Abstract
In this work, we have applied non-linear time series techniques to the Nordic spot electricity market data. The time series are given in two periods, from May 1992 to December 1998 in Norwegian Kröne per MWh and from January 1999 to January 2007 in EUR per MWh. Our main interest was on trying to classify these series and analysing if their dynamical behaviour were in some way correlated with known events, e.g. the evolution of the Nord Pool and the climatic factors. First, a preliminary study was carried out with the aim of characterising the time series in terms of power spectral distribution, long term memory (R/S analysis), stationarity (space-time separation plots) and tails (stable distributions). Then, we used two types of surrogate time series, the first type was generated by a Gaussian linear random process with the same FFT of the real data set, whereas the second type consisted on a shuffled version of the original data series i.e. with the same statistic properties but without any correlation. We used these surrogates to check if RQA measures were able to detect differences with the real data. Finally, we used two RQA measures, %determinism and %laminarity, for developing a new measure of volatility which was able of detecting important historical and meteorological events with better resolution than by measuring the time series standard deviation.