Using Survey Expectations In FX Analyses

by: Econbrowser
Summary

At horizons of up to six months, a variety of "models" or "dynamics" were used (economic fundamentals is roughly interest rates).

At over six months, interest rates and other economic fundamentals dominate, while at short run other dynamics dominate.

Finally, as pointed out in Jeff Frankel's comments during the session, even if survey data are imperfect, we have even more information that the rational expectations hypothesis (forecast errors are true innovations) is also imperfect, imparting a lot of noise in our estimates of expectations.

Editor's note: This article was originally published on July 11, 2019, by Menzie Chinn here.

At the NBER IFM Summer Institute session on exchange rates yesterday, the debate over the use of survey data rekindled. In Exchange Rate and Interest Rate Disconnect, Şebnem Kalemli-Özcan and Liliana Varela used survey data on exchange rate depreciation. The discussant Adrian Verdelhan (MIT) and audience members questioned whether such data actually measured what we thought they measured market expectations.

I was somewhat surprised at this debate. Thirty years ago, the rational expectations hypothesis dominated. Now, survey data on inflation expectations is regularly cited, and used in academic studies. Market practitioners look at confidence indicators as a means of tracking economic activity. A few months ago, the NBER Reporter had an article entitled "The Return of Survey Expectations."

One particularly common refrain is that in forex, the survey respondents either just use interest rate parity, or are so poorly informed that they use a random walk forecast. In my survey of forex traders, Yin-Wong Cheung and I asked how traders formed their expectations. At horizons of up to six months, a variety of "models" or "dynamics" were used (economic fundamentals is roughly interest rates).


Source: Cheung and Chinn (2001).

At over six months, interest rates and other economic fundamentals dominate, while at short run other dynamics dominate.


Source: Cheung and Chinn (2001).

If indeed forex traders merely read off the interest differential their expected depreciation, then in a regression of survey-based depreciation on the relevant interest differential, the coefficient would be one.


Figure 1: OLS regression coefficient of 3-month expected depreciation on 3-month interest rates. Source: Forthcoming revision of Chinn and Frankel (2019).

These estimates are obviously different from the value of unity, and significantly so in the cases of Norway, Britain, Hong Kong currencies (against the dollar).

Finally, as pointed out in Jeff Frankel's comments during the session, even if survey data are imperfect, we have even more information that the rational expectations hypothesis (forecast errors are true innovations) is also imperfect, imparting a lot of noise in our estimates of expectations.

As an example, consider these two measures of the exchange risk premium for the euro.


Figure 2: Excess returns (risk premium using rational expectations hypothesis) (blue) and risk premium calculated using survey data (red), both for euro/dollar at the 3-month horizon. Source: Chinn and Frankel (2019).

If you thought risk premia were persistent and time-varying, I think you'd consider the survey-based version more convincing than the ratex-version…

Editor's Note: The summary bullets for this article were chosen by Seeking Alpha editors.