Home / Use cases

Valcision use cases

Credit-risk modeling

Connect borrower data, default assumptions, and model performance in a visible workflow. Develop analysis your team can review, challenge, and revisit as portfolio conditions change.

Understand what drives credit outcomes.

A credit model is most useful when the path from source data to a result is clear. Valcision combines data preparation, statistical modeling, and reporting so analysts can examine the relationships behind their estimates.

  • Which borrower or loan characteristics are associated with observed defaults?
  • How does model performance vary across segments and observation periods?
  • How sensitive are estimated losses to changes in default, recovery, or exposure assumptions?
  • What changes when new data or revised assumptions enter the workflow?

Build a defensible analytical dataset.

Borrower and exposure information

Bring relevant borrower attributes, loan balances, product classifications, and observation dates. Define the unit of analysis and align records before fitting a model.

Outcomes and assumptions

Supply a consistent default definition, an observation horizon, and historical outcomes. For loss estimates, add exposure and recovery assumptions that match that horizon.

Check missing values, duplicated records, unusual observations, and differences between segments. Keep information available only after the prediction date out of model features.

Explore the demo workflow

This flowchart is rendered from a saved Valcision application demo.

Import synthetic loans, create seeded development, validation and test partitions, fit a CHAID tree on development data, then score and report the held-out samples.
Import synthetic loans, create seeded development, validation and test partitions, fit a CHAID tree on development data, then score and report the held-out samples.

Illustrative workflow using synthetic demo inputs. Click the chart to view it at full size.

From records to a model you can examine.

  1. Prepare and segment. Connect data sources, clean and transform fields, and create analytical segments using the visual workflow and available Python or SQL capabilities.
  2. Develop a default model. Use statistical modeling, including logistic regression for binary outcomes, to examine relationships between selected features and observed defaults.
  3. Evaluate predictions. Review results against actual outcomes. With supplied binary probabilities, back-testing diagnostics include log loss, Brier score, and AUC where the test sample supports them.
  4. Connect probabilities to loss assumptions. Build an expected-loss calculation using probability of default, loss given default, and exposure at default, with explicit units and time horizons.
  5. Compare and communicate. Summarize results by segment, compare alternative assumptions, and retain model snapshots when using supported model providers.

Each stage makes a different question visible: data quality, model fit, predictive performance, loss assumptions, or the interpretation of results.

Illustrative use case

A credit analytics team wants to understand how estimated one-year losses on a commercial-loan portfolio change when default and recovery assumptions become more adverse.

The fictional portfolio

The portfolio contains 1,000 loans with a combined exposure at default of $100 million. For this simplified example, all loans share the same probability-of-default and loss-given-default assumptions. Exposure is held constant across scenarios.

All figures below are invented to illustrate the calculation. They are not fitted model results, a customer case study, or a forecast.

1. Prepare data and define the outcome

The team imports loan identifiers, observation dates, balances, borrower characteristics, and historical default outcomes. It defines a one-year default horizon, checks missing and duplicate records, and separates development data from a later evaluation period. Features must have been available at the prediction date.

2. Develop and evaluate a default model

The team develops a logistic-regression model using the selected borrower and loan features. It scores later observations and reviews probability estimates against actual defaults using appropriate performance diagnostics. It examines results by segment before deciding whether the model is suitable for the intended analysis.

The scenario probabilities below are supplied assumptions for the illustration; they do not imply that a model has been trained or validated on this fictional portfolio.

3. Compare baseline and adverse loss assumptions

One-year assumption or resultBaselineAdverse
Exposure at default (EAD)$100 million$100 million
Probability of default (PD)1.5%3.0%
Loss given default (LGD)40%50%
Expected loss$600,000$1,500,000

Follow the calculation

Expected loss = EAD × PD × LGD

Baseline: $100,000,000 × 0.015 × 0.40 = $600,000.

Adverse: $100,000,000 × 0.03 × 0.50 = $1,500,000.

The adverse estimate is $900,000 higher, a 150% increase from the baseline. Holding baseline LGD constant while increasing PD gives $1.2 million; applying the adverse LGD then raises the estimate to $1.5 million. This ordered comparison makes the effect of each assumption visible.

4. Review the drivers and communicate the result

In Valcision, the team connects data preparation, model scoring, the loss calculation, and reporting in a visual workflow. Reviewers can trace the estimate to the exposure, probability, and recovery assumptions rather than seeing only a final number.

  • Compare segments: repeat the calculation with loan-level or segment-specific assumptions and sum the individual expected losses.
  • Challenge assumptions: examine why PD or LGD changes, whether the horizon is consistent, and how exposure might change in practice.
  • Document the comparison: report baseline and adverse estimates alongside the evaluation findings and scenario definitions.

Limitations of the example

This example connects credit-model outputs and loss assumptions to a portfolio estimate. Expected credit loss, however, is an average under the chosen assumptions. It does not measure the full loss distribution, unexpected loss, tail risk, or regulatory capital.

Valcision can help bridge the gap through its Monte Carlo simulation engine.

Make the findings useful for review.

  • Prepared and scored datasets that preserve the connection to source records.
  • Model diagnostics and performance summaries for the chosen evaluation sample.
  • Segment-level comparisons of probabilities and estimated losses.
  • Tables, visualizations, and reports showing how results change with assumptions.

Use these outputs to support model review, portfolio analysis, and discussions about exposure. The interpretation remains tied to the data, default definition, horizon, and methods your team selects.

Common questions

Does Valcision supply a ready-made credit model?

The workflow supports model development and analysis using your data and assumptions. A suitable model must be specified, evaluated, and reviewed for the portfolio and purpose at hand.

How is expected loss different from a tail-risk estimate?

Expected loss summarizes an average loss under the selected assumptions. Tail-risk analysis examines adverse outcomes in a loss distribution. Explore the capital-at-risk use case to see how scenario aggregation and allocation fit into that analysis.

What does back testing establish?

It compares predictions with observed outcomes over the chosen test periods. It helps assess performance but does not establish that external predictions were trained without future information or that a model will perform similarly in new conditions.

Bring your analytical need into view.

Share your data, modeling question, and the decisions your team needs to support.

Discuss your use case