Regulatory tracker dashboard

The 10 Jan 2025 OJK takeover — now estimable, with a caveat

The Bappebti monthly pre-period (Jan 2023 – Dec 2024) was recovered from reported monthly figures and reconciled cumulative totals, giving n = 45 monthly observations (24 pre, 21 post). Six pre-period months are imputed residuals (see Market series tab). Two specifications were run on this exact series; the second is the right one.

Specification 1: level-shift ITS — misspecified, do not quote

Log values with the naive fit (grey). The level-shift form produces a huge "effect" at any date you choose — the steep 2024 recovery makes every date look like a break, and fake event dates prove it. The number is deliberately not quoted here; the full table sits in the data catalog, with the placebo test that disqualifies it.

Specification 2: growth-rate break — the defensible one REAL ESTIMATE

Month-over-month growth. Green: pre-transfer; orange: post; dashed lines: each period's mean. The gap between them is the "break" — and it is not statistically significant.

Instead of a level shift, test whether the monthly growth rate changed at the transfer. Outcome: Δlog(value), n = 44 month-over-month growth observations, Newey-West (lag 3) standard errors:

Placebo test passes this time. Fake event dates across the pre-period produce shifts between −20.4 and +8.9 pp with p between 0.12 and 0.63 — none stands out, unlike Specification 1.

Result: no statistically detectable change in the growth trajectory at the 10 Jan 2025 takeover. This matches the design prior: the transfer was built for continuity (POJK 27/2024 adopted Bappebti's rules with refinements). The visible post-2025 cooling exists in the raw series but cannot be attributed to the transfer — it coincides with the tax change, unrest, and global cycle, and is not distinguishable from noise at monthly frequency.

Second estimable event: the 1 Aug 2025 tax change REAL ESTIMATE

PMK 50/2025 raised the final income tax on domestic-exchange crypto sales from 0.1% to 0.21% and removed the 11% VAT on purchases (net effect on trading cost ambiguous; foreign-exchange sales taxed at 1%). Interrupted time series on log monthly value, OJK-era months only (n = 21), Newey-West (lag 1) standard errors:

Raw headline: implied level shift +64.7% (95% CI +9.7% to +147.1%), p = 0.016.

But the placebo test fails — treat this as weak evidence

Re-run the same regression at fake event dates inside the pre-period. If the design works, fake dates should produce small, insignificant shifts. They do not:

Fake dates produce shifts as large as the real one with small p-values. With only 7 pre-period months and a rising trend, the design cannot separate a level shift from the trend. Robustness: excluding imputed months gives +41.5% (p = 0.029); cutting the window at Mar 2026 gives +172.6% (p < 0.001) — the estimate moves wildly with sample choices, confirming fragility.

Write-up language: "The estimated effect of the August 2025 tax change is positive and statistically significant in a naive specification (+65%, p = 0.02), but placebo event dates produce similar shifts, so the design does not identify an effect. The data are consistent with a trend break, a tax effect, the August 2025 civil unrest, or noise — the monthly series is too short to distinguish them."

Concurrent events to code as dummies or discuss

Aug 2025 tax change (above) · Aug–Sep 2025 domestic unrest, which OJK commented on · Mar 2026 ICEx launch (new competing exchange/clearing/custody infrastructure) · Jul 2025 formal start of DFA trading operations under the new OJK framework · global crypto price cycle (no control series in this estimate — a global volume or Bitcoin-price control would change all numbers).

The method in one line

We test whether Indonesian crypto transaction volumes changed at the Bappebti→OJK handover (10 Jan 2025) using an interrupted time series on 45 months of data (Jan 2023 – Sep 2026), following the validity framework of Baicker & Svoronos (2019). No control country is in the regression — so every result is labelled single-ITS, not causal.

How the design protects itself

• Fake interruption dates. The model is re-run at made-up event dates; if fakes reproduce the real estimate, the result is discarded as unidentified. This killed the level-shift form and the tax-change ITS — and the handover growth-break passed.
• Two functional forms, both shown. The 2024 recovery breaks the level form; the growth form is the defensible one. The failure is published, not hidden.
• Confounders on the timeline. Tax change, unrest, ICEx launch — coded explicitly and listed beside every estimate.
• Pre-specified plan. Question, variables and placebo strategy were written before looking at results (analysis_plan.md in the repository).

What the data can and cannot say

• Can say: "no change was detectable at the handover, and the design was tested against fake interruption dates."
• Cannot say: "the handover had no effect." That needs a control series; this tracker says so wherever a number appears.

Data limits

• 45 monthly observations (24 pre / 21 post) — standard errors are wide.
• 14 of 45 months are derived from cumulative reports (flagged on charts); 2022 is annual-only.
• Jan–Jun 2025 cumulative and summed monthlies differ by ~2.7T; disclosed, not smoothed.
• All figures are press-transmitted official statements — swap in OJK's first-hand statistics before publication-grade use.

Three takeaways the charts support

Protected by policy, or resilient to it?

Crypto isn't protected by policy — no lender of last resort, no deposit insurance, no bailout for a failing exchange. But that's also why its prices show lower sensitivity to policy shocks: valuations aren't anchored to interest rates and credit, so a single jurisdiction's policy moves transmit weakly.

Traditional finance is the mirror image: protected by policy, and therefore sensitive to it.

Two sides of one mechanism — no safety net, no dependency. Preliminary: consistent with takeaway 1, but a proper test needs the control-series upgrade in Method & limits. "Resilient" applies to market prices only, not institutional stability.

How far these conclusions reach

Everything above rests on 45 monthly observations, single-ITS with passing placebos — "no change was detectable at the interruption, and the design was tested against fake interruption dates", not "the intervention had no effect." 14 of 45 months are derived residuals and all figures are press-transmitted official statements; intervals are wide. The strongest supported reading today is about market structure: supervision moved without repricing the market, and the rulebook is now complete (PADK 3/2026) with a second exchange competing — volume follows the global cycle, structure follows policy.