Handbook of Exchange Rates

20.1 Introduction

The availability of large amounts of tick-by-tick data, in excess of 50,000 data points per day (Glattfelder et al., 2010), oanda, ebs has opened up new opportunities for model building. It is now possible to follow an empirical approach and develop models bottom up by analyzing empirical data and searching for statistical properties.

The analysis of high frequency data is nontrivial: ticks (i.e., quoted prices) are irregularly spaced in time in an intricate sequence. The available literature essentially suggests two ways to handle this issue (Dacorogna et al., 2001; Engle and Russell, -NIL-). The first method suggests aggregating price information by interpolating prices between fixed and predetermined times. The drawback of this method is the loss of intratime information during active periods and the multiplication of price information during quiet periods, when insufficient data is available. With the second approach, one can consider a time series made of ticks and times between their occurrences (i.e., duration); this is referred to as a point process (Bauwens and Hautsch, 2009). Point processes are valuable because they incorporate durations and allow analytical results to be derived; however, they have the disadvantage that time is measured in terms of physical units, and therefore, point processes neither adapt to the changing market activity nor differentiate between a minute of early morning calm and a minute during a hectic news announcement.

We propose ...