Handling Bankrupt Stocks in Survivorship-Bias-Free Backtests
Summary
The document explains how to handle companies that fail during a historical stock study, such as an analysis comparing returns by corporate social responsibility ratings. Its answer recommends defining an investable universe using information available at each point in time, based on a benchmark membership rule or explicit criteria such as market capitalization and liquidity. This avoids selecting firms using knowledge of their future survival.
When calculating forward portfolio returns, the researcher should include the loss associated with bankruptcy during the holding period and stop including the firm in later periods once it no longer meets the investability rule. The example describes a yearly rebalance and a company that fails before the next rebalance, with a near-total loss reflected for that holding period. The treatment should match the study’s rebalance frequency and investment policy. The example is illustrative; actual delisting and bankruptcy returns depend on the available data and event details, so simply carrying post-bankruptcy observations forward is not justified by the answer.
Key ideas
- Define the investable universe using rules that could have been applied at each historical date.
- Use benchmark membership or explicit size and liquidity criteria to guide universe construction.
- Include bankruptcy losses in returns for positions held when the failure occurs.
- Remove a failed company from later portfolio periods when it no longer meets the universe rule.
- Avoid using future survival to determine historical eligibility, which creates look-ahead and survivorship bias.
Tags
Full text
# Analyzing stock performance - keep companies after bankruptcy? # Analyzing stock performance - keep companies after bankruptcy? I am currently analyzing the performance of stocks with high/low corporate social responsibility rating. Some companies went bankrupt during the observation period and I wonder how long I should keep them in my sample. For example, if a company went bankrupt in September 2009, but return data is provided up until 2015, should I keep it in the sample or should I drop it after September 2009? I want to mitigate the survivorship bias, but am afraid that keeping a bankrupt company might distort results. Thank you for your help and have a nice day! ## Answer by NegativeJo (score 3, accepted) https://quant.stackexchange.com/a/36569 It could be useful to set ex-ante what is your investable universe. It is pretty typical that at any period of time you would track only names wich are part of a Benchmark (SP500, R1000, FTSE, Nikkei 225 etc). Or set your own rule for what would consider an investable company. Usually the main criteria are : Market Cap, Market Cap accessible to global investors and median or average liquidity in the last 6 (or 1,3,12) months. As you observed it is critical that you have no survivorship bias. That's why Benchmarks constructed by reputable vendors is usually a safe way to go. Now, apply your investment policy/rule to analyze the performance of stocks with high/low corporate social responsibility rating that are in your investable universe. In the forward returns (holding a stock) make sure you take into account any bankruptcy and stocks moving out of your investable universe. For example: You buy company "FOO" at the end of December 2015 and you have a yearly rebalance process. The company might go bankrupt in August 2016. So when you are holding this name for a year you will likly have a -99.99% return for that name when you rebalance at the end of December 2016. At the same time the likely outcome is that that name is not in your investable universe anymore and would not impact again your performance. This is very similar to how most portfolio managers would operate in general. You can adjust this methodology for a different frequency. The essential is to have a sense of what is an investable stock at point t, independently of what will happen in the future (no survivorship bias and no look ahead bias).
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.