Credit risk analyst role: roll rates, vintages and policy rules

Where the role sits inside a lender

Lenders typically split risk work across a few teams. Underwriting decides on individual applications. Policy and portfolio risk, where most credit risk analysts sit, decides the rules underwriting follows and watches how approved loans behave. Model development builds scorecards and, where the company follows Ind AS 109, expected credit loss models. Collections analytics works on recovering overdue amounts. Titles vary, so read the responsibilities.

A month, not a week

Credit risk runs on a monthly cycle because loans are billed monthly.

• Days 1–5: month-end data lands. Refresh delinquency buckets (current, 1–30, 31–60, 61–90, 90+ days past due) for the whole book.
• Days 5–10: update vintage curves, which show what share of loans disbursed in each month reached 30+ or 90+ days past due by each month on book.
• Days 10–15: roll-rate analysis and the monthly risk review for the chief risk officer.
• Rest of the month: policy work. A business team wants to lend to a new segment; you estimate its risk from comparable past loans or bureau data, and propose guardrails.

The artefact: a roll-rate matrix

A roll rate shows, for loans in each bucket last month, where they are this month. It is the quickest early warning in the portfolio. The example below uses synthetic data and runs with pandas.

import pandas as pd

# Synthetic snapshot: each loan's delinquency bucket in two consecutive months
data = {
    "loan_id": range(1, 21),
    "bucket_aug": ["0", "0", "0", "0", "0", "0", "0", "0", "1-30", "1-30",
                   "1-30", "1-30", "1-30", "31-60", "31-60", "31-60", "61-90", "61-90", "0", "0"],
    "bucket_sep": ["0", "0", "0", "0", "0", "0", "1-30", "1-30", "0", "1-30",
                   "31-60", "31-60", "0", "31-60", "61-90", "0", "90+", "61-90", "0", "1-30"],
}
df = pd.DataFrame(data)
order = ["0", "1-30", "31-60", "61-90", "90+"]
df["bucket_aug"] = pd.Categorical(df["bucket_aug"], categories=order, ordered=True)
df["bucket_sep"] = pd.Categorical(df["bucket_sep"], categories=order, ordered=True)

counts = pd.crosstab(df["bucket_aug"], df["bucket_sep"], dropna=False)
rates = counts.div(counts.sum(axis=1), axis=0).round(2)  # 90+ row is empty here, so it shows NaN
print(counts)
print(rates)

# Forward roll: share of each bucket that moved to a worse bucket
def forward_roll(row_bucket):
    row = counts.loc[row_bucket]
    worse = row[order[order.index(row_bucket) + 1:]].sum()
    return round(worse / row.sum(), 2) if row.sum() else None

for b in ["0", "1-30", "31-60", "61-90"]:
    print(b, "forward roll:", forward_roll(b))

On this data, 3 of 10 current loans rolled into 1–30 (a forward roll of 0.3), and 2 of 5 loans in 1–30 rolled to 31–60 (0.4). In a real review you would compare these to previous months and split by product, channel and sourcing branch. A jump in the current-to-1–30 roll for one sourcing channel is often the first sign of a quality problem there, months before losses show up.

Amounts matter as much as counts. Real roll-rate reports usually show both, because a few large loans can dominate the rupee picture.

Loss rates, approval rates and policy calls

A credit risk analyst is judged on the accuracy and timeliness of the monthly risk pack, whether early warnings were raised before losses materialised, the quality of policy recommendations (did an approved change perform as projected?), and how clearly trade-offs between approval rate and risk were explained to the business.

Risk heads, collections and the credit committee

You usually report to a risk manager or head of portfolio risk, under the chief risk officer. You work with business and product heads who want higher approvals, underwriting and collections teams, finance for provisioning, and internal audit, who will ask you to reproduce any number in the pack.

Scorecards, bureau data and portfolio monitoring

Learning the book before changing the policy

Month one: reproduce last month's delinquency buckets from raw tables and reconcile to the official pack. Month two: own the vintage curves and write short commentary on anything unusual. Month three: take one policy question end to end, from data pull to a recommendation with a projected approval and risk impact.

Policy manager, model developer or validator

Senior risk analyst, policy manager, risk model developer or validator, or collections strategy lead. Analysts who add strong modelling often move towards fraud analytics or data science roles in lending. The Data Science programme covers the classification, calibration and evaluation methods behind scorecards, with live online sessions and optional Noida labs.