Markov chain migration model across 6 delinquency states — 27 months including COVID stress period
Monthly balance-weighted transition matrices computed for CD (Credit Card, unsecured), TW (Two-Wheeler, motorbike collateral),
and CLW (Cash Loan Working Capital, hybrid). WriteDown states are absorbing.
COVID spike visible in Jul–Oct 2021 across all I/P ratio trends — CLW deteriorated earliest (Jun 2021), consistent
with working capital/SME borrowers feeling lockdown stress before retail products.
ECL computed on full outstanding (Principal + Interest + Penalty) using 12-month cumulative PD from T¹².
CD — Portfolio Composition Over Time
Principal balance by state (B VND)
TW — Portfolio Composition Over Time
Principal balance by state (B VND)
CLW — Portfolio Composition Over Time
Principal balance by state (B VND)
Interest + Penalty / Principal Ratio — All Products
ECL as % of total EAD (P+I+N) · includes 12m PD × LGD
Average monthly transition matrix — balance-weighted roll-rate across 26 consecutive month pairs
Each cell T[i→j] = fraction of balance that was in state i and migrated to state j the following month.
WriteDown states are absorbing (once written off, stays written off).
T¹² = matrix raised to the 12th power = cumulative probability of reaching each state after 12 months from any starting point.
Use T¹² rows to read: "if a loan starts in Current today, what is the probability it reaches WD in 12 months?"
Monthly T — One-Step Transition (avg over 26 months)
T¹² — 12-Month Cumulative Transition
Monthly T — Heatmap (diagonal = stay rate)
Darker = higher probability
T¹² — 12-Month Heatmap
Cumulative probabilities after 12 months
Write-Down Probability in 12 Months — By Starting State
12-month forward projection from Jul 2022 baseline — applying T¹² to current portfolio balances
Starting from the latest snapshot (Jul 2022), the portfolio balance in each state is projected forward
12 months by repeatedly applying the average transition matrix. This gives an expected portfolio
composition in Jul 2023 under stable transition behaviour (no macro shocks assumed).
WriteDown balances are cumulative and absorbing — they only increase.
Portfolio State Distribution — Month 0 vs Month +12
Balance by state at baseline vs projected 12m forward
Portfolio Flow — All 12 Months
How each state evolves month by month
WriteDown Balance Growth — All 3 Products (+12 Months)
Projected WD accumulation from Jul 2022 baseline · absorbing state compounds each month
ECL = PD (T¹²) × LGD × EAD — computed on total outstanding including interest and penalty
PD = probability of reaching WriteDown from each IFRS 9 stage within 12 months, derived from T¹².
LGD from provision rates: CD=88.3%, TW=84.7%, CLW=93.9% (from internal New WD sheet).
EAD computed twice: Principal only (conservative) and Principal+Interest+Penalty (full exposure).
Including I+P adds +137B VND (+6.8%) to total ECL — TW shows largest uplift (+10.1%) due to
high interest concentration in delinquent buckets (12–13% of total balance).
ECL Breakdown — Principal vs Total EAD Base
Δ = incremental ECL from including interest + penalty
Penalty & Interest Concentration by State
% of total balance that is interest or penalty — by state & product
I/P ratios are statistically elevated in months preceding portfolio deterioration
For each product, months where balance migrated to a worse bucket were identified.
The Interest/Principal and Penalty/Principal ratios in those months were compared to stable months.
CLW shows the strongest combined signal: interest ratio +36.1% and penalty ratio +33.9% above normal
in migration months. For CD, only the penalty signal is significant — interest stays flat because
revolving credit accrues interest uniformly regardless of DPD status.
Practical use: monitor these ratios monthly as an early warning dashboard.
I/P Ratio — Migration vs Stable Months
Grouped comparison: ratio in months before deterioration vs stable months
Penalty/Principal Ratio — COVID Stress Visible
Monthly trend · spike = deterioration signal
Signal Strength — % Elevation Before Migration
How much higher the ratio is in migration months vs stable · larger = stronger signal