Three machine layers that turn scattered data — into edge. First read the relationships cleanly, then open the web to feel how forces transmit. Forward View forecasts every driver behind your portfolio. Earth Signal reads physical activity from space and maps it to the SET names exposed to it.
MARKET CONSENSUS
4 of 4 signals voted · 2026-09-15
The desk collects forward-looking signals from every model, feed, and book. Each one votes BULL, BEAR, or abstains. Signals are grouped by horizon — what a 20-day forecast says should not cancel a 90-day baseline. The verdict is the weighted median, the confidence is how strongly the voters agreed, and the cautionary note names any high-weight dissenter.
Track-record grading for the 20d and 90d verdicts began 2026-09-15 — a grade appears here once the first horizon realizes (never before). Confidence above is voter agreement, not probability.
LIVE PIPELINE · WHAT'S FLOWING INTO THE MARKET NOWas of 2026-07-02 · 2026-09-14
Each row is a real signal read from live data, translated into what it means for your money and the SET names it touches. The graph below is the full web of connections behind it.
Fed Funds Rate3.63% (no live cache — last reading 2026-07-02, 74d stale)WATCH
Fed held at 3.63% since the Jul 2 brief — yield curve still +0.33 pp (10y–2y, live Sep 11), a calm curve regime. Banks still earn a wide NIM either way; the watch is on whether the cut cycle starts.
LANDS ON →SETKBANKSCBBBL
Oil (WTI)$67.77 (last live 2026-07-02, 74d stale)TAILWIND
Oil last read at $67.77 on Jul 2 — Thailand is an energy importer, so cheap oil is a tailwind for transport, airlines, and consumer real wages. PTT and PTTEP are the natural short side; AOT/MINT the long side.
LANDS ON →AOTMINTPTTPTTEP
SET ValuationP/E 22.2 vs 17.89 avg · +24% premiumWATCH
SET trades at 22.2× — a +24% premium to its 17.89 long-run average. Today's market-state is 'Range-bound calm', consistent with FRAG stage. Less margin of safety — favour quality and patience over chasing.
VIX 17.8 — a calm regime. The Forward 90d view says +4.6%, Kronos 20d says -13% (zero-shot, less reliable). The disagreement is the headline; the SET itself has gained 1.8% in a month and the technical-edge ledger says EMA9 > EMA21 +0.35 pp 30d edge. Calm is honest.
The hyperscaler spend cycle is still the demand behind chips — NASDAQ +2.3% on the week and SOXX 90-day forward at +9.3%. DELTA tracks Nasdaq; SK Hynix (KR) is +20% on the month on HBM tightness.
Gold's 90-day forward is +3.2% with VIX calm — a real-rate hedge bid, not a panic bid. The technical-edge ledger says RSI>70 (overbought) retains momentum — gold's bid persists when the crowd is already long.
LANDS ON →Gold
US 10Y Yield10y–2y +0.33 pp · un-invertedWATCH
Curve still +0.33 pp (live Sep 11) — calm, un-inverted. The signal-edge ledger says an inverted curve has been the strongest 90d forward-return edge on the S&P (786 obs, +3.0 pp 90d). We're NOT inverted today — the edge is dormant.
LANDS ON →SET
USD/THB33.22 · 90d exp +0.8% · flatNEUTRAL
USDTHB around 33.22 with the 90-day forward barely moving (+0.8%). Baht is steady, no devaluation pressure to chase. Foreign flow is the marginal SET buyer and a steady baht keeps that channel open.
LANDS ON →SETKBANKPTT
Bitcoin90d exp +2.5% · ann vol 60% · risk-apetite proxyNEUTRAL
Bitcoin is in the structural edge: p 0.054 (just outside the 5% bar) on a 1-day lag — a risk-appetite proxy, not a SET driver. No SALIENCE for SET itself; included so the halo shows what the risk channel is doing.
LANDS ON →Risk
Dollar Index (DXY)Granger p 0.177 · lag 5d · weak evidenceNEUTRAL
DXY registers a weak Granger signal on SET at lag 5 — the baht is a more direct dollar proxy on the engine. Below the 10% p<0.10 bar; reported for transparency.
Readings drawn from forward-view + market-mood + granger + prices + market-states (some items still sourced from the older daily-brief cache). Staleness flags mark any reading older than 7 days. The causal map is curated trader knowledge — not investment advice.
TREND TAPE· ranked by expected return
2026-09-15◐CACHED · —
ASSETSPOTEXP RETVOLMODEL
SOXX
Semiconductors
$497.4
+9.3%
46%
vol-cone · 90d
QQQ
US tech · Nasdaq-100
$709.18
+3.4%
20%
vol-cone · 90d
USD/THB
Baht vs dollar
33.22
+0.7%
7%
vol-cone · 90d
GLD
Gold
$392.84
+2.7%
30%
vol-cone · 90d
VNM
Vietnam equity
$17.34
-1.0%
26%
vol-cone · 90d
ADVANC
Advanced Info
฿353
-0.9%
21%
Kronos-small · 20d
MCHI
China equity
$53.32
-3.0%
20%
vol-cone · 90d
SCB
SCB
฿151
-12.1%
18%
Kronos-small · 20d
SET
Thai SET
฿1,578.66
-12.7%
16%
Kronos-small · 20d
BBL
Bangkok Bank
฿192.5
-16.6%
19%
Kronos-small · 20d
PTT
PTT
฿42.5
-17.4%
19%
Kronos-small · 20d
KBANK
Kasikorn Bank
฿252
-22.8%
19%
Kronos-small · 20d
Illustrative forecasts. Walk-forward calibration on ^SET.BK: APPROXIMATELY CALIBRATED — 80% band coverage 73% (target 80%), direction 43%, skill vs naive -11.6%. Kronos-mini is zero-shot on SET — not validated until walk-forward passes. Expected ranges, not promises.
· Walk-forward backtest, not live tracking — no guarantee of future calibration.
Forces connected to Thai equities — ranked by evidence, not volume.
NET PRESSURE-9494% of linked force magnitude
S&P 500→SET
CAUSALLIFTSp 0.028 · lag 1dFLOWS
GOLD↛SET
CAUSALPRESSURESp 0.039 · lag 1dCOMMODITIES
BITCOIN→SET
CAUSALLIFTSp 0.054 · lag 1dRISK
USD/JPY→SET
CAUSALLIFTSp 0.068 · lag 1dFX
OIL WTI→SET
CAUSALLIFTSp 0.011 · lag 5dCOMMODITIES
VIX↛SET
CAUSALPRESSURESp 0.106 · lag 1dRISK
NIKKEI 225→SET
CAUSALLIFTSp 0.139 · lag 1dFLOWS
DXY↛SET
CAUSALPRESSURESp 0.177 · lag 5dFX
Correlation is not causation. Granger edges measure predictive lead; structural edges encode known transmission (rates → FX → flows). Tap the web below to trace a node.
Granger precedence and correlation are not causation. Forecasts are model pressure, not prophecy. · Granger 2026-09-14
DIRECT layer · GraphicalLassoCV · 2026-09-10 · Sparse direct links only — transitive co-movement (e.g. via SPX) is removed.
TRACE ANY FORCE TO SET · THE EXPLANATION ROOM
When FED RATE rises, this chain pushes SET down
01 · FED RATE → US 10Ycurated model
When the Fed raises its policy rate, yields on US government bonds follow it upward.
02 · US 10Y → DXYcurated model
Higher US bond yields pull global money into dollars to earn them — the dollar strengthens.
03 · DXY → THB/USDcurated model
A stronger dollar means the baht buys fewer dollars — USD/THB rises.
04 · THB/USD → FOREIGN FLOWcurated model
A weakening baht eats foreign investors' returns in their own currency — they pull money out of Thai stocks.
05 · FOREIGN FLOW → SETcurated model
Foreign money is the largest marginal buyer of Thai stocks — when it leaves, the index falls.
A chain is only as strong as its softest link — softest here: clear mechanism.
HOW THE WEB'S NUMBERS ARE MADE · ZERO HALLUCINATION
NODE POSITION / MASS
Layout is physics (repulsion + ropes). Mass ∝ log market cap for SET names; fixed mass for curated drivers. Not a forecast.
GRANGER EDGE (CLEAR)
F-test lags 1–10 vs SET returns · significant at p<0.05 · from ingestion/granger.py → granger.json · run 2026-09-14. Precedence ≠ tradable edge.
STRUCTURAL EDGE
Curated transmission (Fed→yields→USD→THB→flows→SET). Mechanism text is hand-authored in signal-trace.ts. Stat shown as “curated model”.
CORRELATION / PARTIAL
Raw ρ from aligned daily returns in prices.json. DIRECT layer = Graphical Lasso partials · glasso as-of 2026-09-10. Spurious = strong ρ with no partial edge.
NET PRESSURE ON SET
Weighted sum of signed edge strengths into SET (current value -94). Model pressure, not a price target.
PARTICLES / COLORS
Green = positive sign (lifts); red = negative (pressures). Channel hues are taxonomy only — never bull/bear/amber wayfinding.
If a number cannot name its file or method, it does not appear. Trace any force above for the economic sentence behind each hop.
EVENTS → MARKETS · THE RECEIPTS
Mean-adjusted abnormal returns · estimation [-280,-31] · headline event window [0,+5] with [-1,+1], [-5,+5], [0,+20] robustness checks · first event per 10-day cluster.
LOW EVIDENCE · N=0
The current committed price cache is shorter than the required 280-session estimation window. No CAR is shown until the history is long enough.
GDELT EVENTS 100 · YAHOO PRICES AS OF 2026-09-14 · RESEARCH PROMPTS, NOT CAUSATION
WHEN PATTERNS INTERSECT · TODAY'S SPARKS
THE ODDITIES LEDGER
3 ODD · 0 CONFLUENCE
01
2026-09-14SETPRICE SHOCK
SET dropped 3.17% — a 3.22σ move for this stock
02
2026-09-14PTTMODEL DISAGREEMENT
Model expects PTT down 17.44% but the tape ran up 6.33% over 20 days
03
2026-09-14BBLMODEL DISAGREEMENT
Model expects Bangkok Bank down 16.56% but the tape ran up 1.05% over 20 days
An oddity is a question, not an answer.
Statistics over a 60-day window — extremes are 3-month extremes, not 52-week.
HOW WE GOT THIS NUMBER — Each oddity is a z-score or percentile of the latest bar against the stock's own recent history — volume z ≥ 2.5 vs 20-day baseline, price-shock |z| ≥ 2.5 vs its return history, correlation-break |Δρ| ≥ 0.8 between window halves. · AS OF 2026-09-14
FORMULA · WINDOW · SOURCE
Detectors: src/lib/anomaly.ts over the committed Yahoo OHLCV cache (~60 sessions). Confluence = ≥2 independent families on one subject. An oddity is a question, not an answer — correlation is not causation.
PREDICTION MARKETS● LIVE
Real-money market consensus on macro events that move SET50.
HOW WE GOT THIS NUMBER — Each cone is a 90-day forward distribution — median path plus 10/90 volatility band from the vol-cone (drift+σ baseline) model. Expected ranges, not promises.
FORMULA · WINDOW · SOURCE
Median path anchored to spot with model drift; band = empirical volatility cone scaled by √t around the median. Source: forward-view.json (ingestion/forward_view.py). As of 2026-09-15.
^SET.BK
Thai SET
+4.4%
vol 19%
URTH
Global brands / world
+2.8%
vol 13%
ACWI
Global all-world
+3.0%
vol 14%
SOXX
Semiconductors
+9.3%
vol 46%
EWJ
Japan equity
+4.1%
vol 21%
QQQ
US tech · Nasdaq-100
+3.4%
vol 20%
THB=X
USD/THB
+0.7%
vol 7%
GLD
Gold
+2.7%
vol 30%
BITQ
Blockchain equities
+2.2%
vol 60%
VNM
Vietnam equity
-1.0%
vol 26%
INDA
India equity
-1.8%
vol 15%
MCHI
China equity
-3.0%
vol 20%
Each cone is a 90-day forward distribution for the index that drives a Dream-Portfolio holding — median path + 10/90 band. Baseline is a transparent drift+volatility cone; run ingestion/forward_view.py in the M3 TimesFM venv to swap in the learned model. Expected ranges, not promises — not investment advice.
FORECAST CALIBRATION^SET.BK · 120 windows · 2026-09-13
APPROXIMATELY CALIBRATEDtarget 80% | actual
80% BAND COVERAGE — amber line = target
73%
COVERAGE
73%
80% target
DIRECTION
43%
50% baseline
SKILL VS NAIVE
-11.6%
vs yesterday
LAST 20 WINDOWS — ● inside 80% band, ○ outside
· Walk-forward backtest, not live tracking — no guarantee of future calibration.
· ARIMA fitted on returns, compounded for terminal price bands.
· Model is refitted each window — real-time performance may differ.
· Small positive skill vs naive (yesterday's close) is expected; large skill claims should be suspect.
Statistical Decoupling. The Kelly sizing and Probability of Profit below are computed from the zero-shot expected return and volatility forecast. Since this model is not fine-tuned on SET-specific regulations, treat allocations as risk ceilings.
HOW WE GOT THIS NUMBER — 20-day expected return and volatility come from the Kronos-mini zero-shot forecast (NeoQuasar, 4.1M params, pretrained on 45+ exchanges); PoP = normal CDF of drift/horizon-vol; Kelly f* = μ/σ².
FORMULA · WINDOW · SOURCE
Window: 20-day path from cached inference. Source: kronos-signals.json (ingestion Kronos pipeline). As of 2026-09-15. Zero-shot on SET — illustrative until walk-forward calibration passes.
SET
SET Index
฿1,578.66
20d Drift-12.7%
Ann. Vol16.0%
HOVER TO INSPECT TRAJECTORY
REDUCE / SHORTPoP: 0%
Negative bias detected. Tighten stops or allocate to cash hedges.
graded in public: direction 4/12, beat naive drift 1/12 → /forecast
REC. STOP (1D SD)
฿1558.28
KELLY STAKE
0.0%
PTT
PTT
฿42.5
20d Drift-17.4%
Ann. Vol18.9%
HOVER TO INSPECT TRAJECTORY
REDUCE / SHORTPoP: 0%
Negative bias detected. Tighten stops or allocate to cash hedges.
graded in public: direction 4/12, beat naive drift 1/12 → /forecast
REC. STOP (1D SD)
฿41.85
KELLY STAKE
0.0%
KBANK
Kasikorn Bank
฿252
20d Drift-22.8%
Ann. Vol19.4%
HOVER TO INSPECT TRAJECTORY
REDUCE / SHORTPoP: 0%
Negative bias detected. Tighten stops or allocate to cash hedges.
graded in public: direction 4/12, beat naive drift 1/12 → /forecast
REC. STOP (1D SD)
฿248.05
KELLY STAKE
0.0%
ADVANC
Advanced Info
฿353
20d Drift-0.9%
Ann. Vol20.8%
HOVER TO INSPECT TRAJECTORY
HOLD / CASHPoP: 44%
Unclear directional bias. Retain current allocation limit.
graded in public: direction 4/12, beat naive drift 1/12 → /forecast
REC. STOP (1D SD)
฿347.07
KELLY STAKE
0.0%
SCB
SCB
฿151
20d Drift-12.1%
Ann. Vol18.4%
HOVER TO INSPECT TRAJECTORY
REDUCE / SHORTPoP: 1%
Negative bias detected. Tighten stops or allocate to cash hedges.
graded in public: direction 4/12, beat naive drift 1/12 → /forecast
REC. STOP (1D SD)
฿148.75
KELLY STAKE
0.0%
BBL
Bangkok Bank
฿192.5
20d Drift-16.6%
Ann. Vol19.0%
HOVER TO INSPECT TRAJECTORY
REDUCE / SHORTPoP: 0%
Negative bias detected. Tighten stops or allocate to cash hedges.
graded in public: direction 4/12, beat naive drift 1/12 → /forecast
REC. STOP (1D SD)
฿189.55
KELLY STAKE
0.0%
MATHEMATICAL SPECIFICATIONS
· Zero-shot transfer from global pretrained weights — SET-specific dynamics not learned.
· 8 sample paths averaged — reduces noise but not systematic bias.
· Not validated against SET forward returns yet. Treat as illustrative.
· Bands are vol-cone (95% CI), not model uncertainty — they frame the range, not confidence.
· Expected Returns & Volatility calculated over 20-day path. Continuous-time Kelly Criterion sizing assumes log-normal returns.
EARTH SIGNAL· AlphaEarth Foundations (DeepMind, 2025) → SET tickers
○ AWAITING EARTH ENGINE PULL
64-dimensional per-pixel embeddings fuse optical, radar, LiDAR and climate. Year-over-year embedding shift over an area = physical change on the ground. Each economic site below maps to SET tickers; the change magnitude is a leading, satellite-derived activity signal — alt-data that precedes the financial print.
Planted area + harvest cycle → sugar / cassava output
→ KSL · BRR · KTIS
Utility-scale solar sitesEnergy
—
New PV-array footprint → renewable capacity additions
→ GULF · GPSC · BGRIM
Dataset GOOGLE/SATELLITE_EMBEDDING/V1/ANNUAL. To populate the change metrics: pip install earthengine-api, earthengine authenticate, then run ingestion/alphaearth.py on the M3. No numbers are shown until real embeddings are pulled — never fabricated.