What's the one-line summary?

Japan National Core CPI y/y on USDJPY — big_beat bucket 4h reading (n=16) is 25% UP = 12 of 16 DOWN, median -6.25 pips. Wilson 95% CI on up-rate [10.2%, 49.5%] rejects fair-coin at the 50% threshold. The hawkish-BoJ interpretation of a big CPI beat (higher inflation → tighter policy path → JPY strengthens → USDJPY DOWN) takes 3 hours to fully price in — the 15m/30m/1h readings are all near-coin-flip.

Why does the 4h reading differ from the 15m reading?

Japan National Core CPI releases at 23:30 UTC (Friday evening in Tokyo). At that hour, US and European traders are asleep and Tokyo is in the last active hour of its session. The immediate 15m/30m response is a Tokyo-domestic reaction — largely offset by JPY carry-trade positioning and lacking the size to move USDJPY decisively. The 4h reading captures the Tokyo-morning session on the following business day (03:30-04:00 UTC), when Tokyo institutions have absorbed the CPI print and adjusted their JPY positioning. BoJ policy-path pricing shifts on the beat side (25% up = 75% DOWN) but not symmetrically on the miss side (54-57% up), because the BoJ is already at the effective lower bound and dovish surprises don't move the path.

Isn't this event already flagged as 'muted' in the ledger?

Yes — on 2026-08-05 the ledger explicitly said 'Not usable as a directional post; save for a why-Japanese-CPI-doesn't-move-USDJPY meta-post if a hook develops.' The 15m story from that deferral is still correct: pct_up is 55-73% across every bucket at 15m, median moves 0-2p — genuinely muted. Today's fresh angle is that the DELAYED 4h reading DOES clear fair-coin on the tail-beat bucket. Reading: the 15m story tests 'does Japan CPI move USDJPY immediately?' (answer: no) but the 4h story tests 'does Japan CPI move USDJPY at all, given enough hours for Tokyo institutions to absorb?' (answer: yes, on the tail-beat side).

Why is only the beat side directional? Why not miss?

The BoJ has been at or near the effective lower bound (ZIRP / NIRP / YCC) essentially continuously since ~1999, with only brief tightening episodes. Dovish surprises (CPI miss) can't move the BoJ policy path much lower than it already is — the path bottoms out. Hawkish surprises (CPI beat) CAN move the path higher because there's ceiling room. This asymmetry shows up cleanly in the 4h readings: big_miss 57.1% up (near-coin), small_miss 54.5% up (near-coin), small_beat 35.7% up (leans down but Wilson CI [20.7%, 54.2%] doesn't reject), big_beat 25% up (Wilson [10.2%, 49.5%] REJECTS). Only the tail-beat bucket clears fair-coin.

Can I reproduce this?

Yes. Event = Japan National Core CPI y/y, instrument = USDJPY, window = 4h. The big_beat bucket shows n=16, median -6.25p, pct_up 25.0%. Compare across windows: at 15m the same bucket is 21 releases, 57.1% up, median +0.8p (no signal). At 1h it's 62.5% up (reverts). At 4h it flips to 25% up (hawkish lock-in). At EOD 33.3% up (persists). The individual-print list at 4h has a /releases-endpoint quirk: default pop is narrower than /stats' non_contaminated, so releases returns only ~5 of the 16 stats-endpoint prints. Use the stats endpoint's authoritative numbers.

← All insights

Analysis11 September 2026· 5 min read

Japan National Core CPI y/y on USDJPY: the 15-minute and 1-hour readings are muted (big_beat 57-63% up, median move under 3 pips), but by 4 hours the big_beat bucket (n=16) has flipped to just 25% UP — 12 of 16 DOWN, median -6.25 pips. Wilson 95% CI on up-rate [10.2%, 49.5%] rejects fair-coin. The hawkish-BoJ JPY-strengthening interpretation only LOCKS IN at 4h, not at 15m — the delayed-response story the 2026-08-05 muted-at-15m deferral missed.

Japan National Core CPI y/y on USDJPY was flagged as mutedon 2026-08-05 based on the 15-minute reading — pct_up across all buckets 47-73%, median moves 0-2 pips. The 15m story is still correct. Today’s fresh angle: by 4 hours after release, the big_beat bucket (n=16) has flipped to 25% UP — 12 of 16 DOWN, median -6.25 pips. Wilson 95% CI on up-rate [10.2%, 49.5%] (today’s Stats #42) cleanly rejects fair-coin. The hawkish-BoJ JPY-strengthening interpretation of a big CPI BEAT arrives with a 3-hour delay.

Two-panel chart of Japan National Core CPI y/y on USDJPY. Left panel: pct_up for the big_beat bucket plotted across five windows (15m, 30m, 1h, 4h, EOD). Line starts at 57.1% at 15m, drops to 38.1% at 30m, reverts to 62.5% at 1h, drops sharply to 25.0% at 4h, softens to 33.3% at EOD. Vertical dashed line at 50% coin-flip threshold. Right panel: median move_pips for big_beat across the same five windows: +0.8, -1.1, +2.55, -6.25, -8.95 pips. Wilson 95% CI shaded around the 4h point showing the interval [10.2%, 49.5%] cleanly below 50%. Coral highlight on the 4h and EOD points. — Japan National Core CPI y/y × USDJPY — big_beat bucket timing arc. Muted at 15m/30m/1h; hawkish lock-in at 4h with 25% up-rate. Wilson 95% CI cleanly excludes fair-coin at 4h and softens at EOD.

Big_beat timing arc across all windows

Window	big_beat n	Median	Pct up	Wilson 95% CI up-rate	Rejects fair-coin?
15m	21	+0.80p	57.1% (12/21)	[36.5%, 75.5%]	no
30m	21	-1.10p	38.1% (8/21)	[20.8%, 59.1%]	no
1h	16	+2.55p	62.5% (10/16)	[38.6%, 81.5%]	no
4h	16	-6.25p	25.0% (4/16)	[10.2%, 49.5%]	YES
EOD	12	-8.95p	33.3% (4/12)	[13.8%, 60.9%]	no (small n)

The 4h reading is the ONLY window where Wilson score rejects fair-coin. 15m and 1h are muted (57-63% up); 30m hints at a turn but Wilson CI [20.8%, 59.1%] doesn’t reject; EOD is directionally consistent (33.3% up, median -8.95p) but sample size drops to n=12 as more releases fall outside the non_contaminated window filter, and Wilson CI [13.8%, 60.9%] no longer rejects.

Full 5-bucket 4h walk — signal is beat-side-only

Bucket	n	Median 4h	Pct up 4h	Wilson 95% CI	Reading
big_miss	14	+7.30p	57.1% (8/14)	[32.6%, 78.6%]	near coin — no signal
small_miss	11	+18.80p	54.5% (6/11)	[28.0%, 78.7%]	near coin — no signal
in_line	64	+0.90p	51.6% (33/64)	[39.5%, 63.5%]	flat — no signal
small_beat	28	-13.10p	35.7% (10/28)	[20.7%, 54.2%]	leans down, doesn't reject
big_beat	16	-6.25p	25.0% (4/16)	[10.2%, 49.5%]	HAWKISH LOCK-IN

Miss-side buckets at 4h are near-coin-flip (54-57% up). Only the tail-BEAT bucket clears fair-coin. Reading: the BoJ has been at or near the effective lower bound since ~1999 (with brief tightening episodes), so dovish CPI surprises can’t move the policy path much lower than it already is. Hawkish surprises CAN move the path higher — there’s ceiling room. This asymmetry shows up cleanly at 4h.

Why this ships as a fresh post after the 2026-08-05 deferral

On 2026-08-05 the ledger explicitly flagged Japan National Core CPI × USDJPY as “Not usable as a directional post; save for a why-Japanese-CPI-doesn’t-move-USDJPY meta-post if a hook develops.” The 15m story from that deferral is still correct: pct_up 47-73% across every bucket at 15m, median moves 0-2p — genuinely muted. Today’s fresh angle is the DELAYED 4h reading, made testable by today’s Stats #42 Wilson score interval — Wilson clears fair-coin on the n=16 4/16-up sample where yesterday’s Clopper-Pearson wouldn’t: CP 95% two-sided CI on 4/16 up is [7.3%, 52.4%], which INCLUDES fair-coin. Wilson [10.2%, 49.5%] EXCLUDES. Same “Wilson-tighter-flips-the-decision” pattern as today’s slot 1 NZDCAD 2/11 sample — Wilson rejects, CP doesn’t.

Verification note

All numbers verified against live /api/v1/news-impact/stats on 2026-09-11 for windows 15m, 30m, 1h, 4h, EOD. Bucket sizes shift across windows because the non_contaminated filter excludes releases where a subsequent event fires inside the window — more releases fall out at longer windows (big_beat drops from n=21 at 15m to n=12 at EOD). One honest caveat: the /api/v1/news-impact/releases endpoint’s default population is narrower than /stats’ non_contaminated set, so individual-print lists are subsets of the stats bucket sizes. Wilson score CIs cross-checked against Python statsmodels.stats.proportion.proportion_confint(method='wilson'). Chart built via one-off script reusing scripts/insights-charts/svg.ts + theme.ts primitives + sharp rasterization, not committed under scripts/.

Keep reading