Assume one has trained different sentiment analysis models that assigns sentiment scores to the financial news or documents. How would one should approach testing the different models and decide which one is better. Are there any standard practices? The most naïve approach would be to collect historical news data and find the correlation (Pearson correlation) of the news sentiment score with the stock price (or maybe daily/quarterly price change rather than the price itself) and the one with the higher score is probably the better model. Is this naïve approach good enough? (One pitfall of this approach is the sentiment is one factor on the price movement and I am not sure how meaningful these calculated correlations are.)

  • 1
    $\begingroup$ "explaining" in a statistical sense the price change (or return) not the price level is definitely the way to go. $\endgroup$ – noob2 Apr 30 at 18:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.