Greg Goodman suggested that I should add a better low-pass filter than the ordinary moving average.
Different digital low-pass filters are described on wiki here
I chose a so called Blackman filter, which seems to behave very well. The better performance of a good designed low-pass compared with moving average is evident in this plot of RSS data:
We see here that the moving average, in red, has tops just before 1990, in 2000 and around 2008. The raw data has low values in the same places.
These tops are all artefacts due to the effects of high values on each end of the averaging interval.
The low-pass filter, in blue, gives a more real representation of the raw data.