Abstract:
This dissertation presents the approach of fractal markets, whose founder was Benoit Mandelbrot, and which could serve as a counterweight to Efficient Market Hypothesis.
A key component of this work is the Hurst exponent, which is the main tool to quantify the existence of long-range correlations in the study timeline, and as supportive tools neural networks and technical analysis are utilized. In the empirical implementation, at first the fact that the euro/dollar higher frequency data have a smaller Hurst exponent is proved, confirming the view that higher frequency data are more "noisy", and also it is being displayed that long-range correlations are observed only for small windows of data.
Subsequently, an automated trading algorithm is generated combining exponent Hurst, technical indicators MACD and an adaptive moving average, and neural networks and found, giving it as an input the data of the last 2.5 years, that performance improves in case of applying the Hurst exponent with hourly data, unlike the case of 5-minute data, where there is no improvement.
Finally, a multifractal indicator is implemented so as to forecast volatility by using the technical indicator ATR, which is being applied to 15-minute data in the euro / dollar, sterling / dollar and Swiss franc / dollar for the last ten years, with pretty good results and then a change to the formula is applied, which significantly improves the results.