US interest rates'overnight reaction to macroeconomic announcements is of tremendous importance trading fixed income securities. Most of the empirical studies achieved so far either assumed that the interest rates' reaction to announcements is linear or independent to the state of the economy. We investigate the shape of the tern structure reaction of the swap rates to announcements using several linear and non-linear time series models. The empirical results yield several not-so-well-known stylized facts about the bond market. First, and although we used a daily dataset, we find that the introduction of non linear models leads to the finding of a significant number of macroeconomic figures that actually produce an effect over the yield curve. Most of the studies using daily datasets did not corroborate so far this conclusion. Second, we find that the term structure response to announcements can be much more complicated that what is generally found : we noticed at least four types of patterns in the term structure reaction of interest rates across maturities, including the hump-shaped one that is generally considered. Third, by comparing the shapes of the rates' term structure reaction to announcements with the first four factors obtained when performing a principal component analysis of the daily changes in the swap rates, we propose a first interpretation and classification of these different shapes. Fourth, we find that the existence of some outliers in the one-day changes in interest rates usually leads to a strong underestimation of the reaction of interest rates to announcements, explaining the different results obtained between high-frequency and daily datasets : the first type of study seems to lead to the finding of fewer market mover announcements.