From CFAI book pg463 Question 10.
If exchange rate JPT/USD is random walk function. The answer mentions that the change in exchange rate is not predictable. (3rd last line)
from book “We have concluded that the differenced regression is the model to choose. Now we can see that we would have been seriously misled if we had based our model choice on an R2 comparison. In Table 9, the R2 is 0.9902, whereas in Table 10 the R2 is 0.0026. How can this be, if we just concluded that the model in Table 10 is the one that we should use? In Table 9, the R2 measures how well the exchange rate in one period predicts the exchange rate in the next period. If the exchange rate is a random walk, its current value will be an extremely good predictor of its value in the next period, and thus the R2 will be extremely high. At the same time, if the exchange rate is a random walk, then changes in the exchange rate should be completely unpredictable. Table 10 estimates whether changes in the exchange rate from one month to the next can be predicted by changes in the exchange rate over the previous month. If they cannot be predicted, the R2 in Table 10 should be very low. In fact, it is low (0.0026). This comparison provides a good example of the general rule that we cannot necessarily choose which model is correct solely by comparing the R2 from the two models.”
but Xt - Xt-1 = Et where Et IS PREDICTABLE AND CONSTANT hence covariance stationary. what am i missing?