A follow up on the comments I made two months ago on the high levels of the reputation scores in general and how it might be a bad time to stake. Due to limited data at the time, my estimate suggests that 0.056 is the steady state level at least for the nasdaqjockey model, and reputation at the time is more at the 70 percentile estimate. Staking NMR while the reputation is higher than usual can be bad because at some point the score will fall and you will lose a good part of your stake.
Reputation has moved even higher since, up to late January of 2020. So the estimate using limited data might have been too low. Rather than using everything's data, I continue to think that it's better to use your own data and estimate the reputation curve for your own model. For example, the top staked model as of March 1 still has a high reputation of 0.148 (wilfred), a model that was ranked 153 on December 31, 2019.
Since I estimated the curve back in December 31 on nasdaqjockey's model data, I will focus on that. The model's reputation change is noted in the figure above. It had a high of 0.19 on January 25, and since then has fallen to a low of 0.055 reputation on February 26. All of this blurbish is to say that there is some use to estimating the reputation curve, and it should factor into your decision to add/subtract stakes.