(1) any could be used. In fact, indicators you've mentioned will often produce similar signals if applied to the same timeframe with the same period. Difference in signals of MACD/RSI/Stochastics is usually explained by difference in default period parameter. (2) machine learning can not explain anything. You can use machine learning to combine signals from various indicators. Almost all would suffice, from ensemble learning to Kohonen maps.

Thanks Alexander. I was specific for short-time data forecasting like daily or weekly. Which indicator you think is better under such circumstances.

Well, actually, today weekly or even daily believed to be long- or at least medium-term :) One of the most actively advertised (by TA proponents) property of TA is its applicability to any timeframe. Clearly, if we are speaking about prices and volumes - which are inputs to any TA indicator - statistical properties of weekly, daily and intraday returns do not differ much (except for some extremes, like tickdata). That's why there can not be any recommendation to prefer some indicator over another conditioning on the timeframe only. Choice would depend only upon your goals; otherwise - select several most popular (like, say, MACD, stochastics, RSI, Williams indicators), or several, covering various aspects of price data: choose one oscillator - given that both RSI and MACD are also "oscillators"; one "filter" - EMA/SMA/etc; one "band" - Bollinger/Donchian/etc; one of "new highs-new lows" kind, like Aroon, etc. As I said, most indicators inside one group produce signals close to each other. Perhaps, you can even make a spinoff research by determining the set of technical indicators which tend to produce the most diverse signals, as, speaking in financial terms, making trading algorithm which relies on many similar indicators would result in system, having too much "beta" exposure to benchmark set of securities, instead of some significant "alpha" intercept (which directly opposes initial goal of constructing any algorithm - making source of alpha).

MACD, RSI, ROC, MA and Stochastic generate quite good signals of opening and closing position when there is strong upward or downward trend in price movement but in the horizon trend they generate late signals of opening and closing position. In horizon trend you should use Willimas %R indicator.

With machine learning there problem is that or you tell them what trend dominate in the price pattern or they have to guess it. In the latter case they need some time to discover (or learn) what trend is observed on financial markets. You should also be aware of the overlearning phenomena with machine learning – the program is adopting itself and responding to the given data set but no to the real market conditions.

I have written following three papers explaining the profitability of moving averages and oscillators (RSI, Stochastic, etc)

Article The Profitability of Five Popular Variations of Moving Avera...

Article Technical Analysis of Indian Financial Market with the Help ...

Article Profitability of Oscillators Used in Technical Analysis for ...

Hi there,

please refer to this article, I hope it helps.

regards

Article ANALYSIS OF A STANDALONE USAGE AND LIMITATIONS OF A RELATIVE...