Event studies are a useful and tractable method of making inference about the effects of macroeconomic policy on the economy. However, they rely on the identification assumption that the event in question was a surprise to financial markets. If this identification assumption is violated, the results of an event study can be biased. This thesis first discusses the difficulty posed by these identification assumptions, and how they have become more difficult to satisfy with the recent increased use unconventional monetary policy tools (especially asset purchases). The empirical component of this thesis examines the dynamics of a fully unexpected policy surprise in asset markets, to examine if event study practitioners can gain insight from the data when choosing events. If the data can help to identify a fully unexpected policy surprise, selecting events will become easier, and the difficulty of the identification problem will be ameliorated. This work is conducted in a time-varying parameter vector autoregression context, with the application of Bayesian estimation.