In a series of papers based on analogies with statistical physics models, we have proposed that most financial crashes are the climax of so-called log-periodic power law signatures (LPPS) associated with speculative bubbles. In addition, a large body of empirical evidence supporting this proposition have been presented Along a complementary line of research, we have established that, while the vast majority of drawdowns occurring on the major financial markets have a distribution which is well-described by a stretched exponential, the largest drawdowns are occurring with a significantly larger rate than predicted by extrapolating the bulk of the distribution and should thus be considered as outliers. Here, these two lines of research are merged in a systematic way to offer a classification of crashes as either events of an endogenous origin associated with preceding speculative bubbles or as events of an exogenous origin associated with the markets response to external shocks. We first perform an extended analysis of the distribution of drawdowns in the two leading exchange markets (US dollar against the Deutsmark and against the Yen), in the major world stock markets, in the U.S. and Japanese bond market and in the gold market, by introducing the concept of coarse-grained drawdowns, which allows for a certain degree of fuzziness in the definition of cumulative losses and improves on the statistics of our previous results. Then, for each identified outlier, we check whether LPPS are present and take the existence of LPPS as the qualifying signature for an endogenous crash: this is because a drawdown outlier is seen as the end of a speculative unsustainable accelerating bubble generated endogenously. In the absence of LPPS, we are able to identify what seems to have been the relevant historical event, i.e., a new piece of information of such magnitude and impact that it is seems reasonable to attribute the crash to it, in agreement with the standard view of the efficient market hypothesis. Such drawdown outliers are classified as having an exogenous origin. Globally over all the markets analyzed, we identify 49 outliers, of which 25 are classified as endogenous, 22 as exogenous and 2 as associated with the Japanese anti-bubble. Restricting to the world market indices, we find 31 outliers, of which 19 are endogenous, 10 are exogenous and 2 are associated with the Japanese anti-bubble. The combination of the two proposed detection techniques, one for drawdown outliers and the second for LPPS, provides a novel and systematic taxonomy of crashes further substantiating the importance of LPPS.