R-squared - Believe it or not!

It is not news that news moves financial markets. This blog will publish research on how, when, why, and which news moves what financial markets.

Most people are skeptical when I tell them that up to 80% of the variation in an exchange rate immediately after an announcement can be explained by news (defined as the actual release minus the market expectation derived from a derivatives auction for the announcement). A previous study using survey expectations data and a longer window around the announcement found R²’s of up to 25% (Faust, Jon, Rogers, John H., Wang, Shing-Yi B., Wright, Jonathan H., (2003) “The high-frequency response of exchange rates and interest rates to macroeconomic announcements” U.S. Board of Governors of the Federal Reserve System, International Finance Discussion Papers, number 784.)

By shrinking the window to a minute after the announcement and using derivatives-based expectations gives much bigger news impact coefficients and much higher R²’s. Comparing the size of the news impact with Faust et. al. (2003) the sign of the coefficients is the same (EUR/DM for Faust et. al., EUR for Parker and also GBP). Coefficients are larger as are the R²’s (with the exception of those for Retail Sales - which may suffer from a lack of expectations data as auctions were not held for a six month period between March and August 2004. The same may be the case for the GDP data in my study since there are only a few auctions to date). I attribute this to the more precise announcement window and the better expectations data.

Here’s an example Faust et, al. find the news coefficient for the International Trade Balance announcement on the EUR (DM) to be -10.09 with an R² of 0.24 ( Faust et. al. exchange rate returns are continuously compounded and multiplied by 10,000. The coefficient can be interpreted as the effect of a one unit surprise on the exchange rate in basis points.). I find that the coefficient is as significant (1% level) but the value is orders of magnitude bigger and the R² is 0.78.

Not only that but these results can be improved upon by adding other variables into the model.