WAKE - Enhancement to public releases of Goldman, JPM and Amgren's Institutional Algorithms
- Sandra Wakefield
- 3 days ago
- 5 min read
Download the Full Whitepaper
WAKE is PSRC's institutional order-flow continuation engine. It watches the same benchmark large execution algorithms are measured against — session VWAP — and reads the order-flow signature (Cumulative Volume Delta) left behind as price moves away from it. The name reflects the core thesis: a large institutional order moving through the market leaves a wake, the same way a vessel does moving through water. WAKE is built to read that wake and trade the continuation, not to guess at direction from price alone.
This build is tuned and validated specifically for XAUUSD (Gold Spot / U.S. Dollar) on the 15-minute chart, New York session hours. It is not yet validated for other instruments or timeframes — see Section 7.
Beta Status This is a signal/alert engine (Pine Script indicator), not an auto-execution system. It places no orders. It generates alerts and a simulated position tracker intended for demo-account forward testing ahead of a planned live cTrader Ninjatrader, Python Automate port.
This system is derived and improved from VWAP / TWAP — baseline scheduling algos, publicly documented by every prime broker (Goldman, MS, JPM all publish white papers on their versions) |
2. Core Logic
2.1 The Institutional Benchmark Thesis
Large funds executing size are typically measured against VWAP — it is the standard institutional performance benchmark, formalized in execution literature such as Almgren & Chriss's optimal execution framework. When an institutional participant is actively working an order, price tends to gravitate around VWAP during the execution window, and the manner in which price departs from it — with or against real order-flow participation — carries information about whether that participation is continuing or exhausted.
2.2 Deviation Bands
WAKE computes session VWAP and a cumulative, session-anchored standard deviation band around it, resetting each trading day at 00:00 New York time. Bands are built from a running cumulative variance calculation rather than a fixed lookback, which keeps the calculation non-repainting — every value is fully determined by bars that have already closed. A move beyond the deviation band (default: 1.5 standard deviations) is the first trigger condition.
2.3 Order-Flow Confirmation (CVD)
Cumulative Volume Delta is reconstructed from confirmed 1-minute bars beneath the working chart, summing signed volume (up-close volume minus down-close volume) since the session open. This is the mechanism that distinguishes a deviation-band breach driven by genuine participation from one driven by a thin, low-conviction spike. WAKE requires CVD slope to be moving in the same direction as price (confirmation) before treating a deviation as tradeable.
2.4 Continuation Setup
A long setup requires price above the upper deviation band with CVD confirming (both price and order flow rising together). A short setup requires price below the lower band with CVD confirming downside participation. This is deliberately a continuation model, not a reversal model — WAKE trades with the confirmed institutional flow, not against it.
Why not fade the extremes? An earlier mean-reversion ("fade") branch — betting on reversion back to VWAP off a single-bar CVD divergence — was tested extensively and shelved. Across three rounds of testing it produced win rates as low as 0-20% with near-instant stop-outs. Continuation, not reversion, is where this system's validated edge lives. |
2.5 Exit Architecture
● Initial stop: VWAP-anchored, with a minimum-distance floor (ATR-based) so a compressed VWAP distance can never produce an unrealistically tight stop.
● Partial take-profit: a portion of the position is taken at a configurable R-multiple, with the remaining stop moved to breakeven.
● Trailing stop: once a trade reaches the trail-activation R-multiple, the fixed profit target is replaced by an ATR-based trailing stop that ratchets in the trade's favor only. This exists because early testing showed winners were being cut short by a fixed profit cap — trades that ran past the cap and got flattened at session end were, on average, still the most profitable bucket in the dataset.
● Session-end flatten: any open position is closed if it's still open when the execution window ends, removing overnight/gap risk entirely.
2.6 Session & Liquidity Structure
Trading is restricted to a defined New York-session execution window, with the final hour of that window excluded by default (institutional participation algorithms are typically "catching up" to fill remaining size in the last hour, producing mechanical, low-signal noise). Two additional blackout windows are excluded by default based directly on historical entry-hour performance: 06:00-07:00 NY (a pre-NY-open liquidity lull) and 09:00-10:00 NY (NY cash-open whipsaw).
2.7 Inter-Market & Macro Filters
Gold's dominant macro driver is the U.S. dollar and real yields. WAKE requires DXY (U.S. Dollar Index) to be moving in the direction that supports the trade thesis — DXY strength confirming a Gold short, DXY weakness confirming a Gold long — as an inter-market confluence check specific to this instrument. A recurring macro-release lockout also suppresses new signals around the clock-time windows CPI/NFP/PPI and FOMC releases typically occur in, plus an automatic volatility-spike lockout as a proxy for unscheduled news.
3. Why This Is An Institutional-Grade Approach
"Institutional-grade" is a specific claim, and it rests on four things this system was built to satisfy, not on a marketing label.
Grounded in published execution and microstructure research
The VWAP-as-benchmark thesis is not a retail indicator repurposed — it reflects how institutional execution quality is actually measured (Almgren-Chriss optimal execution framework). The order-flow confirmation layer draws on the same family of ideas as Kyle's Lambda and related market-microstructure research on informed order flow. WAKE does not claim to replicate any specific fund's proprietary signal — those are not public — but its foundations are the same public literature every serious execution desk is built on.
Engineering discipline: non-repainting by construction
Every signal-generating calculation in WAKE is built to resolve only on confirmed, closed bars. Session VWAP and its deviation bands use cumulative running sums rather than a repainting lookback window. CVD is reconstructed only from confirmed lower-timeframe bars. This matters because a system that repaints looks better in a backtest than it will ever perform live — the two most common ways retail indicators inflate backtest results are repainting and lookahead bias, and WAKE was explicitly audited against both.
A tiered protection stack, not a single stop-loss
Tier | Protections |
Tier 1 — Capital | Max daily loss, max consecutive losses, max trades/day, kill switch |
Tier 2 — Market Condition | ATR volatility floor, spread/noise proxy filter, session & blackout windows |
Tier 3 — Signal Quality | CVD order-flow confirmation, DXY inter-market confluence, macro news lockout |
Validated through an iterative, adversarial testing process — not accepted on the first backtest
WAKE went through multiple rounds of structured audit before reaching this build, including a position-sizing defect that was allowing individual trades to reach 75-100% of account equity notional (found and fixed before any live consideration), a stop-placement bug producing near-instant stop-outs on countertrend entries, and — most importantly — a Daily Trend Filter that looked like an improvement in isolated testing but was later found, on full-history data, to have silently suppressed nearly all trading activity through 2024-2025. That filter was shelved rather than kept for a better-looking backtest number. This history is disclosed deliberately: an institutional process is one that catches and corrects its own mistakes before capital is at risk, and shows its work rather than only its conclusions.
4. Performance Summary — Read This Honestly
This is historical, in-sample data, not a forward-tested result. The figures below come from the current frozen configuration run against the full available XAUUSD history. This is disclosed transparently because it's the best evidence available so far — but it is not the same as forward performance, and it should not be read as a guarantee. The demo forward test you are about to run is the real validation gate, not this table. |
XAUUSD, Short-Only Continuation, Current Configuration (2020 – 2026 available history)
Metric | Value |
Trades | 579 |
Win rate | 72.% |
Average win / average loss | $92.55 / -$23.61 |
Expectancy per trade | +$60.03 |
Profit Factor | 10.08 |
Max drawdown | 1.89% |
Download the Full Whitepaper
Comments