Research on alternative credit signals, thin-file scoring methods, and lending market dynamics across Vietnam, Indonesia, the Philippines, and Japan.
Vietnam's credit bureau covers fewer than one-third of adults, yet mobile wallet transaction history is near-universal among urban borrowers. We ran the numbers on what cash-flow feature sets tell you that bureau data can't.
Read articleOJK-licensed neobanks are growing fast in Indonesia, but most still rely on bureau lookups that miss 60% of the addressable market. Here is what alternative signals can and can't fix.
ReadGig workers in Southeast Asia have irregular income that confuses standard credit models. This post explains how feature engineering around income periodicity changes the picture.
ReadMany alternative credit providers ensemble multiple models to reduce variance. We don't. This post explains why market-specific signal extraction outperforms cross-market averaging for thin-file populations.
ReadTransfer-and-repay velocity in mobile wallets encodes financial discipline that no bureau captures. Here is the evidence from 18 months of backtesting across three markets.
ReadThe Philippines Credit Information Corporation has improved bureau coverage significantly, but for OFW dependents and rural informal workers, the gap remains large. Here is the size of the problem and what fills it.
ReadAgricultural workers in Vietnam and the Philippines have income concentrated in harvest windows. Standard credit models penalize this volatility unfairly. Here is how we adjust for seasonality.
ReadCredit models trained on historical data degrade faster in emerging markets due to rapid economic shifts. Here is how Panthera detects and corrects for distribution shift before it affects lender portfolios.
ReadElectricity and gas payment history from TEPCO and Tokyo Gas gives a reliable behavioral signal for Japan's growing population of gig and freelance workers outside formal employment.
ReadJapan's micro-SME lending market is growing as regional banks retreat from small-ticket commercial lending. Here is how alternative data can fill that gap while meeting FSA documentation expectations.
ReadAlternative data sources have widely varying refresh cycles. This piece quantifies how data staleness affects model Gini and what minimum freshness thresholds look like per signal type.
ReadAn overview of the credit gap in Asia as of early 2025: bureau coverage by market, the role of mobile-first financial infrastructure, and why building for the unbanked requires a different architecture.
ReadRegulatory bodies in all four markets require lenders to explain credit decisions to applicants. Here is how we engineer reason codes that are both statistically grounded and regulator-legible.
Read
