Addressing Equity Backtest Survivorship Bias with Delisting Data
Summary
The document asks whether corporate bankruptcy rates can be related to valuation ratios such as price-to-book or price-to-earnings, as a way to estimate survivorship bias in equity backtests. The answer recommends using a historical stock database that retains delisted securities, citing CRSP as an example. A dataset that preserves delisted firms directly addresses the missing-security problem more reliably than estimating losses from bankruptcy rates.
The response cautions that bankruptcy is not equivalent to delisting: some bankrupt firms remain listed, and delistings can occur for other reasons. Using bankruptcy rates as a proxy could therefore introduce more error than it removes. It mentions that Compustat contains extensive corporate finance data, but does not provide a bankruptcy-rate study or a valuation-based adjustment method. The main practical lesson is to source data that includes firms that disappeared from the market. The answer also notes that commonly used free data sources may have survivorship bias, though it does not assess specific datasets in detail.
Key ideas
- Survivorship bias can distort backtests when historical data omit delisted firms.
- A database that retains delisted securities can address this data problem directly.
- Bankruptcy is an imperfect proxy for delisting because the events do not always coincide.
- Estimating a bankruptcy adjustment from valuation ratios is challenging and is not developed in the response.
- Check whether a historical equity dataset preserves securities after they leave the market.
Tags
Full text
# Data on US bankruptcy rate vs. standard valuation ratios # Data on US bankruptcy rate vs. standard valuation ratios Does anyone know of any research or data on US corporate bankruptcy rates as a function of standard valuation ratios, such as P/B, P/E, etc.? I'm trying to adjust the results of backtests to account for survivorship bias. My first thought was to put an upper bound on the impact of survivorship bias by assuming a certain percentage of holdings go bankrupt each period. I was looking for some data on what that percentage should be. I'm aware of values like Altman's Z-score, but that doesn't quite apply to what I'm trying to do. That's meant to predict bankruptcy, but I'm looking for a typical bankruptcy rate. Even an overall rate would be better than nothing, but it would be more useful if it was broken down cross-sectionally in some way. Alternatively, are there any better methods of dealing with survivorship bias when working with incomplete data sets? ## Answer by Richard Herron (score 3, accepted) https://quant.stackexchange.com/a/734 I get my stock data from Univ of Chicago's Center for Research in Security Prices (CRSP), which keeps stocks after delisting, so there's no survivorship bias. I assume that Bloomberg and Reuters do the same. But if you use Yahoo! or Google, there will be a survivorship bias. I think that trying to look at bankruptcies to remove a survivorship bias would be really challenging (not all bankruptcies result in delisting), and might insert an error bigger than survivorship bias. But Compustat provides more corporate finance data than almost anyone knows what to do with. I am not sure that there's a clean work-around for survivorship bias short of moving to a data set like CRSP. HTH.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.