Mean reversion and momentum are the workhorses of quantitative investing. Most systematic trading technologies use a variation of these time-tested techniques. In this article I will be discussing the recent concept of using data from Twitter and the social Internet to augment mean reversion strategies.
A recent paper in Nature entitled Quantifying Trading Behavior in Financial Markets Using Google Trends uses search data from Google to construct an extremely profitable historical trading strategy. Kudos to the authors Tobias Preis, Helen Susannah Moat & H. Eugene Stanley for being one of the first researchers to bring into the mainstream the concept that Google data has predictive signal for the stock market. Mainstream press like Forbes, CNN, BBC, and Time reported on the paper with titles like “Google Predicts the Stock Market”. My firm, AlphaGenius, is predicated on extracting signal from Internet sources like Google and Twitter. As a practitioner, I can confirm the researchers finding that there is signal in Google data. I will be speaking at Quant Invest Chicago in June to discuss this and other interesting topics related to extracting alpha from the social Internet.
I am not going to give a thorough peer review of this paper, but rather, my take on it as a practitioner who has been doing similar studies and trading on them for the past three years.
AlphaGenius is celebrating our one year anniversary of live trading. We launched September 1st, 2011. Our primary strategy has yielded ~19% with very low correlation to the overall market and a Sharpe ratio of 1.7. We spoke at a lot of great conferences and are seeing a growing acceptance that sentiment investing using data from the social Internet should be a core strategy in any diversified portfolio. We have a busy month coming up with speaking engagements at Battle of the Quants in London and Thomson Reuters Hedgeworld in New York.
We also had a nice mention on CNBC today http://www.cnbc.com/id/
Thanks to all of our supporters,
The AlphaGenius Team
Terrapinn, the organizers of Quant Invest, took a video of my Chicago presentation. The video is 30 minutes and does not include the presentation slides. You will see me pointing to the slides off camera. If you are interested the video is below.
A fascinating study on using Facebook to evaluate job candidates is being published in the Journal of Applied Social Psychology that demonstrates that a rule-based analysis of a candidates Facebook page can be a better screener for job applicants than typical interviews, resumes, and questionnaires.
One of the head researchers is Donald Kluemper, a management professor at Northern Illinois University. “I think one of the differences is that you change the frame of reference,” Kluemper said. “You’re asking the rater, ‘Is this person a hard worker?’ On a personality test, the employee would be asked, ‘How hard a worker are you?’ One of the criticisms of self-reporting personality testing is that it can be faked. On a Facebook page, that’s a lot harder to do.”
At AlphaGenius we would wholehearted agree with this logical systematic technique for interviewing. Many “expert” interviewers would claim you need to look at your candidate in the eye to evaluate what kind of worker she will be. I have hired dozens of people over the years, and I can attest that very little can be gleaned from the formal interview to predict that candidates success in that position. Interviewing, like standardized testing, is a skill that can be trained for. There are good interviewers and bad interviewers. Much like standardized testing is only a marginal measure of how someone will perform in college, good interviewing is only a marginal measure of how a candidate will perform over a career. The resume and interview have had decades (centuries?) to evolve into a very well understood and gameable process. The best interviewers get jobs, not the best applicants.
Using the social Internet for interviewing is simmilar in concept to AlphaGenius using the social Internet for building investment models. They both share the same distinct set of advantages.
1. The respondents don’t know they are being measured. They know they have voiced an opinion online, but they don’t know you are using it for interviewing/investing. Therefor, the respondents are very genuine in their response as opposed to a contrived interview or survey. It eliminates biases, gaming, pride and shame inherent in interviews and surveys.
2. There is a lot of data. Facebook profiles have much more data than can be gleamed from a resume and an hour interview. Twitter and
blogs have a ton of data for investing.
3. Not a lot of people are doing it. Interviewing and investing differently from the pack both have their advantages.
This is just further proof that data from the social Internet will fundamentally change the way human decision making will evolve. Many areas of expertise that were previously believed to be the sole realm of human can be replaced by rules and computers. The social Internet is providing a new, unique data-set to guide and give advantages in human to human evaluations: job interviewers, psychologists, investors. I am looking forward to a future study that shows certain peoples’ psychological disorders can be better evaluated from their Facebook profile than a psychiatric session.
If you want a headache, read the comments in an article like this. Note that I am not knocking the article, but the bantering of the commenters.
What I believe at this point is that you cannot know the future of the world well enough to predict future earnings of a company past a few months. The researcher Phillip Tetlock performed a seminal twenty-year study in which 284 experts in many fields, from professors to journalists, and with many opinions, from Marxists to free-marketeers, were asked to make 28,000 predictions about the future, finding that they were only slightly more accurate than chance, and worse than basic computer algorithms.
This is classic Behavioral Economics to have various “logical” commenters vigorously fight over predictions of an unknowable nature. Behavioral Economists would say, they made an instinctive conclusion about the stock/company, then used their logical mind to back-in and justify their instinctive feelings. Humans often confuse intuition for rationality. Humans often feel like they have thought long and hard about a decision when they are really just reinforcing the conclusion they intuitively came to. You will see the same behavior whenever you see two sports fanatics with differing opinions on the same upcoming game. They may as well argue over what the weather will be in 365 days.
Predicting earnings of companies far into the future is a crazy waste of human energy. For further reading, check out what Daniel Kahneman has to say on the subject of flawed intuition: What was I thinking?
I love the Freakonomics book, and I think it is a great introduction to economics. The book really shows how hard numbers can be used to debunk some commonly held biases and heuristics. The author did a nice spot on Yahoo! today to talk about commonly held thoughts about the Superbowl and stocks. If the Giants win it will not be a good year for stocks because they were one of the original NFL teams. The danger when using hard numbers to back-test predictions is that they often lead to spurious relationships. Any time there is a relatively small data set like a yearly event, there will be correlations that are not actually causal. The Freakonomics author does a pretty good explanation in the video, so I will not go into it here. What I will say is that all hope for using statistics is not lost because some relationships turn up spurious. It all comes down to the law of large numbers.
If you flip a coin three times, there is a pretty decent chance that you will get heads all three times. If you flipped the same coin 3,000 times, you will get almost exactly 50% heads and 50% tails. Whenever you hear people or economists (they are different) make predictions, consider if they have enough data. There is nothing wrong with a slight amount of predictive guidance if there isn’t enough data (it is better than nothing), but don’t bet the house on it. Any prediction based on an annual event like the Superbowl will not likely have enough data. Presidential elections are even worse because they only happen every four years.
So, how much is enough data? At AlphaGenius we believe the more the better, but sometimes that is not realistic. As a rule of thumb, and that’s all this is, we do not look at a relationship unless we have over 500 data points. So, that unfortunately rules out predicting Superbowls because there has not been enough of them! Us humans love making predictions though, so I am going to go ahead and say the Patriots will win even though I can’t be statistically certain.
Daniel Kahneman’s new book Thinking, Fast and Slow is a must read for anyone who is interested in sentiment investing. It is not a pure investing book and mostly focuses on psychology. Most of the book focuses on the seminal ideas he won his Nobel Prize for: biases and heuristics. Humans make irrational decisions, especially in matters of money. What I was most impressed with is chapter 21 where he states that “rule based” decision-making is often superior to human expert. Obviously, there are many many investing professionals who take issue with this.
Kahneman gives very detailed and methodical reasoning and evidence to prove the case for “rule based” decision-making and why experts are often over rated. Philip Tetlock’s twenty-year study in which 284 experts in many fields, from professors to journalists were asked to make 28,000 predictions about the future, finding that they were only slightly more accurate than chance, and worse than basic computer algorithms.
Kahneman also gives a fun story about Orley Ashenfelter, a Princeton professor who came up with a simple rule based model for Predicting the Quality and Prices of Bordeaux Wines. The wine community is still in denial that their fine training is inferior to an algorithm.
Investing rules are important. You don’t want to mess up your interpretation of sentiment from news, blogs, Twitter, etc. for investing with your own biases and heuristics.
Investors should use Twitter and the social Internet as a tool to help make investment decisions. The social Internet has the major advantage of providing the collective consciousness of the world. For the first time in human history the well proven concepts of behavioral economics can be applied to mass psychology to help predict securities prices. We believe that behavioral investing will some day be as ubiquitous as fundamental and technical analysis.
Websites like StockTwits.com are free and can be used by any investor to supplement their investment thesis. AlphaGenius is a quantitative technology company that applies econometrics and machine learning algorithms in order to build investing models. The social Internet is a rich data source for fundamental, technical, and quantitative investors.