The application of sentiment analysis to the financial sector is a field that has been revamped thanks to social media, which has unleashed a trove of data to analyze. In particular, text analysis techniques have benefited from the attention that large part of data science researchers have devoted to it. However, as demographics evolve, so do the communication forms on social media. In particular, the usage of emojis to carry whole concepts is more and more diffused, though research on the topic is lacking. That is exactly the gap that we intend to cover with this contribution. In particular, after collecting more than 18.5 million posts from StockTwits, we use different supervised learning models in order to determine the role of emojis in sentiment analysis of financial posts on social media. We assess model accuracy, training/prediction speed, and sensitivity to training data set size for both emojis-only and text-only data, using logistic regression and BiLSTM models. Our main findings are staggering; we are the first to show that, when training sentiment analysis models exclusively on emojis, compared to a text-only approach: (i) achieved accuracy is competitive; (ii) training is 32 times faster; (iii) prediction times are reduced to a third; and, (iv) 40 times less data is needed to train the model. Additionally, we show some interesting patterns regarding emoji usage in financial microblogs. The cited contributions, other than being interesting on their own, also pave the way for further research in the field.