Avoiding Dividend Cutters to Build a Better Dividend Yield Strategy
We provide a high-level overview on a recent strategy we developed that leverages our core technology and machine learning to deliver promising results
"Dividends are the critical factor giving the edge to most winning stocks in the long run." Jeremy Siegel
At AlphaLayer, we uncover investment edges in the markets at scale with a repeatable research process that leverages core technology, data, and artificial intelligence to develop differentiated investment strategies.
Some examples of the edges that we’ve developed include, but certainly not limited to:
Predicting credit rating changes of publically traded companies to identify securities at risk of a near-term downgrade
Factor timing models to create dynamic allocations across factor exposures
Machine learning models applied to a factor universe to improve return forecasts
Recently, we completed work on a dividend yield strategy that aims to drive additional performance through smarter factor exposures and a machine learning model.
This article will give an overview of our research process and techniques used in developing the strategy.
Building a Dividend Yield Strategy
We’ve typically developed institutional strategies but we wanted to take the technology and signals we’ve developed to a retail strategy.
One of the first retail strategies we developed was a dividend yield strategy that gives investors exposure to the market and dividend yield income:
Selection Universe
Dividend-paying stocks
Top 800 US and Top 100 Canadian Companies by Market Cap
All sectors excluding REITs
Rebalance Frequency
Quarterly
Our goal with this strategy was to provide investors with dividend yield income while reducing volatility by limiting exposure to downside risks associated with companies that payout higher dividends:
To mitigate some of the downside risk created by value traps and cutters, we wanted to lean into companies that had solid fundamentals and tailwinds that were unlikely to reduce their dividends in the near term.
Base Strategy (Dividend Factor Version)
The first step that we took was to build a baseline strategy that simply took exposure to a dividend yield, factor portfolio, which would be our starting point to compare the impact of additional enhancements.
We leveraged our factor library to select a set of dividend yield-related factors, which resulted in the following strategy returns:
Now that the Dividend Factor strategy is in place, we moved to enhancing performance through addressing common pitfalls in this type of strategy.
Avoiding Value Traps (Enhanced Factor Version)
“Wow, this stock is paying a 15% dividend! Buy, buy, buy!”
A common trap among dividend payers is companies with a high yield that are attractive from a yield perspective but can be a sign of a company in trouble (high payout ratios, declining cash flows, debt, earnings issues, low growth, etc.).
So while you may get a strong dividend yield, it is likely to be cut or come with significant downward pressure on the stock price resulting in the potential of a flat or a negative total return.
To avoid these traps and ensure that we select dividend payers with stronger businesses, the strategy development effort focused on quant factors from our factor library that emphasized information that might augment strictly looking at dividends only.
We employed the following factor category styles (each category will contain a range of factors):
Dividend Yield (same as used in the Dividend Factor version)
Quality
Momentum
This delivered significantly stronger performance than the original Divided Factor version alone, which signaled that we are avoiding the dividend payers that drag down performance.
While the Enhanced Factor strategy leveraging a broader set of factors was significantly better than just leaning into Dividend Yield factors, we also wanted to layer in machine learning to boost performance.
Avoiding Dividend Cutters with Machine Learning (Enhanced + ML Version)
A dividend cut not only reduces the income that you generate out of an investment but will typically result in downward pressure on stock prices as it is a sign of financial issues with the firm cutting.
This is something that any dividend investor wants to avoid.
We wanted to experiment around if we could predict companies that cut their dividend in the next rebalance period of the strategy. If we avoid these cutters, we are likely to reduce our downside risk.
At a high level, these are the steps we took to build and test our model:
Generate labels on whether a given company will cut its dividend over the upcoming holding period
Extensive feature engineering-based firm fundamentals, technical indicators, and macroeconomic trends
Train the model on past examples of divided cuts
Use the trained model ahead of each rebalance period to remove or avoid an investment in dividend cutters
There are a lot of details that go into the model and its evaluation that we can’t cover in a single article but here are the results of the model we developed to boost the performance of the Enhanced Factor model:
Our model can predict dividend cuts with a reasonable degree of precision - these events are relatively rare but have an outsized negative impact on security prices so any ability to catch them in advance has a performance boost.
While the model doesn’t catch all dividend cutters, when added to the strategy, the annualized returns of our base strategy increased by ~80 basis points with minimal increase in volatility leading to a ~6% increase in the Sharpe ratio.
In Conclusion
In this strategy example, we illustrate how we go about building a retail dividend yield strategy leveraging quant factors and machine learning to deliver a strategy with yield and market exposure with reduced volatility.
There are many more details about how strategies are developed and techniques used that we will cover in future articles.
If you are interested in learning more about the specifics of this strategy and the ML model - feel free to reach out. We’d be happy to dive into more details.
Cheers!