Journal of International Money and Finance (1993), 12, 543-560 The sources of GARCH: empirical evidence from an intraday returns model incorporating systematic and unique risks PAUL A. LAUX AND LILIAN K. NG* Department of Finance, CBA 6.222, University of Texas at Austin, Austin, TX 78712-1179, USA GARCH models propose that volatility is time varying and persistent. As a parsimonious statistical description of the process driving returns in many financial markets, these models have been very successful. However, GARCH lacks a substantiated economic motivation. This paper provides new and more definitive evidence on the mixture of distributions hypothesis, a prominent potential economic explanation. The mixture hypothesis posits that autocorrelation in the time-varying rate of information arrival leads to the volatility dependencies captured by GARCH models. Previous investigations of this hypothesis have been inconclusive. This paper makes use of a more theoretically appealing risk specification, more appropriate measures of the rate of information arrival, and higher frequency data from the currency futures market. The results show that the mixture hypothesis explains substantial portions of contracts' unique risks, but systematic GARCH remains. Judging by the findings in voluminous recent literature, Generalized Autoregressive Conditional Heteroscedasticity (GARCH) is ubiquitous? Its key characteristics--autocorrelated volatility and contiguous periods of volatility and quiescence--as well as associated characteristics such as excess kurtosis, are exhibited by a wide variety of financial time series, including stock, futures and foreign exchange returns. The excellent review paper by Bollerslev et al. (1992) provides many examples and citations. As a parsimonious statistical description of the process driving returns in many financial markets, GARCH modeling is a great success. The lack of a substantiated economic motivation for GARCH in returns is thus lamentable. This paper provides new and more definitive evidence on one prominent potential economic explanation, the mixture of distributions explanation. Various authors, including Diebold (1986), Diebold and Nerlove (1989), and Gallant et al. (1991) have suggested that autocorrelation in the * We appreciate comments of Eric Chang, Yin-Wong Cheung, Prem Jain, Barry Schachter, A.J. Senchack, Richard Sweeney, Bill Taylor and seminar participants at the Texas Finance Symposium, Georgetown University and the Commodity Futures Trading Commission (CFTC). Laux was on leave at the CFTC while part of this work was done, and the CFTC's research support is appreciated. 0261-5606/93/05/0543-18 © 1993 Butterworth-Heinemann Ltd