Capital Flow Methodology
Gazzetta di Kyiv tracks institutional capital flows across asset classes, sectors, and geographies. Every story, bet, and signal on this site is derived from verifiable flow data — not sentiment, not narrative, not "what if." The thesis is simple: capital moves first, stories follow, prices confirm.
1. Data Sources
Flow data is aggregated from:
- EPFR Global — Weekly fund flows across equity, bond, and sector ETFs (institutional-grade, $50T+ tracked AUM)
- Morningstar Direct — Mutual fund and ETF flow data with style-box segmentation
- Internal aggregation — Cross-referenced with public filings (13F), options flow (OI skew), and derivatives positioning (COT reports)
Flow velocity is measured against a 4-week rolling average. "Normal pace" = within 0.8-1.5× the 4w average. Values above 2× are flagged as anomalous — these are where the signal is strongest.
2. Projection Confidence Model
The confidence percentage displayed on each flow claim is computed, not hardcoded. The model uses four inputs:
confidence = min(50 + flow_score + pace_score + positioning_score + contradiction_score, 95)
| Input |
Range |
Logic |
| Flow magnitude ($B) | +3 to +15 pts | $5B+ = +15, $3-5B = +12, $1-3B = +8 |
| Pace multiplier | +3 to +12 pts | 3×+ = +12, 2-3× = +10, 1.5-2× = +7 |
| Institutional positioning | +2 to +10 pts | Accumulating = +10, Distributing = +8, Hedging = +5 |
| Contradiction score | +2 to +8 pts | Score ≥70 = +8, 50-69 = +5 |
Confidence is capped at 95%. A claim with perfect inputs still can't be 100% — capital flows lag, reclassifications happen, and data revisions are real. The floor is 50% — a flow claim without external corroboration has at least coin-flip odds of containing signal.
The hero stat "Model confidence" is the arithmetic mean of all active flow-item confidences, rounded to the nearest integer.
3. Contradiction Score (0-100)
Every story receives a Contradiction Score measuring the tension between what consensus says and what flows reveal:
contradiction = 30 + narrative_tension (0-30) + flow_divergence (0-25) + extremum_quality (0-15)
- Narrative tension (0-30): Contrast markers in "They Say vs Reality" text. Each marker (but, however, not, yet, despite, etc.) adds 5 pts. Substantive pushback (reality >50 chars, >70% of they-say length) adds bonus.
- Flow divergence (0-25): Bullish consensus + outflow = +20 pts. Bearish consensus + inflow = +20 pts. Flow magnitude $5B+ adds +10.
- Extremum quality (0-15): Winner/Loser pair = +5, Idiot/Genius pair = +5, both present = +5.
4. ATR-Based Stop Levels
All 14 anchor assets use volatility-adjusted stops derived from approximate 14-day Average True Range (ATR):
stop(BUY) = entry_price − (entry_price × ATR% × multiplier)
stop(SELL) = entry_price + (entry_price × ATR% × multiplier)
Multipliers range from 2.0× (high-volatility assets like crypto, NVDA) to 3.0× (low-volatility like DXY). This ensures low-vol trades aren't stopped out by noise and high-vol trades have stops wide enough to breathe. The ATR percentage and multiplier are displayed on hover for each asset stop level.
5. Triangulation Score (0-100)
The Signal (Container 3) cross-references three layers:
triangulation = flow_alignment (0-50) + bet_conviction (0-30) + event_strength (0-20)
- Flow alignment (max 50): Amount tier + velocity tier + positioning boost. $5B+ = +20, 3× pace = +15, accumulating = +10. Capital is the prime mover.
- Bet conviction (max 30): Directional bias (BUY/SELL vs WATCH) = +15, HIGH conviction = +10, MED = +5.
- Event strength (max 20): High story confidence = +10, contradiction present = +5, extremum present = +5.
- Alignment bonus: Inflow+BUY or Outflow+SELL = +5. Divergence (Inflow+SELL or Outflow+BUY) = flagged as divergent signal.
6. Track Record
Every ANCHOR_ASSETS prediction is timestamped daily to localStorage. Settled predictions are scored when the bet hits target or stop. The track record displays:
- Win Rate: % of settled bets that realized positive P&L
- Total P&L: Sum of all realized returns (%)
- Expectancy: (Win Rate × Avg Win) + ((1 − Win Rate) × Avg Loss)
- Avg Win / Avg Loss: Mean return of winning and losing bets
All stats are computed client-side from localStorage. No cherry-picking is possible — the record is cumulative and auditable by inspecting browser storage.
7. Limitations
- Flow data lag: EPFR/Morningstar data is weekly with a 3-5 day reporting delay. Timestamps reflect week ending Friday.
- Static site constraint: Asset prices update with site rebuilds. Real-time pricing would require an API layer (planned for premium tier).
- Survivorship: Track record only includes bets settled locally. It does not account for early exits, position adjustments, or sizing decisions.
- No backtesting: The confidence model and triangulation scoring have not been backtested against historical outcomes. They represent a structured framework, not a validated quantitative model.
Disclaimer: Gazzetta di Kyiv is an editorial intelligence product, not investment advice. Capital flow data is descriptive, not predictive. Past flows do not guarantee future performance. All bets, conviction levels, and stop levels represent editorial analysis through a gambler's framework — they are not recommendations to buy or sell any security. Consult a qualified financial advisor before making investment decisions.