How many FX rules were tested, and how many held up?

22 rules were tested on the FX face as of 6 Nov 2026: 7 validated, 2 provisional, 1 descriptive and 12 found no edge.

What does validated mean here?

A rule is validated only when its 95% range excludes zero, both halves of the history agree, it has at least 20 separate windows and 10 episodes, and it survives a false-discovery check (q at or below 0.10) across all 108 tests. In-sample: these rules were written down after a first look at the same history, so every label here is in-sample. Only results logged from 2026-09-25 onward (the forward ledger) are truly out-of-sample.

Which currency signals were tested, and which held up

Every rule the FX face reads, written down before it was tested: the dollar cushion, currency translation, range cones, positioning and strong-dollar rules, with the sample behind each.

Data as of 6 Nov 2026.

Open the FX evidence in the cockpit

As of 6 Nov 2026, 22 rules were tested on the FX face: 7 validated, 2 provisional, 1 descriptive and 12 found no edge (plus 2 arithmetic facts).
Across all four faces the false-discovery check covers 108 tests: 29 validated, 19 provisional, 12 descriptive and 48 with no edge.
Validated (7) on this face: Range-cone calibration (INR); Range-cone calibration (JPY); Range-cone calibration (CNH); USD cushion vs INR; and 3 more.
Provisional (2): Range-cone calibration (KRW); Range-cone calibration (SGD).
No edge found (12): COT crowding (AUD, crowded LONG); COT crowding (AUD, crowded SHORT); COT crowding (EUR, crowded LONG); COT crowding (EUR, crowded SHORT); COT crowding (JPY, crowded LONG); and 7 more.
In-sample: these rules were written down after a first look at the same history, so every label here is in-sample. Only results logged from 2026-09-25 onward (the forward ledger) are truly out-of-sample.

Rules that held up or came close

Effect = the gap against a volatility-matched or ordinary day (percentage points of odds, or % of return; a plain coefficient for betas and rank correlations). Windows = separate, non-overlapping periods the rule could be judged on.
Rule tested	Result	Effect	95% range	Windows	Episodes	Halves
Range-cone calibration (CNH): the +/-2 sigma one-month band, sigma = 20-session realised vol x sqrt(21/252), contains the next 21-session USD/CNH move a share of the time that differs from the normal-curve 95.45%. Target: share of 21-session USD/CNH moves inside +/-2 sigma.	Validated	-6.9 pp	-10.2 pp to -3.7 pp	183	183	agree
Range-cone calibration (INR): the +/-2 sigma one-month band, sigma = 20-session realised vol x sqrt(21/252), contains the next 21-session USD/INR move a share of the time that differs from the normal-curve 95.45%. Target: share of 21-session USD/INR moves inside +/-2 sigma.	Validated	-6.4 pp	-8.8 pp to -3.9 pp	283	283	agree
Range-cone calibration (JPY): the +/-2 sigma one-month band, sigma = 20-session realised vol x sqrt(21/252), contains the next 21-session USD/JPY move a share of the time that differs from the normal-curve 95.45%. Target: share of 21-session USD/JPY moves inside +/-2 sigma.	Validated	-4.5 pp	-6.7 pp to -2.5 pp	283	283	agree
USD cushion vs INR: in SPY's worst 10% of 21-session windows (decile set within the span), USD/INR's change over the SAME 21 sessions is above its all-windows average (a same-window hedge fact, not a forecast). Target: USD/INR % change over the same 21 sessions.	Validated	+1.71%	+1.19% to +2.22%	65	46	agree
USD cushion vs INR, recent regime: the same rule measured on 2016+ only (decile recomputed within 2016+). Target: USD/INR % change over the same 21 sessions.	Validated	+0.91%	+0.35% to +1.47%	28	20	agree
USD cushion vs SGD: in SPY's worst 10% of 21-session windows (decile set within the span), USD/SGD's change over the SAME 21 sessions is above its all-windows average (a same-window hedge fact, not a forecast). Target: USD/SGD % change over the same 21 sessions.	Validated	+1.08%	+0.70% to +1.46%	65	46	agree
USD cushion vs SGD, recent regime: the same rule measured on 2016+ only (decile recomputed within 2016+). Target: USD/SGD % change over the same 21 sessions.	Validated	+0.86%	+0.42% to +1.26%	28	20	agree
Range-cone calibration (KRW): the +/-2 sigma one-month band, sigma = 20-session realised vol x sqrt(21/252), contains the next 21-session USD/KRW move a share of the time that differs from the normal-curve 95.45%. Target: share of 21-session USD/KRW moves inside +/-2 sigma.	Provisional	-2.4 pp	-4.3 pp to -0.7 pp	283	283	agree
Range-cone calibration (SGD): the +/-2 sigma one-month band, sigma = 20-session realised vol x sqrt(21/252), contains the next 21-session USD/SGD move a share of the time that differs from the normal-curve 95.45%. Target: share of 21-session USD/SGD moves inside +/-2 sigma.	Provisional	-2.7 pp	-4.8 pp to -0.9 pp	283	283	agree
Strong Asian currency → own market (context, pooled): on days when any of KRW/TWD/SGD/INR/THB/IDR is up 5% or more vs USD over 63 sessions, the average USD return of those markets over their next 63 sessions, vs the equal-weight… Target: Triggered markets' USD return, next 63 sessions (else the basket).	Descriptive	+1.44%	-1.39% to +4.25%	39	23	agree

Arithmetic facts (no statistics)

Hedge cost (arithmetic, covered-interest-parity approximation): hedging US$ forward for 3 months earns or costs the home 3-month bill yield minus the US 3-month bill yield, per year (TVC:SG03MY / TVC:IN03MY minus TVC:US03MY). Negative = hedging costs that much a year; bank forward spreads come on top.
Home-currency translation (arithmetic, not a forecast): return in S$ = (1 + local return) x (USD/local at start / USD/local at end) x (USD/SGD at end / USD/SGD at start) - 1; the same with USD/INR for ₹. Each market is read at its own last final session; FX legs are the TradingView FX_IDC closes on or before those…

Tested, no edge

These rules were stated in advance and found nothing on this history; they stay on the record.

Rule tested	Effect	95% range	Windows
COT crowding (AUD, crowded LONG): leveraged funds' net futures position as a share of open interest at or above the 90th percentile of its trailing 3 years → the AUD currency index (TVC:AXY) over the…	+0.04%	-1.49% to +1.61%	18
COT crowding (AUD, crowded SHORT): leveraged funds' net futures position as a share of open interest at or below the 10th percentile of its trailing 3 years → the AUD currency index (TVC:AXY) over the…	-0.55%	-3.17% to +1.54%	20
COT crowding (EUR, crowded LONG): leveraged funds' net futures position as a share of open interest at or above the 90th percentile of its trailing 3 years → the EUR currency index (TVC:EXY) over the…	+0.45%	-0.94% to +1.74%	19
COT crowding (EUR, crowded SHORT): leveraged funds' net futures position as a share of open interest at or below the 10th percentile of its trailing 3 years → the EUR currency index (TVC:EXY) over the…	-0.89%	-3.71% to +2.16%	15
COT crowding (JPY, crowded LONG): leveraged funds' net futures position as a share of open interest at or above the 90th percentile of its trailing 3 years → the JPY currency index (TVC:JXY) over the…	+0.21%	-2.01% to +2.56%	18
COT crowding (JPY, crowded SHORT): leveraged funds' net futures position as a share of open interest at or below the 10th percentile of its trailing 3 years → the JPY currency index (TVC:JXY) over the…	-0.11%	-2.68% to +2.29%	19
Won-weak rule: USD/KRW up 5% or more over 63 sessions → KOSPI in USD over the next 63 Korean sessions differs from its all-days average (two-sided).	+1.45%	-3.05% to +6.55%	26
Currency-strength tilt: each day, rank 7 Asian markets (KOSPI, TAIEX, STI, Nifty, SET, JCI, Nikkei) by their currency's 63-session change vs USD; the 3 strongest-currency markets minus the 3 weakest…	-0.19%	-0.49% to +0.12%	283
Strong-dollar rule: when the dollar index (DXY) is up 3% or more over 63 sessions, an equal-weight Asia ex-Japan basket (KOSPI, TAIEX, STI, Nifty, SET, JCI, each in USD) returns LESS than its all-days…	-1.69%	-4.47% to +1.04%	38
Trend-chip fade (replaces the unreproducible 't -6.9' claim): when DXY's trend chip reads 'up' (DXY above its 50-session average, which is above its 200-session average), DXY's change over the next 21…	+0.15%	-0.16% to +0.46%	115
Taiwan-dollar-weak rule: USD/TWD up 3% or more over 63 sessions → TAIEX in USD over the next 63 Taiwan sessions differs from its all-days average (two-sided).	+0.95%	-3.99% to +6.40%	26
Yen-spike rule (carry unwind): USD/JPY down 3% or more over 10 sessions → Nikkei 225 (in JPY) returns LESS than its all-days average over the next 21 Tokyo sessions.	+0.86%	-0.68% to +2.30%	59

How old is each input?

Every input the face reads, and how old it is.
Input	Latest data	Age	On time
FX_IDC:USDSGD	6 Nov 2026	1 business day	yes
FX_IDC:USDINR	6 Nov 2026	1 business day	yes
TVC:DXY	6 Nov 2026	1 business day	yes
AMEX:SPY	6 Nov 2026	1 business day	yes
TVC:SPX	6 Nov 2026	1 business day	yes
NSE:NIFTY	6 Nov 2026	1 business day	yes
TVC:STI	6 Nov 2026	1 business day	yes
HSI:HSI	6 Nov 2026	1 business day	yes
KRX:KOSPI	6 Nov 2026	1 business day	yes
TWSE:IX0001	6 Nov 2026	1 business day	yes
TVC:NI225	6 Nov 2026	1 business day	yes
SSE:000001	6 Nov 2026	1 business day	yes
TVC:US03MY	6 Nov 2026	1 business day	yes
TVC:SG03MY	6 Nov 2026	1 business day	yes
TVC:IN03MY	6 Nov 2026	1 business day	yes
CNY daily closes	6 Nov 2026	1 business day	yes
HKD daily closes	6 Nov 2026	1 business day	yes
Rule tests (re-measured weekly)	3 Nov 2026	4 business days	yes

How to read the labels

Validated: the 95% range excludes zero, both halves of the history agree in sign, there are at least 20 separate windows and 10 episodes, and it survives the false-discovery check (q at or below 0.10) across every macro test.
Provisional: the 95% range (or the 90% range) excludes zero but one of the other validation checks fails: a tilt, not a rule.
Descriptive: context only: the range includes zero or there were too few cases to test.
No edge: a test stated in advance that found no edge on this history.
Fact: arithmetic (currency translation, hedge cost), not a statistical claim.

Method: each effect is the gap against a volatility-matched or ordinary day, with a 95% range from a circular moving-block bootstrap (2,000 resamples, fixed seed). Prices and indicators are TradingView closes; how the numbers are made; the cockpit's forecasting models are validated separately on the model research page.

Questions

How many FX rules were tested, and how many held up?: 22 rules were tested on the FX face as of 6 Nov 2026: 7 validated, 2 provisional, 1 descriptive and 12 found no edge.
What does validated mean here?: A rule is validated only when its 95% range excludes zero, both halves of the history agree, it has at least 20 separate windows and 10 episodes, and it survives a false-discovery check (q at or below 0.10) across all 108 tests. In-sample: these rules were written down after a first look at the same history, so every label here is in-sample. Only results logged from 2026-09-25 onward (the forward ledger) are truly out-of-sample.

More from the Macro tab: all five faces · Risk gauges · Commodities · Crypto · Bonds.