Modernizing the Academic Teaching and Research Environment by Jorge Marx Gómez & Sulaiman Mouselli

Author:Jorge Marx Gómez & Sulaiman Mouselli
Language: eng
Format: epub
Publisher: Springer International Publishing, Cham

1.3 Sample-Selection Bias

It is the bias that results from the elimination of certain observations (such as firms, funds or assets) from a study due to different reasons. The work of Heckman (1979) argues that there are two reasons for sample-selection bias to appear in practice; “First, there may be self selection by the individuals or data unit being investigated. Second, sample selection decisions by analysts or data processors operate in much the same fashion as self selection” [9].

This bias could also be a database-specific that results from using historical information and suffers from a type of sample-selection bias known as survivorship bias. Survivorship bias results from using a dataset that only includes the survivors during the sample period, not the complete set of firms that were available over the period [10]. Many databases suffer from this bias especially those that only list firms or funds that currently alive (for example, Compustat for United States accounting information), which means firms that have failed are not included in the database. Consequently, results obtained from the study will be biased towards firms or funds that are successful and may not accurately reflect the true picture.

Sample selection bias is even worse in studies of credit rating and hedge fund returns. If firms that would have received lower credit ratings (because they have weak financials) opt not to request a rating, results of the determinants of credit rating will only be valid for firms with high credit rating, and coefficients will be inconsistent [5]. Similarly, because hedge funds are not required to publicly disclose their performance data, only funds with good performance will choose to disclose their performance. Hence, hedge fund returns will be overestimated and coefficient estimates will be inconsistent.

To control for sample-selection bias in hedge fund studies, Heckman (1979) suggests two-step procedure that involves first estimating a 0–1 probit model of whether the fund solicits to disclose its performance and second estimating the ordered probit model for the determinants of the performance of hedge funds [9].

A common bias in sample selection occurs also when a researcher selects the sample from the currently available observations and ignores firms that disappear during the sample period. This type of bias mixes the sample selection bias with look-ahead bias because it only includes firms with available information at a certain recent point of time. One way to avoid this bias is through considering all dead and live cases (firms) during the whole sample period.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.