My aim is to forecast the one-quarter ahead EUR/USD exchange rate. I have constructed a regression model with the following as explanatory variables: exchange rate in the previous quarter, EUR/USD futures price, American and European GDP growth rate, inflation rate, return on stocks, interest rate on 3-month govt. securities, current account balance, net financial account balance and dummies for the inversion of EU and US yield curve. All the explanatory variables are at one-quarter lag. The input data file could be found here.

I am getting spurious results - all these variables are coming out to be insignificant but surprisingly the R^2 is more than 90%.

Is OLS not a good technique for forecasting exchange rates? If not, then which other techniques could be used?

  • $\begingroup$ Have you considered just using the prices on the current 3 month futures contracts? $\endgroup$ Dec 24, 2019 at 6:06
  • $\begingroup$ That is a tell-tale sign of multicollinearity. What do the VIFs say? $\endgroup$
    – ahorn
    Dec 27, 2019 at 9:25

1 Answer 1


There are many methods to forecast exchange rates- look at the series of influential papers by Meese and Rogoff for a primer (https://scholar.harvard.edu/files/rogoff/files/51_jie1983.pdf).

To answer your particular question- what you describe seems to be driven by multicollinearity. For instance, returns on stocks will be correlated with the inflation rate if they are nominal, CAB and FA balance will be related to both the GDP growth rate as well as the interest rates etc. To test whether your results are susceptible to multicollinearity in the variables, try Variance Inflation Factor type tests.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.