What you don’t know can hurt you.
There are 3,845 companies listed on U.S. stock exchanges today. Additionally, we follow another 679 companies listed on Canadian exchanges, for a total of just over 4,500 publicly-listed companies. That’s a lot of companies for one investment manager or even a group of analysts to effectively cover. However, it is only the tip of the iceberg.
Any solid investment strategy needs to have a rationale that not only can be implemented, but also explained to investors. Every successful manager has “an edge” that he or she brings to the table. That details of that edge may be widely disseminated and public knowledge (like a Warren Buffett), or may be black-box and proprietary (like a DE Shaw). Even an investor that just “buys stocks that I like” has an embedded methodology of what screams “goodness” or “badness” from a particular candidate pool.
The best investment strategies have history and an understanding of how they perform over different economic and investing cycles. One needs to be wary of a strategy that only has a short history of positive returns, or no history at all. I can certainly show you a strategy that has solid returns over the last few years, but miserable returns over an extended period time. But to review any strategy over multiple years, capturing those 4,500 stocks are not enough. Since 1987, there are 35,000 stocks that once traded, the no longer do. The fate of those 35,000 names – whether it be from acquisition, bankruptcy or delisting – plays a major role in how well the strategy would have actually performed.
Bloodhound maintains a comprehensive 26-year point-in-time database, meaning it has all the data (and only that data) that was available to an investor on the days being evaluating. It includes all the 4,500 active stocks, plus all the 35,000 now-defunct ones. As such, the database is free of look-ahead and survivorship biases.
Stocks tend to be correlated to each other, so how important is eliminating survivor bias? Let’s look at a few strategies we have created.
Shareholder Yield Strategy
We built a basic strategy for a well-known quant blogger. We called it the Shareholder Yield Strategy. It was relatively simple function, dividends + net share buybacks + net debt paydown divided by market cap. With a few exceptions (this year included), the strategy has decent outperformance compared to the S&P 500 in the last 1-, 3-, 5-, 10- and 20- years. It is a strategy with a seemingly low risk profile from a company selection standpoint. Based on the criteria, these are not companies with any significant signs of distress. However, when we compare the results of those using the whole universe available at each point in time, and only companies that are still trading today (thus building in a survivorship-bias), the results are markedly different.
In this particular strategy, the difference is real, even if not consistent. A portfolio without survivorship bias performs better in the medium term than one with only companies of today. However, over the longer term such traditional backtesting clearly overstates the return of this Yield strategy. Even considering the near-term outperformance, over 20 years, a portfolio that generates an average 17% return is nearly two times the dollar return of one that returns 13.3%.
Companies with solid fundamental growth stories often make interesting acquisition candidates. As such, a database that does not consider companies that are no longer listed, will exclude take-out premiums by companies ultimately acquired. It will also exclude those names that flamed out, but based on the table below, it appears that building in a survivorship-bias hurts the returns of this strategy.
After ten years, a portfolio that excludes those companies no longer listed is almost a quarter lower.
The best example is probably the Low P/E strategy. Any Value investor will tell you that sometimes companies are cheap for a reason. A strategy that invests in low price-to-earnings ratios, but fails to consider those companies that ultimately never make it, will drastically overstate its results.
Low P/E companies of the last few years that have survived through today have likely appreciated considerably in this current bull market; however, as that time period is expanded, it becomes clear that having a survivorship bias in your data makes the strategy look better than it would have looked in real life.
Having the point-in-time database that we do allows Bloodhound to run a simulation not a typical backtest. Backtesting can often come with negative connotations due to its misuse and/or abuse. However, a simulation re-creates and emulates a strategy as if it were implemented on the date it was run. A simulation free of look-ahead and survivorship biases therefore provides a realistic record of how that strategy would have performed as is ultimately the standard bearer for historical performance evaluation.