The project was a difficult ¡­

The project was a difficult one, but overall I was very satisfied with the answers you gave. In my own evaluation, the great majority got more than 30 out of 50 possible points, some very close to 50.

In the last part of a) it was easy to show that consecutive scores were orthogonal; the general case demanded a more complicated argument. Point b) was easy. Point c) was essentially a question of comparing two different notations. Many of you made point d) correctly.

In point e) the expression for the second regression coefficient was really messy, and to verify the stated formula for this coefficient in point f) was extremely difficult. Yet some of you were able to solve this point.

Most of you did the simulations in point g) OK. In point h) some of you found that the solution m=1 was best, and some found that the solution m=2 was best. It can be shown that when looking at the expected prediction error (a large training set and a very large test set), the solution m=1 will be best. The reason is that in the simulation, there is only one 'relevant component' for prediction: the z which occurred both in x and in y.

Since the project was handed in anonymously, it cannot be returned.