September, 2019 / Issue 2 - Editorial on nontraditional data for Israeli Investors
In this post we discuss how overlaying sentiment to a fundamental valuation model, which identifies undervalued U.S. large-caps, can improve overall returns (+1.6/2.4pp* per annum)
*pp - percentage points (absolute delta in %-returns)
Market
Country/Region: North America, USA
Asset Class: Equities
Investment Universe: Russell 1000
Research & Application Context
Alt. Data Factor: Trailing Sentiment
Method of Data Integration: Overlay
Underlying Inv. Strategy Type: Fundamental
Period of Analysis: 01/01/2002-01/07/2018
Investment Direction: Long
Avg. Holding Period: month(s)
Avg. Portfolio Size: 140
Investment Process
Underlying investment logic
The original strategy follows a fundamental valuation framework that tries to gauge when a stock is attractively valued, using fundamentals and price data. Cheapest 20% of U.S. large-caps are selected on a monthly basis.
Motivation for Sentiment Incorporation
The reason for a seeming undervaluation isn't known -> risk of "value traps".
Basic idea: if a stock is surrounded by bad sentiment in the media, then there is likely a reason for it being undervalued -> better to avoid such undervalued candidates.
Sentiment Overlay
Overall shorter-term trailing sentiment indicator, aggregated from various types of events and sources (news, social media), is used to rank and group stocks into quintiles
From the fundamentally "cheapest", undervalued stocks remove the low-sentiment ones (bottom 20%) and overweight instead the high sentiment stocks (top 20%)
Main Results

Results - added value of sentiment
Evaluation has been performed for two time periods: the full period and the most recent one.
Reallocating undervalued low-sentiment stocks capital into undervalued high-sentimen stocks results in extra returns
2002-2018 +1.62pp ann.
2013-2018 +2.4pp ann.
>> this is the pure value added by the sentiment data to the underlying fundamental investment approach.
Overall performance (fundamental valuation + sentiment)
"Low valuation-High sentiment" stocks on average outperformed the market by
2002-2018 +3.5pp ann.
2013-2018 +0.5pp ann.
"Low valuation-Low sentiment" stocks on average underperformed the market by-1.9/-7.5 percentage points per year.
2002-2018 -1.9pp ann.
2013-2018 -7.5pp ann.
Remarks & Conclusion
Correlation to price momentum
Sentiment measured over shorter past periods (i.e. 1 to 3 months) records low correlation to price momentum returns. Longer "look-back" windows for trailing sentiment result in much higher correlation with the price momentum(!)
Conclusion
Avoiding low-sentiment stocks among the fundamentally undervalued companies, and increasing allocation to the high-sentiment ones, can provide uncorrelated (to value and price momentum factors) alpha source and improve overall portfolio returns.
Source: Empirical Research Partners, "Stock Selection: Research and Results", August 2018