stockback / Experiment result
EXP-003 — bnf_daily_c4_baseline daily validation
21 years of Samsung daily data: 590 trades, -94.70% versus +2539% buy-and-hold, with unreachable targets and costs dominating the signal.
Original record: 2026-08-20 · Migrated: 2026-10-27 · Translated from Korean: 2026-10-27
A historical record, not rerun in this migration.The figures, hypotheses and verdicts of the time are preserved. A past positive result or candidate mention does not mean validated profitability today or a live-trading recommendation. Validation periods that were already used are not treated as new independent validation.
This is an English translation of a preserved Korean original. It describes conditions at the time of writing, not current provider specifications, costs or regulations.
EXP-003 — bnf_daily_c4_baseline daily validation
Status: draft. A BNF daily baseline that isolates only the C4 MACD rebound and a conservative daily exit
Grounds for the verdict
- Blocking
TARGET_UNREACHABLE — the target was reached in only 18/590 trades (3.1%). The setup requires an average stop distance of 3.56% × 3.0R = +10.68% within 3 trading days. reward_to_risk and max_holding_days contradict each other, so this backtest effectively measured a '3-day hold then time-exit' strategy, not the configured 3.0R strategy. - Blocking
BENCHMARK_SHORTFALL — simply holding the same symbol over the same period would have turned final capital into 26.39x. The strategy ended at 0.053x, trailing by 497.8x on a multiple basis (strategy -94.70% / benchmark 2539.09%). A long-only strategy trailing its underlying this much means the signal systematically misses up-moves or costs exceed the size of the signal. Note: simple subtraction of the two values is meaningless over long compounding periods, so they are compared as multiples. - Blocking
COST_DOMINATED — the expected gross return per trade is -0.06%, while the round-trip cost is 0.40%. Costs exceed the average move the signal produces, so this result measures the cost structure more than the merit of the signal. At this frequency and holding period, whether the signal has an edge cannot be judged at all. It must be measured again after lengthening the holding period or raising the signal threshold so that the expected move per trade is several times the cost. - Blocking
NEGATIVE_EXPECTANCY — expectancy per trade is negative at -0.46%. The structure loses more the more it trades, so the signal and exit rules themselves must be reviewed, not the parameters searched. - Warning
SIGNAL_ATTRITION — 50.7% of 1,196 generated signals did not become trades (590 because a position was held, 16 invalidated by gap-down). The single-position constraint determines a large part of strategy performance, so signal quality and position-management effects must be measured separately. - Warning
LOSS_STREAK — maximum losing streak of 16 trades. Surviving such a stretch in live trading requires sizing and stop rules to be defined first.
Reproducibility
| Item | Value |
|---|
| experiment_id | EXP-003 |
| strategy_id / version | bnf_daily_c4_baseline / 0.1.0-draft |
| code_revision | 37ec9e0+D009..D011 |
| config_hash | 11f7092194687a6e |
| data_manifest_hash | b10f653a2ee6c1f9 |
| input | data/raw/creon/A005930-D-creon-20260811T060800Z.csv |
| instrument | A005930 |
| period | 2005-01-03~2026-08-11 (5,331 bars / 21.60 years) |
| data quality | 0 duplicates · 0 excluded |
| execution_model | Next-day open entry · same-bar stop first · 3-trading-day time exit · target 3.0R |
| cost_model | 20.0bp one-way + 0.0bp slippage |
| random_seed | n/a (deterministic run) |
| generated_at | 2026-08-20T09:50:30+00:00 |
Results
| Metric | Strategy | Buy-and-hold |
|---|
| Total return (net) | -94.70% | 2539.09% |
| Final capital multiple | 0.053x | 26.39x |
| CAGR | -12.71% | 16.36% |
| Max drawdown | -94.91% | -46.66% |
| Metric | Value |
|---|
| Completed trades | 590 |
| Wins / losses | 232 / 358 |
| Win rate | 39.3% |
| Profit factor | 0.645 |
| Expectancy / trade | -0.46% |
| Average winning trade | 2.13% |
| Average losing trade | -2.14% |
| Payoff ratio | 0.996 |
| Average realized R | -0.224 (target 3.0R) |
| Average stop distance | 3.56% |
| Maximum losing streak | 16 trades |
| Maximum drawdown period | 2005-08-12 → 2026-08-03 |
| Longest drawdown duration | 5,178 bars |
| Market exposure | 29.5% |
| Average hold | 2.5 days |
| Annual volatility | 14.6% |
| Sharpe-like value (rf=0) | -0.88 |
Signal attrition
| Stage | Count | Share |
|---|
| Signals generated | 1,196 | 100.0% |
| Entries filled | 590 | 49.3% |
| Discarded because a position was held | 590 | 49.3% |
| Invalidated by gap-down | 16 | 1.3% |
| No entry bar | 0 | 0.0% |
Breakdown by exit reason
| Exit reason | Trades | Wins | Win rate | Total net P&L | Mean | Best | Worst |
|---|
| stop | 168 | 0 | 0.0% | -422.36% | -2.51% | -0.40% | -15.33% |
| target | 18 | 18 | 100.0% | 59.87% | 3.33% | 5.57% | 0.63% |
| time | 404 | 214 | 53.0% | 91.04% | 0.23% | 9.10% | -9.96% |
Breakdown by year
| Year | Trades | Wins | Win rate | Total net P&L | Mean | Best | Worst |
|---|
| 2005 | 25 | 12 | 48.0% | -1.76% | -0.07% | 4.63% | -3.71% |
| 2006 | 29 | 8 | 27.6% | -27.15% | -0.94% | 4.41% | -4.85% |
| 2007 | 26 | 14 | 53.8% | 4.94% | 0.19% | 9.10% | -4.92% |
| 2008 | 28 | 12 | 42.9% | 4.46% | 0.16% | 8.97% | -6.93% |
| 2009 | 29 | 8 | 27.6% | -32.68% | -1.13% | 4.81% | -5.10% |
| 2010 | 23 | 9 | 39.1% | -10.27% | -0.45% | 3.61% | -3.82% |
| 2011 | 25 | 9 | 36.0% | -14.72% | -0.59% | 4.27% | -4.79% |
| 2012 | 26 | 13 | 50.0% | 6.46% | 0.25% | 5.51% | -3.86% |
| 2013 | 30 | 13 | 43.3% | -6.44% | -0.21% | 4.54% | -4.83% |
| 2014 | 31 | 7 | 22.6% | -25.83% | -0.83% | 3.69% | -3.84% |
| 2015 | 35 | 9 | 25.7% | -34.52% | -0.99% | 3.19% | -6.36% |
| 2016 | 27 | 12 | 44.4% | 5.15% | 0.19% | 5.83% | -4.81% |
| 2017 | 30 | 16 | 53.3% | 8.92% | 0.30% | 5.32% | -3.29% |
| 2018 | 30 | 12 | 40.0% | -15.25% | -0.51% | 4.30% | -4.49% |
| 2019 | 24 | 13 | 54.2% | 6.03% | 0.25% | 5.79% | -4.40% |
| 2020 | 23 | 17 | 73.9% | 12.21% | 0.53% | 5.25% | -5.51% |
| 2021 | 26 | 6 | 23.1% | -34.65% | -1.33% | 2.82% | -6.68% |
| 2022 | 27 | 7 | 25.9% | -38.18% | -1.41% | 2.18% | -4.82% |
| 2023 | 30 | 10 | 33.3% | -8.72% | -0.29% | 5.50% | -3.78% |
| 2024 | 27 | 7 | 25.9% | -19.53% | -0.72% | 5.49% | -3.65% |
| 2025 | 24 | 13 | 54.2% | 10.43% | 0.43% | 5.08% | -6.01% |
| 2026 | 15 | 5 | 33.3% | -60.37% | -4.02% | 6.26% | -15.33% |
Parameter sensitivity
Total return (net, single-position compounding) · trade counts in parentheses.
| max_holding_days \ reward_to_risk | 1R | 1.5R | 2R | 3R | 5R |
|---|
| 2 | -97.5% (749) | -97.2% (738) | -96.7% (731) | -96.0% (724) | -95.7% (723) |
| 3 | -96.7% (624) | -94.3% (606) | -95.1% (597) | -94.7% (590) | -94.1% (586) |
| 5 | -94.3% (520) | -88.8% (496) | -85.5% (479) | -84.6% (469) | -84.4% (464) |
| 10 | -86.2% (438) | -77.8% (411) | -73.7% (394) | -55.6% (384) | -48.4% (377) |
| 20 | -88.9% (403) | -85.3% (375) | -85.5% (354) | -75.4% (344) | -74.1% (337) |
This table is not for choosing optimal parameters. It shows in which direction and by how much results move when only the exit rule changes on the same signals; adopting the best cell here would be overfitting to a single symbol and a single period.
Interpretation limits
- A single-symbol, single-account result. Survivorship, delisting and universe bias are not addressed.
- It assumes single-position compounding allocating 100% of capital to each trade. Sizing, diversification and multi-strategy capital competition are not reflected.
- Based on adjusted-price OHLCV, so real order-book depth, queues, partial fills, VI and unfilled orders at price limits are not verified.
- Walk-forward splits and out-of-sample validation have not been performed yet.
- This result alone does not decide strategy adoption or a move to live trading.
Elements not included: C1, C2, C3, sector thresholds, regime controller, multi-symbol universe
Source and migration evidence
Source: reports/experiments/EXP-003-bnf_daily_c4_baseline.md
Original SHA-256: d4da90eaff784e3195d5b113216ad7d10719e0e1cfb422cf007b16101df3b039
Raw data and account information are not part of the public package. The original results do not replace statistical or scientific revalidation. The English text is a translation; numbers, tables and charts are carried over unchanged.