Over the past three decades, the number of papers published on Load Forecasting (LF) has increased exponentially, largely due to the advance of Artificial-Intelligence/Machine Learning techniques. Most research has focused on short-term load forecasting (STLF), hours or days in advance. The rise of the Smart Grid and Microgrid concepts require load demand control at shorter lead times, at a resolution of minutes, leading to the need for Very Short Term Load Forecasting (VSTLF). There is not a significant body of research on this topic. Additionally, attention needs to be paid to small power systems, far smaller than those studied in most of the literature. Previous work has used statistical techniques to characterize power systems and studied univariate methods for accurate VSTLF. This study builds upon the previous research and investigates the relationship between system characteristics and the achievable VSTLF accuracy. The results presented here are based on study and simulated forecasting of three years' worth of real load data obtained from the New York Independent System Operator (NYISO).