In the last post we discussed how to model a financial time series with ARIMA plus GARCH models. In that post we discussed a number of articles that have been published on how to model financial time series with ARIMA plus GARCH. GARCH models are used when the market volatility is variable. ARIMA models are used with the assumption that market volatility is constant which hardly it is in case of most of the financial time series especially EURUSD time series or the GBPUSD time series. GARCH models work well on the daily timeframe but are not good for the intraday timeframe due to the high level of seasonality present.
Statistical Learning Theory (SLT) is a new branch of statistical analysis that is making rapid progress now a days due to the tremendous increase in the crunching power of the computers. Quants are using these techniques to develop powerful market modelling algorithms. This is a new article on forecasting GBPUSD daily rates using Statistical Learning Theory new method called Kernel Based Regularized Least Squares known as RLSR in short. You can download the GBPUSD Forecasting Using RLSR PDF and go through it.
The problem is that GBPUSD daily time series show non linear behavior. This new method called Kernel Based Regularized Least Squares tries to solve this non linear model by mapping sample onto a higher dimension space using a kernel. This paper claims that this Kernel Based Regularized Least Squares Model works better than than the random walk, linear regression, ARIMA and neural network modelling in forecasting GBPUSD daily price.
But we don’t believe statistical analysis will be able to replace technical analysis anytime soon. Technical analysis is much more intuitive as compared to statistical learning theory which many day traders will find too hard to learn. Technical analysis is much more easy to learn and in the end it works most of the time.