Abstracts

Most papers fitting Hawkes processes to financial data do not obtain statistically significant results. Significance on the activity of EBS limit order books cannot be achieved on mid-price changes. We argue that this is due to the limited time resolution (0.1s) of such data that prevents many events to be detected. This limitation can be self-consistently lifted for transactions on the bid and ask sides by using the volumes of transactions on both sides during a time slice. Fitting one hour of activity yields fits that pass Kolmogorov-Smirnov tests provided that at least two exponentials are used; all hours of the day have an endogeneity factor of about 0.7. We are able to fit accurately single days by accounting for the intra-day variability of activity, which however suggest larger endogeneity factor (0.8). We therefore argue that seasonalities and limited time resolution are major obstacles when fitting Hawkes processes to financial market data.