Happiness is a fundamental societal metric. Normally it is measured through self-reported answers to questionnaires which do not necessarily reflect experienced utility of respondents. In this paper, we examine instead messages posted on the social networking service Twitter, investigating idiosyncratic shocks of happiness. Our data set comprises over 43 millions of tweets posted on a daily basis in all the 110 Italian provinces and collected during 2012. In measuring happiness, we construct an original index using innovative statistical techniques to quantify the level of happiness at the provincial level related to a specific amount of tweets. We then explore the determinants of happiness in our sample. Variables capturing the overall quality of institutions seem to matter only marginally in affecting the level of happiness of the Italian provinces. On the contrary, meteorological variables and extemporaneous events related to specific days, including the variability of the spread between German and Italian Bonds, have the largest impact.