Algorithm 2: Best subset selection of M co-integrated nodes from N − 1 candidates for each of N nodes

1: # Search for the best subset of M sensors for each individual sensor, i

2: M ← number of sensors in the subset

3: for each sensor i do

4:    searchspace ← set of all sensors minus sensor i

5:    bestsubset[i] ← NULL

6:    for j = 1 to M do {add one more sensor to best subset for i}

7:     lowest estimation error ← infinity

8:     for each sensor k in searchspace

9:      fit linear model to sensor i using (k + bestsubset[i])

10:      if estimation error from linear model < lowest estimation error

11:       lowest estimation error ← estimation error from linear model

12:       bestsensor ← k;

13:      end if

14:     end for

15:     searchspace ← searchspace − bestsensor

16:     bestsubset[i] ← bestsubset[i] + bestsensor

17:    end for

18: end for