The foreign exchange market is the largest and most liquid financial market in the world. Traders are the ones who influence the market, and they dictate if the price goes up or down. Proteus Software comes in to show you exactly what the traders are thinking, exactly when they are thinking it.
Proteus 5 uses the old values of a few standard indicators (RSI, MACD, prices etc.) to predict the future values of other indicators. Currently supported are the MACD indicator and the Stochastic oscillator, with excellent accuracy, while also providing a nice visual histogram for easy interpretation (see screenshots). It works due to the fact that it learns patterns of price movement which happened in the past, and applies those patterns to the current price providing a very good guess about how the price will move in the near future. No need to learn hundreds of candlestick patterns, this indicator has already learned them and is applying them for you!
Due to the indicators' learning algorithm (which uses artificial neural networks), the prediction accuracy varies with each pair and depends on current market conditions and volatility, but on average is around the following values:
| Bar no. | MACD(20,200,20) main |
MACD(20,200,20) signal |
MACD(12,26,9) main |
MACD(12,26,9) signal |
Stoch(10,10,20) main |
Stoch(10,10,20) signal |
|---|---|---|---|---|---|---|
| #1 | 90% | 95% | 82% | 93% | 84% | 94% |
| #2 | 88% | 92% | 80% | 91% | 80% | 90% |
| #3 | 85% | 92% | 77% | 91% | 78% | 88% |
| #4 | 80% | 92% | 75% | 90% | 72% | 88% |
| #5 | 75% | 91% | 69% | 90% | 67% | 88% |
The reason why the signal line is generally more accurate is because it reacts slower to price changes.
The accuracy was measured in terms of: the number of times that the indicators predicted the indicator value/signal will be larger or smaller than the previous bar, and the real value/signal was actually larger or smaller, as predicted. So basically, if the indicator predicted the value will go up, and the value really did go up, then it was successful. The average difference between the predicted values and the real values was also very small but varies slightly with each pair.
Percentages are based on 150 random tests (tested multiple times on each pair; test data is separate from training data), while training the indicator was done using 20.000 bars (2+ years of data in case of H1 time frames); this time span provides the optimum results.
It's obvious that the prediction accuracy moves down the further away in time we try to predict, which is normal: more bars mean more time, which means real world events have a greater chance of affecting the price.
