Altair RISE

A program to recognize and reward our most engaged community members

Nominate Yourself Now!

FOREX PREDICTION

Hi,

I'm a trader and I'm trading forex market. After years I'm looking for something automated or something that would help me make my decisions better and my life easier. This is very hard task itself, it's probably harder to create consistently profitable automated system than to trade manually. I've came across a website http://www.neuralmarkettrends.com where the author describes how he uses Rapid-I in his trading. Tutorials helped, but I still have a lot of question. I'm new to this platform, so it's not possible to know everything. I've managed to go thru his tutorials pretty fast and they were pretty easy to understand and to do.

So as I've managed to understand concept, the best will be to move to the actuall problem and to start getting to things I'm not sure of. My objective is to predict tomorrows close or some event in the future on a given time-series data. Data is basically simple OHLCV (open, high, low, close, volume) csv sheet that I can use. For inputs I will use several relevant sources that might be supporting predictions. If I will create some model that could possibly work, I need much more than just several periods ahead prediction. I will need to backtest this model and see how it would behave on historical data. Backtests can be misleading and can not compete with forward testing, but they have some sort of reporting value. We can better understand our current model if we know it's success rate over period of time. So my first question is... If I build the model, how will I be able to backtest it?

I have excel data sheet with f.e. daily or 4 h data consisting of the following columns: Date, Time, Open, High, Low, Close, Volume. After successful model creation I can only do prediction X bars (time periods) ahead f.e. 5 days. But if I have f.e. 10 years of the data... I will use 2 years for training and then I want to predict the following X periods... how do I move on so I can automate this so it will keep retraining and retraining with fixed 2 years (x periods) training data and proceed with X period prediction and write this all to some file or summarize my model broader time frame performance without having to manually retrain it and test X period and repeat all over for all 10 years?

Would someone be able and willing to help me with this?

Find more posts tagged with

AI Studio

Accepted answers

All comments

IngoRM

Hi,

well, predicting future values (forecasting) from time series is of course a complex topic which hardly can be explained in a few sentences. If you search this forum, you will notice that several users have similar goals and probably there are some people who are able to help. Here is an example for a discussion:

http://rapid-i.com/rapidforum/index.php/topic,16.0.html

(this is actually one of the most often read threads here).

If you are from the US, you might also be interested in the training courses we are giving in New York at the beginning of Oktober:

http://rapid-i.com/content/view/106/125/

The first two days are a good introduction into data mining techniques and RapidMiner in several different application fields, the following three days concentrate on advanced topics leading to prediction of series data and financial data forecasting.

Hope that helps,
Ingo

fxhst329

well I'm not interested in trading details, input selections or method, I pretty much know what I want to use, it's just this thing. I will create data set and models and everything and maybe someone here can show me how do I apply windowing to it. will post the files later

Legacy User

You can create your model from 2 years of forex data and then test 5 days of current data to get a prediction for next week. It will only work if your prediction data high and lows are contained within the trained model high and lows. If your prediction data set is outside of your trained model, then you need to retrain it.

I highly advise that you retrain your model with new data ever week because the Forex markets are so volatile and often new highs or lows are created daily.

Regards,
Tom
www.neuralmarkettrends.com

fxhst329

yes I know forex is very volatile market, if you look at it now, we're experiencing historical events... volatility is unbelievable... so if I would traing model on a historical data...it would not work now...

I realise all this, but it will be extremelly hard to manually backtest, week by week for like 4 years period. I would also like to try different timeframes, maybe 4H. I have a specific problem here I want RapidMiner solve and I'm very curious of how the results would be. so I need to apply this windowing to the project so it's possible to backtest it further back and automatically.

can we get in touch maybe?

brandt_martiN

Hello,

Did you get any results from your project.

Thanks