He is just saying that you need an adequate number of samples but that you also need a wider sample interval which is true of any stochastic process.The number of data points is more important. Let's say your test horizon is 12 years. Well that's only 144 data points, if you go by months. 6000 is better.