What Is Stock Screening Survivorship Bias?

Survivorship bias fails to take into account the non-survivors in a study, sometimes leading to inflated figures and inaccuracy.

Stock screening survivorship bias is a statistical error that can materialize in stock analysis. It occurs when failed stocks from the past are not included in the analysis of present stocks, and the results of the analysis are skewed towards overestimating performance and toward overly optimistic projections for the future. More generally, survivorship bias is any statistical error that results from not fully incorporating the statistical impact of non-survivors into statistical analysis.

Survivorship Bias in Finance

A smart statistician will correct for survivorship bias by including failed stocks in the analysis of performance.

Survivorship bias can sometimes arise in finance studies during stock screening procedures, especially those conducted by amateur investors and non-statisticians. When a study is being undertaken of past stocks and performance of those stocks, the performance of failed companies (those whose stock price is now zero and who do not exist anymore) is sometimes not included. This tends to weight the results, skewing them higher because only successful companies (those that survive until the end of the period) are included.

Debate on Prevalence

Using a large sample size whittled eventually down to a few individuals, Joseph Banks Rhine exemplified the error of survivorship bias in his study on ESP.

The prevalence of stock screening survivorship bias, and survivorship bias generally, is debated among academics. Some studies allege that a large majority of finance studies and, indeed, a large number of studies in other fields as well, suffer from survivorship bias and other types of selection bias. Famous examples abound, such as Joseph Banks Rhine's finding of ESP in certain individuals amongst a very large sample. After finding a select few participants who seemed to be achieving ESP in a card-guessing game, he published a paper on his findings, which largely ignored the many participants who had not shown promise. By eliminating those who had shown little promise, Rhine selected for those who by chance showed accuracy, and the resulting study imputed telepathic powers to people who had most likely just been luckier than others. On the other hand, other studies of survivorship bias suggest that if it occurs in many studies, the effect is likely to be small.

Correcting for Surivorship Bias in Stock Screening

Analyzing the S&P 500's historic returns should include firms that are no longer in the index fund but were at one time.

The key to avoiding the error of survivorship bias is to always include the non-survivors, in any context. For example, imagine a study to determine the annual rate of return of the S&P 500, a common index fund. One method would be to take the 500 US companies which are currently in the fund, and chart their performance over the past 50 years or any other desired time period. However, the S&P 500 is an index fund, so it has changed over the years, as unprofitable companies are phased out and more profitable companies are included. This method of looking at the current 500 companies through the past would surely overstate the average return of the S&P 500 during the time period under investigation. Instead, a proper statistical technique would be to include all firms that have ever been a part of the S&P 500 over that time period, keeping track of their dates of entry into the fund and the date (if ever) that they exited the fund. This would thereby include non-survivors and free the study from survivorship bias.

Theory of Research

Survivorship bias may manifest itself in the types of articles that are published in journals, such that some findings might be merely chance.

The effect of survivorship bias relates to a fundamental problem in modern academia and research. If enough researchers attempt to study a phenomenon (like ESP, as in the example of Joseph Banks Rhine, or mutual fund performance), some of the researchers studying it will, by sheer chance, observe statistically significant results. The vast number of other researchers studying the phenomenon can fail to find this conclusion, but the one statistically significant paper gets sent into a scientific journal and is published --- in other words, the non-survivors are in this case papers that prove "negative findings", and the survivors are those that by chance declare a positive finding which is false and occurs only by chance. From this logic, it makes sense that a possibly high number of articles in scientific journals are outliers, displaying nothing more than chance, as alleged by the controversial 2005 paper "Why Most Published Research Findings are False" by Dr. John Ioannidis. To solve this problem, some scientific journals are now encouraging the submission of so-called "negative findings."

References

• "Survivorship bias in Performance Studies"; Brown, Goetzmann, Ibbotson, and Ross; 1992
• "Survivor Bias and Mutual Fund Performance"; Elton, Gruber, Blake; 1996
• "Managerial Economics and Business Strategy"; Michael R. Baye; 2009
• "Why Most Published Research Findings are False"; John Ioannidis;

Resources

• "Most Scientific Papers Are Probably Wrong" by Kurt Kleiner; NewScientist 2005