purchasing power parity, revisited recall that the principle of
TRANSCRIPT
Purchasing power parity, revisited
Recall that the principle of purchasing power parity said that, assuming the model
Average annual change inthe exchange rate = β0 + β1 ×
Difference in averageannual inflation rates
+ random error,
PPP is consistent with β0 = 0 and β1 = 1. We examined that before by looking at two
t–statistics, but that wasn’t quite correct; this really is a partial F–test problem, testing
the hypotheses
H0 : PPP is correct
vs.
Ha : a general linear relationship different from PPP is correct.
The appropriate F–statistic has the form
F =(Residual SSPPP − Residual SSgen)/2
Residual SSgen/(n − 2),
on (2, n − 2) degrees of freedom, where Residual SSPPP is the residual sum of squares
assuming the PPP model holds, while Residual SSgen is the residual sum of squares from
the general regression model (the one based on predicting Exchange rate change from
Inflation difference). The residual sum of squares based on the PPP model is not
obtained from a regression run in this case, since there are no estimated parameters;
rather, PPP implies that the fitted value for Exchange rate change for any case is exactly
Inflation difference, so the residual sum of squares is simply
∑
i
(Exchange rate changei − Inflation differencei)2,
which equals 5.677 for the industrialized country data. Recall the regression results for the
general model:
Regression Analysis: Exchange rate change versus Inflation difference
Analysis of Variance
Source DF Adj SS Adj MS F-Value P-Value
Regression 1 61.850 61.8496 183.49 0.000
Inflation difference 1 61.850 61.8496 183.49 0.000
c© 2015, Jeffrey S. Simonoff 1
Error 14 4.719 0.3371
Total 15 66.569
Model Summary
S R-sq R-sq(adj) R-sq(pred)
0.580583 92.91% 92.40% 90.91%
Coefficients
Term Coef SE Coef T-Value P-Value VIF
Constant 0.086 0.147 0.59 0.567
Inflation difference 0.9024 0.0666 13.55 0.000 1.00
Regression Equation
Exchange rate change = 0.086 + 0.9024 Inflation difference
The residual sum of squares for this model is 4.719, so the partial F–test has the form
F =(5.677 − 4.719)/2
4.719/14= 1.42;
on (2, 14) degrees of freedom, this has a tail probability of .274. Thus, this simultaneous
analysis says that we don’t have evidence that PPP doesn’t hold for the developed countries
(recall that Greece was an unusual country in this analysis, but it turns out that omitting
it doesn’t change this implication; in fact, the evidence against PPP becomes weaker).
We can do a similar calculation for the developing country data (I’ve already omitted
Brazil):
Regression Analysis: Exchange rate change versus Inflation difference
Analysis of Variance
Source DF Adj SS Adj MS F-Value P-Value
Regression 1 1309.89 1309.89 1359.33 0.000
Inflation difference 1 1309.89 1309.89 1359.33 0.000
Error 22 21.20 0.96
c© 2015, Jeffrey S. Simonoff 2
Total 23 1331.09
Model Summary
S R-sq R-sq(adj) R-sq(pred)
0.981645 98.41% 98.33% 98.06%
Coefficients
Term Coef SE Coef T-Value P-Value VIF
Constant -0.641 0.287 -2.23 0.036
Inflation difference 0.9637 0.0261 36.87 0.000 1.00
Regression Equation
Exchange rate change = -0.641 + 0.9637 Inflation difference
The residual sum of squares based on the general model is 21.2; using the PPP model,
it is 26.1 (this is determined by direct calculation). The partial F–test is then
F =(26.1 − 21.2)/2
21.2/22= 2.54;
this has a tail probability of .10, so we see marginal evidence to reject the PPP hypothesis,
which we see comes from the intercept term. Omitting Mexico, however, drops the p-value
of the F -test to less than .01, reinforcing that notions of whether PPP holds or not can be
strongly dependent on accidents of how the “long run” is defined.
c© 2015, Jeffrey S. Simonoff 3