# Bankruptcy Predictions

{% hint style="info" %}
Monthly corporate bankruptcy predictions arrive the **2nd of every month***.*
{% endhint %}

{% hint style="success" %}
Dataset contains 4762+ tickers, available from 1998-03-31 onwards.
{% endhint %}

`Tutorials` are the best documentation — [<mark style="color:blue;">`Corporate Bankruptcy Tutorial`</mark>](https://colab.research.google.com/github/sovai-research/sovai-public/blob/main/notebooks/datasets/Bankruptcy%20Prediction.ipynb)

<table data-view="cards" data-full-width="false"><thead><tr><th></th><th></th></tr></thead><tbody><tr><td><strong>Input Datasets</strong></td><td>SEC Bankruptcies, Delistings, Market Data, Financial Statements</td></tr><tr><td><strong>Models Used</strong></td><td>CNN, LightGBM, RocketModel, AutoEncoder</td></tr><tr><td><strong>Model Outputs</strong></td><td>Calibrated Probabilities, Shapley Values</td></tr></tbody></table>

## Description

The model predicts the likelihood of bankruptcies in the next 6-months for US publicly listed companies using advanced machine learning models.

With an accuracy of around 89% and ROC-AUC of 85%, these models represent a large improvement over traditional methods of bankruptcy prediction for equity selection.

Advanced modeling techniques used in this dataset:

* **The Boosting Model**: Utilizes LightGBM technology, integrating both fundamental and market data for accurate predictions.
* **The Convolutional Model**: Employs a Convolutional Neural Network (CNN) for efficient pattern recognition in market trends.
* **The Rocket Model**: Specializes in time series data, using random convolutional kernels for effective classification and forecasting.
* **The Encoder Model**: Combines LightGBM with CNN autoencoders, enhancing feature engineering for more precise predictions.
* **The Fundamental Model**: Focuses solely on fundamental data via LightGBM, without extra architectural layers, for straightforward financial analysis.

## Data Access

### **Monthly Probabilities**

**Specific Tickers**

```python
import sovai as sov
df_bankrupt = sov.data('bankruptcy', tickers=["MSFT","TSLA","META"])
```

<figure><img src="/files/PJVVR92TkmGxZ3TanzwH" alt=""><figcaption></figcaption></figure>

**Specific Dates**

```python
import sovai as sov
df_bankrupt = sov.data('bankruptcy', start_date="2017-01-03", tickers=["MSFT"])
```

**Latest Data**

```python
import sovai as sov
df_bankrupt = sov.data('bankruptcy')
```

**All Data**

```python
import sovai as sov
df_bankrupt = sov.data('bankruptcy', full_history=True)
```

### Daily Probabilities

```python
import sovai as sov
df_bankrupt = sov.data('bankruptcy/daily', tickers=["MSFT","TSLA","META"])
```

The daily probabilities are experimental, and have a very short history of just a couple of months.

### Feature Importance (Shapleys)

```python
import sovai as sov
df_importance = sov.data('bankruptcy/shapleys', tickers=["MSFT","TSLA","META"])
```

Feature Importance (Shapley Values) calculates the contribution of each input variable (features) such as Debt, Assets, and Revenue to predict bankruptcy risk.

<figure><img src="/files/2BRMfai6pJy6r1vgYCpp" alt=""><figcaption></figcaption></figure>

## Reports

### Sorting and Filtering

```python
import sovai as sov
sov.report("bankruptcy", report_type="ranking")
```

<figure><img src="/files/LYnZs07yCgPRUWbPI3sc" alt=""><figcaption></figcaption></figure>

Filter the outputs based on the top by **Sector**, **Marketcap**, and **Revenue** and bankruptcy risk. You can also change <mark style="color:blue;">`ranking`</mark> to <mark style="color:blue;">`change`</mark> to investigate the month on month change.

```python
sov.report("bankruptcy", report_type="sector-change")
```

## Plots

### Bankruptcy Comparison

```python
import sovai as sov
sov.plot('bankruptcy', chart_type='compare')
```

<figure><img src="/files/Xg91Fa8MlL8kdNM1J8ky" alt=""><figcaption></figcaption></figure>

### Timed Feature Importance

```python
import sovai as sov
df = sov.plot("bankruptcy", chart_type="shapley", tickers=["TSLA"])
```

<figure><img src="/files/Qy8o0xUNpsKJEXf0oXUa" alt=""><figcaption></figcaption></figure>

### Total Feature Importance

```python
import sovai as sov
sov.plot("bankruptcy", chart_type="stack", tickers=["DDD"])
```

<figure><img src="/files/7JSco8tEWm7lqbfgIcoA" alt=""><figcaption></figcaption></figure>

### Bankruptcy and Returns

```python
import sovai as sov
df= sov.plot("bankruptcy", chart_type="line", tickers=["DDD"])
```

<figure><img src="/files/ZGIKwcXNem0A1AtxNrli" alt=""><figcaption></figcaption></figure>

### **PCA Statistical Similarity**

```python
import sovai as sov
df= sov.plot("bankruptcy", chart_type="line", tickers=["DDD"])
```

<figure><img src="/files/U6tLTMrT7o2jLRiEeUlf" alt=""><figcaption></figcaption></figure>

### Correlation Similarity

```python
import sovai as sov
sov.plot("bankruptcy", chart_type="similar", tickers=["DDD"])
```

<figure><img src="/files/jEd4uCpBI7EXjQ1HoO85" alt=""><figcaption></figcaption></figure>

### Trend Similarity

```python
import sovai as sov
sov.plot("bankruptcy", chart_type="facet", tickers=["DDD"])
```

<figure><img src="/files/9eg2jqVMOnxioke6ZIZd" alt=""><figcaption></figcaption></figure>

## Model Performance

### **Confusion Matrix**

```python
import sovai as sov
sov.plot("bankruptcy", chart_type="confusion_global")
```

<figure><img src="/files/StjFBmNI4w9fYnBUCZM0" alt=""><figcaption></figcaption></figure>

### **Threshold Plots**

```python
import sovai as sov
sov.plot("bankruptcy", chart_type="classification_global")
```

<figure><img src="/files/OqNVP5RFasv6FpkP78mR" alt=""><figcaption></figcaption></figure>

### **Lift Curve**

```python
import sovai as sov
sov.plot("bankruptcy", chart_type="lift_global")
```

<figure><img src="/files/vXMZrt5jW73gcyIev6bx" alt=""><figcaption></figcaption></figure>

### Global Explainability

```python
import sovai as sov
sov.plot("bankruptcy", chart_type="time_global")
```

<figure><img src="/files/X4h80rEAUsf1TY5SnSKO" alt=""><figcaption></figcaption></figure>

## Computations

Leverage advanced computational tools for deeper analysis:

* **Distance Matrix:**

  ```python
  sov.compute('distance-matrix', on="attribute", df=dataframe)
  ```

  Assess the similarity between entities based on selected attributes.
* **Percentile Calculation:**

  ```python
  sov.compute('percentile', on="attribute", df=dataframe)
  ```

  Calculate the relative standing of values within a dataset.
* **Feature Mapping:**

  ```python
  sov.compute('map-accounting-features', df=dataframe)
  ```

  Map accounting features to standardized metrics.
* **PCA Calculation:**

  ```python
  sov.compute('pca', df=dataframe)
  ```

  Perform principal component analysis for dimensionality reduction.

**For more advanced applications, see the tutotrial.**

## Data Dictionary

<table><thead><tr><th width="293">Name</th><th width="246">Description</th><th width="89">Type</th><th>Example</th></tr></thead><tbody><tr><td><code>ticker</code></td><td>Stock ticker symbol.</td><td>TEXT</td><td>"TSLA"</td></tr><tr><td><code>date</code></td><td>Record date.</td><td>DATE</td><td>2023-09-30</td></tr><tr><td><code>probability_light</code></td><td>LightGBM Boosting Model prediction.</td><td>FLOAT</td><td>1.46636</td></tr><tr><td><code>probability_convolution</code></td><td>CNN Model prediction for bankruptcies</td><td>FLOAT</td><td>0.135975</td></tr><tr><td><code>probability_rocket</code></td><td>Rocket Model prediction for time series classification</td><td>FLOAT</td><td>0.02514</td></tr><tr><td><code>probability_encoder</code></td><td>LightGBM and CNN autoencoders Model prediction.</td><td>FLOAT</td><td>0.587817</td></tr><tr><td><code>probability_fundamental</code></td><td>Prediction using accounting data only.</td><td>FLOAT</td><td>1.26148</td></tr><tr><td><code>probability</code></td><td>Average probability across models.</td><td>FLOAT</td><td>0.553823</td></tr><tr><td><code>sans_market</code></td><td>Fundamental prediction adjusted for market predictions.</td><td>FLOAT</td><td>-0.20488</td></tr><tr><td><code>volatility</code></td><td>Variability of model predictions.</td><td>FLOAT</td><td>0.62934</td></tr><tr><td><code>multiplier</code></td><td>Coefficient for model prediction calibration.</td><td>FLOAT</td><td>1.951868</td></tr><tr><td><code>version</code></td><td>Model/data record version.</td><td>INT</td><td>20240201</td></tr></tbody></table>

{% hint style="info" %}
When `sans_market` is <mark style="color:green;">positive</mark>, it means that the fundamentals show a larger predicted bankruptcy than what the market predicts **(stock might go down in medium term)** , when `sans_market` is <mark style="color:red;">negative</mark>, the market might have overreacted, and predict a larger probability of bankruptcy than what the fundamentals suggest **(stock might go up in medium term)**.
{% endhint %}

## Use Cases

1. **Bankruptcy Prediction Analysis**: Offer insights into predicted corporate bankruptcies and identify key factors, clarifying main drivers across different cycles.
2. **Variable Impact Breakdown**: Analyze how each individual variable affects bankruptcy predictions, providing in-depth feature contribution insights.
3. **Temporal Feature Distribution Analysis**: Reveal how variables contribute to predictions over time, emphasizing key features in forecasting models.
4. **Correlation Discovery**: Identify stocks with similar bankruptcy probability trends, revealing correlated market behaviors.
5. **Probability Shift Overview**: Showcase changes in bankruptcy probabilities among correlated stocks, providing a comprehensive market perspective.
6. **Sentiment Inversion Analysis**: Convert bankruptcy predictions into positive sentiment indicators to gauge potential impacts on stock returns.
7. **Behavioral Similarity Mapping**: Locate stocks with similar behaviors to a selected reference, based on bankruptcy trends and PCA feature analysis.


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.sov.ai/realtime-datasets/equity-datasets/bankruptcy-predictions.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.