Foreign Exchange, Fractional Cointegration and the Implied-Realized Volatility Relation

Abstract

Almost all relevant literature has characterized implied volatility as a biased predictor of realized volatility. This paper provides new time series techniques to assess the validity of this finding within a foreign exchange market context. We begin with the empirical observation that the fractional order of volatility is often found to have confidence intervals that span the stationary/non-stationary boundary. However, no existing fractional cointegration test has been shown to be robust to both regions. Therefore, a new test for fractional cointegration is developed and shown to be robust to the relevant orders of integration. Secondly, employing a dataset that includes the relatively new Euro markets, it is shown that implied and realized volatility are fractionally cointegrated with a slope coefficient of unity. Moreover, the non-standard asymptotic distribution of estimators when using fractionally integrated data is overcome by employing a bootstrap procedure in the frequency domain. Strikingly, tests then show that implied volatility is an unbiased predictor of realized volatility!