@Grant, the real question is: What has sustainable value in our “short-term” trade decision making?
We could choose any combination of stocks from its humongous tradable universe based on whatever criteria. But, we have this little mathematical problem, we want the stocks we pick to really outperform the market averages in some manner. That it be by outperforming over the long term or by trading on all its price swings for whatever reasons we might devise using anything we find of significance.
The decision making is different when considering short-term price moves. Short-term, there is a lot of randomness in the making of price movements. It needs to be accounted for and dealt with. I consider the degree of randomness in price movements plays a major role in the mechanics of the short-term trade.
Like, I accept a high degree of randomness in stock prices (even 95%+). Evidently, it has its set of implications. For one, I might not know what the price of any stock might be tomorrow (even if the price might be biased upward over the long term). This, whatever criteria I might want to devise or adhere to from some other source.
I opted to just make the bet based on an “excuse” of a parameter that I identify as possibly a medium to long-term drift. I don't even ask it to be precise, just to have been there in the past. It is after taking the bet that I will manage and game the trade.
Say stock prices are “totally” random, or almost like it. Then, the wisdom of crowds, machine learning, deep learning, artificial intelligence, indicators, parameters, factors, residuals, principal component analysis, wavelets, multiple regressions, quadratic functions, and on and on, cannot help you decipher a game approaching a heads or tails type of game where upcoming odds and biases even change all the time.
The pure randomness of it all precludes finding anything significant in [sentdex.sentiment_signal] for instance, or in [stocktwits] for that matter. Oh, you could always find something on some past, smoothed, interpreted, tailored and doctored sentiment data. But, going forward that is not the data that will be presented. You will have a new ball game. What was there at some time might not be there the next and there is no way to figure out in advance which is which.
No one on this planet, using whatever tools they had at their disposal, ever beat the game of heads or tails except by pure luck of the draw. Sometimes you win, sometimes you lose, but it does not change your average expectation which remains zero for every long-term game you take. No expected gain, but also no expected loss.
However, the nature of the game changes if you add some long-term drift. It makes it an upward biased game where you can win all the time just by sitting around for a long time.
So, short-term, how should you trade? What kind of decision process will fit your temperament, your style and your understanding of the game?
Say, you can identify the best cluster of similarly behaving stocks. These would be the highest performers in risk-return space. Shouldn't most of the efforts be concentrated on trading methods that would enhance their respective performance even more?
Note that by changing the set of factors, your clusters will be moving around in risk-return space. And, also, clusters do move around in time relative to others, disassemble, rearrange, even morph as if in other sectors. All things I consider, and more, in the design of a trading strategy.