Risk Automatic Prediction

Predict the risk of bankruptcy of a Social Economy Enterprise (SEC).

Risk Automatic Prediction

Predict the risk of bankruptcy of a Social Economy Enterprise (SEC).
Inspect every SEC is not possible.

Risk Automatic Prediction

Predict the risk of bankruptcy of a Social Economy Enterprise (SEC).
Inspect every SEC is not possible.
Supervise each SEC is a time consuming task.

Risk Automatic Prediction

Predict the risk of bankruptcy of a Social Economy Enterprise (SEC).
Inspect every SEC is not possible.
Supervise each SEC is a time consuming task.
How to automatically evaluate the risk of each SEC?

Risk Automatic Prediction

The components of CAMELS are:

(C)apital adequacy
(A)ssets
(M)anagement capability
(E)arnings
(L)iquidity
(S)ensitivity

Risk Automatic Prediction

The components of CAMELS are:

A scale of 1 implies that a bank exhibits a robust performance, is sound, and complies with risk management practices.
A scale of 2 means that an institution is financially sound with moderate weaknesses present.
A scale of 3 suggests that the institution shows a supervisory concern in several dimensions.
A scale of 4 indicates that an institution has unsound practices, thus is unsafe due to serious financial problems.
A rating of 5 shows that an institution is fundamentally unsound with inadequate risk management practices.

Risk Automatic Prediction

The problems with CAMELS computation:

We do not have the estimation until the company send the information to review
We can not predict what would be the CAMELS value of a company in the next period

Enterprise Properties Variables

Type of organization
Whether it is a cooperative, fund or other type of organization.
Type of company: Multi-active, employee funds.
Type of supervision: 1,2,3.
Group Niif of the organization Department of the organization
Municipality of the organization
Category of the organization

Number of associates
Number of employees
Number of offices
Number of correspondents
Number of savers
Number of debtors
Total assets of the organization

Risk and Camels Variables

Consolidated risk rating
Risk rating
Camel rating
Credit risk

Liquidity risk
Operational Risk
Sarassoft risk

Variational Variables

Variation between two consecutive periods of associates
Variation between two consecutive periods of employees
Change between two consecutive periods in the number of offices
Change in the number of savers between two consecutive periods
Variation between two consecutive periods in the number of debtors.

Change between two consecutive periods in total assets
Variation between two consecutive periods of gross portfolio
Variation between two consecutive periods of the consumer portfolio
Variation between two consecutive periods of the housing portfolio

Bankruptcy Risk Model

Experimental Setup: Dataset

Window Model

Final Proposed Model

Accuracy Results

\[ \begin{array}[c]{|c|c|c|c|c|} \hline Model & Random & Neural & SVM & Logist \\ &Forest & Networks & & Regression\\ \hline \text{Chart of account (CA)} & \bf{72\%} & 69.3\% & 70.1\% & 65.5\% \\ \text{Descriptive variables (DV)} & & & & \\ \text{CA + DV} & \bf{70.95\%} & 68.3\% & 68.2\% & 63.5\%\\ \text{Variation between periods (VBP)} & & & & \\ \text{CA + DV + VBP} + & \bf{76.66\%} & 75.5\% & 74.5\% & 70.3\%\\ \text{CAMELS/RISK + MI} & & & & \\ \hline \end{array} \]

Confusion Matrix Results

\[ \begin{array}{|c|c|c|c|c|c|} \hline & \textbf{1} & \textbf{2} & \textbf{3} & \textbf{4} & \textbf{5} \\ \hline \textbf{1} & \textbf{14%} & 79\% & 7\% & & \\ \hline \textbf{2} & & \textbf{80%} & 20\% & & \\ \hline \textbf{3} & & 33\% & \textbf{67%} & 1\% & \\ \hline \textbf{4} & & 7\% & 84\% & \textbf{9%} & \\ \hline \textbf{5} & & 100\% & & & \textbf{0%} \\ \hline \end{array}\]

Developed API

Conclusions

A machine learning model is proposed to assess risk

Conclusions

A machine learning model is proposed to assess risk
The use of inter-period variations improves the accuracy of the model

Conclusions

A machine learning model is proposed to assess risk
The use of inter-period variations improves the accuracy of the model
The addition of macroeconomic variables improves the accuracy and recall of the model

Conclusions

A machine learning model is proposed to assess risk
The use of inter-period variations improves the accuracy of the model
The addition of macroeconomic variables improves the accuracy and recall of the model
Oversampling and undersampling is a neccesary step