How Vintage Cohort Analysis Transforms Financial Risk Management

Background

In modern fintech and consumer finance, the ability of marketing to acquire users and guide traffic directly impacts downstream risk control and approval processes. Balancing acquisition cost, marketing precision, and long‑term repayment behavior is essential for reducing overall operational risk.

What is Vintage Analysis

Vintage (or cohort) analysis tracks the performance of loans or assets over their lifecycle by grouping them into batches based on the month or quarter of issuance. By comparing delinquency rates, recovery rates, and other risk metrics across cohorts, firms can assess asset quality over time. The term originates from wine aging, where older vintages indicate more stable quality.

Key Terminology

Month on Book (MoB): the number of months since a loan was issued.

DPD (Days Past Due): days between the actual repayment date and the due date.

DPD0+, DPD30+: indicators of any delinquency and delinquency over 30 days.

M0‑M6: delinquency buckets (e.g., M0 = 1‑30 days, M6 = >360 days).

Application Scenarios

Vintage analysis is used in consumer finance, internet user retention, mortgage lending, and supply‑chain finance to monitor asset performance, identify risk spikes, and validate strategies. Specific use cases include monitoring new‑customer conversion rates (T0‑T30), credit‑line utilization, and sales‑channel effectiveness.

Framework Construction

The framework consists of three dimensions:

Row dimension : batch (e.g., loan issuance month).

Column dimension : MoB (account age).

Data area : conversion metric such as delinquency rate.

These map to a table (rows = batches, columns = MoB) and a corresponding curve (x‑axis = MoB, y‑axis = metric).

Table Decomposition Example

Using the wine analogy, the table maps industry, batch, time, quality, maturity, and influencing factors to their financial equivalents (loan month, MoB, delinquency rate, risk strategy, etc.).

Building the Vintage Record Table

Record each loan with its issuance month, MoB, and relevant metrics (e.g., delinquency flag). Example data: 202301 0.50% 1.20% 2.90% 4.20% Interpretation: values rise with MoB, indicating increasing delinquency as assets age.

Constructing Vintage Curves

Plot the table data with:

Label: issuance month (row label).

X‑axis: MoB.

Y‑axis: metric (e.g., delinquency rate).

The resulting curve shows a peak in the first 1‑3 months, a slight dip, then a gradual rise stabilizing after month 12.

Implementation Examples

Example 1 – First‑Loan Vintage

Build a report that tracks the conversion of first‑loan customers to repeat loans, capturing dimensions such as loan amount, interest rate, credit limit, and channel.

Example 2 – Login Retention

Treat user login events as a Vintage analysis: row dimension is the first login of a user in a period, column dimension is the retention interval (next day, 7‑day, 30‑day, etc.), and the metric is retention count or rate. The following SQL snippets illustrate table creation and data extraction:

-- create table mid_flow_login_app_remain_di ( ... )

select a.user_id, c.login_date, c.login_chan_no, c.login_chan_name, c.login_app_id, c.login_app_name, c.user_type, a.login_date as tn_login_date, DATEDIFF(a.login_date, c.login_date) as app_date_diff from ( ... ) a inner join ( ... ) c on a.user_id = c.user_id;

Effectiveness

Deploying Vintage analysis across multiple business lines (self‑served loans, platform loans, AB‑tests, dynamic MoB monitoring, login retention) provides a standardized, multi‑dimensional view of customer lifecycle value, enabling data‑driven strategy adjustments and improved resource allocation.

Issues and Outlook

Challenges include maintaining flexibility as new marketing scenarios emerge, avoiding report proliferation, and ensuring reusable, extensible MoB dimensions. Future plans focus on a modular data‑model approach that allows incremental dimension addition and better integration with the metrics platform.