Hi, looking for a partner(s) for data analysis

[erich]

Hi, my name is Erich and I am new to this forum.

I am a mathematician working in the field of data mining / statistics.
In recent years I have developed a new data mining algorithm which
in many ways is better than the standard text book algorithms like
neural networks, univariate decision trees, at least so I think.

The algorithm was not developed primarily with trade in mind.
One of my test problems I constructed with Yahoo trade data.
I did not really expect great results because I believed that
stock movements on the small scale are mostly Brownian motions and
hence can not really be predicted. But to my surprise I found that that is not true.
In the model the movement of the Dax-Index on a day is predicted using
movement key data from previous days of the Dax, Nikkei, Dow Jones
and HangSeng.

I am now looking for a partner with access to historical intraday data.
I could show him/her the results that I get so far with the Yahoo data using
a screen-sharing session on the internet. I am open to all kinds of collaborations.

Best regards, Erich

 

[t2wza]

I am interested

Hi Erich, Dragisa is here. I'm willing to help You in Your improvements. My interest is FOREX , and I can test Your algorithm. In the past few months my main interest became neural network, so we can together work on. But as I said my interest if forex.
Best regards and good luck.

 

[erich]

Hi Erich, Dragisa is here. I'm willing to help You in Your improvements. My interest is FOREX , and I can test Your algorithm. In the past few months my main interest became neural network, so we can together work on. But as I said my interest if forex.
Best regards and good luck.

Hi Dragisa,

thanks for your reply. I think we should have a session on the internet and
telephone or Skype together.

I have heard about Forex, but I have no idea about the details how it works
when actually trading and what kind of data you have at hand.
For data mining algorithms the data is the important part and you should
explain to me how you trade and what kind of data you have.

Because I am afraid of spam please send me an email
to my "play-email" with your contact details.
Please send an email to steinererich at hotmail dot com.

I am looking forward to showing you my results and working out
an application to Forex-trading together with you.

Best regards, Erich

 

[ronblack]

But to my surprise I found that that is not true.
In the model the movement of the Dax-Index on a day is predicted using
movement key data from previous days of the Dax, Nikkei, Dow Jones
and HangSeng.

I'm always interested in data mining algorithms applied to the markets and I'm willing to pay for custom applications. It appears there is a lot of potential in such techniques.

What was the probability of the predictions you mentioned, their historical success rate and frequency? Can you identify the conditions that lead to the predictions or what you have is a black-box algorithm that only does forecasting?

Are you aware of this program that does data mining but it is essentially a white-box?

Discover trading systems automatically with APS Automatic Pattern Search

Best wishes for the Holiday Season!

Ron

 

[erich]

Hi Ron

Hi Ron,

thanks for your post and your interest. It is a black box but also shows you
the key features of a pattern. So your question must be answered two-fold.
But this enters mathematical details which we should talk about in a
screen-sharing session.

At about 20% to 25% of the days it predicts that the Dax will move up
und the historical odds are than about 2:1 to 3:2 that it really goes up.
So setting a sell-order at plus 1% and at -1% of the opening Dax-value
essentially creates a casino game with the winning probabilities on your side.

Similarly for moves down. So at 50% - 60% of the days it predicts something
symmetrical about moving up and down. At these days it would be best
to not trade.

I think that a similar model could be built with all kinds of other
indices or also commodities. I am not really an expert on the
trading details in practice. I have never traded myself.
My expertise is the maths.
I know about instuments like options, futures and the like.
But I do not know how big the trading costs are for each instrument
for example.

Best regards, Erich

 

[Adamus]

Hi Erich

what you're doing is known as inter-market analysis. I don't know the exact proportion but most people stick to intra-market analysis because the set-up is easier. That makes it more likely that you've found something that could be profitable in reality.

The trading costs for the DAX DTB future would be around 5 to 8 Euros per trade depending on your broker, and the amount of trades. You would also have to account for the spread which the market makers take. Plus some slippage (Nachgeben in German I think - the difference between your theoretical ideal execution price - the opening price for instance - and the price of the fill on your trade that your broker gives you).

Once you add those costs, you might find your system's profitability disappears - but maybe not.

Have you calculated whether your theory would be profitable if you simply buy on open and sell on close?

 

[erich]

Have you calculated whether your theory would be profitable if you simply buy on open and sell on close?

Hi Adam,

thank you very much for your reply.
The prediction of the closeing value is not so successful.
The big success is with predicting if the day-max is higher than 1% with
respect to the opening value and at the same time the day-min is higher
than -1% with respect to the opening value.
Similarly for downward movements of the Dax.
So the trading strategy would be to buy and then set immediately
trade-closing sells at opening-value + 1% and opening value - 1%.
Going long or short respectively to the up- or downward prediction.

There is nothing special about the number 1. It works just as fine with
other thresholds but the probabilities and the proportion of days in
clear up or down-predictions change.

Best regards, Erich

 

[Adamus]

Hmm. Yes OK then you really do need the intra-day data for the Dax. Don't have it myself, sorry.

 

[leovirgo]

There are some practicalities when testing with historical data.
1) Yahoo data was not accurate when I used it three years ago. O,H,L,C are less than 100% accurate

2) Gap day trades may be impossible to take due to opening gap ups/downs. Theoretically, you buy/sell at opening price, but in reality, the market may gap up and gap down due to previous day's volatility. Mostly, when INDU closed up or down, dax and ftse are already gapped up/down. I have a very good system on INDU that produces v.high performance in back-testing but in reality I can't get an entry at opening price.