@Jeff. Sorry but I can only but disagree.

In your statement "1/ Sentiment can only be measured by price" you are simply repeating what you've been taught, and most probably you never looked at any data produced by us or some of our competitors.

Price based sentiment indicators are perfectly valid, but we offer alternative approaches. To put it differently, we consider that available information is not always discounted efficiently into the market prices. I see plenty of examples every day. Precisely because markets are complex animals.

With statement 2/ you seem to indicate that you are an advocate of the random walk hypothesis, which is a highly debated one. And clearly not one I follow.

3/ It is in fact possible to derive profitable algorithmic strategies out of dumb signals - be it Twitter sentiment, or simple moving averages - through for instance Machine Learning - I'd be happy to share more details to you if need be. An alternative way to look at it is that we can indeed score some of the noise. And those noise levels indicate and correlate quite well with areas of volatility.

For the rest, and as we stated previously in that stream of discussion, they are many other fields of application of sentiment analysis technics.

I truly believe that many market professional today don't understand what professionals in the market sentiment place are trying to do simply because they were not taught at school and/or because they do not understand the technology.

This is a genuine concern, a challenge that we face, but that does not negate the validity of our approach.

To take an analogy, consider Black and Scholes in the 70's. Was it right? No. Was it perfect? No. Yet it became a standard.

We are trying to do the same thing in our space: set standards to assess and rate unstructured information be it in Twitter, News, Social Media etc.

Is it to be followed blindly? No Is it perfect? No. Yet it has tremendous potential added value.

Please look at our data before criticizing.

@Frederic

Thank you for disagreeing with me and for debating my points.

1)I would argue that just as the available information is not always discounted efficiently into market prices, I would also agree that available information is not always discounted efficiently into sentiment. Case after case, individuals who trade based on their own individual sentiment would admit that their "sentiment" caused them to buy when they should have been selling. Therefore, if you aggregate sentiment among equally uninformed market participants, then market participants as a whole would also admit that sentiment was wrong. When market participants realize that their sentiment was wrong (in hindsight), their sentiment quickly changes.

2) No, I do not advocate the random walk hypothesis. What I meant is that no discrete (isolated) event or measure can be correlated with price. Only the aggregate of all events and measures - known and unknown - will correlate with price.

3) Although I have not seen the data myself, others on this thread have said that sentiment correlates with high levels of transaction volume and volatility. What does that tell us about sentiment and the predictability of price? Transaction volume represents purely neutral sentiment, because that means a buyer sentiment caused a purchase and a sellers sentiment caused a sale. The direction of price after the transaction only reveals which transactor's sentiment was wrong.

I am not familiar with the data that your company produces, so cannot comment and would not criticize.

But I am interested. How could your company's sentiment data have been used to predict the September 18 market decline in the S&P? How could your company's sentiment data have been used to predict the spike in 30 year Treasury bond prices last week? Can the data quantify the predicted price movements and their time frame? Based on the current data, which direction will the S&P and 30 year Treasury prices go next? How far up or down will they go?

Nice Jeff, very nice! I especially liked the content of your first paragraph. Collective sentiment is very similar to crowd mentality. But I can see someone might say we study sentiment to trade against the crowd not with the crowd.

@Muhammed

I think that you hit the nail on the head - we should attempt to trade with the ones that are trading against the crowd.

So if there are two types of sentiment data - the sentiment data of the crowd and the sentiment data of those that trade against the crowd, how can sentiment data be parsed and segregated? Wouldn't it be easy to corrupt or manipulate sentiment data so that those that think that they are trading with the ones that are trading against the crowd are actually trading with the crowd? Is published sentiment data regulated?

Sentiment by definition needs large numbers which is equivalent to a crowd of people. Traders make profit by buying before the crowd or at the early stages of crowd buying, and sell before or at the early stages of crowds selling.