Global Financial Cycle Transmission to
Indonesian Commercial Bank Balance Sheets

Bank-Level Panel Evidence, 2002–2025

Ahmad Fatikhul Khasan

HAMPARAN Institute · Jember, Indonesia

ORCID 0000-0003-0209-834X · [email protected]

Bali · 31 July 2026

The Puzzle

Why did the largest US tightening since 1980
produce credit expansion in Indonesia?

GFC 2008–09

The textbook risk-off

Credit growth: −11.5 log points**
NPL gross: +1.22 pp**
Foreign capital outflow
Bank-level signature: contraction

Fed hike 2022–23

Largest US tightening since 1980 — and yet…

Credit growth: +9.0 log points*
NPL gross: −1.29 pp**
No outflow, no panic
Bank-level signature: expansion

Polar opposite signatures. This paper asks why — and what it means for the framework.

Start with a puzzle. Between March 2022 and July 2023 the Federal Reserve raised the federal funds rate by five-and-a-quarter percentage points. That is the largest tightening cycle in forty years. What should we expect for a small open economy like Indonesia? The textbook answer — supported by the Global Financial Crisis of 2008 and by COVID-19 in 2020 — is contraction. Capital flows out, dollar funding tightens, credit growth drops, NPLs rise. Look at the numbers in the slide. In the 2008–09 episode, Indonesian commercial bank credit dropped by 11.5 log points. NPLs rose by 1.22 percentage points. That is the textbook signature of risk-off transmission. Now look at the 2022–23 Fed hike. Credit expanded by 9 log points. NPLs fell by 1.29 percentage points. The polar opposite. That is the puzzle this paper sets out to explain.

Contribution

Four contributions

1

First single-country, bank-level test

of asymmetric macroprudential effectiveness for Indonesia
2

23-year panel · 97 BUK · 2,055 bank-years

covers all six major risk-off episodes (GFC, Taper, CNY, EM 2018, COVID, Fed hike)
3

Identifies the operative shock

the corporate credit spread (BAA − 10Y), not the VIX
4

Quantifies the asymmetry

post-2017 sensitivity ≈ 4× the pre-2014 magnitude

Where the paper sits

Literature: the asymmetry test missing for Indonesia

Three strands frame the question:

▸ Cross-country evidence of asymmetric macroprudential effectiveness

Cerutti, Claessens & Laeven (2017); Akinci & Olmstead-Rumsey (2018) — tighter on downside than upside

▸ Bank-level transmission of global liquidity

Bruno & Shin (2015); Avdjiev, Du, Koch & Shin (2017) — leverage cycle at sub-annual frequency

▸ Indonesia-specific evidence

Warjiyo (2016); Basri (2017); Soedarmono et al. — aggregate-level or pre-2015 only

The gap → No single-country bank-level test of asymmetric effectiveness for Indonesia using the full post-Basel III regime.

Data

A 23-year bank-level panel from OJK CFS

commercial banks (BUK)

2,055

bank-year observations

17,377

parsed Excel files

99.93%

parse success rate

Outcomes

Balance sheet: log assets, log credit, log DPK (third-party deposits)
Risk: Capital Adequacy Ratio (CAR), Non-Performing Loan gross ratio
Exposure: FX off-balance-sheet commitments / total assets

Bridging (documented, validated)

Pre-2010 ↔ 2010+ label conventions for balance-sheet items
Pre-2021 ↔ post-2021 OJK templates for CAR/NPL: Pearson r = 0.9995 / 0.98
Bulanan ↔ triwulanan FX exposure cross-validation: Pearson r = 0.925

The Global Financial Cycle Factor

Composite factor: PC1 of five US risk-off indicators

PC1 loadings

(sign-aligned: positive = risk-off)

VIX+0.41

Broad USD−0.21

Fed Funds rate−0.55

US 10Y yield−0.41

BAA spread+0.56

48.2% variance

BAA spread is the largest positive loading → accommodation phase signature, not tightening

Empirical Strategy

Local projections, bank fixed effects

y_i,t+h − y_i,t = α_i + β_h · GFC_t + γ_h′ X_i,t−1 + ε_i,t+h

Outcomes: log credit, log assets, NPL gross, CAR · horizons h = 1, 2, 3, 5 years
Bank fixed effects α_i absorb time-invariant heterogeneity (business model, ownership, region)
Lagged controls X: log assets (size), DPK share (funding), credit share (asset mix)
SE clustered by bank · two-way bank + year as robustness
Shock: Q1 reading of GFC factor — aligns with April BUK snapshot date
Identification: GFC factor built entirely from US series → exogenous to single Indonesian bank

Baseline Result

Slow-build response: peak at h = 3, triad signature

Outcome	h = 1	h = 2	h = 3	h = 5
Δ log credit	−0.003 (0.005)	+0.016* (0.009)	+0.046*** (0.012)	+0.049*** (0.012)
Δ log assets	−0.001 (0.003)	+0.008 (0.005)	+0.030*** (0.007)	+0.030*** (0.009)
Δ NPL gross (pp)	+0.055 (0.040)	−0.034 (0.064)	−0.254** (0.126)	−0.307*** (0.117)
Δ CAR (pp)	+0.162 (0.217)	+0.120 (0.381)	−0.795** (0.333)	−0.498 (0.371)

At h = 3 — a one-SD GFC shock yields:

+4.6 log pts credit · +3 log pts assets · −25 bp NPL · −80 bp CAR

Triad: credit expansion + asset-quality improvement + capital compression. GFC factor here behaves as accommodation indicator (cheap $ funding + elevated risk premium).

Table 2 reports the baseline. Read the third column — that is the three-year horizon, which is when the response peaks. A one-standard-deviation move of the GFC factor is associated with a 4.6 log point increase in cumulative credit growth, a 3 log point increase in asset growth, a 25 basis point decline in the NPL ratio, and an 80 basis point compression of CAR. All four are significant at one or five per cent. Three things are striking here. First, the transmission clock is two to three years from shock to peak. That is slower than the leverage-channel literature would predict, and it reflects the annual frequency of the OJK panel plus the bank-intermediated structure of Indonesian credit. Second — and this is interpretively crucial — the GFC factor in this regression behaves like an accommodation indicator, not a tightening indicator. Cheap dollar funding plus elevated risk premium. Banks expand credit, asset quality improves because the lift-up reaches even weak borrowers, and CAR compresses because the lending boom outpaces capital generation. Third, there is a clean signature: simultaneous credit expansion, NPL improvement, and CAR compression. Call this the accommodation-phase triad.

Heterogeneity

Ownership dominates size — implication for D-SIB design

OWNERSHIP

State, BPD, private domestic

CAR support +1.6 to +2.0 pp during stress (vs foreign omitted) — signature of uniform regulator rekapitalisasi, not bank-specific bailout

FX EXPOSURE

Quartile 3–4

Assets contract −0.038 to −0.046 log pts — but loan book is protected. Trading book adjusts first, loan book last.

SIZE QUINTILE

Insignificant across all outcomes

Two banks of identical size but different ownership behave very differently. Current D-SIB designation is partly size-based — the data say size alone misses the relevant heterogeneity.

Heterogeneity tells a story that matters for the design of the macroprudential framework. First — the ownership channel. State banks, regional state banks, and private domestic banks all show a uniform CAR-support response of between 1.6 and 2 percentage points during stress, relative to foreign-owned banks. This is the signature of regulator-driven recapitalisation that operates uniformly across domestic ownership types. It is not a bailout specific to state-owned banks. Second — the FX channel. Banks in the top quartile of foreign-currency exposure contract their balance sheets — assets fall by about 4 log points — but their loan books are protected. This is sequencing: the off-balance-sheet trading book adjusts first; the loan book adjusts last. Third — and this is the most policy-relevant finding — size quintile is statistically insignificant across all four outcomes. Two banks of identical size but different ownership type respond very differently to global cycle shocks. The current OJK Domestic Systemically Important Bank designation is partly size-based. The data suggest size alone misses the relevant heterogeneity.

Centerpiece — The Episode Anomaly

Three regimes across six risk-off episodes

Episode	Regime	Δ log credit	Δ NPL (pp)
GFC 2008–09	Textbook	−0.115**	+1.22**
Taper tantrum 2013	Muted	+0.111***	—
CNY / commodity 2015–16	Muted	+0.065	—
EM contagion 2018	Muted	−0.033	—
COVID 2020	Textbook	−0.156***	—
Fed hike 2022–23	INVERSE ⚡	+0.090*	−1.29**

Reading the table

GFC 2008 + COVID: textbook risk-off contraction
2013 / 2015–16 / 2018: muted or absorbed
Fed hike 2022–23: polar opposite — credit expansion, NPL improved

The polar opposite of 2008–09.

Largest US tightening since 1980 — and the Indonesian banking system expanded credit and improved asset quality.

This is the centerpiece. Table 4 reports the bank-level event-study for each of the six pre-specified risk-off episodes, all at the two-year horizon. The episodes sort into three regimes. Two are textbook contractions. The GFC of 2008–09 produced an 11.5 log point drop in credit and a 1.22 percentage point rise in NPLs. COVID-19 in 2020 produced an even larger 15.6 log point credit contraction. Three episodes are statistically muted — Taper, CNY, and EM 2018. And then there is the Fed hike of 2022–23. Credit expanded by 9 log points. NPLs fell by 1.29 percentage points. CAR fell by 3.5 percentage points — not statistically significant but pointing in the same direction. The polar opposite of 2008–09. The largest tightening cycle since the 1980s, and the Indonesian banking system expanded. This is what asymmetric macroprudential effectiveness looks like at the bank level.

Robustness

Pre-2014 → Post-2017: ~4× amplification

6 robustness checks on β(h=3)

Baseline: +0.046*** (0.012)

R1 + macro controls+0.047***

R3 two-way SE+0.046***

R4b VIX alone+0.002 (n.s.)

R4c BAA spread alone+0.047**

R5a pre-2014+0.033***

R5b post-2017+0.130***

R6 trim p99+0.040***

BAA spread = operative shock
Post-2017 ≈ 4× pre-2014

Interpretation

The asymmetry: what the instruments bind on

✓ Downside contained

2022–23 Fed hike: credit expanded, NPL improved
CAR support uniform across state, BPD, private domestic
Orderly FX deleveraging — trading book first, loan book last
LTV, CCyB, loss-provisioning bind on losses & on CAR

✗ Upside not contained

Pre-2014 β ≈ 0.033 → Post-2017 β ≈ 0.130 (≈ 4× amplification)
During accommodation: NPLs fall, CAR comfortable
Median bank CAR = 22% against 8% regulatory floor — tools do not bind
Credit boom propagates; exposures build up

Why does this asymmetry exist? It is a question of what the instruments bind on. The post-2016 framework — and this is also the broad shape of Basel III plus emerging-market variations — relies on instruments that bind when losses crystallise or when capital ratios approach minimums. The loan-to-value ratio binds at origination but only matters when prices fall. The countercyclical capital buffer binds when reserves are depleted. Loss-provisioning standards bind when NPLs are high. During an accommodation phase — cheap dollar funding, low credit spreads — NPLs are falling. CAR is comfortable. The median bank in this sample carries a CAR of 22 per cent against an 8 per cent regulatory floor. None of those instruments bind. So the credit boom propagates. The exposures build up. And when the cycle turns, the instruments engage on the way down — successfully, as 2022–23 shows. But the build-up of those exposures during the up-phase was not contained. Downside management is the framework's strength. Upside management is its gap.

Policy Implications

Four instruments to close the upside gap

1

CCyB triggered by BAA spread

Use the corporate credit spread (the operative shock identified by the data), not VIX, not domestic credit-to-GDP gap alone
2

Exposure-cap tools (GWM Valas)

Bind on exposure, not loss — exactly what loss-recognition tools fail to do during accommodation
3

D-SIB designation by ownership

Ownership type dominates size in heterogeneity — current size-based designation misses the relevant differentiation
4

Preserve forbearance budgets

The uniform CAR-support signature in the data is consistent with forbearance and should be retained as a deliberate framework feature

Four concrete instruments, each addressing the upside gap. First, calibrate a surge-cyclical countercyclical buffer that triggers on the BAA corporate credit spread, not on the VIX, and not on domestic credit-to-GDP alone. The BAA spread is the leading indicator the data identify as the operative shock. Second, expand exposure-cap tools — Bank Indonesia's GWM Valas, foreign-currency commitment caps, surge-cyclical rate hikes on FX liabilities. These bind on exposures, which is exactly what loss-recognition tools fail to do during accommodation. Third, redesign D-SIB designation around ownership type, not size. The data show ownership dominates size in heterogeneous response. State banks, BPDs, and private domestic banks behave differently from foreign-owned banks regardless of size. Fourth, preserve forbearance budgets that allow uniform loss-recognition timing across domestic ownership types. The CAR-support signature in the heterogeneity table is consistent with this practice and should be retained as a deliberate framework feature.

Limitations · Next Steps

Honest limitations point to follow-up work

Limitations

Annual frequency: CAR/NPL only at Q1 (April) snapshot — sub-annual dynamics blurred
2016 structural gap in OJK public archive — absorbed into year fixed effects
Ownership classification static (2021–2024 top-owner average) — pre-2021 M&A not captured
No loan-level granularity — borrower-side mechanism unobserved

Next steps

Quarterly extension once 2021+ triwulanan sample matures
Historical-ownership robustness using OJK annual reports
Syariah banking system extension (BUS panel from 2010)
Loan-level micro-data via SLIK access (regulatory permission required)

Conclusion

Indonesia’s post-2016 macroprudential framework
contains the downside of the global cycle
but not the upside.

The build-up of exposures during accommodation phases
remains an open margin for policy design.

Thank you. I welcome your questions.

Ahmad Fatikhul Khasan · HAMPARAN Institute

[email protected] · ORCID 0000-0003-0209-834X