Foreign Exchange Volatility - Effect of Speeches

Jean Claude Trichet In Frankfurt am Main, the first Thursday of each month the President of the European Central Bank gives a press conference in order to say whether or not interest rates have been changed, and why. To test the hypothesis that foreign exchange market volatility is affected by the speech, three databases have been constructed: (i) seventy monthly speeches given since January 1999, (ii) thirty time series of foreign exchange rates between the euro and other currencies, and (iii) intraday future contracts on the EURO/US dollar. The investigation has found evidence of the impact of the speech on the financial markets. Further research aims at improving volatility predictions by including word structures in the model used for forecasting.

Deutsche Borse The results of this investigation are likely to have a number of implications. By taking into account the market perturbations caused by these speeches, it will be possible to evaluate and compare a number of different forecasting strategies. Furthermore the analysis will facilitate principal economists when preparing speeches as they might wish to take into account the fact that putting too much emphasis on certain words in the past have been interpreted with concern by financial markets.

Other investigations, for example [1] and references therein, did not take into account the text itself when constructing their models. In this investigation the information contained within the speech is analysed by means of mutual information and model selection by a superimposed Hidden Markov Model (HMM) in [2]. In addition, a stochastic volatility model driven by a recurrent neural network developed in [3] including word counts and will be tested on intraday data at the Oxford Supercomputing Centre (OSC).

Availability of constructed databases:

European Central Bank press conferences,
Daily foreign exchange time series provided by the European Central Bank, and
Intraday data future contracts on EURO/US dollar from Tick Data.

The software environment is R, linked to C, for computational intensive statistics and Perl, for practical text extraction. The volatility models employed are based on those developed in [4] and [5]. Multivariate analysis techniques have also been employed.

[1] R.S. Gürkaynak, B. Sack, E.T. Swanson (2005), Do Actions Speak Louder Than Words? The Response of Asset Prices to Monetary Policy Actions and Statements. International Journal of Central Banking, May 2005: pp. 55-93.

[2] P.E. McSharry, L.J. Carbonaro (2006). (article in progress).

[3] B.D. Ripley, L.J. Carbonaro (2006). (article in progress).

[4] R.F. Engle (1982), Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of United-Kingdom Inflation. Econometrica, 50(4): pp. 987-1007.

[5] S. Taylor (1986), Modelling financial time series. Chichester [West Sussex] ; New York: Wiley. xvi, 268 p.