We reveal properties of electricity spot prices that cannot be captured by the statistical models, commonly used to model financial asset prices, that are increasingly used to model electricity prices. Using more than eight years of half-hourly spot price data from the New Zealand Electricity Market, we find that the half-hourly trading periods fall naturally into five groups corresponding to the overnight off-peak, the morning peak, daytime off-peak, evening peak, and evening off-peak. The prices in different trading periods within each group are highly correlated with each other, yet the correlations between prices in different groups are lower. Models, adopted from the modeling of security prices, that are currently applied to electricity spot prices are incapable of capturing this behavior. We use a periodic autoregression to model prices instead, showing that shocks in the peak periods are larger and less persistent than those in off-peak periods, and that they often reappear in the following peak period. In contrast, shocks in the off-peak periods are smaller, more persistent, and die out (perhaps temporarily) during the peak periods. Current approaches to modeling spot prices cannot capture this behavior either. (C) 2007 Elsevier Ltd. All rights reserved.