How To Design A Bias-Free Dividend Strategy

by: Fred Piard

In a search for probabilistic gains (my methodology can be found here), it is impossible to ignore the power of dividends. My approach is different from similar ones that use a Dividend Champions universe. This kind of universe is defined from a historical survey, so it is subject to "survivor" bias and "look-ahead" bias. Both are dangerous traps in simulating a strategy. Using a simulation software for a calculation doesn't make it a simulation. The problem is that when we hold a hammer in our hand, every problem looks like a nail.

A simulation should use only data that were available at every decision point in the past. It starts with the stock universe. Simulating a strategy for 10 years on a Dividend Champions list built with data unavailable 10 years ago is questionable. Calculating a Sharpe ratio, a probability, a gain-loss ratio makes no sense. A Dividend Champions list is a great tool. I just don't expect anything from a backtest on it. The good news is that designing a bias-free dividend strategy may be easier than you think.

My Bias-Free Dividend Universe

I start with a set of stocks whose price is above $10 and has a market capitalization above $100 million. OTC stocks are excluded, ADRs are included. Stocks whose average daily volume of the latest 20 days in U.S. dollar is below 500,000 are excluded. Then, I keep only the stocks whose average yield for the previous five years is above 5%. It is a universe of 229 stocks at the time this article was written. The number and content are variable. It has no conceptual survivor bias and no look-ahead bias. The database I use has no technical survivor bias.

A "Dummy" Strategy

This is just an example, so I've made it as simple as possible: I selected in this universe the 20 companies with the highest current yield.


In this simulation, the portfolio is rebalanced every quarter from Jan. 1, 2002, to April 1, 2012.

  • A 1% slippage is applied for every trade.
  • Capital is equally allocated between the stocks.
  • I added a rule so that a position cannot be more than 1/20 of the total (useless here, because the 20 positions are always filled).


  • Average quarterly gain (dividends included): 3.76%
  • Average gain/average loss (quarterly): 0.95
  • Probability of gain (quarterly): 70%
  • Drawdown: 60%

I confess that I wouldn't sleep well having money in a strategy that has had such a drawdown (it's very similar to the SPDR S&P 500 (NYSEARCA:SPY) in fact). But if the strategy looks like a dummy, an annualized return of about 15% and a Kelly criterion of 40% are not.

Here are the 20 stocks at the time this article was written: AI, BBEP, BCF, CH, CMO, CNSL, CODI, CWH, ERF, FGP, FTE, MVO, NLY, NMM, OIBR, PBI, PSEC, SJT, TEF, and YPF.

The description and analysis of these stocks is out of scope here. Interested readers can explore the list; I just want to point out that almost half of it is in the financial sector, the rest being in energy, services and transportation.


I have shown here my first steps in designing a bias-free strategy. The strategy provided as an example is not an end product, and more research (not described here) can improve the drawdown and the return. Nevertheless, its rough performance is far from ridiculous. If you are a dynamic investor ready for some homework, feel free to take that as a framework. In this case, you can also read basic principles of risk management here that may help you set up your own strategy.

Disclosure: I am long NLY.