with Bayesian Networks.
- how do you define your target variable - is it just one bar GainAhead?
This question is a little ambiguous. At the level of an individual decision to buy/sell a security, I have a "target variable" which is just increase in price over a holding period. At the level of the system, there is an objective function that ML tries to optimize which is a fractal stability measure. It makes random perturbations to trading period and to parameter values and measures the stability of results and summarizes that in a single risk/reward number.
- what is the trade frequency and time frame which is likely to work for retail trades (like us) considering commissions, slippage and other cost?
There is no "trade frequency". At a portfolio level, the algorithm measures the market each day to listen for a rebalancing signal. Shares move whenever reallocations occur. It can go long periods without reallocating or it might reallocate daily over an extended period of time. It just depends. On average, rebalancing occurs about once a month but the variation around that mean is very high.
__._,_.___
Posted by: rosenberggregg@yahoo.com
| Reply via web post | • | Reply to sender | • | Reply to group | • | Start a New Topic | • | Messages in this topic (11) |
**** IMPORTANT PLEASE READ ****
This group is for the discussion between users only.
This is *NOT* technical support channel.
TO GET TECHNICAL SUPPORT send an e-mail directly to
SUPPORT {at} amibroker.com
TO SUBMIT SUGGESTIONS please use FEEDBACK CENTER at
http://www.amibroker.com/feedback/
(submissions sent via other channels won't be considered)
For NEW RELEASE ANNOUNCEMENTS and other news always check DEVLOG:
http://www.amibroker.com/devlog/
This group is for the discussion between users only.
This is *NOT* technical support channel.
TO GET TECHNICAL SUPPORT send an e-mail directly to
SUPPORT {at} amibroker.com
TO SUBMIT SUGGESTIONS please use FEEDBACK CENTER at
http://www.amibroker.com/feedback/
(submissions sent via other channels won't be considered)
For NEW RELEASE ANNOUNCEMENTS and other news always check DEVLOG:
http://www.amibroker.com/devlog/
.
__,_._,___
EmoticonEmoticon