KARISMA OJK Paper Award 2026 · Jakarta · October 2026

Substitutes, Not Complements

Digital Adoption and Formal Financing in Indonesian Micro-Small Manufacturing
Ahmad Fatikhul Khasan
HAMPARAN Institute · Jember, Indonesia
[email protected] · ORCID 0000-0003-0209-834X
MSMEs in the Digital Ecosystem to Accelerate Economies of Scale
Motivation

The complementarity premise

Since 2020, Indonesian MSME policy has quietly bundled two levers on the assumption they are complements.

  • KUR (Kredit Usaha Rakyat) — subsidised bank credit for MSMEs
  • OJK financial-inclusion roadmap
  • Kemenkop UMKM digitalisation initiatives
  • Kartu Prakerja + digital-adoption modules

All bundle digital adoption and formal financing, implying the two reinforce each other in the field.

The unanswered question

"Does the interaction of digital adoption and formal financing actually deliver more than the sum — or does one substitute for the other?"

Nobody has tested this at the firm level in Indonesian manufacturing.

Literature

Three strands, one gap

Digital adoption & MSE scale

Goldfarb & Tucker (2019); Hjort & Poulsen (2019); Viollaz (2019); Falentina et al. (2021).

Documents positive scale effects. Rarely conditions on financing access.

Access-to-finance & growth

Beck, Demirgüç-Kunt & Levine (2006); Banerjee & Duflo (2014); Tambunan (2017, 2019).

Documents positive growth effects. Rarely conditions on digital adoption.

Digital × Finance interaction

Suri & Jack (2016); Jack & Suri (2014); Tang (2019); Cornelli et al. (2023).

Studies interaction — but in African households or U.S. small business, not Indonesian manufacturing MSEs.

The Indonesian gap: no published firm-level test of the digital-times-financing interaction across the full manufacturing MSE population. Our paper closes that gap.

Contribution

Three contributions

  • 1
    Data contribution
    Firm-level analysis on the near-universe of Indonesian manufacturing IMK — 94,779 firms across 23 two-digit KBLI categories from BPS VIMK 2022 Semester 2 dissemination microdata.
  • 2
    Method contribution
    Five-channel decomposition of digital adoption (promotion, sales, purchasing, fintech P2P, information search), plus KBLI × KLAS sector-size fixed effects with cluster-robust standard errors.
  • 3
    Identification contribution
    Baseline OLS replicated under propensity-score matching, inverse probability weighting, and a doubly-robust estimator — three causal-style estimators converging on the same coefficient bounds observable selection.
Data

BPS VIMK22-S2 firm-level microdata

94,779
Firms in analytical sample — the near-universe of Indonesian manufacturing IMK
23
Two-digit KBLI categories (manufacturing 10–33)
122
Variables covering digital adoption, financing structure, workers, revenue, cost, profit, controls
4.34M
Population weight target — projected via BPS sampling weights

Two data constraints shape identification:

  • BPS suppresses province and lower geographic codes in the dissemination file (confidentiality) — no spatial FE below the sector-size cell.
  • VIMK is cross-sectional; retrospective items give only limited within-firm temporal variation.

Source: BPS Survei Industri Mikro dan Kecil 2022 Semester 2 dissemination microdata.

Method

Baseline specification (Eq. 1)

log(Yi) = β·di + γ·fi + δ·(di × fi) + Xi'θ + αk(i),s(i) + εi

  • Yi — outcome: revenue, cost, profit, employment, or revenue per worker
  • di — digital-adoption indicator (composite or per-channel)
  • fi — formal-financing indicator (any_formal_loan, kur_user, or bank_user)
  • δ — the coefficient of primary interest: the digital × formal interaction

Fixed effects and SE structure

  • αk(i),s(i) — KBLI × KLAS sector-size cell FE (46 cells)
  • Xi — 11 firm-level controls (entrepreneur, firm, investment)
  • SE — cluster-robust at KBLI × KLAS; wild-cluster bootstrap for selected specs
Method (cont.)

Three selection-robustness estimators

PSM

Nearest-neighbour with caliper 0.05, one-to-one matching on 12 controls + KBLI × KLAS dummies.

Balance target: |SMD| < 0.10 on all covariates.

IPW

Inverse probability weighting using logistic propensity score, truncated at 1% tails to bound extreme weights.

Common support: [0.02, 0.99].

Doubly Robust

Regression adjustment on the IPW-weighted sample — consistent if either the outcome model or the propensity model is correct.

AUC of propensity model: 0.733.

Balance diagnostic: 99.97% of firms within common support; all 12 covariates achieve |SMD| < 0.10 after matching.

Results — Descriptive

The digital and financing landscape

Digital adoption rate by channel
Figure 1. Digital adoption rate by channel, Indonesian IMK 2022.
IndicatorShare
Any digital adoption38%
Online sales29%
Online purchase20%
Fintech P2P borrowing0.3%
KUR user4.9%
Non-KUR bank loan3.5%
Self-financed87%

WhatsApp and marketplace dominate the sales-platform mix. Fintech remains marginal.

Results — Baseline

Baseline OLS on log(Revenue)

Variable M1 M2 M3 M4 (main)
digital_any +0.691 +0.657 +0.690 +0.483
any_formal_loan — +0.634 +0.795 +0.537
digital × formal — — −0.316 −0.258
Firm controls (11) — — — ✓
KBLI × KLAS FE (46) ✓ ✓ ✓ ✓
R² 0.201 0.280 0.328 0.457

All coefficients in M4 significant at p < 0.001. Cluster-robust SE at KBLI × KLAS (46 clusters).

The Central Finding

Substitutes, not complements

Firm-typeEffect on log(Revenue)Interpretation
Digital only+62%Independent effect
Formal financing only+71%Independent effect
Both (observed)+114%Empirical joint effect
Both (log-additive counterfactual)+177%If truly complementary
−63pp
The substitution gap — the joint intervention absorbs ~19pp of the additive benchmark.

Digital and formal financing have independent scale-enhancing channels — but they do not compound as policymakers have been assuming. They partially substitute.

Robustness

27 specifications, 27 concur

Robustness variantδp-value
Main (M4 baseline)−0.258<0.001
Mikro subsample (88,918)−0.230<0.001
Kecil subsample (5,627)−0.2670.018
Sample-weighted (pop. 4.34M)−0.2440.002
Leave-one-KBLI-out (5 sectors)−0.223 to −0.307all sig.
Wild cluster bootstrap−0.2580.001

Selection-robustness estimators

  • OLS + controls: +0.483 (digital main)
  • PSM matched ATT: +0.397
  • IPW-ATE: +0.559
  • Doubly-robust: +0.499

All four estimators lie within the same magnitude. Selection on observables is modest.

Nuance

Not all digital is equal

Per-channel effect forest plot
Figure 3. Per-channel effect on log(Revenue) — forest plot.

Transactional channels dominate

ChannelOn RevenueOn Employment
Online sales+31%+8%
Online purchase+30%+11%
Digital promotion+9%n.s.
Fintech P2Pn.s.+9.1%
Info searchn.s.n.s.

Fintech P2P raises workers but not revenue — survival capital, not growth capital.

Limitations

Three limitations we own

  • 1
    No sub-national geography
    BPS suppresses province and lower codes in the dissemination file. Sub-provincial policy inference requires follow-up with restricted-access data.
  • 2
    Manufacturing only
    VIMK covers KBLI 10–33. Results do not generalise to trade, agriculture, mining, services, or transportation — which together form a larger share of Indonesia's MSE population.
  • 3
    Selection on observables
    We control 11 firm characteristics and 46 sector-size cells; PSM, IPW, and doubly-robust converge on OLS. But no IV, no natural experiment. Estimates are tightly-controlled correlations, not point-identified causal effects.
Policy

Four concrete levers for OJK

  • 1
    Abandon the universal-complementarity premise
    Programme design should stop bundling digital + financing on the assumption that stacking multiplies gains. Both are independently valuable, but the combination delivers less than the sum.
  • 2
    Target by firm-type, not universal push
    Firms with neither digital nor formal financing gain the most from any first intervention. Segmentation — not universal push — maximises the marginal Rupiah of programme spending.
  • 3
    Reformulate fintech P2P as employment-stabilisation
    Fintech P2P raises workers but not revenue — a survival-capital signature. Product design, disclosure, and consumer-protection frameworks should acknowledge this rather than market it as a growth tool.
  • 4
    Invest in transactional-digital channels
    Online sales and online input purchase carry the bulk of the digital scale premium. Promotion and info search are smaller levers — programme budget should reflect the per-channel effect asymmetry.
Synthesis

One finding, actionable at the programme level

94,779 firms. One specification. 27 robustness variants. Three selection-robustness estimators.
All converge on one finding: digital adoption and formal financing are partial substitutes, not complements. The joint intervention delivers less than the sum. Programme design has been assuming otherwise — and the marginal Rupiah of MSME spending can be redirected accordingly.

Thanks

Badan Pusat Statistik for the VIMK dissemination programme; KARISMA OJK organising committee; HAMPARAN Institute colleagues for the analytical infrastructure.

Correspondence

Ahmad Fatikhul Khasan
HAMPARAN Institute, Jember, Indonesia
[email protected]
ORCID 0000-0003-0209-834X

Replication package: HAMPARAN research repository.