| May 23, 2019
In this study, data scientists at RavenPack wanted to see whether news sentiment scores, generated by scanning high-quality financial news articles, could be used as a screener for better-performing stocks.
The expectation was that sentiment could be used to filter out higher-performing stocks with applications both for systematic and fundamental traders.
After experimentation, it was found that those stocks subject to good or bad news flow, and extreme sentiment readings as well (measured by RavenPack’s ‘Sum Excess Sentiment Indicator’ (SESI)), outperformed a control group of random stocks with a neutral sentiment.
The conclusion of the white paper was that the degree of sentiment as much as the direction was the determining factor in screening for high-performing stocks.
“We found that higher quintiles in Sentiment Strength outperform lower quintiles. Encouraged by this result, we want to build dynamic portfolios that only select a small number of equities that rank high in Sentiment Strength,” says Peter Hafez, Chief Data Scientist at RavenPack.
Portfolios optimized in this way achieved high annualized returns of up to 80% and high per trade returns of up to 33bps. They also enjoyed higher Information Ratios (IRs) - a measure of return consistency - of up to 4.0.
When compared to a simple screener based solely on momentum, the RavenPack approach showed 4x higher returns over an 8-year test period.
The team then looked at other ways of optimizing the basic screener model, with some interesting results.
Experiments on portfolio size, for example, suggested smaller selections of stocks with more extreme sentiment enhanced returns significantly compared to larger portfolios, which might contain stocks with milder sentiment scores.
This is evidenced in the charts below, which show cumulative returns by portfolio size (PS) and holding period (HP), in units of 100%.
It was found that there was little point in focusing on specific sectors, for whilst financials provided the best returns, with a key variable being the high volume of news in the sector, the best results came from screening all sectors.
A further test to see if the market cap could provide the basis of enhanced optimization revealed that small-caps significantly outperformed mid or large-cap companies.
Finally, ‘the icing on the cake’, came from adding a further discretionary overlay, which attempted to reflect the skill of an experienced trader in making adjustments to the portfolio selection.
“We will include an additional allocation step that tries to reflect the expertise of the trader, by randomly avoiding some bad performing stocks in the selection process,” says Hafez.
This was done by recreating, in a systematic way, the skill of a trader in weeding out ‘bad apple’ stocks after the initial screen.
RavenPack recreated ‘trader skill’ by using the following day’s performance as the basis of separating stocks into a ‘good’ and a ‘bad’ basket, and then dropping stocks from the ‘bad’ basket with varying degrees of probability depending on level of trader skill, numerically defined as ‘skill probability’.
A trader with ‘zero’ skill, for example, would “select a bad trade 50% of the time and a good trade the other 50% of the time,” says Hafez. “However, if we introduced a skill probability of 25%, it implies that the trader will correctly drop a bad trade 62.5% of the time vs. 37.5% for a good trade.”
The results of adding the discretionary overlay to the stock screening process were positive, although results varied depending on the skill level modeled.
“Our random sentiment portfolios with discretionary overlay already show promising results: if the trader is able to successfully include new valuable information in the stock selection process, a statistically significant boost in performance can be achieved compared to Extreme Sentiment Portfolios (vertical line) across all HPs and PSs. Figure 9 compares the annualized return and Information Ratio distributions for different skill probabilities, built from the stock screen with S = ALL (using all sentiment signals),” says Hafez.
Using trader skill to drop only 2 out of 22 daily trades, for example, can lead to annualized returns and IRs that are considerably higher. A 0.5 skill probability, for example, can increase returns by up to 100%.
The study is not the only one to prove RavenPack can have useful applications for fundamental investors.
Separate research conducted by an independent portfolio and quant research firm showed how news and social media
sentiment could improve the timing of stock picking
, not only in the U.S but also internationally, both within systematic and discretionary methodologies.
Please use your business email. If you don't have one, please email us at email@example.com.
We will process your personal data with the purpose of managing your personal account on
RavenPack and offering our services. You can exercise your rights of access, rectification,
erasure, restriction of processing, data portability and objection by emailing us at firstname.lastname@example.org. For more information, you can
Your request has been recorded and a team member will be in touch soon.
We consider incorporating sentiment signals from news, earnings call transcripts, and insider transactions to
boost the risk-adjusted returns, and revive factor performance.
We find stronger, more predictable market reactions when the words of company executives agree with their actions.
We have gathered 12 insights from 2021 research that can be leveraged in 2022.