Re: Mix Oracle(OLAP, Data Mining-SVM) and Ruby With ETF Price Data
Date: Mon, 15 Jun 2009 09:06:43 -0700 (PDT)
On Jun 14, 3:39 pm, "dan.bi..._at_gmail.com" <dan.bi..._at_gmail.com> wrote:
> I just wrote up some notes about my efforts to use Data Mining-SVM to
> pick optimal times for an ETF trader to act in the market.
> The only attributes I fed to SVM were slopes of moving averages of ETF
> Result: SVM did not find slopes of moving averages of ETF prices to be
> very predictive.
> What I might study next are the relationships between Asian markets
> and US markets.
> This is based on my inspection of this google search:
> It boils down to this, "What's the dog and what's the tail?"
> It seems plausible that the trend of the Nikkei might be somewhat
> correlated with the trend of the S and P 500. Perhaps the same is
> true for the dollar value of the Yen.
> Comments anyone?
You've discovered what many day-traders have, there's a lot more dimension than you can express predictively in a moving average. You really need a minimum of several hundred variables to effectively model the problem, and even then you have edge situations that are off the chart, like, what happens when the whole system freezes up because too much information has been obscured to properly evaluate risk and everything goes to hell. S&P and similar averages lose information in the aggregation, and in a more global sense, skew the information arbitrarily. The Japanese were screwed for 15 years after the early 90's. Have fun.
-- _at_home.com is bogus. "You also may not combine the Oracle program with programs licensed under the GNU General Public License ("GPL") in any manner that could cause, or could be interpreted or asserted to cause, the Oracle program or any modifications thereto to become subject to the terms of the GPL." - oracle xe license.Received on Mon Jun 15 2009 - 11:06:43 CDT