We study the long-term (≈30 years) radio variability of 43 radio bright AGNs by exploiting the data base of the University of Michigan Radio Astronomy Observatory (UMRAO) monitoring program. We model the periodograms (temporal power spectra) of the observed lightcurves as simple powerlaw noise (red noise, spectral power P (f ) ∝ f −β ) using Monte Carlo simulations, taking into account windowing effects (red-noise leak, aliasing). The power spectra of 39 (out of 43) sources are in good agreement with the models, yielding a range in power spectral index (β) from ≈1 to ≈3. We fit a Gaussian function to each flare in a given lightcurve to obtain the flare duration. We discover a correlation between β and the median duration of the flares. We use the derivative of a lightcurve to obtain a characteristic variability timescale which does not depend on the assumed functional form of the flares, incomplete fitting, and so on. We find that, once the effects of relativistic Doppler boosting are corrected for, the variability timescales of our sources are proportional to the accretion rate to the power of 0.25 ± 0.03 over five orders of magnitude in accretion rate, regardless of source type. We further find that modelling the periodograms of four of our sources requires the assumption of broken powerlaw spectra. From simulating lightcurves as superpositions of exponential flares we conclude that strong overlap of flares leads to featureless simple power-law periodograms of AGNs at radio wavelengths in most cases. Subject headings: Galaxies: active -radiation mechanisms: non-thermal -methods: statistical 4 We use the term radio bright AGNs because not all of our sources might be radio loud, i.e., have a radio-to-optical flux density