A common joke among economist is: Why has god created meteorologists? To make the forecasts of economist look less bad! At the heart of this joke stands the critique that economic forecasts are notoriously inaccurate. Prediction Markets are an attempt to improve these forecasts by aggregating the knowledge of many. The present article takes a closer look at these Prediction Markets. By analyzing the existing literature in terms of the relevant theoretical as well as empirical basis, it is shown that an adapted version of the model by Kyle (1985) with noise and insider traders is able to explain the high degree of predictive accuracy, i. e. informational efficiency, of prediction markets. At the same time such a model is able to cope with the Grossman-Stiglitz Paradox (1976) or the No-Trade Theorem (Milgrom & Stokey, 1982), both are common theoretical arguments against informational efficiency. This allows the interpretation of market prices as event probabilities. Even though some empirical artefacts (e. g. the favorite-longshot bias) exist and more research, especially in terms of prediction markets covering economic events, is needed, the overall verdict on these forecasting tools has to be that they are roughly semi-strong efficient. They hence provide an interesting, very accurate and additional tool in forecasting.