#55 — Dynamic position sizing

2026-05-27/28 (Session 078); extended to accounts 001-004 in Session 079 and later to 009/010 at x1.25-1.3

Verdict: mixed · 🟡 mixed — works mechanically, scales hidden risk

Sizing to settled balance (units = max(1, round(balance x 1.25)), refreshed hourly) keeps the margin-utilization fraction constant — but on a TP-only/no-SL pipeline it compounds a hidden catastrophic tail.

Hypothesis — what it tests

Whether computing position size as a constant fraction of settled account balance keeps the per-trade risk fraction and margin utilization constant as the account compounds.

Description

An infrastructure change rather than a signal experiment: the live momentum services (fx-price-mom-live on 011, fx-sma-live on 012, then the four Session-079 variant services) switched from hardcoded unit counts to dynamic sizing computed from the OANDA settled balance — units = max(1, round(balance x 1.25)), refreshed every 3600s. Using balance rather than NAV avoids a feedback loop where unrealized drawdown on open positions cuts size mid-cycle; per-trade units are stored on the PairState so DB records stay exact. Mechanically it worked (011 $20.22 → 25u; 012 $22.25 → 28u), and margin utilization stays roughly constant (~58-60%) as the account grows or shrinks. The audit-table verdict is mixed because the thing being sized was the TP-only/no-SL momentum book: constant-fraction sizing on a strategy family with an unbounded adverse tail scales the hidden risk right along with the wins — the tail the finite-margin re-validation (entry 57) later exposed. The same formula (at 1.25-1.3x) was later reused on the bounded-risk retrace (009) and SMA-Stack (010) services.

Key result

units = max(1, round(balance x 1.25)) ≈ 25u per $20 settled balance; margin util held ~58-60% constant; on restart 011 $20.22→25u, 012 $22.25→28u (down from manual 35u)

Algorithms

constant-fraction position sizing rule (balance-proportional, hourly refresh)

Code

services/strategy_price_mom_live/main.py services/strategy_sma_live/main.py

Interactive version (search, filters, figures): the experiment explorer.

← #54 MTF/wave/wavelet confluence + Harvester #56 Portfolio variation campaign (8 variants) →

This page is one experiment from the audited record behind the book The Edge That Wasn't — get the book · about the project.