Volatility homogenisation decomposition for forecasting

We explore the idea that by modeling a financial time series at regular points in space (i.e. price) rather than regular points in time, more predictive power can be extracted from the time series. We will term this concept of modeling time series at regular points in space as “volatility homogenisation”. Our hypothesis is that if we select the correct quantum in terms of regular steps in space, we replace noise which can normally interfere with prediction methods and thus uncover the underlying patterns in the time series. Furthermore, this technique can also be viewed a way of decoupling spatial and temporal dependence, which again, can replace unnecessary noise. We apply this decomposition to nine different financial time series and then apply support vector classification in order to make our predictions on the decomposed time series. Our results show that in the majority of cases, this technique yields better predictions than applications to data that has regular points in time, with applications of techniques such as support vector regression and Autoregressive Integrated Moving Averages models. The contribution of this paper is that it demonstrates the efficacy of this new methodology known as “volatility homogenisation”.