Constructing leading indicators (using technical or fundamental analysis) is very difficult especially for short-term prediction as in stock market prediction. Maybe a fuzzy DSS can deal in such situation. Therefore, I am asking for any help in this regard. What indicators are more important?
Few of my students did try it using neural networks with no success however (the prediction probability was somewhere between 0.5-0.6). I didn't try with FIS however. The approach for daytrade was opening/closing course and relation between candlestick chart (bar-chart). For the short / medium term prediction students improved prediction by including MACD and Stochastic Oscillators but they do practiced the prediction based on the course. If you do consider technical analysis as a manner of trend prediction you can include the volume, because it used to be the acknowledge or rejection of the bull/bear market when rapid course change occurs.
In any case this is hard cookie as the amount of the indicators is large and the problem complexity grows rapidly thus you may hit computational problems. You should also be aware that carious markets behave differently (it is particularly related to the size of the market and its dynamic), so the set of the indicators for i.e. Forex may be inequivalent to the stock market (i.e. NASDAQ). Good luck anyway, I do wish you to be the millionaire!
Hope You are ;-).
You're welcome.
I'm do interested on employing FIS on financial-related time series prediction so let me hear the results as I'm in the medical data mining at the moment and out of resources to do some extra job unfortunately. Thus I kindly ask for sharing the results (event if they're not promising).
For short-term predictability, the forecasting ability should be low compared to transaction costs related to trading strategies. Any gain could be dried out due to the transaction costs of trading. However, the leading indication should be changing from time-to-time. However, long-term predictability is usually topic of research in finance (see new fact in finance). and the predictive power of each variable could be investigate using either traditional econometric or alternative method (such as fuzzy logic).
First of all, let me stress that stock-market time series are in principle non-predictable because they are highly influenced by very random events, such as speach of some politician or top manager, nature or other kind of disaster, etc. If the general situation is more or less stable, some predictability of trend or tendency may be possible. We have developed a very effective method for analysis and forecasting of time series using soft computing methods (combintation of Fuzzy Transform and Fuzzy Natural Logic). Try our demo of the software tool LFL Forecaster
http://irafm.osu.cz/en/c110_lfl-forecaster/
- Badges
- Science topic