A Filtering Process to Remove the Stochastic Component from Intraday Seasonal Volatility
Author | Robert T. Daigler,Jang Hyung Cho |
DOI | http://doi.org/10.1002/fut.21585 |
Published date | 01 May 2014 |
Date | 01 May 2014 |
AFILTERING PROCESS TO REMOVE THE
STOCHASTIC COMPONENT FROM INTRADAY
SEASONAL VOLATILITY
JANG HYUNG CHO and ROBERT T. DAIGLER*
The intraday seasonal variance pattern contains stochastic as well as deterministiccomponents.
Therefore, the estimation of information arrivals in the associated volatility process requires the
proper filtering of both of these seasonal components. However, popular current models remove
only the deterministic part of the typical U‐shape volatility. Here, we provide the first empirical
results of the importance of the stochastic component, as developed by Cho and Daigler (2012).
We show that a highly significant additional 8.5% to 12.9% of the total seasonal variance is
explained by the stochastic seasonal variance component for S&P500 futures, live cattle futures,
and the Japanese yen‐U.S. dollar spot exchange rate. Moreover, we show that the stochastic
seasonal filtering model implemented here does not create any statistical distortions of the
filtered series, as occurs with deterministic‐based seasonal adjustment processes, as well as
comparing the model examined here with the most popular current deterministic model. As part
of our analysis we examine the application of the model to macroeconomic news and out‐of‐
sample results for the model. © 2012 Wiley Periodicals, Inc. Jrl Fut Mark 34:479–495, 2014
1. INTRODUCTION
An understanding of the volatility process of a time series is important, because the volatility
process reflects the impact of information arrivals (Andersen & Bollerslev, 1997b; Gross-
man, 1976), as well as impacting the price of derivative instruments. However, it is difficult to
analyze the volatility arising from intraday information events because the informational
volatility is obscured by the inherent (daily) seasonal volatility component. In particular,
empirical findings show that a consistent daily U‐shaped seasonal volatility pattern exists for
various financial markets (Wood, McInish, & Ord, 1985; Lockwood & Linn, 1990; Daigler,
1997). Consequently,it is crucial to employ a proper seasonal adjustment procedure to identify
the size of both the seasonalized and the deseasonalized volatility components, especially
surrounding an information event.
Several methods exist to filter the seasonal variance components from the total variance
process, as discussed in section 2. We classify these methods into deterministic and stochastic
Jang Hyung Cho is an Assistant Professor of Finance, Department of Accounting and Finance, BT 854,
College of Business, San Jose State University, San Jose, California. Robert Daigler is Knight Ridder Research
Professor of Finance, Chapman Graduate School of Business, Florida International University, Miami,
Florida. We would like to thank Paul Pfleiderer for his encouragement and Zhiyao Chen and Zhiguang Wang for
suggestions on earlier versions. We also wish to thank the reviewer, whose extensive comments substantially improved
the quality of the study. An earlier draft of this study was presented at the Financial Management Meetings.
*Correspondence author, Department of Finance, RB 206, College of Business, Florida International University,
Miami, FL 33199. Tel: 305‐348‐3325, Fax: (305) 348-4245, e‐mail: daiglerr@fiu.edu
Received November 2011; Accepted September 2012
The Journal of Futures Markets, Vol. 34, No. 5, 479–495 (2014)
© 2012 Wiley Periodicals, Inc.
Published online 26 November 2012 in Wiley Online Library (wileyonlinelibrary.com).
DOI: 10.1002/fut.21585
seasonal adjustment procedures. The deterministic seasonal adjustment models, such as the
flexible Fourier form model (FFF), are widely used in empirical analysis, as explained below.
However, the intraday seasonal volatility pattern also possesses stochastically time‐varying
components, causing these stochastic seasonal components remaining in a deterministic only
seasonally adjusted series to create statistical distortions. Therefore, one must filter out the
stochastic components as well as the deterministic components from an intraday volatility
series. Given the importance of filtering both types of seasonal variance components, it is
surprising that no models or empirical studies exist that examine the stochastic seasonal
variance component. In fact, important economic and statistical reasons exist to adjust for the
significant seasonal component inherent in the U‐shape volatility curve. In particular,
Granger (1976) points out that the seasonal component obscures the movements of
economically important variables, as well as making it difficult to determine local volatility
trends. In addition, the seasonal component causes nonstationarity (Lutkepohle, 2007) and a
spurious relation between two time series if they share a common seasonal component
(Granger, 1976; Hylleberg, 1986).
This study estimates the stochastic seasonal variance components in a financial time
series, as well as separating the stochastic seasonality from the deterministic seasonality. In
addition, we compare the seasonal adjustment performance of the stochastic filtering
model presented here to the most commonly used deterministic model, namely the FFF
model. In order to estimate both the stochastic and deterministic seasonal variance series
simultaneously, we employ the Autoregressive Conditional Seasonal Variance (ARCSV)
theoretical process developed by Cho and Daigler (2012). Unlike other seasonal adjustment
procedures, the ARCSV model captures the stochastic (as well as the deterministic)
components in the seasonal pattern by assuming that the intraday seasonal variance in each
intraday time interval follows a unique autoregressive moving average (ARMA) process.
Moreover, Cho and Daigler (2012) prove that the ARCSV model does not impound statistical
distortions into the seasonally adjusted series from the filtering process, as found in other
filtering methods.
This study adds to the literature by (1) showing how to implement and use the Cho and
Daigler (2012) model for filtering seasonality, (2) empirically determining the amount of
variance that is stochastic, as well as deterministic portion, in validating the ARCSV model, (3)
illustrating by direct comparison how the ARCSV model is superior to the flexible Fourier form
(FFF) model, (4) applying the ARCSV model to macroeconomic news and providing out‐of‐
sample results, and (5) thereby showing how the AACSV model is an appropriate tool to
analyze the volatility components of high frequency data.
Our results provide the first empirical evidence for the ARCSV model, showing that the
stochastic component accounts for an average of 10.14% of the total seasonal variance
processes for the S&P500 and live cattle futures and the yen versus dollar exchange rate
(specifically 9.04%, 12.90%, and 8.48%, respectively). We also provide a unique finding that
the importance of the stochastic seasonal variance component increases when the effects of
macroeconomic news are incorporated into the filtering model. Testing for potential
distortions using spectral tests shows no statistical distortions in the resultant filtered time
series. We also provide the first empirical evidence that nonstochastic modifications to
deterministic filtering models (without including a seasonal shock term to capture the time‐
varying seasonal patterns) leads to statistical distortions in the resultant series. In particular,
the FFF model creates statistical distortions for the filtered time series if the interaction terms
between the daily volatility and the seasonal sinusoid terms are included to capture the time‐
varying seasonal pattern. Furthermore, purely deterministic filtering models that do not
include interaction terms (the FFF model in Andersen and Bollerslev (1997a) can be modeled
without these terms) and the variance mean filtering model in Engle, Sokalska, and Chanda
480 Cho and Daigler
To continue reading
Request your trial