Profile of a Sentiment Trader

December 01, 2020

RavenPack data is used in an increasing number of academic fields for research work that involves quantifying market sentiment. The article below highlights one particular study from 2020 that makes original use of our data as part of a social profiling exercise.

Whilst it might seem a little passe to trade the news nowadays, with black box automation making the practice largely redundant, it is surprising how many retail traders still jump on the back of breaking news story to try to profit from the volatility. A recent research study from a student at the Massachusetts Institute of Technology (MIT) sought to use RavenPack news analytics data to profile these sorts of “sentiment traders” and generate models with the foresight to predict their next move.

How much can be inferred from observing the transactions of a retail investor? As it turns out, quite a lot - to the point of predicting their next brokerage order. According to one particular study from 2020, that is, which makes use of our data as part of a social profiling exercise. It is just one example of many studies making use of our data in a growing number of academic fields.

“All too often studies consolidate and conceptualize investors as a “crowd” or single entity that moves the markets, many also tend to focus on the impact of sentiment on actual asset prices, without considering the actors involved,” says Sophia Luo, the author of the thesis.

Luo sought to define predictive characteristics for sentiment traders, such as age, job, number of dependents, investment experience, investment knowledge, and marital status.

Trading Numbers

The first step in the profiling process was to use the demographic and self-reported trading data from 63,814 brokerage accounts between 2003 to 2015 and cross-reference that with RavenPack’s news analytics dataset.

The RavenPack platform is useful in quantifying sentiment as it scores the “positivity” of financial news events by scanning a wide range of quality online news sources using the latest natural language processing and machine learning techniques.

The study went on to isolate days where “sentiment events” occurred -- substantial changes in sentiment as highlighted by the RavenPack data -- and cross-referenced these to investor positioning using the brokerage account data, to highlight trader responses.

Based on the results each of the 63,814 eligible accounts was classified as either belonging to a sentiment trader or not, according to whether they had shown more sentiment trades than non-sentiment trades.

They found that of all the accounts in the sample a subset of 9,010, or 14.1%. met the criteria for belonging to “sentiment traders”.

Profile of a Sentiment Investor

So what if any were the defining characteristics of this subset of traders?

Careful analysis of the 9,010 subset found several demographic characteristics.

Firstly they tended to have good jobs: financial professionals, executives, managers, business owners, real estate workers, CPAs, attorneys, retired, skilled laborers (e.g. scientists, government workers, engineers, and paralegals, etc.), self-employed, consultants, medical professionals and physicians.

Secondly, they tended to self-report their level of trading experience as either “Good” or “Excellent”

Sentiment Investor

When it actually came to investment knowledge, however, they were more likely to report having “Limited “ knowledge, although they were also less likely to report having “None” whatsoever, compared to non-sentiment investors.

Thirdly, sentiment investors were more likely to report information about the number of their dependents - and most sentiment investors had zero dependents.

They were more likely to be either divorced, married, single, or widowed, and less likely to be separated, unmarried or a minor.

Finally, sentiment traders were more likely to be male than female.

Predicting Reactivity

Next, the researcher analyzed whether different profile characteristics could be used to help predict how sentiment traders would react to a given sentiment event.

A whole host of variables were tested for predictability, including demographics, the level of event sentiment itself, whether the event occurred before during or after the financial crisis, whether the event was preceded by a large number of “positive” sentiment events in the previous week and how many events had transpired in the previous month.

Luo found almost all the variables had a high level of predictability for whether a sentiment trader would react to a given event or not.

“We conclude that our models have strong predictive power of whether a sentiment investor will react to a given event. In addition, we find that for almost every coefficient in every model, the coefficient’s average across all simulations is significant at the 1% level. We attribute the large proportion of significant coefficients to the large number of Monte Carlo simulations that we run.” Says Luo.

Despite showing a lower coefficient owing to measurement constraints (being a non-binary demographic variable), age was the most critical demographic variable for deciding whether a trader would react to a sentiment event or not.

It was also found that a sentiment investor was much more likely to react to an event concerning a given asset if the investor already had a history of trading that asset.

“In general, the investor’s age, reporting of having good or limited experience, reporting of having excellent or good investment knowledge, proportion traded of the given asset across all previously traded assets, proportion of sells of the asset made on sentiment, and the fraction of positive events in the week prior to the current data-point have slightly positive effects on the log-odds of reacting to a sentiment event. On the other hand, having three or less dependents, reporting to have excellent investment experience or limited investment knowledge, and being married as well as the event occurring before 2009 have slightly negative effects on the log-odds of reacting to a sentiment event.” Concluded the study.

A Stomach For Risk

Another focus of the study was whether variables could predict how much risk sentiment traders were willing to take when trading a given event.

The magnitude of the sentiment of the event, for example, was a key predictor of the proportion of wealth they were willing to risk.

The proportion of “positive” (sentiment) events leading up to the event was another factor, with a one unit increase in the proportion of positive events in the seven day period before a reaction raising the expected proportion of wealth traded during the event by about 0.016.

Whether the sentiment investor had more or less than three dependents was another determinant, with those having less than three dependents trading a slightly higher-than-expected proportion of wealth.

The study also showed that if a sentiment investor was married they were expected to trade a proportion of wealth that was -0.004 units less than if the investor were single.

Sentiment investors who reported having “excellent” investment experience tended to trade a smaller proportion of their wealth compared to those who reported to have no investment experience.

If an investor had a history of trading an asset on sentiment, they were likely to have a larger reaction to the current event than if they had never traded the asset on sentiment before.

Finally, during the financial crisis, sentiment traders were observed as trading a slightly smaller proportion of wealth compared to events post-financial crisis.

Find out more about using RavenPack data for quantitative research

Request a Trial

Fill out the form below and see RavenPack in action.