Understanding the dynamics of extreme observations, so-called spikes, in realtime electricity prices has a crucial role in risk management and trading. Yet the contemporaneous literature appears to be at the beginning of understanding the different mechanisms that drive spike probabilities. We reconsider the problem of short-term, ie, half-hourly, forecasts of spike occurrence in the Australian electricity market and develop models, tailored to capture the data properties. These models are variations of a dynamic binary response model, extended to allow for regime-specific effects and an asymmetric link function. Furthermore, we study a recently proposed approach based on the autoregressive conditional hazard model. The proposed models use load forecasts and lagged log prices as exogenous variables. Our in-sample and out-of-sample results suggest that some specifications dominate and can therefore be recommended for the problem of spike forecasting.