P-Hacking Via Academic Finance Research Conferences

Documentation of the File Drawer Problem at Finance Conferences: A Follow-Up Study

• Manoela N. Morais and Matthew R. Morey
• Journal of Investing
• A version of this paper can be found here
• Want to read our summaries of academic finance papers? Check out our Academic Research Insight category

What are the research questions?

This research is an update to “Documentation of the File Drawer Problem in Academic Finance Journals” published by the same authors in the Journal of Investment Management in 2018. A summary of that article can be found here. The “file drawer problem” refers to the idea that journal editors are predisposed to accepting articles for publication, only if they contain statistically significant results.  Since editors are motivated by improving journal impact numbers and citation counts, this bias is not surprising.  Articles with significant results are more likely to be cited and thus improve journal impact.  Articles with nonsignificant results end up hidden away in the researchers’ file drawer and not submitted anywhere at all.  Putting numbers to the problem in academic journals, the authors reported only 2.1% of 29 finance journals published nonsignificant results.  Five of those 29 journals published no studies with insignificant results. This update examines the degree to which finance conferences exhibit a similar pattern.

1. Is there a significant file drawer problem with respect to academic financial conferences?

What are the Academic Insights?

1. YES. The file drawer problem was observed to be at least as serious at finance conferences as it is in finance journals. The authors constructed a database of 3,425 empirical articles presented at the annual Financial Management Association for 5 years.  The FMA is the largest academic conference by number of papers.  Each paper examined was a stand-along research article.  Roundtables, panel sessions, pedagogy series and debates were not included. Of the 3,425 articles, only 14 (or 0.41%) had nonsignificant results over the five year period.  This is in comparison to the 2.1% of articles published in academic journals.  It also appears that the problem within the FMA intensified between 2014 and 2018. Stunning.

Why does it matter?

As with journal publications, this article provides evidence that the file drawer problem is alive and well with respect to academic financial research conferences.  It appears that potential presenters should avoid submitting analyses that have nonsignificant results otherwise risk rejection by the conference. As a result, conference attendants see a biased set of research presentations comprised of only those papers that exhibit statistical significance. The important question here how much this bias contributes to the use of p-hacking or datamining practices in order to achieve significant results. We have seen increasing attention paid to the practice of p-hacking, datamining, and other “bad habits” and the negative impact they have on the credibility of the discipline.

In 2017, Campbell Harvey (his Presidential Address for the Am Finance Assoc) took the issue one step further into the intentional misuse of statistics in empirical research.  He defines intentional p-hacking as the practice of reporting only significant results when the researcher has conducted a myriad of statistical methods, empirical approaches or data manipulation.  The underlying motivation for the use of such practices is the desire to be published in a world where finance journals are biased towards publishing significant results almost exclusively. The underlying risk to p-hacking and datamining, especially in the investments area, is the identification of significant results when they are likely just random events.   Since random events by definition, do not repeat themselves in a predictable manner, the investment results are likely to fail on a going-forward basis. Datamining and p-hacking go a long way in explaining why investment strategies fail out-of-sample, or even worse when they are implemented in the real world.

This criticism can now be extended to finance conferences.

The most important chart from the paper

Abstract

The file drawer problem is a publication bias where journal editors are much more likely to accept empirical papers with statistically significant results than those with statistically nonsignificant results. As a result, papers that have nonsignificant results are not published and relegated to the file drawer, never to be seen by others. In a previous paper, Morey and Yadav (2018) examined the file drawer problem in finance journals and found evidence that strongly suggests that such a publication bias exists in finance journals. In this follow-up study, we examine the prevalence of the file drawer problem at finance conferences. As such we are the first article in finance that we know of to attempt such an analysis. To do this, we examine every single empirical paper presented at the annual Financial Management Association (FMA) conference from 2014–2018. In an examination of 3,425 empirical papers, we found less than 0.5% of these papers had statistically nonsignificant results. These results suggest that there is also a significant file drawer problem at finance conferences.