I have one dataset with currency exchange ratios for several years. The exchange ratios are for two Asiatic currencies and the Canadian dollar, all against American dollars. My (perhaps naïve) thinking is that I can observe which economies are more strongly tied by looking to the association strength between the ratios. So, I have got an Excel add-in that can calculate distance correlation and maximal information coefficient, plus with some tweaking it can even give p-values of Pearson. However, when I calculate the three measures for all pairs of variables and their p-values against the null hypothesis that the variables are independent, they give utterly discordant results.

Which measure should I believe better? Is there anything wrong with my analysis?

Screenshot with the problem http://dl.dropbox.com/u/5363697/currency_screeshot.png

It is possible to check both the Excel worksheet and a PDF version of it.

Thanks in advance for any help.

  • $\begingroup$ The link to the plot does not work anymore $\endgroup$ Apr 17 '19 at 6:49

Actually the results look pretty consistent. But based on the plots I do not believe the result for CAD vs CNY. The Pearson correlation looks like it should be practically 0 not 0.57. Also since these are time series these analyses are ignoring any time dependence if there were any. There is no apparent time dependence based on the plots.

  • $\begingroup$ Hi! Thanks for your answer! I will carefully look to that 0.57 value, it is true that it should not be there. How can I take into account the time for this analysis? $\endgroup$
    – dsign
    Jun 7 '12 at 10:59
  • $\begingroup$ You could fit time series models and compute estimates of autocorrelation and cross-correlation. Given the plots though there does seem to be any time dependence. So such analysis may not help. $\endgroup$ Jun 7 '12 at 11:06
  • $\begingroup$ For CAD vs CNY the Pearson Coefficient of 0.57 is correct unless you delete the outliers. $\endgroup$
    – user52729
    Jul 24 '14 at 15:59
  • $\begingroup$ @guest Thank you for correcting this answer. However, in spirit Dr. Chernick is right--as you hint--because the correlation coefficient is a terrible description of this bivariate relationship. The high value is due to a small number of low outliers; the remaining data exhibit little correlation. $\endgroup$
    – whuber
    Jul 24 '14 at 16:18

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.