Forex Automaton Black Box response to mean-reversion and trending

User Rating: / 3
PoorBest 
Written by Forex Automaton   
Sunday, 28 October 2012 12:09
Article Index
Forex Automaton Black Box response to mean-reversion and trending
Synthesizing Monte Carlo samples with autocorrelations
Testing Black Box on the Monte Carlo samples
All Pages

 Three types of input time series (open, close, high, low) have been simulated: one based on independent returns, a trending one, and one with an oscillating autocorrelation (with an effect dubbed "mean reversion" incorporated). The current version of the Forex Automaton Black Box algorithm with same parameters has been applied to each time series in the usual fashion, where after initial learning time, forecasts are generated for the coming bar using only past information, for the rest of the time series. 

The Monte Carlo study of the trending sample produces Black Box forecasts resulting in positive correlation coefficients between predicted and real returns for daily maxima, minima, and close.

 The Monte Carlo study of the oscillating autocorrelation ("mean reversion") sample reproduces the features of real-life (Danica) Black Box performance when applied to the daily FX bars aggregated with an opening at 9am US Eastern time, namely, positive correlation coefficients between predicted and real returns for daily maxima and minima in combination with negative correlation coefficients between predicted and real returns for daily close.

Geometrical random walk, independent returns, bar chart 1.1 Geometrical random walk, trending returns, bar chart 1.2 Geometrical random walk, oscillating returns, bar chart 1.2

Fig.1. Open-High-Low-Close bar time series. Monte Carlo. 1.1: Independent returns. 1.2: Trending returns. 1.3: Oscillating returns. The figure shows only a sub-sample of generated data.

The chars in Fig.1 probably look psychologically untrue to experienced technical traders and chartists, nevertheless they form an interesting quantitative playground to investigate and showcase capabilities of the algorithms.



Last Updated ( Saturday, 01 December 2012 14:47 )