JCR will help you build a mathematical model that fits your portfolio in the following steps:

[Step 1] Requirement Definition

You define what information to input into the model and what events the model will predict. At this stage, we also consider what tools will be needed when using the prediction results for decision-making.

If there is already an existing model or if there are many unknowns about the modeling method to be adopted, it is important to prototype a model based on sample data as a PoC (Proof of Concept) to get an idea in advance of what level of performance the model is likely to achieve.

[Step 2] Data Collection

Collect the data required for the model. Events such as bankruptcy do not occur very often, making the data valuable. It is important to prepare as many bankruptcy scenarios as possible, including digitizing paper documents if necessary.

[Step 3] Model Building

Based on the data collected in step 2, we define financial indicator formula for the explanatory variables of the model and build an analytical database for calculating the indicators. As part of the analysis, we first determine how much data is available for each model. Next, based on the index database that implements the financial indicator formula, we perform data cleansing to address missing or outliers in the data, and analyze the bankruptcy prediction ability of each indicator alone. Finally, we use statistical optimization software to fit the model to the data (multivariate regression) and finalize the model by exploring the combination of indicator, model parameters, and hyper-parameters that result in the best model performance.

[Step 4] Model Operation and Maintenance

We prepare a tool system so that end users can use the constructed model, and if necessary, we link it to be able to obtain input data from the company's existing financial database. We provide end users with an Model Construction Flow overview of the model so they can understand what's inside before they use it. In addition, even after the model goes live, we will regularly check the model's performance using the latest data.