All four strategies
Same v23+v32 entry signals, different sizing rules — Sharpe / return / drawdown trade-offs locked in by the 2026-05-17 sizing experiment.
Equity comparison · since inception
All four portfolios started with 100M IDR on the same day. Each line is the live cash + mark-to-market for that mode.
Side-by-side
Current state of each strategy. Click any row to drill into that mode.
| Mode | Style | Current equity | P&L since Day 1 | Open | Pending | Closed |
|---|---|---|---|---|---|---|
| Loading… | ||||||
Equity since inception
Total portfolio value over time for this mode. Starts at 100M IDR — every change is a real trade outcome at live IDX prices.
Live portfolio
What this mode's bot is doing right now — current holdings plus orders queued for the next market open / close.
Open positions 0
All in cash — waiting for the next entry signal.
Pending buys 0
No buys queued.
Pending sells 0
No sells queued.
Recent trades
Last 20 closed positions for this mode. Honest about both winners and losers — every closed trade appears here.
| Symbol | Entry | Exit | Entry price | Exit price | P&L | Reason |
|---|
Today's BUY signals
Stocks the model is recommending right now. All four modes pick from this same list — only their sizing differs.
How it works
Two ML models run on closing data for 55 trained-universe IDX stocks plus a growing list of user-added symbols. A BUY signal fires when both agree:
- v23 — supervised MLP predicting 5-day forward returns from 35 hand-crafted features (price/volume, broker flow, market regime).
- v32 — RL meta-policy trained via Evolution Strategies. Takes v23's prediction plus context features and outputs a single decision: BUY or SKIP.
The four modes share that signal exactly. They differ only in position sizing and slot count:
- Conservative — 5.625% per position × 16 slots, fixed sizing on initial 100M. Lowest variance.
- Balanced — 10% × 8, equity-scaled (compounds with portfolio growth, auto-de-risks during drawdowns).
- Aggressive — 15% × 6, equity-scaled. Highest Sharpe in the menu.
- Extreme — 22.5% × 4, equity-scaled. Maximum return at the cost of -42% drawdowns in backtest.
All four use the same v32-driven exits + 10-day hard cap + -15% close-based stop. Fills use live prices from Stockbit and Yahoo Finance — no simulated slippage or hindsight optimization.
Historical statistics on unseen stocks (15-stock holdout × 24 random sim starts, 2018–2026): Balanced ~+22%/yr, -24% max DD, Sharpe 1.40. Extreme ~+40%/yr, -42% DD, Sharpe 1.34. Live frictions (slippage, market impact) typically erode 2–4 pp/yr — quote conservative numbers when reasoning about future returns. Past sim performance does not predict future live performance.