Faculty of law blogs / UNIVERSITY OF OXFORD

Predicting Bankruptcy: Ask the Employees


John Knopf
Professor Emeritus at the University of Connecticut
Kristina Lalova
Assistant Professor at Michigan State University


Time to read

4 Minutes

In one of our recent papers, Predicting Bankruptcy: Ask the Employees, we test the predictive performance of established bankruptcy prediction models against a new model inclusive of employee information. We track employees’ attitudes from before bankruptcy filings to after bankruptcy filings and test their predictability of bankruptcy across three various phases of the bankruptcy process—two and three years before (1), one year before (2), and from the time of filing to the time of liquidation/reorganization (3). We show that before bankruptcy filings, employees have insider information on where the company is headed. Well before bankruptcy or even negative financial performance, managers and employees may be aware of significant problems within their companies. Although managers may be reluctant to disclose this information, employees may reveal problems through dissatisfaction with their jobs and the firm, although we don’t necessarily know what the underlying cause for their dissatisfaction is and how they are going to express their dissatisfaction. This employee insider information, however, shows up as a more powerful predictor of bankruptcy, in comparison to financial statement data, two and three years prior to bankruptcy filings. In the year before the bankruptcy, employee insider information is overwhelmed by financial statement data. From the time of bankruptcy filing to the time of liquidation/restructuring, employee insider information overwhelms financial and market data to predict whether the company will emerge from bankruptcy.  We empirically test our model for predictability, not causation, similar to what prior bankruptcy prediction literature does. Whether employees are less satisfied because of an impending bankruptcy or whether employee satisfaction impacts the chances of bankruptcy is an interesting topic for further studies.

In this paper, we define employee satisfaction as employees’ attitudes toward and perceptions of the tasks employees have in the companies they work for and various firm dynamics. Those perceptions toward various firm dynamics include perceptions toward career opportunities, compensation and benefits, culture and values, work-life balance, senior leadership, and overall organizational performance. All these firm dynamics in the dataset we use are determinants of employee satisfaction. To put it in other words, leadership styles (senior leadership as in our data), organizational culture (culture and values as in our data), work-life balance (work-life balance as in our data), compensation and benefits (compensation and benefits as in our data), and opportunities for career advancement (career opportunities as in our data) are all determinants of employee satisfaction. We take the level of the perceptions toward those various firm dynamics to build employee satisfaction. While employee satisfaction can influence bankruptcy risk through its effect on organizational performance, it is not the sole predictor of bankruptcy. Researchers have determined that financial and market information is a predictor of bankruptcy in various models and settings. We improve on prior and established research to track how employee satisfaction shows up as a predictor of bankruptcy in various settings over various phases/years of the bankruptcy filing process.

We document that employee satisfaction is a strong predictor of bankruptcy in all three phases of the bankruptcy process—from two and three years before, one year before, and while in bankruptcy. Specifically, we find that the employee satisfaction model predicts bankruptcy more accurately than any of the existing financial information-based models in all years other than the year immediately prior to a bankruptcy filing. While already-established models’ predictive power increases the closer we get to bankruptcy filings, our model’s predictive power, in comparison to other models, is higher the further we move from bankruptcy filings. With this finding, we improve on one of the drawbacks of already-established models, more specifically that already-established models don’t predict bankruptcy as accurately when moving years back from bankruptcy filings. We additionally find that close to the bankruptcy filing date, models with inclusion of both financial statement and employee satisfaction data outperform models with inclusion of financial data only, according to Adjusted R-Squared, ROC curves, optimal threshold points with the highest sensitivity and specificity, and Type I, Type II, and Total classification error rate analysis. Separately, we hypothesize that if a company is more likely to emerge from bankruptcy, employees are more likely to keep their jobs and be more satisfied with their jobs which is reflected in higher chances of such companies surviving bankruptcy and emerging from it. We document that employee satisfaction predicts bankruptcy emergence and that companies with higher employee satisfaction are more likely to emerge from bankruptcy.

We split our results’ analyses into three main sections—testing bankruptcy prediction models with information one year before bankruptcy filings (1), testing Altman’s (the most accurate statistical model with financial information one year before) and employees’ model with information two and three years before bankruptcy filings (2), and testing a bankruptcy emergence model (3). In the first results section of the paper, we test four key statistical bankruptcy models from the literature using a dataset from 2008 to 2020 to show that each one contains unique information regarding the probability of bankruptcy filings. We also build a new model to reflect employees’ attitudes before bankruptcy filings and include key variables from each of the four already-established bankruptcy models in the literature in our model. We perform several analyses including parameter estimation tests, bankruptcy classification rate tests, out-of-sample analyses, and boosting machine learning analyses.

Overall, our paper provides novel insights into how employee satisfaction shows up as a predictor of bankruptcy in various years (phases) of the bankruptcy process. Since prior literature has proved that companies’ financial and market information can predict bankruptcy, we don’t contradict other papers’ findings, but rather improve on prior and established research by arguing that employee satisfaction shows up as a predictor of bankruptcy prior to financial and market information, that employee satisfaction improves other models’ performance in the year prior to the bankruptcy (where their predictive power is the highest), and that employee satisfaction is a more powerful predictor of bankruptcy emergence than financial and market information. The paper shows that two and three years before bankruptcy employees sense issues within their companies, while one year before bankruptcy financial and market information already reflect what employees knew two and three years prior. From the time the company files for bankruptcy to the time the company either liquidates or restructures, employee satisfaction up to three years prior to bankruptcy filings reflects whether the company would emerge from bankruptcy or not. The results signify that employees hold information on companies’ financial health and prospects throughout various years of the bankruptcy process, but that information is expressed in a different way throughout the bankruptcy process.

The authors’ full paper can be found here.

John Knopf is Professor Emeritus at the University of Connecticut.

Kristina Lalova is a Fixed-Term Assistant Professor at Michigan State University.


With the support of