Bankruptcy Prediction Using Financial Ratios

Ozge Yilmaz and Azra Ozcirpan

2026-03-26

Bankruptcy Prediction Using Financial Ratios

ECON 465 – Final Project

Özge Yılmaz & Azra Özçırpan

Economic Question and Motivation

Why Do Firms Go Bankrupt?

  • Bankruptcy affects investors, lenders, employees, and economic growth.

  • Early identification of financial distress can reduce economic losses.

  • Financial ratios are commonly used as warning indicators of firm failure.

Economic Question

Can financial ratios predict whether a firm will go bankrupt?

Dataset Description

Polish Company Bankruptcy Dataset

This dataset contains financial ratios of firms over multiple years together with a binary bankruptcy indicator.

1-year company data subset

Source: UCI Machine Learning Repository

3194 observations

1-year subset of the Polish Companies Bankruptcy Dataset

64 financial ratios

Target variable:

is_bankrupt Yes / No

Variables Used

Attr1 → Profitability

Attr2 → Leverage

Attr5 → Liquidity

Attr7 → Operating Performance

Attr10 → Asset Efficiency

Finding from First Stage

Probability & Distribution Analysis

  • Only about 3.9% of firms were bankrupt showing that bankruptcy is a rare event.

  • The dataset was highly imbalanced, making prediction more difficult.

  • Profitability (Attr1) showed strong skewness and extreme values. Firms with lower profitability tended to be closer to financial distress.

  • Financial ratios showed substantial variation across firms.

  • Log transformation improved the distribution.

Implication

Bankruptcy is a rare event, making prediction more challenging.

Finding from Second Stage

Initial Classification Results

  • Bankrupt firms were less profitable and more leveraged.
  • Leverage was the strongest bankruptcy indicator.
  • Class imbalance made bankruptcy prediction difficult.
  • Accuracy alone was not a reliable performance measure. 

Initial Classification Results

Financially weaker firms face a greater risk of bankruptcy.

Models Built and Compared

Model 1(Full Logistic Regression)

  • Profitability (Attr1)

  • Leverage (Attr2)

  • Liquidity (Attr5)

  • Operating Performance (Attr7)

  • Asset Efficiency (Attr10)

Model 2: Simple Logistic Regression

  • profitability + leverage

  • Smaller set of predictors

  • More stable coefficient estimates

  • Easier economic interpretation

Both models were evaluated using training and test data.

Model 2 produces more stable estimates and has the advantage of interpretable, theoretically grounded coefficients.

Why We Chose the Final Model

Reasons

  • Better test-set performance

  • Stronger economic interpretation

  • Included key financial indicators

  • Simple and transparent model structure

The final model was simpler, easier to interpret, and performed well on test data.

Main Results

Key Results

  • Bankruptcy rate: 3.9%

  • Bankrupt firms were less profitable.

  • Higher leverage increased bankruptcy risk.

  • Financial ratios helped identify distressed firms.

Economic Interpretation

What do the results mean?

  • High debt levels increase financial vulnerability.

  • Financially weak firms are more likely to fail.

  • Investors and lenders can use financial ratios as warning signals.

  • Bankruptcy prediction can support better financial decisions.

Limitations and Future Improvements

Limitations

  • Highly imbalanced dataset

  • Only five financial ratios used

  • Logistic regression may miss complex patterns

Future Improvements

  • Use imbalance-handling techniques

  • Include additional financial indicators

  • Experiment with different prediction thresholds and maybe with different predictors.