that is a wonderful question. I encountered similar issues, when I used sentiment analysis for real estate related text documents. The issue with financial related documents is that they are extremely driven by numbers. So a market increase is different to a market increase by 25%. My suggestion would be to run a supervised learning approach on your text, assumed you have labelled data. But if there is another solution I am eager to hear about that.
Interesting question and depending on the context the answer can be yes or no, or positive and negative, as you suggest. As I am in finance I suggest discerning the differences from the following: