Academic journal article Journal of Financial Management & Analysis

Problems Encountered When Using High Frequency Financial Market Data: Suggested Solutions

Academic journal article Journal of Financial Management & Analysis

Problems Encountered When Using High Frequency Financial Market Data: Suggested Solutions

Article excerpt

Introduction

In the last decade frequency data has become available for many financial markets, leading to a very rapid expansion in the number of empirical studies using such data. High frequency data is defined as occurring when the trading process is sampled between the open and close, i.e. intraday observations. Such data offers many opportunities for a more detailed analysis of market activity, but also presents new problems. The new opportunities are still being discovered*, and many are featured in recent publications (see ap Gwilym and Sutcliffe1; Dacorogna, et al2; Madhavan3). The contribution of this paper is the first to systematically document the multitude of problems encountered when using high frequency financial market data**, together with suggested solutions. We focus on recognising and dealing with the new, largely hidden, problems associated with this type of data. In highlighting the many pitfalls encountered, the paper offers a resource for empirical research in financial markets.

In the physical and biological sciences, when the level of analysis is at the molecular or atomic level, phenomena which could safely be ignored at the macro level become important (see McMahon and Bonner). A similar effect exists in financial markets, where market microstructure effects that can be ignored at a daily or lower frequency, can no longer be ignored when using high frequency data. In essence, as the time period over which returns are computed (the differencing interval) is changed from daily to intraday returns are not self-similar fractals, and the "size" of the differencing interval matters. The shift from daily to intraday returns means that microstructure features of a fixed absolute size become of substantial relative importance. In addition, some institutional features only affect returns with a short differencing interval, e.g., trading halts at the open are irrelevant to annual data*. In consequence, as will be argued later, shortening the differencing interval from daily to intraday is quite different from any other shortening of the differencing interval, e.g., from monthly to daily.

Why is High Frequency Data Different ?

Spot, futures and options markets exhibit a very wide variety of institutional features, and these induce patterns in trading activity. There are also strong regularities in these markets for which there is no generally agreed cause. When modelling financial markets, it is sensible to control for the effects of both explained, and known (but unexplained) regularities in the data. However, it is important to.include in the model only those microstructure features or empirical regularities which play a significant role in explaining the dependent variable, and this choice changes with the differencing interval. For example, as the differencing interval shortens from annual to quarterly, monthly, weekly, daily or intraday, the relevant institutional features change, e.g., tax year effects are relevant to annual and possibly higher frequency data, while derivatives contract expiration is relevant for quarterly and possibly higher frequency data. In addition, while many of the causes are still under debate, empirical studies have found evidence of annual, quarterly, monthly, weekly and daily seasonals. Higher frequency patterns and microstructure effects are irrelevant to data sampled at a lower frequency, and so the shorter the differencing interval, the larger is the number of potentially relevant patters and microstructure. However, although many microstructure features operate with a daily frequency, this argument need not lead to a sharp discontinuity between daily and intraday data; merely that more aspects of market microstructure and empirical regularities are potentially relevant to high frequency data. The discontinuity between daily and intraday data is due to the conjunction of three factors :

the linear relationships between the length of the differencing interval and the level of returns and variance of returns

the empirical usage of a limited number of differencing intervals (annual, quarterly, monthly, weekly, daily and intraday)

the small absolute size of the microstructure effects and regularities in the data. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.