Explanation: Noise

1. the constant + noise hypothesis doesnt work -- even a very small amount of noise will cause correlations to go to zero.

e.g. t1 = 0.5 * ones (1000,1) + 0.00005 * randn(1000,1) t2 = 0.5 * ones (1000,1) + 0.00005 * randn(1000,1)

corrcoef(t1,t2) --> close to zero

2. it is independent of how many samples we take (i.e. the timescale)

e.g., changing 1000 --> 10000 (or to 100/10) in the above example doesnt matter.