Stress signals · Q2 2026 · Jul 2026
Stress signals from funding and trade together
Seven signals, each a money quantity from money-flow set against a trade quantity from trade-flow, scored every quarter for the 36 board economies with nothing read from later periods. Each one fires at a threshold; the count says how many fired. They are shown beside money-flow’s vulnerability score V, and every one carries its test against past stress.
No composite is published. Under rule G19 a signal joins a composite only if it beats the better of its two single-side inputs, with the 90% interval of the gain above zero. None of the seven did on this build’s funding episodes, so the page shows the seven signals beside V and the count, as the plan said it would.
The test against funding episodes
Positives: the highest value in the 4 quarters before each of 29 economy-episodes. Calm: quarters more than two years from any episode and outside the global windows of 2008–09 and 2020. Thresholds that come from validation were chosen on quarters to Q4 2020; the later quarters are the hold-out.
| Signal | AUROC (90% interval) | Money input alone | Trade input alone | Gain over the better side (90%) | In composite | Threshold | Hit / false rate | Hold-out false rate |
|---|---|---|---|---|---|---|---|---|
| J1 Import cover 27 / 1298 | 0.512 (0.406–0.620) | 0.655 | 0.394 | −0.143 (−0.230–−0.066) | context | ≤ 3 Threshold by convention | 15% / 12% | 13% |
| J2 Financing need ÷ exports 27 / 2171 | 0.625 (0.550–0.696) | 0.483 | 0.675 | −0.049 (−0.101–0.000) | context | ≥ 50 Threshold chosen by validation | 59% / 35% | 22% |
| J3 Double hit 17 / 1844 | 0.734 (0.653–0.804) | 0.754 | 0.640 | −0.020 (−0.047–+0.011) | context | — Fires by definition | 6% / 13% | 6% |
| J4 Partner stress 18 / 1791 | 0.661 (0.545–0.769) | 0.677 | 0.534 | −0.016 (−0.100–+0.064) | context | ≥ 50.85 Threshold chosen by validation | 28% / 3% | 0% |
| J5 Dual dependence 27 / 2171 | 0.379 (0.308–0.450) | 0.428 | 0.450 | −0.071 (−0.187–+0.054) | context | ≥ 43.5 Threshold chosen by validation | 4% / 11% | 7% |
| J6 Dollar channel 17 / 1854 | 0.756 (0.674–0.824) | 0.760 | 0.641 | −0.003 (−0.029–+0.023) | context | ≥ 4 Threshold chosen by validation | 59% / 38% | 34% |
| J7 Commodities and buffers 27 / 2312 | 0.777 (0.727–0.823) | 0.722 | 0.680 | +0.055 (−0.035–+0.148) | context | ≥ 9.1 Threshold chosen by validation | 59% / 19% | 21% |
| The count (default thresholds) | 0.751 (0.675–0.827) | V alone (money-flow): 0.725 (0.642–0.807) | +0.026 (−0.088–+0.130) | K = 3 | 41% / 9% | 5% | ||
The count scores an AUROC of 0.751 (0.675–0.827) against 0.725 for V on the same positives; the difference +0.026 has an interval of −0.088 to +0.130, which includes zero. Leaving one episode out moves it between 0.727 (without taper-2013) and 0.803 (without gfc-2008).
Composite designs written before the run
- D1 Half level rank, half own-history robust z, mean of admitted signals (the plan's design) · not tested: no admitted signal
- D2 Level rank only, mean of admitted signals · not tested: no admitted signal
- D3 Own-history robust z only, mean of admitted signals · not tested: no admitted signal
A design passes if its interval’s lower bound is above V alone on the same positives. None is tested when no signal is admitted.
Expectations written before the build
- E1Korea's count is 3 or more in 2008 Q3 · fail count 0
- E2Türkiye's count is 3 or more in 2018 Q2 · fail count 1 (J2)
- E3Argentina's count is 3 or more in 2018 Q1 · fail count 2 (J1 J2)
- E4Russia's count is 3 or more in 2014 Q4 · fail count 2 (J6 J7)
- E5Czechia's count is 1 or less through 2017–2019 · pass max 1
- E6Saudi Arabia's J7 fires in 2015 and in 2020 · pass 2015: 2015-12; 2020: 2020-03 2020-06 2020-09 2020-12
- E7J4 for Hungary and Poland rises in 2008 Q3 while their own V does not yet · fail HUN: J4 50.8→54.6, V 82.5→88.5; POL: J4 44.8→47.2, V 62→65
- Locational banking statistics (LBS) · BIS, open, with attribution · Q1 2005 to Q2 2026, quarterly · read 30 Oct 2026 · via money-flow method
- Consolidated banking statistics (CBS) · BIS, open, with attribution · Q1 2005 to Q2 2026, quarterly · read 30 Oct 2026 · via money-flow method
- Debt securities statistics (DSS) · BIS, open, with attribution · Q1 2005 to Q2 2026, quarterly · read 27 Oct 2026 · via money-flow method
- International Reserves and Foreign Currency Liquidity (IRFCL) · IMF, open, with attribution · Q1 2005 to Q2 2026, quarterly · read 29 Oct 2026 · via money-flow method
- International Investment Position (IIP) · IMF, open, with attribution · Q1 2005 to Q2 2026, quarterly · read 27 Oct 2026 · via money-flow method
- Balance of Payments (BOP) · IMF, open, with attribution · Q1 2005 to Q2 2026, quarterly · read 27 Oct 2026 · via money-flow method
- Credit to the non-financial sector (GDP for every ratio) · BIS, open, with attribution · Q1 2005 to Q2 2026, quarterly · via money-flow method
- FRED, Federal Reserve H.4.1 and H.10 series · Federal Reserve Board, via FRED (St. Louis Fed), public domain, citation requested ·
DTWEXBGS· read 31 Oct 2026 · via money-flow method - UN Comtrade, goods trade by reporter and partner · UN Statistics Division, derived records only (shares × world total) · 12-month window, Aug 2025–Jul 2026 · read 2 Nov 2026 · via trade-flow method
Board quarters to Q2 2026; trade windows to Jul 2026. Calculations: datasteg global-network (scripts/stress.py).